United States
           Environmental Protection
           Agency
Office of Pollution
Prevention and Toxics
Washington. DC 20460
EPA 747-R-00-004
December, 2000
EPA     Risk Analysis to Support Standards for
           Lead in Paint, Dust, and Soil
           Supplemental Report
           VOLUME I: Chapters 1 to 7
            110
            Its-
            5-
                 \
      I'd

-------

-------
                                      EPA 747-R-00-004
                                       December, 2000
RISK ANALYSIS TO SUPPORT STANDARDS
  FOR LEAD IN PAINT, DUST, AND SOIL

        SUPPLEMENTAL REPORT
     VOLUME I: CHAPTERS 1 TO 7
                    U.S. EPA Headquarters Library
                          Mail code 3201
                    1200 Pennsylvania Avenue NW
                      Washington DC 20460
National Program Chemicals Division (7404)
  Office of Pollution Prevention and Toxics
   U.S. Environmental Protection Agency
         Washington, DC 20460

    EPA Contract Number 68-W-99-033

-------
                             DISCLAIMER

The material in this document has been subject to Agency technical and policy
review and approved for publication as an EPA report. Views expressed by the
individual authors, however, are their own and do not necessarily reflect those
of the U.S. Environmental Protection Agency.   Mention of trade  names,
products,  or services does not convey, and should hot be interpreted as
conveying, official  EPA approval, endorsement, or recommendation.

                 This report is copied on recycled paper.

-------
                         CONTRIBUTING ORGANIZATIONS

      This report is a supplement to EPA 747-R-97-006 ("Risk Analysis to Support Standards
for Lead in Paint, Dust, and Soil"). Efforts to produce this report were funded and managed by
the U.S. Environmental Protection Agency.  The risk analysis was conducted by Battelle
Memorial Institute under contract to the U.S. Environmental Protection Agency. Each
organization's responsibilities are listed below.
                        Battelle Memorial Institute (Battelle)

      Battelle was responsible for performing the additional data analyses, literature reviews,
and documentation presented in this report. Battelle was also responsible for preparing this
report.
                    U.S. Environmental Protection Agency (EPA)

      The Environmental Protection Agency was responsible for providing direction on the
technical issues to be presented in this report, providing relevant information for the report,
reviewing the report, contributing to the development of conclusions, and managing the peer
review and publication of the report. The EPA Work Assignment Manager was Mr. Ronald
Morony. The Deputy Work Assignment Managers were Mr. Brad Schultz and Mr. Dave
Topping. The EPA Project Officer was Ms. Sineta Wooten.
                                         iii

-------
This page intentionally blank.
             IV

-------
                                 TABLE OF CONTENTS
                                                                                     Page
EXECUTIVE SUMMARY	xxi

1.0    INTRODUCTION  	 1
       1.1    Peer Review	 2

2.0    HAZARD IDENTIFICATION	 5
       2.1    Review of the Adverse Health Effects of Lead Exposure, With
              a Focus on Neurological Effects, As Observed in Animal Studies	 5
              2.1.1   Approach to Reporting on the Findings of Animal Studies 	 6
              2.1.2   Neurological, Behavioral, and Developmental Health Effects	 7
                     2.1.2.1   Physiological Effects of Lead on the
                              Neurological System  	  11
                     2.1.2.2   Behavioral and Developmental Effects of Lead	  18
              2.1.3   General Health Effects  	  24
                     2.1.3.1   Death  	  25
                     2.1.3.2   Systemic Effects	  25
                     2.1.3.3   Immunological Effects 	  26
                     2.1.3.4   Reproductive and Genotoxic Effects	  26
                     2.1.3.5   Carcinogenicity	  26
              2.1.4   Conclusions from Animal Studies Investigation	  27
       2.2    Support for  the Causality of Adverse Health Effects Due to
              Lead Exposure	  29
              2.2.1   Principles of Causality	  29
              2.2.2   Causality as Addressed in Longitudinal Studies  	  31
              2.2.3   Conclusions on Causality  	  35
       2.3    The Association Between Blood-Lead  Concentration and
              IQ Score  	  35
              2.3.1   Linearity and Slope Assumptions	  36
              2.3.2   Threshold Assumption  	  40
              2.3.3   Verifying the Results of Schwartz (1994)  	  46
       2.4    Impact of Certain Residential Dust Characteristics
              on Dust-Lead Exposure  	  46
              2.4.1   Review of Literature: Effects of Chemical Composition
                     on Lead Bioavaitability in Dust	  47
                     2.4.1.1   Research on Lead Bioavailability in
                              Controlled Animal Studies	  47
                     2.4.1.2   Research on Lead Bioavailability in Soils	  48
              2.4.2   Review of Literature: Effects of Particle Size
                     on Lead Bioavailability in Dust	  49
              2.4.3   Information Gaps, Issues and Conclusions	  49

3.0    EXPOSURE ASSESSMENT  	  51
       3.1    The National Survey of Lead and Allergens in Housing	  52
       3.2    Comparison of Environmental-Lead Levels in the HUD
              National Survey with Those of Other Key Studies	  61
                                                           US EPA Headquarters Library
                                                              "    Mail code 3201
                                                           1200 Pennsylvania Avenue
                                                              Washington DC 204eu

-------
                           TABLE OF CONTENTS
                                 (Continued)
       3.2.1   Characterizing Dust-Lead Loadings on Floors and
              Window Sills	 63
              3.2.1.1   Data Summaries for the §403 Risk Analysis
                       Versus the Interim NSLAH  	 64
              3.2.1.2   Data Summaries for the §403 Risk Analysis
                       Versus Three Other  Studies  	 75
              3.2.1.3   Calculating National Exceedance Percentages
                       For Household Average Floor Dust-Lead Loading	  . 89
              3.2.1.4   Interpreting the Observed Differences with
                       Other Studies	 96
              3.2.1.5   Conclusions of the Dust-Lead Data Comparisons	 97
       3.2.2   Characterizing Soil-Lead Concentrations  	 98
              3.2.2.1   Data Summaries for the §403 Risk Analysis
                       Versus the Interim NSLAH  	  102
              3.2.2.2   Data Summaries for the §403 Risk Analysis
                       Versus Other Studies	  110
              3.2.2.3   Calculating National Exceedance Percentages
                       For Yardwide Average Soil-Lead Concentration	  128
              3.2.2.4   Interpreting the Observed Differences with
                       Other Studies	  142
              3.2.2.5   Conclusions of the Soil-Lead Data Comparisons	  145
3.3    Evaluation of Soil Pica in Children	'. .  . . .  146
       3.3.1   What is Soil Pica?  	  146
       3.3.2   How Does the §403 Risk Analysis Account for Soil Pica?	  151
       3.3.3   Prevalence of Soil Pica Behavior	  152
              3.3.3.1   Literature Review  	  152
              3.3.3.2   Prevalence of Soil Pica  Separate from
                       Paint Pica	  153
       3.3.4   Estimating the Frequency of Ingestion and Amount
              of Soil Ingested by Children Who Exhibit Soil Pica   	  155
       3.3.5   Conclusions on Soil Pica	  158
3.4    Characterizing the Population of Children  in  the Nation's
       Housing Stock   	  158
3.5    Summaries of Dust-Lead Levels on Surfaces Other than Uncarpeted
       Floors and Window Sills	  160
       3.5.1   Distribution of Dust-Lead on Surfaces Other than Floors
              And Window Sills	  161
       3.5.2   Evidence of a Relationship Between  Children's Blood-Lead
              Concentrations and Dust-Lead  on  Surfaces Other Than
              Floors and Window Sills  	  173
       3.5.3   Implications of the Available Information for
              Regulatory Standards	  176
3.6    Distribution of Childhood Blood-Lead	  178
       3.6.1   Evaluation of the HUD Lead-Based Paint Hazard Control
              Grant Program ("HUD Grantees")	  179
                                     vi

-------
                                 TABLE OF CONTENTS
                                       (Continued)
              3.6.2  Evidence of the Impact of Mousing Age/Condition
                    on Blood-Lead Concentration	  185

4.0    DOSE-RESPONSE ASSESSMENT	  187
       4.1     HUD Model  	  188
              4.1.1  Form of the HUD Model  	  188
              4.1.2  Development of the HUD Model	  189
              4.1.3  Interpreting Results of Fitting the HUD Model  	  194
              4.1.4  Conclusions	  196
       4.2    Alternative Multimedia Models for Predicting a Geometric
              Mean Blood-Lead Concentration Based on Environmental-
              Lead Levels		  196
       4.3    Supplemental Information on Model-Based Approaches in
              the §403 Risk Analysis	  199
              4.3.1  The "Scaling' Algorithm Used to Determine a
                    Post-Intervention Blood-Lead Concentration Distribution 	  199
              4.3.2  Adjusting the Empirical Model Parameter Estimates
                    To Reflect  Measurement Error	  202

5.0    RISK CHARACTERIZATION	  205
       5.1     Risk Characterization Sensitivity and Uncertainty Analysis	  206
              5.1.1  Estimates of Individual Risks from Applying the HUD Model ........  206
              5.1.2  Estimates of Individual Risks from Applying the
                    Alternative Rochester Multimedia Model	  215
              5.1.3  Considering Potential Declines in Blood-Lead Concentration
                    from NHANES III Phase 2 Measures	  218
              5.1.4  Considering How Baseline Environmental-Lead  Levels May
                    Have Changed Since the HUD National Survey	  220
              5.1.5  Impact on the Estimated Incidence of IQ Point  Decrement
                    Assuming Certain Thresholds on the IQ/Blood-Lead
                    Relationship	  223

6.0    ANALYSIS OF EXAMPLE OPTIONS FOR THE §403 STANDARDS 	  227
       6.1     Performance Characteristics Analyses	  228
              6.1.1  Data Used  in the Performance Characteristics Analysis	  228
              6.1.2  Analysis Approach	  231
              6.1.3  Results Cited in the §403 Proposed Rule	  235
              6.1.4  Results of Analysis on Specified Sets of Standards	  238
                    6.1.4.1  Analyses performed on 41  Combinations of
                              Candidate Standards, in Three Iterations	  238
                    6.1.4.2   Considering Only Soil and Dust Standards   	  253
                    6.1.4.3   Analysis Involving Only Dust-Lead Standards and
                              a Standard on the Amount of Deteriorated Paint	  261
       6.2    Investigating Incidence of Elevated Blood-Lead Concentration
              in Housing Units Meeting All Example Options for Standards  	  268
              6.2.1  The Model-Based Approach	  268
                                           VII

-------
                                  TABLE OF CONTENTS
                                       (Continued)
                                                                                    Page
              6.2.2  Examples of Applying the Model-Based Approach	  269
       6.3    Review of Published Information on Post-Intervention
              Dust-Lead Loadings	  271
              6.3.1   Post-Intervention Floor Dust-Lead Loadings	  272
              6.3.2  Post-Intervention Window Sill Dust-Lead Loadings	  272
       6.4    Sensitivity and Uncertainty Analyses for Risk Management
              Analyses	  275
              6.4.1   Considering How Baseline Environmental-Lead Levels
                     May Have Changed Since the HUD National Survey	  275
              6.4.2  Impact on the Estimated Incidence of IQ Point
                     Decrement Assuming Certain Thresholds on the
                     IQ/Blood-Lead Relationship  	  278
              6.4.3  Considering Alternative Assumptions on Post-Intervention
                     Dust-Lead Loadings  	  278
              6.4.4  Characterizing the  Post-Intervention Blood-Lead Distribution
                     Based on Relative Change from Baseline in the Geometric
                     Mean and the Probability of a Child's Blood-Lead
                     Concentration Exceeding 10 //g/dL	  283
       6.5    Lead Exposure Associated with Carpeted Floor-Dust	  285

7.0    REFERENCES	  291


                             LIST OF APPENDICES (VOLUME II)

Appendix A.   Glossary to Section 2.1	A-1
Appendix B.   Calculating Average IQ Decrement Assuming a Non-zero
              Threshold on the IQ/Blood-Lead Concentration Relationship  	B-1
Appendix C.   Method to Imputing Household Average Environmental-Lead
              Levels for Housing Units in the National Survey of Lead and
              Allergens in Housing (NSLAH)  	C-1
Appendix D1.  Summaries of Interim Dust-Lead Loading Data from the
              National Survey of Lead and Allergens in Housing (NSLAH),
              Where Imputed Data are Excluded	D1-1
Appendix D2.  Summaries of Interim Yard-Wide Average Soil-Lead Concentration
              Data from the National Survey of Lead and Allergens in Housing
              (NSLAH), Where Imputed Data are Excluded  	D2-1
Appendix E.   Method to Estimating Total Soil-Lead Concentration from
              Analytical Results for Fine and Coarse Soil Fractions	E-1
Appendix F.   Comparison and Contrast of  Risk Estimates from the HUD
              Model and the Rochester Multimedia Model Developed in
              the §403 Risk Analysis	F-1
Appendix G.   Performance Characteristics  Analysis Cited in the
              §403 Proposed Rule	G-1
Appendix H.   Review of Published Information on Post-Intervention Wipe
              Dust-Lead Loadings on Floors and Window Sills  	H-1
                                           VIII

-------
                                 TABLE OF CONTENTS
                                       (Continued)
                                                                                    Pane
Appendix I.    An Assessment of Dust-Lead Levels in Carpeted Floors and
              Their Relation to Children's Blood-Lead Concentration, Using Data
              from the Rochester Lead-in-Dust Study and the HUD Grantees
              Program Evaluation	  1-1

Appendix J.   Additional Performance Characteristics Analyses, Where
              Candidate Standards for Lead in Play-area Soils Are Considered  	J-1

                                     LIST OF TABLES

Table 2-1.     Summary of Lead Exposure Levels and Key Findings for Selected
              Animal Studies	  8

Table 2-2.     Summary of Key Findings from Studies that Investigate the
              Relationship Between Blood-Lead Concentration and IQ Score  	  38

Table 3-1.     Differences in Approaches and Outcomes Between the HUD
              National Survey of Lead-Based Paint in Housing and the
              HUD National Survey of Lead and Allergens in Housing	  54

Table 3-2.     Estimated Number of Occupied Housing Units in the U.S.
              Housing Stock Within Year-Built Categories, According to
              Four Recent Surveys and/or Analyses  . .	  59

Table 3-3.     Estimated Number of Occupied Housing Units in the U.S.
              Housing Stock Within Each Census Region, According to
              Four Recent Surveys and/or Analyses  	  60

Table 3-4.     Descriptive Statistics of Area-Weighted Average Floor Wipe
              Dust-Lead Loadings for Households, As Reported in the
              §403 Risk Analysis Versus the Interim NSLAH Data 	  65

Table 3-5.     Descriptive  Statistics of Area-Weighted Average Window Sill
              Wipe Dust-Lead Loadings for Households, As Reported in the
              §403 Risk Analysis Versus the Interim NSLAH Data 	  65

Table 3-6.     Descriptive  Statistics of Area-Weighted Average Floor Wipe
              Dust-Lead Loadings for Households, Presented by Housing Age
              Category. As Reported in the §403 Risk Analysis Versus
              the Interim NSLAH Data  	  69

Table 3-7.     Descriptive  Statistics of Area-Weighted Average Window Sill
              Wipe Dust-Lead Loadings for Households, Presented by.
              Housing Age Category. As Reported in the §403 Risk Analysis
              Versus the Interim NSLAH Data 	  70
                                            IX

-------
Table 3-8.
Table 3-9.
Table 3-1 Oa.
Table 3-1 Ob.
Table 3-11 a.
Table 3-11b.
Table 3-12.
Table 3-13.
                                  TABLE OF CONTENTS
                                        (Continued)
Descriptive Statistics of Area-Weighted Average Floor Wipe
Dust-Lead Loadings for Households, Presented by Census
Region. As Reported in the §403 Risk Analysis Versus the
Interim NSLAH Data  	
                                                                                      73
Descriptive Statistics of Area-Weighted Average Window Sill
Wipe Dust-Lead Loadings for Households, Presented by Census
Region. As Reported in the §403 Risk Analysis Versus the
Interim NSLAH Data  	
                                                                                      74
Descriptive Statistics of Area-Weighted Average Floor Wipe
Dust-Lead Loadings for Households, Presented bv Housing Age
and Census Region, As Reported in the §403 Risk Analysis
Versus the Interim NSLAH Data Where No Adjustments Were
Made to Not-Detected Results	
                                                                                      78
Descriptive Statistics of Area-Weighted Average Floor Wipe
Dust-Lead Loadings for Households, Presented bv Housing Age
and Census Region. As Reported in the §403 Risk Analysis
Versus the Interim NSLAH Data Where Not-Detected Results
Were Replaced bv LOD/2	
                                                                                      79
Descriptive Statistics of Area-Weighted Average Window Sill
Wipe Dust-Lead Loadings for Households, Presented by.
Housing Age and Census Region. As Reported in the §403
Risk Analysis Versus the Interim NSLAH Data Where No
Adjustments Were Made to Not-Detected Results	
                                                                                      80
Descriptive Statistics of Area-Weighted Average Window Sill
Wipe Dust-Lead Loadings for Households, Presented by.
Housing Age and Census Region. As Reported in the §403
Risk Analysis Versus the Interim NSLAH Data Where Not-
Detected Results Were Replaced bv LOD/2  	
                                                                                      81
Descriptive Statistics of Area-Weighted Average Pre-
Intervention Floor Wipe Dust-Lead Loadings for Households,
As Reported in the §403 Risk Analysis, the HUD National
Survey, and Other Studies	
Descriptive Statistics of Area-Weighted Average Pre-
Intervention Window Sill Wipe Dust-Lead Loadings for
Households, As Reported in the §403 Risk Analysis, the
HUD National Survey, and Other Studies	
                                                                                      86

-------
Table 3-14.
Table 3-15.
Table 3-16.


Table 3-17.



Table 3-18.



Table 3-19.
Table 3-20.
Table 3-21 a.
Table 3-21b.
                                  TABLE OF CONTENTS
                                       (Continued)
Descriptive Statistics of Area-Weighted Average Pre-
Intervention Floor Wipe Dust-Lead Loadings for Households,
Presented by Housing Age Category. As Reported in the §403
Risk Analysis, the HUD National Survey, and Other Studies . .
                                                                                    Page
                                                                                       90
Descriptive Statistics of Area-Weighted Average Pre-
Intervention Window Sill Wipe Dust-Lead Loadings for
Households, Presented bv Housing Age Category. As Reported
in the §403 Risk Analysis, the HUD National Survey, and
Other Studies	
Estimated Percentages of 1997 U.S. Housing Exceeding
Specified Thresholds of Household Average Dust-Lead Loading

Information on Soil Sampling and Analysis Protocols for
Studies Whose Soil-Lead Data Were Compared to Results
from the §403 Risk Analysis and the HUD National Survey . .
Descriptive Statistics of Yard-Wide Average Soil-Lead
Concentrations for Households, As Reported in the
§403 Risk Analysis Versus the Interim NSLAH Data  .
                                                                                       92
 96
 99
103
Descriptive Statistics of Yard-Wide Average Soil-Lead
Concentrations for Households, Presented bv Housing Age
Category. As Reported in the §403 Risk Analysis Versus
the Interim NSLAH Data  	
                                                                                     106
Descriptive Statistics of Yard-Wide Average Soil-Lead
Concentrations for Households, Presented bv Census
Region. As Reported in the §403 Risk Analysis Versus the
Interim NSLAH Data 	
                                                                                     108
Descriptive Statistics of Yard-Wide Average Soil-Lead
Concentrations for Households, Presented bv Housing Age
and Census Region. As Reported in the §403 Risk Analysis
Versus the Interim NSLAH Data Where No Adjustments Were
Made to Not-Detected  Results  	
                                                                                     111
Descriptive Statistics of Yard-Wide Average Soil-Lead
Concentrations for Households, Presented by.
Housing Age and Census Region. As Reported in the §403
Risk Analysis Versus the Interim NSLAH Data Where Not-
Detected Results Were Replaced bv LOD/2  	
                                                                                     112
                                            xi

-------
                                  TABLE OF CONTENTS
                                        (Continued)
Table 3-22a.   Descriptive Statistics of Yard-Wide Average Soil-Lead
              Concentrations. According to Study and Within Specific
              Subsets of the Sampled Housing Within a Study	  117

Table 3-22b.   Descriptive Statistics of Average Soil-Lead Concentrations
              in Specific Yard Areas and/or for Certain Subsets of the
              Sampled Housing Within a Study 	  121

Table 3-23a.   Descriptive Statistics of Yard-Wide^verage Soil-Lead
              Concentrations. According to Study and Housing Age
              Category and Within Specific Subsets of the Sampled
              Housing Within a Study	  129

Table 3-23b.   Descriptive Statistics of Average Soil-Lead Concentrations
              in Specific Yard Areas and/or for Certain Subsets of the
              Sampled Housing Within a Study, Presented by Housing
              Age Category	  134

Table 3-24.    Estimated Percentages of 1997 U.S. Housing Exceeding
              Specified Thresholds of Yardwide Average Soil-Lead
              Concentration	  141

Table 3-25.    Results of Literature Review on Children's Exposure to
              Lead Through Soil Pica 	  148

Table 3-26.    Estimated Rates of Paint and Soil Pica Behavior Reported in the
              USLADP Studies, the Rochester Lead-in-Dust Study,  and the
              Baltimore R&M Study  	  155

Table 3-27.    Alternative Estimates of the Average Number of Children Per
              Unit in the 1997 National Housing Stock,  by Age of Child	  159

Table 3-28.    Alternative Estimates of the Average Number of Children in
              the 1997 National Housing Stock, by Age of Child  and
              Year-Built Category, Based on Data  Obtained Since the
              §403 Risk Analysis	  160

Table 3-29.    Studies for Which Dust Samples Have Been Collected from
              Exterior  Areas, Air Ducts, Window Troughs, and Upholstery
              for Lead Analysis	  162

Table 3-30.    Summary of  Data from Studies Where Exterior Dust Samples
              Were Collected for Lead Analysis 	  163

Table 3-31.    Summary of  Data from Studies Where Air Duct Dust Samples
              Were Collected for Lead Analysis	  167
                                            XII

-------
Table 3-32.
Table 3-33.
Table 3-34.
Table 3-35.
Table 4-1.
Table 5-1 a.
Table 5-1 b.
Table 5-2.
Table 5-3.
                                 TABLE OF CONTENTS
                                       (Continued)
Summary of Data from Studies Where Window Trough Dust Samples
Were Collected for Lead Analysis	
Summary of Data from Studies Where Upholstery Dust Samples
Were Collected for Lead Analysis	
Summary of Children's Pre-lntervention Blood-Lead Concentration
in the HUD Grantees Evaluation According to Blood Collection
Method, Child Age Category, and Grantee (ages 1-2 years only)  .
Page


 168


 170



 180
Percentage of Children with Elevated Blood-Lead Concentration
(at Pre-lntervention) in the HUD Grantees Evaluation According
to Blood Collection Method, Child Age Category, and Grantee
(ages 1-2 years only)	
Parameter Estimates and Associated Standard Errors for the
Three Alternative Multimedia Models Fitted to Rochester Study
Data to Predict Log-Transformed Blood-Lead Concentration  . . .
                                                                                     182
                                                                                     198
Yard-Wide Average Soil-Lead Concentration at Which the
Percentage of Children Aged 1-2 Years With Blood-Lead
Concentration At or Above 10 i/g/dL is Estimated by the
IEUBK Model at 1, 5, or 10%, for Three Assumed Dust-Lead
Concentrations (Table 5-5 in §403 risk analysis report)
                                                                                     209
Yard-Wide Average Soil-Lead Concentration at Which the
Percentage of Children Aged 1-2 Years With Blood-Lead
Concentration At or Above 10 jt/g/dL is Estimated by the
HUD Model at 1, 5, or 10%, for Eight Assumed Dust-Lead
Loadings and Two Assumed Geometric Standard Deviations  .

Floor Dust-Lead Loadings at Which the Percentage of Children
Aged 1 -2 Years With Blood-Lead Concentration At or Above
10 //g/dL is Estimated by the HUD Model at 1, 5, or 10%, for
Five Assumed Soil-Lead Concentrations and Two Assumed
Geometric Standard Deviations	
                                                                                     209
                                                                                     213
Floor Dust-Lead Loadings at Which the Percentage of Children
Aged 1 -2 Years With Blood-Lead Concentration At or Above
10 //g/dL is Estimated by the Rochester Multimedia Model at
1, 5, or 10%, for Five Assumed Soil-Lead Concentrations, Two
Assumed Window Sill Dust-Lead Loadings, and Two Assumed
Geometric Standard  Deviations (expanded version of Table 5-6
in §403 risk analysis report)	
                                                                                     214
                                           XIII

-------
Table 5-4a.
Table 5-4b.
Table 5-4c.
Table 5-5.
Table 5-6.
Table 5-7.
Table 6-1.
Table 6-2.
Table 6-3.
                                 TABLE OF CONTENTS
                                       (Continued)
Uncaroeted Floor Dust-lead Loadings at Which the Percentage
of Children Aged 1 -2 Years With Blood-Lead Concentration At
or Above 10//g/dL is Estimated by the Alternative Rochester
Multimedia Model (A) at 1, 5, or 10%, for Fixed Levels of
Yardwide Average Soil-Lead Concentration and Window Sill
Dust-Lead Loading  	
                                                                                      217
Window Sill Dust-Lead Loadings at Which the Percentage
of Children Aged 1 -2 Years With Blood-Lead Concentration At
or Above 10 //g/dL is Estimated by the Alternative Rochester
Multimedia Model (A) at 1, 5, or 10%, for Fixed Levels of
Yardwide Average Soil-Lead Concentration and Uncarpeted
Floor Dust-Lead Loading	
                                                                                      217
Yardwide Average Soil-Lead Concentration at Which the
Percentage of Children Aged 1 -2 Years With Blood-Lead
Concentration At or Above 10//g/dL is Estimated by the
Alternative Rochester Multimedia Model (A) at 1, 5, or 10%,
for Fixed Levels of Dust-Lead Loadings for Uncarpeted Floors
and Window Sills  	
                                                                                      218
Sensitivity Analysis for the Estimated Baseline Number and Percentage
of Children Aged 1 -2 Years Having Specific Health Effect and
Blood-Lead Concentration Endpoints, Assuming Various Percentage
Declines in Blood-Lead Concentration Since Phase 2 of NHANES III . . .
                                                                                      219
Sensitivity Analysis on How Changes in Household Average
Baseline Dust-Lead Loadings/Concentrations and Soil-Lead
Concentration Impact Pre-5403 Estimates of Health Effect and
Blood-Lead Concentration Endpoints for Children Aged 1-2 Years  .

Sensitivity Analysis on the Assumed Blood-Lead Concentration
Threshold on IQ Decrement and Its Impact on the Pre-§403
Estimates of IQ Decrement Endpoints for Children Aged 1 -2 Years

Definitions of Performance Characteristics  Used to Evaluate
How Various Combinations of Environmental-Lead Standards
Classify Housing Units in the Rochester Lead-in-Dust Study	
Set of 21 Housing Units in the Rochester Study in Which No
Standard Was Exceeded in at Least One of the 864 Combinations
of Candidate Standards	,
Results of Performance Characteristics Analysis Performed on
Data for 177 Units in the Rochester Lead-in-Dust Study for
Specified Sets of Standards	
                                                                                      222
                                                                                      225
                                                                                      232
                                                                                      237
                                                                                      240
                                           XIV

-------
Table 6-4.
Table 6-5.
Table 6-6.
Table 6-7.
Table 6-8.
Table 6-9.
Table 6-10.
Table 6-11.
Table 6-12.
Table 6-13.
                                  TABLE OF CONTENTS
                                        (Continued)
Results of Performance Characteristics Analysis Performed'on
Data for 184 Units in the Rochester Lead-in-Dust Study for
Specified Sets of Standards	
                                                                                     Page
Numbers of Housing Units with Missing Data for Four Endpoints
and the Imputed Data Values Assigned to These Units in
This Analysis	
Results of Performance Characteristics Analysis Performed on
Data for 205 Units in the Rochester Lead-in-Dust Study for
Specified Sets of Standards	
Estimates of Sensitivity and Negative Predictive Value Presented
in Tables 6-3, 6-4, and 6-6  	
                                                                                      244
                                                                                      248
                                                                                      249
                                                                                      252
Results of Performance Characteristics Analysis Performed on
Data for Housing Units in the Rochester Lead-in-Dust Study, for
Specified Sets of Candidate Standards for Lead in Dust and
Soil Only	

Results of Performance Characteristics Analysis Performed on
Data for Housing Units in the Rochester Lead-in-Dust Study, for
Specified Sets of Candidate Standards for Dust-Lead Loadings
and Observed Amount of Damaged Paint on a Tested Surface  .
                                                                                      255
                                                                                      262
Results of Performance Characteristics Analysts Performed on
Data for Housing Units in the Rochester Lead-in-Dust Study, for
Specified Sets of Candidate Standards for Dust-Lead Loadings
and Observed Amount of Damaged Lead-Based Paint on a
Tested Surface	
Summaries of Pre- and Post-Intervention Floor Wipe Dust-Lead
Loadings for Housing Groups Within Seven Studies	
Summaries of Pre- and Post-Intervention Window Sill Wipe
Dust-Lead Loadings for Housing Groups Within Seven Studies
                                                                                      265
                                                                                      273
                                                                                      274
Sensitivity Analysis on How Changes in Household Average
Baseline Dust-Lead Loadings/Concentrations and Soil-Lead
Concentration Impact Post-§403 Estimates of Health Effect and
Blood-Lead Concentration Endpoints for Children Aged 1-2  Years
Under a Spe'cified Set of Example Standards  	
                                                                                      277
                                            XV

-------
Table 6-14.
Table 6-15.
Table 6-16.
                                  TABLE OF CONTENTS
                                        (Continued)
Sensitivity Analysis on the Assumed Blood-Lead Concentration
Threshold on IQ Decrement and Its Impact on the Post-§403
Estimates of IQ Decrement Endpoints for Children Aged 1 -2 Years,
Under a Specified Set of Example Standards  	
                                                                                     Page
                                                                                      279
Sensitivity Analysis on How Changing the Assumption on the
Post-Intervention Household Average (Wipe) Dust-Lead Loadings
on Floors and Window Sills Impact Post-§403 Estimates (Based
on the Empirical Model) of the Health Effect and Blood-Lead
Concentration Endpoints for Children Aged 1 -2 Years Under a
Specified Set of Example Standards  	
Estimated Post-§403 Health and Blood-Lead Concentration
Endpoints Under the Original and Alternative Scaling Algorithms
for Characterizing the Post-§403 Blood-Lead Distribution
                                                                                      281
                                                                                      284
Figure 3-1.
Figure 3-2.
Figure 3-3.
Figure 3-4.
Figure 3-5.
                       LIST OF FIGURES

Boxplots of Area-Weighted Average Floor Wipe Dust-Lead Loadings
Usg/ft2), As Observed in the §403 Risk Analysis (Using HUD National
Survey Data) and in the Interim NSLAH (under 2 approaches to
handling not-detected values)	
                                                                                       67
Boxplots of Area-Weighted Average Floor Window Sill Wipe
Dust-Lead Loadings (//g/ft2). As Observed in the §403 Risk
Analysis (Using HUD National Survey Data) and in the Interim
NSLAH (under 2 approaches to handling not-detected values) .
                                                                                       68
Boxplots of Area-Weighted Average Floor Wipe Dust-Lead Loadings
(//g/ft2), by Housing Age Category, As Observed in the §403 Risk
Analysis (Using HUD  National  Survey Data) and in the Interim
NSLAH (under 2 approaches to handling not-detected values)	
                                                                                       71
Boxplots of Area-Weighted Average Floor Window Sill Wipe
Dust-Lead Loadings {//g/ft2), by Housing Age Category,
As Observed in the §403 Risk Analysis (Using HUD National
Survey Data) and in the Interim NSLAH (under 2 approaches
to handling not-detected values)	
                                                                                       72
Boxplots of Area-Weighted Average Floor Wipe Dust-Lead Loadings
(//g/ft2), by Census Region, As Observed in the §403 Risk
Analysis (Using HUD National Survey Data) and in the Interim
NSLAH (under 2 approaches to handling not-detected values)	
                                                                                       76
                                            XVI

-------
                                 TABLE OF CONTENTS
                                       (Continued)
                                                                                    Page
Figure 3-6.    Boxplots of Area-Weighted Average Floor Window Sill Wipe
             Dust-Lead Loadings (pg/ft2), by Census Region, As Observed
             in the §403 Risk Analysis (Using HUD National Survey Data)
             and in the Interim NSLAH (under 2 approaches to handling
             not-detected values)  	  77

Figure 3-7.    Boxplots of Area-Weighted Average Pre-lntervention Floor Wipe
             Dust-Lead Loadings (pg/ft2) for Houses in the HUD National Survey,
             Baltimore R&M Study, Rochester Lead-in-Dust Study, and Grantees
             Within the HUD Grantees Evaluation	  83

Figure 3-8.    Boxplots of Area-Weighted Average Pre-lntervention Window Sill
             Wipe Dust-Lead Loadings (//g/ft2) for Houses in the HUD National
             Survey, Baltimore R&M Study, Rochester Lead-in-Dust Study, and
             Grantees Within the HUD Grantees Evaluation  	  84

Figure 3-9.    Boxplots of Area-Weighted Average Pre-lntervention Floor Wipe
             Dust-Lead Loadings (//g/ft2) for Houses in the HUD National Survey,
             Baltimore R&M Study, Rochester Lead-in-Dust Study, and HUD
             Grantees Evaluation, by Age of House Category (pre-1979 only)	  87

Figure 3-10.  Boxplots of Area-Weighted Average Pre-lntervention Window Sill
             Wipe Dust-Lead Loadings U/g/ft2) for Houses in the HUD National
             Survey, Baltimore R&M Study, Rochester Lead-in-Dust Study, and
             HUD Grantees Evaluation, by Age of House Category (pre-1,979 only)	  88
Figure 3-11,
Estimated Distribution of Household Average Floor Dust-Lead
Loading in the Nation's Housing Stock, and Corresponding Estimates
of the Percentage of Homes Exceeding Specified Thresholds (with
95% Confidence Intervals on the Corresponding Number of Homes,
in Millions), Based on Data from the HUD National Survey (top plot)
and the Interim NSLAH (bottom plot)	
                                                                                       95
Figure 3-12.
Boxplots of Yard-Wide Average Soil-Lead Concentration U/g/g),
As Observed in the §403 Risk Analysis (Using HUD National
Survey Data) and in the Interim NSLAH (under 2 approaches
to handling not-detected values)	,
                                                                                     104
Figure 3-13.
Boxplots of Yard-Wide Average Soil-Lead Concentration 0/g/g),
by Housing Age Category, As Observed in the §403 Risk
Analysis (Using HUD National Survey Data) and in the Interim
NSLAH (under 2 approaches to handling not-detected values) . .
                                                                                     107
                                          XVII

-------
Figure 3-14.
Figure 3-15.
Figure 3-16.
Figure 3-17.
Figure 3-18.
Figure 3-19.
Figure 3-20.
Figure 3-21.
                                 TABLE OF CONTENTS
                                       (Continued)
Boxplots of Yard-Wide Average Soil-Lead Concentration U/g/g),
by Census Region, As Observed in the §403 Risk Analysis
(Using HUD National Survey Data) and in the interim NSLAH
{under 2 approaches to handling not-detected values)	,
                                                                                    Page
                                                                                     109
Boxplots of Household Average Soil-Lead Concentrations (//g/g)
for Houses in the HUD National Survey, Baltimore R&M Study,
Rochester Lead-in-Dust Study, and Grantees Within the HUD
Grantees Evaluation  	
Summary Statistics of Average Household Soil-Lead Concentrations
(pg/g) for Selected Studies as Compared to Summaries Based
on Data from the HUD National Survey	
                                                                                     113
                                                                                     115
Boxplots of Household Average Soil-Lead Concentrations (jt/g/g)
for Houses in the HUD National Survey, Baltimore R&M Study,
Rochester Lead-in-Dust Study, and HUD Grantees Evaluation,
by Housing Age Category (pre-1979 only)	
                                                                                     127
Estimated Distribution of Yardwide Average Soil-Lead Concentration
in the Nation's Housing Stock, and Corresponding Estimates
of the Percentage of Homes Exceeding Specified Thresholds (with
95% Confidence Intervals on the Corresponding Number of Homes,
in Millions), Based on Data from the HUD National Survey  (top plot)
and the Interim NSLAH  (bottom plot)	
                                                                                     140
Estimated Distribution of Yardwide Average Soil-Lead Concentration
Among Urban Housing in the HUD National Survey, and Corresponding
Estimates of the Percentage of Homes Exceeding Specified Thresholds
(with 95% Confidence Intervals on the Corresponding Number of Urban
Homes in the Nation, in Millions)	
                                                                                     143
Fitted Regression Models Predicting Children's Blood-Lead
Concentration as a Function of Area-Weighted Arithmetic Average
Floor Dust-Lead Loading (Wipe Collection Method), for the
Various Grantees in the HUD Grantees Evaluation and for
the Rochester Lead-in-Dust Study	
                                                                                     183
Fitted Regression Models Predicting Children's Blood-Lead
Concentration as a Function of Area-Weighted Arithmetic Average
Window Sill Dust-Lead Loading (Wipe Collection Method), for
the Various Grantees in the HUD Grantees Evaluation and for
the Rochester Lead-in-Dust Study	
                                                                                     184
                                           XVIII

-------
Figure 3-22.
Figure 5-1.
Figure 5-2.
Figure 5-3.
Figure 6-1.
Figure 6-2.
                                 TABLE OF CONTENTS
                                       (Continued)
Geometric Mean Blood-Lead Concentration Versus Child Age,
As Reported Within the Cincinnati Prospective Lead Study and
Presented According to Housing Age and Condition  	
                                                                                      186
Percentage of Children's Blood-Lead Concentrations, as Predicted
by the IEUBK and HUD Models, That Will Exceed or Equal lOpg/dL
as a Function of Yard-Wide Average Soil-Lead Concentration and
at Fixed Levels of Dust-Lead Concentrations or Loadings	
                                                                                      208
Percentage of Children's Blood-Lead Concentrations, as Predicted
by the HUD Model and the Rochester Multimedia Model, That Will
Exceed or Equal 10/yg/dL  as a Function of Floor Dust-Lead Loading
for Five Soil-Lead Concentrations and Two Window Sill Dust-Lead
Loadings (Geometric Standard Deviation = 1.6)	
                                                                                      211
Percentage of Children's Blood-Lead Concentrations, as Predicted
by the HUD Model and the Rochester Multimedia Model, That Will
Exceed or Equal 10//g/dL as a Function of Floor Dust-Lead Loading
for Five Soil-Lead Concentrations and Two Window Sill Dust-Lead
Loadings (Geometric Standard Deviation = 1.72)	
Example of an Ideal Situation for Establishing Potential Dripline
Soil-Lead Standards	
Example of a Situation Where the Negative Predictive Value and
Sensitivity Equal 100%, but the Positive Predictive Value and
Specificity are Less than 100%  	
                                                                                      212
                                                                                      233
                                                                                      234
                                           XIX

-------
This page intentionally blank.
             xx

-------
EXECUTIVE SUMMARY

       This report is a supplement to the EPA report "Risk Analysis to Support Standards for
Lead in Paint, Dust, and Soil" (USEPA, 1998a), which presented the methods and findings of a
risk analysis that supported efforts by the U.S. Environmental Protection Agency (EPA) to set
regulatory standards for lead levels in dust and soil and to control lead-based paint hazards within
most pre-1978 housing and child-occupied facilities. These regulatory standards were mandated
through §403 of the Toxic Substances Control Act (TSCA), as specified within Title X, the
Residential Lead-Based Paint Hazard Reduction Act of 1992 (42 U.S.C. 4851). The §403 risk
analysis provided EPA with a scientific foundation for establishing the regulatory standards.  On
June 3,1998, EPA proposed a regulation to establish these standards (40 CFR Part 745); this
regulation is referred to as the "§403 proposed rule."

       In 1998, the Environmental Health Committee of EPA's Science Advisory Board (SAB)
conducted a technical review of the §403 risk analysis. In response to this review, EPA deemed
that a supplement to USEPA (1998a), referred to as the "§403 risk analysis report," was
necessary to provide additional technical analyses and to clarify certain key analyses and findings
presented within the report. EPA used the summaries and analyses of data considered in the
original §403 risk analysis to prepare responses to the public comments on the §403 proposed
rule and to prepare the final rule.  Although this supplement also presents summaries and
analyses of data made available to EPA since the §403 proposed rule was released, such as
interim data from the National Survey of Lead and Allergens in Housing (NSLAH), this type of
information is meant only to provide an alternative to the findings presented in the §403 risk
analysis report. EPA has  not used these new data in efforts to select the hazard standards or
levels of concern that are  included within the final rule. Furthermore,  this supplement does not
replace any parts of the original §403 risk analysis, but rather supplements this risk analysis with
more detailed analyses  on selected topics.

       The format of this supplement follows that of the §403 risk analysis report, with separate
chapters for each phase of the risk assessment (hazard identification, exposure assessment, dose-
response assessment, and risk characterization) and  a final chapter on risk characterization under
example options for standards.  The remainder of this executive summary summarizes the key
issues and findings presented within each chapter of this supplement.

Hazard Identification  (Chapter 21

       Additional evidence that lead-based paint hazards pose a health risk to children and the
magnitude of such risk was provided in the following four areas within this supplement:

       •      Adverse health effects of lead exposure, with a focus on neurological effects, as
              observed in animal studies (to address the point raised in the SAB review that
              animal data can support causality effects over and above the available data on
              humans).
                                          XXI

-------
       •      Evidence supporting causality between lead exposure and adverse health effects.

       •      The association between blood-lead concentration and reduction in intelligence
              quotient (IQ) score.

       •      The role that dust particle size and the chemical composition of lead compounds
              in lead-contaminated dust may play in determining the extent to which lead in
              residential dust is bioavailable to humans.

       Adverse health effects, especially neurotoxicitv. observed in animal studies.  Lead has
been observed to have widespread neurotoxic effects, as well as behavioral and cognitive
symptoms, in humans.  These observations are largely consistent with the findings that have been
demonstrated in controlled, dose-response studies on animals (e.g., rodents, dogs, non-human
primates).  Animal studies are also congruent with observations of lead exposure in humans,
suggesting an increased susceptibility of the young brain to lead poisoning (Banks et al., 1997).
For example, lead-induced alteration of protein kinase C (PKC) activity in the brain has been
shown to correlate with poor performance in several learning tasks in animal studies, and
researchers have suggested that some of the learning and memory deficits observed in children
are likely to be causally related to the types of PKC activity alterations exhibited in these studies
(Chen et al., 1998).  In addition, animal studies have provided physiological evidence that many
of the effects of lead on the differentiation of the developing nervous system, such as synaptic
and dendritic development, and myelin and other nerve  structure formation, have the potential to
be long-lived (USEPA, 1986). Although animal models do not duplicate the human response to
lead exposure, they do serve to provide strong support for expecting certain health effects to
occur when humans are exposed.

       Cause-effect relationship of lead exposure and adverse health effects.  The combined
weight of human and animal studies provides evidence that lead  may be assumed to cause
adverse neurological effects in young children. For example, longitudinal studies in humans
(e.g., Boston, Port Pirie) have shown that disturbances occur in neurobehavioral development
early in life even at low lead exposure levels. These studies have observed effects of lead
exposure even after accounting for other demographic factors (e.g., socioeconomic status,
maternal IQ) that could affect neurological development.  While these other demographic factors
tend to be highly correlated with blood-lead levels early in life, the influence that lead exposure
has on blood-lead tends to increase with a child's age. This is because over time, measures of a
child's lead exposure tend to change at a faster rate than the child's demographic measures, and
more recent lead exposures continue to be predictive of a child's current health consequences.
The §403 risk analysis assessed these adverse neurological effects in children through measuring
average IQ score decrements in the population due to lead exposure.

       Association between blood-lead concentration and IP score decrement. This supplement
provides additional technical justification regarding the  assumptions made in the §403 risk
analysis on the association between blood-lead concentration and decrement in IQ score.
                                          xxn

-------
       While prior investigations into this relationship have considered both linear and log-linear
associations, a linear relationship (with positive slope) was used in the §403 risk analysis based
on the evidence taken from these investigations and the desire not to unduly over-estimate the IQ
decrement associated with children having low blood-lead concentrations.

       Recent studies and meta-analyses investigating the presence of a threshold hi the blood-
lead/IQ relationship (e.g., Schwartz, 1994) have concluded either that no non-zero threshold
exists, or if one does exist, if may be very low (e.g., less than 1.0 ug/dL; Schwartz, 1993).
Researchers who disagree with this conclusion have not reached a clear consensus on a value for
this threshold. Several older studies that have suggested high thresholds (e.g., over 10 ug/dL)
involved few, if any, children at low blood-lead levels, thereby preventing their ability to provide
information on potential health effects at low levels. Other researchers have used visual
inspection of data summaries to conclude that thresholds exist at relatively high levels, rather
than using statistical inference techniques that would yield more scientifically defensible
conclusions. As it would have been necessary to have clear, scientifically defensible evidence of
a particular non-zero threshold to justify its adoption within the §403 risk analysis, and based on
the findings of recent meta-analyses, the approach taken in the §403 risk analysis was to assume
that no threshold exists. Nevertheless, sensitivity analyses documented in this supplement have
investigated the impact of assuming a positive threshold on the risk analysis estimates.

       Effect of residential dust characteristics on the bioavailabilitv of lead in dust. It has been
suggested that lead speciation and particle size may affect the bioavailability of lead in dust
through their influence on solubility. Therefore, this supplement included an investigation of
current knowledge about bioavailability of lead in dust and how this knowledge may impact the
rulemaking process.

       There is relatively little in the literature which examines the relationship between
bioavailability and chemical  composition specifically for household dust. Animal studies  have
concluded that different lead compounds are associated with different rates of absorption (e.g.,
metallic-lead was associated with low absorption). Otherwise, most of the investigations into the
effect of lead speciation on bioavailability primarily address lead in soil. However, soil can have
a considerable influence on dust-lead levels.  Evidence exists that correlation between blood-lead
and soil-lead levels can be influenced by both particle size and chemical composition.  Generally,
the smaller the particle size, the greater the absorption of lead due to more rapid dissolution in
the gastrointestinal tract.  Some researchers have  also hypothesized that smaller particle sizes can
contain higher lead concentrations.

       While evidence does  suggest that particle  size and chemical composition can influence
the level of bioavailability of lead in dust, the current information base may be inadequate to
determine how such factors can reasonably be incorporated into the rulemaking effort.
Furthermore, needing to characterize dust in this manner within a risk assessment would likely
add to the expense of dust analyses, and dust standards that distinguish between these various
characterizations could add considerable complexity to the rule.
                                           XXIII

-------
Exposure Assessment (Chapter 31

       When using model-based procedures to estimate health risks associated with lead-based
paint hazards, the §403 risk analysis used data from the 1989-1990 HUD National Survey of
Lead-Based Paint in Housing to characterize lead levels in dust and soil within the nation's
housing stock. As mentioned above, more recent data have been made available to EPA since
the §403 risk analysis report and the proposed rule were published, and some commenters on the
proposed rule suggested that EPA should use these data when available. These data include
interim data (collected in 1998 and 1999) for 706 housing units from the NSLAH, an on-going
national survey of lead levels in dust and soil in the nation's housing.  HUD assigned interim
sampling weights to these 706 surveyed units in such a way as to allow interim results that
properly incorporate these sampling weights to be nationally representative of occupied housing
in which children can possibly reside. Other recently-acquired data that are summarized in
Chapter 3 of this supplement include additional data from the Evaluation of the HUD Lead-
Based Paint Hazard Control Grant Program ("HUD Grantees") and data from the 1997 American
Housing Survey.

       Comparing dust-lead loadings in the HUD National Survey with those of other studies.
Before dust-lead loadings measured in the HUD National Survey could be compared to dust-lead
loadings from other studies, the loadings needed to be adjusted to reflect the fact that the HUD
National Survey used a vacuum technique rather than a wipe technique to collect the dust
samples. The §403 risk analysis included a procedure for performing this adjustment.  The
resulting dust-lead loadings, adjusted to represent dust samples collected using wipe techniques,
tended to be lower than the wipe dust-lead loadings reported in other recent lead exposure studies
(e.g., Rochester study, HUD Grantees evaluation), even when taking into account housing age
category and Census region.  One major exception, however, were data from the interim
NSLAH, whose distribution of dust-lead loadings was lower than in the HUD National Survey
for both floors and window sills. For example, household average floor dust-lead loadings based
on the interim NSLAH data had a median of less than 2 ug/ft2, compared to approximately 5.3
ug/ft2 as estimated hi the §403 risk analysis based on the HUD National Survey data. The
estimated percentage of housing units with average floor dust-lead loadings that exceed SO ug/ft2
(i.e., the proposed floor dust-lead standard) was 6.4% based on HUD National Survey data used
in the §403 risk analysis, and 0.9% based on interim data from the NSLAH.

       Comparing soil-lead concentrations in the HUD National Survey with those of other
studies. Data on yardwide average soil-lead concentrations from the HUD National Survey were
compared with interim NSLAH data, as well as data for 22 other studies that characterized soil-
lead concentrations in urban areas prior to any lead abatement. While geometric mean  yardwide
average soil-lead concentration was lower in the HUD National Survey relative to most of the
other studies reviewed, the distributions of yardwide average soil-lead concentration were similar
between the HUD National Survey and the interim NSLAH (although the estimated median was
nearly 50% lower in the HUD National Survey versus the interim NSLAH for homes built prior
to 1940). Regardless of which national  survey is considered, the risk analysis supplement
estimates that the yardwide average soil-lead concentration exceeds 2000 ppm in approximately

                                          xxiv

-------
1.7% of housing, approximately 3.3% of housing exceeds an average soil-lead concentration of
1200 ppm, and between 11% and 12% of housing exceeds an average soil-lead concentration of
400 ppm.

       Soil pica in children.  Because the impact of paint pica (i.e., the purposeful ingestion of
paint chips) on blood-lead concentration in the presence of deteriorated lead-based paint is not
represented within other environmental exposures to lead, the §403 risk analysis accounted for
paint pica as a separate factor when estimating risks. While the analysis did not consider the
independent impact of soil pica (i.e., the purposeful ingestion of soil) over and above paint pica,
it considered the impact of soil pica as part of the relation between soil-lead concentration and
blood-lead concentration. While this supplement does not change the approach taken in the
original risk analysis, it documents information obtained on the component of soil-lead exposure
that may be attributable to soil pica.

       Based on what has been found in the literature on studies in which paint pica and soil pica
behaviors were characterized and could  be separated, approximately 10% to 20% of study
children appeared to exhibit soil pica behavior in the absence of paint pica. The frequency of soil
pica episodes depends on many factors,  including climate, access to bare soil, socioeconomic
standing, age of child, and parental supervision. While estimates of the amount of soil ingested
during pica episodes can vary widely among mass balance studies (i.e., from 500 to 13,000
mg/day), average daily ingestion over a year may be much lower. Because soil pica behavior
tends to be episodic in nature, it is currently uncertain whether the amount of lead in soil ingested
on an infrequent basis would be sufficient to elevate blood-lead concentration to unsafe levels.
Again, it should be noted that the original §403 risk analysis did include an effect of soil pica in
the relationship between soil-lead and blood-lead levels; the additional information presented in
this supplement addresses how the effect of soil pica is separated from  the effect of paint pica.

       Dust-lead levels on surfaces other than floors and window sills. Comments were received
on the  §403 proposed rule recommending that EPA establish dust standards for surfaces other
than floors and window sills. In response, EPA conducted further analysis on whether surfaces
other than floors and window sills might provide significant additional information for risk
assessments.  EPA reviewed what has been published regarding dust-lead loadings from exterior
locations, air ducts, window troughs (also known as window wells), and upholstery, with special
attention given to studies that investigated the correlation between such dust-lead levels and
blood-lead concentrations.

       It is difficult to use data from lead exposure studies to characterize the effect of lead
exposure from exterior dust on children's blood-lead levels, primarily because studies differ
considerably in how they collect exterior dust and how they report the results of lead analysis.
For example, some studies have collected both exterior dust and soil for lead analysis but report
results that are combined across both media. Also, studies often differ in their sampling
approaches (e.g., surface scrapings, vacuum sampling) and in the locations at which the samples
are taken. Furthermore, studies frequently have not investigated the specific influence of exterior
dust-lead on blood-lead level in the presence of lead in other environmental media. For these
                                          XXV
                                                        U.S. EPA Headquarters Library
                                                               Mai) code 3201
                                                        1200 Pennsylvania Avenue NW
                                                                           orv/icn

-------
reasons, along with the fact that lead in exterior dust can be highly correlated with soil-lead, a
scientifically defensible standard for lead in exterior dust was not identified, nor was such a
standard deemed essential over and above the planned standards for lead in soil and interior dust.
This latter conclusion becomes more acceptable if risk assessors are made aware that lead in
exterior dust can be associated with adverse health effects in children, and as a result,
recommend corrective action in cases where lead is  present in exterior dust and is suspected to be
an important pathway of exposure (e.g., when children spend a considerable amount of time on
hard surfaces immediately outside of the residence).

       For air ducts and upholstery, there is insufficient data upon which to characterize the
role that lead in dust from these surfaces plays in a child's total lead exposure, and therefore, to
develop a hazard standard for lead on these surfaces.

       Because dust from window troughs (i.e., window wells) has historically been sampled in
lead exposure studies (along with floors and window sills), and because risk assessors sample
dust from window troughs in determining clearance following a lead-based paint intervention,
reports of window trough dust-lead levels are generally prevalent in the literature.  However,
several studies have reported that the association between window trough dust-lead and blood-
lead is not statistically significant after taking into account the effects of dust-lead on floors and
window sills (for which standards have been proposed under §403). For this reason, along with
the likelihood that exceeding a window sill dust-lead standard will prompt a cleaning of leaded-
dust from window troughs as well as sills, it does not appear that an additional standard for
window troughs is necessary either to identify a home with a hazard or to guide corrective
actions. Also, given the correlation in lead levels between window troughs and window sills, it
is likely that if more sampling is to be done beyond  a minimal risk assessment, more benefit
would be obtained from sampling more windows at only the sill rather than fewer windows at
both the sill and trough.

       Distribution of childhood blood-lead. Information on the distribution of blood-lead
concentrations in children based on data from the HUD Grantees evaluation was updated to
reflect additional data collected through January, 1999. These data provided a means by which
estimates based on data from Phase 2 of the Third National Health and Nutrition Examination
Survey (NHANES IE) (i.e., the data used in the §403 risk analysis to characterize blood-lead
levels in the nation) could be evaluated.  The HUD Grantees evaluation data (under venipuncture
blood collection) had a geometric mean of 9.3 jig/dL for children aged 1-2 years and 8.0 ug/dL
for children aged 3-5 years.  In contrast, the geometric means based on data from Phase 2  of
NHANES ffl were 3.1 ug/dL for children aged 1-2 years and 2.5 ug/dL for children aged 3-5
years. Also, 51 percent of children aged 1-2 years sampled via venipuncture methods had blood-
lead concentrations at or above 10 jig/dL, compared to the estimates of 5.9% for Phase 2 of
NHANES HI, 53.8% for the Baltimore R&M study  (pre-intervention), and 23.4% for the
Rochester Lead-in-Dust study.  The trend toward high blood-lead levels in the HUD Grantees
evaluation reflects, among other factors,  the HUD Grantees program's procedure of selecting
high-risk children for monitoring. While NHANES ffl provides the most nationally
                                          XXVI

-------
representative data on children's blood-lead concentration, it does not provide environmental-
lead data that could be used to investigate the effect of environmental-lead exposure on blood-
lead concentration. Therefore, other data sources such as the HUD Grantees evaluation must
provide this information.

       Regression modeling of the blood-lead concentration data suggests that the relationships
between blood-lead concentration and household average dust-lead loading were relatively
consistent across grantees.  In particular, these relationships were similar to that observed for data
from the Rochester study (i.e., the data used to develop the empirical model presented in Chapter
4 of the §403 risk analysis). This conclusion is important in that the data from the HUD
Grantees evaluation reflect a much larger geographical area than the Rochester study and
represent several types of exposure conditions.

       This risk analysis supplement includes evidence that housing age and condition play
important roles in the likelihood of a resident child having an elevated blood-lead concentration.
The association between older housing and the prevalence of lead hazards has been well-
documented and is accepted by many experts in residential lead exposure. The level of
deterioration is an important variable in the accessibility of lead-based paint hazards to children.

Dose-Response Assessment (Chapter 4)

       The objective of the dose-response assessment in the §403 risk analysis was to develop a
statistical procedure to characterize the relationship between environmental-lead exposure and
the resulting adverse health effects in young children. This characterization would then be used
to estimate health risks at specified environmental-lead levels or over the entire population. The
modeling tools used in this characterization were EPA's Integrated Exposure, Uptake, and
Biokinetic (IEUBK) model and an empirical model developed especially for the §403 risk
analysis from data collected in the Rochester Lead-in-Dust study. In this supplement, additional
models were considered to quantify this characterization: a new model developed from
epidemiological data collected from 12 lead exposure studies and made available to EPA after
the §403 risk analysis was completed, and revisions to the  multimedia model developed for the
§403 risk analysis using the Rochester study data ("Rochester multimedia model").  In addition,
this supplement provides additional detail on specific aspects of the model-based analysis
employed within the §403 risk analysis, such as how post-intervention blood-lead concentration
distributions are characterized and how measurement error was handled when fitting the
empirical model.

       HUD Model.  An additional model for predicting blood-lead concentration as a function
of environmental-lead levels became available after the §403 risk analysis report was published.
Some commenters on the §403 proposed rule suggested that EPA use this new model. This new
model is a log-linear regression model developed on behalf of the U.S. Department of Housing
and Urban Development (HUD) from epidemiological data collected from 12 studies. Thus this
model is referred to in this supplement as the "HUD Model." The goal of this  model was to
                                         xxvii

-------
"estimate the contribution of lead-contaminated house dust and soil to children's blood-lead
levels" (Lanphear et al., 1998).

       When using the HUD model to predict blood-lead concentration .as a function of lead
levels in various environmental media, this risk analysis supplement has noted several caveats
associated with interpreting the predicted blood-lead concentration:

       •      Risks associated with exposure to specific environmental-lead levels (as estimated
              from the HUD model) are generally not comparable to population-based risks (as
              estimated by the IEUBK and empirical models in the §403 risk analysis).

       •      The prediction parameters in the HUD model are not independent. Therefore, it is
              not appropriate to interpret the parameter estimates hi the HUD model (or in the
              models developed for the §403 risk analysis) in isolation.

       •      The HUD model has adjusted for measurement error in certain environmental-
              lead measures used as input. Therefore, the model assumes that these input values
              represent "actual" exposure levels. In contrast, the models developed for the §403
              risk analysis use measured levels as input that would be reported from a risk
              assessment. Because the §403 standards will be compared to lead exposure
              measures that are subject to being measured with error, this latter approach is
              more relevant for rulemaking purposes.

Further discussion of the HUD model is provided in Chapters 4 and 5 of this supplement.

Risk Characterization (Chapter 51

       Health risks associated with current (i.e., baseline) lead exposures for children aged 1 to 2
years were characterized in the §403 risk analysis. Both individual risk estimates (i.e., risks
associated with specific environmental-lead levels) and population-based risk estimates (i.e.,
average risks over the entire nation) were presented. In this supplement, additional sensitivity
and uncertainty analysis associated with the baseline risk characterization was performed, where
possible alternatives to various approaches taken and assumptions made in the risk
characterization were identified and incorporated into the analysis, and the resulting impact on
the risk estimates was evaluated.

       When predictions under the Rochester multimedia model (developed in the §403 risk
analysis to characterize individual risks) were compared to predictions under the HUD model,
the following general findings were observed:

       •      At  very low floor dust-lead loadings (i.e., 1-5 ng/ft2), the HUD model and the
              Rochester multimedia model yield similar predictions for the geometric mean
              blood-lead concentration, which also results in similar predictions for the health-
              effect endpoints that are calculated directly from this geometric mean (e.g.,

                                          xxviii

-------
             percentage of children with blood-lead concentration at or above a specified
             threshold; average IQ decrement resulting from lead exposure).

       •      The predicted geometric mean blood-lead concentration under the HUD model
             ranges from 20% to nearly 60% higher than the prediction under the Rochester
             multimedia model as floor dust-lead loadings increase from 15 to 100 fig/ft2 and
             as soil-lead concentrations decrease from 2000 ppm to 10 ppm (assuming, for the
             Rochester multimedia model, that window sill dust-lead loadings are at their
             estimated national median level). Note that for a fixed value of the geometric
             standard deviation (GSD) for the blood-lead distribution, the average IQ
             decrement in the population that is associated with lead exposure is a multiple of
             the geometric mean (as calculated in the §403 risk analysis). Therefore, similar
             differences in predictions between the two models would occur for average IO
             decrement.

       •      If the geometric standard deviation (GSD) associated with the blood-lead
             distribution is fixed, then as floor dust-lead loadings increase beyond 10 jig/ft2,
             the predicted percentage of children with blood-lead levels at or above 10 mz/dL
             increases at a much faster rate under the HUD model (at a constant soil-lead
             level). For example, if window sill dust-lead loading is at its estimated national
             median and soil-lead concentration is below 2000 ppm, the predicted percentage
             under the HUD model is at a minimum twice as large as the prediction under the
             Rochester multimedia model.  This difference  in predictions gets even greater as
             the assumed soil-lead concentration gets lower. For example, at a GSD of 1.6, a
             floor dust-lead loading of 100 ug/ft2, and a soil-lead concentration of 10 ppm, the
             prediction is over 7  times higher for the HUD model compared to the Rochester
             multimedia model (13.1% versus 1.76%).

       Other findings within the additional sensitivity analyses performed to support the baseline
risk characterization were as follows:

       •      If it is assumed that a 50% across-the-board decline in blood-lead concentration
             has occurred relative to the distribution portrayed by data from Phase 2 of
             NHANES ffl, the estimated number of children whose blood-lead concentration
             was at or above 20 jig/dL declined by 95% (from 46,800 to 2,130), while the
             estimated number at or above 10 ug/dL was reduced by nearly 90% (from 458,000
             to 46,800). The estimated average IQ decrement in the population due to lead
             exposure is cut in half under this assumption (from  1.06 to 0.53 points).

       *      Model-based baseline risk estimates were calculated for various alternative
             assumptions on the percentage decline in dust-lead and soil-lead levels that may
             have occurred in the housing stock since the HUD National Survey was
             conducted. Risk estimates under the empirical  model seemed to be more sensitive
             to these changes than the estimates under the EUBK model, and reductions in

                                          xxix

-------
              soil-lead concentration seemed to have more of an impact on reducing these risk
              estimates than reductions in dust-lead levels.  Baseline estimates for the
              percentage of children with blood-lead concentrations at or above 10 |ig/dL were
              45% lower under the empirical model when both dust-lead and soil-lead levels
              were decreased by 50%, with smaller declines occurring for less drastic total
              reductions in the environmental-lead levels.

       •      In an effort to determine whether an assumption of no threshold made in the §403
              risk analysis was particularly sensitive to the risk estimates, baseline estimates of
              the IQ-related health effect endpoints were calculated under the assumption that
              specified non-zero thresholds exist in the relationship between blood-lead
              concentration and IQ score decrement. While the §403 risk analysis estimated an
              average IQ decrement of 1.06 points occurs due to lead exposure across the
              population of children aged 1-2 years, this average declines by approximately 44%
              under a assumed threshold of 2 ug/dL (0.588 points) and by 90% under a
              threshold of 8 ug/dL (0.103 points).

Analysis of Example  Options for the 8403 Standards (Chapter 6)

       Prompted by public comments on the §403 proposed rule and risk analysis, this
supplement included the following on methods used in the §403 rulemaking process to evaluate
candidate hazard standards and levels of concern:

       •      Detailed information on performance characteristics analyses (also known as
              sensitivity/specificity analysis), used by EPA to help establish levels of concern
              within the §403 rule.

       •      Characterizing the extent to which children with elevated blood-lead
              concentrations reside in homes where no candidate standard is met or exceeded
              (i.e., children who would be "missed" by a specified set of candidate standards),
              as part of the candidate standards evaluation process.

       •      An additional investigation into the assumptions made in the risk management
              study on post-intervention dust-lead loading (40 ug/ft2 on floors, 100 ug/ft2 on
              window sills).

       •      Additional sensitivity and uncertainty analyses for the analyses performed and
              documented within Chapter 6 of the §403 risk analysis, including alternative
              assumptions on baseline and post-intervention environmental-lead levels.

       Performance characteristics analysis. Performance characteristics analysis is a non-
modeling approach (based on calculating conditional probabilities) to assessing how often a
specified set of candidate hazard standards would "trigger" interventions in housing units within
the studies in question and the extent to which these units contained a child with an elevated

                                          xxx

-------
blood-lead concentration (a 10 ug/dL). Data from the Rochester Lead-in-Dust study were used in
these analyses. This supplement contains a detailed discussion of how to interpret performance
characteristics analysis and provides additional information on analysis results that were cited in
the §403 proposed rule. Furthermore, this supplement presents the results of follow-on
performance characteristics analyses which EPA considered when responding to public
comments and in preparing the final rule. While the analysis presented in the proposed rule was
based on data for 77 housing units in the Rochester study, additional assumptions made to the
soil-lead data permitted up to 184 units to be represented among the data analyzed in the follow-
on performance characteristics analyses. One goal of these analyses was to identify those sets of
candidate dust-lead loading standards for which the analysis estimated that no more than 5% of
children living in housing units with environmental-lead levels below the standards would have
elevated blood-lead concentrations.

       Investigating incidence of elevated blood-lead concentrations in homes where no
candidate standard is met or exceeded. As an alternative to the performance characteristics
analysis, a model-based approach was developed to determine the likelihood of a child with
elevated blood-lead concentration residing in a housing unit that exceeds none of a given set of
candidate standards. This approach was designed to use data from the Rochester study and to
yield results that would be directly comparable to those from the performance characteristics
analysis.

       Review of published information on post-intervention dust-lead loadings. To evaluate
the performance of a given set of candidate standards in reducing population-based health risks
to lead exposure, the §403 risk analysis needed to make assumptions on lead levels hi dust and
soil that would occur after performing interventions that would be prompted by exceeding the
example standards. Assumptions made on post-intervention dust-lead loadings (40 ug/ft2 for
floors, 100 ug/ft2 for window sills) within the §403 risk analysis were evaluated in this
supplement in a detailed review of results from studies that evaluated abatement effectiveness.

       fa the reviewed studies, geometric mean or median floor dust-lead loadings were
generally at or below 41 ug/ft2 over periods ranging from 6 months to 6 years post-intervention,
with several studies reporting levels below 21 ug/ft2 at follow-up periods ranging from
12 months to 2 years.  Of the eight grantees participating in the HUD Grantees evaluation that
had post-intervention floor dust-lead loadings available at 12 months post-intervention, four had
median values for these loadings that were at or below 10 ug/ft2.  Median pre-intervention floor
dust-lead loadings for these four grantees ranged from 9 to 26 fig/ft2.  For post-intervention
window sill dust-lead loadings, geometric means or medians ranged from 24 ug/ft2 to 958 jig/ft2
in the reviewed studies. Most of the study groups  had geometric mean or median post-
intervention window sill dust-lead loadings below 100 ug/ft2, while a few were at or below 51
       As a result of the post-intervention dust-lead loadings review, a sensitivity analysis
documented in this supplement applied the methods developed in the risk analysis under
alternative post-intervention floor dust-lead loadings of 10 and 25 ug/ft2 and post-intervention
                                          xxxi

-------
window sill dust-lead loadings of 50 and 75 ug/fi2. Results of this sensitivity analyses indicated
that while more housing units may be assumed to achieve reductions in average dust-lead loading
as a result of lowering the post-intervention dust-lead loading assumptions, the corresponding
reduction in the estimated blood-lead concentration and health effect endpoints appeared to be
modest, especially compared to the reduction that occurs from pre-intervention conditions.

       See page I-i of Appendix I for an executive summary of an investigation on the
relationship between lead levels in carpet-dust and children's blood-lead concentration and on
how extending the floor dust-lead loading standard in the §403 proposed rule to include carpeted
floors might impact the performance of these proposed standards.
                                           XXXII

-------
1.0    INTRODUCTION

       On June 3,1998, the U.S. Environmental Protection Agency (EPA) proposed regulation
to establish standards for lead-based paint hazards in most pre-1978 housing and child-occupied
facUities (40 CFR Part 745, "Lead; Identification of Dangerous Levels of Lead; Proposed Rule").
EPA proposed these standards in accordance with Section 403 of the Toxic Substances Control
Act (TSCA), as amended by Title X, the Residential Lead-Based Paint Hazard Reduction Act of
1992. The proposed standards are as follows:

       *      Dust-lead hazards: Household average dust-lead loadings equal to or exceeding  50
              [ig/ft2 on uncarpeted floors and 250 ug/ft2 on window sills, assuming wipe
              collection techniques for dust;

       •      Soil-lead hazards: Total lead levels equal to or exceeding 2,000 ppm based on a
              yard-wide average soil-lead concentration

       •      Hazardous lead-based paint: Lead-based paint in poor condition, defined as
              follows:
              •     More than 10 ft2 of deteriorated paint on exterior components with large
                    surface areas
              •     More than 2 ft2 of deteriorated paint  on interior components with large
                    surface areas
              •     Deteriorated paint consisting of more than 10% of the total surface area of
                    exterior or interior components with small surface areas.

These standards, a focal point of the Federal lead program,  identify the presence of lead-based
paint hazards, defined within TSCA Section 401 as the condition of lead-based paint and the
levels of lead-contaminated dust and soil that "would result" in adverse human health conditions.

       To provide a scientific basis  for selecting the §403 standards, EPA conducted a risk
analysis to assess the health risks to young children (aged 1-2 years) from exposures to lead-
based paint hazards, lead-contaminated dust, and lead-contaminated soil in the nation's housing.
This risk analysis also documented EPA's approach to estimate the reduction in these risks
following promulgation of the §403  standards and applied this methodology to evaluate example
options for the §403 standards. Finally, the risk analysis provided estimates of the numbers of
homes and children that would be affected by various example standards. EPA published this
risk analysis in June, 1998, in a document hereby referred to as the "§403 risk analysis report"
(USEPA, 1998a).

       A period of public comment  followed publication of the §403 proposed rule, extending to
March, 1999. Several comments received during this period requested additional analyses and
investigation.  In addition, the Environmental Health Committee of EPA's Science Advisory
Board (SAB) performed a review of the technical aspects of the §403 risk analysis, the  §403
economic analysis, and the proposed rule. While the SAB concluded that many approaches taken

                                           1

-------
in the §403 risk analysis were technically sound and scientifically defensible, their final report
provided detailed comments and recommendations for additional investigation and analysis to be
considered when preparing a final rule (USEPA, 1998b).

       This report is a supplement, or addendum, to the §403 risk analysis report. It contains
additional information obtained since the report was published that further supports the findings
and conclusions made hi that report. It also contains the results of additional analyses requested
by the SAB and by key public comments. This supplement does not replace any parts of the
original risk analysis, but rather supplements the original risk analysis with more detailed
analyses on selected topics.

       Reflecting its close ties to the original §403 risk analysis report, this supplement contains
the same chapters as those found hi the §403 risk analysis report. These chapters represent the
different components of the risk analysis: Hazard Identification, Exposure Assessment, Dose-
Response Assessment, Risk Characterization, and Analysis of Example Options for the §403
Standards.  The additional analyses and investigations presented in this supplement are found
within the specific chapters to which their findings contribute. Each analysis or investigation is
presented as an independent module within each chapter.  See the chapter introductions for the
contents of each chapter and the motivation for each analysis being presented.  Furthermore, the
reader is referred to the §403 risk analysis report for details on the risk analysis approach and
findings.

1.1    PEER REVIEW

       This report was reviewed independently by members of a peer review panel who,
together, had considerable knowledge on the subject areas addressed in this report. The three
reviewers on this panel who provided EPA with comments on this report were:

       Dr. Ruth Chen, Tennessee Department of Health
       Joseph Schirmer, MS., Wisconsin Division of Health, Bureau of Public Health
       Nellie K. Laughlin, Ph.D., Harlow Center for Biological Psychology, University of
              Wisconsin-Madison.

Two of these reviewers (Dr. Chen and Mr. Schirmer) also provided peer review comments on the
§403 risk analysis report prior to its 1998 publication. Therefore, they were previously aware of
the issues addressed and approaches taken in the original §403 risk analysis. The third reviewer
(Dr. Laughlin), while not involved in the review of the §403 risk analysis report, was provided a
copy of this latter report for reference while reviewing the present report. Dr. Laughlin was
selected primarily based on her involvement in research to investigate the health effects of lead
exposures in various animal species, which is addressed in Section 2.1 of this report.

       EPA asked the peer reviewers to provide general comments and suggestions concerning
this report. In addition, EPA suggested that the reviewers provide more detailed comments on

-------
those sections of the document addressing specific components of the risk analysis in which the
reviewer had a particular technical expertise.

       The peer reviewers were appreciative in regard to the additional information being
provided in this supplement report. They did provide some useful suggestions for additional
additions and revisions, however, which were considered when finalizing this report. The
remainder of this section discusses how the peer review comments led to report modifications.

       One reviewer questioned the usefulness of measuring IQ decrements greater than 1,2, or
3 resulting from lead exposure, which the risk analysis has included among the health effect
endpoints that were estimated. This reviewer pointed out that in an individual child, the
variability associated with a measured IQ score is generally greater than the decrement measures
of 1,2, or 3 being considered in the risk analysis.  While this may be true, the risk analysis is
targeting the decrement in IQ score that occurs on average, across an entire population of
children, rather than in an individual child. The variability associated with estimating the mean
IQ in a population of children is much smaller than the variability of an individual child's IQ
score, thereby allowing the risk analysis to consider these types of IQ decrements. To emphasize
and clarify this point to the reader, initial introductions of these health effect endpoints in this
report have been revised to refer to "IQ score in the population of U.S. children."

       Peer reviews of Section 2.1 of this report, which addressed the adverse health effects
associated with lead exposure as observed in animal studies, resulted in considerable revision to
this section. Suggestions were followed to re-format Section 2.1.2 to resemble the format of
Section 2.1.3, where detailed results of specific studies were incorporated within the general
discussion of observed health effects across studies. Additional published articles on specific
studies that were suggested and/or provided by the reviewers were obtained and reviewed, and
their relevant results were added to the discussion within Section 2.1.2. The articles suggested
by the reviewers tended to focus on the effects of lead exposure on the visual and auditory
systems and on social development and behavior.  Requests for additional information on
selected discussion items were addressed by re-reviewing the relevant references that had
previously been cited and augmenting the discussion. Finally, general clarification points raised
by the reviewers on the content of the section were incorporated when revising the section.

       EPA has established a public record for the peer review of this report under
administrative record AR-188, "Risk Analysis to Support Standards for Lead in Paint, Dust, and
Soil: Peer Review." The record is available in the TSCA Nonconfidential Information Center,
which is open from noon to 4 PM Eastern time Monday through Friday, except legal holidays.
The TSCA Nonconfidential Information Center is located in Room NE-B607, Northeast Mall,
401 M Street SW, Washington, DC.

-------
This page left blank intentionally.


-------
2.0    HAZARD IDENTIFICATION

       Chapter 2 of the §403 risk analysis report presented information on the toxicity of lead,
through a discussion of how body-lead burden is measured, how lead works in the body, the
resulting adverse health effects, and populations at risk. This chapter introduced the endpoints
used in the risk analysis to represent the adverse health effects resulting from lead exposure and
to estimate the benefits of the §403 rule. These endpoints included the likelihood of exceeding a
specified blood-lead concentration threshold (10 or 20 ug/dL), the likelihood of achieving a
specified IQ score decrement as a result of lead exposure (1,2, or 3 points on average across the
population), the likelihood of achieving an IQ score in the population of less than 70 points due
to lead exposure, and the average IQ score decrement in the population that results from lead
exposure. The blood-lead concentration thresholds were among those established by the Centers
for Disease Control and Prevention (CDC) as levels of concern. The IQ-related endpoints
represented measures of the neurological effects of lead exposure.  The representative population
upon which the §403 risk analysis focused was children aged 1-2 years, as it was considered the
most appropriate age range for the estimation of health effects.

       In this chapter of the supplemental report, the results  of four additional investigations into
hazard identification are presented:

       •       Section 2.1: A review of the adverse health effects of lead exposure, with a focus
              on neurological effects, as observed in animal studies.

       •       Section 2.2: Support for the causality of adverse health effects due to lead
              exposure.

       •       Section 2.3: Characterizing the relationship between blood-lead concentration
              and IQ score.

       •       Section 2.4: Documenting what is known about the role that dust particle size and
              the chemical composition of lead compounds  in lead-contaminated dust may play
              in determining the extent to which lead in residential dust is bioavailable to
              humans.

The motivation for including each of these sections into this chapter is presented within the
introduction to each section.

2.1    REVIEW OF THE ADVERSE HEALTH EFFECTS OF LEAD EXPOSURE.
       WITH A FOCUS ON NEUROLOGICAL EFFECTS.
       AS OBSERVED IN ANIMAL STUDIES

       The §403 risk analysis used data from human exposure studies to characterize the
relationship between environmental-lead levels and measures of the various blood-lead
concentration and health effect endpoints. However, as the SAB's review of the §403 risk

-------
analysis points out (USEPA, 1998b), causality is difficult to establish using human studies alone,
due to the potential for confounding factors being present.  For example, while average IQ score
may differ significantly between one group of children with low blood-lead concentration and
another group with elevated blood-lead concentration, the reason for this.difference may not be
solely due to lead exposure, but to demographic and other factors that cannot be controlled
completely by the researcher. The factors left uncontrolled in the analyses of data from human
exposure studies contribute to large uncertainties associated with these analyses. While ethical
considerations preclude the use of humans hi controlled lead-exposure experiments, a substantial
amount of published literature is available on controlled lead-exposure experiments involving
animals.  Such studies use animals that are chosen from a homogeneous population, reared under
identical conditions and randomly assigned to groups where the mode, duration, and amount of
lead exposure is controlled within each group (these groups can include one or more control
groups). Therefore, in a well-controlled animal study, the presence of significant differences
between dose groups can be inferred to be the result of lead exposure at certain doses.

       To address the SAB's recommendation to consider the findings of "... animal data, since
they support human data by establishing causality, due to the absence of confounding variables,
and potential mechanisms for adverse health effects" (USEPA, 1998b), this section presents the
key findings of animal studies that have investigated the impact of lead exposure on adverse
health effects, especially at low doses.  In these studies, animals were typically exposed to lead
either in utero, during infancy (through mother's milk/formula), during maturation or adulthood,
or a combination of these life phases. The major lead-induced adverse health effects noted in
humans, including neurological, neurodevelopmental, immunological, and systemic (e.g.,
cardiovascular, hematological, and renal effects) have been demonstrated in controlled, dose-
response studies in rodents, dogs, and/or non-human primates.  While this section recognizes the
variety of adverse health effects, its primary focus is on the neurological effects of lead, as the
§403 risk analysis has recognized that young children are most susceptible to neurological effects
due to their developing central nervous systems.

       A glossary of selected terms used in this section can be found in Appendix A.

2.1.1 Approach to Reporting the Findings of Animal Studies

       This subsection reviews the incidence of adverse health effects associated with lead
exposure, as reported in published animal studies. The emphasis of this review is on the
neurological, developmental, and neurobehavioral effects of lead.

       In preparing this review, it was not desired to duplicate the previous efforts of others who
have prepared excellent published literature reviews that have been peer reviewed and are easily
available to the general public. These include two articles  cited by the SAB (USEPA, 1998b) as
"important references" on animal studies data: Rice (1996) and Cory-Slechta et al. (1997).
Other important review documents include USEPA (1986) and USDHHS (1999). This section
frequently references the content of these  and other study review documents.

-------
       To help identify any additional articles that may have been published since these key
review references were prepared and published, a search of the scientific literature over the last
five years was conducted. The strategy for this literature search was on the key words "lead,"
"effect," "exposure," "neurologic," "behavior," "development," "teratology," and "animal" as
whole or root words. Upon review of abstracts identified from this literature search, articles
found to be relevant to the objectives of this report were obtained and reviewed, with high
priority placed on study reviews.

       Certain results of recent key studies identified within the review publications and the
literature search are also discussed in detail in this section. Overviews of these studies are
presented in Table 2-1. Note that these selected studies represent only a subset of all studies
whose results and conclusions provide important contributions to the knowledge base on adverse
health effects associated with lead exposure.

       Section 2.1.2 focuses on the findings of animal studies on the neurological,
developmental, and neurobehavioral effects of lead. This subsection contains an overview of the
physiological consequences of these effects, along with a review of general findings of studies
investigating these types of health effects.  In addition, certain animal studies are discussed in
greater detail with summaries of their design and conclusions.

       Lead has been documented to have considerably more effects on the health of humans
and animals than just neurological effects.  Therefore, Section 2.1.3 presents a brief overview of
general health effect information and how it relates to lead exposure, as observed in animal
studies. Information in this subsection is organized according to the type of health effect.

2.1.2 Neurological, Behavioral,  and Developmental
       Health Effects

       Lead has been observed to have widespread neurotoxic effects, as well as to cause
behavioral and cognitive symptoms, in humans. These effects are largely consistent with results
of morphological, electrophysiological, biochemical, and behavioral studies on animals.
Although lead toxicity research has isolated many of the specific neurological effects of lead, it is
generally considered to be a relatively indiscriminate toxin within the neurological system. This
consideration is largely due to the ability of lead to disturb several fundamental biotic processes
such as cellular metabolism and energy production, ion transport across membranes, and protein
function. In addition, the neurological effects of lead frequently have been observed to occur as a
series of interrelated events. Thus, lead poisoning is likely to cause simultaneous and interrelated
disturbances in a number of processes within the nervous system. As an additional consequence,
the ability to separate the direct and indirect effects of lead on the neurological system is often
difficult (Banks et al. 1997).

-------
Table 2-1.    Summary of Lead Exposure Levels and Key Findings for Selected Animal
               Studies
    Authors
Subject
           " Lead Exposure
                    '
          Key Findings / _,'
      Effects of Lead Exposure
 Attmann et at.
 (1993)
  Rat
0 or 750 ppm lead acetate in diet at
various life stages (/r? utero, pre-
weaning, and post-weaning)
Active avoidance learning and long-
term hippocampal potentiation were
impaired when exposure occurred
prior to  16 days postnatally.
 Burger et al.
 (1998)
 Turtle
0.25, 1.0, or 2.5 mg/g lead acetate in
deionized water, by injection; one-time
exposure; controls received an isotonic
saline solution injection
Death with high dose; dose-
dependent righting response
impairment with low and moderate
doses.
 Bushnell et al.
 (1977)
Monkey
control, lower-dose (targeted blood-lead
of 55 /sg/dL), and higher-dose (targeted
btood-lead of 85 j/g/dL) of lead acetate
in milk formula in the first year of life
At age 18 months, visual
discrimination in the higher-dose
group was impaired under dim light
compared both to their performance
under bright light and to the other
groups under various luminescence
levels.
 Bushnell and
 Bowman
 (1979a, b)
Monkey
control, tower-dose (targeted blood-tead
of 50 fjg/dl), and higher-dose (targeted
blood-lead of 80 //g/dL) of lead acetate
in milk formula and food in the first
year of life
In the first year of life, suppressed
play and increased social clinging, as
well as various social delays that
occur when play environment is
abruptly changed (primarily occurring
in animals dosed post-natally only;
1979b). At age 4 years, when a
subset of the animals was further
tested, diminished performance on
spatial-cue  reversal learning sets was
observed in the higher-dose group
and, to  a lesser extent, in the lower-
dose group (1979a).
 Chen et al.
 (1998)
  Rat
0.2% lead acetate as drinking water;
exposed in utero and during nursing via
dams, and postweaning directly;
controls received 0.145% sodium
acetate in drinking water
Altered protein kinase C (PKC)
distribution in the hippocampus.
 Cory-Slechta
 (1997)
 (review)
  Rat
50 or 250 ppm lead acetate as drinking
water; exposed postweaning; control
groups received no lead exposure
Disruption of neurotransmitter
systems (i.e., dopamine and
glutamine systems); selective
learning deficits (e.g., impaired repeat
acquisition performance); dose-
dependent alteration of fixed interval
schedule-controlled response rates.
                                                  8

-------
                                        Table 2-1.  (cont.)
   Authors
       "
Cutler (1977)
  Rat
0 or 0.1% lead acetate exposed pre-
weaning, and 0 or 0.1 % lead acetate
exposed post-weaning in drinking water
At 8 weeks of age, duration of non-
social activity was significantly
greater in the exposure group versus
the controls for males but not
females. Across both sexes,
frequency and duration of social and
sexual investigation was decreased in
the exposure group.
Fox et al.
(1997)
  Rat
0, 0.02%, or 0.20% lead acetate in
diet at various life stages (in utero,  pre-
weaning, and post-weaning)
Significant association between
retinal degeneration and both age and
lead exposure level.
Hastings et
al. (1977)
  Rat
0, 0.02%. or 0.10% lead acetate in
diet pre-weaning
At 60 days of age, aggressive
behavior was significantly reduced in
the exposed groups, but no
significant differences were observed
in visual discrimination tasks.
Kuhlman et
al. (1997)
  Rat
750 or 1000 ppm lead acetate, as feed;
exposed in utero, in utero to adulthood,
or postweaning to adulthood; controls
received no lead exposure
Performance impairment in water
maze for all rats exposed in utero; no
impairment for rats exposed post-
weaning.
Lasky et al.
(1995)
Monkey
Controls or "modest' exposure to lead
via lead acetate in utero, pre-weaning,
and/or post-weaning within the first
year of life
Markedly abnormal distortion product
otoacoustic emissions (DPEs) for the
two animals with the highest blood-
lead concentrations.  Otherwise,
DPEs did not differ significantly
between the groups, although
auditory brain stem evoked response
(ABR) did differ significantly.
Lilienthal et
al. (1990,
1994)
Monkey
0, 350, or 600 ppm lead acetate in diet
of mothers and their offspring (in utero,
pre-weaning, and post-weaning to age
10 years)
At age 7 years, offspring in the
exposed groups had significantly
higher latencies in flash-evoked and
brain stem auditory evoked potentials
(BAEP) which increased with
increasing click rates (1990).  At age
12 years, offspring in the exposed
groups had significantly increased
amplitudes of the scotopic b-wave in
electroretinogram (ERG) recordings
(1994).
Mello et al.
(1998)
  Rat
1.0 mM lead acetate as drinking water;
exposed in utero and during nursing via
dams; control dams received deionized
water only
Selective motor skill impairment
(accelerated fist eye opening, startle
reflex, and free-fall righting; impaired
spontaneous alternation performance
in maze).

-------
                                    Table 2-1.  (cont.)
   Authors
Subject
Species
          Lead Exposure
     '' ,  ' Key Findings /
     Effects'of Lead Exposure
 Nagymajtenyi
 etal. (1998)
  Rat
80, 160, 320 mg/kg/day lead acetate in
distilled water, by gavage; exposed pre-
or post-natally; controls received same
volume of distilled water by gavage
Dose dependent increase in
behavioral and bioelectric aberrations
(e.g., hyperactivity).
 Rice and
 Gilbert
 (1990a,
 1990b)
Monkey
1.5 mg/kg/day lead acetate, in
capsules; exposed continuously after
birth, postnatally until 400 days, or
after 300 days from birth; controls
received vehicle only
Nonspatial discrimination tests
(1990a): group dosed continuously
from birth onward exhibited greatest
degree of impairment, followed by
group dosed after infancy only (no
impairment observed in group dosed
during infancy only).

Delayed alternation task <1990b): all
exposed groups showed impairment
to an approximately equal degree.
       Research, based on both human observation and animal studies, indicates that relatively
low doses of lead can adversely affect both the peripheral and central nervous systems, while
high lead exposures can result in acute lead encephalopathy, and may ultimately lead to death.
Lead induced damage to the brain and nervous system may be manifested as various and diverse
developmental symptoms, including both behavioral and cognitive impairments.

       In the following subsections, the physiological effects of lead on the brain and nervous
system (Section 2.1.2.1) and subsequent effects on development and behavior (Section 2.1.2.2)
are discussed. The findings of specific studies listed in Table 2-1 above are included in this
discussion.  In general, some caution must be taken when extrapolating the findings in these
studies to humans and to other species. For example, lead neuropathy in rats is primarily
characterized by demyelination of nerves, while cats, rabbits, and humans generally show
damage to the axons of nerves (Davis et al., 1990). Developmental age of the brain also varies
between animal species and humans. For example, at birth, the rat brain is relatively less
developed and is roughly equivalent to the human brain at 5-6 months of gestation (Winneke et
al., 1996). Furthermore, although rats and possibly monkeys tend to have higher tolerances for
the general toxic effects of lead exposure relative to humans (i.e., they require a higher exposure
level to reach an equivalent blood toxicity level), the lowest blood-lead levels at which lead-
induced developmental/neurobehavioral effects have been observed in  animals and in humans are
reported to be similar in magnitude (Banks et al., 1997; Davis et al., 1990;  USDHHS, 1999).
                                            10

-------
       2.1.2.1  Physiological Effects of Lead on the Neurological System.

       Overview. As lead is known to adversely affect several universal processes within
biological systems, the symptoms of lead poisoning have been observed in many cell types,
tissues, and organs within the neurological system.

       On the cellular level, lead has been observed to cause disruption of mitochondria!
function (i.e., cellular metabolism), damage to cell structural components (e.g., microtubules),
and damage to glial cells and the myelin sheaths and axons of nerve cells (USEPA, 1986).
Electrophysiological and biochemical processes at the cellular level may also be disrupted by
lead exposure. Specific alterations of these processes reported in animal studies include
impairment of synaptic events and neuron function, interference with neurotransmitter function,
and protein activity inhibition (e.g., enzymes and hormones) (USEPA, 1986). Much of lead's
role in cellular level dysfunctions is suggested by many researchers to be attributed to its
interference with calcium-mediated processes (Banks et al., 1997). Calcium is an important ion
in many biological systems and is specifically involved in neurological phenomena such as
enzyme-protein activation, secondary messenger regulation of metabolic pathways, membrane
potential/ion channel regulation, and neurotransmitter release.  Lead ions, as they are similar to
calcium ions in both size and charge, can substitute for calcium and thus competitively interfere
with these types of calcium-mediated cell processes.

       Disruption of cellular processes can eventually result in damage to tissues and organ
systems within the neurological system. Reported effects of lead in animals at the organ and
system level within the central nervous system include compromise of the blood-brain barrier,
disruption of the limbic system and cerebral cortex, and damage to the cerebellum (Banks et al.,
1997; USEPA, 1986). Animals exposed to lead in early post-natal life have also exhibited
reductions and delays in development of various brain regions, including the hippocampus and
cerebral cortex (Banks et al., 1997; USDHHS, 1999). The observed effects of lead exposure on
specific organ systems and processes, as observed in animal studies, are now discussed.

       Blood-brain barrier. Banks et al. (1997) reviewed animal studies that examined the
effect of lead exposures on the blood-brain barrier that regulates the movement of chemical
substances in and out of the brain. In studies of rats exposed to relatively high lead levels, higher
concentrations of lead were observed in the barrier capillaries than were observed in the brain as
a whole. USEPA (1986) reviewed studies which provided evidence that lead transport within the
brain is by the same mechanisms as calcium transport. Thus, as the capacity for calcium
transport (specifically, into neural cell mitochondria) is known to be much higher in the brain
than other body tissues, a lead accumulation in brain capillaries is not unexpected (USEPA,
1986). Acute lead toxicity as lead accumulations in brain capillaries may disrupt barrier
permeability, allow greater influxes of water, ions, and other substances, and result in swelling of
the brain (encephalitis) (USEPA, 1986). Although lower lead exposure levels (< 40ug/dL) have
not been reported to result in specific damage to the barrier or in disproportionate accumulation
of lead in the capillaries of the blood-brain barrier (Banks et al., 1997), developing brains have
been suggested to be particularly susceptible to the transfer of even very low levels of lead from

                                           11

-------
the blood into the brain since the blood-brain barrier is not yet fully functional (Altmann et al.,
1993; Cory-Slechta, 1997).

       General cellular processes. Several in vitro studies with animal cells have provided
evidence that a major site of lead interference is with cellular metabolism and energy transfer via
disruption of the normal mitochondria! ion gradient (Banks et al., 1997; USEPA, 1986). m
addition, mitochondria in the cerebellum and in developing brains of animals have been observed
to show a greater sensitivity to lead disruption than mitochondria in other body tissues or at other
ages (USEPA, 19S6). This has been suggested to provide a possible explanation for one of the
root causes of the greater sensitivity of the neurological system, and of the young in particular, to
lead poisoning (USEPA, 1986). Impairment of mitochondria! energy production subsequently
can affect many other energy-requiring cellular processes, such as protein synthesis, lipid
synthesis, and membrane integrity. In animal studies, the normal development of proteins in
neurons was altered in rats exposed to lead perinatally (USDHHS, 1999). Other studies reviewed
by Banks et al. (1997) indicate that moderate levels of lead can interfere with microtubule
formation (in vitro animal cell studies) and the formation of the myelin sheath in neurons in both
the central and peripheral nervous systems of lead exposed rodents.

       Lead can also interfere directly with calcium-mediated cell processes. This may include
disruption of protein function (e.g., enzyme regulation of cell growth and differentiation), ion
transport systems across membranes, and membrane potentials (Banks et al, 1997; USEPA,
1986). For example, a study performed by Chen et al. (1998) found that hippocampal protein
kinase C (PKC) activity, which has been correlated with performance in several learning tasks,
was altered by relatively low internal lead exposures (<30 ug/dL) in postnatal rats. More
information on this study and its findings can be found in the discussion below on the limbic
system and hippocampus.

       Neuron and neurotransmitter function.  Lead-induced disruption of cellular processes
may result in neuron dysfunction. According to a body of experimental animal research
reviewed by Banks et al. (1997), evidence exists that moderate to low levels of lead exposure can
impair synapse formation in the hippocampus during postnatal development of rats, in the visual
cortex of primates, and in the frontoparietal cortex of guinea pigs, and also appears to interfere
with synaptic transmission in fetal rat hippocampal neurons by blocking postsynaptic receptors.
USEPA (1986) reviewed numerous studies in which rats exhibited morphological effects such as
decreased glial cell and synaptic density, and delayed maturity of synapses in neurons of the
cerebral cortex with lead exposure. Several of the reviewed studies also reported abnormal
development of neuron dendrites and persistent impairment of electrical activity (i.e., reduced
firing rates) in the cerebellum of cats and rats exposed perinatally to low levels of lead (Banks et
al., 1997). The cerebellum is responsible for the regulation and coordination of complex
voluntary muscular movement, and is also implicated in cognitive attention-switching activity.
The cerebral cortex is largely responsible for higher brain functions, including sensation,
voluntary muscle movement, thought, reasoning, and memory. Thus, damage to neurons in these
areas may explain some attention deficits, learning and memory impairments, and disturbances in
                                           12


-------
motor coordination that have been observed hi behavioral studies of lead exposure (Banks et al.,
1997).

       Lead has also been observed to interfere with the release and uptake of neurotransraitters
in various in vitro and in vivo animal studies, including the inhibition of neurotransmitter release
from calcium sensitive voltage channels in snails, and alteration of synthesis and turnover rates
of various neurotransmitters hi different regions of the brain (hippocampus, cerebellum,
hypothalamus, brainstem) in lead-exposed rats (Banks et al., 1997; USEPA, 1986; Nagymajtenyi
et al., 1998).  Altered activity of various neurotransmitters has been observed in studies involving
rats exposed to lead prenatally and postnatally (USDHHS, 1999).  In the reviewed literature, lead
has generally been reported to have highly variable and non-specific effects on the various
neurotransmitter systems, possibly due to its more general effect on metabolic processes (Banks
etal., 1997).

       Cory-Slechta (1997) conducted a review of studies linking disruptions in neurotransmitter
systems (i.e., dopaminergic (DA) and glutamatergic (GLU) systems) with behavioral and
cognitive impairments hi lead exposed animals. The author cites previous research which
indicates that the DA and GLU neurotransmitter systems are critical to various cognitive
functions, and also are sensitive to lead-induced disruptions. In addition, the  author reviews
studies reporting non-lead related disruptions of these systems and the concurrent appearance of
behavioral symptoms that are very similar to symptoms of lead toxicity. However, the author
also concedes that much of the historical research Unking behavioral impairments to biochemical
effects is based solely on correlations and that relationships are complicated by the fact that any
given behavioral symptoms may have multiple physiological mechanisms.  Cory-Slechta (1997)
summarized several studies conducted in her own lab on rats exposed, postweaning  (i.e., at 21
days of age), to lead (0,50 and 250 ppm lead as lead acetate) in drinking water for varying
durations of time.  Results of these studies indicated that the DA neurotransmitter system is
vulnerable to lead-induced modifications, such as unpaired regulation of DA  synthesis and
release. These modifications to DA system function were suspected to contribute to alterations
in response control (fixed interval schedule-controlled behavior) that were observed hi lead-
exposed rats in the reviewed studies.  Results indicated that the GLU system was also involved in
lead-induced impaired learning, primarily as manifested by an increase in perseverative errors hi
lead-exposed rats, relative to controls. There was no evidence that the DA system was involved
hi learning accuracy. The author suggests that confirmation of the involvement of the GLU and
DA systems in lead-induced effects, could also have implications extending beyond cognitive
concerns. For example, similar patterns of neurotransmitter disruptions have been associated
with schizophrenia, drug  addiction, and psychosis (Cory-Slechta, 1997).

       Lead-induced disruption of neuron and neurotransmitter function can result in altered bio-
electric activity in the brain and nervous system. This was seen, for example, in a study by
Nagymajtenyi et al. (1998) in which 120 female and 60 male rats and their 120 male offspring
were administered a lead acetate solution by gavage at a concentration of either 0,80,160, or 320
mg/kg. There were three variations on the treatment schedule: (1) pregnant females were dosed
only during the 5th -15th day of pregnancy; (2) pregnant females were dosed during pregnancy and

                                           13

-------
lactation; or (3) pregnant females were dosed during pregnancy and lactation, and their weaned
offspring were dosed for 8 weeks. Behavioral observations of the offspring were made at 12
weeks. The study observed electrophysiological disruptions, including changes in
electrocorticogram (ECoG) indices, cortical evoked potential, and slowed nerve conduction
velocity, in the somatosensory area of the cerebral cortex in the rats exposed to lead both pre- and
post-natally relative to controls.  Electrophysiological functions showed both dose- and
treatment-dependent changes, including decreased mean amplitude and increased frequency of
the ECoG, and lengthened latency and duration of the evoked potentials. The observed changes
in electrophysiological functioning depended on the dose and timing (i.e., age of animal at
exposure) of lead administration. The authors suggested that non-invasive monitoring of
electrical disturbances in the nervous system may provide a valuable and early indicator of low-
level lead poisoning.

       The limbic system and hippocampus. The limbic system (e.g., hippocampus) is of
particular interest in lead toxicity studies as it is key in many of the processes that appear to be
affected by lead poisoning, including cognition, emotion, motivation, behavior, memory, and
various autonomic functions. Some researchers have even suggested that symptomatic
similarities (e.g., learning and memory impairments)  between lead toxicity and other
experimental limbic system disruption indicate that the limbic system is a target site for lead
toxicity in the brain (Walsh and Tilson, 1984).

       Reported behavioral changes that may be attributable to bippocampal damage include
increased aggressiveness, seizures, inappropriate responsiveness, reversal problems, visual
discrimination deficits, impaired motor coordination, and other types of learning deficiencies
(Petit et al., 1983). Furthermore, the developing hippocampus may be more susceptible to
functional injury by lead exposure compared to the mature hippocampus. For example, Altmann
et al. (1993) conducted a study where 88 female rats were dosed with lead acetate in their diet at
either 0 or 750 ppm for 50 days prior to mating through day 16 following birth of a litter,  at
which time 2-3 male offspring were taken from each  litter and dosed under the same regimen
until sacrifice. Half of the animals were tested for two-way active avoidance learning. This
study observed a correlation between hippocampal disruptions and active avoidance learning
deficit for rats exposed to lead during the prenatal and early postnatal stages (i.e., during
hippocampal development), but not when lead exposure occurred only after 16 days postnatally.

       Some studies reviewed in the literature differ  on  the extent to which the hippocampus is a
target organ for lead.  For example, relative to other regions of the brain, greater impairment of
neuron function has been observed in hippocampal cells of rats exposed to lead perinatally
(Banks et al., 1997). Studies reviewed by Petit et al.  (1983) indicated that higher levels of lead
may tend to accumulate in hippocampal cells of rat brains. However,  other studies reviewed by
Banks et al. (1997) did not report a preferential accumulation of lead in the hippocampus of
young lead-exposed rats.  For example, a 1994 study  by D.V. Widzowski and D.A. Cory-Slechta
(as cited in Banks et al., 1997) exposed rats postnatally to lead from dams' milk (dosed with four
levels of lead from 100 to 2000 ppm lead) and measured lead levels in 12 brain regions after 7-60
days. The study found that lead tended to accumulate in similar patterns across brain regions and

                                           14

-------
exposure levels.  Some researchers have hypothesized that the particular vulnerability of the
hippocampus to lead poisoning may be due more to sensitivity rather than increased lead
accumulation in this region (Banks et al., 1997). In addition, because the most rapid phase of
development of hippocampus is known to occur postoatally (late compared to other brain
regions), a particular sensitivity of the hippocampus to early-life lead exposure is plausible (Petit
et al, 1983; Altmann et al., 1993).

       Specific lead-associated physiological effects in the hippocampus, as reported in animal
studies reviewed by Banks et al. (1997) and Petit et al. (1983), include significant reductions in
size and weight of the hippocampus and reductions in hippocampal cell layer thickness with
relatively high perinatal lead exposure in rats. Damage and stunting of hippocampal structural
cells (glial cells/astrocytes) was observed in rats and monkeys exposed to lead prenatally and
postnatally.  Perinatally exposed rats also exhibited reductions in development of hippocampal
neuronal dendrites and mossy fiber pathways, which are both involved in the transmission of
nerve impulses (Petit et al., 1983).

       Initially mentioned in the overview of general cellular processes above, Chen et al. (1998)
performed a study to investigate the effects of developmental lead exposure on protein kinase C
(PKC) activity in the hippocampus of rats at various postnatal ages.  This study attempted to
elucidate some of the physiological mechanisms of lead-induced learning deficits. In this study,
lead was administered orally as 0.2% lead acetate in drinking water to pregnant and lactating
female rats and then directly to their weanling pups (weaned at postnatal day 21) in drinking
water. Controls received 0.145% sodium acetate in drinking water.  Four to six rat pups were
randomly selected for necropsy from different dams at postnatal days 7,14,28, and 56, and PKC
activity was measured in both the membrane and cystolic fractions of the hippocampi. Results
showed that lead exposure increased PKC activity in the cystolic fraction at postnatal day 56, and
decreased PKC activity in the membrane fraction at postnatal day 7. The ratio of membrane to
cystolic PKC activity, which is indicative of PKC distribution, decreased at postnatal days 28 and
56.

       A review of studies in Chen et al. (1998) indicated that PKC activity has been associated
with various brain functions (e.g., ion channel function, receptor function, and neurotransmitter
release) and that alteration of hippocampal PKC, in particular, has been correlated with poor
performance in several learning tasks. Therefore the authors hypothesize that  the lead-induced
alterations of PKC activity and distribution observed in their study may have caused functional
changes in the animal brain, including modulation of ion channels, desensitization of receptors,
and enhancement of neurotransmitter release. Chen et al. (1998) also suggested that some of the
learning and memory deficits observed in children are likely to be causally related to the types of
PKC activity alterations exhibited in this study.

       The visual system.  Some researchers contend that the retina serves as a good model for
studying the effects of lead on the central nervous system.  Because most retinal cells, like the
central nervous system, develop during gestation and, in the rat, for up to two weeks postnatally,
researchers have studied the effects of lead exposure on retinal cell development in post-natal

                                          15

-------
rats in order to characterize such effects for humans during early gestation and post-natal periods.
Fox et al. (1997) studied the effects on retinal development of low and moderate lead exposure
via mother's milk in female Long-Evans hooded rats aged 0 to 21 days (i.e., from parturition to
weaning). These rats were partitioned into three dose groups (6-14 rats per group) based on the
concentration of lead (0, 109, or 1090 ppm) in the lead acetate solution of drinking water that the
dams were provided. At 21 days of age, lead levels in the blood and the retinas of rats in both the
low and moderate dose groups averaged significantly higher than the control group, while
significantly higher results were seen at 90 days of age in only the high dose level group and in
only retina-lead levels. Blood-lead levels averaged 19 and 59 ug/dL in the low and high dose
groups at 21 days of age compared to  1 ug/dL in the control group.  Significant retinal
degeneration (i.e., rod and bipolar apoptic cell death) was associated with age and lead exposure
levels in this study. The higher loss rate of rods in the lead-exposed groups was associated with a
loss of rhodopsin content, implying that the loss was directly due to the presence of lead. The
authors concluded from these results (when considered with the results of other researchers) that
the developing retina may be more sensitive to lead exposure in pre-weaned rats than the
hippocampus.

      Also adopting the hypothesis that the effects of lead on the central nervous system leads
to adverse effects on the eye, Bushnell et al. (1977) document the findings of experiments to
characterize the relationship between high food-lead exposures early in life and unpaired
scotopic visual function (i.e., night blindness). In their experiments, baby formula was spiked
with lead acetate and given daily to six rhesus monkeys in their first year of life. Lead
consumption was regulated to allow blood-lead levels to be maintained at an average of 55 ug/dL
for three monkeys and 85 ug/dL for three monkeys. Four other monkeys whose formula was not
spiked with lead served as a control group.  Approximately  18 months after this feeding
paradigm was ended, monkeys in all three groups averaged nearly normal blood-lead levels. At
this time, the monkeys were administered a discrimination procedure at various light levels to
test their ability to select an option which provided reinforcement (i.e., food) versus an option
which did not provide reinforcement.  Animals exposed to the higher levels of lead in the first
year of life (i.e., the 85 ug/dL group) performed significantly worse in this procedure compared
to animals in the control and lower lead groups as light levels were reduced. Various controlling
factors within this experiment allowed the researchers to conclude that the primary reason for the
degraded performance in the higher-dosed animals was most likely a loss of scotopic function.
Thus, the researchers concluded that lead exposure early in life was associated with impaired
scotopic visual performance later in life, even when blood-lead levels in these animals were
allowed to return to normal levels.

      In  a study performed by Lilienthal et al. (1994),  15 rhesus monkeys  were pre- and post-
natally exposed to lead at one of three levels (0, 350,600 mg/kg lead acetate) until they were
nearly 10 years of age. At approximately 12 years of age, electroretinogram (ERG) recordings
were made of each eye in these monkeys. Five animals were in the control  group (average blood-
lead concentration of 0.44 ug/dL), four in the lower-dosed group (average blood-lead
concentration of 4.55 ug/dL), and six  in the higher-dosed group (average blood-lead
concentration of 8.26 ug/dL). Significant differences (at the 0.05 level)  were observed across

                                           16

-------
dose groups in the amplitudes (but not latencies) of the scotopic b-wave, with increased
amplitudes seen in the two lead-dosed groups, and the nature of the differences being dependent
on luminescence level. The lead-induced effects were similar to those observed earlier in these
animals and were similar to the effect of dopamine antagonists. This suggests that lead may be
permanently affecting dopaminergic processes.

       Over many years, significant visual impairment associated with lead exposure has been
observed at high lead levels in rabbits, rats, and monkeys, both at the retina and visual cortex
(Otto and Fox, 1993). Exposure to moderate levels of lead by rats can result in decreases in rod
cells, thinning of retinal layers, reductions in the number of axons in the optic nerve, and necrosis
of photoreceptors and cells in the inner'retinal layer (Banks et al., 1997; USEPA, 1986).  While
such damage is less frequently associated with low levels of lead exposure, functional and
neurochemical effects on the retinal system can be associated with low-level lead exposures in
rats (i.e., blood-lead concentrations below 20 ug/dL; Otto and Fox,  1993), along with persistent
decreases in visual acuity and spatial resolution (USDHHS, 1999).  While considerable data exist
to allow the neurotoxicity of lead to be characterized, considerably less data exist on the
morphological effects of lead on the visual system, especially at low exposure levels. In
addition, Otto and Fox (1993) have concluded that the effects of lead are more likely to adversely
affect rod cells compared to cone cells.

       Davis et al. (1990) reviewed several studies in which lead exposed rats and monkeys
exhibited decreased responsiveness of neurons to visual stimuli, as assessed by parameters (i.e.,
visual evoked potentials) which measure nerve conduction.  Ulienthal et al. (1990) found that
central visual processing, as reflected in measurements of visual evoked potentials (VEP), was
clearly affected in lead exposed monkeys. Monkeys were exposed to dietary lead pre- and
postnatally (low lead group at 350 ppm and high lead group at 600 ppm per day), and VEP was
measured in response to visual flash stimulation beginning at age seven. Results showed that
amplitudes of sensory-evoked potentials were smaller and latencies were longer in lead-exposed
monkeys relative to controls, even in monkeys with the lower lead exposure regime.

       The study by Nagymajtenyi et al. (1998) introduced earlier in this subsection also
observed disruptions in the functioning of optical nerves in prenatally and postnatally lead-
exposed rats. In their study, effects of lead exposure were most often significant in the middle
and high lead exposure groups, and included dose dependent changes in visual evoked potentials
and slowed nerve conduction velocity, as measured in response to flash stimulation.

       The auditory system. In a review of the literature (on both human and animal studies) on
the effects of lead exposure on the auditory system, Otto and Fox (1993) concluded that lead
exposure is more likely to adversely affect that portion of the auditory system that resides within
the central nervous system (e.g., cochlear nerve) compared to more peripheral sites where
sensory transduction processes occur. However, more data were deemed necessary to more
definitively characterize the effects of lead on specific sites within the auditory system and how
such auditory dysfunction contributes to a child's overall learning impairment that can be
attributed to lead exposure.
                                           17
                                                                U.S. EPA Headquarters Library
                                                                       Mail code 3201
                                                                1200 Pennsylvania Avenue NW
                                                                   Washington DC 20460

-------
       Certain animal studies have reported an association between lead exposure and
disruptions of the auditory system. For example, Lilienthal et al. (1990) observed lead-induced
alterations in auditory functioning in monkeys, as reflected in measurements of brainstem
auditory evoked potentials (BAEP). In their study, monkeys were exposed to dietary lead pre-
and postnatally (low lead group at 350 ppm and high lead group at 600 ppm per day), and BAEP
was measured in response to auditory click stimulation beginning at age seven. Results showed
that BAEP latencies increased with increasing click rates in lead-exposed monkeys relative to
controls, although the increase tended to be consistent in the group of monkeys with the higher
lead exposure regime (600ppm). The study by Nagymajtenyi et al. (1998) (cited earlier in this
subsection) reported other types of disruptions in the auditory pathway in prenatally and
postnatally exposed rats, including changes of auditory evoked potential and slowed nerve
conduction velocity, as measured in response to click stimulation.

       To assess the long-term auditory effects of chronic lead exposure at low levels, Lasky et
al. (1995) assessed auditory functioning in two groups of 11-year-old rhesus monkeys, where one
group contained 11 monkeys who were exposed to lead either pre- or post-natally (within the
first year of life), and the other group contained 8 monkeys were not exposed. Auditory function
in these monkeys was assessed by measuring distortion product otoacoustic emissions (DPEs),
auditory brain stem evoked responses (ABRs), and middle latency evoked responses (MLRs).
The two animals with the highest blood-lead concentrations during their first four years had
''markedly abnormal" DPEs, where this result was found not to be due to canal obstruction nor
middle ear problems. Among the remaining animals, DPEs did not differ significantly between
the two groups, although DPE amplitudes increased more rapidly (given the stimulus level) in the
control group compared to the exposed group.  The ABRs, but not the MLRs, differed
significantly between the two groups of animals.  The exposed animals tended to have slightly
longer latency ABRs.

       2.1.2.2  Behavioral and Developmental Effects of Lead.  Several recent review papers
have summarized previous research and advances in the area of neurotoxicological effects of
lead. The findings of key studies and of study reviews by Banks et al. (1997), Cory-Slechta
(1997), Davis et al. (1990), and Rice (1996), among others, are presented below. When possible,
attention was focused on health effects at low levels of lead exposure or at low blood-lead levels.
Many animal studies have investigated the neurobehavioral effects of lead exposure, that is,
effects upon learning and performance activities. Some studies also addressed related issues
such as: (1) what blood-lead levels were associated with learning and performance deficiencies in
test animals;  (2) whether other signs of toxicity were present; (3) what biochemical mechanisms
produce toxicity; and (4) how blood levels in exposed animals correlate with specific blood-lead
levels in humans at which analogous effects upon learning/IQ are observed. Many experiments
utilize rats or monkeys as subjects since they respond to motivational factors, usually food
rewards, and can be induced to learn certain behaviors of interest to researchers.

       To synthesize the neurobehavioral effects of lead across species, Davis et al. (1990)
conducted a review of rodent, primate, and human studies. The lowest levels of internal lead
                                           18

-------
exposure during early development at which neurobehavioral effects have been observed were
reported to be <20 ug/dL for rodents (Cory-Slechta et al., 1985) and 15-25 jig/dL for primates
(Rice, 1985). These ranges are similar to the lowest range (10-15 ug/dL) in which adverse
neurobehavioral effects have been observed in children (Davis et al., 1990; Fulton et al., 1987;
Silva et al., 1988; USEPA, 1999).

       Rice (1996) has summarized results from both human epidemiology and animal studies of
the potential behavioral and developmental effects of lead exposure as follows: "Increased
distractability, inability to inhibit inappropriate responses, perseveration, and inability to change
response strategy are common themes that may be extracted from both literatures." These
observations are noted in the findings presented throughout this subsection.

       Learning deficiencies and behavioral test performance. To investigate the association
between lead exposure and learning deficiencies, Rice (1996) conducted an extensive review of
animal studies, interpreted in conjunction with learning disabilities in children as reported in
human epidemiological studies. In experiments with rats or monkeys, a general learning
deficiency was frequently observed at high lead exposure levels. However, for monkeys exposed
to low or moderate lead levels, the majority of the impairment  was evident only with the
performance of more complex tasks, such as non-spatial discrimination reversal tasks. This
phenomenon was observed in (a) monkeys exposed to lead from birth with preweaning blood-
lead levels of 50 ug/dL and adult blood-lead levels of 30 ug/dL (Rice, 1988); (b) monkeys with
blood-lead levels of 30-35 ng/dL from infant formula and postweaning levels of 19-22 ug/dL
(Rice and Gilbert, 1990a); and (c) monkeys with blood-lead levels of 15-25 ug/dL during infancy
and steady-state levels of 11-13 ug/dL during adulthood (Rice, 1985).  In these studies, lead
exposure had ceased by the time performance tests were conducted.

       Kuhlman et al. (1997) conducted an experiment with 10 rats reared in each of five groups.
The control group received no lead exposure.  The "maternal exposure" group was exposed to
lead in utero and during lactation (750 ppm lead acetate in feed via dams), but was moved to a
control diet after weaning. The permanent group was exposed to lead both in utero and
continuously afterward into adulthood (750 ppm lead acetate in feed).  Finally, there were two
post-weaning groups, in which no exposure occurred until after weaning, and then pups were fed
diets containing two different lead concentrations (750 or 1000 ppm lead acetate in feed). Rats
were performance-tested using a water maze at about 100 days of age, which tested their ability
to locate a submerged platform, and their blood-lead levels were measured at that time. Even
though average blood levels in  that group (1.8 ug/dL) had returned to control levels by the time
of the test, a highly significant impairment in performance (e.g., longer time to find platform,
longer pathway to platform) was observed in both the maternal and permanent exposure groups.
The post-weaning exposure groups did not show any significant performance impairment, even
though average blood-lead levels exceeded 20 ug/dL at the time of testing. The authors
suggested that the results seen in the maternal group demonstrated the impact of lead upon early
development.
                                           19

-------
       In a study documented by Mello et al. (1998), rat pups were exposed to lead in utero and
during nursing, using 1.0 mM lead acetate administered to dams as drinking water. Eleven litters
from control females and nine litters from lead-exposed females were used, for a total of 160
pups. The pups were observed for physical development and tested for reflexes/behavior at aged
17-19 days. While lead exposure appeared to significantly accelerate the appearance of first eye
opening, startle reflex, and free-fall righting, it significantly impaired spontaneous alternation
performance in a maze. No explanation was given for the seemingly contradictory effects in this
study, though the authors suggested that any lead-induced alterations in animal development or
behavior, regardless of direction, must be considered deleterious.

       Operant discrimination and reversal tasks. Rice (1996) describes experiments that were
conducted with visual discrimination problems with  the addition of "reverse performance"
requirements, and/or the addition of irrelevant distracting signals.  In the operant discrimination
reversal task, the researcher changes the pattern of rewards so that previously-learned correct and
incorrect responses become switched. Lead-exposed animals sometimes perform as well as
controls in the original learning acquisition, but perform poorly when the change of rules requires
learning a change in strategy.  Also, lead-exposed animals tend to be distracted by irrelevant
details more than do controls (although in some cases they may perform similar to controls in the
absence of such distraction). For example, Gilbert and Rice (1987) report an experiment in
which monkeys exposed to low lead levels (50 and 100 ug/kg/day) through age 10 years tended
to perform poorly, relative to controls, on spatial discrimination reversal tasks with unfamiliar
distracting cues, but adequately on tasks with familiar distractions. Furthermore, the lower dose
group was impaired only during the tasks immediately after the introduction of the irrelevant
stimuli, but not after the irrelevant stimuli became familiar.  Under these dosing regimes, steady-
state blood lead concentrations in these mature adult monkeys (13.1 ug/dL for the higher dose
group and 10.9 ug/dL for the lower group) approximate levels typical for humans in
industrialized environments (Gilbert and Rice, 1987).

       Bushnell and Bowman (1979a) reported diminished performance on spatial
discrimination reversal tasks by adult monkeys (4 years of age) that were exposed to dietary lead
in the first year of life (either 0.287 mg/kg or 0.880 mg/kg per day as lead acetate in formula).
This finding was observed despite average blood-lead levels in each group being essentially
normal at the tune of testing.

       Impairments in learning and performance have also been noted in experiments with lead-
exposed rats. Cory-Slechta (1997) found that selective learning deficits were present after lead
exposures of 50 and 250 ppm (as lead acetate in drinking water) with resulting blood-lead levels
as low as 20-25 ng/dL. Lead-exposed rats performed as well as controls during the performance
component of the experiment (i.e., the correct sequence of responses remained constant across
trials), but less accurately during the repeat acquisition component (the correct sequence changed
in an unpredictable way with each new set of trials).  Rats also displayed perseverative behavior,
pressing the same lever repeatedly, even though the experiment precluded this pattern of
response from generating a food reward.  The author discusses the research attempting to
elucidate some of the biochemical mechanisms underlying these results, including evidence

                                           20

-------
which strongly suggests the existence of a link between learning impairments and lead-induced
disruptions of neurotransmitters, particularly those of the glutamine system.

       Cory-Slechta (1997) also reviewed the reported effects of lead on fixed interval (FT) test
performance (delayed response operant schedule of food reinforcement). The FI test requires the
animal to bar-press only at specific minimum time intervals before a food reward can be
provided.  Studies conducted by Cory-Slechta showed that rats exposed to low doses of lead (25-
300 ppm with postweaning exposures) tended to respond more rapidly than did controls, even
though this behavior resulted in withheld rewards. In contrast, animals exposed to higher doses
(500 ppm and above) showed decreased rates of FI responding, at least initially.  It is
hypothesized that the increased response rates may actually be a form of perseverative behavior.
The author notes that, in other studies, lead-induced interference with the dopamine system has
been suggested as a possible mechanism for perseverative behavior.  Also of significance, Cory-
Slechta reports that dose-dependent patterns in FI performance, like those in her own study, have
consistently been observed across a wide range studies and methodological conditions, including
species (i.e., reported effects described in rats, monkeys, sheep, pigeons, and mice) and
developmental period (i.e., prenatal, postnatal, postweaning, adult, old adult) during which lead
exposure occurs. Thus, evidence in the literature strongly suggests that changes in FI schedule-
controlled behavior seem to be one of the most reliable parameters, relative to other measures,
for assessing the behavioral effects of experimental lead exposures (Cory-Slechta, 1997).

       Rice and Gilbert (1990a, 1990b) conducted a series of behavioral impairment studies on
52 infant monkeys.  Shortly after birth monkeys were assigned in equal numbers to one of four
feeding groups: (1) a control diet; (2) a lead-containing diet continuously after birth; (3) a lead-
containing diet from birth until age 400 days, followed by a control diet; and (4) a control diet
from birth until age 300 days, followed by a lead-containing diet. Lead was administered orally
in gelatin capsules as lead acetate in 0.05 M sodium carbonate, equivalent to 1.5 mg/kg/day.
Feeding regimes were maintained up to and through the time of behavioral  testing, which
occurred from about 5 through 7 years of age. When monkeys were 5 to 6 years old, they were
tested on a series of nonspatial discrimination reversal tasks, including form, form with irrelevant
color cues, color with irrelevant form cues, and alternating form and color (Rice and Gilbert,
1990a). Based on this series of tests, the group dosed continuously from birth exhibited the
greatest degree of impairment, followed by the group dosed after infancy only. The group dosed
during infancy only did not exhibit significant impairment  on these tasks. The results of this
study provided evidence that while exposure to lead after infancy can produce impairment, this
effect is exacerbated if the animal was also exposed during infancy. Approximately one year
later, the same group of monkeys were tested on a spatial delayed alternation task in which they
were required to alternate responses between two push buttons, with an increasing delay in
response time (Rice and Gilbert, 1990b). In this series of tests, all three exposure groups showed
performance impairment relative  to controls and .were impaired to an approximately equal
degree, exhibiting "perseverative  behavior" and an inability to suppress inappropriate responses
during test delay intervals. The authors conclude that that there is not an early critical period for
lead-induced impairment on spatial delayed alternation tasks, and that lead exposure only during
                                           21

-------
infancy (i.e., < 400 days) can result in impairment comparable to exposure that continues beyond
infancy (Rice and Gilbert, 1990b).

       General neurobehavioral effects. A review of animal studies which evaluated the
persistence of lead-induced effects on cognitive development was conducted by long (1998).
Neurobehavioral toxicity was reported to persist for up to 10 years of age in monkeys exposed to
low levels of lead and was strongly suggested in some studies to be an irreversible neurotoxin.
The physiological evidence of lead disruption of the developing brain (e.g., neuron and synapse
formation), and the fact that intracellular lead may not be removable from neural cells, also
supports the plausibility of enduring, and possibly irreversible, deficits in neurobehavioral
function with lead exposure.

       Burger et al. (1998) conducted an experiment on 48 slider turtle hatchlings which were
randomly assigned in equal numbers to a control group and three lead exposure groups.  The
exposed groups experienced a single injection (intramuscularly) of one of three doses of lead
acetate (0.25,1.0, or 2.5 mg/g).  No survivors remained in the high dose group by 120 days of
age.  Behavioral observations, growth (weight and length), and survival data were taken both
prior to and at various times after lead injection (i.e., weekly during the first month, and at 4
weeks, 4 months, and 6 months post-exposure). The primary behavioral measure was a righting
response (i.e., latency before attempting righting and time required to turn over), as the animals
naturally tended to right themselves whenever placed on their backs, and other behavioral
measures were not reliably exhibited in this species.  By the age of 6 months, righting response
adjusted for body weight was significantly impaired by lead dosing. Both the low and medium
dose groups performed significantly worse than did controls, and the medium dose group
performed significantly worse than did the low dose group, indicating a dose-response effect.
The range of lead doses used in the study had such a marked impact on survival that "the
threshold for behavioral effects is on the same order of magnitude as the LD^" (i.e.,  the lowest
lead dose at which only 50% of animals would survive). The authors noted that although these
results suggest hatchling turtles are vulnerable to lead exposure, behavioral effects due to lead
may be very subtle in the field for this species.  In addition, when effects do become evident, the
levels of lead in turtles may be dangerously close to the threshold for lethality.

       USEPA (1986) contains an extensive review of experimental studies using rats exposed
to lead. A variety of behavioral and physiological responses affected by lead neurotoxiciry, as
well as some social interactions in rats, were examined. Although it was generally not possible
in the review to standardize results across all studies due to different dose levels and
administration modes being utilized across studies at different life stages in the animals, adverse
effects were often observed in rats with blood-lead levels around 30 ug/dL, and some effects
upon learning were detected even when maximum blood-lead levels were below 20 ug/dL.
Many studies (e.g., Angeil and Weiss, 1982; Cory-Slechta and Thompson, 1979; Geist et al.,
1985) have observed greater behavioral and physiological effects in rats exposed after weaning or
during maturation than in those exposed prenatally or during infancy. Generally lead-exposed
rats have been observed to acquire performance skills on discrimination tests more slowly than
controls, and to commit more errors.

                                           22

-------
       Social behavior and development. While lead exposure has been linked with the
incidence of certain behavioral difficulties in children (e.g., irritation, aggressiveness) since the
1940's, more recent studies have investigated whether such a link exists at low-level lead
exposures.  One such study investigated the development of social behavior in monkeys and has
been documented in Bushnell and Bowman (1979b). In this study, 21 newborn monkeys of both
sexes were randomized into three groups of 7 animals each, where the control group received no
added lead to their diet, while lead was spiked into the diet of animals in the lower and higher
dosed groups over a one-year period in order to achieve targeted average blood-lead
concentrations of SO and 80 ug/dL, respectively. Social tests were administered to the animals
once per week for 30-39  weeks, either alone, in the company of the other animals in its group, or
with animals of the same age from each group. The frequency and duration of "rough-and-
tumble play" was reduced in the lead-exposed groups compared to the control group (p < 0.05),
while the duration of "contact cling" was significantly higher in these groups. The researchers
suggested that although lead-exposed monkeys did occasionally respond to others' prompts to
play, they were less likely to initiate such invitations to others (although their tests did not
differentiate between initiating and responding to such play). The lead-exposed monkeys were
also observed to be less adaptable to a sudden change in their environment for social interaction
(which occurred at approximately 21 weeks of age in this study) compared to control animals.

       Bushnell and Bowman (1979b) followed up this experiment with one in which 16
monkeys were placed into four groups of four animals each, where one group served as a control,
one was dosed with dietary lead over a one-year period to achieve an average blood-lead
concentration of 80 ug/dL ("chronic"), one was dosed with lead for only two weeks at two
months of age ("pulse"), and one was dosed under both "chronic" and "pulse" regimens.
Animals were tested in a play cage until week 40 (when interactive play behaviors have typically
been developed), when they were moved to a play room. This study found that animals exposed
chronically to lead had reduced frequency and duration of "rough-and-tumble play," the
development of "initiated social explore," and the frequency of approach, compared to the
control and "pulse" groups. This suggested that reduced desire to play is associated with chronic
lead exposure during the period of social development.  Meanwhile, animals in the "pulse" group
did show significant behavioral effects when the play environment was changed at week 40,
suggesting a latent effect associated with an acute exposure early in life and which occurred upon
introducing a new and potentially intimidating environment. Furthermore, such effects were less
apparent among monkeys exposed to lead only in utero.

       In a study testing electric shock-elicited aggression  (Hastings et al., 1977), lead-exposed
animals (through post-natal lead exposure in mother's milk, where the dams were provided
drinking water at concentrations of either 0.02% or 0.20%) showed significantly less aggressive
behavior than did controls. This finding was the opposite of what the authors expected at the
time. At 60 days of age (i.e., the time of testing), the lead-exposed animals averaged blood-lead
levels of 5-9 ug/dL, and exposure levels were generally low in the study. Other tests performed
in this study (e.g., wheel-running, visual discrimination task with reversal) did not see significant
differences across exposure groups.

                                          23

-------
       A study (Cutler, 1977) which investigated gender differences with respect to rodent
behavior found that lead-exposed male mice showed significantly reduced levels of aggressive
behavior (even though body weights and overall activity levels were not affected) compared to
controls. In addition, both sexes showed significantly increased social/sexual investigative
behavior in the lead-exposed group compared to controls.  Total activity and the occurrence of
tremors or other jerky movements did not differ significantly between the exposure and control
groups.

       As cited in Section 2.1.2.1 above, the study performed by Nagymajtenvi et al. (1998)
observed an increase in bio-electric aberrations in the brains of rats exposed to lead pre- and
post-natally compared to control rats.  In addition, behavioral aberrations were observed in this
study among the lead-exposed rats. Results showed that lead dosing during pregnancy was
related to a significant, dose-dependent increase in hyperactive behavior (e.g., as measured by
higher ambulation rates and increased  grooming) in the offspring. The authors conclude that
low-level lead exposure during prenatal and postnatal development can interfere with normal
development and bioelectric functioning within the  nervous system and is associated with
behavioral changes. In addition, the authors suggest that the functional and behavioral effects of
lead, which occur without other overt signs of lead toxicity, are much more harmful than has
been previously supposed.

2.1.3 General Health Effects

       Although the health effects of lead are diverse, and in general depend on the duration and
degree of lead exposure, they are all thought to originate from lead's ability to interfere with
fundamental biochemical processes (i.e., mitochondria! energy production, calcium-mediated
processes, and protein function). All of the major types of health effects of lead that have been
observed in humans, including hematological, neurodevelopmental, immunological,
cardiovascular, and renal effects, have been demonstrated in controlled, dose-response studies in
rodents, dogs, and/or non-human primates.  Although conclusive evidence for carcinogenicity is
still lacking in human studies, several animal studies have indicated that there is an association
between high levels of oral exposure to several lead compounds and renal tumors in various
species (USDHHS, 1999). The U.S. Department of Health and Human Services has determined
that lead acetate and lead phosphate may reasonably be expected to be capable of causing cancer,
based on sufficient evidence from animal studies (USDHHS, 1999).

       While the focus of this report is on the neurobehavioral effects of lead exposure, any type
of health effect that is demonstrated to have resulted from lead exposure in animal studies could
be of interest, especially at low lead doses. Therefore, an overview of experimental animal
studies which have investigated the general physical health effects that result from lead exposure
is presented below. This overview will provide only brief statements of findings across animal

-------
studies. Much of the information presented here has previously been cited in USDHHS (1999)1
and USEPA( 1986).

       2.1.3.1  Death. At high levels of exposure, lead is known to cause death in humans
following severe lead encephalopathy, and has been suggested to be a causative agent in sudden
infant death syndrome (USDHHS, 1999). The data are minimal, however, regarding high
exposures to lead and death in animals. Increased mortality has been observed in studies with
rats and mice given lead in food or drinking water, although in some cases mortality did not
occur in a dose-related manner (USDHHS, 1999).

       2.1.3.2  Systemic Effects.
       Hematological Effects. There have been numerous studies in animals demonstrating the
adverse effects of lead on heme (hematin) biosynthesis, which can in turn affect many organ
systems. In acute and intermediate-duration studies, the activities of several enzymes involved in
the heme biosynthetic pathway have been observed to be altered by administration of lead to rats.
Adverse hematological effects have also been observed in rats and dogs in longer term, lower
dose studies (USDHHS, 1999).

       Renal Effects. The effects of lead on the renal system have been documented in animal
studies involving rats, dogs, monkeys, and rabbits.  Symptoms of renal insufficiency, including
both transient and irreversible kidney lesions, tubular dysfunction, and increased excretion of
amino-acids and nitrogen compounds in urine, have been observed at high lead exposures in
these studies (USDHHS, 1999; USEPA, 1986).

       Cardiovascular Effects. Exposure to lead has been associated with adverse
cardiovascular effects in studies with laboratory animals. While older animal studies concluded
that hypertension was clearly associated with extremely high doses of lead, the direct effects of
lead on blood pressure and the secondary effects (i.e., hypertension as a result of renal damage)
of lead were difficult to separate (USEPA, 1986). However, USDHHS (1999) includes  areview
of several more recent chronic-duration experiments in rats and found that lower lead exposures
(i.e., levels that were otherwise non-toxic) were associated with sustained increases in blood
pressure as compared to controls. Other adverse cardiovascular effects, such as structural and
functional changes  relative to controls (e.g., degeneration of the myocardium and aorta), have
also been observed in rats following lead ingestion (USDHHS, 1999).

       Hepatic Effects. Indications of possible lead toxicity to the hepatic system, as suggested
by increased liver weight, morphological changes, and changes in liver enzyme activity, have
been observed in animal studies involving mice, rats, dogs, and baboons (USDHHS, 1999;
USEPA, 1986). In general, adverse effects of lead on the liver in animal studies were observed at
high exposure levels.
         A 1998 draft of this document was also used in obtaining information for this section.

                                           25

-------
       Respiratory Effects. There is limited evidence from inhalation studies that prolonged
exposure to lead may cause respiratory system irritation in mice. Data suggest that lung
irritation, if significant, is largely dependent on the duration of exposure (USDHHS, 1999).

       Other Systemic Effects. Lead, at relatively high exposure levels, has also been associated
in experimental animal studies with other various systemic effects. These include general
impairment of cellular function (via interference with membranes, calcium ion transport, and
mitochondria! respiration), long-term visual system deficits in animals exposed during post-natal
development, weight loss, impairment of vitamin D metabolism, and endocrine disruption
leading to a decrease in thyroid function (USDHHS, 1999; USEPA, 1986).

       2.1.3.3 Immunoloaical Effects. The literature provides good evidence that lead can
have an irnmunosuppressive effect on animals at relatively low doses. Although additional
research is required to elucidate the exact mechanisms of action, the macrophage is postulated to
be the primary immune system target cell for lead (USEPA, 1986). Studies reviewed in USEPA
(1986) and USDHHS (1999) showed that in rats, mice, and rabbits, both oral and airborne lead
exposures contributed to increased susceptibility to bacterial and viral infections, decreased
antibody counts, decreased cell-mediated immunity, depressed lymphocyte function, and
suppressed macrophage-dependent immune response, as compared to control animals.  In many
of the studies cited, lead-induced immunosuppression occurred at low exposure levels that
otherwise induced no overt symptoms of systemic lead toxicity, such as increased blood pressure
or weight loss.

       2.1.3.4 Reproductive and Genotoxic Effects. The observed effects of lead on the
reproductive system are mixed across animal studies. Adverse effects observed included
decreased pregnancy rates, decreased male fertility, damage to ovaries and testes, and altered
pubertal progression, all presumably by interference with hormone production.  High doses of
lead have also been associated with fetal stunting and fetotoxicity. Lead was not observed to be
teratogenic in limited rodent studies, except when lead was administered via injection
(USDHHS, 1999). In various animal studies on the genotoxic effects of lead, chronic exposures
to lead produced either slight or no significant increases in chromosomal aberrations in mice or
monkeys, except in one study (a 1977 study by Deknudt and associates, as cited in USDHHS,
1999) in which monkeys were given a calcium-deficient diet (USDHHS, 1999).

       2.1.3.5 Carcinoqenicitv.  Most animal studies conducted to investigate the
carcinogenicity of lead compounds have considered only one or two doses, preventing detailed
dose-effect characterizations from being performed (USDHHS, 1999).  Available data generally
suggest that lead acetate and lead phosphate can be carcinogenic following ingestion by
laboratory animals, and that the most common tumor is renal.  There  are currently no animal data
on the carcinogenicity of lead by inhalation or dermal exposure.

       In its review of animal studies performed over a 30-year period, USDHHS (1999) has
reported that high levels of oral exposure (approximately 25-100 mg/kg/day) to lead acetate and
                                          26


-------
lead phosphate have been shown to increase the incidence of renal tumors in rats and mice
(USDHHS, 1999). However, due to the extremely high cumulative doses of lead used in these
studies, and the uncertainties regarding mechanisms by which lead induces tumors in the rat
kidney (i.e., by acting on species-specific or trans-species proteins), extrapolation to low-level
exposure hi humans is difficult. Therefore, observations of lead-induced kidney tumors in rats
may not be relevant to humans (USDHHS, 1999). The carcinogenicity of lead at lower doses in
animals is yet to be determined.

2.1.4 Conclusions from Animal Studies Investigation

       Lead has been observed to have widespread neurotoxic effects, as well as behavioral and
cognitive symptoms, in humans.  These observations are largely consistent with the findings of
morphological, electrophysiological, and biochemical studies on animals. Animal studies are
also congruent with observations of lead exposure in humans, suggesting an increased
susceptibility of the young brain to lead poisoning (Banks et al., 1997). In addition, animal
studies have provided physiological evidence that many of the effects of lead on the
differentiation of the developing nervous system, such as synaptic and dendritic development,
and myelin and other nerve structure formation, have the potential to be long-lived effects
(USEPA, 1986). Although animal models do not duplicate the human response to lead exposure,
they do serve to provide strong support for expecting certain health effects to occur when humans
are exposed, hi addition, animal studies allow for more specific determination of dose-response
relationships. Single-route exposure and dose-response data are generally not available for
humans. Animal studies provide an opportunity to elucidate some of the physiological
mechanisms of lead toxicity, and thus they are a valuable tool for assessing the potential risks of
lead exposure for human health.

       A large resource of published literature also exists in which animal experiments were
conducted to investigate the neurobehavioral effects caused by lead dosing.  Most of these used
monkeys or rats hi experiments to assess the effects of lead dosing upon learning and
performance activities and were conducted using low to moderate doses of lead.  In many studies
using monkeys, no other overt signs of toxicity (e.g., weight loss) were observed even when
learning impairments were observed.

       Several experiments have demonstrated that lead-dosed animals perform similarly to
controls on some simple tasks (e.g., visual discrimination of simple shapes), but perform worse
than controls on more complex tasks such as reverse discrimination tasks (where the animals
have to change a previously-learned strategy) or fixed-interval/delayed response tests.
Perseverative behavior is a recurring phenomenon in the latter type of test, whereby the animal is
unable to inhibit inappropriate, repetitious movements despite having rewards withheld as a
consequence.

       Experiments to compare the severity of impairments in animals dosed only during early
development (i.e., prior to birth or during infancy) versus those dosed only during maturation did
                                           27

-------
not yield consistent results. Generalizations about dosing schedules could not be made across all
studies or across species, as to which groups of animals faired worse.
       There are general similarities in studies and across species, however, with regards to the
lowest levels of internal lead exposure during early development at which neurobehavioral
effects are commonly observed. For example, the review by Davis et al. (1990) of earlier
research findings concluded that internal lead exposures associated with neurobehavioral effects
were reported to be as low as <20 ug/dL for rodents, <15  ug/dL for primates, and 10-15 ug/dL
for children. Although direct comparisons of blood-lead levels are generally not considered to be
accurate across species, evidence that rodents and primates may actually tolerate higher
exposures to lead before reaching a given blood-lead level, suggests that exposures and blood-
lead levels of concern would even lower in humans than in animals (Davis et al., 1990).

       The published literature varied in the amount of detail given to explaining experimental
designs used. For example, it was often implied but not always explicitly stated that subjects
were randomized to exposure groups according to some statistical criteria, and that the
experimenters observing the performance tests were blind to the dosing status of subjects.
Articles also provided varying levels of detail on experimental procedures, such as the amount of
lead provided through food or water.  Blood-lead levels at the time of performance testing were
usually measured and reported.

       The choice of response endpoint was clearly of particular importance hi the design of
performance tests.  For example, as noted in Burger et al. (1998), behavioral tests need to include
procedures in which the given species (absent of lead exposure) will readily participate,
otherwise there is a risk that the response endpoint will not be sensitive enough to allow the
researcher to detect significant differences between dosed  and control animals.  Also, lead
exposure may delay or accelerate the physical development of some reflexes, so peak
performance may occur at different times in different exposure groups. Therefore, the time at
which statistical comparisons are made may produce seemingly paradoxical results. In
particular, any apparent non-monotonic dose-response relationships must be interpreted with
caution.

       Despite the aforementioned caveats, animal studies do provide an opportunity to elucidate
some of the physiological mechanisms of lead toxicity, as  well as generate single-route exposure
and dose data that are generally not available in humans. Although animal models cannot claim
to duplicate the human response to lead exposure, the resulting ability to control confounding
factors permits these models to succeed in providing substantial evidence that supports the
existence of a causal relationship between low level lead exposure and neurological impairment,
especially in the young.
                                           28

-------
2.2    SUPPORT FOR THE CAUSALITY OF ADVERSE HEALTH
       EFFECTS DUE TO LEAD EXPOSURE

       Chapter 2 of the §403 risk analysis report documented previously-published information
on adverse health effects associated with lead exposure to humans. Subsequent chapters
characterized how environmental-lead exposure impacts selected blood-lead concentration and
health effect endpoints in children. Specifically, IQ-based health endpoints were selected to
represent the neurological effects associated with lead exposure.

       A concern, particularly with the IQ-based health endpoints, is whether lead can be
assumed to cause the adverse health effects. For ethical reasons, the controlled lead exposure
studies necessary to test this hypothesis can not be performed on humans. However, animal
studies can be used to supplement the evidence of human studies in this area. To this end,
Section 2.1 of this document presented key findings of animal studies that investigated the
impact of lead exposure on adverse health effects. This section examines whether the combined
evidence of human and animal studies suggests that lead exposure causes neurological damage
that can be measured through intelligence testing.

2.2.1  Principles of Causality

       On the issue of causality, Needleman & Gatsonis (1990) make the following assertions
and present the following principles of causality (citing Kenny, 1979, as a reference):

       "Epidemiologic studies cannot, by themselves, establish causal relationships.
       Causality is not subject to empirical proof, whether in the field or in the
       laboratory. Given that direct demonstration of proof of a low-dose lead effect in
       a naturalistic setting is not achievable, epidemiologists rely on canons that, if
       satisfied, permit the conservative drawing of causal inferences. They are (1) time
       precedence of the putative cause, (2) biologic plausibility, (3) nonspuriousness,
       and (4) consistency."

Needleman & Gatsonis (1990) also introduce a fifth principle, biologic gradient, which is part of
biologic plausibility. Needleman (1998) indicates that these principles originate from
investigations of whether tobacco use causes cancer and attributes them to the British statistician
Sir Austin Bradford Hill. Together, these principles can be applied to the findings of existing
studies (both human and animal studies) to imply that lead exposure causes neurological damage
that may be measured through IQ score decrements.

       The  following paragraphs discuss the five principles of causality relative to the evidence
in human and animal studies.

       Time precedence. Time precedence means that the proposed cause must exist before the
proposed effect occurs. That is, a child must become exposed to lead before neurological deficit
is observed.

                                          29

-------
       In cross-sectional studies in humans, it is impossible to establish the time precedence of
lead exposure and neurological deficit. Longitudinal studies in humans, however, have shown
that disturbances in early neurobehavioral development occur even at low lead exposure levels in
early life. For example, in Boston, 4 to 8 point differences in performance on the Bayley Mental
Development Index were reported at 6,12,18, and 24 months, after adjusting for other
covariates, when children with low prenatal blood-lead levels (mean of 1.9 ug/dL) were
compared to children with modestly elevated prenatal blood-lead levels (mean of 14.6 ug/dL)
(Bellinger etal., 1985a, 1985b, 1986a, 1986b, 1987a; 1991; 1992). Additional detail on this and
other longitudinal studies is presented in Section 2.2.2 below and Section 2.3.1 of the §403 risk
analysis report.

       Controlled animal studies further support the time precedence of lead exposure. For
example, groups of monkeys exposed to lead either continuously from birth or only after infancy
showed learning impairment on a series of non-spatial discrimination reversal tasks, relative to a
control group that were not exposed to lead (Rice and Gilbert 1990a).  These lead-exposed
monkeys, along with a group exposed only during infancy, also showed performance impairment
relative to controls in a spatial-delayed alternation task at age 6-7 years. Rats exposed to lead in
utero and during lactation displayed a highly significant impairment in performance on a water
maze test at 100 days of age, relative to controls, even though the average blood-lead level had
declined to 1.8 Ug/dL by this time (Kuhlman et al., 1997). Additional animal research described
in Section 2.1 also supports the conclusion that lead exposure precedes the neurological deficit.

       Biologic plausibility. Biologic plausibility means that the causal relationship between
exposure and adverse health effects must be consistent with known biological function.

       While investigation of the mechanisms of lead toxicity remains the subject of active
research, it is known that lead can interfere with cell function by competing with essential
minerals, such as calcium and zinc, for binding sites on membranes and proteins. Lead binding
to membranes or transport proteins can inhibit or alter ion transport across the membrane or
within the cell.  In the brain, lead can substitute for calcium and zinc in ion transport events at the
synapse (i.e.,  the junction where the axon of one neuron terminates with the dendrite of another
neuron, through which nerve impulses must travel to move from one nerve cell to another).  The
normally-developing brain appears to delete synapses that are unused and to keep and strengthen
synapses that are used. Goldstein (1990,1992) suggested that lead may disrupt, or delay, the
development  of synapses and that, perhaps, the resulting connections in the brain are "poorly
chosen,"  leading to functional impairment. Silbergeld (1991) found that exposure of fetal
animals to lead affects both regional growth and synaptogenesis, with synaptogenesis being the
more sensitive.  Although some of these conclusions are speculative, it is biologically plausible
that exposure to lead causes neurological damage.

       Biologic gradient.  Biologic gradient means that a dose-response relationship must be
present (i.e., increased doses of lead should cause increased impairment).
                                           30

-------
       It is well known that high-level exposure to lead produces encephalopathy in children,
starting at blood-lead levels of 80 to 100 ng/dL. At lower exposure levels, IQ decrements, fine
motor dysfunction, and disturbances in neurobehavioral development have been related to
varying levels of lead exposure. These effects are summarized hi Section 2.3.1 of the §403 risk
analysis report. As summarized hi Section 4.4 and Appendix D2 of the §403 risk analysis report,
many studies have focused on estimating the magnitude of this dose-response relationship.

       Nonspuriousness. Nonspuriousness means that confounding factors associated with
adverse health effects must be ruled out.

       Studies that investigate the relationship between children's IQ and blood-lead
concentration have adjusted for other demographic factors that may affect neurological
development, such as Home Observation for Measurement of the Environment (HOME) score,
maternal IQ, and socioeconomic status. These studies are summarized in Appendix D2 (Tables
D2-1 and D2-2) of the §403 risk analysis report. In many of these studies, the level of lead
exposure remains a highly-significant factor after adjusting for these potential confounding
factors. Because the method of adjusting for confounding factors hi human exposure studies
does not necessarily remove all confounding, however, animal research is required to supplement
human subject research. Animal studies minimize confounding by administering lead in
controlled doses to randomly-selected subjects that are genetically similar and are otherwise
treated similarly.  Controlled animal studies described in Section 2.1 of this document support
the conclusion that lead exposure precedes neurological deficit hi addition, these studies
demonstrate that the effects are nonspurious by establishing study designs that include control
groups and that attempt to control for confounding factors.

       Consistency. Consistency, also known as coherence, requires that the phenomenon be
demonstrated in different studies under similar, but not identical conditions.

       Human studies investigating the relationship between blood-lead concentration and IQ
scores have generally associated IQ decrements with increases in blood-lead concentration in
various populations around the world, as summarized in Appendix D2 (Tables D2-1 and D2-2)
of the §403 risk analysis report.  The magnitude of the estimated IQ decrement varied from study
to study, as may be expected, and the relationship was not always statistically significant.
However, the relationship was consistently negative (i.e., increased blood-lead concentration was
associated with IQ decrement in more than ten studies). Furthermore, a loss of approximately 2-
3 IQ points was associated with an increase hi blood-lead concentration from 10 to 20 ug/dL in
several of these studies.

2.2.2 Causality As Addressed in Longitudinal Studies

       The consistency principle of causality implies that causality can not be concluded from
the findings of a single human monitoring study. In turn, the time precedence principle indicates
that repeated data collection over time for study subjects within longitudinal studies provides an
important component of an investigation into the specific role of lead exposure as the cause of

                                          31

-------
certain adverse health effects. Therefore, the findings of multiple longitudinal studies on the
association between lead exposure and diminished performance on cognitive function or
intelligence testing can be an important contribution to the argument of causality (while the
ability to conclude causality exclusively from such findings remains very^limited). The level of
importance of the contribution increases when consistent findings are observed across studies
(and across different cohorts having different demographic characteristics and lead exposure
potential).

       This section provides some key findings from two longitudinal studies (Boston and Port
Pirie) that can be used in evaluating the hypothesis of causality. This is not meant to be an
exhaustive presentation of all results (significant or otherwise) across all longitudinal studies, but
instead is a presentation of only key findings from selected studies.  For example, results from
other longitudinal studies that monitored lead exposure (e.g., Cleveland, Cincinnati, Sydney,
Yugoslavia) have not been reviewed.  In addition, the potential for reverse causality (i.e., children
with lower levels of intelligence are more prone to elevated blood-lead concentrations) is not
addressed.

The Boston Prospective Study (Bellinger et al.. I985a, 1985b, 1986a, 1986b,  1987a; 1991; 1992)

       The Boston prospective study considered infants born at the Brigham and Women's
Hospital hi Boston, MA, from August, 1979, to April,  1981.  Cord blood-lead levels were
measured for 9489 infants (97% of all available infants), and those whose cord blood-lead levels
were within one of the following three categories were considered for the study:  < 3 ug/dL
(low), 6-7 ug/dL (mid), and 210 ug/dL (high).  These categories represented the 10th, 50*. and
90th percentiles of the cord blood-lead distribution observed in the first three months of the study.
A total of 249 infants in these categories were enrolled in  the study (85 in the low category
having mean  1.5 ug/dL, 88 in the mid category having mean 6.5 ug/dL, and 76 in the high
category having mean 14.6 ug/dL).

       The cohort was considered to be of high socioeconomic standing (e.g., 87% of the
enrolled infants were white; 92% were considered to come from intact families). While such a
cohort tends to have a lower likelihood of lead exposure than more disadvantaged children (e.g.,
those in poor, inner-city neighborhoods), thereby restricting the ability to generalize results to a
wide population, the low occurrence of certain demographic conditions that are typically highly
correlated with lead exposure gives this study a greater opportunity to isolate the effect of lead
exposure (especially at low levels) on cognitive function.

       Post-natal blood-lead concentrations were measured on the study cohort at ages 6,12,18,
24, and 57 months, and at age 10 years.  Capillary blood was obtained through 24 months of age,
and venous blood was obtained at ages 57 months and 10  years.  Cognitive function was
measured by the Mental Development Index (MDI) of the Bayley Scales of Infant Development
at 6,12,18, and 24 months of age, by the McCarthy Scales of Children's Abilities at age 57
months, and by the Wechsler Intelligence Scale for Children-Revised (WISC-R) and Battery
                                           32

-------
Composite scores on the Kaufman Test of Educational Achievement -Brief Form (K-TEA) at age
10 years.

       Investigations on the association between lead exposure (as measured by cord-lead or
blood-lead concentration) and intellectual functioning within the study cohort were performed at
various time points during the study.  The extent of association was measured by multiple
regression modeling, adjusting for parameters that represent potential confounding factors.  This
resulted in the following key findings:

        1.     At ages 6,12,18, and 24 months, children in the high cord-blood group had
              significantly lower MDI scores relative to each of the other two groups (4.8 points
              lower than the low group and 3.8 points lower than the mid group) at the 0.05
              level. Furthermore, the MDI scores at these ages were not significantly associated
              with post-natal blood-lead concentrations measured up to that time (Bellinger et
              al., 1987).  The level of association increased upon adjusting for potential
              confounders within the multiple regression equation.2

       2.     By age 57 months, the association between cord-blood grouping and cognitive
              function diminished considerably from what was observed at  earlier ages and was
              no longer statistically significant, except in the instance where only children with
              blood-lead concentrations at or above 10 ug/dL at age 57 months was considered.
              However, a significant (inverse) association was observed between blood-lead
              concentration at 24 months of age and score on the McCarthy scales at 57 months,
              upon adjusting for potential confounders3 (Bellinger et al., 1991). The perceptual-
              performance subscale  of the McCarthy scales, which measures visual-spatial and
              visual-motor integration skills, was especially sensitive to post-natal lead
              exposure.

       3.     The association between increased blood-lead concentration at 24 months of age
              and observed deficits in full-scale and verbal IQ scores was statistically significant
              at age 10 years, even after adjusting for confounding variables4 (p=0.007 for full-
              scale IQ, 0.004 for verbal IQ, but only 0.091 for performance  IQ). Blood-lead
              concentrations  at other ages, however, were not significantly associated with such
              deficits  at the 0.05 level.
         Mother's age, mother's race (white vs. nonwhite), mother's IQ (as measured by the Peabody Picture vocabulary test),
mother's education level, # of years mother smoked, # alcoholic drinks per week by mother in the third trimester of pregnancy,
Hollingshead Four-Factor Index measure of family social class, quality of care-giving environment, child's sex, child's birth
weight, child's gestaiional age, child's birth order.

         Family social class, maternal IQ, marital status, preschool attendance, HOME score, # hours per week of "out-of-
home" care, # family residence changes, recent medication use, # adults in household, gender, race, birth weight, birth order.

       4 HOME score at 10 years, total HOME score at 57 months, child stress, maternal age, race, maternal IQ,
socioeconomic status, sex, birth order, maternal marital status, and # family residence changes prior to 57 months.

                                              33

-------
Finding #1 shows that despite a child's lead exposure during the first two years, cognitive
function during this period is more significantly associated with pre-natal exposure, which is
generally less prone to behavioral and environmental confounding than is direct, post-natal
exposure.  As a child ages, as seen in Findings #2 and #3, post-natal lead exposure (especially its
peak at approximately two years of age) becomes more dominant than pre-natal exposure in its
association with the child's performance on intelligence tests. This suggests that the effect of
cumulative post-natal exposure through approximately age 2 years eventually outweighs pre-
natal exposure, hi particular, children can eventually see reduced performance on intelligence
testing if their post-natal lead exposure becomes significant, despite their pre-natal lead exposure.
Furthermore, a child's post-natal lead exposure tends to change at a faster rate than other
demographic variables that are highly correlated with blood-lead levels early in life, with the
more recent lead exposures being predictive of a child's current health consequences. These
findings support the hypothesis of causality, especially in the relatively homogeneous and highly-
privileged cohort considered in this study.

       Typically, in environments having nigh lead levels (e.g., inner-cities, smelter/mining
communities), correlations between blood-lead concentrations measured at different ages are so
highly correlated that it is difficult to separate out the age effects. However, the lower age-to-age
correlations observed in the Boston study (resulting from relatively low lead exposure potential)
allowed for investigating age-specific vulnerabilities within this study.

The Port Pirie Cohort Study (Baghurst et al., 1992; Tong et al., 1996; Bums et al., 1999)

       This study consisted of 723 subjects bom in the lead smelting community of Port Pirie,
Australia (and surrounding rural communities) from May, 1979, to May, 1982.  Cord-blood was
obtained and analyzed for lead (geometric mean = 8.3 ug/dL). In addition, capillary blood
samples were collected at ages 6  and 15 months and annually from ages 2 through 7 years.  A
venous blood sample was collected at age 11-13 years. From the measured blood-lead levels,
average lifetime blood-lead concentration was calculated for each child using trapezoidal
integration. Geometric mean blood-lead concentrations increased to 21.2 ug/dL by age 2, then
declined to 7.9 ug/dL by age 11-13, when 375 children remained in the cohort.  Children from
more advantaged backgrounds were more likely to remain in the cohort through age 11-13 years
than children from disadvantaged backgrounds.

       Measures of developmental status were made at ages 2 years (Bayley scales), 4 years
(McCarthy scales), and at 7 and 11-13 years (Wechsler intelligence scale). In addition,  emotional
and behavioral problems were assessed by their mothers using a Child Behavior Checklist, and
other demographic parameters were measured via questionnaire.

       The key findings within this longitudinal study on how lead exposure is associated with
decreased developmental status measures were as follows:
                                            34

-------
       1.     The inverse relationships between IQ at age 7 years and average lifetime blood-
              lead concentration measured at 15 months and at 2, 3, and 4 years were
              statistically significant at the O.OS level after adjusting for potential confounding
              factors5. This result held for both full-scale and verbal IQ measures, but not
              performance IQ.

       2.     Despite blood-lead concentration declining after age 2 or 3 years, the association
              with cognitive development continued into later childhood. Furthermore, lifetime
              blood-lead concentration was significantly associated with childhood emotional
              and behavioral problems (after adjusting for such confounding factors as HOME
              score, maternal psychopathology, and child's IQ) at ages 11-13 years.

Finding #1 was consistent with the Boston prospective study in that no significant relationship
was observed between IQ and pre-natal blood-lead concentration at approximately age 7 years,
while the relationship was significant when considering blood-lead concentration at 2 years of
age.

2.2.3 Conclusions on Causality

       The combined weight of human and animal studies provide evidence, consistent with the
principles of causality presented in Section 2.2.1, that lead may be assumed to cause adverse
neurological effects in young children. In particular, longitudinal studies in humans have shown
that disturbances occur in neurobehavioral development early in life even at low lead exposure
levels. These studies have observed effects of lead exposure even after accounting for other
demographic factors (e.g., socioeconomic status, parents' IQ) that could affect neurological
development. While these other demographic factors tend to be highly correlated with blood-
lead levels early in life, the influence that post-natal lead exposure has on blood-lead tends to
increase with a child's age. This is because over time, measures of a child's lead exposure tend
to change at a faster rate than the child's demographic measures, and more recent lead exposures
continue to be predictive of a child's current health consequences. For the §403 risk analysis,
adverse neurological effects are assessed through IQ score decrements.  Reasons for selecting IQ
score decrement as a health effect endpoint are presented in Section 2.5.2 of the §403 risk
analysis report.

2.3   THE ASSOCIATION  BETWEEN BLOOD-LEAD CONCENTRATION
       AND IQ SCORE

       In its risk characterization, the §403 risk analysis used IQ score  decrement associated
with lead exposure as the basis for measures of neurological effects. As the risk analysis used
         Sex, parents' level of education, maternal age at delivery, parents' smoking habits, socioeconomic status, quality of
home environment, maternal IQ, birth weight, birth order, feeding method (bottle, breast, both), duration of breast feeding,
whether the child's natural parents were living together.

                                            35

-------
blood-lead concentration as its primary measure of body lead burden to quantify environmental-
lead exposure, it was necessary to determine IQ score as a function of blood-lead concentration
and to characterize the extent to which a change in IQ score occurs when blood-lead
concentration changes within a child.  The following assumptions were made in this analysis on
the association between blood-lead concentration and IQ score for the representative population
of 1-2 year old children:

       •      The relationship between blood-lead concentration and IQ score decrement was
              assumed to be linear.

       •      The risk characterization assumed a loss of 0.257 IQ points per 1 ug/dL increase
              in blood-lead concentration (with alternatives of 0.185 and 0.323 considered in
              sensitivity analyses).

       •      No threshold was assumed in this relationship (i.e., no blood-lead concentration
              exists below which a relationship between blood-lead concentration and IQ score
              is not apparent), although selected non-zero thresholds have been assumed in
              sensitivity analyses presented in Sections 5.1.5 and 6.4.2 of this report.

While the §403 risk analysis report discusses the basis for making these assumptions (e.g., see
Section 4.4 and Appendix D2), this section presents additional information that is necessary to
judge the correctness and accuracy of the assumptions.  Section 2.3.1 addresses the linearity and
slope assumptions, while Section 2.3.2 addresses the threshold assumption.

2.3.1  Linearity and Slope Assumptions

       How was such an assumption made? As researchers have used primarily linear and log-
linear models to characterize the relationship between blood-lead concentration and IQ scores,
these two types of models were considered for use in the risk analysis.  The log-linear model
predicts IQ score as a linear function of log-transformed blood-lead concentration (plus other
important confounding variables, such as maternal IQ and HOME score), while the linear model
does not take a log transformation  of the blood-lead concentration.  The scientific community
does not appear to have reached a consensus on which form is more appropriate. For example,
the meta-analysis in Schwartz (1994) included three studies that employed log-linear models and
four studies that employed linear models.

       To obtain a single measure of the relationship that would be comparable across studies,
despite the different model forms used, Schwartz (1994) used the change in IQ score associated
with a doubling of blood-lead concentrations from 10 to 20 ug/dL.  The meta-analysis (Schwartz,
1994) yielded an estimated decrease of 2.57 IQ points for an increase in blood-lead concentration
from  10 to 20 ug/dL. This was the slope estimate used in the §403 risk analysis.

       Using the measure discussed in the previous paragraph, Schwartz (1994) provides some
evidence that the log-linear relationship may be more appropriate than the linear relationship. In

                                           36

-------
an analysis to investigate the presence of a threshold (see Section 2.3.2), Schwartz (1994)
estimated an IQ point decrement of 3.23 IQ points for the three studies with mean blood-lead
concentrations below 15 ug/dL, compared to a 2.32 IQ decrement for the four studies with mean
blood-lead concentrations at or above 15 ug/dL. Thus, if anything, a trend toward greater IQ loss
associated with lower blood-lead concentrations was observed. This result is consistent with a
log-linear relationship.

       Despite this evidence, a linear relationship was applied in the §403 risk analysis.  The
assumption of a linear model reduces the likelihood of overestimating the number of children
with low blood-lead concentrations at risk, or who may benefit from actions taken in response to
the §403 standards. See Section 4.2.1 and Appendix D2 of the §403 risk analysis report for
additional information.

       Additional information: Tables D2-1 and D2-2 in Appendix D2 of the §403 risk analysis
report summarize a total of 18 studies that report the relationship between children's blood-lead
concentration and IQ. Each of these studies was used in at least one of the meta-analysis studies
reviewed in Appendix D2 of the §403 risk analysis report.  Table 2-2 provides a subset of the
information previously reported in Tables D2-1 and D2-2 of the §403 risk analysis report for
these studies and also reports the type of model (linear or log-linear) which each study used to
predict IQ as a function of blood-lead concentration.

       A few of the studies included in Table 2-2 used a log-linear model rather than a linear
model to characterize the effect of blood-lead concentration on IQ.  However, the overall
evidence that these studies provide regarding a log-linear relationship was more limited than the
existing evidence on a linear relationship.  Furthermore,  if EPA had adopted a log-linear model
approach, the risk analysis would have estimated that blood-lead concentration had a greater
impact on IQ at lower levels than at higher levels. This would have resulted in a greater
possibility that the risk analysis would have overestimated benefits at lower levels, compared to
underestimating benefits. For these reasons, EPA felt that a linear model was the better approach
over a log-linear model. However, the §403 risk analysis did include a sensitivity analysis which
considered the effects of a steeper slope in the linear model, in order to evaluate the possibility of
underestimating the relationship between blood-lead concentration  and IQ.

       A recent article by Marais and Wecker (1998) has suggested that researchers who have
characterized IQ as a function of blood-lead concentration using linear regression techniques
have often reported biased estimates for the effect of blood-lead concentration on IQ for one or
both of the following reasons:

       •      by not having all four of the following predictor variables in the model:  blood-
              lead concentration, mother's intelligence, father's intelligence, and socioeconomic
              status.

       •      by not taking into account measurement error in these predictor variables.
                                           37

-------
 Table 2-2.   Summary of Key Findings from Studies that Investigate the Relationship
              Between Blood-Lead Concentration and IQ Score
"•; .;''••
.Study
Hatzakisetal. (1987)
Hatzakisetal. (1989)
Bellinger et a!. (1991)
Bellinger etal. (1992)
Baghurst etat. (1992)
Ernhart et al. (1989)
Cooneyetal. (19911
Schrocder et al. (1985)
Hawk »t al. (19861
Dietrich etal. (1993)
Yule at at. 0981)
Lansdown et al. (1986)


Winneke etal. (1990)



Silva (1988)
Harvey et al. (1988)
Wang at al. (1989)
Winneke et al. (1985a)
Fulton et al. (1987)
. ' •"
Type of
Study ~
Prospective
Prospective
Prospective
Prospective
Prospective
Prospective
Prospective
Prospective
Replication
of Schroeder
Study
Prospective
Pilot Study
Replication
of Yule
Study


Multi-Center,
Cross -
Study


Cross -
Sectional
Cross -
Sectional
Cross -
Sectional
Cross -
Sectional
Cross -
Sectional
-
Location
Lavrion, Greece
Lavrion, Greece
Boston, MA
Boston, MA
Port Pirie,
Australia
Cleveland, OH
Sidney, Australia
Wake County,
NC
Lenoir & New
Hanover
counties, NC
Cincinnati, OH
London, England
London,
England
Bucharest
Budapest
Moden
Sofia
Dusseldorf
Dusseldorf
Dunedin, New
Zealand
. Birmingham,
England
Shanghai, China
Nordenham,
Germany
Edinburgh,
Scotland

: N
509
509
150
147
494
212
175
104
75
231
166
166
301
254
216
142
109
109
579
177
157
122
501
- ->• ; •-;
PfaB
Mean (SO)
(pgMU
23.7 (9.2)
23.7 <9.2)
6.4(4.1)
6.5 (4.9)
20
16.7(6.45)
14.2

20.9 (9.7)
15.2(11.3)
13.52(4.13)
12.75(3.07)
GM **18.9 4 1.3)
GM~18.2(1.7)
GM»11.0(1.3>
6M = 18.2 (1.6)
CM = 8.3 (1.4)
7.4(1.3)
11.1 (4.91)
12.3(0.2)
21.1 (10.11)
8.2(1.4)
GM = 11.5
'. - .
IQ Score
Mean (SE»

87.7 (14.8)
115.5(14.5)
119.1 (14.8)
104.7
87.5 (16.6)

Range <•
45-140
Range =
59-118
86.9(11.3)
98.21 (13.44)
105.24 (14.2)




116

108.9(15.12)
105.9(10.6)
89
120.2(10.3)
1 12 (13.4)
Association Be
~ r~,..,'-i»
Change fold
as PbB '
increases
TVOOI
1030 figlOL
-2.7
-2.7
-1.6
-5.8
-3.3
-1.1
+0.4
-2.0
-2.6
-1.3
-5.6
+ 1.5






-1.5

-9

-2.6
iMneatQa
.dUveb
'P-VakM
<0.001
<0.001
0.23
0.007
0.04
<0.01

<0.01
<0.05
•cO.10
0.084
0.63
<0.1
<0.1
<0.1
<0.1
<0.1
<0.1



<0.1
0.003
nd Blood-
Model
KOCEf!
linear
linear
log-linear
linear
log-linear
linear
linear
linear
linear
linear
log-linear
Jog-linear
linear
linear
linear
linear
linear
linear
log-linear
linear
linear
linear
log-linear
PbB = blood-lead concentration (ug/dL); SD = standard deviation: GM = geometric mean
                                             38

-------
By analyzing data from four case studies, the authors imply that the bias overestimates the effect,
thereby making it likely that the researcher would declare that blood-lead concentration
(especially at low levels) has a significant effect on IQ, when in reality, such an effect is
insignificant.  The authors show how to arrive at an estimate of the blood-lead effect that is not
subject to this bias; this estimate is a function of the correlations among the four predictor
variables and the measurement variability associated with these variables. The article prompted
two responses that were published simultaneously with the article, both of which challenged the
article's conclusions.

       Despite their claims, the findings in Marais  and Wecker (1998) have not resulted in any
change to the approach taken by the §403 risk analysis to characterize the relationship between
blood-lead concentration and children's IQ for the following reasons:

        •       Sensitivity analyses performed by other researchers6 and documented in a
               response published with the article by Marais and Wecker (1998) have shown that
               the approach to obtaining an "unbiased" estimate for the effect of blood-lead on
               IQ is highly sensitive to the values of the estimates for the correlations and
               variability among the four predictor  variables that are input to the calculation,
               thereby implying that input values that do not represent the target population can
               lead to a highly inaccurate estimate for this effect under this approach.

        •       The authors have not shown that any overestimation associated with this bias is
               always significantly large enough to warrant concern or will result in an incorrect
               declaration that blood-lead has a significant effect on IQ.

        •       The meta-analysis documented in Schwartz (1994), which the §403 risk analysis
               used to characterize the relationship between blood-lead concentration and IQ,
               utilizes the estimated blood-lead effects on IQ that were reported by seven studies,
               six of which estimated these effects  after taking into account both parental IQ and
               socioeconomic status (i.e., the predictor variables of most concern to Marais and
               Wecker). Furthermore, the §403 risk analysis considered not only the outcome of
               this meta-analysis, but also the sensitivity analyses associated with this analysis,
               when investigating the effect of deviation from the meta-analysis outcome on the
               risk analysis results.

        •       The need to adjust for measurement  error in the predictor variables is not relevant
               to the §403 risk analysis, as the goal is to predict how a measured blood-lead
               concentration (after adjusting for the measured values of other potentially
               important variables) is associated with IQ.
       6 Watemaux, C, Petkova, E., and DuMouchel, W. "Comment: Problems with Using Auxiliary Information to Correct
for Omitted Variables When Estimating the Effect of Lead on IQ." Journal of the American Statistical Association. 93:505-513.

                                             39

-------
       The issue of whether a linear model is appropriate over the entire range of blood-lead
concentration must address the presence of a threshold in the relationship. This is discussed
further in the next section.

2.3.2  Threshold Assumption

       Despite the claims of some researchers on the presence of a threshold in the blood-
lead/IQ relationship, the majority of findings across studies and in meta-analyses have failed to
find sufficient evidence of a non-zero threshold. Furthermore, when claims of a non-zero
threshold were made, the value of this threshold (when suggested) differed considerably across
these claims. Therefore, the approach taken in the §403 risk analysis was to assume that no
threshold exists (although risk calculations assuming certain non-zero threshold values have been
included in sensitivity analyses found in Sections 5.1.5 and 6.4.2 of this document).

       In Section 3.3 of USEPA (1998b), the SAB concurred that "available data have not
identified a clear threshold," and, therefore, "the assumption of no threshold for lead effects on
IQ score  is both defensible and appropriate statistically." However, it was desired to document
the technical justifications for this assumption more thoroughly.  Furthermore, the investigation
into the presence of a threshold could be addressed by evaluating whether the dose-response
function is linear across the entire range of blood-lead concentration.

       How was such an assumption made? The assumption of no threshold made in the §403
risk analysis was based on the findings of Schwartz (1994), who noted that the presence of a
threshold would result in a decline in the estimated slope associated with blood-lead
concentration as the range of blood-lead concentrations declined across studies. However, as
mentioned in Section 2.3.1 above, a larger effect size was observed in the four studies with mean
blood-lead levels of 15 ug/dL or lower (-0.323 ± 0.126)  compared to the other three studies
(-0.232 ± 0.040). This observed trend toward higher slopes at lower concentrations discounted
the likelihood of a threshold.

       Also, Schwartz (1994) examined data from the Boston prospective lead study (discussed
in Section 2.2.2 above) specifically to investigate the presence of a threshold. This study was
selected as it had the lowest mean blood-lead concentration at two years of age (6.5 (Jg/dL;
n=133) of the studies considered in the meta-analysis, thereby allowing thresholds  at low blood-
lead levels to be identified if present. In addition, the study cohort's high socioeconomic (SES)
standing  may have limited the likelihood of certain confounding, and the Boston study
coordinators found relatively weak association between blood-lead concentration at two years of
age and various sociodemographic characteristics and psychosocial environment parameters
(Bellinger et al., 1986; Bellinger et ah, 1992).

       Schwartz's examination of the Boston study data involved fitting two separate regression
curves to the same set of covariates: one using IQ score (at 10 years) as the dependent variable,
and the other using blood-lead concentration (at 2 years) as the dependent variable. The
covariates included age, race, stress and HOME scores, maternal IQ, educational level and

                                           40


-------
occupational status for each parent, mother's time working out of the house, marital status,
gestational age, birth weight, mother's use of alcohol during pregnancy, otitis media history, birth
order, and SES. Then, a nonparametric smoothed curve (LOESS) was used to characterize the
residuals from the IQ score regression as a function of the residuals from the blood-lead
regression. The residuals were used in this curve-fitting exercise as they represent blood-lead
and IQ score measures after any effects of the above covariates have been removed. The LOESS
technique allowed for nonlinear curve fits, such as those that would result if a threshold was
present. The curve fit suggested that IQ score decrement was associated with declines in blood-
lead concentration even when blood-lead levels were below 5 ug/dL, supporting the hypothesis
that a blood-lead threshold on IQ score decrement was essentially not present.

       In Schwartz (1993), this nonparametric smoothing approach was performed on McCarthy
index data collected at age 57 months and blood-lead concentration data collected at 24 months,
as recorded hi the Boston prospective lead study (Bellinger et al., 1991). Again, after adjusting
for potential confounding variables, a definite relationship was observed even at levels below 10
Hg/dL, with no evidence of a threshold (Schwartz, 1993). To allow any potential threshold to be
identified, a piecewise-linear regression model was fitted to these data which allowed the
relationship to resemble a "hockey-stick" (i.e., the fit resembled two lines of different slopes that
meet at some point representing the potential threshold, with the line below the threshold having
nearly a zero slope, and the line above the threshold having a larger, positive slope). This model
fit suggested that any potential threshold would be less than 0.0001 ug/dL (Schwartz, 1993).

       The meta-analysis by Pocock et al. (1994) involving 26 studies concluded that no single
study has collected a sufficient amount of information to make definitive statements on the
presence of a threshold, and contradictory results (due to chance) on the presence of a threshold
can be observed for different studies. Thus, the analysis did not have enough evidence to reject
the hypothesis that no threshold exists.

       Identifying a threshold.  If a statistical hypothesis is used to determine the presence of a
threshold, the test should take the following form:

       Null hypothesis: No threshold exists (i.e., the "threshold" is at 0 fig/dL)
       Alternative hypothesis: A non-zero threshold exists.

If a statistical test is used as a scientific basis for making a decision, one either rejects the null
hypothesis or fails to reject it. One never says that the null hypothesis is "true." Therefore, given
a set of data and the statistical methods being applied, one either rejects the hypothesis that no
threshold exists or cannot reject it.

       The statistical method used to test the above hypotheses can also vary from study to
study. In general, the method is applied as part of an investigation  into a dose-response
relationship between blood-lead concentration and IQ.  Two examples of statistical approaches
are as follows:
                                           41

-------
       •      Several studies (e.g., Dietrich et al., 1993; Hatzakis et al., 1989) investigated
              dose-response by placing the study cohort into from 5 to 10 groups according to
              blood-lead concentration, determining the predicted IQ score associated with the
              mean blood-lead concentration in each group (using some.pre-determined
              regression model), calculating confidence intervals associated with the prediction,
              and determining how the groups differ in their predictions (as well as any patterns
              among the groups).

       •      Another approach focuses on attempting to fit the piecewise-linear "hockey-stick"
              regression model discussed above that predicts IQ score as a function of blood-
              lead concentration (and other confounding variables), where the fitted line has a
              different (larger, positive) slope once blood-lead concentration achieves a certain
              level, which is interpreted as a threshold value.

       Problems associated with suggesting that a threshold exists:  Determining whether a
threshold exists in the relationship between blood-lead concentration and IQ score is problematic
due to the difficulties in accurately characterizing the blood-Iead/IQ relationship and the inability
to generalize findings across studies and to the nation as a whole.  Major sources of these
difficulties include the following:

       •      Different protocols for measuring IQ and different IQ measures (e.g., performance
              IQ, verbal IQ, full-scale IQ measures are all associated with the Wechsler
              protocol) are used in different studies.

       •      Study designs differ, as do the methods used to make inferences from the data.

       •      Children's IQ can be difficult to measure and can be more variable than adult IQ.

       •      Outcomes are often highly dependent on the given set of confounding variables
              being considered. This set differs from one study to the next. Furthermore, when
              multiple studies consider the same confounding variables, these variables are
              often measured differently, using different protocols, from study to study.

       •      Different ages of children and different ranges of blood-lead concentration are
              found across studies.

       A non-zero threshold would result in reduced estimates of the likelihood of adverse
health effects, as children with blood-lead concentrations below the threshold would no longer be
labeled as experiencing an exposure-related IQ decrement. The level of reduction would depend
on how large the value of the threshold is.  Thus, if a  decision on a non-zero threshold was made
in error, the incidences of adverse health effects would be underestimated. The impact that a
non-zero threshold has on reducing the risk estimates calculated in this risk analysis is addressed
in sensitivity analyses presented in Sections 5.1.5 and 6.4.2 of this document.
                                            42

-------
       Examples of Possible Non-zero Thresholds Concluded from Study Findings.  Relatively
high thresholds (e.g., 10 ug/dL or above) have been suggested in some older studies conducted
10 or 15 years ago. However, some of the higher suggested thresholds appear to have lost their
legitimacy as they are higher than the levels for which more recent studies have observed some
type of health effect.  For those older studies that did not report a possible threshold, many
involved children with a range of blood-lead concentrations that would be considered high by
today's standards. Thus, the findings from these studies cannot be used to determine whether
thresholds exist at lower lead levels (i.e., levels below the observed ranges). A study's design
must allow for a sufficiently large range of blood-lead concentrations, and in particular, cover a
sufficient range of lower-lead levels (i.e., below 5  or 10 ug/dL), to ensure that any threshold
value would occur within the observed range.

       Some researchers attempting to prove the existence of a threshold have reviewed results
of applying the first statistical approach above (i.e., making predictions within groups of the
cohort), but have made conclusions based upon simple plots of the results rather than by citing
the outcome of statistical comparisons. For example, Kaufman (1996) has concluded that
threshold effects may exist at about 20 ug/dL from data presented in Dietrich et al. (1993), at
from 10-15 ng/dL from data presented in Bellinger et al. (1992), and at from 25-35 ug/dL in
Hatzakis et al. (1989). However, each conclusion  was based on visually interpreting selected
figures within these articles rather than on the results of controlled statistical hypothesis tests.
Furthermore, the following must also be considered when interpreting these conclusions:

       •       Dietrich et ah (1993): From this prospective study conducted in Cincinnati, OH,
              Kaufman (1996) cites the authors' presentation of predicted Wechsler Scale
              Performance IQ (PIQ) for four groups of children approximately 6.5 years of age,
              where the groups are determined by lifetime mean blood-lead concentration (i.e.,
              average blood-lead concentration measured at 3-month intervals from age 3 to 60
              months, and at ages 66 and 72 months). The predicted PIQ score was lower in the
              group with the highest lifetime mean blood-lead concentration (>20 ug/dL)
              compared to the other groups.  However, the authors caution against interpreting
              this finding as evidence of a threshold effect, as most children in this group had
              one or more individual measurements above 30 ng/dL. Furthermore,  in his meta-
              analysis, Schwartz (1994) considered a different relationship cited by the authors:
              full-scale IQ as a function of average blood-lead concentration through age 3 years
              (and other covariates, including HOME score and material IQ). This relationship,
              cited in Appendix D2 of the §403 risk analysis report, is more relevant to the
              representative population in the risk analysis.

       •       Bellinger et al.  (1992): From this prospective study conducted in Boston, MA,
              Kaufman (1996) cites the authors' presentation of predicted WISC-R full-scale IQ
              and K-TEA Battery Composite scores for four groups of 10-year old children,
              where the groups are determined by blood-lead concentration at 24 months of age
              (PbB^). Kaufman (1996) indicates that differences were apparent only between
              the two lowest and the two highest  groups, suggesting an apparent threshold

                                          43

-------
              between them (i.e., 10-15 ug/dL).  However, little difference among any of the
              four groups would have been identified if the assertion was based on confidence
              intervals associated with the predictions, rather than standard errors for the
              individual groups. Meanwhile, the regression model used to predict IQ from
              PbB24 indicated a highly significant linear trend (p=0.007) across the entire range
              of observed values of PbB24. This trend was present even among the three groups
              having the lowest blood-lead concentrations, suggesting that the trend is in fact
              present at the lower range of the observed concentrations (i.e., below IS ug/dL).

       •      Hatzakis et al. (19891: From this study conducted in Greece within a city in which
              lead mining and smelting occurred, Kaufman (1996) cites how the authors present
              predicted full-scale IQ for primary school-aged children grouped by blood-lead
              concentration. Blood-lead concentrations in this study were high: the average
              blood-lead concentration in this study was 23.7 ug/dL, no child had blood-lead
              concentration below 7 ng/dL, and more than 90% of the children exceeded 10
              ug/dL. The mean predicted IQ in the first two groups (s!4.9 jJg/dL, 15-24.9
              ug/dL) appeared to be statistically equivalent, then steadily declined for the
              remaining three groups, suggesting the presence of a threshold around 25 ng/dL.
              However, many other more recent studies (including animal studies) have
              observed neurological and developmental effects at lower blood-lead
              concentrations, making the concept of a threshold at 25 ug/dL highly unlikely. In
              fact, the linear regression model developed hi this study (which included 17
              covariates) had a highly significant slope for blood-lead concentration (p < 0.001)
              across the entire range of data in this study, even though more than 50% of the
              data occurred from 7-25 |ig/dL. Furthermore, it is unclear if different conclusions
              would have been made if the groups of children were defined differently.

These examples illustrate the complexities associated with characterizing the dose-response
relationship and the ability to conclude that a threshold exists in this relationship.

       The findings of a study by Fulton et al. (1987), which was included in the Schwartz
(1994) meta-analysis, appear to discount the high threshold level suggested by the findings of
Hatzakis et al. (1989). This study was conducted on 501 children aged 6-9 years in Scotland.
The predicted B ASC score was calculated for ten groups of children determined by log-
transformed blood-lead concentration. No evidence of a threshold was found among these data,
and the estimated slope associated with log-transformed blood-lead concentration was significant
(p=0.003). These findings were observed despite having only 10 study children with blood-lead
concentrations exceeding 25 (ig/dL.

       Concluding Possible Thresholds for Tooth-Lead Concentration:  Some studies (e.g.,
Bellinger and Needleman, 1983; Rabinowitz et al., 1992) have observed the potential for
thresholds in  tooth-lead concentration when relating tooth-lead concentration to IQ score.  Based
on their investigation of the relation between tooth-lead concentration  and IQ score, Rabinowitz
et al. (1992) suggested that a threshold for blood-lead concentration exists at approximately 8

                                            44

-------
ug/dL. Their investigation was centered around their 1989-1990 study of 764 children in grades
1-3 in Taiwan (an average age of 6.7 years). In this study, teeth shed by these children were
analyzed for lead. In addition, the children were administered Raven's Colored Progressive
Matrices (CPM) test, the score of which is considered a measure of IQ (average=25, SD=5.7).  A
model was developed which found CPM test score to be highly correlated with selected non-lead
predictors (parental education level, sex, grade level, and whether or not the child is
ambidextrous). The difference between the model-predicted and observed test scores for a child
(the "CPM score deficit") was interpreted as a measure of the change in the test score that results
from lead exposure.

       Each of the 380 children for which CPM score deficit could be calculated was placed into
one of two groups according to whether or not their tooth-lead level (|Jg/g) exceeded a specified
value.  Then, a Mann-Whitney test was performed to determine whether the mean CPM score
deficit differed significantly between the two groups. This was done for a series of grouping
values for tooth-lead, from 2 to 6 ug/g.  Significant differences between the two groups (p <
0.05) were seen at grouping levels of tooth-lead at 3.5 ng/g or above, but not at 3 ug/g or below.
Therefore, the authors concluded that a tooth-lead threshold for intelligence deficit existed at
approximately 3.25 ug/g. Finally, the authors relate this tooth-lead threshold value to blood-lead
by applying a modeled relationship between tooth-lead and blood-lead levels (formulated from
data for 88 Boston children aged 57 months), and concluding that a tooth-lead threshold of 3.25
Hg/g corresponded roughly to a blood-lead threshold of 8 ug/dL (± 2 ug/dL).

       While the cohort was considered to  have low tooth-lead concentrations, the following
must be considered when interpreting the above conclusion on the presence of a threshold and its
relevance to the §403 risk analysis:

       •      A non-zero threshold existing for tooth-lead concentration does not necessarily
              imply that one exists for blood-lead concentration, as tooth-lead may impact
              children's health in a different way from blood-lead.

       •      When noting the lack of significant difference between two groups defined by
              whether or not  tooth-lead exceeds a given threshold when this threshold gets low
              enough, it is uncertain of the extent to which the lack of significance is actually
              due to reduced power to detect differences as the sample size in the non-
              exceedance group declines with the threshold being considered.

       •      It is uncertain whether the model to predict blood-lead based on tooth-lead from
              the Boston study, which was used to obtain the blood-lead threshold estimate of 8
              ug/dL, can be applied directly to the findings of the Taiwan study without needing
              to consider certain statistical issues. For example, measurement error associated
              with tooth-lead levels may differ between the Boston and Taiwan  studies.

       •      Rabinowitz et al. (1992)  state that,  when attempting to fit a "hockey-stick"
              regression model to the data, "this data shows no change in the slope (or intercept)

                                           45

-------
             of the lines across any trial threshold." While this statement appears to support
             the hypothesis that no tooth-lead threshold exists, the authors do not provide any
             further information on the outcome of this model-based analysis.

2.3.3  Verifying the Results of Schwartz (1994)

       Schwartz (1994) applied a random effects modeling approach suggested by DerSimonian
and Laird (1986), a highly-regarded reference on meta-analysis. As a result, it was considered
among the best encountered by the §403 risk analysis, and therefore, was an important
contributor to how the §403 risk analysis characterized the relationship between blood-lead
concentration and IQ score in children. For this reason, the meta-analysis findings were verified
as part of the §403 risk analysis.  Using either the weighted noniterative method or the weighted
maximum likelihood method suggested by DerSimonian and Laird (1986), the §403 risk analysis
obtained the same finding as Schwartz (1994): that a decrease of 0.257 (± 0.041) IQ points was
associated with an increase in blood-lead concentration of 1.0 ng/dL within the range of 10-20
ug/dL. In addition, it was noted that heterogeneity of variance among the seven studies
considered by Schwartz (1994) was not significant, and the random effects model gave the same
results as a fixed effects model.

2-4    IMPACT OF CERTAIN RESIDENTIAL DUST
       CHARACTERISTICS ON DUST-LEAD EXPOSURE

       The bioavailability of lead can be an important factor in determining the toxic effects of
lead exposure to children within a specific environment.  Because lead is found in a variety of
chemical and physical forms depending on its source, the bioavailability of lead has been studied
as a function of chemical make-up (i.e., the particular form of lead present) and particle size in
various environmental matrices (e.g., dusts and soils, mining wastes).  Generally, the literature
concludes that the bioavailability of lead can depend on, among other things, the particular lead
species present (which varies depending on the source of lead), the size of the lead-containing
particles, the matrix incorporating the lead species, and the types of nutrients or other compounds
ingested with the lead (Freeman et al., 1992; USEPA, 1994). It has been suggested that lead
speciation and particle size may affect the bioavailability of lead through their influence on
solubility (USEPA, 1994). For example, lead bioavailability appears to be lower in mining areas
relative to urban and smelter areas.  Some authors (e.g., Rieuwerts and Farago, 1995; Davis et al.,
1995; Freeman et al., 1992) have suggested that this difference may be, in part, explained by
variations in chemical form (dissolutions rates) and particle size. Studies have also shown that
correlations between soil-lead and blood-lead levels are influenced by particle size and
composition of the lead compounds (USDHHS, 1992).

       The purposes of this section are:

       1.     To present a brief review of the some of the available literature which specifically
             examines the bioavailability of lead in dust, as a function of particle size and
             chemical composition of the dust.

                                          46

-------
2.4.1
2.     To determineif there is evidence which warrants the consideration of particle size
       and lead speciation. as related to lead bioavailability in dust, in the §403
       rulemaking; and, if warranted, determine if there is sufficient information
       available in the literature to allow for a thorough consideration.

3.     If relevant, to identify significant information gaps and potential issues that may
       warrant further research.

Review of Literature: Effects of Chemical Composition on
Lead Bioavailability in Dust
       There is substantial evidence in die scientific literature that the particular chemical
species, as well as the matrix (e.g., mineralogy, organic matter content) within which the lead
compound is found, are important hi determining lead bioavailability (USDHHS, 1999; USEPA,
1994). Many of the studies in the literature have been based on comparisons of relatively simple
lead compounds in controlled animal feeding studies or have focused on lead bioavailability in
urban and mining-associated soils. With respect to household dust in particular, there is
relatively little hi the literature which specifically examines the relationship between
bioavailability and chemical composition.

       The literature does recognize, however, that the composition of interior dust is
substantially influenced by soil and exterior dust (Diemel et al., 1981; USEPA, 1994). For
example, USEPA (1994) characterizes total lead in household dust as being comprised of soil-
lead, air-lead, lead from outside sources (e.g., workplace, school), and lead from household paint.
As a default value to the IEUBK exposure model, EPA has set the ratio of household dust-lead
concentration to soil-lead concentration at 0.70, which was considered appropriate for
neighborhoods or residences where loose particles of surface soil are readily transported into the
house (USEPA, 1994). Thus, soil particles have the potential to be a significant contributor to
lead levels in household dust.

       The following sections will provide a brief review of scientific findings related to the
general physical and chemical principles of lead and how they relate to bioavailability differences
in controlled environments and in soil studies.  Because the literature recognizes that the
composition of interior dust is influenced by soil and exterior dust, discussion of these general
bioavailability factors will, in the absence of more dust-specific data, serve as a starting point in
understanding the relationship between lead bioavailability in household dust and chemical
composition and particle size.

       2.4.1.1 Research on  Lead Bioavailabilitv in Controlled Animal Studies.  The relative
bioavailabilities of simple lead compounds have been studied under controlled conditions in
animal studies. For example, Barltrop and Meek (1975) (as cited hi USDHHS, 1992) compared
the absorptions of 12 different lead compounds in rats by measuring the kidney contents
following oral exposure. They found that the absorption of metallic-lead (particle size 180-250

                                            47

-------
     was the lowest of the lead compounds tested. Data also suggested that the absorption of
lead sulfide (particle size <50 /zm) was significantly less than the oral bioavailability of other
lead salts (oxide, acetate). Lead carbonate had the highest absorption, which was suggested to be
due to its high solubility in gastric juice.

       Dieter et al. (1993) also found differing blood-lead, bone-lead and kidney-lead levels in
rats fed different lead compounds, indicating variability in bioavailability.  For example,
maximum blood-lead levels were higher (80 jUg/dl) in rats fed lead acetate and lead oxide, hi
comparison to rats fed lead sulfide and a lead ore concentrate. Similar differences were observed
in bone-lead and kidney-lead levels between the rats receiving the more soluble (e.g., lead acetate
and oxide) and less soluble (e.g., lead sulfide and ore) lead compounds.

       2.4.1.2  Research on Lead Bioavailabilrtv in Soils.  The literature also suggests that the
soil matrix itself can be an important factor in determining the bioavailability of lead. For
example, Freeman et al. (1992) found that tissue-lead concentrations were lower in rats fed lead-
contaminated mining waste soils from Butte, Montana, as compared to rats fed comparable doses
of soluble lead acetate.  It was suggested that the inherent chemical properties of soil-adsorption
sites and the alteration of lead-bearing solids (e.g., encapsulation processes which inhibit
dissolution) may reduce the bioavailability of soil-lead, as compared to lead ingested without
soil. In general, the fate and bioavailability of lead in soils are affected by the species of lead
incorporated into the soil, the degree of absorption at mineral interfaces, precipitation of solid
phases, and the formation of relatively stable complexes/chelates with organic matter, as well as
other complex soil matrix factors such as pH (USDHHS, 1999; McKinney, 1993; Freeman et al.,
1992). This has been suggested to be largely due to the influence of these factors on solubility,
although it is important to note that solubility is but one factor in the bioavailability of lead to
humans or animals (USEPA, 1994).

       m a study of lead bioavailability in soil, Laperche et al. (1997) found that apatite (calcium
fluoride phosphate) amendments to a lead-contaminated soil lowered the bioavailability of soil
lead (as determined by plant uptake) by inducing the formation of geochemically-stable lead
phosphate compounds.  Similarly, in a study of an old mining village with elevated lead levels in
both garden soils and house dust, Cotter-Howells (1994) identified the predominance of lead
phosphate compounds (of limited bioavailability) as probable explanation of why blood-lead
levels in the village were not elevated in an otherwise contaminated area. In a study of mining-
associated soils in Butte, Montana, Davis et al. (1995) suggested that the predominance of lead
sulfide/sulfate and oxide/phosphates in soil and mine waste samples might provide an
explanation for the limited lead bioavailability that was observed when the Butte soils were fed
to rats in a previous study (Freeman et al., 1992).

       Urban area soils are typically contaminated with alkyl lead species originating from
combustion of leaded gasoline; lead halides (chlorides and bromides) from auto exhaust
participates; or lead carbonate, chromate, and octoate (as chips, flakes, and dusts) from exterior
and interior lead-based paint (USEPA,  1994). Lead halides in soils  are quickly transformed to
                                            48

-------
(or associated with) oxides or sulfates (USEPA, 1986 as cited in USEPA, 1994). In many lead-
mining districts, the predominant form of lead is galena or lead sulfide (USDHHS, 1992).

2.4.2 Review of Literature:  Effects of Particle Size on Lead Bioavailability in Dust

       Data in the literature are limited with specific regards to how particle size of lead-
contaminated house dust influences the bioavailability of lead in the dust. However, studies have
been conducted to examine the general relationship in soils between particle size and
bioavailability, as well as particle size and lead concentration.  For example, Barltrop and Meek
(1979) found that the bioavailability of lead in the intestinal tract of rats fed metallic-lead of
various particle sizes increased fivefold as particle size decreased from 197 microns to 6
microns. Particle size, due to kinetic limitations that control dissolution rates in the
gastrointestinal tract, has also been hypothesized to contribute to the lower bioavailability of lead
observed in mining waste soils relative to urban and smelter soils (Davis et al., 1995). In general,
the smaller the particle size, the greater the absorption of lead due to more rapid dissolution
(small particles have higher surface area to mass) in the gastrointestinal tract (Freeman et al.,
1992).

       Que Hee et al. (1985) found that when lead concentrations were measured in dust samples
categorized by size fraction, lead concentration was generally independent of the particle size.
However, most of the dust particle mass (about 75%), and thus most of the lead (about 77%),
was present in the <149 /zm size fraction. Lead concentration in smaller particle size ranges may
possibly maximize intestinal absorption, and thus increase bioavailability (USDHHS, 1992).
Duggan and Inskip (1985) performed an extensive literature review oh the variation of lead
concentration with particle size and reported that higher lead concentrations are usually found in
the smaller-sized fractions of soil and dust.  As reported by the Agency for Toxic Substances and
Disease Registry, numerous studies have also observed the lead content of soil, street dust, city
dust, and house dust to increase with decreasing particle size (USDHHS, 1992).

2.4.3 Information Gaps, Issues and Conclusions

       Although the literature is generally lacking in data which directly address the
bioavailability of lead in household dust as a function of chemical composition and particle size,
by recognizing that ulterior dust composition can be greatly influenced by outside soil, it can
reasonably be expected that the factors which affect lead bioavailability in soils will also
influence the bioavailability of lead in household dust. Therefore, based on general knowledge of
the bioavailability of simple lead compounds and studies of lead compounds in soil matrices,
evidence suggests that particle size and chemical composition have the potential to significantly
affect lead bioavailability in dust.  Nonetheless, the current information base which specifically
addresses particle size and chemical composition of dust as factors in lead bioavailability may be
inadequate to determine  how such factors can reasonably be incorporated into the rulemaking
effort.  Furthermore, needing to characterize dust by particle size and lead by chemical speciation
within a risk assessment will likely add to the expense of dust analyses, and dust standards that
                                           49

-------
distinguish between these various characterizations could add considerable complexity to the
rule.

       Specific uncertainties that remain concerning bioavailability of lead in household dust
include the following:  physical and chemical properties that may be unique to dust versus soil;
whether the effect of lead speciation in dust is significant enough to affect dust standards for
lead; distributions of lead across particle sizes found in household dust (e.g., whether dust is
enriched with the smaller size fraction relative to outside  soil) and whether particle size
differences are significant enough to affect standards; and possibly variances in exposure
mechanisms that may occur across particle sizes.

       The need for further research in these and related areas has been supported by several
authors.  For example,  Freeman et al. (1992), based on comparisons of mining waste soils and
other soil types in reviewed studies, emphasized the importance of evaluating the soil mineralogy
and lead species present when predicting bioavailability values for lead in soils.  In addition,
USEPA (1994) in the Guidance Manual for the Integrated Exposure Uptake Biokinetic Model
for Lead in Children, notes that adequate characterization of lead contaminated media, for the
purpose of estimating bioavailability, should include assessment of physical and chemical
parameters, such as particle size and media solubility.
                                            50

-------
3.0    EXPOSURE ASSESSMENT

       According to Chapter 3 of the §403 risk analysis report, the goal of the exposure
assessment was to document the important sources of lead in the environment, to document the
major pathways by which children are exposed to lead, to characterize the current (baseline)
distribution of environmental-lead levels in the nation's housing stock, and to characterize the
current distribution of average blood-lead concentration among the nation's children.

       In particular, Chapter 3 introduced those data sources used to characterize environmental-
lead levels in the nation's housing stock and presented summaries of household average lead
levels in dust and soil as reported in these studies. The U.S. Department of Housing and Urban
Development (HUD)'s National Survey of Lead-Based Paint in Housing ("HUD National
Survey", Section 3.3.1.1 of the §403 risk analysis report) was selected as the data source for
characterizing baseline environmental-lead levels in the nation's housing stock. Pre-intervention
data from other selected studies, such as the Rochester Lead-in-Dust study and the ongoing
Evaluation of HUD Lead-Based Paint Hazard Control Grant Program ("HUD Grantees") were
also summarized in Section 3.3.1 of the §403 risk analysis report to provide supporting
information on environmental-lead levels and to obtain information on the relationship between
these levels and blood-lead concentration in children.

       Since the §403 risk analysis report was published, additional data on environmental-lead
levels in the nation's housing stock have been made available to EPA. These data include interim
data from the National Survey of Lead and Allergens in Housing, and additional data from the
HUD Grantees evaluation.  In  addition, updated data from the U.S. Census Bureau are available
on numbers of young children associated with the various types of lead exposures found in the
national housing stock.  Some comments on the §403 proposed rule suggested that EPA use these
additional data when available. Therefore, EPA has investigated these new data to document
additional,  more recent information on lead levels in the nation's housing stock and, when
available, blood-lead levels in children exposed to these lead levels. For example, it was of
interest to document more recent information on the distribution of lead levels in dust deposited
on interior uncarpeted floors and window sills (i.e., the surfaces included in the proposed §403
standards), as well as on other types of surfaces (e.g., exterior surfaces, window troughs) to help
evaluate their potential contribution to overall lead exposure at a residence. It was also of
interest to characterize the national distribution of residential soil-lead levels and percentages of
the housing stock whose soil-lead levels exceed specified thresholds. Therefore, this chapter
provides additional information on lead exposure within the following sections:

       *     Section 3.1:  Information on the National Survey of Lead and Allergens in
             Housing (NSLAH), a national survey begun in 1997 of lead levels in dust and soil
             in U.S. residential housing.

       •     Section 3.2:  Comparison of the HUD National  Survey data summaries for dust-
             lead loading and soil-lead concentration with summaries from other lead exposure
             studies, including interim data (for 706 households) from the NSLAH and pre-

                                          51

-------
             intervention data from the HUD Grantees evaluation that have been revised and
             augmented since the §403 risk analysis report was published).

             Section 3.3: Information on the prevalence of soil pica tendencies in young
             children and how such tendencies may occur over and above paint pica
             tendencies.

             Section 3.4: Updated information on numbers of children in the nation's housing
             stock, using interim data from the NSLAH.

             Section 3.5: Distribution of dust-lead levels on surfaces other than uncarpeted
             floors and window sills.

             Section 3.6: Revised summaries of pre-intervention blood-lead concentration
             based on updated data from the HUD Grantees evaluation.
3.1   THE NATIONAL SURVEY OF LEAD AND ALLERGENS IN HOUSING

      The National Survey of Lead and Allergens in Housing (NSLAH) is a currently-ongoing
survey sponsored by the U.S. Department of Housing and Urban Development (HUD) and the
National Institute of Environmental Health Sciences (NIEHS) to assess the lead and allergen
burden in that portion of the regularly-occupied U.S. housing stock that can potentially include
young children among its residents. In particular, the survey is assessing lead burden by
characterizing levels of lead-contaminated dust, lead-based paint, and lead-contaminated soil in
housing and residential areas. HUD initiated this survey in 1997 and has been approved by the
Office of Management and Budget to collect information through April 2001 for up to 1000
housing units.

      The NSLAH provides a more recent nationally-representative characterization of
environmental-lead levels in the U.S. housing stock than the 1989-1990 HUD National Survey
and involves sampling in considerably more housing units. In addition, dust samples in the
NSLAH are collected using wipe techniques (i.e., the technique assumed in the §403 rule) rather
than the Blue Nozzle vacuum method used in the older survey, and the NSLAH did not restrict
the sampling frame to only housing built prior to 1980.  Therefore, the information collected in
the NSLAH is very important for the §403 risk analysis to consider. However, the survey's
scheduled completion date and the expected date for finalizing the survey's database do not fall
within the time frame necessary to complete the risk analysis. Therefore, in order to utilize data
from the NSLAH, the risk analysis could only consider data collected up to an interim point in
the survey.

      Interim NSLAH data for 706 housing units, collected from 1998-1999, were made
available to the §403 risk analysis in August, 1999. This is a preliminary subset of the survey's
final database that will represent an expected 825 housing units. To allow the data for these 706

                                          52

-------
units to be considered a nationally-representative characterization of lead levels in the housing
stock, the interim database included sampling weights assigned to each unit based on its set of
selection probabilities within each stage of the multi-staged sampling design and adjusted for
nonresponse. These are interim sampling weights as they were generated by only considering the
706 units represented in the interim database. As the final sampling weights to be assigned at the
end of the survey will reflect all housing units hi the survey, and as there is a potential for
additional correction of the existing data before the survey database is finalized, any analysis
results based on the interim database of 706 housing units will likely differ from those to be
based on the final database.

       Table 3-1 contains key design specifications and approaches of the NSLAH, such as the
types of rooms in which dust samples were collected and paint-lead levels were measured, the
approach to taking soil samples, and laboratory analytical methods. Also included for
comparison purposes in Table 3-1 are the design specifications and approaches taken in the older
HUD National Survey.  Note that in both surveys, dust samples were taken from the same types
of surfaces (floors, window sills, and window troughs,  also known as window wells) and
analyzed under similar methods, and soil sampling occurred in the same areas of the yard. The
method for analyzing soil samples was changed from ICP-AES in the older survey to FAA in the
NSLAH due to the need to reduce detection limits associated with the method. Specific focus
was made in the NSLAH to ensure that rooms in which children frequently reside are more
dominantly represented in the sampling design.

       Various types of data are being collected from housing units participating in the NSLAH.
Household questionnaire data are collected at two time points: at the initial contact with the
household during recruitment (to screen for eligibility and to perform an inventory on interior
rooms) and during  an interview with residents during the study (to obtain information on the
building, household, and residents). Allergen dust levels are measured by collecting and
analyzing vacuum dust samples. Lead levels in the unit are characterized through the following
types of measures:

       •       Dust samples: Dust-lead loadings (jig/ft2, assuming wipe collection techniques)
              for floors, window sills, and window troughs (also known as window wells)

       •       Soil samples: Soil-lead concentrations (ug/g) at entryway, dripline, and mid-yard

       •       Lead on painted surfaces: X-ray fluorescence (XRF) measurements (mg/cm2)

       To determine the numbers of housing units represented by the interim NSLAH sampling
weights within certain housing categories and how these numbers compare with estimates made
in the §403 risk analysis and by the U.S. Census Bureau, Tables 3-2 and 3-3 provide estimated
numbers of occupied housing within specified housing  age categories and the four Census
regions, respectively. These totals are presented based  on data from the NSLAH as well as from
the following additional surveys/analyses:
                                          53

-------
o
a
a.
r
n
£
A
a
S
           S
             a
         a 5 -o .c *i a CD
         o o 45 to a a .£
                 .
                2
ousing units selected from yi
he 50 states and the District
ldren were pi
ected 825
il
xp
   £ «
   0 i
   i-:
   S 3
   c S
   O _
   O O>
   i_ 0>
  ^£2
sample represents 67 of
(PSUs). Institutional and
         &£
         Ii
         !i
         o£
         z§
         W '^
         Si
        . -a "o
   -  to >  jc c
2
O
         »   a    s £
         €£|3i*5£
         -(D^eeCO)*
         E a «? § -S -| .c -o
         S8l<.fiaS-g
         goioCLCinox
         iO.CUJDr-~.Ca>
         t«-ii
        ^- E '•= = *
        1!FJ
        •c o tt o »
          ^ *- ~ —. >
          ,- > a « w
        ^ c "D " E

        I 1 g | |
        i Illll
          "S
        A .<
          •s C • OT
          ® s s
          »r5
                IO
          c s a 3 -5
M. _*   -. «} Jt
> 2 o > > .S
r~ JC «•- >  o
             a. «- »- a.
             •o
             c
               §
specified
number i
s
m
i
I:

Us
S: Q.O>
  P O>
O CO *~
£. "> c
& i IK
8«{
7.2 S
£•« T»
III
  E °
  fc o
   0
  c •
                        I
                           • ^
                        " ? S1
                        S | E
                        §
                        O)
                        i
Novembe
                        «- t o
                        o e c
     *= > E
     Q §5 S
                           54

-------
from the
loor dus
seed roo
ere collected
ect
                           I
                           s|
                           §^
                           7 c
                           =5 «
                           •o c
                           « «
1,
-O »
sis
°»  -S
i i?i
H*
85*
•Sgo
g = *
ffSS
«S =
s-g10

II!
• « a
*JS
Iff
2 e =
*§£

ISl
5"S =
_ CD u
"C O> O
* S *
r- *ft _C
JE "o t
E
ip
r t
                                                         (D
                                                         5
                                                         1
                                                         S
                                                         CD
                                                         a
                                                         o
                                                         2
                                                         0
                                                         (0
                                                         w CD
                                                         «*- ***
                                                         CD <=
                                                         •= o
                                                         a. a.
                                                         E o
                                                         c tr
                                                         O J5
o
o
       "S
       ID

       jj   CB
       2   E

       *   $

       3.   S

       S - « ?

       E o -o co
       s P .2 S
       "a. ^ S= a,
       0..2-8 i
             u o
         *J 
om,

om
bat
Drv
                    I

                    o
                    2

                    •g

                    1
                                                         o s

                                                         t §
                                                         « o

                                                         E?
                                                         i °-
                                                         


ni £
                               O *
                               ** O)
                               T3 C
                               0 =


                               « i
  1

15
a «
             o a
             "§
                                                     eg
                                                         03 P

                                                         88
                                                   3*. 9
                                                   Zoo.
                               55

-------
    1
      '

o
•g
5
o
•a
a
rough sampled
and
oom
Entire sil
selected
                f
         IE
         ll
                c o
                0 IS
                 *
               11
                 *
                s
                o
ti
S 3
lori
st
tal
Diges
STM E
ch
pe
d
A
A
ric
Meth
Envir
FA
or
nit
o
u
c?
          "8
       o
       o >
         5> TB **
         t's
_ a ^ o
m  £


"> § -n 52
           o
           s
         o S
         •fe "
         I i

         a 5
                51

       * i   * -
       •I 1 £ -§ 2
o .E

o o
•fc T,

if
E-o
SS
a> _v
O S
                     TS
               o
               a>
               TJ
                o >
0 c o >
ff> « r- £-
SS*i
                             a 3
                             •£ o
                             c €
                             q> ^
                             15
       "
111
c » a
S 3 Q.
> "O <0
                u a> o

                g^ a
                Ȥ s
                •? S'Q.
                 ^- ra
                Z O «9
              ,>
              ID
              C W
              ID 
-------
CO
_£
13
.•:

*^^-
l;
. o
:;I.!i
• £:

«-:!
|
1
'*':


'.•»•
-"§:•;
O
=
1
HUD National Survey of lead-Based




§'




„
- -I ?
_ _. 4^
S« -5 £ j
2 * ~Hi
O o o .£ _
<">«•= 4: E
0 | = 1 5>
*^ *£ > Q b_
i ? 1 | c
® s e « .2
co "s '1 *• S
> " IT o £
«> § S 00
o « o 55
CP-AES
digestion method: modification
VSTM ES 36-94 (hot-plate dige
HCI/H2Oj), or SW-846 Method :
ligestion)
Method must be that used in pr
invironmental Lead Laboratory ,
«iii> U "*. ^ U b UM!





i
.0

SO) ~
to
a>
•- °
-a a
If
• \ C
« 0
S io
la o
1 !
2 <6
c £
r ^
5 2
"he final results as reported by
he database (i.e., not-detected
vith detection limits.
r~ >* ^

TJ
0 f S
-C « ja
CD _
.£ £ |
IB II
a « e —
As log-transformed lead amounts are re
database, only positive measurements
No indication is given as to when data i
truncated due to being below detection
•o
0} >
^3 — M
1 -1
^ m —
1 S.I
« E s
c "2 £ 2
~ 2 3 C
•o* g-a
111*
ij. CO C -M





^^
id
1
CD
>
4>
£
ig

4^
•s
10
O)
1
O
CO
X
o
4M*
z
LL.
5
^
O
1&
_ g *-
C *f
3 ±! °
ll s
»".*>•
T> 3 "g
f* U O
MAP-3 XRF instrument (single 60-secoi
reading" measurement using a 40 millii
Measurements were adjusted to statist!
measurement bias.



CD
_C
io **
r g
fi-a |
E^ 5
f A ca
'S C CB
0> ~ ffi
5 S. E

•o
c
 >
*^ E
S o
o •-
§• M «
I | .2
o « a>
>- 0 <8
.2 Q. _
5 «, E > 'E
« I E ^ 1
1 s M 1
1 i "1 1
« | a, 1 J | c
^-o ^igcoES 9-
O  Q "C O C
**T3 ^C O UL S» Q 1— CL o
o> 3 Q
— "o =

i
I » > § 1
g o o | g -g
-i T. co -a a. c
5 2r c c te •=
o | 1 2 i £
I II! If
€ «- -o 
-------
n

£

r
o
i
JJ5
£
: ._C
..fj.
c
t.

<'
I

1
*a
I
• -3"
- ^g.-
1
S-
1
^^••-



-' C"
.1
o
X ,
•£
{
*
1
I
§
X












•o
c
03

"3
>
TJ
CD
i
m


1
8
o
1
3
—
s
o
^
in
tN
•s
+*
(ft


lull
3
0
••"•
O)
"i
ted surfaces were categorized into the follo\
.£
(0
a.
05
c
S

Q.
o
(O
2
0)
c
a.
E
o



















a
(75







Metal substrate within the selected wall

•


(0
4^
ID
£
W


i
o
t»
8|
§ |
M 5
w ^3
W T3
•b C
S> ID
52
i °
£ a






32
Non-metal substrate within the selected wa
Other surfaces within the selected wall

• •






























El
£Z





£
>•
painted components were selected random!
ID
^
IT



































-o
t5
^
•M
«di
1
8
^
0>
'
t*>
I

S














1
en
S
£
o
5
"5
S
"8
a
i
CO
o
o
E
a.
(0
a>
ompon
o
<"~


TJ
S
O
JD
"o
I
i
1
0
E
i
o
31
I



































O
4*
o
4^
t
§
2
3
(0
JC
2
2
4M
(O
"5
to
k.
***
"O
5


































0
1
Is
« TJ
osive measurements were taken from paint
le exterior of the unit that may be suspecte
Q. *:
•3 C
Q. 0






































T>
ID
'9
4-*
O
o











                                                          <0
                                                          an

                                                          t>
                                                          I
                                                          «J
                                                          o

                                                          I
                                                          I


                                                          I

                                                          w

                                                          5
                                                          r
                                                          a
                                                          a.
                                                          u

                                                          «
                                                          a



                                                          o
                                                          «^
                                                           58

-------
Table 3-2.     Estimated Number of Occupied Housing Units in the U.S. Housing Stock
                 Within Year-Built Categories, According to Four Recent Surveys and/or
                 Analyses
Year in Which the
Unit Was Built
Prior to 1 940
1940-1959
1960-1979
O 960-1 977 for NSLAH>
After 1979
(Aft«r 1977 for NSLAH)
Not specified5
Total
Number of Units in the National Housing Stock (and Percentage
of Total Units), as Estimated by the ...
1995
American
Housing
Survey1
19,308,000
(20%)
19,885,000
(20%)
35,300,000
(36%)
23,201,000
(24%)
-
97,693,000
§403 Risk
Analysis2
19,676,000
(20%)
19,718,000
(20%)
34,985,000
(35%)
24,893,000
(25%)
-
99,272,000
1997
American
Housing
Survey3
19,441,000
(20%)
19,797,000
(20%)
34,884,000
(35%)
25,367,000
(25%)
-
99,487,000
National
Survey of
Lead and
Allergens in
Housing
(NSLAH)4
14,412,000
(18%)
16,886,000
(21%)
25,688,000
(32%)
24,076,000
(30%)
8,089,000
89,151,000
# Units Surveyed
1989-90
HUD
National
Survey
77
87
120
-
-
284
NSLAH
(interim)
114
145
201
180
66
706
' Estimates represent only year-round occupied housing in the 1995 national housing stock and were obtained from
information within Table 1A-1 of 'American Housing Survey for the United States in 1995" (Current Housing Reports
H150/95RV, published by the Bureau of the Census and HUD's Office of Policy Development and Research). This national
survey was conducted on about 55,000 surveyed units from August 1995 through February 1996.  An updated report
reflecting the 1997 American Housing Survey data has not yet been published.

1 Estimates were obtained from Table 3-5 of the 5403 risk analysis report. Estimates are based on data from the 1989-90
HUD National Survey of Lead-Based Paint in Housing, augmented by other Census information in order to represent the 1997
housing stock (see the §403 risk analysis report for details).

* Estimates were obtained from Table 2-1 of U.S. Census  Bureau (1999). The estimates represent total year-round occupied
units in the 1997 national housing stock.

* This survey, conducted from 1998-1999, characterized only occupied housing in which a young child could reside.
Information in this table is based on an interim dataset for 706 surveyed housing units.  Year-built information was
determined from responses given in the survey's resident questionnaire. If no year-built information was available from the
resident questionnaire, any year-built information provided from the recruitment questionnaire {when available) was used.
Note differences in how year-built categories were defined in this survey.  The specified percentages are relative to the total
minus the number of units represented by surveyed  units with no year-built information specified (i.e., 89,151,000 -
8,089,000 - 81,062,000).

5 Total sampling weights for surveyed units where either no housing age information was provided, or responses of 'Don't
Know' or 'Not Ascertained*  were given, on both the resident and recruitment surveys.
                                                     59

-------
Table 3-3.    Estimated Number of Occupied Housing Units in the U.S. Housing Stock
                Within Each Census Region, According to Four Recent Surveys and/or
                Analyses
Census Region
Northeast
Midwest
South
West
Total
Number of Units in the National Housing Stock (and Percentage
of Total Units), as Estimated by the ...
1995
American
Housing
Survey1
19,200.000
(20%)
23,662,000
(24%)
34,236,000
(35%)
20,596,000
(21%)
97,693,000
§403 Risk
Analysis2
15,878,000
(16%)
22,313,000
(22%)
41,733,000
(42%)
19,348,000
(19%)
99,272,000
1997
American
Housing
Survey3
19,484,000
(20%)
23,951.000
(24%)
34,808.000
(35%)
21,245,000
(21%)
99,487,000
National
Survey of
Lead and
Allergens in
Housing
(NSLAH)4
14; 977,000
(17%)
22,202,000
(25%)
32,519,000
(36%)
19,453,000
(22%)
89,151,000
# Units Surveyed
1989-90
HUD
National
Survey
53
69
116
46
284
NSLAH
(interim)
109
150
265
182
706
1 Estimates represent only year-round occupied housing in the 1995 national housing stock and were obtained from
information within Table 1A-1 of "American Housing Survey for the United States in 1995" (Current Housing Reports
H150/95RV, published by the Bureau of the Census and HUD's Office of Policy Development and Research). This national
survey was conducted on about 55,000 surveyed units from August 1995 through February 1996. An updated report
reflecting the 1997 American Housing Survey data has not yet been published.

1 Estimates were obtained from Table 3-5 of the S4O3 risk analysis report. Estimates are based on data from the 1989-90
HUD National Survey of Lead-Based Paint in Housing, augmented by other Census information in order to represent the 1997
housing stock (see the §403 risk analysis report for details).

3 Estimates were obtained from Table 2-1 of U.S. Census Bureau (1999). The estimates represent total year-round occupied
units in the 1997 national housing stock.

4 This survey, conducted  from 1998-1999, characterized only occupied housing in which a young child could reside.
Information in this table is based on an interim dataset for 706 surveyed housing units.
                                                   60

-------
      •      the 1995 American Housing Survey (i.e., the last survey in which estimates of
             these totals were published hi documents issued by the Census Bureau)

      •      the §403 risk analysis (which characterized the 1997 housing stock by revising
             the sampling weights from the 1989-90 HUD National Survey)

      •      the 1997 American Housing Survey (based on information obtained from the
             HUD web-site)

As noted in these tables, the sum of the interim sampling weights in the NSLAH (89,151,000) is
over ten million units lower than the corresponding sums from the §403 risk analysis and the
1997 American Housing Survey. It is possible that this difference is due to the NSLAH's
exclusion of housing that forbids resident children (i.e., adult-only housing), while the §403 risk
analysis and the 1997 American Housing Survey results reflect the entire regularly-occupied
housing stock.

      When reviewing the sum of sampling weights by housing age category (Table 3-2), the
interim NSLAH data represent a slightly smaller percentage of pre-1940 housing compared to the
other surveys. The housing age categories for the NSLAH differ slightly from the categories in
which the other surveys are represented and what was used in the 1989-90 HUD National
Survey.  Approximately eight million housing units in the  U.S. housing stock are represented by
66 units in the interim NSLAH dataset that do not have a housing age specified.  The final two
columns of Table 3-2 present numbers of surveyed units by housing age category in both HUD
surveys.

      The percentages of housing units within Census regions (Table 3-3) are similar between
the interim NSLAH and the 1997 American Housing Survey except for the Northeast, where the
percentage in the interim NSLAH was lower than hi the 1997 American Housing Survey.
Differences relative to the 1997 American Housing Survey were even greater for the §403 risk
analysis, where the adjustments made to the sampling weights in the 1989-90 HUD National
Survey to represent the 1997 housing stock did not take into account Census region.

      Summaries of the interim NSLAH data and comparison to the 1989-90 HUD National
Survey data summaries (citedin the §403 risk analysis report) are provided in the next section.

3.2   COMPARISON OF ENVIRONMENTAL-LEAD LEVELS IN THE HUD NATIONAL
      SURVEY WITH THOSE OF OTHER KEY STUDIES

      As discussed in Sections 3.2 and 3.3 of the §403 risk analysis report, the risk analysis
used data from the HUD National Survey to represent baseline (pre-§403) environmental-lead
levels (paint, dust, soil) in the nation's housing stock. To help evaluate how accurate this
representation may be and how environmental-lead levels may have changed since the HUD
National Survey was conducted (1989-1990), the survey data were compared with data from
other environmental field studies that were conducted more recently and that measured

                                         61

-------
environmental-lead levels in a large number of housing units. This section also summarizes how
housing selection, sample collection techniques, laboratory testing practices, and the distribution
of environmental-lead levels reported in the HUD National Survey differ from those in these
other studies.

       The studies whose dust-lead and soil-lead data were used in the comparisons in this
section included the ongoing National Survey of Lead and Allergens in Housing (NSLAH,
introduced in Section 3.1 above), the Baltimore Repair & Maintenance (R&M) Study, the
Rochester Lead-in-Dust Study, and the various portions of the ongoing HUD Grantees evaluation
(design information and data for the latter three studies were summarized in Section 3.2.2 of the
§403 risk analysis report).  These studies were conducted since 1993 in locations within the
United States where a specific point source of lead was not necessarily present.  The latter three
studies provided the §403 risk analysis with the most useful and available data on the
relationship between environmental-lead levels (paint, dust, and soil) and childhood blood-lead
concentration. In particular, dust samples in these studies were collected from floors and
window sills using either a wipe technique or a method whose resulting dust-lead loadings could
be converted to wipe-equivalent loadings using methods  such as those documented in Section 4.3
of the §403 risk analysis report.  Data summaries for the HUD Grantees evaluation were updated
from the §403 risk analysis report summaries to reflect data collected through February, 1999.

       The risk associated with elevated soil-lead concentrations and intervention practices
designed to alleviate that risk are more frequently debated in the scientific literature than are the
risk from and the intervention practices targeting elevated dust-lead loadings. As a result, this
section supplements the comparison of the HUD National Survey's characterization of soil-lead
concentrations to the interim NSLAH and the aforementioned three recent studies with the results
of other relevant studies.

       Boxplots were used in this section to summarize household average dust-lead and  soil-
lead levels graphically. A boxplot, also known as a box-whisker plot, portrays the distribution
visually by using a box to represent data falling within the 25* and 75* percentiles and using
different graphical symbols for the remaining data values according to their distance from the
box. The following features are included within the boxplots presented in this section:

       •      A horizontal line within the box corresponds to the median.
       •      A dot within the box corresponds to the geometric mean.
       •      The bottom and top edges of the box correspond to the 25* and 75* percentiles,
              respectively.
       •      Central vertical lines ("whiskers") extend to 1.5 interquartile ranges (IQR, equal
              to the difference between the 75* and 25* percentiles on a log scale) of the box.
              However, if the data extend to less than 1.5 IQRs of the box, the whiskers extend
              only as far as the data exist.
       •      Open circles represent data values that exceed 1.5 IQRs but no more than 3 IQRs
              from the box.
       •      Asterisks represent data values that exceed 3.0 IQRs  from the box.

                                           62

-------
The boxplots were plotted on a logarithmic scale to improve the readability of the data
distributions, due to the tendency of the data to be skewed toward the lower end of these
distributions. Selected information portrayed within the boxplots have also been included within
tables of descriptive statistics presented throughout this section.

       Dust-lead loading data comparisons are provided in Section 3.2.1, while soil-lead
concentration data are addressed in Section 3.2.2.

3.2.1  Characterizing Dust-Lead  Loadings on Floors and Window Sills

       Household area-weighted average dust-lead loadings (assuming wipe techniques) as
calculated in the §403 risk analysis were the basis for the comparisons made in this section. This
average, calculated for each building component sampled for dust (i.e., floor, window sill),
represented a single dust-lead measure for the component within a housing unit and was
calculated by weighting each dust sample's result by the area that was sampled.

       While the household average dust-lead loadings assumed wipe collection techniques, the
dust collection device differed among the studies:

       •     HUD National Survey: Blue Nozzle vacuum
       •     Baltimore R&M studv: BUM vacuum
       •     NSLAH. Rochester study, and the HUD Grantees evaluation: wipes.

To obtain wipe-equivalent dust-lead loadings for samples taken in the HUD National Survey and
the Baltimore E&M study, the reported loadings were entered into the conversion equations
presented in Sections 4.3.1 (Blue Nozzle vacuum to wipe) and 4.3.2 (BRM vacuum to wipe) of
the §403 risk analysis report. Note that dust-lead loadings for samples collected by other
collection methods in these studies were not included in determining the area-weighted
averages.7

       In the §403 risk analysis, the household averages were calculated on wipe-equivalent
sample loadings associated with the 284 units in the HUD National  Survey, with imputed
averages assigned to those units having no available data (Section 3.3.1.1 of the §403 risk
analysis report).  When characterizing the distribution of these averages across units, the §403
risk analysis weighted each unit by its 1997 sample weight as calculated for the §403 risk
analysis (Appendix Cl of the §403 risk analysis report), and each unit built between 1960 and
1979 and without lead-based paint also represented post-1979 housing (Section 3.3.1.5 of the
§403 risk analysis report). The resulting data distribution was used  in the §403 risk analysis to
characterize the distribution of average dust-lead loadings in the nation's housing stock.
          The HUD National Survey database included a few wipe dust-lead loadings that were used as reported in
          determining household area-weighted averages.

                                           63

-------
       3.2.1.1.  Data Summaries for the §403 Risk Analysis Versus the Interim NSLAH.
Descriptive statistics of household average dust-lead loadings for floors and window sills as
calculated in the §403 risk analysis using the HUD National Survey data are presented in this
subsection as they compare with the same statistics calculated on interim data for 706 housing
units in the NSLAH. Note that these statistics reflect the sampling weights used in the §403 risk
analysis and the interim NSLAH sample weights, thereby allowing these summaries to be
nationally representative of the 1997 housing stock.

       The interim NSLAH summaries include imputed average dust-lead loading data values
which are assigned to households when no such data are available for a given surface (floors,
window sills).  Assigning an imputed dust-lead loading average to a household that has no dust-
lead loading data ensures that it (and its corresponding sampling weight representing a given
portion of the national housing stock) is represented in the risk analysis. The method used to
impute data closely follows the method used in the §403  risk analysis for housing units in the
HUD National Survey; this method is detailed in Appendix C. This appendix also gives the
imputed data values and how they were assigned to housing units. Summaries of the interim
NSLAH dust-lead loading data with imputed data excluded are found in Appendix Dl.

       When using the interim NSLAH data to calculate a household's average dust-lead loading
for floors or window sills, five different approaches were considered for handling individual
sample results that fell below the instrument's detection limit. These five approaches, which
include censoring the not-detected results, are presented in Appendix Dl. The data summaries
that exclude imputed data values, found in Appendix Dl, were performed and presented for each
of these five approaches. Of these five approaches, two were specifically identified as most
likely to be applied in the supplemental risk analysis involving the interim NSLAH data:

       •      making no adjustment to not-detected data values, and
       •      replacing not-detected data values with one-half of the detection limit.

The first approach eliminates potential bias that can be introduced when an adjustment is made to
a reported data value, but it also permits a household's average to be zero or below, preventing
the data from being used as input to the empirical model within the §403 risk analysis. (As the
survey's analytical method adjusted for potential analytical bias by subtracting a specified
amount from a given sample result, reported results of less than zero were possible. Such results
were included in the survey database used in this analysis.) The second approach prevents this
problem from occurring and represents the best estimate  of a sample's actual lead amount value
when the analytical result is only known to fall somewhere between zero and the instrument's
detection limit. Interim NSLAH data summaries under both approaches are presented in this
section to illustrate the impact that any one approach has on the characterized distribution.

National comparisons

       Tables  3-4 and 3-5 present descriptive statistics of average household dust-lead loadings
for floors and window sills, respectively, for the 1997 national housing stock. These summaries

                                          64

-------
Table 3-4.    Descriptive Statistics of Area-Weighted Average Floor Wipe Dust-Lead
               Loadings for Households, As Reported in the §403 Risk Analysis Versus the
               Interim NSLAH Data
Study
How, Not-
Detected
and
Negative
Data were
Handled
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced
by LOD/2
Vi": ;••:• - Area-Weighted Average floor Dust-Lead Loading (pg/ft2)1
** ' " •"•*
.#
Surveyed
Units with
Positive
Averages
284
633
706
Arith-
metic
Mean
16.5
10.4
10.8
Gee-
instnc
Mean2
6.27
1.22
1.82
Geo- '
rncrtnc "
Std. .-
Dev.*
3.49
4.57
2.78
Minimum
0.508
-1.23
0.750
25*
Pereen-
tfle
2.65
0.300
0.950
••truirarh
mBaum
5.32
1.05
1.31
75*
Peroen-
tile ,
12.2
2.30
2.46
Maximum
375
5940
5950
1 All statistics are calculated by weighting each household by its sampling weight.
2 Only household averages greater than zero are used to calculate this value (data for all units with floor dust-lead data are
used to calculate the remaining statistics).
3 Summaries include imputed data for households having no floor wipe dust-lead loading data. The method for imputation is
presented in Appendix C.
Table 3-5.     Descriptive Statistics of Area-Weighted Average Window Sill Wine Dust-
                Lead Loadings for Households, As Reported in the §403 Risk Analysis
                Versus the Interim NSLAH Data
Study
How Not-
Detected
and
Negative
Date were
Handled
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced
by LOD/2
Area-Weighted Average Window SRI Dust-Lead Loading (fig/ft2)1
•'•:# :
Surveyed
Units with
Positive
Averages
284
690
706
Arith-
roetic
Mean
550
137
137
Geo-
metric
Mean2
23.0
14.5
15.8
Geo-
metric:
Std.
Dev.2
15.8
7.83
6.57
Minimum
0.0118
-9.43
0.445
25*
. Per con-
tBe
4.35
2.90
3.35
Median
19.5
12.8
13.6
75*
PercGft-
tile
198
51.3
51.0
Msximum
43700
11100
moo
1 All statistics are calculated by weighting each household by its sampling weight.
2 Only household averages greater than zero are used to calculate this value (data for all units with window sill dust-lead
data are used to calculate the remaining statistics).
3 Summaries include imputed data for households having no window sill wipe dust-lead loading data. The method for
imputation is presented in Appendix C.
                                                65

-------
imply that the average dust-lead loadings for both floors and window sills based on the interim
NSLAH data are considerably lower than that reported in the §403 risk analysis (based on the
HUD National Survey after converting to wipe-equivalent loadings). For example, the median
floor dust-lead loading is less than 2 ug/ft2 based on the interim NSLAH data compared to 5.3
ug/ft2 from the §403 risk analysis, and the median window sill dust-lead loading is less than 12
ug/ft2 based on the interim NSLAH data compared to nearly 20 ng/ft2 from the §403 risk
analysis.

       Median detection limits for dust-lead loadings in the interim NSLAH were 1.5 pg/ft2 for
floors and 3.6 ug/ft2 for window sills. When considering all dust samples in the interim NSLAH
that had lead amounts reported, approximately two-thirds of the floor dust-lead samples and one-
third of the window sill dust-lead samples had results below the detection limit.

       Boxplots of the data distributions presented in Tables 3-4 and 3-5 are found in Figures
3-1 and 3-2, respectively. Appendix Dl contains these tabular summaries and boxplots after
excluding imputed data values.

       In addition to these data summaries that are based solely on the observed data and the
sampling weights, it was desired to characterize the national distribution of household average
floor dust-lead loading in such a way that the percentage of housing where this average exceeds a
specified threshold could be estimated. This was done for both the HUD National Survey and
interim NSLAH data by assuming that these data originate from a lognormal distribution. Then,
the fitted distributions and corresponding estimated exceedance percentages were compared
between the two surveys. These results are presented in Section 3.2.1.3 below.

Comparisons by housing age category

       While the summaries in Tables 3-4 and 3-5 represent the entire nation, Tables 3-6 and 3-7
present descriptive statistics according to the housing age category scheme defined in Table 3-2
above. Considerable declines in the geometric means and medians from the §403 risk analysis to
the interim NSLAH data were observed in all  four age categories.

       Boxplots of the data distributions presented in Tables 3-6 and 3-7 are found in Figures
3-3 and 3-4, respectively. Appendix Dl contains these tabular summaries and boxplots after
excluding imputed data values.

Comparisons bv Census region

       Tables 3-8 and 3-9 present descriptive statistics according to Census region.  Declines in
the geometric means and medians were observed from the §403 risk analysis to the interim
NSLAH data for all regions but the West region, where very slight increases in these estimates
were observed. The greatest declines were observed in the Northeast and Midwest.
                                          66

-------


























.












o 5 2 " °
2 *
(,»/W) Butpon pDTHsna JOOLJ

rf .**

SS £ a>

2 *o
«»§
*0
c
"~ 8
si
k_ (0
Jo
o|
g<»
f-D
3.2^
f!
! IS
8 •? «
^J JC
M *"*
5£
&"O
§1
O £^

l^i ^^
o S
2 =
0(0
"2 o
O 3
2|
*a> Q
^£ ^3
^t f
(0 '•^
£ ? S
2 «|
z«r ** *^- >
^ "S"* S
Q.^-t>
^^ fM CD
o c "S
CD < -0

^
CO
1
en
il
C <0
5 "8

S|
E §
3 N
" "§
•S 5
•§£
If
E«
|l
.= —
05 ••-
3 O
•§ c
t|
^1
1^
11
m (p
£ O
co
j= .
ii
•3 «
Sf.S
® '•§
.5- ra
0 =
|i
8 o>

g S
ffl
®2
2 o

67

-------
                   rlf
                         C 0
68

-------
Table 3-6.    Descriptive Statistics of Area-Weighted Average Floor Wipe Dust-Lead
               Loadings for Households, Presented bv Housing Aae Category. As Reported
               in the §403 Risk Analysis Versus the Interim NSLAH Data
Study
• • fjnu* •tjit
'. now N0i~
Detected
and
Negative
Data were
Handled
Area-Weighted Average Ffcx>r Dust-Lead Loading (^g/ft2)1
8
Surveyed
Units with
Positive
Averages
" : Arith- • '
rhetic
Mean
Gob-
ntotnc
Mean2
Gee-
iR&bnc.-
Std.
Dew.2
fwunmium
2S».^
Percen-
tite
Median
75*
Percon-
tile
Maximum
Units Bunt Prior to 194O
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
77
111
114
47.9
36.9
37.0
22.6
3.74
4.00
3.63
4.53
3.97
Units Built from 1940
5403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
87
134
145
18.1
4.11
4.38
8.74
1.90
2.31
3.34
3.57
2.64
0.991
-0.600
0.750
-1959
0.508
-0.720
0.750
8.84
1.30
1.45

4.07
0.719
1.05
17.7
2.42
2.71

7.81
1.80
1.99
79.7
9.50
9.50

22.4
4.00
4.00
375
5940
5950

171
71.0
71.0
Units Built from 1960-1977 (1960 - 1979 for the 5403 risk analysis)
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2

5403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
120
176
201
Units
28
151
180
6.74
1.50
1.96
Built After
4.16
1.20
1.71
4.14
0.912
1.46
2.45
3.47
1.92
0.657
-O.733
0.750
2.25
0.236
0.900
3.62
0.900
1.20
7.59
1.68
1.92
106
28.5
28.8
1977 (after 1979 for the §403 risk analysis)
3.14
0.545
1.14
2.06
3.35
1.72
1.06
-1.05
0.750
1.76
0.146
0.750
2.84
0.400
1.00
5.66
1.08
1.35
12.9
265
265
NSLAH Units with Unspecified Year-Built Indicator
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
61
66
31.7
32.1
1.37
2.20
6.64
3.92
-1.23
0.750
0.300
1.00
1.24
1.40
2.72
2.56
1040
1040
1 All statistics are calculated by weighting each household by its sampling weight.
2 Only household averages greater than zero are used to calculate this value (data for all units with floor dust-lead data are
used to calculate the remaining statistics).
3 Summaries include imputed data for households having no floor wipe dust-lead loading data. The method for imputation is
presented in Appendix C.
                                                69

-------
Table 3-7.    Descriptive Statistics of Area-Weighted Average Window Sill Wipe Dust-
               Lead Loadings for Households, Presented bv Housing Age Category. As
               Reported in the §403 Risk Analysis Versus the Interim NSLAH Data
Study
How Not-
Detected
and
Negative
Data were
Handled
Area-Weighted Average Window Sffl Dust-Lead Loading Uig/ft?)1 . . „ •',
. #
Surveyed
Units with
Positive
Averages
Arith-
metic
Mean
Geo-
•••-. metric
Mean2
Geo-
rnotnc
Std.
Dev.1
Mu*wnunt
25*
Percen-
tife
I Median
-7.5*"
Percen-
tite
MflXIRKRIt
Units Buift Prior to 1940
§403 Risk Analysis
(HUD Nat). Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
77
113
114
2060
400
400
168
77.5
76.8
16.7
6.59
6.44
Units Buift from 1940
1403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
87
144
145
285
129
129
22.0
24.5
26.1
10.7
6.80
5.97
0.0155
-0.152
1.03
• 1959
0.0118
-1.73
0.923
35.6
21.2
21.2

6.47
6.35
6.58
198
79.8
79.8

19.1
23.0
22.0
1220
294
294

107
88.6
88.6
43700
11100
11100

16100
3630
3630
Units Built from 1960-1977 (1960 - 1979 for the 5403 risk analysis)
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2

§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
120
195
201
Units
28
174
180
184
36.6
36.9
Built After
83.0
15.6
16.0
16.2
10.7
11.3
14.6
4.71
4.18
0.0164
-2.32
1.02
2.05
2.89
3.17
16.6
9.40
9.54
217
29.0
29.3
5790
1390
1390
1 977 (after 1 979 for the 5403 risk analysis)
8.17
3.56
4.57
9.94
5.27
3.79
0.0164
-9.43
0.445
2.58
0.916
1.72
8.11
3.19
3.67
57.8
10.3
9.99
1590
426
427
NSLAH Units with Unspecified Year-Built Indicator
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
64
66
367
367
39.8
40.2
7.32
6.72
-0.629
0.720
18.6
18.8
36.4
36.4
118
118
9030
9030
1  All statistics are calculated by weighting each household by its sampling weight.
2  Only household averages greater than zero are used to calculate this value (data for all units with window sill dust-lead
data are used to calculate the remaining statistics).
3  Summaries include imputed data for households having no window sill wipe dust-lead loading data.  The method for
imputation is presented in Appendix C.
                                                70

-------
.!•
                           -If

                           -25
                                ll
                                o<
                                o>cn

                                n
                                ||
                                £5
                                s*
                                (8 •-
       £o

      *• C (A
      o— £





      O J3 £
      CQ O «
            d>

            .s J

            la
            II
            o '5
            •82

              ?
              o
            = B
            ® «
            (0 52
            ll

            i|
            v "D

            1 1
      5 «    -S-o)
      oz    s i
      0)0    - o
      fta    *o c
      sj 5--S-  ^ -g
      3* ^M CO  ^— ••?

      sf|  II
      ^K "?2 CO  u O*
      £3"  =S
      «  "S  |1


      111  II
                                       Q!
if
2£
                                       JB "5
(,41/M) Sujpwn p»Hsno jooy
      «

      3
      m
Of

Is
           71

-------
  * * » »H-
   • •> * •M«Mii4«.
      »<-
•• •  * * «<4-
   (jW/W) Buipoff] pwrnsna IK wpujM
                                              O a.
                                              0)0
                                          *   v*~
                                       ~ M 5
                                      o
                                      o
                                              I!
                                              ?§
                                              ll
                                              3
                                                  3
                                              m< «
                                              3
                                              D>
«•• .53

(P ™
«J
II
                                                       <0 -t
                                                       3 O
                                                       a •£
                                                       if
                                                       •Jg-3
                                                       •§s
text
n the §
See
« 2


3 S
H
T3 £
5s
a> 43

ii

i
xa «
11
                                                       03
                                                       € 0-
z|

11


-------
Table 3-8.     Descriptive Statistics of Area-Weighted Average Floor Wipe Dust-Lead
                Loadings for Households, Presented by Census Region. As Reported in the
                §403 Risk Analysis Versus the Interim NSLAH Data
Study
How Not-
Detected
and ,
Ne$|alivo
Data were:
Handled
Area-Weighted Average
#
Surveyed
Units with
Positive
Averages
Arith-
metic
Mean
Geo-
metric
Mean2
Geo-
ftlfitfJC
Std.
Dev.1
Floor Dust-lie»i Loadina fiia/ft*)

MfftiiiuiiYi
25*
, P6fC6ft"
tfe
Median
t
75*
- Perceft- •
tite
Maximum
Northeast
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
53
103
109
35.6
10.0
10.3
14.9
2.28
2.90
3.95
4.42
3.15
0.632
-0.620
0.750
4.79
0.800
1.20
11.0
1.90
2.13
76.3
6.00
6.00
375
617
617
Midwest
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2

!403 Risk Analysis
{HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2

§4O3 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
73
136
150

134
235
265

52
159
182
14.7
14.7
15.0

13.3
2.65
3.07

9.81
18.7
19.1
6.32
1.34
2.04

5.01
0.981
1.55

4.97
0.949
1.46
3.26
5.81
3.39
South
3.28
3.94
2.25
West
2.75
3.66
2.31
0.508
-0.733
0.750

0.735
-1.05
0.750

1.06
-1.23
0.750
2.83
0.283
0.760

2.00
0.254
0.970

2.65
0.255
0.800
6.32
1.20
1.29

3.89
0.940
1.21

4.01
0.800
1.20
11.0
2.48
3.25

10.0
1.76
1.94

8.43
1.67
1.88
173
1040
1040

236
265
265

197
5940
5950
1  All statistics are calculated by weighting each household by its sampling weight.
2  Only household averages greater than zero are used to calculate this value (data for all units with floor dust-lead data are
used to calculate the remaining statistics).
3  Summaries include imputed data for households having no floor wipe dust-lead loading data. The method for imputation is
presented in Appendix C.
                                                73

-------
Table 3-9.    Descriptive Statistics of Area-Weighted Average Window Sill Wipe Dust-
               Lead Loadings for Households, Presented bv Census Region. As Reported in
               the §403 Risk Analysis Versus the Interim NSLAH Data
Study
How Not-
Detected
and
Negative
Data were
Handled

#
Surveyed
Units with
Positive
Averages


Arith-
metic
Mean
Geo-
metric
Mean2
Vverage Window Sffl Dust-Lead Loading (pg/fl?)? ,- -. .,-, •
Geo-
metric
Std.
Dev.*
Mimniurn
" -25*>
Percen-
t3e
Median.
"Si ,•* '
twv
.POfCOfl-
tite
Maximum
Northeast
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
53
107
109
1440
172
172
92.2
21.5
22.6
16.1
8.01
7.06
0.0155
-1.89
0.578
15.3
5.94
5.94
173
16.0
16.0
335
89.5
90.0
14600
5530
5530
Midwest
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3

No
adjustment
Replaced by
LOD/2

§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2

§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
73
145
150

134
259
265

52
179
182
564
218
218

432
115
116

62.2
54.3
54.4
48.5
21.0
21.6

19.6
13.8
15.6

4.45
7.73
8.72
13.2
7.25
6.49
South
12.4
8.11
6.42
West
12.7
6.65
5.59
0.0706
-2.32
1.12

0.118
-9.43
0.646

0.0118
-0.115
0.445
7.76
4.00
4.75

4.60
2.88
3.06

1.68
2.07
2.30
83.0
16.6
16.4

15.0
12.8
13.9

5.40
7.54
7.76
309
60.1
60.1

127
53.8
53.8

28.0
29.0
29.3
43700
9630
9630

284OO
11100
11100

14OO
3630
3630
1  All statistics are calculated by weighting each household by its sampling weight.
*  Only household averages greater than zero are used to calculate this value (data for alt units with window sill dust-lead
data are used to calculate the remaining statistics).
3  Summaries include imputed data for households having no window sill wipe dust-lead loading data. The method for
imputation is presented in Appendix C.
                                                74

-------
       Boxplots of the data distributions presented in Tables 3-8 and 3-9 arc found in Figures
3-5 and 3-6, respectively. Appendix Dl contains these tabular summaries and boxplots after
excluding imputed data values.

Comparisons bv combination of housing age and Census region

       Tables 3-10a and 3-10b present descriptive statistics for household average floor dust-
lead loadings according to the 16 combinations of Census region and housing age category.
Table 3-10a considers no adjustment to the interim NSLAH data when not-detected results were
observed, while Table 3-10b summarizes data where not-detected data were replaced by one-half
of the detection limit.  Tables 3-1 la and 3-1 Ib present the same descriptive statistics for
household average window sill dust-lead loadings. As the central tendency of the dust-lead
loading data was of primary interest to compare across the different combinations, these tables
only contain estimates of the arithmetic and geometric means, geometric standard deviation
(GSD), and median. Appendix Dl contains these tabular summaries after excluding imputed
data values.

       Due to the small number of housing units within certain combinations, caution is
warranted when making inferences based on the numbers in these tables.

       3.2.1.2. Data Summaries for the §403 Risk Analysis Versus Three Other Studies.
This subsection provides descriptive statistics of household average dust-lead loadings for floors
and window sills for the HUD National Survey (both as collected and as used in the §403 risk
analysis), comparing these summaries to those for the three studies identified in the introduction
to this section that provided the most useful and available information to the §403 risk analysis
on the relationship between environmental-lead levels and childhood blood-lead concentration:
the Baltimore R&M study, the Rochester Lead-in-Dust study, and the ongoing HUD Grantees
evaluation (data collected through February 1999).

       Summaries of the reported dust-lead loadings in the HUD National Survey and the
Baltimore R&M study were performed on wipe-equivalent dust-lead loadings using conversion
methods presented in the §403 risk analysis report. In addition, the household averages based on
HUD National Survey data were summarized in two different ways:  by ignoring the sample
weights assigned to the surveyed housing units and any imputed data for households with
missing data, and by handling the data as used in the §403 risk analysis (described earlier in this
section).

       Because the HUD Grantees program emphasizes local control of the individual programs,
each grantee participating in the HUD Grantee evaluation is responsible for designing and
implementing lead-hazard reduction approaches applicable to its specific needs and objectives.
These responsibilities include the recruitment methods, enrollment criteria, and intervention
strategies. However, to enable comparison of results from the various approaches, grantees
participating in the evaluation follow the same sampling protocols and use standard data
collection forms developed specifically for this evaluation. Table 3-4 of the §403 risk analysis

                                          75

-------
  .
                              o 2
                              <£
                               .«
                              §2
                              cc §
                              Ii
                              13
                              -
                              1>£
                              5®
                              (0 *§•*
                              JE ~
                              ii

                              if
                               >t
                              O 3
                              1;
                              i§
                              oz
                              U.Q
                              •OS- 3



                              !|i
                              $3 J2"S
                              tf SS i
                              111
                                    •gs
o c
ii
•o -c
(A O
•O £
•|^

1 "

ll
51



ll
^ §
* 2
IS
*3 GJ
II
t*
« 2

I
l|
•? <"
i •-

If
E &
CO .

ft
if
(jH/Bri) 6ujpo
-------
    » •  ••
                                               •


^C4
-^ ^^
*- ffi
s-s
° c
?3
55
_J
«0>
cz
(5E
^1

ff^~B
to
4M>
^B
^^* C
&i
£ §
"S-s*
S|
JS
•o >,
CO m
3!

«?s
II!
tto co
S,|
®« al
I<1
Isi
^ȣ?
ISi
5 "* c
i CO) (B
S S-c
% £ O
< ** *5
*- C «
Boxplots a
Observed 1
approache


•
f &
2
ID
f_ -
Si —
C <8
1 0
3 .a
8 o
tJ -C
•85
o •*"
.C D)
"SS C
II
11
»i
.£ —
'5 •«:
3 O
03 «
t"I
•"5
t
*?
'= 0
8^
s^
jS
£co
c .
«g
Is
(Q Q
3 S
?s
i's
.8- o>
sg
SB
"S s
« 3
(D A
s s
It
Is
Ji <8
02
S5
is
il
<»1
It
•s £
««
"5
/"> 3
li

-------
Table 3-1 Oa.  Descriptive Statistics of Area-Weighted Average Floor Wipe Dust-Lead
              Loadings for Households, Presented by Housing Age and Census Region. As
              Reported in the §403 Risk Analysis Versus the Interim MSLAH Data Where
              No Adjustments Were Made to Not-Detected Results
Census
Region
Northeast
Midwest
South




West


Study
$403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
5403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Housing Age
Category
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
Prior to 1940

1940-1959

1960-1977
11 960-79 for §403)
After 1977
( 1979 for §403)
Prior to 1940
1940- 1959
1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1940
1940-1959
1960-1977
{1960-79 for §403)
After 1 977
(1979 for §403)
' Area-Weighted Average Ftoor Oust
#
Surveyed
'Units
26
41
17
21
10
19
15
19
33
21
35
29
32
4
25
19
26
33
42
64
71
18
72
13
11
16
36
17
54
6
39
Arithmetic
.VMean ,
63.5
23.7
13.2
3.75
7.00
3.34
1.12
31.3
8.49
15.8
5.48
6.33
1.52
3.32
. 0.913
50.7
11.0
25.4
3.66
8.06
1.16
4.19
1.04
34.9
264
14.6
2.86
4.50
1.16
4.60
1.75
-Geometric
Mean
36.5
5.02
8.84
2.37
4.73
1.72
0.714
14.7
2.62
6.69
2.05
4.58
0.737
2.77
0.545
20.8
3.66
10.3
1.63
4.13
0.825
3.16
0.549
16.2
3.84
9.04
1.70
3.53
0.949
3.36
0.454
-Lead Loadina

Geometric
Std.Dev.
3.39
4.31
2.54
3.36
2.23
3.76
2.78
3.01
4.47
3.95
4.16
2.35
4.77
1.83
3.86
4.01
3.93
3.91
3.40
2.74
3.04
2.05
3.12
3.51
6.17
2.46
2.92
2.03
2.42
2.21
3.67
(ml*?!
Median
76.3
4.20
7.81
2.38
4.76
1.46
0.867
8.94
2.16
5.79
1.59
4.44
1.12
. 2.80
0.320
19.0
2.74
10.0
1.77
3.39
0.880
2.84
0.480
17.2
2.30
7.47
1.36
3.35
0.990
3.00
0.270
Note: Summaries include imputed data for households having no floor wipe dust-lead loading data. The method for
imputation is presented in Appendix C.
                                           78

-------
Table 3-1 Ob.  Descriptive Statistics of Area-Weighted Average Floor Wipe Dust-Lead
              Loadings for Households, Presented by Housing Age and Census Region. As
              Reported in the §403 Risk Analysis Versus the Interim NSLAH Data Where
              Not-Detected Results Were Replaced bv LOD/2
Census
i?9*011
Northeast
Midwest
South
West
Study ."
.'<&
it
§403 Risk Anal.
Interim NSLAH
§403 Risk Ana).
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
5403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
$403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Ana).
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Housing Age
Category
Prior to 1940

1940-1959

1960-1977
|196O-79for §403)
After 1977
Prior to 1940

1940- 1959

1960-1977
(1960-79 for §403)
After 1977
<1 979 for §403)
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
' Area-Weighted Average Floor Dust-Lead Loading big/ft2) - -
#-,
Surveyadl
Units
26
41
17
23
10
21
16
19
36
21
36
29
37
4
30
19
26
33
48
64
81
18
84
13
11
16
38
17
62
6
SO
- j *»*
Anttvnetic
., Mean
63.5
23.8
13.2
4.03
7.00
3.58
1.68
31.3
8.79
15.8
5.80
6.33
2.00
3.32
1.31
50.7
11.1
25.4
3.94
8.06
1.67
4.19.
1.54
34.9
264
14.6
3.07
4.50
1.62
4.60
2.34
Geometric -
Mean
36.5
5.47
8.84
2.86
4.73
2.16
1.43
14.7
2.88
6.69
2.57
4.58
1.50
2.77
1.09
20.8
3.87
10.3
1.99
4.13
1.31
3.16
1.13
16.2
4.03
9.04
1.99
3.53
1.4O
3.36
1.07
- ^ A&M«UV*M«*
"laooiiifluic
Std. Dev.
3.39
3.91
2.54
2.23
2.23
2.60
1.72
3.01
3.41
3.95
3.20
2.35
2.03
1.83
1.67
4.01
3.76
3.91
2.35
2.74
1.73
2.05
1.57
3.51
5.91
2.46
2.34
2.03
1.6S
2.21
1.95
Median
76.3
4.35
7.81
2.40
4.76
1.68
1.29
8.94
2.19
5.79
1.53
4.44
1.20
2.80
0.938
19.0
2.70
10.0
1.54
3.39
1.18
2.84
1.06
17.2
2.19
7.47
1.52
3.35
1.38
3.00
0.900
Note: Summaries include imputed data for households having no floor wipe dust-lead loading data. The method for
imputation is presented in Appendix C.
                                          79

-------
Table 3-11 a.  Descriptive Statistics of Area-Weighted Average Window Sill Wipe Dust-
              Lead Loadings for Households, Presented bv Housing Age and Census
              Region. As Reported in the §403 Risk Analysis Versus the Interim NSLAH
              Data Where No Adjustments Were Made to Not-Detected Results
Census
Region .
Northeast
Midwest
South
West
Study
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
1403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
. Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
interim NSLAH
Housing Age
Category
Prior to 1940

1940-1959
1960-1977
(1960-79 for §403)
After 1977
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1940

1940- 1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1940

1940- 1959

1960-1977
(1960-79 for §403)
After 1 977
(1979 for §403)
Area-Weighted Average Window SHI Dust-Lead Loading (fig/ft2)
#--
Surveyed
Units
26
40
17
23
10
20
16
19
36
21
35
29
33
4
30
19
26
33
48
64
80
18
80
13
11
16
38
17
62
6
48
Arithmetic
'Mean .
2700
396
98.5
62.7
499
13.9
18.3
1660
361
98.2
103
223
27.9
62.5
21.0
2450
600
657
160
149
55.4
112
18.2
125
47.6
107
186
58.7
26.1
9.66
5.64
Geometric
Mean
265
99.4
32.6
20.1
38.9
7.88
3.28
435
72.5
17.7
20.0
20.9
9.94
27.5
6.57
64.0
112
38.9
30.7
24.0
14.3
9.09
3.93
11.5
14.2
7.35
29.0
3.83
8.34
2.65
1.99
Geometric
Strf. Dev.
15.8
6.33
5.55
4.31
20.8
2.67
5.69
5.79
6.15
11.6
6.33
11.6
4.75
6.78
3.64
23.1
5.87
9.93
8.58
12.6
5.44
8.60
6.00
14.7
5.17
13.2
7.21
11.5
4.19
11.6
4.08
Median
176
91.7
50.7
18.5
217
6.49
2.06
542
67.3
17.4
17.1
48.3
9.54
83.0
5.86
24.4
115
26.2
32.0
32.0
15.4
7.58
3.89
7.05
17.1
6.96
33.8
4.35
7.51
5.94
1.63
Note: Summaries include imputed data for households having no window sill wipe dust-lead loading data. The method for
imputation is presented in Appendix C.
                                          80

-------
Table 3-11b.  Descriptive Statistics of Area-Weighted Average Window Sill Wine Dust-
              Lead Loadings for Households, Presented bv Housing Age and Census
              Region, As Reported in the §403 Risk Analysis Versus the Interim NSLAH
              Data Where Not-Detected Results Were Replaced bv LOD/2
Census
Region
Northeast
Midwest
South
West
Study
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anat.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§4O3 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
,' ' Housing Age
Category
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
.After 1977
(1979 for §403)
Prior to 1940

1940- 1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1940

1940- 1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Area-Weighted Average Window SB Dust-lead Loading (pgfft2)
#
'Surveyed
Units
26
41
17
23
10
21
16
19
36
21
36
29
37
4
30
19
26
33
48
64
81
18
84
13
11
16
38
17
62
6
50
Arithmetic
Mean
2700
396
98.5
62.7
499
14.7
18.6
1660
361
98.2
103
223
28.4
62.5
21.4
2450
600
657
160
149
55.7
112
18.8
125
47.8
107
186
58.7
26.0
9.66
5.77
Geometric
••___
mean
265
90.1
32.6
19.6
38.9
8.39
4.80
435
75.7
17.7
20.9
20.9
10.3
27.5
7.01
64.0
112
38.9
35.5
24.0
15.3
9.09
5.21
11.5
15.9
7.35
30.6
3.83
8.77
2.65
2.57
Geometric
Std. bev.
15.8
6.91
5.55
4.49
20.8
2.55
3.80
5.79
5.65
11.6
5.49
11.6
3.81
6.78
3.54
23.1
5.86
9.93
6.78
12.6
4.88
8.60
3.86
14.7
4.23
13.2
6.51
11.5
3.88
11.6
3.14
M— -K— —
, median
176
91.7
50.7
18.9
217 •
7.37
3.73
542
67.3
17.4
17.6
48.3
9.54
83.0
6.20
24.4
115
26.2
32.0
32.0
15.8
7.58
4.00
7.05
17.2
6.96
33.8
4.35
7.51
5.94
1.85
Note: Summaries include imputed data for households having no window sill wipe dust-lead loading data. The method for
imputation is presented in Appendix C.
                                           81

-------
report documented the differences between grantees in their enrollment/recruitment criteria.  As a
result, the HUD Grantees data summaries in this subsection are presented by grantee.

Overall data summaries

       Figures 3-7 and 3-8 present boxplots of the area-weighted household average dust-lead
loadings for floors and window sills, respectively. Each of these two figures contains a boxplot
for each study, along with separate boxplots for each grantee in the HUD Grantees evaluation8.
Each figure also includes three boxplots associated with the HUD National Survey data:

       •      "HUDNS (U)" summarizes the data without regard to sampling weights

       •      "HUDNS (403)" summarizes the data as used in the §403 risk analysis (e.g., using
              sampling weights reflecting the 1997 housing stock; incorporating imputed data
              assigned to housing units with missing data)

       •      "HUDNS (OW)" summarizes the data weighted according to the original weights
              assigned in the survey.

       Tables 3-12 and 3-13 present values of the statistics presented in the boxplots (geometric
mean, minimum, median, maximum, 25* and 75th percentiles), along with other important
information not explicitly observable from the boxplots (number of houses whose data enter into
these statistics, geometric standard deviation) that is necessary when comparing distributions
across studies.  The GSD reported for the overall HUD Grantees evaluation is the exponentiation
of the square root of the weighted average of log-transformed variances for the different grantees,
where the weights correspond to the numbers of units with data.

Comparisons bv housing age category

       Figures 3-9 and 3-10 contain boxplots on pre-1980 housing data (floors and window sills,
respectively) from the HUD National Survey, Baltimore R&M, and Rochester studies, and pre-
1978 data from the HUD Grantees evaluation (data combined across grantees) according to three
housing age categories (pre-1940,1940-1959,1960-1977/79). As in the overall summaries
above, the HUD National Survey data are presented within three boxplots for each age category.
Caution is warranted when interpreting results in these figures for the Rochester study, as the
actual age of certain bouses may be older than what was specified in the Rochester study
database (see Section 3.3.1.3 of the §403 risk analysis report). Also for this reason, and since the
other studies surveyed few, if any, post-1979 homes, boxplots were not created for homes built
after 1979. Boxplots for non-control houses in the Baltimore R&M study, all of which were built
prior to 1941, are also included in these figures and are displayed in the "pre-1940" category.
        "Alam"=AIameda County; "Balt"=Baltimore; "Bos"=Boston; "CA"=Califomia; "Cle"=CleveIand;
"MA"=Massachusetts; "MN"=Minnesota; "NJ"=New Jersey; "Rl"=Rhode Island; "WI"=Wisconsin; "Milw"=MHwaukee;
"Chic"=Chicago; "NYC"=New York City; "VT'=Vennont.

                                           82

-------
                     - S
                     hi



                       I



                       s



                     - e
                        e
                        o
                       5§
                      -!J
                      - I?
                           0

                           £
                            o£

                           z c
                              01
|0


Is

V 5*
fOTJ
O 3
£

s
                            o
^ °


ii
£ CO


1 fl?*=


« 3 "5
£(03

<« §

*5§ui
                            1
                            O)
0>


u>


O)



I


1

Q.



O
I
T3_^



ll



ii
                                     •c

                                     "3
                                    o 2
                                    IS
                                    — .2

                                    m'c
                                    •or
                                    ||


                                    E i
                                    O5C
     O   §t
         « c

         -1^

         to 2


         O 
-------











• — <= — •







(jW/W) 6qp«n pwn-jsno ms wpuiM
.. j

* £"2
-1 **
1 ll
il
« —
. B fcj ®
J ||
j =|
"'* It
"I !!
- il
ii
.1 J, ±£
1 Is
1 £2*0
~ 1 O> 
li.
«
•o
o
1
1
1
1
1
i
I
•o ^_
> s
§1
§.§
1 £
e R&M study
>1 labels along
o *r
11
00 C
11
>> O

12
Z Q_
Q £
J.|
II
s-s
w^
(Note; Dust-lead lo
documented in the
84

-------
Table 3-12.   Descriptive Statistics of Area-Weighted Average Pre-lntervention Floor Wipe
                Dust-Lead Loadings for Households, As Reported in the §403 Risk Analysis,
                the HUD National Survey, and Other Studies
Study
HUD
National
Survey
Approach/
Grantee
unweighted1
orig. weights2
S403RA3
Rochester Lead-in-Dust
Baltimore R&M*
HUD
Grantees
Alameda Co.
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
New Jersey
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
All Grantees
Area-Weighted Average Pro-Intervention ROOT Dust-Lead Loading (figlit*}
# Units
• with
Data
281
281
284
205
90
168
402
114
90
190
229
212
45
203
236
291
158
399
354
3091
Arith-
metic
Mean
21.0
21.0
16.5
110
54.3
127
642
205
130
232
408
202
308
530
172
247
234
462
515
366
Geo-
metnc
Mean
8.19
7.97
6.27
17.7
40.9
31.4
149
61.3
24.6
39.4
64.4
27.3
68.2
60.6
24.8
32.2
29.4
52.6
67.9
50.1
Geo-
ntotnc-
Std. Dev.
3.66
3.70
3.49
3.20
2.27
5.78
5.48
4.79
4.89
6.51
6.47
5.14
6.71
5.85
6.92
4.61
4.40
5.90
6.70
5.76
MinnnuiTk
0.508
0.508
0.508
1.21
4.48
0.250
0.250
1.00
2.75
1.00
0.521
0.333
1.75
0.250
0.400
1.50
0.200
0.0880
0.750
0.0880
25*
Percen-
tfle
3.23
3.17
2.65
10.4
29.1
8.59
53.2
18.8
7.95
10.3
17.0
10.9
10.5
17.7
5.99
11.0
11.5
18.5
15.8
14.3
Median
7.27
6.94
5.32
16.1
45.2
31.0
167
55.3
15.6
36.4
59.8
19.2
93.4
54.0
16.9
27.5
28.2
32.9
49.9
40.2
73*
Percen-
t3e
17.3
17.0
12.2
26.6
70.4
98.0
456
170
59.3
134
234
62.4
298
187
79.1
76.3
69.2
94.4
219
165
Maximum
375
375
375
8660
266
3730
89100
2490
2650
8800
16600
13800
4250
59200
2780
31900
26400
22200
15600
89100
1 Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized without weighting by sample weights.
* Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized by weighting each average by the original sample weights assigned in the survey.
3 Area-weighted average dust-lead loadings, as calculated in Chapter 3 of the §403 risk analysis, are summarized by
weighting each average to reflect the 1997 U.S. housing stock and imputing averages for units with missing data.
4 BRM dust-lead loadings are converted to wipe-equivalent loadings prior to summary in this table.
                                                  85

-------
Table 3-13.   Descriptive Statistics of Area-Weighted Average Pre-lntervention Window
                Sill Wipe Dust-Lead Loadings for Households, As Reported in the  §403 Risk
                Analysis, the HUD National Survey, and Other Studies

Study
HUD
National
Survey
Approach/
Grantee
unweighted1
orig. weights2
§403 RA3
Rochester Lead-in-Dust
Baltimore R&M*
HUD
Grantees
Alameda Co.
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
New Jersey
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
All Grantees
. Area-Weighted Average- Pre^irtenrention Window SB Dust-Lead loading 0/g/ft1)
# Units
with
Data
245
245
312
196
90
178
402
95
81
185
206
193
51
192
234
271
146
382
318
2934
Aritn-
VttOttC
Mean
678
721
550
558
627
677
669O
4090
909
4050
2990
3160
758
4930
2790
3520
16OO
1580
374O
3360
~ Geo-
metric
"*' Mean '
21.7
24.9
23.0
196
356
118
1560
452
316
259
425
308
93.7
659
279
536
260
267
246
380
Geo-
IIKBtflC
Strf.Dev.
15.4
17.9
15.8
3.96
3.55
9.14
7.39
9.87
4.60
16.2
7.13
6.17
27.8
11.9
8.44
6.89
5.71
5.58
14.6
8.68
Mmimum
0.0118
0.0118
0.0118
2.83
20.6
0.0016
 .
Maximum
43700
43700
43700
14900
2330
19700
220000
106000
9630
241000
76100
300000
8450
132000
142000
88000
50500
57100
98100
30OOOO
1 Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized without weighting by sample weights.
2 Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized by weighting each average by the original sample weights assigned in the survey.
3 Area-weighted average dust-lead loadings, as calculated in Chapter 3 of the §403 risk analysis, are summarized by
weighting each average to reflect the 1997 U.S. housing stock and imputing averages for units with missing data.
4 BRM dust-lead loadings are converted to wipe-equivalent loadings prior to summary in this table.
                                                  86

-------
                  0  ODCOO—C3ID	1
   0    OOO
                 OD  oj


                                                  if
                                                  •i" >
                                                  30
                                                  CO O
                                                  fl

                                                  ll
5*
si
                                                  d>
                                                  jjjj
                                                    o
                                                    o

JP


|

O)

                                                          o>.E
                                                         S £
                                                         c «
                                                         » o
        o i


        "§•'
        1|


        E«

        gi
        ®.ffi

        i«
        C N
        a> -c



        o o>
        = «
        « m

        51

                                                          C 0)

                                                          3 *
        — tf
        ta o
        z g.


        S«
        z§.
        • 15 ' •

"5§|   |-
3^5

• • £ 
-------

CO











nn o n ana — i 	 re 	 — i 	 1
j I I ! i • - i i

r «a? tar •— * *jsa-
id Loadings (ug/ft2) for
ust Study, and HUD Grantees
P
3 *§
0 fl>
J 11
L il
1 PS:
_ || £ 5 "E
ll-l
. O) o" O»
. |L 2 E 5
"f ill
. 1 o z o
* S> IE *?
*& (o 3
X 3 «
. IsJ 0 0 >
p CQXUJ
o
O)
£
0>
•sf
o ..
(Note: Dust-lead loadings from the HUD National Survey and Baltimore R&M study have been converted from vacuum to wipe-equivalents using the math
documented in the §403 Risk Analysis report. See text for definitions of labels along the horizontal axis. Caution must be taken when categorizing houses
Rochester study by age of house.)
88

-------
       Values of the statistics entering into the boxplots in Figures 3-9 and 3-10 are included
within Tables 3-14 and 3-15. While not included in the figures, these tables include summary
statistics for homes labeled as post-1979 (although the Rochester study units may not have
actually been built in this time period, as mentioned in the previous paragraph). The post-1979
results labeled as "HUD National Survey (§403 RA)" represent surveyed homes built from 1960-
1979 that contain no lead-based paint (Section 3.3.1.5 of the §403 risk analysis report).

       3.2.1.3  Calculating National Exceedance Percentages for Household Average Floor
Dust-Lead Loading.  With respect to the national summaries of household average floor dust-
lead loading presented in Section 3.2.1.1 above, it was desired to estimate the percentage of
housing with average floor dust-lead loadings at or above specified thresholds (i.e., "exceedance
percentage"), with separate estimates originating from data for each of the two national surveys
(i.e., HUD National Survey and the interim NSLAH). This was done by fitting a lognormal
distribution to the household average floor dust-lead loadings summarized in Section 3.2.1.1  and
calculating the exceedance percentages based on this distribution.9 If the household averages
from the two surveys could each be considered a sample from their respective fitted lognormal
distributions, with the probability of selection for the sample determined by the sampling
weights, then the estimates based on these fitted distributions would be considered representative
of actual percentages for the nation. The fitted lognormal distributions and the resulting
exceedance percentage estimates are now presented for both surveys.

       For both surveys, normal probability plots prepared on the log-transformed average floor
dust-lead loadings indicated that a lognormal distribution did not adequately represent data in the
upper tails of the distribution (i.e., typically the upper quartile).  This was because the fitted
distribution was heavily influenced by the considerable amount of data at the lower end of the
distribution. Because  it was necessary in this exercise to characterize the upper tail of the
distribution as accurately as possible (due to calculating exceedance percentages from the
distribution), the actual values of the data at the lower end of the distribution did not need to
influence the fitted distribution to the extent that they were. Under these considerations, the
procedure to fit a lognormal distribution was as follows:

       •      For values of P from 5 to 50 (in multiples of 5), the value of the log-transformed
              average floor dust-lead loading (call this value X) was identified for which P% of
              the (weighted) data fell below.

       •      For each value of P, log-transformed data values falling below the value X were
              considered to be censored at X. That is, rather than using these actual log-
              transformed data values, the procedure assumed that each of these values was
              somewhere at or below X.
         For the interim NSLAH, household averages calculated from data where no adjustment was made when below
detection limits were used in this exercise.

                                           89

-------
Table 3-14.  Descriptive Statistics of Area-Weighted Average Pre-lntervention Floor Wipe
             Dust-Lead Loadings for Households, Presented by Housing Age Category.
             As Reported in the §403 Risk Analysis, the HUD National Survey, and Other
             Studies

Study

Approach/
Grantee
A
# Units
with
Data


Arith-
metic
f nffOOfl
Geo-
in0tnc
"Mo&n
M Average Pro-Intervention Floor Dust-Lead Loading (pgftt*) .-
Geo-
metric
Std. Dev.
:minununi
25*
Percen-
tie
Median
75*
Percen-
tife
Maximum
Houses Built Prior to 1940
HUD
National
Survey
unweighted*
orig. weights2
§403 RA*
Baltimore R&M4
Rochester Lead-in-Dust
HUD
Grantees
Alameda Co.
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
New Jersey
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
All Grantees
76
76
77
74
172
138
345
71
35
173
146
182
26
123
214
262
144
375
288
2522
43.9
47.9
47.9
63.6
127
118
672
222
269
209
147
171
511
197
183
254
60.7
470
478
328
19.5
22.4
22.6
55.5
19.8
30.6
153
55.4
48.4
34.7
31.7
21.3
215
44.2
28.4
30.7
25.6
50.0
63.7
45.9
3.68
3.65
3.63
1.65
3.18
5.40
5.10
5.11
6.75
6.30
5.03
4.72
4.19
4.68
6.77
4.42
3.92
5.93
6.81
5.45
0.991
0.991
0.991
22.0
1.66
0.250
1.00
1.00
2.75
1.00
0.521
0.333
10.5
2.00
0.400
1.50
0.200
0.0880
0.750
0.0880
8.45
8.84
8.84
38.7
11.3
10.3
54.6
16.0
8.38
9.50
11.9
10.0
134
16.4
7.26
11.0
10.7
18.1
15.8
13.7
17.1
17.7
17.7
54.3
16.9
31.0
164
35.0
35.0
31.0
26.5
16.8
239
38.7
18.5
26.3
25.4
31.4
49.0
36.1
47.1
79.7
79.7
76.0
30.0
97.7
456
151
250
121
83.1
40.0
513
106
99.5
71.0
62.4
84.4
197
145
375
375
375
266
8660
3730
89100
2490
2650
8800
4540
13800
4250
6050
2780
31900
668
22200
15500
89100
Houses Built From 1940-1959
HUD
National
Survey
unweighted1
orig. weights2
§403 RA»
Rochester Lead-in-Dust5
HUD
Grantees
Alameda Co.
Baltimore
Boston
California
Massachusetts
Minnesota
87
87
87
19
19
43
4
51
5
1
19.8
18.1
18.1
11.8
153
494
57.3
41.7
55.5
149
9.20
8.74
8.74
8.36
32.1
120
26.6
15.4
46.5
149
3.53
3.34
3.34
2.61
7.15
9.13
4.46
3.29
1.93
-
0.508
0.508
0.508
1.21
2.00
0.250
5.00
2.75
22.5
149
4.20
4.07
4.07
3.54
5.75
39.5
10.0
6.25
30.0
149
8.32
7.81
7.81
11.1
17.0
197
27.0
10.1
39.8
149
22.5
22.4
22.4
19.2
157
648
105
33.3
70.3
149
171
171
171
26.9
909
4170
170
825
115
149
                                        90

-------
                                             Table 3-14.  (cont.)
Study
Approach/
Grantee
: '' Area-Weighted Household Average, Pre-tatervention Floor
# Units
with
Data
•, Arith-
nwtic
Mean
Gee-
inotnc
M«___
^MQafi
Get*
• metric ,
Std. Dev.
iMkiinwm
25*
Percen-
tDe
Dust-Lead Loading fira/ft*) " -• -

Median
75*;
Percen-
tBe
Maximum
Houses BuiJt From 194O - 1959 (com.)
HUD
Grantees
Rhode Island
Wisconsin
Milwaukee
Chicago
Vermont
All Grantees
34
15
&
5
31
213
81.3
87.0
14.0
5300
38.4
276
27.3
7.22
6.78
102
26.4
30.1
5.47
6.99
4.14
23.8
2.23
5.39
0.250
0.800
1.50
16.4
8.00
0.250
7.60
1.60
2.25
17.8
15.0
8.00
36.6
5.72
4.88
19.2
17.8
24.3
77.3
17.1
22.5
75.8
45.5
89.3
617
1050
38.8
26400
219
26400
Houses Built From 1960 - 1979 (1960 - 1977 for HUD Grantees)
HUD
National
Survey
unweighted'
orig. weights2
S403RA3
Rochester Lead-in-Dust5
HUD
Grantees
Boston
Cleveland
New Jersey
Wisconsin
All Grantees
118
118
120
4
1
1
16
6
24
7.14
6.74
6.74
9.65
18.8
9.25
32.6
4.42
24.0
4.30
4.11
4.14
7.84
18.8
9.25
13.6
4.01
10.0
2.50
2.46
2.45
2.40
-
-
3.70
1.61
3.14
0.657
0.657
0.657
2.13
18.8
9.25
1.75
2.40
1.75
2.26
2.25
2.25
6.38
18.8
9.25
6.58
2.50
4.45
3.85
3.62
3.62
11.6
18.8
9.25
10.0
3.84
8.88
7.59
7.59
7.59
12.9
18.8
9.25
34.6
5.93
24.6
106
106
106
13.2
18.8
9.25
245
8.02
245
Houses Buih After 1979 (After 1 977 for HUD Grantees)
HUD National Survey
{1403 RA>3
Baltimore R&M4
Rochester Lead-in-Dust*
HUD
Grantees
Minnesota
Rhode Island
All Grantees
28
16
10
1
3
4
4.16
10.9
37.2
32.4
984
746
3.14
9.97
15.0
32.4
838
372
2.06
1.55
3.34
-
2.00
2.00
1.06
4.48
3.48
32.4
440
32.4
1.76
7.13
5.57
32.4
440
236
2.84
10.5
16.8
32.4
763
602
5.66
14.7
21.2
32.4
1750
1260
12.9
17.4
250
32.4
1750
1750
' Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized without weighting by sample weights.
2 Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized by weighting each average by the original sample weights assigned in the survey.
3 Area-weighted average dust-lead loadings, as calculated in Chapter 3 of the §403 risk analysis, are summarized by
weighting each average to reflect the 1997 U.S. housing stock and imputing averages for units with missing data.
4 BRM dust-lead loadings are converted to wipe-equivalent loadings prior to summary in this table.
* Some houses in this housing age category may belong to an earlier age category, as some houses may have actually been
built earlier than the year specified within the study's database.
                                                       91

-------
Table 3-15.  Descriptive Statistics of Area-Weighted Average Pre-lntervention Window
             Sill Wipe Dust-Lead Loadings for Households, Presented bv Housing Age
             Category. As Reported in the §403 Risk Analysis, the HUD National Survey,
             and Other Studies
-:Stwp-':

HUO
National
Survey
Approach/
Grantee

unweighted1
orig. weights2
1403 RA3
Baltimore R&M"
Rochester Lead-in-Dust
HUD
Grantees
Alameda Co.
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
New Jersey
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
All Grantees
Area-Weighted Household Average Pre-lntervention Window SB Dust-Lead Loading (fig/ft2)
# Units
with
Data

71
71
77
74
164
148
347
71
35
172
146
177
26
123
211
261
140
368
269
2494
ArraV
matte
Mean

1610
2060
2060
751
613
767
7070
5150
1530
4120
1770
3320
1080
5780
3020
3610
1630
1530
3860
3480
Geo-
metric
Mean
Geo-
nwtnc
SttLDev.


54.7
146
168
555
234
138
1600
577
506
233
322
282
328
816
294
543
259
258
272
391
19.6
16.8
16.7
2.41
3.67
9.34
7.69
7.89
5.70
16.9
5.81
6.30
5.08
6.49
8.91
6.89
5.78
5.51
15.9
8.22
Minimum
1940
0.0155
0.0155
0.0155
44.0
2.85
0.0016
<0.0001
14.0
28.0
< 0.0001
2.60
5.66
21.3
12.0
0.0008
1.00
3.02
0.320
<0.0001
<0.0001
25*
PwcBfl-
tfe

7.05
35.6
35.6
399
95.3
44.7
451
135
159
63.8
93.8
71.0
99.6
192
81.4
127
85.0
95.6
72.0
106
Median

67.1
198
198
628
223
164
1690
425
524
270
296
190
276
709
258
413
267
175
275
351
75* -
* P0TC6ft-
«e

442
1220
1220
989
475
566
6140
2410
2440
876
1090
945
1170
2500
1090
2110
852
543
1340
1470
Maximum

43700
43700
43700
2330
14900
19700
220000
106000
9630
241000
63400
300000
8450
132000
142000
88000
50500
57100
98100
300000
Houses Built From 1940-1959
HUD
National
Survey
unweighted1
orig. weights2
1403 RA3
Rochester Lead-in-Dust9
HUD
Grantees
Alameda Co.
Baltimore
Boston
California
Massachusetts
Minnesota
79
79
87
18
20
43
4
42
4
1
430
285
285
399
152
4310
382
395
142
289
23.1
17.9
22.0
72.0
47.7
1330
150
203
59.7
289
11.4
10.5
10.7
6.16
8.04
5.39
5.20
3.41
8.20
-
0.0118
0.0118
0.0118
2.83
0.140
33.0
39.4
11.0
2.79
289
6.47
6.47
6.47
23.0
14.5
256
39.6
89.9
47.9
289
21.7
19.1
19.1
56.0
71.1
1600
160
190
123
289
107
107
107
194
260
4820
724
565
237
289
16100
16100
16100
4390
580
29400
1170
1850
321
289
                                        92

-------
                                            Table 3-15.   (cont.)
Study
• — >
— ApproacH/
Grantee _,
* - Aran-IAfa&iRtowt t>

93****
''with
•Data
Arith-
metic
Mean
tousehold AveragevPre^nterventton Window SOI Dust-Lead Loading fc/g/ft2} ;
Geo-
metric'.
Mean
Geo-
'IflOtflC*'
Std.Dw.
Houses Built From 1940 -
HUD
Grantees
Rhode Island
Wisconsin
Milwaukee
Chicago
Vermont
All Grantees
34
16
6
5
30
205
1520
497
552
835
52.4
1350
416
148
140
449
40.4
222
7.53
4.24
7.09
3.84
2.08
4.94
**"~<~ 1*
Inuumum
25*
PfeFccn*
tile
* v&'~
Median
^75* '
Pctfoon-
- «e
Maximum
1959 (cont.)
0.500
24.0
18.0
111
7.00
0.140
144
47.4
28.8
120
31.0
45.0
617
105
123
521
45.0
205
1120
338
797
1170
45.0
814
9970
4750
2220
2250
212
29400
Houses Butt From 1960 - 1979 (1960 - 1977 for HUD Grantees)
HUD
National
Survey
unweighted'
orig. weights2
§403 RA3
Rochester Lead-in-Dust*
HUD
Grantees
Boston
Cleveland
New Jersey
Wisconsin
All Grantees
95
95
120
4
1
1
20
6
28
190
184
184
54.4
289
409
59.8
209
112
10.3
9.10
16.2
52.3
289
409
12.9
153
27.8
13.3
14.5
14.6
1.38
-
-
63.1
2.90
39.3
0.0164
0.0164
0.0164
36.2
289
409
<0.0001
21.0
<0.0001
1.68
2.05
2.05
40.0
289
409
17.8
105
20.9
8.69
16.6
16.6
55.2
289
409
29.6
240
44.4
51.3
217
217
68.7
289
409
72.7
289
179
5790
5790
5790
70.7
289
409
333
359
409
Houses Built After 1979 (After 1977 for HUD Grantees)
HUD National Survey
(1403 RAJ*
Baltimore R&M4
Rochester Lead-in-Dust*
HUD
Grantees
Minnesota
Rhode Island
All Grantees
28
16
10
1
1
2
83.0
50.8
134
2350
816
1580
8.17
45.6
113
2350
816
1390
9.94
1.65
1.95
-
-
-
0.0164
20.6
26.9
2350
816
816
2.58
27.1
75.7
2350
816
816
8.11
52.6
125
2350
816
1580
57.8
66.5
159
2350
816
2350
1590
85.9
320
2350
816
2350
1 Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized without weighting by sample weights.
2 Area-weighted average dust-lead loadings, as reported in the HUD National Survey but converted to wipe-equivalent
loadings, are summarized by weighting each average by the original sample weights assigned in the survey.
* Area-weighted average dust-lead loadings, as calculated in Chapter 3 of the 5403 risk analysis, are summarized by
weighting each average to reflect the 1997 U.S. housing stock and imputing averages for units with missing data.
* BRM dust-lead loadings are converted to wipe-equivalent loadings prior to summary in this table.
5 Some houses in this housing age category may belong to an earlier age category, as some houses may have actually been
built earlier than the year specified within the study's database.
                                                       93

-------
       •      For each value of P, a normal distribution was fitted to the log-transformed data,
             taking into account the censoring of the lower P% of the data and the sample
             weights, using the LIFEREG procedure in the SAS® System.

       •      The value of P (and its corresponding cut-off X) was identified that resulted in the
             best fit for normality in the upper tail of the distribution (based on review of
             normal probability plots).  The exceedance percentages were estimated based on
             this final distribution, using normal probability theory.

This procedure was applied separately to HUD National Survey data and interim data from the
NSLAH. Exceedance percentages were estimated for each of the following floor dust-lead
loading thresholds: 5,10,20,30,40, and 50 fig/ft2.

       Figure 3-11 contains the fitted distributions based on the HUD National Survey data (top
plot) and the interim NSLAH data (bottom plot). (The top plot is labeled "Section 403 risk
analysis" as it reflects sample weights adjusted for the 1997  housing stock and dust-lead loadings
converted to wipe-equivalents, both done within the §403 risk analysis.)  Each plot contains a bar
chart of the observed data, onto which the fitted lognormal distribution curve is superimposed.
Note that the same floor dust-lead loading (horizontal) axis is used for both plots, so that the two
plots can be directly compared. As can be noted in this figure (and which was seen in the
summaries in Section 3.2.1.1), the distribution based on the interim NSLAH data covers a
considerably lower range compared to the distribution based on the HUD National Survey data
used in the §403 risk analysis. Thus, the estimated exceedance percentages for each of the six
thresholds, also annotated within each plot, are considerably lower based on the interim NSLAH
data, especially as the threshold increases.

       Each estimated exceedance percentage within Figure 3-11 is accompanied by an
approximate 95% confidence interval on the number of homes in the U.S. housing stock that
exceeds the threshold.  These intervals were calculated based on the estimated total number of
housing units in the housing stock, as determined by the sum of the sampling weights for the
given survey (which is specified within each plot).

       In Figure 3-11, the distribution based on the HUD National Survey data used in the §403
risk analysis was determined by censoring data values below 3.81 fig/ft2 (i.e., the bottom 40
percent of the data, taking into account the sample weights). The distribution based on the
interim NSLAH data was determined by censoring data values below 0.2025 ug/ft2, which
corresponds to the bottom 20 percent of the observed weighted distribution, including negative
values.

       For both surveys, the estimated exceedance percentages specified within Figure 3-11 for
household average floor dust-lead loading, based on the fitted lognormal distribution, are also
included within Table 3-16 (columns 2 and 4) for the same six thresholds.  Also included in
Table 3-16 (columns 3 and 5) are  estimated exceedance percentages that were determined solely
by the proportion of total sampling weights in the survey that corresponded to surveyed units

                                           94

-------
                                          a
                                              ... £
                                              V S
                                       w
95

-------
Table 3-16.   Estimated Percentages of 1997 U.S. Housing Exceeding Specified
              Thresholds of Household Average Dust-Lead Loading
Dust-
Lead
Loading •
Threshold '
tog/ft2)
5
10
20
30
40
50
§403 Risk Analysis - Based on Data from the
HUD National Survey (n = 284)
Based on the Fitted
, Lognormal Distribution
(i.e., the curve in
Figure 3-11)
51.9%
33.6%
18.5%
12.0%
8.5%
6.4%
Based on the
. Weighted Observed
Data (i.e., the bar
chart in Figure 3-11)
51.1%
30.6%
15.6%
11.9%
9.6%
8.3%
Data from the Interim NSLAH (n= 706J
Based1 on the Fitted
Lognormat
Distribution (i.e., the
curve in Figure 3-11)
15.8%
7.9%
3.4%
2.0%
1.3%
0.9%
Based on the
Weighted Observed
Data (i.e., the bar:
chart in Figure 3-11)
13.2%
7.2%
4.0%
2.0%
1.4%
1.2%
Note: Data are imputed for those surveyed units with missing data prior to calculating the above statistics (3 observations in
the HUD National Survey and 9 observations in the interim NSLAH had imputed data). The estimates based on the weighted
observed data are simple weighted percentiles that do not originate from a fitted distribution.
whose household average floor dust-lead loadings exceeded the given threshold (i.e., Information
from the bar charts within Figure 3-11).  These results are included to evaluate the similarity
between the lognormal-based estimates and those generated from an approach that uses only the
observed data without an underlying distribution assumption.  As Table 3-16 shows, the
lognormal-based estimates are slightly lower for the lower thresholds and slightly higher for the
higher thresholds, while the two approaches yield nearly equivalent estimates at the threshold of
30 ug/ft2.  It should be noted that the lognormal-based estimates for the exceedance percentages
(which were also  portrayed in Figure 3-11) should be used when making inferences on the
nation's housing stock.

       3.2.1.4  Interpreting the Observed Differences with Other Studies. In order to make
proper interpretations from the results portrayed in this subsection, in particular why differences
exist between the studies, one must be aware of how the housing selection procedure and sample
collection and analysis procedures differ between the studies and can contribute to the
differences observed in the boxplots and tables. For the studies highlighted in the §403 risk
analysis report, this information was summarized in Tables 3-3a through 3-3f of that report.
Some of the differences among these studies that may contribute to differences in the reported
data are as follows:

       •      All non-control housing units in the Baltimore R&M study, approximately 88
              percent of units selected in the HUD Grantees evaluation, and at least 84 percent
              of the Rochester study units were built prior to  1941.  In contrast, only 27 percent
              of the housing units in the HUD National Survey were built prior to 1940.
                                            96

-------
      •      The neighborhoods surveyed within the Baltimore R&M study and HUD Grantees
             evaluation had a high prevalence of homes with lead-based paint hazards, along
             with a history of children with elevated blood-lead concentrations and/or
             considered at high-risk for lead poisoning.

      •      The HUD National Survey targeted occupied permanent housing throughout the
             48 contiguous states. The units were selected via a statistically-based sampling
             design to represent the national housing stock built prior to 1980. Excluding two
             grantees from California, the HUD Grantees (as well as the Rochester and
             Baltimore R&M studies) sampled housing from the Northeast and Midwest
             census regions. Approximately two out of every five homes sampled in the HUD
             National Survey were from the South.

      •      For the HUD Grantees evaluation, 28 percent of the homes were single-family
             buildings (16 percent were single-family detached, and 12 percent were single-
             family attached, or rowhouses). All homes in the R&M intervention group within
             the Baltimore R&M study were urban rowhouses (single-family attached). Eighty
             percent of the homes in the HUD National Survey were single-family dwellings.

      •      From 8 percent to 67 percent of the dwelling units for any one Grantee were
             vacant prior to sampling. Out of the 5,265 dwelling units in total that were
             enrolled as of January 1999,1524 units were vacant prior to pre-intervention
             sampling. Overall vacancy rate was 29 percent for the Evaluation. On the other
             hand, the HUD National Survey contained dwelling units which were permanent
             and occupied, with the potential for containing children.

      •      The dates of environmental sampling were 11/89-3/90 for the HUD National
             Survey, 12/93-1/99 for the HUD Grantees evaluation (pre-intervention), 8/93-
             11/93 for the Rochester study, and 3/93-11/94 for the Baltimore R&M study.
             Therefore, the HUD National Survey performed sampling roughly three years
             before each of the other studies and during the late fall and winter months.

Section 3.1 discussed differences in approaches and methods between the HUD National Survey
and the NSLAH that could impact observed differences in the reported data.

      3.2.1.5 Conclusions of the Dust-Lead Data Comparisons. The following
conclusions could be made upon review of the dust-lead loading summaries within Tables 3-4
through 3-16 and Figures 3-1 through 3-11:

      •      For both floors and window sills, the interim NSLAH data are considerably lower
             than that reported in the §403 risk analysis (and based on the HUD National
             Survey data), as well as for all other sources of data available to the risk analysis.
             Household average floor dust-lead loadings had a median of less than 2.0 ug/ft2

                                          97

-------
             across the interim NSLAH data, while household average window sill dust-lead
             loadings had a median of approximately 12.0 ng/ft2. Approximately two-thirds of
             the floor-dust samples and one-third of the window sill-dust samples had lead
             measurements below the detection limit in the interim NSLAH. Further
             investigation is necessary to determine the reasons for such low dust-lead loadings
             in the interim NSLAH.

       •      Compared to the other lead exposure studies whose data were considered in the
             §403 risk analysis (e.g., Rochester study, Baltimore R&M study, HUD Grantees
             evaluation), geometric mean dust-lead loadings tended to be lower in the HUD
             National Survey. However, all of these studies had similar ranges of observed
             dust-lead loading data. This suggests that 1) the conversions to wipe-equivalent
             dust-lead loadings performed on the HUD National Survey data in the §403 risk
             analysis did not lead to extreme adjustments overall, and 2) there is not sufficient
             evidence that data from the HUD National Survey are higher than what is
             representative of the 1997 housing stock simply because it was performed some
             years earlier.

       •      The importance of housing age is evident hi the summaries within the four
             housing age categories. Older housing is more likely to contain higher average
             dust-lead loadings compared to newer housing.  However, within an age category,
             the summaries were quite consistent across studies (with the exception of the
             interim NSLAH).

       •      The percentage of housing units with average floor dust-lead loadings that exceed
             50 ug/ft2 (i.e., the proposed floor dust-lead standard) was 6.4% based on data used
             hi the §403 risk analysis, and 0.9% based on interim data from the NSLAH.

3.2.2  Characterizing Soil-Lead Concentrations

       This subsection summarizes observed soil-lead concentrations in the HUD National
Survey and how these data were used to characterize soil-lead levels in the §403 risk analysis,
and compares these summaries with summaries of the interim NSLAH data (Section 3.1), as well
as data for 22 other studies that characterized soil-lead concentrations in urban areas prior to any
lead abatement. These 22 studies include the three recent studies included in the dust-lead data
summaries of the previous section (Baltimore R&M study, Rochester study, and HUD Grantees
evaluation) and other studies dating to the early 1970s (e.g.,  Omaha, Charleston).  Sampling and
laboratory protocols for the 22 additional studies are summarized in Table 3-17. The soil-lead
data summaries for these 22 studies were either calculated directly from the available data set or
culled from the published scientific literature.

       Household mass-weighted average soil-lead concentration for a specific portion of the
yard was the basis for the comparisons made in this section.  This average was calculated by
weighting the result for each soil sample taken at that location by the sample's mass. If this

                                          98

-------
Table 3-17.   Information on Soil Sampling and Analysis Protocols for Studies Whose Soil-
                 Lead  Data Were Compared to Results from the §403 Risk Analysis and the
                 HUD  National Survey
     Study
  [Reference]
            So3 Samp&ng and Analysis Details
SoB-Lead ParametertsJ Used in This Section
  for Comparison to HUD National Survey
   Baltimore
     R&M
   {USEPA,
    1996c)
1993-94.  Three 0.5" core samples per composite, taken
from randomly determined areas along the driptine using a 6"
stainless steel recovery probe and collected into a polysterene
liner.  Samples were sieved and homogenized and digested
using SW 846-3015 and SW 846-3051.  GFAA (SW
846-7421) laboratory analysis method. Only data for
occupied units were used.
Soil-lead concentration for each composite
sample (one composite sample per housing
unit, taken from the dripline).
   Baltimore
 Urban Garden
 Soil (Mielke et
   al., 1983)
1982.  Samples were from garden soil in random locations
within a 30-mile radius of downtown Baltimore. Samples
were air-dried and sieved with a 2mm stainless steel mesh
screen and digested in nitric acid.  Extracts were filtered and
analyzed using a Varian atomic absorption spectrophotometer
with deuterium background correction.
Soil-lead concentration for each collected
sample.
     3-Crty
  (Baltimore,
    Boston,
  Cincinnati)
    (USEPA,
    1996a)

  (Also known
  as the Urban
   Soil Lead
  Abatement
  Demonstra-
  tion Project)
               Only round 1 (pre-abatement) measurements were used.

               Baltimore/Boston: Only results for the top 2 cm of a 15 cm
core sample were considered.
Cincinnati: Soil samples were collected within neighborhoods.
as well as within the yards of surveyed housing units.
Baltimore: Yard-wide average for a unit,
equal to the unweighted arithmetic average
of the unit's average dripline, average mid-
yard and average boundary soil-lead
concentrations within a property (set to
missing if any of these measurements were
missing). The location averages were also
summarized.
Boston: Yard-wide average for a unit, equal
to the average across all samples
associated with that unit.
Cincinnati: Yard-wide average for a unit,
equal to the unweighted arithmetic average
of average building and average play area
soil-lead concentrations for the unit (set to
missing if either of these measurements are
missing). These and other location
averages were also summarized.
    Boston
  Brigham and
    Women
 (Rabinowitz et
   al., 1985)
3 samples were collected one meter apart and at least 3
meters from any road structure (preference given to obvious
play areas). These samples were composited prior to
analysis. Soil sampling occurred twice: when the resident
child of interest was 18 and 24 months of age. Laboratory
analysis method was atomic absorption spectrophotometry
(AAS).
Soil-lead concentration for each unit, equal
to the unweighted arithmetic average of
soil-lead concentrations for composite
samples taken at the 18 and 24 month
visits.
  CAP Study
   (USEPA,
    1996b)
1990.  Soil samples taken from Denver units that were
abated in 1989 during the HUD Abatement Demonstration
Study.  Samples were collected from the dripline, entryway,
and remote areas of the yard with a soil recovery probe (1"
diameter liner and 12" core sampler).  At each location, a
composite sample consisted of 3 cores, each 0.5" in depth.
The sample preparation method was EPA SW846 Method
3050 (included use of nitric acid and hydrogen peroxide for
digestion). The laboratory analysis method was ICP-AES.
Soil-lead concentration for the dripline,
entryway, and remote areas of the yard.
(One composite sample per location per
unit.)
                                                     99

-------
                                             Table 3-17 (com.)
    Study
  [Reference]
             Sofl Sampling and Analysis Deta3s
                                                         SoD-L(
               rterfs) Used in This Section
                                                           foe Comparison to HltDvNationai Survey
  California
(Sutton et al.,
    1995)
1987-91. Older units in Oakland, Los Angeles, and
Sacramento.  Composite soil samples (of 4 subsamples) were
collected at each of the front, side, and rear yards.  In
addition, units in Oakland and Los Angeles had a composite
soil sample collected from a secondary structure (e.g.,
garage) and a single sample collected from rain drains. All but
rain drain samples were composites. Samples were < 1" in
depth and were collected using a trowel (visible paint chips
removed first).  The laboratory analysis method was AAS.
                                                                       Soil-lead concentration for each collected
                                                                       sample (from 3 to 5 per unit).
  Cincinnati
 Longitudinal
(Bornschein et
 a)., 1985a;
 1986; Que
  Hee et al.,
    1985)
1980-87. Surface scrapings rather than soil cores were
taken. Laboratory analysis method was AAS.  Enrolled
expectant mothers residing in areas with a history of child
residents with elevated blood-lead concentrations.
Not determined'
  Cincinnati
  Roadside
 (Tong, 1990)
1990.  Samples were collected near highways, boulevards,
and cul-de-sacs in two neighborhoods (not industrial areas nor
poor neighborhoods with deteriorated housing) within the
Greater Cincinnati Metropolitan District. Samples were from
a depth of 0-5cm and were analyzed using a Leeman plasma
spectrophotometer with background correction.
                                                                       Not determined1.
  Charleston
 (Galke et al.,
    1975)
1973.  Soil samples taken from a child's primary play area.
Laboratory analysis method was AAS.
Not determined'
Corpus Christ!
  (Harrison,
    1987)
1984.  Samples were collected from parks, schools, and
roadside embankments within the city limits of Corpus
Christ!, Texas from vegetated, non-sandy soil.  The top 2 cm
of soil was sampled with a Teflon knife. The laboratory
analysis method was AAS.
Soil-lead concentration for each collected
sample.
    1-880
  (Alameda
   County)
(Teichmean et
  al., 1993)
1990. Samples were collected from homes, parks,
playgrounds, and public housing developments within one
mile east or west of 1-880.  The top 0.50" to 0.75" of soil
was sampled.  The laboratory analysis method was AAS.
Soil-lead concentration for each collected
sample.
    HUD
  Abatement
  Demonstra-
  tion Study
  (USHUD,
    1991)
Dripline samples were taken from 1 to 3 feet from an exterior
wall and were composites of 5 subsamples.  Soil sampling
(and compositing) occurred  twice: prior to and following lead-
based paint abatements performed in this demonstration. The
laboratory analysis method was AAS.
Soil-lead concentration for each dripline
composite sample (one composite sample
per housing unit collected at pre-
intervention, and one sample per unit
collected at post-intervention)
HUD Grantees
  Evaluation
  (USHUD,
    1998)
1994-97. Pre-lntervention phase only. From 5-10 core
samples were taken at 0.5-1" depths at a given location and
composited. Locations were the dripline (samples taken from
all sides of the unit, 2' from foundation and 2* from each
other) and play areas (samples collected along x-shaped grids
at least 1' from each other).
Yard-wide average for a unit, equal to the
unweighted arithmetic average of dripline
and play area soil-lead concentrations
within a unit (set to missing if either of
these measurements were missing).
 Maine Urban
  (Krueger et
  al., 1989)
1988.  Samples collected from units at least 30 years of age
and from parks/playgrounds in Portland, Maine.  A single
composite sample, consisting of 4 cores taken 2' from the
foundation, was associated with each housing unit.
Laboratory analysis method was AAS.            	
Soil-lead concentration for each composite
sample (housing units) and each sample
collected from parks/playgrounds.
                                                      100

-------
                                             Table 3-17 (cont.)
    Study
  (Reference]
                             Analysis Details
              Soil samples collected from perimeter and play areas at each
              housing unit.
 Sod-Lead Pararneterts) Used in This Section
  for Comparison to HUD National Survey ';
  Milwaukee
  (Pendleton)
                                                         Yard-wide average for a unit, equal to the
                                                         unweighted arithmetic average of perimeter
                                                         and play area soil-lead concentrations (set
                                                         to missing if either of these measurements
                                                         are missing).
  Minneapolis
   Clean-Up
 (Mielke et al.,
    1992)
Only pre-cleanup data were considered. Deep scrape samples
were taken at a depth of 2.5 cm, air-dried and sieved with a
2 mm stainless steel mesh screen, and digested in nitric acid.
Extracts were filtered and analyzed using a Varian atomic
absorption spectrophotometer with deuterium background
correction.
                                                                        Not determined1.
  Minnesota
  (Schmitt et
  al., 1988;
 Mielke et al.,
    1989)
1986-87.  Only results for St. Paul and Minneapolis were
considered (except results labeled "Whole Study* also
included Duluth, Rochester, St. Cloud and rural areas).
Foundation samples were taken within  1.5 m of building.
Yard samples (front, side, and back) were taken at the
midpoint of the yard and at least 1.5 m from the foundation.
Street samples were taken within 1.5 m of a curb.  Samples
were from the top 2 cm of soil. The laboratory analysis
method was 1CP-AES.
                                                                        Soil-lead concentration for each collected
                                                                        sample.
 New Orleans
   (Mielke,
 1995; 1993)
1983.  Samples taken from residential neighborhoods within
283 census tracts in the New Orleans metropolitan area.
Foundation samples were taken within 1 m of a house.
Streetside samples were taken from within 1 m of a street.
Open area samples were from vacant lots or parks.  The
laboratory analysis method was AAS with deuterium
background correction.
                                                                        Soil-lead concentration for each collected
                                                                        sample.
 New Haven,
 Connecticut
 (Stark et al.,
    1982)
1974-77. Samples (5-10 g) collected from homes of children
who lived at the same address for at least one year. Only the
top 0.5" of soil was analyzed.
Not determined1.
    Omaha .
 (Angle et al.,
    1979)
1971-77. Soil core samples (2* depth) self-selected from
halfway between the building and lot line on four sides of the
selected units.
Yard-wide average for a unit, equal to the
arithmetic average soil-lead concentration
across all collected samples at the unit.
  Rochester
 Lead-in-Dust
   (USHUD,
    1995a;
  Lanphear et
  al., 1996a)
1993.  Two composite samples, one from the dripline (12
samples per composite} and one from play areas (8-10
samples per composite). Core samples were taken at a depth
of 0.5". Composites were mixed and sieved into fine and
coarse fractions and analyzed separately. Digestion method
was SW 846-3050, and the laboratory analysis method was
FAA (method 239.1).
Total soil-lead concentrations were computed as 0.25'Fine
Soil Fraction + 0:75*Coarse Soil Fraction (see Appendix E).
Yard-wide average for a unit (for both total
soil and fine soil only), equal to the
unweighted arithmetic average of the unit's
dripline and play area soil-lead
concentrations (set to missing if either of
these measurements was missing). The
soil-lead concentration for the dripline
sample at each  unit was also summarized
(for both total soil and fine soil only).
 Washington,
     DC
 (Elhelu et al.,
    1995)
Housing units were randomly selected from each of the 8
wards of Washington, DC. Soil samples were collected from
unpaved front yards approximately 1 m from the unit and at a
depth of  15cm. Average dwelling distance from the road was
4.5 m. Fine soil samples were analyzed with a Perkin Elmer
2100 Atomic Absorption Spectrophotometer, with one result
associated with each surveyed unit.                	
                                                                        Soil-lead concentration for each collected
                                                                        sample (i.e., each housing unit).
' Most likely soil-lead concentration for each collected sample.
                                                     101

-------
average could not be calculated for a given study due to insufficient data, then alternative
statistics were calculated. For example, if mass weights were not available, the arithmetic
average soil-lead concentration was instead calculated.

       When possible, a yard-wide average soil-lead concentration was calculated in a manner
that attempted to be consistent with the §403 risk analysis. This involved taking a weighted
arithmetic average of the soil-lead concentrations reported at the dripline, unit entryway, and
remote areas of the yard, with remote concentrations weighted twice as much as the dripline and
entryway concentrations. (When only one of the dripline or entryway concentrations was
available at a housing unit, the yard-wide average was the unweighted arithmetic average of that
one concentration and the remote soil-lead concentration.) Thus, the yard-wide average was
essentially an arithmetic average of two measures: the average soil-lead level at the dripline and
unit entryway (i.e., "near" the housing unit) and the soil-lead level at a remote area of the yard
(i.e., "far" from the housing unit). It was assumed that "play areas" represented remote areas of
the yard.  Imputed data values replaced missing values for a housing unit in the §403 risk
analysis summaries, where imputation methods discussed in Section 3.3.1.1 of the §403 risk
analysis report were used.

       3.2.2.1   Data Summaries for the §403 Risk Analysis Versus the Interim NSLAH.
Descriptive statistics of yard-wide average soil-lead concentrations as calculated in the §403 risk
analysis using the HUD National Survey data are presented in this subsection as they compare
with the same statistics calculated on interim data for 706 housing units in the NSLAH. Note
that these statistics reflect the sampling weights used in the §403 risk analysis and the interim
NSLAH sample weights, thereby allowing these summaries to be nationally representative of the
1997 housing stock.  In addition, the interim NSLAH summaries do not include any data that
may have been imputed within the revised §403 risk analysis when missing data for key
parameters were encountered for a housing unit.

       As in the dust-lead loading summaries (Section 3.2.1.1), the interim NSLAH summaries
include imputed values of yard-wide average soil-lead concentration for those housing units
having no reported soil-lead concentration data. As discussed in Appendix C, the imputation
method involved imputing values for average dripline/entryway soil-lead concentration and for
average mid-yard soil-lead concentration, then averaging these two imputed values together. If
data existed for one of the two locations but not the other, the yard-wide  average for that unit
equaled the average soil-lead concentration at the location represented by the available data.
Appendix C also gives the imputed data values and how they were assigned to housing units.
Summaries of the interim yard-wide average soil-lead concentration data from the interim
NSLAH excluding any imputed data can be found in Appendix D2.

       Also, in the same manner as the dust-lead loading summaries (Section 3.2.1.1), Appendix
D2 presents soil-lead concentration summaries for the interim NSLAH under five different
approaches (including data censoring) to handling sample results that were below the detection
limit. The summaries in this subsection were calculated under two of these approaches:
                                         102

-------
       •      making no adjustment to not-detected data values
       *      replacing not-detected data values with one-half of the detection limit.

These two approaches, the same two used in the dust-lead loading data summaries in Section
3.2.1.1, were included together in the summary tables to illustrate the impact that any one
approach has on the characterized distribution of yard-wide average soil-lead concentration.

National comparisons

       Table 3-18 presents descriptive statistics of yard-wide average soil-lead concentrations
for the 1997 national housing stock.  These results indicate that only a slight downward shift in
the distribution of soil-lead concentrations was observed from the §403 risk analysis to the
interim NSLAH data, (e.g., a decline in the geometric mean from 62 ug/g to approximately 53
ug/g). This decline was much smaller than that observed for dust-lead loadings.

       Boxplots of the data distributions presented in Table 3-18 are found hi Figure 3-12.
When not-detected data hi the NSLAH were replaced by one-half of the detection limit, the
observed distribution of yard-wide average soil-lead concentration appears similar to what was
characterized in the §403 risk analysis.  Appendix D2 contains tabular summaries and boxplots
after excluding imputed data values.
Table 3-18.   Descriptive Statistics of Yard-Wide Average Soil-Lead Concentrations for
               Households, As Reported in the §403 Risk Analysis Versus the Interim
               NSLAH Data
1 Study
How Not-
Defected
and
Negative
ii Data were
Handled
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced
by LOD/2
; Yard-Wide Average Soil-Lead Concentration 0*8/9)'
*
Surveyed
Unto with
Positive
Averages
284
689
706
Arith-
metic
Mean
235
200
200
Geo-
metric
Mean2
61.9
53.0
52.6
Geo-
metric -
Stt.
Dev.2
4.46
5.09
4.73
MiniifKini
4.63
0.00
4.62
25*
Percen-
tife
21.3
16.6
16.8
Median
49.2
41.8
41.4
75*
Percen-;
tfle
142
158
158
Maximum
7030
9270
9270
1  All statistics are calculated by weighting each household by its sampling weight.
1  Only household averages greater than zero are used to calculate this value (data for all units with soil-lead data are used to
calculate the remaining statistics).
3  Summaries include imputed data for households having no soil-lead concentration data. The method for imputation is
presented in Appendix C.
Note: The yard-wide average for a household is the average of the following two statistics: 1) the average of the mid-yard
sample results, and 2) the average of results for the drtpline and entryway samples.
                                             103

-------
                                   I
                                 51
                                 « s>
                                  SB
                                 .0 «
                                 OCM
                             I
                                 ^^
                                 2 £

                                 «i
                                 = 0,
                                   £    §
                                  0) CO
                                  2?°
                                  II
                                  *;
                                  ^§
   8     5



(iudd) uogtaifUi




         104

g-


O


CM
       2
       !
       b
       s
       1
       1
       I
       A
                                        •o
                                        1
                                        Q.
  I  I
2<>  |

     I
     I
     CO
o>
E

-------
      The detection limit for soil-lead concentrations in the interim NSLAH ranged from 7.2 to
12.4 ug/g, with a mean (and median) of 9.9 ug/g. Of those soil samples in the interim NSLAH
with soil-lead concentrations reported, approximately 22% (covering approximately 38% of
housing units reporting soil-lead concentrations) had soil results below the detection limit.

      In addition to these data summaries that are based solely on the observed data and the
sampling weights, it was desired to characterize the national distribution of yardwide average
soil-lead concentration in such a way that the percentage of housing where this average exceeds a
specified threshold could be estimated. Like what was done in Section 3.2.1.3 above for floor
dust-lead loading, this was done for both the HUD National Survey and interim NSLAH data by
assuming that these data originate from a lognormal distribution. Then, the fitted distributions
and corresponding estimated exceedance percentages were compared between the two surveys.
These results are presented in Section 3.2.2.4 below.

Comparisons bv housing age category

      The distribution of yard-wide average soil-lead concentrations is portrayed for each study
according to housing age category in Table 3-19.  The importance of housing age on yard-wide
average soil-lead concentration is  seen in both surveys, as the geometric mean and median
concentrations tend to increase with the age of house. The method to handling not-detected
values in the interim NSLAH dataset affected the data summaries only slightly, if at all.

      Boxplots associated with the data distributions portrayed in Table 3-19 are found in
Figure 3-13. Appendix D2 contains tabular summaries and boxplots after excluding imputed
data values.

Comparisons bv Census region

      The distribution of yard-wide average soil-lead concentrations is portrayed for each study
according to Census region in Table 3-20. Geometric mean estimates declined from the §403
risk analysis to the interim NSLAH data for each Census  region, but the magnitude of the
declines were typically small. Observed median values increased from the §403 risk analysis to
the interim NSLAH data for the Midwest and West, but these increases were likely due to
random chance. No changes from the §403 risk analysis in the pattern of the yard-wide soil-lead
concentration distributions across  Census regions were observed, with the Northeast continuing
to be associated with somewhat higher concentrations compared to the others (although the
ranges of observed soil-lead concentrations are comparable across all Census regions).

      Boxplots associated with the data portrayed in Table 3-20 are found in Figure 3-14.
Appendix D2 contains tabular summaries and boxplots after excluding imputed data values.
                                          105

-------
Table 3-19.   Descriptive Statistics of Yard-Wide Average Soil-Lead Concentration for
                Households,  Presented bv Housing Aoe Category. As Reported in the §403
                Risk Analysis Versus the Interim NSLAH Data
Study
How Not-
Detected
.---and ,-
Negative -
Data were
Handled
- , VX«rd-Wi*
-> #
Surveyed
Units with
Positive
Averages
Aritn^
fMtlC
Mean ,
Geo-
metric
Mean1


- Gib* ~
metric
SttL
Dev.z
Muiunuin
i- ^
Nicentration (pg/g)1
25*-^
Pwccn-
fle
Median.
™
••s
, A " •$••

'•" >B*<
Percan-
tite .
* "****

Maximum
Units BuiK Prior to 1940
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH*
No
adjustment
Replaced by
LOD/2
77
114
114
761
646
646
463
297
297
3.09
3.56
3.56
Units Built from 1940
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
87
145 -
145
287
264
264
92.6
112
114
3.15
3.43
3.33
17.4
12.8
10.8
-1959
5.40
1.65
4.62
259
135
135

44.3
45.2
45.2
569
294
294

77.3
110
110
1030
711
711

162
273
273
4620
9270
9270

7030
4340
4340
Units Built from 1960-1977 (1960 • 1979 for the §403 risk analysis)
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
120
198
201
55.0
76.7
77.2
Units Built After
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
28
168
180
31.3
27.4
28.2
32.8
31.8
33.3
1977(afte
22.4
15.7
16.2
2.56
3.65
3.24
4.63
0.00
4.83
19.7
14.0
14.7
29.7
29.4
29.4
61.6
58.3
58.3
996
1120
1120
r 1979 for the S403 risk analysis)
2.31
3.19
2.65
5.35
0.00
4.65
13.6
6.07
6.34
21.2
16.0
14.9
45.0
28.7
28.7
97.4
474
475
NSLAH Units with Unspecified Year-Buitt Indicator
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
64
66
175
175
72.9
68.9
4.15
4.13
0.00
4.74
22.3
22.4
63.8
64.4
211
211
2290
2290
1  All statistics are calculated by weighting each household by its sampling weight.
2  Only household averages greater than zero are used to calculate this value (data for all units with soil-lead data are used to
calculate the remaining statistics).
3  Summaries include imputed data for households having no soil-lead concentration data. The method for imputation is
presented in Appendix C.   .
Note: The yard-wide average for a household is the average of the following two statistics: 1) the average of the mid-yard
sample results, and 2) the average of results for the dripline and entryway samples.
                                                 106

-------









• M • ••^^— -J ml ""1 1
























t
"
|^
s||
m{\
-
1 ^
" «• 5
-If1
2*
-
XlJ
- Jo'
.1 1
S '
?!>""
-Is
-

£
l'
s s
»
•r



-


if
*I
9
~ I a
* &
j
-Is
" |s
OQOOO^*;^
82"
(udd) uoyoJiwauoa pD»i-|]<>5

2
*-
C
TO o
°—.
(J MB^ (Q
|a|
^^ «^» ^^
A 60 ^_
S^3
*J
S CO O
Tl "3 ®
B-^§
£ c ^
*5^ c
2 M g>
°ffi^
^ M T3
gee
CD <0>£
m
S
£
O)














8'
1
&
S
o
1
O)

c
"3
c
3
A
(9
1
•o
0)
o.
•^
•o
"1
1
1
m

-------
Table 3-20.   Descriptive Statistics of Yard-Wide Average Soil-lead Concentration for
                Households,  Presented  by Census Region. As Reported in the §403 Risk
                Analysis Versus the Interim NSLAH Data
Study
How Not*
Detected
and
Negative
Data were
Handled
' ~ Yard-Widf
,#
Surveyed
Units with
..Positive
Averages-
Arith-
~ metic
Mean
, Geo-
nvotnc
Mean1


Geo-
inotiic
Std.
Dev.2
Mtfitiraim
mcentration (pg/g)1 " • • : ," ', ''
' :25*
Percen-
tie
Itttn irtnn
IVI6Ctt3n
"75*
r0fC6Jl-
tile
- '""•"" "".

Northoast
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
53
109
109
437
423
423
206
160
162
3.58
4.24
4.16
14.8
3.92
6.24
60.1
52.3
52.9
279
176
176
569
396
396
4320
3460
3460
Midwest
§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH*
No
adjustment
Replaced by
LOD/2

§403 Risk Analysis
(HUO Natl. Survey)
Interim
NSLAH*
No
adjustment
Replaced by
LOD/2

§403 Risk Analysis
(HUD Natl. Survey)
Interim
NSLAH3
No
adjustment
Replaced by
LOD/2
73
149
150

134
258
265

52
173
182
404
220
220
•
125
162
163

112
68.2
69.0
81.4
65.5
65.8

44.5
37.3
36.4

34.4
30.5
31.7
6.33
4.97
4.71
South
2.94
4.62
4.38
West
3.92
4.36
3.55
4.63
0.00
4.90

5.22
0.00
4.65

4.79
0.00
4.62
19.7
22.1
22.1

22.6
11.9
13.1

14.2
12.5
12.8
51.6
63.2
63.2

40.8
27.6
27.9

27.2
29.4
29.4
264
206
206

79.3
79.2
79.2

61.6
77.5
79.3
2750
7070
7070

7030
9270
9270

2020
776
776
'  All statistics are calculated by weighting each household by its sampling weight.
2  Only household averages greater than zero are used to calculate this value (data for all units with soil-lead data are used to
calculate the remaining statistics).
3  Summaries include imputed data for households having no soil-lead concentration data. The method for imputation is
presented in Appendix C.
Note: The yard-wide average for a household is the average of the following two statistics: 1) the average of the mid-yard
sample results, and 2) the average of results for the dripline and entryway samples.
                                                 108

-------

age
atio
                                J°
                                <0
                                 ® =
                                lg"3T
                                >.£ 3
                                •633
                                 fe^-g
                                >.2^   «
                                "o §.S   -g
                                   ••• 7*   w

                                111   1
S
s
3
v.
O
s
1

I

I

A
OJ
                          Sff

                         -
I       §       s


   (Uldd) UOl|OJ|l«K>U03 PM1-1WS
Boxpl
Risk A
not-de
                                CO

                                 I
                                 en
a.



|

73
c
CQ

I
109

-------
Comparisons bv combination of housing age and Census region

       Tables 3-21a and 3-21b present descriptive statistics for yard-wide average soil-lead
concentration according to the 16 combinations of Census region and housing age category.
Table 3-2la reflects the data when no adjustment to not-detected results, while not-detected
results are replaced by one-half of the detection limit prior to performing the summaries in Table
3-2 Ib. As the central tendency of the soil-lead concentrations was of primary interest to compare
across the different combinations, these tables only contain estimates of the arithmetic and
geometric means, geometric standard deviation, and median.

       Due to the small numbers of housing units entering into each summary within Tables
3-2la and 3-2Ib, caution must be taken when making inferences from the results portrayed in
these tables. Appendix D2 contains these tabular summaries after excluding imputed data
values.

       3.2.2.2  Data Summaries for the §403 Risk Analysis Versus Other Studies.  This
subsection presents data summaries for the 22 studies in Table 3-17 that characterized soil-lead
concentrations in urban areas and how these summaries compare to that for the HUD National
Survey and to the distribution of yard-wide average soil-lead concentration characterized in the
§403 risk analysis.  The soil-lead concentration parameters that are summarized in this
subsection were specified for each study in Table 3^17.

       The 22 studies whose data are considered in this subsection include the three recent
studies included in the dust-lead data summaries in Section 3.2.1:  Baltimore R&M study (pre-
intervention), Rochester Lead-in-Dust study, and HUD Grantees evaluation (pre-intervention
data available through 1/99).  Figure  3-15 contains boxplots of household average soil-lead
concentration for these three studies and the HUD National Survey ("HUDNS")- These boxplots
represent yard-wide averages in all cases except the Baltimore R&M study, where only dripline
soil samples were collected. Separate boxplots are included for each grantee in the HUD
Grantees evaluation10.

       As in Figures 3-7 through 3-10, the left-most three boxplots in Figure 3-15 represent
yard-wide average soil-lead concentration data from the HUD National Survey:

       •      "HUDNS (U)" summarizes the data without regard to sampling weights

             "HUDNS (403)" summarizes the data as used in the §403 risk analysis (e.g., using
             sampling weights reflecting the 1997 housing stock; incorporating imputed data
             assigned to housing units with missing data)
       10 "Alam"=Alameda County; "Balt"=Baltimore; "Bos"=Boston; "CA"=Califomia; "Cle"=Cleveland;
"MA"=Massachusetts; "MN"=Minnesota; "NJ"=New Jersey; "Rl"=Rhode Island; "Wl"=Wisconsin; "Miiw"=Milwaukee;
"Chic"=Chicago; "NYC"=New York City; "VT"=Vennont.

                                          110

-------
Table 3-21 a. Descriptive Statistics of Yard-Wide Average Soil-Lead Concentrations for
               Households,  Presented  bv Housing Age and Census Region. As Reported in
               the §403 Risk Analysis Versus the Interim NSLAH Data Where No
               Adjustments Were Made to Not-Detected Results
Census
Region
Northeast
Midwest
South
West
Study1
f4O3 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
1403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
$403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Housing Age
Category ' ,
Prior to 1 940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
Prior to 1940

1940- 1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1940

1940-1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Yard-Wide Averagi
# Surveyed
Units
26
41
17
23
10
21
16
19
36
21
36
29
37
4
29
19
26
33
48
64
79
18
81
13
11
16
38
17
61
6
42
Arithmetic
Mean
542
877
573
290
79.1
132
57.8
1310
498
127
236
42.7
93.8
13.0
34.1
417
684
327
364
54.6
68.7
38.5
22.2
594
155
96.8
143
56.2
47.4
21.7
17.3

> son-Lead concentration ty
Geometric
Mean3
491
499
136
199
60.7
65.5
40.0
941
224
92.6
110
27.1
38.3
11.5
12.9
174
278
83.1
96.6
36.5
26.9
29.7
15.7
295
122
72.1
86.9
23.8
24.7
15.0
10.6
Geometric
Std.Dev.3
1.57
3.22
4.40
2.24
2.15
2.95
2.63
2.68
3.34
2.41
3.14
2.32
3.34
1.66
3.92
3.68
3.74
3.27
4.40
2.30
3.60
2.11
2.45
3.76
2.23
2.19
3.08
3.02
3.81
2.34
3.54
*/S»
Median
444
S69
60.1
273
69.7
50.9
38.8
1390
238
123
82.0
23.4
34.6
12.4
9.36
159
186
81.0
77.9
34.7
26.1
25.0
15.0
394
158
60.4
90.3
20.0
26.9
13.6
9.53
' All statistics are calculated by weighting each household by its sampling weight.
2 Summaries include imputed data for households having no soil-lead concentration data. The method for imputation is
presented in Appendix C.
3 Only household averages greater than zero are used to calculate this value (data for all units with soil-lead data are used to
calculate the remaining statistics).
Note: The yard-wide average for a household is the average of the following two statistics: 1) the average of the mid-yard
sample results, and 2) the average of results for the dripline and entryway samples.
                                                111

-------
Table 3-21 b. Descriptive Statistics of Yard-Wide Average Soit-Lead Concentrations for
               Households,  Presented bv Housing Age and Census Region. As Reported in
               the §403 Risk Analysis Versus the Interim NSLAH Data Where Not-Detected
               Results Were Replaced bv LOD/2
Census
Region
Northeast
Midwest
South
West
Study2 ;
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
5403 Risk Anat.
Interim NSLAH
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
§403 Risk Anal.
Interim NSLAH
Category
Prior to 1940
1940-1959
1960-1977
(1960-79 for §403)
After 1977
Prior to 1940
1940-1959

1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1 940

1940-1959
1960-1977
(1960-79 for §403)
After 1977
(1979 for §403)
Prior to 1 940

1940- 1959
1960-1977
(1960-79 for §403)
After 1977
1 1979 for §403)
. Yard-Wife Average
# Surveyed
Units •
26
41
17
23
10
21
16
19
36
21
36
29
37
4
30
19-
26
33
48
64
81
18
84
13
11
16
38
17
62
6
50
* Arithmetic*
Mean
542
877
573
290
79.1
132
58.1
1310
498
127
236
42.7
94.1
13.0
34.7
417
684
327
364
54.6
69.4
38.5
22.7
594
155
96.8
143
56.2
48.0
21.7
18.9

i dOtt-Leaa voncentrauon y.
.** -_. .- -
'Mean "
491
497
136
199
60.7
65.4
42.0
941
224
92.6
111
27.1
39.0
11.5
14.0
174
278
83.1
97.7
36.5
27.8
29.7
15.4
295
122
72.1
89.8
23.8
27.8
15.0
12.1
Geometric
Std. Dev.
1.57
3.26
4.40
2.24
2.15
2.96
2.36
2.68
3.34
2.41
3.11
2.32
3.27
1.66
3.06
3.68
3.74
3.27
4.34
2.30
3.24
2.11
2.29
3.76
2.21
2.19
2.78
3.02
2.91
2.34
2.42
»3/9»
it *
Median -
444
569
60.1
273
69.7
50.9
38.8
1390
238
123
82.0
23.4
34.6
12.4
9.67
159
186
81.0
77.9
34.7
26.1
25.O
14.7
394
158
60.4
90.3
20.0
26.9
13.6
11.2
1  All statistics are calculated by weighting each household by its sampling weight.
1  Summaries include imputed data for households having no soil-lead concentration data. The method for imputation is
presented in Appendix C.
Note: The yard-wide average for a household is the average of the following two statistics: 1) the average of the mid-yard
sample results, and 2) the average of results for the dripline and entrvway samples.
                                              112

-------
ICD»
  !l
  s
-I
                                                                   9 U
                                                                   (0  45
-1!

  a
  a
  j
  1
 (uidd) uoj|3ui>3 po«~l— Ij
                113

-------
       •     "HUDNS (OW)" summarizes the data weighted according to the original weights
             assigned in the survey.

As soil samples were sieved into tine and coarse fractions in the Rochester study, Figure 3-IS
includes two boxplots for the Rochester soil-lead data. The boxplot labeled "Rochester (FINE)"
summarizes household average soil-lead concentration considering only the fine-sieved traction
of the collected soil samples. The boxplot labeled "Rochester (TOTAL)" summarizes estimated
household average soil-lead concentration assuming the total soil sample was analyzed. Total
soil-lead concentration for each sample was estimated as the average of the reported
concentrations for the fine and coarse fractions of the sample, with the coarse fraction result
weighted three times that of die fine sample result (see Appendix E for the derivation of this
estimate using data from the Milwaukee study). Estimating soil-lead concentration in the total
soil sample was intended to allow soil-lead data from the Rochester study to be more comparable
to data from the other studies in which no sieve-fractions were calculated.

      Figure 3-15 shows that while the ranges of average soil-lead concentrations among the
study households tended to overlap from study to study, the distributions based upon the HUD
National Survey data (including the §403 risk analysis) tended to be shifted lower than for the
other studies.

      Figure 3-16 contains a graphical presentation of how the distribution of household
average soil-lead concentration in other selected studies listed in Table 3-17 compare with the
distributions based upon the HUD National Survey data (i.e., the same three distributions
portrayed in the boxplots labeled "HUDNS" in Figure 3-15). The studies selected for Figure
3-15 were among those in which an average soil-lead concentration for a particular area could be
determined. As only summary  statistics for many of the studies in Table 3-17 were available
from the references or prior literature reviews, boxplots like those in Figure 3-15 could not be
created for these other studies.  Instead, specific descriptive statistics (when cited in the
references) are plotted in Figure 3-16 for each study by using plotting symbols that indicate the
type of statistic. These statistics, with their plotting symbols following in parentheses, are the
minimum (MIN), 25th percentile (25th), median (50th), 75* percentile (75th), maximum (MAX),
and geometric mean (GM) soil-lead concentrations. In studies where the arithmetic mean is
specified instead of the geometric mean, the arithmetic mean (AVE) was plotted. The vertical
dashed line in Figure 3-16 separates results based on the HUD National Survey data from the
results for the other studies.

       Yard-wide average soil-lead concentration (or an average that is not specific to a given
location) were available or could be calculated within eight of these studies (one being the 3-
Cities study, which consisted of three sub-studies). Table 3-22a presents values of descriptive
statistics (e.g., geometric mean, minimum, maximum, selected percentiles) for yard-wide average
soil-lead concentration within these studies. This table also includes the estimated number of
averages represented in the descriptive statistics that exceed a given soil-lead concentration
threshold (400,1200,2000, and 5000 ug/g). The following features can be found within this
table:

                                           114

-------




X
2


X
2





X



X
2








X
2




X
2



X

2





i" • ' ' ' • • r ' • '
o o
§ §
o o
o —




£ £ £ z
m o m 5
I- in w 2


2
£ ° £ *
K £ 3 *
a
in


2
£ £ 2



£ £ £ 2
A o m 5
N m N •*
2
x £ « £ z
5 « *» tf» TB
2 is* £ M 2
in



£ a£ £ z
I^ lO (M *




£ * £ z
m r in ^
r» £ « *
o
m

S 3I S -

rv m «N





TJ
1
O
1 W
"1 I
° 1
CO
.III
o*? ffi

^5
CO
o
C X —^
Jr o
'11 1
n
e
c ,O
|1 S
"!| f >
*x ^ CO
o-j (88
" ™ i *% CD
5 K) ^^ C
CD » O
O -ff
CO a
•o*
% is
3® 0} «•»
z 3 a>
I ^
« E
0) o
""'«'«
' °S 3r ^

*. o
0 C
t, li

^ 5 •
CO CO
£ E
E E
3 3
CO CO
• • ^ ' 	 - 	 	 ' 	 • 
-------
	    V   Hfc

  ,-  «   5
=   5
    3
 :

 1
                                                              I!
             i
-II
          I  S§
                                                             • IS
                                                              u I
                                                                  o
                                                                  o
    (uJdd)suoi|DjtuaouoQ p031-1105
     
-------
O
o
o.

CO
 0


•o
 o
«*

 O)
 (0
 tu


N
n
 9
ft
I
"E J

|«
fl
|
*
£ -



"Z
1
S-
•s
5
I

1
*••


:
•s
j
S

'-•S
:l
"" f
:l
• -j?
!
=
4
1
"\
m
fl

I
m
*1

• Mlnlmiin

Oeometrii
Stc^iBIv
o
!|
; z
*f
If
'". *
#
O
*

CO

o
CM

ID
CN
O
_
IO
0
CD
ID
CM
IX
CD

O

O
rt
8
CM
a
5

|
o
o
0
ID

*
0

to
CO
IX
en
ID
en
o
CO
ID
to
*f
ID
m
co
CD
o
CO
£
o
(J
at
S

o
o
rx

*

O
o

en
o
t
to
o
Xf
O
to
00

CO

o
CN
O
CO
Big City/ Suburb

«*
CN
*

CD
•-
CN
S

1
co
ID
00
ID
O>
.
o
CO
CN
CM
ID
CN
to
O
CM
CN
Small City/ Metro

«*
o
o
*
CN

CM

*
ct>

to
^
*"•
eo
O
IO
CO
**
rx
CD
CD
m
,_
ID
CO
CN
il
4"
:
i
v
"
1
j
c
o
o
CO

*
IO
CO
t
rx

CM
8
CM
CM
ID
CN
CM
ID
|x
CM
CO

CO
to
ID
CO
ID
rx
ID
Non-Metro
| (unweighted) '
en
O
to
CM
to
rx
#
ID
CO

to
s
rx
CO
CN
CO
rx
O
CN
.to
CN
ID
n

IO
o>
2
•

g
o
£
CM
CO
CM
IO
CO

to
to
en
«
CO
to
ID
at
to
CM
CO

|X
CO
00
CN
CD
5
Non-City

o
o
CM
CO
ID
CO
CD
CM

00
to
co
CO
CM
ID
CO
CO
ID
CO
*~
rx
CO
ID
3
to
Northeast

o
o
IS
co
ID
en
rx
CN

rx
CM
co
CM
CD
en
8
|X
CM
CD

CO

O
ID
rx
to
CO
Midwest

o>
o
*
00

*
to

C*
CM
rx

IX
_
ID
|X
s
CO
CO
CM
CN
ID
IO
CO
CO
8
-
i

o
o
*
CM
ex
*

CO
to

O)
i
o
0)
T).
§
en
S
CO

q.

O
ID
IX
ID
W

CN
O
*
U>
CM
rx

CM
m

ID
CM
0
rx
CO
to
8
CD
CD
ID
CO
CM
CO

IO

tn
CO
CO
CN
(A
S

o
o
o
o
10
CM
cn
rx

eo
eo
co
co
ID
en
o
CO
*•
.3.
ID
CN

ID
CO
CO
en
Big City/ Metro

i
O
IX
0
£
o
CN
0

en
co
co
9-
f-
to
co
CO
CN
CO

00
CO
CM
CD
ID
to
rx
Big City/ Suburb

CO

*
IO
co
rx
••
#
en

ID
8
en
en
n
to
CO
00
ID
•*
CM
CN
to
ID
ID
en
rx
IX
o
ID
Small City/ Metro

o
o
•s
CO
00
co
en
*

00
co
en
CM
a»
CM
5
o
CN
r*.
CD
,t
CO
CM
ID
00
CN
p

o
o
*
to
<*
eo
CD
o
r*

CM
§
CN
0
ID
0
en
S
rx
rx
co

en

o>
§
ID
CO
0
O
i
1HUD National Survey
ID
0
IX
CN
?
CD
ID
CO

ID
S
rx
to
CO
CO
rx
S
CO
3
CN
ID
rx

co
eo
CO
CO
1
>
5
t
A
c*
o
o
eo
CN
*
co
to
*
ID
O5

CO
to
to
to
CN
en
Z
*
eo
CM
CO

CM

ID
S
CD
CO
Non-City

o
o
f
(O

rx
CM
to
en
5
CM
CO
rx
rx
CO
•—
CD
to
CD
ID
CO
ID
Northeast

o
O
en
CO
en
—
CM
*-

CN
to
rx
CN
IX
CD
CM
.-
5
T-
CN
CO

CO
CD
,j
to
eo
Midwest

*
O
CO
0
00
o
oo
CO

ID
s
^
rx
en
CN
§
^
ID
CN
CM
IO
en
CM
ID
"*
to
C:
1
CO

o
O
<*
O

t

CO
CO

en
o
CN
CO
en
IX
s
eo
CM
oo

en
eo
^
eo
CM
ID
I

                                        117

-------

9
* f
S g
It
£ -O
li
•5 |
ton
•
I ,;







'•: '';••••&)
•o
2-
1
. >
••H|
:>"




a
-- .:
•s

»
S
•f
o
f
%
1
' *
1
Maximum
„
!'

.:• S.
- x

if

•i

=* i
o
|s
I1

z
or Type of
Soil/cover
1
co
#
CO
O
*
CO
CO
CO
eo
ID
IO
(M
g
.-
IO
*
CO
cn
ID

n
rv
CM
co

ID

_
CO
rv


ID
CM
'c

o
o
O
O
co
CM
CD
CM
eo
I
IV
S
CO
g

eo
ID
eo
^
ID
CO
CO
o
IV


CD
Big City/ Metro

o
o
*

*

CO
^
o
CM
o
rv
CO
o
to

•*
o
eo
CO
*
CD
CO
_
CO
CO


CD
Big City/ Suburb

en

*
o
*
CO
£
CD
CM
ID
CM
g
CO
ID
S
CO
eo
CO

o
CO
CM
CM
IO
IO
CD
O>
CM


•*
Small City/ Metro

*
o
0
CD

*
CO
IO
eo

co
co
,j-
CO
a>
g

CD
CO
CM
|V
CD
CO
CO
is.
O
CO


CM
'j

O
o
*
10
#
o
*
CO
CM
CM
S
CM
ID

rv
CM
CO

,_
10
IO
rv
rv


ID
i
1 HUD National Survey
1 i~. :„:„„! ..._:„!.. .11
CO
O
*

CO
CO
*
cn
IO
CM
rv
eo
CM
CO
rv
rv

o
CM
CO
CM
ID
CO

CD
O
en


^
5
>
p
>
!
1
>
O
O
*
CO
#
IO
CD
CM
CO
eo
cn
IO
eo
CO
to
ID

cn
ID
CM
CO

CO
•*•
ID
o
IV


*
Non-City

o
o
IO
CO
*

IO
M
a
n
CO
i
CM
IO

CO
CO
IO
CO
*~
CO
eo
CM
o
ID


CO
Northeast

o
O
*
co
CO
o
CO
£
S
CM
ID
CM
eo
CM
CO
CO
CO
ID

rv
CM
CD

o
CO
eo
eo


to
Midwest

*
o>
o
m

co

*
CD
CO
to
CM
g
_
to
IV
CM
IO

CO
CD
CM
CM
IO
CO
CO
O>
rv
IO


^
1

O
O
*

r»
CM
O
CD
CO
8
CD
|v
CD
4.
D
to

CO
IO
IN
co

o
•«
eo
5



1

z

z

z

2

8
CD
o
•
o
CM
O
O
o

»
CM
O

<
z
<
z

CM
CM

(Baltimore Urban
Garden
a*
o
o
*
o
o
co
CM
rv
cn
IO

CD
CO
S
£
z
o
S
o
rv
g
O
CO
o
o
CO
CM
O
CO
co
CO
S
iv
CO

en
8
CM
.
0
§
1
9
eo
i
o

a?
rv

5

CO
CO
cn
§
^
CM
o>
3


*
i
o
ss

O
o
*
o
o
g
o
0
o
CO
CO
to
eo
CD
M

CO
rv
rv
ID
rv
ID

rv
ID


CO
01
M
«g
CM
A
1 Cincinnati 3-City
o
o
o
o
o
o
o
IO
CM
CM
.-
I
._
CM

CD
cn
CO
CD
IO
CO
CM
CD
§


*
CO
a

Cleveland

o
o
Jo
IV
*
"
rv
0
rv
!
10
8
ID
CD

IO
cn
CO
eo
ID
o>
^
CM
V-



*
Minnesota

#
IO
rv
ID
3
IO
s
to
IV
CO
ID
CO
IO
to
o
IO
i
CO
CM

IO
8
IO
IO
eo
eo
eo
O
§
"

*
t>
i
a

O
O
«
IV
co
ID
t
*
CM
to
CO
O
rv
CD
CO
CD
CO

O
CO
CO
O
O
co
**
CO
CO


CO
| Wisconsin

118

-------

O
O
CM
C>l

m
o
Z
a
                 s

                       g
                       s
                           I
                              1
soil
-2'S
                                                        a
                                                               1-s
                                                         S 5 5
             c «
            < 5
                                              Z   — CM   CO •» 10 CO


                                                    119

-------
       *      For the HUD National Survey data, which are associated with sampling weights
              from the original survey and revised sampling weights for the §403 risk analyses,
              all results in Table 3-22a are portrayed three times: under each of these two sets
              of weights as well as without regard to weights.

       •      For the HUD National Survey data, which are associated with sampling weights
              from the original survey and revised sampling weights for the §403 risk analyses,
              all results in Table 3-22a are portrayed three times: under each of these two sets
              of weights as well as without regard to weights.

       •      In addition to summarizing results across all housing units or samples in a study,
              results for selected studies are also summarized for specific subsets of housing
              units, soil types, or soil samples. In particular, HUD National Survey results are
              portrayed according to urbanicity and Census region, results from the HUD
              Grantees evaluation are portrayed by grantee, and the Rochester study results are
              portrayed for the fine soil fraction as well as for total soil.

Refer to Table 3-17 to verify the types of results being summarized in Table 3-22a (i.e., housing
unit averages versus averages for single analytical samples).

       Table 3-22b contains the same descriptive statistics as those portrayed in Table 3-22a, but
they represent average soil-lead concentration for specific locations, such as dripline, play areas,
remote areas, geographical areas, and other locations that were considered within the individual
studies. As in Table 3-22a, the statistics in Table 3-22b are given over the entire study, as well as
for specified sets of units that are determined by urbanicity and other factors.

Summary statistics by housing age category

       As housing age category is generally regarded as an important influence on soil-lead
concentrations, the above summaries are also presented according to the housing age categories
considered in the HUD National Survey (pre-1940,1940-1959,1960-1979, post-1979). Figure
3-17 presents boxplots for pre-1980 housing data from  the HUD National Survey and the
Rochester Lead-in-Dust study (total soil and fine soil), non-control houses in the Baltimore R&M
study, and pre-1978 data from the HUD Grantees evaluation (data combined across grantees).
As all non-control units in the Baltimore R&M study were built prior to 1941, the only boxplot
for this study in Figure 3-17 appears in the "pre-1940" category. Caution must be taken when
interpreting results in Figure 3-17 for the Rochester study, as the actual age of certain houses may
be older than what was specified in the Rochester study database (see Section 3.3.1.3 of the §403
risk analysis report).

       Many of the other studies listed in Table 3-17 did not have information readily available
on housing age.  Thus, no corresponding figure portraying distributions according to housing age
was prepared to represent these other studies.
                                           120

-------
Cfl
0
O

£


3



1


a

S





I

o
«5
o
0
a
(0


c
o
O
o
W
  w
a
11
o ^

.2
*•
a
•c a
o

S s
^

M
 I

eo
           i

           i
          if
           i
           £
          P
          li
          i
            K
                   i
                             I
                                               §
                                                         I
                                         S
                                     u>
                                           si
                                                 fi
                                                       a>

                                                       S
                                         0$
                                         to
                                       I
                                       1
                                         o

                                         S*
                                         ca
•§

-------
      ii
      01
              I
                 I
                                10
                                                               g
                                                     cvi
                                                                          5?
                                                                          m
                                                 *
                                                               S
§
CM

CO
 o

1
                                                                             «0

                                                                             o
                                                                                        u>
                                                                                        s
                                                                    u>
                                                                    v»
                                                                    
        5
        I
                                             CO
                                             ui
              I

                                                                  s
                                      o
                                                                                            X
                                                                                            I
                                                                                                   o
                                                                                                         3
                                                                  s
                                                                                        111
ZS--5
Q e s

X«^Z
                                                                                       0.

                                                                                       5
                                                                                                              i
                                                                                                              3
                                                              122

-------


8
(N
CO
i!
i
      §
         Is'
             g
             g

             S
             8
       g g
         g

           8

                       CM
                       si

I g
                              CM
                              3

                              »

                               S
                                     s
                                     
-------
      Si


     Ii
      oj


                                        §
                                               g
                                           i
                                                       §
                                                     I
                                                                     I
                                                                     s
                   §
g
£
8
                   s
8
CSJ

m

£

•§
                                                                           to
                                                                           s
8
                                                              00
                                                              ^~

                                                              CM
                                                                           i
                                               i
S
                         oi
   ii
                                                     a
                                                                               •
                                                                                      §>
                                       »


                                                                        tL
                                                  ¥
                                                  Q
                                                                              g.
                                                                                      CO
                                                                                 »
                               =«
                               To oco
                                                          I
                                        Ii
                                                 124

-------

|

J3
X
n
 «o

c
o
I
to
£
Yard
t of H
      i

                                                                                                     i
                                                     z
                                                                            z
                                                                                          z
          g
                                    1
                                                                                          i
       g
                                                                                   S
                                                                                                         IM
                                                                                                         S


                            i
                      p
                      ui
                                5;
                                CM
                                a
             n
                                §
           S
           1
                                                           §
                                                                    1
3
                                                                                         o
                                                                                         1
                                            S
CM
e>
s
                                                                                                            CM
<9>
S
                                              S
                      S
                             ui
                             5
                                                         CM
                                                   i


                 N
                                CM
                                           S
                         o
                         CM
                         CM
                                                                    si
                                   3
•s
S.
                 S
                                         I
                                         ts
                                         i
                                i
et side
                                                            01
                                                                    a
                                                                    S
                                                                    CO
near
                          1
                          S
the stre
                                                                           1
                   ca



                   I
                                                                                              S
ndation
Area

                                                          O
                                |

                                1

                                I

                                I
                                Q.
                                                       C
                                                                        I§
                                                                        zo
                                                                     u
                                                                     i
                                                                    o
                                                                    i
                                                                                                                          II
                                                       125

-------
       1
   II
   i-
CM
rs|

n

e

3
a
     ;
       fj
                I
                  1
1
               g.


               1
     L
     If
     ii

                       iff
                       II!
    .   .
    SgS

    SSS

    ggg
                                  I
                                  .



                                  I


                                  £

                                  I
                                  2

                                  i
                                 ll
-, |i
                              a
  ^a> 5



  If!
  =5>c
                         o

                         1
  1:
                          8888
                               126

-------
                                                o
                                                o
                                                0
                                                     o>
             o
             ra


             I
          000
          000
       o o
     o o     001



     o o  o
   «(
0  0000
      o oo
      O fnt
                                              — O)
                                              II
                                              ii
 S
 il



 ill
                                        -I!
 «!
hH
O E
^O


fi
      2-0
                                              8"
                                              § »
                                              P 3
                                              _l "O
                                               • «
                                              — o
                                              «
                                              0 -5
                                              > o
                                              < oc
                                                (0
                                              0 m 0>
                                              « 2 IS.



                                              til
             o

             I
             o
             £>
       o

       1
       O
       to
       1
       m

       i
_0
0>
JO
0

«

"S

i
.2

i
ffl


I

•R
                                                      -
                                                     £ a
             1 °
             i&
             5 •
             a|
             ii
              0)
                                                     5-5
                                                     s o
                                              o>
                        127
                                                     o 5
                                                     5..i

-------
       The summary statistics found in Tables 3-22a and 3-22b were calculated according to
housing age category for relevant studies. These summaries are found in Tables 3-23a (for yard-
wide average soil-lead concentration) and 3-23b (for average soil-lead concentration for specific
locations). Note that these tables also include summary statistics for housing units built after
1979 (although the Rochester study units may not have actually been built in this time period, as
mentioned in Section 3.2.1.2). The post-1979 results labeled as "HUD National Survey (§403
RA)" represent surveyed homes built from 1960-1979 that contain no lead-based paint (Section
3.3.1.5 of the §403 risk analysis report).

       3.2.2.3  Calculating National Exceedance Percentages for Yardwide Average Soil-
Lead Concentration. The soil-lead data summaries presented above suggest that the distribution
of measured soil-lead concentrations as reported for the HUD National Survey are reasonably
consistent with the distributions suggested by other studies, including the interim NSLAH data.
Thus, these two national surveys are expected to generate similar national distributions for
yardwide average soil-lead concentration, from which the estimated percentages of housing units
whose yardwide average soil-lead concentrations exceed specified thresholds ("exceedance
percentages") could be calculated. These percentages give some indication of the frequency with
which intervention activities might be prompted by regulations that target alleviating soil-lead
exposure. Soil abatement practices are often recommended both within the literature and by the
HUD "Guidelines for the Evaluation and Control of Lead-Based Paint Hazards in Housing"
(USHUD, 1995b; pages 12-47 to 12-56).

       The methods detailed in  Section 3.2.1.3, which were used to fit lognormal distributions to
household average floor dust-lead loadings based on data from the two national surveys, were
also used to fit lognormal distributions to yardwide average soil-lead concentration data from
these two surveys.  As discussed in Section 3.2.1.3, the key objective to fitting the lognormal
distribution was to use the distribution to estimate exceedance percentages for specified soil-lead
concentration thresholds. Therefore, in order to ensure that the upper tail of the distribution was
as accurately portrayed as possible within the fitted distribution, this method treated a certain
percentage of the lowest data values as censored data when fitting the distribution. In this
exercise, four thresholds were of interest for yardwide average soil-lead concentration: 400,
1200,2000, and 5000 ppm.

       Figure 3-18 contains plots of the fitted lognormal distributions (superimposed on bar
charts of the observed data) and the estimated exceedance probabilities corresponding to these
distributions, for residential yard-wide average soil-lead concentrations, based on the HUD
National Survey data (top plot) and the interim NSLAH data (bottom plot). Recall that the
sampling weights corresponding to the HUD National Survey data were revised in the §403 risk
analysis to reflect the 1997 national housing stock. The same soil-lead concentration (horizontal)
axis is used for both plots, so that the two plots can be directly compared.  The similarity of the
two distributions is noted in this plot, as the fitted distributions are nearly the same shape and
cover approximately the same ranges of data. Furthermore, the estimated exceedance
percentages for a given threshold differ by less than one percentage point between the two
                                           128

-------
a

*«

o


•a

a



I
«*
CO

o
*<

a

=5 >>
5 -a
O 3
O &
O CO

< a
5
 o
   «
   •o
   o
   CO

   o
   §


   I
   CO
   o
•X a
0 co
 a

to
II
(N

CO
I
•8 Al


1!
11
II

           i
           is
               §
                     g
                                               I
                                         I
                                     £
                                     CM
                                                               I
                                                               n
                                                         iri
                                                         OJ
       £
       *
                                     3S
                                                 ffl
                                                       2
                                                       s
                                                  I
                     n
                                     Ol

                                     s
                                8
                                                  iri
                                                      s
                                                    S
CO




t
                                     i
                                                      in

                                                      S
           I


           0
s-
u
                                  O
                                   f

                                   *
                                   O
                                                    I

                                                    &

                                                    O
                                                        o
                                         18
                         z
                         aw 5

                         g  I
                                                       I-
                                                       DC CD
                                   129

-------
i*
I*.
I
 »
CO
cs

«
  1
    !i
   o
   .•E.
   o :

   = ,
I
                                                     II

            *
            8
                                       10
                                       CM
                                                          §
                    5
                       S
                                                   «
                                                   in
                                                         S
                                         A
                                         S
                            S
                                    o
                                    d
                          I
                                           ui
                                            u>

                                                       S
                                                     O
                                                       o
                         o
                                                          u
                       i
                 —^
              PI
              In
              330)
              rwz
                           130

-------
eo
o
       I
            1
            I
     S&
     o
     I
                                        SS
                                        q
                                                                   § I
                                                           §
                                                  g
                                      S
                             g
                                                                        g
                                                                                     u>

                                                                                     $
                                 n
                                 ri
                                       in
                                       ui
                                                                •a.
«

a
(D

oi
2
in
n
K
r»
                                                             §
                                        o>
                                        S
                                                            (D

                                                            S
                                                                03
                                                                T-
                                                                o>
                                                                     ri
                                                    1
                                            I
                                   O
                                O
                   u
                     U

                                                          o
                                        o
                                               I
                                               (0
          IS
          1s*
          fl?
          S*

                                                                                     I
                                                II
                                              131

-------
rd-Wld
     0
a
«
n
o

Subs
Type of


             i
i
                  g
                  CM
                  N
                     CM
                     
Ci

g
                    3
                    S
                                      g
                                      O
                         S
                          5
                         S
                           g
                                           in
g
                             g
                                             ui
            g  g
g
            g
                                5
st
               g
               co-
                                                   3

                 3

                                       g
g

                                       3
            g
                                                          8
                                          ri
                         i
Ci
                                               S
                                                               oJ
                                  Ss-<
                                  c.?1

                                                                  CM
                      g g
g
                    g
                                                                    iri
                                                                    CM
               g
                         g  g
g
                            g
                              i
                                                                            

8 S i? So fflQ^ c: 3 ^? III III 132


-------
e.
o
o
a
m
CM
e

1

           1
          -I
          il
              I
                §
                s

                     IU
ui
          §
                                  §
               tvi
                                         s
                                         to
                               75"


                               o
                                             S
                                             in

                                             10

                                             5
                    ^

                    Ifi
                    x P n
                                            133
                            If:

                            at
                            *

                            5
                                                    u>
                                                    in
                                                   I
                                                   s

-------
CO

•MM



I

 o
a
«

S
<
 o
CO


   •g
o
CO
> •=
< -5
£ «
.2 3
« o
   —
.S Q

a
•2

O CO

   «
o
_

•a

               i

              II
                          s
                             n
                                       s
                                       8
                                       o
                                          u
                               x  3.
§
                                                          S?
                                                            $
                                                               ri
                                                                  I
                                           134

-------
           I
     I!
     o
     s,


     I
     s.
o
o





                                                                                     w
                                                                                     
                                                          o

                                                          s
                                                                              «

                                                                              .
                                                     s
                                                     0
                                                       I
                                               3

                                               t
                                               2

                                               I
                                                                              a


                                                                              Q
                                            il
Si
X
                                                                      I
                                                               o,
                          iw:
                                               135

-------
XI
«
M


                                    £
                                     c>
                                                    &
                                              OJ
                                                      §
                                                csi
                                                           *
                                               z
                                               Q
g
                                                                              S
                                                                      00"
                                                                      S= «"
                                                                      x wz
                                          136

-------
o
o
m
«
•s
I
              s
               3



               I

               S
                     g
                       15
                             g
                                     I
                           JK "*




                           II
                                         i
                                            cti
                                         •2
                                         137
                                                            U.S. EPA Headquarters Library

                                                                   Mail code 3201

                                                            1200 Pennsylvania Avenue NW

-------
    li
    S;

o
o

o
21
a
         I
         I
        II
           s
                              *
                              g
                                          t
                                              «
                                              s
               05
               t»
                   u>
                   oi
                              I
o
                                J
o
                     *
                     o
t
O
*
O

i
                 8
                 *
                 o
a-
o
           ?!
           if

          75Q^
                                                       s
                                                       I
                                                       5

                                                       2
                                                       S
                           i
         w
         I
                     Q
                     i-
          O
          3
          r
   JS

   I

   3
                                     138

-------
CM
n
  S3
  B
  &
    I
       §
       i
                   cvi
                       O
                  S
                            i
                            t
                               g
                                 u>
    
    =>
    oil  I

    I'S  1
                                        1- -.  I-
                                       a- 2 fr  -3
                                       i>li  I
                                       [1:
        i

   it!  !
   £:§>€  =

   fit  I
   iHt  1
   la*  ^
   If*  t
   III  t
   l|r  i
   '§ = 3  S
   z«pz  2

   sis  I
   E £x  a
   3^3  I
                                       i = |  «
                                       HI  1
                                       533  g
                                       111  1
                                       ii

                                       1l
     ^*J
     1-i
     II;
   § i
   1?
                                      ss
                                         •§°i
                                         * g-.s
r=


S

tl

                                   139

-------
                                          03
          O
          w
          o

          E
          a

          O
                   a
                   o
                   EC
                                          0>
                                          .2 c
                                          S= O
                            si
                               c
                               O
                               O
                              O

                              •o
                           P?  CO
                               CD
                                          CO
                                          111
E'O     Z'Q     i'O
       Aouanbsjj aAjjeisy
                         0-0
140


-------
surveys. Each estimated exceedance percentage is accompanied by an approximate 95%
confidence interval on the number of homes in the U.S. housing stock that exceeds the threshold
(given in millions).

       In Figure 3-18, the distribution based on the HUD National Survey data used in the §403
risk analysis was determined by censoring data values below 24.5 ppm (i.e., the bottom 30
percent of the data, taking into account the sample weights). The distribution based on the
interim NSLAH data was determined by censoring data values below 2.01 ppm, which
corresponds to the bottom 5% of the observed weighted distribution, including negative values.

       Among these four thresholds, the estimated percentage of residences that exceed the
threshold vary widely. For a threshold of 2000 ppm, the estimated percentage is 1.6% to 1.7%
for the two surveys, while the percentage increases to from 11.1% to 11.8% for the two surveys
when the threshold is lowered to 400 ppm.

       For both surveys, the estimated exceedance percentages specified within Figure 3-18 for
yardwide average soil-lead concentration, based on the fitted lognonnal distribution, are also
included within Table 3-24 (columns 2 and 4) for the same four thresholds. Also included in
Table 3-24 (columns 3 and 5) are estimated exceedance percentages that were determined solely
by the proportion of total sampling weights in the survey that corresponded to surveyed units
whose household average floor dust-lead loadings exceeded the given threshold (i.e., information
from the bar charts within Figure 3-18).  The two types of estimates are very similar for the
interim NSLAH data except at the highest threshold, while for the HUD National Survey data,
differences between the estimates increase as the threshold decreases.  It should be noted that the
lognormal-based estimates for the exceedance percentages (which were also portrayed in Figure
3-18) should be used when making inferences on die nation's housing stock.

Table 3-24.   Estimated Percentages of 1997 U.S. Housing Exceeding Specified
              Thresholds of Yardwide Average Soil-Lead Concentration
Soil-Lead
Cone.
Threshold
tpprn)
400
1200
2000
5000
§403 Risk Analysis -Based on Data from the
HUD National Survey (n=284)
Based on the Fitted
Lognormal Distribution
(i.e., the curve in
Figure 3-18)
11.8%
3.4%
1.7%
0.4%
Based on the
Weighted Observed
Data (i.e., the bar
chart in Figure 3-18)
13.2%
4.7%
2.5%
0.2%
Data from the Interim NSLAH (n=706)
Based on the Fitted
h Lognormal
Distribution (i.e., the
curve in Figure 3-18)
11.1%
3.2%
1.6%
0.4%
:r: Based on the
'Weighted .Observed
Data (i.e., the bar
chart in Figure 3-18)
11.2%
2.9%
1.7%
0.1%
Note: Data are imputed for those surveyed units with missing data prior to calculating the above statistics (34 observations
in the HUD National Survey had either dripline or remote soil-lead concentration imputed prior to calculating a yardwide
average; 42 observations in the interim NSLAH had an imputed yardwide average). The estimates based on the weighted
observed data are simple weighted percentiles that do not originate from a fitted distribution.
                                           141

-------
       It was also desired to calculate exceedance percentages for only urban residences within
the U.S. housing stock, as urban soil has the potential for being more likely to be contaminated
by lead than non-urban soil (in the absence of a particular lead source). Thus, the procedure used
to fit the distributions in Figure 3-18 was also applied to the HUD National Survey data for only
the 146 surveyed homes labeled as being located in urban areas. (The interim NSLAH data were
not included in this exercise because homes were not characterized by urbanicity.)

       Figure 3-19 plots the distribution and documents the exceedance percentages for urban
residential soil-lead concentrations as estimated using the HUD National Survey data. Based on
the fitted lognormal distribution, this figure indicates that approximately 2.8 percent of the
roughly 40 million homes hi urban areas are estimated to exceed a yardwide average soil-lead
concentration of 2000 ppm11. This corresponds to approximately 1.1 million homes.  However,
because the sampling weights in the HUD National Survey (and revised in the §403 risk analysis)
were not necessarily determined to ensure that the weights assigned to the homes in urban areas
would be representative of the entire urban housing stock, caution must be taken in making
inferences on the national urban housing stock based on these estimates.

       3.2.2.4  Interpreting the Observed Differences w'rth Other Studies.  Contrasting the
measured soil-lead concentrations from one study to another is complicated by differences in
study designs, sampling locations, and sampling and laboratory protocols and practices used by
these studies.  As areal patterns in the lead concentration of residential soil have long been
recognized, different locations within the same yard can have widely different soil-lead
concentrations. For example, levels along the foundation of the residence are typically highest,
reflecting the presence of deteriorated lead-based paint formerly on the residence or deposited
leaded gasoline emissions washed off the roof.  Also, distinct sampling protocols may impact the
amount of lead measured in a collected sample. The Rochester and Milwaukee studies, for
example, partitioned a collected soil sample into fine- and coarse-sieved fractions. Finally,
various laboratory practices and procedures can leach more or less lead from the digested soil
sample. Some studies seek to mimic "bioavailable" lead by using an acidic digestion meant to
mimic human stomach acids.

       Unfortunately,  insufficient data were available from the various studies in Table 3-17 to
consider fully any distinctions in soil-lead concentration that would be prompted exclusively by a
study's collection and measurement practices.  Undoubtedly, soil collection and measurement
practices partially explain the observed differences across the studies, but their effects cannot be
quantified at this stage. The data summaries in Section 3.2.2.2 attempted to express soil-lead
concentrations in the Rochester study as reflecting the total sample (as is done in many studies)
rather than only the fine-sieved portion of the sample by adjusting the data based on relationships
observed in the Milwaukee study among fine-, coarse- and total-sieved soil fraction data.
       11 The sum of the sampling weights (adjusted in the §403 risk analysis to represent the 1997 housing stock) for the
146 urban homes in the HUD National Survey is roughly 40 million.  The fitted lognormal distribution in Figure 3-19 treats the
bottom 20 percent of the HUD National Survey (based on the sample weights) as censored data at 21.3 (Jg/g.

                                           142

-------
      CO
      O
  0
  D
  D"
  
-------
       It is possible to discuss study-specific caveats about how the housing selection procedure
and sample collection and analysis procedures differ between the studies and, therefore, can
contribute to the differences observed in the plots and tables in Section 3.2.2.2. For the
Baltimore R&M, the Rochester, HUD Grantee, and HUD National Survey studies, this
information was summarized in Tables 3-3a through 3-3f of the §403 risk analysis report. For
the interim NSLAH, this information was summarized in Section 3.1  of this report. Some of the
study differences mentioned in Section 3.2.1.4 as possibly contributing to differences in dust-lead
loading data would also be contributors to differences in the reported  soil-lead concentration
data. Other differences among the studies in Table 3-17 include the following:

       •      The neighborhoods surveyed within the Baltimore R&M study, 3-City study,
             Cincinnati Longitudinal study, California Lead study, and HUD Grantees
             evaluation had a high prevalence of homes with lead-based paint hazards, along
             with a history of children with elevated blood-lead concentrations and/or
             considered at high-risk for lead poisoning.

       •      For the HUD Grantees evaluation, 28% of the homes were single-family
             buildings, 32% were single-family detached, and 12% were single-family attached
             (rowhouses). All homes in the R&M intervention group within the Baltimore
             R&M study were urban rowhouses (single-family attached). Eighty percent of the
             homes in the HUD National Survey were single-family dwellings.  In the 3-Cities
             study, 100% of the Boston homes were single-family detached residences, most of
             the Baltimore homes were single-family attached dwellings, and the majority of
             Cincinnati homes were multi-story, multi-family structures.

       •      The dates of environmental sampling were 11/89-3/90 for the HUD National
             Survey, 12/93-1/99 for the HUD Grantees evaluation,  8/93-11/93 for the
             Rochester study, 3/93-11/94 for the Baltimore R&M study, 2/89-2/90 for the
             Baltimore 3-City study, 7/89-12/89 for the Boston 3-City study, and 1/89-8/89 for
             the Cincinnati 3-City study. Therefore, the HUD National Survey performed
             sampling roughly three years before the three major studies in this report, but near
             in time to others (such as the 3-Cities study).

             The New Orleans, Baltimore Garden, Minneapolis Clean-Up and Minnesota
             studies have sometimes been identified as using distinct laboratory practices,
             producing higher soil-lead concentrations than might be otherwise measured. The
             published literature regarding these studies, however, cites nothing unusual.

       Because the HUD Grantees evaluation emphasizes local control of the individual
programs, each grantee is responsible for designing and implementing lead-hazard reduction
approaches applicable to its specific needs and objectives. These responsibilities include the
recruitment methods, enrollment criteria, and intervention strategies.  However, to enable
comparison of results from the various approaches, grantees participating in the evaluation
follow the same sampling protocols and use standard data collection forms developed

                                          144

-------
specifically for this evaluation. Table 3-4 of the §403 risk analysis report documented the
differences between grantees in their enrollment/recruitment criteria. As a result, the summaries
in Section 3.2.2.2 were also presented by grantee.

       3.2.2.5 Conclusions of the Soil-Lead Data Comparisons. The following can be
concluded from review of the boxplots and tables within Section 3.2.2 of this report, especially in
regard to how the reported soil-lead concentration data for various studies compare with data
from the HUD National Survey (as portrayed in the §403 risk analysis):

       •     Geometric mean (yard-wide) average soil-lead concentration was quite lower for
             the HUD National Survey relative to the yardwide estimates for most of the other
             studies cited in Table 3-17. However, the interim NSLAH (Section 3.2.2.1), as
             well as such studies as the Cincinnati 3-City and Baltimore Garden studies, did
             report geometric mean soil-lead concentrations that were comparable to that for
             the HUD National Survey.  Otherwise, the distributions of soil-lead
             concentrations were rather consistent across the studies and available grantees.

       •     Among the housing age categories, the greatest difference in observed soil-lead
             concentration between the HUD National Survey (as portrayed in the §403 risk
             analysis) and the interim NSLAH was for housing built prior to 1940, where
             nearly a 50% decline in the estimated median was seen from the HUD National
             Survey to the interim NSLAH. The two sets of results were comparable among
             the other housing age categories.

       •     The low geometric mean soil-lead concentration in the HUD National Survey
             compared to other studies within Table 3-17 was most dramatic for homes built
             from 1940 to 1959. For homes built prior to 1940, the geometric mean reported in
             the §403 risk analysis (463 ng/g) was within 150 ng/g of that for three grantees
             within the HUD Grantees evaluation: California, Minnesota, and Vermont.
             However, for homes built from 1940 to 1959, the geometric mean soil-lead
             concentration across all units in the HUD Grantees evaluation (492 ug/g) was
             over four times higher than that reported in the §403 risk analysis (92.6 ng/g).
             Insufficient numbers of housing units built after 1959 in the other studies prevent
             reliable comparisons of soil-lead concentrations with these studies.

       •     Overall, the importance of housing age is evident in the summaries within the four
             housing age categories. Older housing is more likely to contain higher average
             soil-lead concentrations compared to newer housing. However, within an age
             category, the summaries were reasonably consistent across studies.

       •     As expected, dripline/entryway soil-lead concentrations consistently exceeded
             yard-wide average levels for all studies with sampling plans permitting such
             comparisons.  That soil-lead concentrations exhibit an areal pattern is well-known
             and documented throughout the scientific literature, and suggests caution when

                                          145

-------
              comparing the HUD National Survey yard-wide average results to those of other
              studies.

      "In general, the soil-lead concentrations observed in the HUD National Survey seem lower
than many studies, but not necessarily beyond reason.  Several of these other studies were
conducted in urban neighborhoods already recognized to either have elevated environmental-lead
levels or high incidence rates of elevated blood-lead concentrations among resident children.  As
such, higher soil-lead concentrations among these residences may be entirely consistent.
Furthermore, soil-lead levels in the HUD National Survey were found to be comparable with
those reported in the interim NSLAH, which reflects the entire nation's housing stock, and in-
other studies such as the Cincinnati 3-City and the Baltimore Garden study. Even studies
conducted within the same urban area can differ considerably in the reported soil-lead
concentrations; for example, the  Baltimore 3-City study had levels about five tunes higher than
the Baltimore Garden study.
3.3    EVALUATION OF SOIL PICA IN CHILDREN

       This section investigates what has been published in the literature concerning the
potential effects that pica for soil may have on children's exposure to lead, over and above the
exposure associated with pica for paint that was considered when estimating risks in the §403
risk analysis.  While the analysis did not consider the independent impact of soil pica over and
above paint pica, it considered the impact of soil pica as part of the relation between soil-lead
concentration and blood-lead concentration. While this section does not change the approach
taken in the original §403 risk analysis, it documents information obtained on the component of
soil-lead exposure that may be attributable to soil pica.

       This section summarizes information on pica behavior for soil and paint for the three
studies constituting the Urban Soil Lead Abatement Demonstration Project (USLADP) (USEPA,
1996a), the Rochester Lead-in-Dust study (USHUD, 1995a; Lanphear et al., 1996a), and the
Baltimore Repair and Maintenance (R&M) study (USEPA, 1996c). The percentage of children
who ingest soil, the frequency of soil ingestion episodes, and the amount of soil ingested by
children with pica are estimated.

3.3.1  What Is Soil Pica?

       Definitions. The literature provides varying definitions of pica. Pica is generally
accepted to be the consumption of non-food items and there are at least nine different types of
pica, including soil pica (Lacey, 1990). Some authors also consider mouthing of non-food items
a pica behavior. Usually, pica is seen as normal behavior in young children, but abnormal in
older children  and adults. Exceptions occur, however, for some individuals, such as children and
pregnant women in certain ethnic groups, the socially disadvantaged, groups of low income and
socioeconomic status, developmentally delayed individuals, and the mentally retarded.
                                          146

-------
       The American Psychiatric Association (DSM-EH-R) has clinically defined pica as the
ingestion of non-nutritive or inedible substances and requires repeated ingestion of a
non-nutritive substance for at least one month before pica is considered a diagnosis.  However, in
practical research, authors tend to use less rigorous definitions of pica. For example,
Shellshear et al., (1975) defined pica simply as an unusual appetite for non-food items.

       Some authors consider pica a common occurrence in young children while others view
pica behavior as abnormal. Sedman (1989) included in his definition of pica the ingestion of
foreign substances by children that occurs during the course of normal development. This is
consistent with Karam et al., (1990) who stated that pica includes the ingestion of some non-food
items and that pica is a relatively common occurrence in small children. Barltrop et al., (1974)
defined soil pica as the habitual insertion of soiled fingers or toys into a child's mouth, in
addition to the direct consumption of soil.  In contrast, Lyngbye et al., (1990) loosely defined
pica as a mouthing habit more pronounced than in other children at the same age.  Calabrese
et al., (1991) defined soil pica as the ingestion of soil in amounts far exceeding those observed in
the average child.

       Pica for soil is considered by most authors to be the purposeful ingestion of soil.  This
definition is used throughout this report. Estimates of intentional soil ingestion, such as would
occur in an actual "pica" episode, range from 500 to 13,000 mg soil/day, according to the studies
cited in Table 3-25.  To put this in perspective, quantitative estimates of inadvertent soil
ingestion by normal children range from 9 to 246 mg/day (see Table 3-25), which are consistent
with the estimates used in the §403 Risk Analysis.

       Methods Used to Measure Soil inoestion. Average daily soil ingestion can be
quantified using a mass-balance approach, in which concentrations of tracer elements in fecal
matter are measured and used to estimate the amount of soil ingested. The tracer elements
typically used in soil ingestion studies include barium (Ba), manganese (Mn), silicon (Si),
aluminum (Al), titanium (Ti), vanadium (V), yttrium (Y), and zirconium (Zr). However, in an
adult validation study investigating the recovery of different tracer elements, Calabrese et al.,
(1989) concluded that the most reliable elements for this type of study are Al, Si, and especially
Y.  In addition, the authors indicated that when using these tracers, 500 mg/day could reliably be
detected, and 100 mg/day could also be reliably detected but with a higher degree of variability.
These levels are greater than most estimates of average daily soil ingestion in children. Tracer
elements are generally selected due to their high concentration in soil relative to food products,
and their low level of absorption in the gastrointestinal tract.  Thus, the quantities of these tracer
elements present in the feces, corrected for "background" or intake levels, can be attributed to the
ingestion of soil (assuming there is no other non-food ingestion occurring, e.g., paint).  Using the
concentration of a tracer element in the bulk soil, the total quantity of soil ingested can be
calculated. Concentrations of the tracer elements in the bulk soil are  determined from  soil
samples around the child's home and play area.  Samples are typically taken from the upper
layers of soil (as this is where children are assumed to play), and finer size fractions  may be
separated out (as this size fraction is preferentially ingested) (Calabrese et al., 1989;  Sheppard,
                                           147

-------
(B
O
£

1
O
§
DC
S
1
in
01

            fo
          0§
          §00
          Ǥ
          §5
          a."
          c «
          § .«
          c a.
1-4 years; pica
l analysis using
nd mass-balance
tudy '
es
ca
64 children, ag
measured by f
chemical tracers a
methodology; 8-day
           a
           S
           e
           to
           « —
           X 0)
           0 C.
3
a
1.
a
f •
                 a. o
                 i S
                 5.S
                 ^ 09
                 ^ <0
                 £ u
£p
                o ~
                      CO
                      it
                      U CD
                      O
      0 3
      a o
• lead consumed in soil:
0.96 /jg/day - 11 .6 ^g/day
soil lead = 22 ppm)
Follow-up study of pi
Calabreseet al. ,11 98
                      
                      o °>
                      O C
          to
                           8
2 —
          55 o C
                                xi -ii
                                C 2 <5
                                <0 A JS" g >
                                » £ S S "i
                                *S CO « O *£
                            __   *** ^TT  ** CO

                     fllili.  Islii
                                            -

                                nu
                                2 I .2 •§
                           148

-------
           >&
           II
           &!
           og
           ~6
           v <-
ica:
n-
hil
n);
n)
00-
pi
ldre
dren
no
5 c
4 c
ica: 53,050 mg/day
20 mg/day)
pi
3
ID
N
CO
A
2
ages 1-3 years; children
exhibiting soil ingesti
1998); ingestion
l analysis using
and mass-balance
2 children,
reselected as
Stanek et al.,
measured by fec
chemical tracers
12
pr
(St
          •5
          0 Z,
ra
{
•
ra
2
                     .0
                     a
  a
                       c*  <
                        eo  »;
                      ~  n»  O
                      t* 
-------
                 I
                 (5
                 I
                 (3
                 •i

10
i
                 If) .O
                   ^ I
                    a '>
                 I
                 oJ  a
                   00
                   o>
                 II
c
0>
9-1
tO V
09 C
O 
       b
       o
                                   15
                                   4M
                                   tt

                                   I?
                                                  2-*
                     C CJ
                     « r»

                     o«
                     CJ C
                                       2

                                       o
                                                                    e
                                                                    o
                                                                                   o
                                                                                   CD


                                                                                   a.
                                                                                   
                                                                                   •
                                                                            " a =

                                                                            S >g
                                             ^ •"*  «l *fc
                                             n] •—  0) A
                                            5 g z 1,
                                           :  i
                                            •••
                                           ' .S .9> o "5
                                           1  Q..C  S 5
                                                                                   •o  «
                                                         150

-------
1998). Most quantitative estimates of the amount of soil ingested that were reviewed in this
report were obtained using the mass-balance/tracer element approach.

       Incidence rates (i.e., prevalence) of pica for soil in young children may be estimated from
parental questionnaires.  This approach can yield biased results, however, as it relies on the
observation and accurate reporting of pica by the adult.  In addition, the response depends on the
wording of the question. Various surveys have asked whether the child eats dirt (Annest and
Mahaffey, 1984; Bates et al., 1995; USEPA, 1996a - Boston and Baltimore portions), whether
the child puts dirt or sand in mouth while playing outside (USEPA, 1996a - Cincinnati portion;
USEPA, 1996c), or whether the child puts fingers or toys in mouth while playing outside
(Barltrop et al., 1974). Clearly, these questions would elicit different responses from the same
caregiver. In addition, response choices may be simply yes or no, may specify a timeframe
(e.g., in the past month), or may be open-ended. These choices, too, would result in differing
responses.  Thus, care must be taken in comparing soil pica prevalence rates originating from
parental questionnaire data.

       While mass-balance studies provide soil ingestion rates to support prevalence data, these
studies are also subject to error and have disadvantages. For example, Calabrese et al., (1989)
acknowledge that analyzing chemical tracers without the use of a mass-balance approach
(i.e., not correcting for intake) can result in soil ingestion estimates that are increased by factors
of 2 to 6. In addition, the particular tracer used, the duration of the study, and the frequency of
sampling may also influence reported results (Calabrese et al., 1989; Calabrese et al., 1997). For
example, the short duration of most mass-balance studies makes it difficult to determine a
"normal" rate of soil ingestion for a child.  Approaches using chemical tracers also have
disadvantages in that they are more expensive and generally have small sample sizes.

3.3.2 How Does the §403 Risk Analysis Account for Soil Pica?

       Within the exposure assessment (Chapter 3) portion of the §403 risk analysis report
(USEPA, 1998a), soil was considered an indirect source of lead exposure, although summary
information on soil pica frequency from two lead exposure studies was presented in Table 3-3b
of the §403 risk analysis report.

       An indicator of soil pica was considered as a candidate predictor variable in the
development of the empirical model for the §403 risk analysis.  The soil pica variable was based
on the parental questionnaire administered hi the Rochester Lead-in-Dust study. This variable
measured the child's tendency to put dirt or sand in the mouth using a scale of 0 (never) to
4 (always). The soil pica variable was borderline significant in single media models, which
assessed the relationship between blood-lead concentration and each predictor variable under
consideration. These single media models were the first step in developing the empirical model.
Variable selection for the multimedia exposure model was based on several properties, including
the strength of the relationship with blood-lead concentration as estimated using the bivariate
statistical models, predictive power of each variable when included into a model with competing
sources of lead exposure, and interpretability of parameter estimates.  The soil pica variable was

                                          151

-------
dropped during this phase of the empirical model development. Additional information on the
development of the empirical model can be found in Appendix G of the §403 risk analysis report.

       Age-dependant soil and dust ingestion rates for the Integrated Exposure, Uptake, and
Biokinetic (BEUBK) model were taken from the IEUBK guidance manual (USEPA, 1994) and
represent central tendencies within the range of values seen in different studies. Combined soil
and dust ingestion amounts ranged from 85 to 135 mg/day, as shown in Table 4-1 of the
§403 risk analysis report, of which 45 percent was assumed to be from soil. Thus, soil ingestion
was assumed to be between 38 and 61 mg/day for children aged 0 to 7 years, with the maximal
ingestion estimated for children aged 1 to 3 years. These ingestion rates are consistent with
estimates of inadvertent soil ingestion presented in this report, but are not representative of pica
episodes. While IEUBK model predicted blood-lead levels were adjusted in the §403 risk
analysis to allow consideration of paint pica in homes with damaged lead-based paint, as
described in Section 4.1 and Appendix Dl of the §403 risk analysis report, no such adjustment
was made for the effect of soil pica.

       It should be noted that while neither model used in the §403 risk  analysis  had explicitly
accounted for soil pica as a separate factor independent of paint pica, the impact of soil pica was
included in the analysis as part of the relation between soil-lead concentration and blood-lead
concentration which the analysis characterized.

3.3.3  Prevalence of Soil Pica Behavior

       Estimates reported in the scientific literature of the percentage of children who ingest soil
are summarized in this section.  From the literature, it was not possible to estimate the percentage
of children who exhibit pica for soil but not paint The §403 risk analysis did account for the
effect of paint pica on blood-lead concentration estimates. For children who ingest both paint
chips and soil, it is reasonable to assume that the effect of soil pica is insignificant compared to
that of paint pica.  Thus, in estimating the percentage of children who ingest soil, it is important
to exclude those who also ingest paint chips. It was possible to estimate the percentage of
children who exhibit soil pica, but not paint pica, using information from parental questionnaires
administered in the USLADP study (USEPA, 1996a), Baltimore R&M study (USEPA, 1996c),
and Rochester Lead-in-Dust study (USHUD, 1995a; Lanphear et al., 1996a). This information is
also summarized in this section.

       3.3.3.1  Literature Review. Most sources in the literature reported prevalence rates for
general pica behavior (mouthing or eating non-food items) or for soil pica (eating dirt). One
source (Stanek et al., 1998) reported pica rates for a variety of specific non-food items (soil, paint
chips, paper, toys, etc), but did not cross-tabulate [e.g., 18 percent of children ages 1-6 years, as
assessed by parent interview, were reported to ingest/mouth dirt at least monthly  and 3 percent to
ingest/mouth paint chips, but information was not provided on how many children eat both soil
and paint chips (Stanek et al., 1998)]. An overview of selected studies estimating pica
prevalence is shown in Table 3-25 above.  It is important to note that in the cited studies, various
definitions of "pica" were used in reporting the prevalence of pica behavior.  For example,

                                          152

-------
Greene et al., (1992) defined soil pica only as the occurrence of soil ingestion and reported the
percentage of children who ingest soil based on caretaker interview.  In comparison, Calabrese et
al., (1997) defined soil pica as consumption exceeding 0.5 grams per day and reported the
prevalence of pica behavior as assessed quantitatively by mass-balance methods. It should be
noted that only primary research findings are reported in Table 3-25, with the exception of
Calabrese and Stanek (1993).  Several review articles were also obtained. These were excluded
from Table 3-25, as insufficient details of the source studies were provided.  General findings of
the review articles are cited in the text.

       Table 3-25 shows that the estimated percentage of children ingesting soil ranged from
1.6 to 62 percent and varied with definition/criteria for soil pica used, age group of children, and
socioeconomic status. In general, 12 of 16 observations in the table report a prevalence of soil
ingestion in children of 13 percent or lower (where, at a minimum, limiting criteria are defined as
ingesting soil at least once). "Normal" mouthing behavior, however, is typically exhibited more
commonly, particularly in the younger age groups.  For example, Barltrop et al., (1974) reported
that 43 percent of children exhibited pica defined to include mouthing behavior, but that
9 percent were estimated to swallow soil.  Stanek et al., (1998) assessed non-food ingestion and
mouthing behaviors in 533 children, ages 1 to 6, by parental interview. Results of the survey
indicated that 38 percent of 1 year old children and 21 percent of 2-year old children
ingest/mouth soil at least monthly. In contrast, at ages 3 to 6 years, less  than 10 percent of
children were observed to ingest/mouth soil at least monthly.  At age 1 year, 11 percent of
children were observed to ingest/mouth soil daily compared to one percent or less among
children aged 2 to 6 years.

       Of the studies that used mass-balance methodology, prevalence of soil pica ranged from
1.6 to 20.8 percent. For studies that employed parent or caretaker interview methodology, soil
pica prevalence ranged from 3.2 to 19 percent, although one study reported a rate of 62 percent,
which appears to be more consistent with studies that monitored general mouthing behavior.  For
both methodologies, the prevalence of soil ingestion tended to be higher in the younger age
groups and for children in families with lower socioeconomic  status. Although, as mentioned in
Section 3.3.1 above, techniques utilizing parental interview are generally considered less reliable
than quantitative methodologies, the issue of consistently defining "pica" when reporting study
results is an issue not only in studies using questionnaire methodology, but also in the
mass-balance/chemical tracer studies. Differences in values reported, both between and within
the various assessment techniques, may largely be due to differences in how pica behavior is
being defined in the study.  Thus, many studies estimating soil ingestion prevalence may not
consistently monitor the actual overall risk associated with soil pica behavior in children.

       3.3.3.2  Prevalence of Soil Pica Separate from Paint Pica. Although the literature
review provided several estimates of the prevalence of soil pica behavior, none of the cited
sources provided information about concurrent paint pica behavior. For the purpose of this
report, the prevalence of soil pica behavior in the absence of paint pica is of interest, as the §403
risk analysis did account for paint pica behavior.  For children who ingest both paint chips and
soil, it is reasonable to assume that the effect of soil pica is insignificant compared to that of

                                           153

-------
paint pica. Unless there is industrial contamination, or the home is in an area with heavy traffic,
where residual leaded gasoline emissions are present, lead in residential soil is usually derived
primarily from lead-based paint. Thus, soil pica can be considered an indirect pathway of
exposure to leaded paint, whereas paint pica is a direct exposure pathway.

       Information on pica behavior for paint, soil, and other objects was collected in the three
USLADP studies (USEPA, 1996a), the Rochester Lead-in-Dust study (USHUD, 1995a;
Lanphear et al., 1996a), and the Baltimore R&M study (USEPA, 1996c), through parental
reporting of observed behaviors. Therefore, it is possible to use these data to estimate the
prevalence of soil pica separately from paint pica. This information is summarized hi Table
3-26.  As can be seen hi this table, rates of soil pica only range from 9.1 to 40.9 percent, while
rates of both soil and paint pica range from 1.4 to 7.4 percent.

       Some of the disparity hi the rates reported in Table 3-26 can be explained by the survey
questions and other factors associated with the study. For example, in the Rochester Lead-in-
Dust, Baltimore R&M, and Cincinnati USLADP studies, parents were asked how frequently the
child put dirt or sand in his or her mouth. In contrast, parents in the Boston and Baltimore
portions of the USLADP were asked how frequently the  child ate dirt or sand. The paint pica
questions were more consistent across studies, querying how frequently the child put paint chips
in his or her mouth.  In the Cincinnati USLADP study, the time-period of observation for both
soil and paint was limited to the previous month, whereas the other studies used open-ended time
periods. Response rates in Rochester were consistent with literature estimates of soil pica that
included mouthing behavior, while the Baltimore R&M and Cincinnati USLADP studies
provided substantially lower estimates. Because most homes in the Baltimore R&M study had
small or no yards, the low estimates of soil mouthing behavior are not unexpected. The lower
response in Cincinnati is probably due to the limited period of observation.

       Since inadvertent soil ingestion due to mouthing behavior was included in the IEUBK
model analysis for the §403 risk analysis, the prevalence of soil ingestion, rather than mouthing
behavior, is of interest hi the context of this report. Thus, the Boston and Baltimore portions of
the USLADP study provide the best estimates of soil  pica behavior in the absence of paint pica.
These estimates are  14.4 and 16.3 percent hi Boston and Baltimore, respectively.  These
estimates are greater than those derived from the mass-balance studies, but consistent with other
studies that rely on parental reporting methods. The prevalence of pica for both paint and soil
was low in Boston (1.4 %), but somewhat higher in Baltimore (6.0 %). Adding these rates to the
reported rates for soil pica alone does not substantially increase the estimates, however, which
remain hi the range of other studies that rely on parental reporting.
                                          154

-------
Table 3-26.   Estimated Rates of Paint and Soil Pica Behavior Reported in the
              USLADP Studies, the Rochester Lead-in-Dust Study, and the
              Baltimore R&M Study
Study
Boston
USLADP
(146 children)
Baltimore
USLADP
(400 children)
Cincinnati
USLADP
(220 children)
Baltimore R&M
Pre-intervention
(165 children)
Rochester
Lead-in-Dust
(203 children)
Type of Pica
Behavior
Soil only
Paint only
Soil and Paint
neither
Soil only
Paint only
Soil and Paint
neither
Soil onty
Paint only
Soil and Paint
neither
Soil only
Paint only
Soil and Paint
neither
Soil only
Paint only
Soil and Paint
neither
StudyiChOdren Exhibiting Such Pica Behavior.
Percent (#) of Y
Study Children1
14.4% (21)
9.6% (14)
1 .4% (2)
74.7% (109)
16.3% (65)
10.5% (42)
6.0% (24)
67.3% (269)
23.2% (51)
2.7% (6)
2.3% (5)
71.8% (158)
9.1% (15)
7.3% (12)
7.3% (12)
76.4% (126)
40.9% (83)
2.5% (5)
7.4% (15)
49. 3% (100)
F Average Age
. (months)
NA
NA
NA
NA
35.7
33.1
26.5
41.7
35.9
19.7
28.0
28.1
27.5
26.7
24.2
31.1
21.0
23.0
20.0
21.4
Geometric Mean,
Blood-Lead Cone. (pg/dL).
12.5
12.7
13.4
11.7
12.0
11.7
15.2
10.3
12.8
15.3
14.7
8.9
10.5
15.3
20.7
9.4
6.1
11.5
8.5
6.2
'  A response of "Unknown* was treated as missing and was not included in the calculation of these percentages.
NA = Not applicable
3.3.4 Estimating the Frequency of Ingestion and Amount
       of Soil Ingested by Children Who Exhibit Soil Pica

       As discussed in Section 3.3.3.1 above, studies reporting soil ingestion prevalence have
the potential to misrepresent the extent of soil pica behavior in children due to differences in
methodology and criteria for defining pica. Therefore, estimates of ingestion quantity and
frequency may also be employed to assess the severity of soil pica behavior.

       Mass-balance studies provide data on the frequency of ingestion and amount of soil
ingested by children who exhibit soil pica.  These studies estimate the typical amounts of soil
inadvertently ingested by normal children as ranging from 9 to 246 mg/day. The estimated
                                           155

-------
quantities ingested in actual "pica" episodes are between 500 and 13,000 mg/day (Table 3-25).
The literature generally reports pica behavior to be episodic hi nature, varying both amongst
different children and within individual children.  In addition, the occurrence (including both
frequency and quantity) of soil ingestion was observed to be influenced by the age of the child
(Stanek et al., 1998; Annest and Mahaffey, 1984), as well as by variety of factors that may alter
the child's access to soil, including seasonal variation and/or climate/vegetation differences
(Simon, 1998; Calabrese and Stanek, 1993), socioeconomic status (Annest and Mahaffey, 1984;
Bhatia, 1988), and parental supervision (Calabrese and Stanek, 1993; Bhatia, 1988). However,
Davis et al., (1990) found that although there was considerable variability in soil ingestion
estimates among children, there was no consistent demographic or behavioral factor that was
predictive of soil ingestion.

       Calabrese et al., (1989) estimated median soil ingestion rates, including those involved in
non-pica behavior, between 9 and 40 mg/day (n = 64). Calabrese et al., (1997) also observed
median soil ingestion rates under 40 mg/day in 12 children (selected from the population
described in Stanek et al., 1998) identified by their parents as likely to ingest soil at a high rate.
These levels of soil ingestion typically would not be considered pica behavior. Each of these
studies, however, did report observations of a child exhibiting extreme soil pica behavior, with
one child ingesting from 5 to 8 grams of soil per day (Calabrese et al., 1989) and another child
ingesting between 0.5 and 3.0 grams of soil per day on 4 of 7 days (Calabrese et al., 1997).
Calabrese et al., (1991) found that the soil pica behavior for the former child occurred only on
two days during the two weeks of observation with an ingestion rate ranging from 10-13 grains
of soil per day, suggesting that the issue of variability in soil pica behavior may be very
important, meriting further research. Implications of these patterns were demonstrated by
Calabrese et al., (1993), who observed that on the two days when the child displayed soil pica
behavior, she also displayed striking increases in fecal lead excretory values. In contrast, the pica
child reported in Calabrese et al., (1997) consistently ingested large quantities of soil
(0.5-3 g soil/day on 4 of 7 days).

       Calabrese and Stanek (1993) presented results of a 4-month mass-balance/chemical tracer
study performed by M.S. Wong of 52 Jamaican children of generally normal intelligence hi an
institutional setting.  The children were partitioned into a younger (0.3-7.5 years) group and an
older (1.8-14 years) group.  One of the children in the older group exhibited mental retardation.
This was the only child in the older group (of 28 children) that exhibited soil pica exceeding one
gram of soil per day. This child had an average ingestion rate of 41 g soil/day over 4 months
(observations on 1 day per month), hi the younger group, 10.5 percent of total observations (n =
84) included soil pica, and five of the 24 children exhibited soil pica on at least one occasion.
The Wong study showed that soil pica occurred more frequently in younger children, and there
was a fairly high degree of daily variation in soil ingestion among the children exhibiting soil
pica. For example, 3 of 6 children displayed pica on only 1 of 4 days.  Furthermore, even for the
children who consumed soil more consistently with regards to frequency, the rates were still
variable (e.g., 1.0-10.3 g/day). Calabrese and Stanek (1993) suggest that although this study
confirms that soil pica, strictly defined as ingestion greater than 1.0 g/day, is likely to be rare in
                                           156

-------
older children, the Wong study is important in that it challenges the idea that pica is a rare event
in younger children.

       Using daily soil ingestion data from their 1989 study, Stanek and Calabrese (1995)
developed annual soil ingestion distribution estimates as follows. First, the mean and variance of
daily soil ingestion were estimated for each of the 64 children in the 1989 study, based on 4 to
8 daily estimates for each child. Then 365 daily soil ingestion amounts for each child were
calculated as percentiles of a log-normal distribution with the estimated mean and variance, in
increments of 1/365. Based on these distributions, Stanek and Calabrese conclude that 33
percent of children are expected to ingest more than 10 grams of soil on 1-2 days per year and
that 16 percent of children are expected to ingest more than 1 gram of soil on 35-40 days per
year. These ingestion levels are consistent with amounts estimated for soil pica episodes. The
median and 95th percentile for average daily soil ingestion resulting from this method were
75 mg/day and 1,751 mg/day, respectively. While the median estimate is similar to previous
estimates, the estimated 95th percentile is substantially greater than most other estimates.

       Assumptions and limitations of this approach include:

       1.    The assumption of a log-normal distribution for daily soil ingestion. Insufficient
             data were available to determine whether this assumption is reasonable.

       2.    The estimation of the mean and variance for each child based on very small
             sample sizes. The annual estimates were strongly affected by the tails of the
             distribution, which are imprecise due to large variability in the estimates of the
             mean and variance.

       3.    The extrapolation of daily soil ingestion estimates from a 2-week period in
             autumn to the remainder of the year,  without regard to possible seasonal effects.
             In addition, the children studied were a nonrandom sample residing in or near an
             academic community in western Massachusetts. Thus, the soil ingestion behavior
             of these children may not be representative of those living in other climates,
             geographic regions, or in inner-city or rural areas.

       4.    The presence of trace elements in fecal matter was assumed to be entirely due to
             soil consumption, after correcting for food consumption, with no contribution
             from indoor dust.

Many of these assumptions and limitations serve to introduce positive bias to the daily soil
ingestion estimate, while the effect of others is unclear.  Nonetheless, this analysis is at present,
the only available source of both frequency of soil pica episodes and amount ingested during soil
pica episodes.
                                          157

-------
3.3.5  Conclusions on Soil Pica

       The following conclusions can be made from the findings presented in this section:

       •     The prevalence of soil pica, exclusive of paint pica, is most likely between 10 and
             20 percent in young children. For the purpose of this report, the Boston and
             Baltimore portions of the USLADP provide the best estimates of soil pica
             behavior in the absence of paint pica (14.4 and 16.3 percent, respectively).

       •     Soil pica behavior is episodic in nature. The frequency of soil pica episodes
             depends on many factors, including climate, access to bare soil, socioeconomic
             standing, age of child, and parental supervision. In one study of 12 children
             identified by their parents to be predisposed to pica for soil, only one child
             displayed soil pica during the two week observation  period (Calabrese et al.,
             1997). Only one study estimated annual rates for pica episodes (Stanek and
             Calabrese, 1995). This study suggested that 33 percent of children would ingest
             more than 10 grams of soil on 1-2 days per year, and that 16 percent of children
             are expected to ingest more than 1 gram of soil on 35-40 days per year.

       *     Estimates of the amount of soil ingested during pica episodes vary widely among
             the mass balance studies, from 500 to 13,000 mg/day. The average daily ingestion
             over a year, however, may be much lower.  Assuming the frequencies estimated
             by Stanek and Calabrese (1995), children who ingest 15 grams of soil on 1-2 days
             per year and 50 mg/day on remaining days would have an average daily soil intake
             of 132 mg/day over the course of a year. Children who ingest 1.5 grams of soil on
             40 days per year and 50 mg/day on remaining days would have an average daily
             soil intake of 209 mg/day. A question, however, is whether the amount of lead in
             soil ingested on the small number of days where pica episodes occurred would be
             sufficient to elevate the blood-lead concentration to unsafe levels.
3.4    CHARACTERIZING THE POPULATION OF CHILDREN
       IN THE NATION'S HOUSING STOCK

       For the §403 risk analysis, it was necessary to estimate numbers of children of specific
age groups who reside within the 1997 national housing stock in order to characterize the extent
to which various environmental-lead levels provide exposures to children and to characterize the
benefits associated with performing interventions under §403 rules. These estimates were based
on numbers of housing units determined by sampling weights within the HUD National Survey
(conducted in 1989-1990), revised to represent the 1997 national occupied housing stock and on
average numbers of children per housing unit determined from the 1993 American Housing
Survey (AHS). The estimates used in the §403 risk analysis were presented in Section 3.3.2 and
Appendix C of the §403 risk analysis report. This section provides alternative estimates using
                                         158

-------
more recent data (i.e., interim data from the NSLAH and data from the 1997 American Housing
Survey).

       The method to calculating the alternative estimates involved determining the numbers of
children in a given age group for each of the 706 housing units surveyed within the NSLAH
whose interim data were made available to this effort. Methods used to obtain these estimated
numbers of children were similar to those presented in Section 1.2 of Appendix Cl of the §403
risk analysis report.  For a given age group of children, the estimated number of children
associated with a given NSLAH-surveyed unit was determined by the following formula:
      # children  = (1997weight)*(Average#residentsperunit)*(#chUdrenperperson)
(1)
The "1997 weight" factor in equation (1) was the interim sampling weight from the NSLAH for
the unit.  The factor "average # residents per unit" in equation (1) was calculated for the housing
group based on information obtained from the 1997 AHS. The 1997 AHS database provided
information on up to 18 residents within each housing unit in the AHS.  Once units surveyed in
the 1997 AHS were placed within the four year-built categories (pre-1940,1940-1959,1960-
1979, post-1979), the average number of people residing in a unit (regardless of their ages) was
calculated for each group. This average ranged from 2.5 to 2.7 across the four year-built
categories. Therefore, a common average of 2.6 residents per unit was used for the entire
national housing stock. The third factor in equation (1), "# children per person," represented the
average number of resident children (of the given age group) in a housing unit. This factor was
determined by dividing the total number of residents in the housing stock of a given age group by
the total number of residents regardless of age, where both totals were calculated from data in the
1997 AHS. The method for calculating this third factor differed from the approach used in the
§403 risk analysis, where forecasted birth rate and population estimates  from the Bureau of the
Census were used.

       Table 3-27 contains estimates of average number of children per unit in the 1997 national
housing stock, according to age group.  These number are the product of the final two factors in
equation (1).  Therefore, these number are multiplied by the sampling weights for each housing
unit in the interim NSLAH to obtain a revised number of children per housing unit. For children
aged 12-35 months, the estimated average of 0.073 children per unit is about 9% lower than the
estimate of 0.080 used in the §403 risk analysis.

Table 3-27.   Alternative Estimates of the Average Number of Children Per Unit in the
              1997 National Housing Stock, by Age of Child
Age Group
12-35 months
12-71 months
Estimated Average Number of
Children Per Unit
2.6*0.0281 = 0.073
2.6-0.0732 = 0.190
                                         159

-------
       By summing the estimates across surveyed units in the interim NSLAH, the updated
number of children aged 12-35 months and 12-71 months residing within the 1997 national
housing stock is obtained by year-built category and for the nation. Table 3-28 provides these
alternative estimates on the number of children residing in the 1997 housing stock according to
age of housing unit and age of child. The overall estimate of approximately 6.51 million children
aged 1-2 years is approximately 18% lower than the estimate of 7.96 million made in the §403
risk analysis.  The lower estimates are due to the lower per-unit estimate from Table 3-27 and on
the lower sample weight total in the interim NSLAH data compared to the HUD National Survey
(Table 3-2). They are also likely to be underestimates of the numbers of children of the given
age category, based upon population projections previously published by the U.S. Bureau of the
Census (e.g., Day, 1993).

Table 3-28.  Alternative Estimates of the Average Number of Children in the 1997
             National Housing Stock, by Age of Child and Year-Built Category, Based on
             Data Obtained Since the §403 Risk Analysis
"Years in Which Housing
Units Were Built
Prior to 1940
1940-1959
1960-1977
After 1977
Unknown1
All Housing2
Age of Child Within These Housing Units
1-2 Year$::-':P '•' :
1,053,000
1,234,000
1,877,000
1,759,000
591,000
6,513,000
1-S Years
2,743,000
3,214,000
4,889,000
4,582,000
1 ,540,000
16,967,000
1 There are 66 units in the interim NSLAH which have missing age of house data.
2 Values in this row may differ from sum of previous rows due to rounding.
3.5    SUMMARIES OF DUST-LEAD LEVELS ON SURFACES OTHER
       THAN UNCARPETEP FLOORS AND WINDOW SILLS

       The exposure assessment in Chapter 3 of the §403 risk analysis report concluded that
even at low to moderate lead levels, lead-contaminated dust can affect children's blood-lead
concentration.  The assessment focused on dust-lead found on floor and window sill surfaces, for
which §403 regulatory standards were proposed. However, dust found on other surfaces, such as
exterior dust and dust hi air ducts, carpeted floors, window troughs (also known as window
wells), and upholstery, may also potentially present a lead exposure hazard. Many issues
concerning potential exposure to dust-lead on these other surfaces were raised throughout the
§403 Dialogue Process as well as in comments received on preliminary drafts of the §403 risk
analysis and on the proposed rule. For example, there was extensive discussion during the
Dialogue Process concerning whether standards were necessary for window troughs (i.e.,
window wells) as long as there are standards for window sills and the window troughs are
                                         160

-------
thoroughly cleaned. Concern was also expressed about sampling on carpeted and upholstered
surfaces.

       The purpose of this section is to supplement information in the original exposure
assessment by assessing the potential exposure to dust-lead found on surfaces other than floors
and window sills. In particular this assessment seeks to answer the following questions:

       1.   What information is available to assess residential lead exposure resulting from dust
           on surfaces other than floors and window sills?

       2.   What does this information say about the distribution of environmental lead levels
           for dust on these other types of surfaces?

       3.   Is there evidence of a relationship between these exposures and children's blood-
           lead concentrations?

       4.   Is the information sufficient to set a regulatory standard and is a standard necessary?

       The exposure assessment is based on a review of the literature to identify studies with
potentially useful data for assessing lead hazards due to dust-lead on these other surfaces. It
should be noted that this section deals only with several specific surfaces other than uncarpeted
floors and window sills. These  surfaces include exterior dust, air ducts, window troughs, and
upholstery. Hazards associated with lead-contaminated dust from carpeted floors are addressed
in Section 6.5 and Appendix I of this report.

       The literature review for this assessment drew upon previous literature reviews conducted
for the §403 risk analysis and reviews conducted for other EPA published reports (e.g., USEPA,
1997b). Most of the studies that were found that addressed the various surfaces are included
here, regardless of whether there is any information specifically relating dust-lead levels and
blood-lead levels. For those studies where blood samples were  collected from resident children,
those results are also presented. Table 3-29 provides a summary of the studies that were
examined. The table indicates the surfaces from which dust samples were collected and in the
case of exterior dust samples, where those samples were collected.

       Table 3-29 indicates that the review of the literature found fourteen studies that have
examined exterior dust as a source of lead exposure and seven studies that similarly assessed
window troughs. Only four studies were located with significant information on dust-lead levels
in air ducts, and four with similar information on upholstery.

3.5.1  Distribution of Dust-Lead on Surfaces Other than Floors
       and Window Sills

       Tables 3-30 through 3-33 present summary information from the studies related to
exterior dust samples, air duct samples, window trough samples, and upholstery samples,

                                           161

-------
Table 3-29.  Studies for Which Dust Samples Have Been Collected from Exterior Areas,
            Air Ducts, Window Troughs, and Upholstery for Lead Analysis

  = medium was sampled in the given study
I.-'..' •:.•'''•'• '. '" . Study:'; , . ,:v. . ./\-
Lead-Based Paint Abatement and Repair & Maintenance

-------
01

M


•5
§
£
"5
O
 10
M
 o.

i
**
 (0
         S
                O> 00 OJ 00 09

                o> co •* 01 •»


                            i^ in ui
                            85*
                                                       *
                             I
                          C9
If
III
                                  &.&.SL


                                  SO TO 00
                            ss
                                                                 at
 0)
 O

1
 *•«
CO


 2
n»


 i
O
 I

(0

6
CO
 a
                     il
                            EC
                            m
 Hi
                                                           I
                             55
                                      163

-------
                                           ep  a>»   ev i

                                           CO * CD CO   CO '
go   n co   to o> co   •** co   ,, -^
a  • en oo • i~ co 03 'uico  • r^ p-
                                                           •r- CM « •» in 
-------
                    CD o> to I-; r-^

                    S.J2.H2.HLS2.
                    'v 2 S3 § I
                                         Su> CM •- r«- r^ r» m
                                         „ v *_ oj »- t- ~
                                        i < oj o a uj u. O
 eo" u£ ^* %? &? &? 35" ^* TO^ 3%  £% *
 s.si^.s2.siaiaj.a!.s2.aia^3
?s-sr
;s,s>.


                                                                                                   .

o
m
 a

            if
      :1
          {]
          &•
                    I
                    CO
                li
                                       a. sL IL £2, si al al si.
                                                                        ;
                                                                        i o a>
                                                                       o ed id co
                                                                       £ u. u_ u.
                                           m o o uj u. O
                                           g g g
 •   w en en

 •g-i-i-i
-S S S^
 ffl CD OQ CD
                                                                    CO
                                CD UJ I C,
                                                     165

-------

                   S2.Ei.QSi
                   to cd ^ cd

Q i

(O.C
                                    .
                                     9cj .  in

                                   —si a. 21
                                 to .


                                 *'
                                 a.
                          166

-------
CO

I
<

"8
5
CD
t»
£
"3
U
e
(0
t!
3
ts
a
i_
<
s

I
•O


w
I
"S
>.
0)
    a
    Si
               5-r«

               *?
                 CO  CO *T CM V
         25i"
         l5Sp
         iiill
            I

            o
 1
            O

            E
co 7

833
cjcjf
                 fi
Fl
             !l
             k
.2
                        £1
                       2-
                       5.
                                            s-
                                            o
                                         |i|
                                         asE
                                                       E
                                                 II
                                         M 25'
                                        w «1.
                                        n SS
                                        zo<:
                                                     CO II


                                                    JQcr
                                       S u
                                       gs
                                 167

-------
*

I
<
•o
a


I
•o
o
o

S
0



S
a
CO
Q


f
2


o
•o

i
S
o

1
*<
0)



S
«0
O

•5


a



CO

ci
n
n
o
S

.2
              e>o>c\i cq oq
              o» CT * cu •*
                         S .SB .a
           i
              lifll
                      g
  iiiii
                           .
                         in in
CO CO S S
—.ass

8SS2E:
                                           O) CO
                                                 CM




                                                 
               .

              *? *f *? S
              ^ ce i- T r
              cn o 


              i
              m
                         i!
                         CD I
                                  J
                                  CD

       II
       II
                       lill
                       nil

                       Illl
llsl!*
iilii!
                              1
                               168
                                                13

-------
CM
CO

n

9
5
a
           3
                                     coco
-------
1
a

             8
     1
§
o
 S
 o
 a


 

_«

a
           S
           tl

           8'
                     *=*w  .  *

                     *^ <0 (O  05 CVI 05»

                     O> CT •* CM •*
                     333S?
                     i-=|||
                 i
                 0
                        _
                      >«   rj-ejl

                       §   So
                     .£--- ~r-e
ent
Ab
                  ' S 3^.




                  ip


                  111
                                  1
                                      -oil

                                       0 " «S
                                      ^ 52 n u) t
                                      O


                                      e
                                  O
B/6

                                                           ^
«?
s
                                                               gf
                                                                I


                                                               1

                                                                        *
PA Supetfund

(1989-1 990)
                                                                             5 ° <  ll S it  ll eS

                                                                             n II -g  "22 M

                                                                             z(3<5 wm a ir
                                                 170

-------
respectively. Included in the tables are sampling location, dust collection methods used, number
of dust samples taken, and distribution statistics available from the literature on dust-lead
concentrations and loadings. Also presented in the tables are simple summaries of blood-lead
concentrations for those studies which also sampled blood.

Exterior Dust

       Table 3-30 summarizes the data from the fourteen studies in which exterior dust samples
were collected and analyzed for lead content. In addition, there are also results for several
studies in which dust samples were collected inside the home in the entryway. These can
occasionally be good representations of exterior dust samples. The summary consists of
descriptive statistics for the dust-lead measures and also for blood-lead measures, when
available. Summary statistics are presented separately for different study groups and different
sampling periods, where appropriate.  As indicated in Table 3-29, exterior dust samples were not
collected at a consistent location across all studies. Sampled locations included the house
entrance  (including doormats and porches),  street, and the child's play area.  Likewise, samples
were collected by a variety of methods including surface scraping, vacuum sampling, and wipe
sampling. The sampling location and method have a significant impact on the lead loading and
concentration estimates.

       Overall, even with the variability in sampling location and method, Table 3-30 indicates
the potential for significant amounts of lead in exterior dust, with concentrations often exceeding
400 ug/g and loadings often exceeding 100 ug/ft2.

Air Ducts

       As indicated in Table 3-31, only four studies were identified that contained information
on lead levels in dust within air ducts in residential housing. While only limited information was
encountered, a consensus across studies was that air ducts can contain high amounts of dust and
lead. This was due partially to the general lack of cleaning of ah* ducts over time and the ability
of lead particles to enter air ducts from outside of the unit via ventilation filters.

       In units with a potential for containing lead hazards, dust-lead loadings in air ducts
typically exceeded 100 ug/ft2, with individual samples often exceeding 1,000 ug/ft2. Lead levels
can vary  considerably among dust samples within the same unit and in different units. Older
ductwork and HVAC systems, as well as vacant units in which no cleaning is performed and
HVAC systems may not be used,  tend to have high dust loadings, and therefore, higher dust-lead
loadings  when a lead source is present. Several methods were used across studies to collect dust
in air ducts. As air ducts often have metal surfaces, issues concerning static electricity must be
considered when sampling dust from air ducts.

       The EPA Comprehensive  Abatement Performance (CAP) study involving occupied
homes assumed to be free of lead-based paint for at least two years, provided the greatest amount
of information on lead in dust within air ducts; levels were relatively low in this study compared

                                          171

-------
to the others. Nevertheless, in a typical housing unit in the CAP study, average dust-lead
loadings from air ducts exceeded all other sampled surfaces except for window troughs and
entryways. In general, air duct dust-lead levels in the Baltimore R&M study and the Renovation
and Remodeling study were considerably higher than in the CAP study, as these studies included
older, vacant units in need of repair and maintenance.

       The relative sparsity of published information indicates that many open questions exist on
the nature of lead contamination of dust within residential air ducts and whether the lead in this
dust is available for exposure to residents (especially children).  Nevertheless, evidence exists
that air ducts can contain some of the highest levels of lead in dust within a housing unit.

Window Troughs

       Table 3-32 presents summary information on seven studies that sampled dust-lead levels
in window troughs (also known as window wells). The HUD Grantees evaluation is a significant
data source on window trough dust-lead levels in high risk housing that is not included in Table
3-32, as these data are still being collected and reported.

       In general, partially because of the published data summarized in Table 3-32, and
partially because a standard for window troughs has been historically used in risk assessments
and to determine clearance (following EPA's Interim Guidance for §403 standards and the HUD
Guidelines), window troughs are widely recognized as a major reservoir of dust-lead in
residences. As shown in Table 3-32, levels often exceed 800 ug/ft2 and it is not uncommon to
see levels above 10,000 ug/ft2 in high risk housing. However, unlike the other surfaces discussed
in this report, national estimates of the distribution of dust-lead in window troughs are available
from the HUD National Survey. The estimated national geometric mean dust-lead loading in
window troughs from the HUD National Survey (as modified in the §403 risk analysis to reflect
the 1997 housing stock and wipe techniques) was 460 ug/ft2 (Table 3-32), with 30% of homes
estimated to have average window trough dust-lead loadings at or above 800 ug/ft2.

Upholstery

       Table 3-33 summarizes the data from the four studies which collected dust-lead samples
from upholstery. In general, dust-lead loadings for these surfaces averaged below 100 ug/ft2.
As the sample sizes in all of these studies were small, and sampling techniques, sampling
locations, and study goals varied considerably from study to study, more information would be
necessary to fully characterize potential lead hazards associated with upholstery.
                                          172

-------
3.5.2  Evidence of a Relationship Between Children's Blood-Lead
       Concentrations and Dust-Lead on Surfaces Other Than
       Floors and Window Sills

       The information available to assess children's exposure to dust-lead on surfaces covered
in this section is discussed in detail for each surface type below. It should be noted, however,
that in general it is difficult to establish the causal link between these surfaces and children's
blood-lead concentrations. This is true for many reasons. Often other important sources of lead
exposure are not well characterized in the studies that provide data on these special surfaces.
Correlations are often estimated based on small sample sizes and without adjusting for other
exposure variables such as lead in floor-dust and soil. Moreover, there is often correlation
between lead levels on these surfaces and lead levels on floors, window sills, and in soil. For all
of these reasons, it must be noted that even significant correlation coefficients should not be
interpreted as the degree to which dust-lead on these surfaces causes a change in blood-lead
concentration. In almost all cases, in order to characterize the pathway of lead from these
surfaces to children's blood, additional data collection or analyses are needed.

Exterior Dust

       Table 3-30 (in the previous subsection) contains a summary of the dust-lead data and
blood-lead data separately for studies which collected exterior dust.  However, it contains no
results providing information about the relationship between exterior dust-lead levels and blood-
lead levels.  The reports describing these analyses were examined to assess the relationship
between exterior dust-lead levels and blood-lead levels. In the Repair and Maintenance Study
(USEPA, 1996c, 1997c), exterior dust samples were collected at five separate times. The
(Pearson) correlation coefficients between blood-lead concentrations and entryway dust-lead
concentrations ranged from 0.23 to 0.49, and the correlation coefficients between blood-lead
concentrations and entryway dust-lead loadings ranged from 0.10 to 0.46. In most cases, those
correlation coefficients were statistically significant at the o=0.01 level. In the Rochester Study
(USHUD, 1995a), the University of Cincinnati Dust Vacuum Method (DVM) and Baltimore
Repair and Maintenance vacuum method (BRM) were used to collect external dust samples.  The
correlation coefficients for the BRM and DVM techniques were 0.21 and 0.27 for the correlation
between blood-lead concentrations and exterior dust-lead concentrations and 0.34 and 0.18 for
the correlation between blood-lead concentrations and exterior dust-lead loadings, respectively.
The correlations were all statistically significant at the o=0.05 level, with the correlation for the
DVM measured loading significant at the a=0.01 level.  On the other hand, the Mexico City
Study (Romieu et. al., 1995), Arnhem Study (Brunekreef et. al., 1981) and the Midvale Study
(Bomschein et. al., 1990) reported the correlations between external dust measurements and
blood-lead levels to be statistically insignificant.

       Multivariate regression and structural equation modeling was used in some of the studies
to examine how multiple sources of environmental lead exposure and other factors affect blood-
lead levels.  Regression analyses were carried out in most of the studies where both external dust-
lead and blood-lead measurements were collected, but the external dust-lead measurements were

                                          173

-------
not included as an explanatory variable in any of the reported regression models. Reasons for
excluding external dust-lead were not clearly stated. Speculatively, such reasons may include a
lack of interest in the relationship, poor data quality in the external dust-lead measurements,
colinearity of external with internal dust-lead measurements and omission of the variable through
step-wise regression. Structural equation modeling was carried out in the (Three Cities) Urban
Soil Lead Abatement Demonstration Project (USEPA, 1996a), Butte-Silver Bow Study (Butte-
Silver Bow DoH et. al., 1991) and Midvale Study (Bornschein et. al., 1990). In the Three Cities
Study, exterior dust was lead considered as a component of the lead exposure pathway in the
general structural equation model, but the component was excluded in the actual implementation
of the model. In the Butte-Silver Bow Study, external dust-lead was also excluded from the
structural equation model, but external dust-lead was observed to be correlated (Pearson
correlation r=0.64) with soil lead, which was included as a component of the lead exposure
pathway in the model.  The Midvale Study was the only one to include external dust-lead hi the
actual implementation of the structural equation model, but the dust-lead to blood-lead
relationship was reported as being statistically insignificant.

       hi summary, there is much difficulty in distinguishing between direct and indirect
exposure in cases where external dust-lead levels is closely related to levels in other sources of
environmental lead. Correlation and univariate regressions with external dust-lead and blood-
lead fail to account for the possibility that external dust-lead by itself may only play a small part
in aggregate lead exposure when other sources of lead and exposure pathways are considered.
Multivariate regressions using lead measurements from multiple sources do not solve this
problem due to problems with colinearity.  The preferred approach would be to use structural
equation models, which allow multiple source and exposure pathways to be modeled in a
reasonable way, but this approach requires more effort in terms of implementation and
interpretation of the model, and is not well-reported in the literature. Therefore, quantitative
estimates of the effect of external dust-lead on children's blood-lead concentrations have not
been well established in the literature.

Air Ducts

       Most of the encountered articles provided only preliminary information on lead exposures
associated with ah* ducts. It is unclear to what extent dust-lead in air ducts is accessible to
children. Children would not typically be expected to encounter the dust lodged in air ducts
directly.  One case study found that dust-lead levels in living areas outside of contaminated air
ducts can be orders of magnitude lower than what is found in the air ducts.  However, if dust in
air ducts is disturbed, it is more likely to be introduced to the air and to nearby surfaces with
which children can come into direct contact. In particular, HVAC ductwork removal can yield
extensive contamination of surfaces in the general area of the ductwork.

       Only one study (the Baltimore R&M Study) estimated (in a quantitative manner) the
association between blood-lead concentrations in children and dust-lead levels found in air ducts.
This relationship was expressed as a simple correlation coefficient.  Unlike correlations between
blood-lead concentrations and dust-lead levels on other surfaces, the correlation coefficient

                                          174

-------
involving dust-lead levels from air ducts was not significant at the 0.05 level.  However, this
analysis was based on a small sample size and did not adjust for the effects of other exposure
variables such as lead in floor-dust and soil. Moreover, as evidence of a significant correlation
was observed between air duct dust-lead levels and lead levels on other surfaces, such as floors,
even significant correlation coefficients should not be interpreted as the degree to which air duct
dust causes a change in blood-lead concentration. In order to characterize the pathway of lead
from air ducts to children's blood, additional data collection and analyses are needed.

Window Troughs

       Of the seven studies listed in Table 3-32 above that collected information on dust-lead in
window troughs (also known as window wells), three also collected blood-lead data from
resident children. Correlation coefficients between blood-lead levels and window-trough dust-
lead concentrations in the R&M study ranged from 0.20 to 0.39, and correlation coefficients
between blood-lead levels and window-trough dust-lead loadings ranged from 0.06 to 0.44. The
correlations between dust-lead concentrations and blood-lead concentrations were statistically
significant in 4 of the 5 sampling campaigns, and the correlation between dust-lead loading and
blood-lead concentration was statistically significant only in the pre-maintenance sampling. In
the Rochester Study, correlation coefficients between blood-lead concentrations and dust-lead
loadings were 0.35 for the BRM samples, 0.31 for the DVM samples, and 0.29 for wipe samples,
while correlation coefficients between blood-lead concentrations and dust-lead concentrations
were 0.23 for both BRM and DVM  samples.

       Previous analyses (Battelle,  1996a; Battelle, 1996b) have examined whether the
predictive ability of a model improves when adding window trough lead levels to a model which
already accounts for dust-lead on floors and window sills. Results of these analyses on the
Rochester Lead-in-Dust Study and Baltimore R&M study data indicated that the estimated effect
of window trough dust-lead on blood-lead was either not statistically significant or only
marginally significant after adjusting for the effects of floor lead, sill lead, and temporal
.variation. A pathways analysis (USEPA, 1998c) using structural equations modeling concluded
that window troughs were a significant pathway for lead exposure, both as a direct pathway of
lead to children's blood-lead concentration (seen when Rochester Lead-in-Dust study data were
analyzed) and as an indirect pathway through window sills and floors to blood-lead
concentrations (seen  when both the Rochester and Baltimore R&M study data were analyzed).

       In summary, the association  between blood-lead concentrations and window trough dust-
lead has been well established in the literature. The more difficult question of the degree to
which window troughs contribute directly or indirectly to children's lead exposure is not well
established.
                                           175

-------
Upholstery

       Table 3-33 in the previous subsection included results for two studies which measured
both children's blood-lead concentrations and dust-lead on upholstery.  In the Baltimore R&M
study, correlation coefficients were calculated between blood-lead concentrations in children and
both the loading and concentration of lead in upholstery dust. These correlation coefficients
ranged from 0.19 to 0.61 for dust-lead concentrations and from 0.06 to 0.47 for dust-lead
loading. These correlations were statistically significant in the pre-intervention sampling.  As
with most of the other surfaces discussed in this section, upholstery dust-lead levels were not
included in any analyses to determine which lead sources were most significantly related to
blood-lead levels. In the Mexico City Study, the correlation between upholstery dust-lead levels
was not statistically significant, resulting in the absence of upholstery dust-lead levels from
models linking blood-lead levels and environmental lead levels.

       The results of the two studies assessing the importance of upholstery dust as a source for
lead exposure in children differ. In one case, the relationship between blood-lead and upholstery-
dust-lead is significant, while in the other it is not. Moreover, as upholstery dust-lead is often
correlated with other lead exposure variables, such as floor dust-lead and soil-lead, as cautioned
earlier, the positive correlation coefficient should not be interpreted as the degree to which
upholstery dust causes a change in blood-lead concentration. In order to characterize the
pathway of lead from upholstery to children's blood (and perhaps hands), additional data
collection and analyses are needed.

3.5.3 Implications of the Available Information For  Regulatory Standards

       Two primary questions related to the need and feasibility of regulatory standards for dust-
lead on surfaces other than floors and window sills are:

        1.     Is there sufficient information available on which to base a standard?

       2.     Is the standard necessary to either identify a lead hazard at a residence or to
              characterize the risk to determine appropriate corrective actions?

       The answers to these questions are discussed for each surface type below.

Exterior Dust

       In general, there is a fair amount of data on exterior dust, including studies where exterior
dust has been measured along with other lead exposure variables and blood-lead concentrations.
The amount of data implies that analyses could be conducted to provide a quantitative basis for
an exterior dust standard. However, implementation and interpretation of such analyses for
exterior dust will face many difficulties. For example, in many of the studies it is difficult to
distinguish between exterior dust and soil samples because of aggregation of the samples or of
the measurements.  Some external sampling for lead was carried out  using surface scrapings

                                           176

-------
which measures lead levels from a mix of both soil and dust-lead and some analyses averaged the
external soil and dust measurements and recorded the value as a single external lead
measurement.  (Hence only studies for which a clear distinction between external soil and dust-
lead levels is possible were included in this summary.)  It is also difficult to determine what
locations for exterior dust should be included. Should the focus be on enclosed spaces or also
include unenclosed areas such as sidewalks, stoops, and unenclosed porches? One primary
reason to focus only on enclosed areas is because exposures to unenclosed areas are not under the
direct control of property owner. Exposure and cleaning scenarios for enclosed versus
unenclosed areas are likely to be very different as well.  In conclusion, decisions on the specific
focus of a standard for exterior dust would impact the feasibility of establishing a good
quantitative basis on which to set the standard.

       The question of whether a standard for exterior dust is necessary is also a difficult one,
for which the literature does not have a clear answer.  While it is reasonable to assume that
measurements of lead in interior dust and exterior soil might capture a lead hazard if one exists,
there is not a strong body of information on which to base this conclusion. A separate standard
may not be necessary if risk assessors are aware of the potential hazard from exterior dust, and
include testing or corrective actions in cases where it is suspected to be an important pathway of
exposure (for example, in the case where a child spends a considerable amount of time on a
paved surface, such as a driveway or patio).

Air Ducts

       There is insufficient data upon which to develop a hazard standard for lead in air duct
dust, or upon which to draw conclusions about the necessity of a standard to either identify a
hazard or determine corrective actions.

Window Troughs

       The fact that regulatory standards have been proposed for dust-lead on floors and window
sills based on data sets (most notably the Rochester Lead-in-Dust Study and the HUD National
Survey) that also include window troughs implies that sufficient data exists on which to base a
standard for window troughs.

       However, while there is sufficient information on which to base a standard, analyses
conducted to assess the necessity of a window trough standard given the existence of a floor and
window sill standard suggest that a window trough standard may not be necessary to identify a
residence with a lead-based paint hazard. These analyses include the sensitivity/specificity
analyses included in a companion §403 report as well as the analyses that examine the effect of
adding window troughs to a statistical model that already includes floors and window sills
(Battelle, 1996a; Battelle, 1996b).  Given the correlation between window trough and window
sill lead levels, it is likely that if more sampling is to be done beyond a minimal risk assessment,
more benefit will be obtained from sampling more windows at the sill rather than sampling fewer
windows but at both the sill and trough. Moreover, cleaning of window troughs is

                                          177

-------
recommended for all homes that require a dust intervention, and clearance standards have been
proposed to guide assessment of the effectiveness of the cleaning. For these reasons, it does not
appear that an additional standard for window troughs is necessary either to identify a home with
a hazard or to guide corrective actions.

Upholstery

       There is insufficient data upon which to develop a hazard standard for lead in upholstery
dust, or upon which to draw conclusions about the necessity of a standard to either identify a
hazard or determine corrective actions.
3.6    DISTRIBUTION OF CHILDHOOD BLOOD-LEAD

       This section updates the information presented in Section 3.4 of the §403 risk analysis
report on the distribution of childhood blood-lead concentration hi the United States, with a focus
on the  1-2 year (12-35 month) age range as the population of interest. In addition to a national
characterization based on data from Phase 2 of the Third National Health and Nutrition
Examination Survey (NHANES ffl), Section 3.4 of the §403 risk analysis report summarized data
from other studies (e.g., the Baltimore R&M study, the Rochester Lead-in-Dust study, and the
HUD Grantees evaluation) to provide supporting information on the prevalence of elevated
blood-lead concentrations in children living in urban locations and in older housing or housing
likely to contain lead-based paint. Blood-lead data from these other studies were also considered
because the NHANES DI did not collect environmental-lead data, despite having the most
nationally representative data on blood-lead levels.

       Section 3.6.1 below is an update of Section 3,4.4 of the §403 risk analysis report. It
contains revised data summaries of pre-intervention blood-lead concentrations hi children
monitored within the HUD Grantees evaluation and revised regression model fits to predict
blood-lead concentration as a function of dust-lead loading for each individual grantee, as well as
for the Rochester Lead-in-Dust study (i.e., the study that provided the data used to developed the
empirical model developed within the §403 risk analysis. These revisions were possible as
additional pre-intervention data from the HUD Grantees evaluation (through 1/99) have been
made available to the risk analysis since the report was released.

       Section 3.6.2 provides information from the Cincinnati Prospective Lead study (Clark et
al., 1985) and summarized by the Centers for Disease Control and Prevention (CDC) on the
relationship between children's blood-lead concentration and housing age/condition, and how
this relationship may change with the age of the child (CDC, 1991).  CDC used this  information
in their recommendations for blood-lead screenings of young children.
                                           178

-------
3.6.1  Evaluation of the HUD Lead-Based Paint Hazard Control
       Grant Program ("HUD Grantees")

       Blood-lead concentrations of children residing in households participating in the
evaluation phase of the HUD Grantees evaluation (Section 3.2.2.3 of the §403 risk analysis
report) were measured, along with environmental-lead levels in various media. The population
of children targeted for participation in the program differed among the fourteen grantee
recipients, due to the different enrollment criteria among the grantees (see Table 3-4 of the §403
risk analysis report). These criteria included targeting high-risk neighborhoods, enrolling only
homes with a lead-poisoned child, and considering unsolicited applications. Pre-intervention
data collected through January 1999 are presented in this section; these data provide some of the
most recent information on the relationship between children's blood-lead concentration  and
environmental-lead levels.

       Across all grantees, pre-intervention blood-lead concentration data through 1/99 were
available for 526 children aged 1-2 years and for 764 children aged 3-5 years. For these children,
Table 3-34 summarizes measured blood-lead concentration for each combination of blood
collection type (venipuncture, fingerstick) and age of child (1-2 years, 3-5 years, and 1-5  years).
Table 3-34 also summarizes measured blood-lead concentration for children aged 1-2 years for
each combination of blood collection type and grantee. Note that fingerstick methods were
predominant for Wisconsin, Milwaukee, and Vermont, while Rhode Island used both methods
for similar numbers of children. The remaining nine grantees (excluding New Jersey) used the
venipuncture  either exclusively or predominantly.

       According to Table 3-34, the geometric mean blood-lead concentration via the
venipuncture  collection method was 9.3 (Jg/dL for children aged 1-2 years and 8.0 ng/dL for
children aged 3-5 years.  In contrast, the geometric means based on data from Phase 2 of
NHANES UI were 3.1 ug/dL for children aged 1-2 years and 2.5 ng/dL for children aged 3-5
years (Table 3-36 of the §403 risk analysis report). The larger values in the HUD Grantees
evaluation reflect the HUD Grantees program's procedure of selecting  high-risk children for
monitoring. The differing enrollment criteria across grantees also contributed to considerable
differences in the geometric mean blood-lead concentration among the grantees.

       Under venipuncture, the geometric means of children aged 1-2 years for Individual
grantees reporting more than three blood-lead results ranged from 4.2 ug/dL (California,  which
only targeted  older units) to 15.9 ug/dL (Cleveland, which targeted units with lead-poisoned
children).

       The geometric mean blood-lead concentration via the fingerstick collection method was
9.4 ug/dL for children aged 1-2 years and 8.9 ug/dL for children aged 3-5 years. When data were
available for more than one child under fingerstick collection methods, the geometric means for
children aged 1-2 years ranged from 5.9 ug/dL (Wisconsin) to 13.5 ug/dL (Milwaukee).
                                          179

-------
Table 3-34.  Summary of Children's Pre-lntervention Blood-Lead Concentration in the
              HUD Grantees Evaluation According to Blood Collection Method, Child Age
              Category, and Grantee (ages 1 -2 years only)
-'..;' '
: Age Category
1-2 Years
3-5 Years
1-5 Years
'"-I" Grantee
Alameda County
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
New Jersey
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
Age Range
1-2 Years
3-5 Years
1-5 Years
: Grantee
Cleveland
Massachusetts
Minnesota
Rhode Island
Wisconsin
Milwaukee
Vermont
Number
of
ChBdren
Hood-Lead Concentration fe/g/dU
Arithmetic
Mean •
; ' '_.
.Mean
Geometric
Standara •
Deviation
Minimum
Rood Collection Method -
361
536
897
12.5
10.6
11.4
9.3
8.0
8.5
2.3
2.2
2.2
0.7
0.0
0.0
25th
Pefcentife

Median

;••:•>,•-• ' :'".
-•ITWir-/
rarcanuie

Maximum
= Vehipuncture ' .
5.4
4.5
5.0
10.0
8.6
9.0
17.0
15.0
16.0
53.0
48.0
53.0
Blood Collection Method = Venipuncture {Children Aged 1-2 Years only}
27
25
20
21
64
43
75
1
14
9
3
28
23
8
6.5
9.3
12.7
5.3
19.3
11.2
14.5
3.0
10.0
10.2
26.0
13.9
5.2
13.5
4.7
7.7
10.4
4.2
15.9
9.1
10.7
3.0
8.1
8.7
25.1
11.7
4.7
12.4
2.2
1.9
2.0
2.0
1.9
1.9
2.4
-
2.0
1.8
1.4
2.0
1.6
1.6
1.4
2.0
3.0
1.4
4.0
3.0
0.7
3.0
2.0
4.0
18.0
1.0
2.0
6.0
: hi Blood I Collection Method
164
232
396
11.7
11.5
11.6
9.4
8.9
9.1
1.9
2.0
2.0
2.0
2.0
2.0
3.0
6.0
6.0
3.2
11.5
6.0
6.0
3.0
6.0
6.0
18.0
9.5
4.0
8.5
4.7
7.0
14.5
3.8
17.0
9.0
11.0
3.0
8.5
8.0
25.0
12.0
5.0
14.5
6.6
10.0
19.0
6.0
28.0
16.0
22.0
3.0
14.0
12.0
35.0
19.0
7.0
17.0
24.8
26.0
27.0
16.9
53.0
40.0
43.0
3.0
21.0
24.0
35.0
35.0
12.0
22.0
= Fingerstick '•. •• ,';:< - ' •" ' <„•'
6.0
5.0
5.0
9.0
9.0
9.0
15.0
15.0
15.0
48.O
62.0
62.0
Blood Collection Method = Fingerstick (Children Aged 1-2 Years only)
1
4
1
9
43
82
24
13.0
7.8
33.0
8.8
6.2
16.0
7.9
13.0
7.3
33.0
8.2
5.8
13.2
7.0
-
1.5
-
1.5
1.4
1.9
1.6
13.0
4.0
33.0
5.0
3.5
2.0
3.5
13.0
6.0
33.0
7.0
4.0
9.0
5.0
13.0
8.5
33.0
7.0
6.0
14.5
6.5
13.0
9.5
33.0
11.0
8.0
20.0
11.0
13.0
10.0
33.0
15.0
14.0
48.0
16.0
Note: All pre-intervention blood-lead concentration data available and collected through 1/99 are included in the above
summaries.
                                             180


-------
       The percentages of children with elevated blood-lead concentrations (Le., concentrations
at or above 10,15,20 or 25 ug/dL) at pre-intervention are summarized hi Table 3-35.  According
to this table, 51 percent of children aged 1-2 years sampled via venipuncture methods had blood-
lead concentrations at or above 10 ug/dL, compared to the estimates of 5.88% for Phase 2 of
NHANES m, 53.8% for the Baltimore R&M study (pre-intervention), and 23.4% for the
Rochester Lead-in-Dust study (Tables 3-37, 3-41, and 3-42, respectively, of the §403 risk
analysis report). For individual grantees having more than three children with a measured blood-
lead concentration, the  percentage of children aged 1-2 years with blood-lead concentrations
(venipuncture) at or above 10 ug/dL varied from 4% (New York City, which targeted housing
and neighborhoods rather than lead-poisoned children) to 80% (Cleveland).  The range of
percentages under the fingerstick method were similar to that under the venipuncture method, but
less data were available to estimate them.

       Figures 3-20 and 3-21 illustrate the nature of the linear relationship observed in the HUD
Grantees evaluation between a child's (log-transformed) blood-lead concentration and the
household's (log-transformed) area-weighted arithmetic average wipe dust-lead loading for floors
and window sills, respectively. The figures portray fitted linear regression models for each
grantee, as well as for the Rochester Lead-in-Dust study and, in Figure 3-21, the Baltimore R&M
study (for comparison purposes). The regression model used only the log-transformed average
dust-lead loading as a predictor variable; the impact of other potentially important predictor
variables on blood-lead concentration was not considered in the model fittings.  The regression
lines span the ranges of the observed area-weighted average dust-lead loadings, except data for
five HUD Grantees households (three from Cleveland and one each from Baltimore and Rhode
Island) were omitted from Figure 3-21 as their average window sill dust-lead loadings were
extremely low (less than 0.05 ug/ft2) compared to the other households and were considered too
influential to the model fittings.

       When fitting the regression models in Figures 3-20 and 3-21 to the HUD Grantees data, it
was desired to have each household having blood-lead and dust-lead data be represented by only
a single data point. This was possible only if blood-lead data were considered for a single child
in that household. In situations where data for multiple children were available for a single
household, only data for the youngest child older than 12 months of age were considered.  This
approach resulted in a single blood-lead result for each household with blood-lead data. In
addition, only data for children meeting the following criteria were included in the regression
modeling:

       •      Children who lived in the sampled housing unit for at least three months and
              before dust and soil samples were collected;

       •      Children whose blood samples were taken within four months of dust and soil
              sample collection;

       •      Children not having medical treatment for lead poisoning.
                                           181

-------
Table 3-35.  Percentage of Children with Elevated Blood-Lead Concentration (at Pre-
              Intervention) in the HUD Grantees Evaluation According to Blood Collection
              Method, Child Age Category, and Grantee (ages 1-2 years only)

Age Range
1-2 Years
3-5 Years
1-5 Years
*"...•'• Grantee ' . •;::!.*
Alameda County
Baltimore
Boston
California
Cleveland
Massachusetts
Minnesota
New Jersey
Rhode Island
Wisconsin
Milwaukee
Chicago
New York City
Vermont
Age; Range
1-2 Years
3-5 Years
1-5 Years
Grantee
Cleveland
Massachusetts
Minnesota
Rhode Island
Wisconsin
Milwaukee
Vermont
Number of
• Chadren
Percentage of C
ilOpg/dL

hQdren mUi Elevate
"I i 15#g/dL
id Blood-Lead Concentration (%) ,-,
220pg/dL:
•» 25jig(dL
Rood Collection Method = Venipunctura
361
536
897
51
43
46
^ Hood Conection Method =
27
25
20
21
64
43
75
1
14
g
3
28
23
8
22
36
55
14
80
47
61
0
36
44
100
75
4
63
35
27
30
18
12
14
12
7
9
= Venipuncture (Ctiiidren Aged 1-2 Years only) -;/-
11
20
50
5
59
30
44
0
21
22
100
39
0
50
4
8
15
0
38
9
31
0
7
11
67
14
0
13
0
4
5
0
30
7
21
0
0
0
67 .
7
0
0
Blood Collection Method « Fingerstick -.
164
232
396
46
44
45
28
26
27
13
15
14
9
8
9
Bood; Collection Method — Rngerstick (Children Aged 1-2 Years onlyl •
1
4
1
9
43
82
24
100
25
100
33
9
70
29
0
0
100
11
0
50
13
0
0
100
0
0
26
O
0
0
100 .
0
O
17
0
Note: All pre-intervention blood-lead concentration data available and collected through 1/99 are included in the above
summaries.
                                            182

-------
     100-
      10
          0.1              1.0             10.0             100.0           1000.0          10000.0

           Carpeted and Uncarpeted Floor Wipe Area-Weighted Arithmetic Average Dust-Lead Loading (
       A A A AtamodB Canty
       C E E Ctevetand
                                                                            ».i	IA..I. rw
                                                                            IWW flNII Mqr
Figure 3*20.  Fitted Regression Models Predicting Children's Blood-Lead Concentration as
              a Function of Area-Weighted Arithmetic Average Floor Dust-Lead Loading
              (Wipe Collection Method), for the Various Grantees in the HUD Grantees
              Evaluation and for the Rochester Lead-in-Dust Study

(Note: Venipuncture blood-lead data were exclusively used in each fitting except for Wisconsin. Milwaukee, and Vermont,
where fingerprick blood-lead data were exclusively used.)
                                             183

-------
     100
      10-
       1-
          0.1
  1.0           10.0         100.0        1000.0       10000.0
Window Sill Area-Wteighted Arithmetic Average Dust-Lead Loading (
                 100000.0
       A A A Afcmdi County
       U O  H BflDlNMI
       f? r*  r- ii.,..,,ii,.,.,di.
       I  I  r MRBaBGnUSBm
       Vialf—i M |«WHj(m
       O	O  O BflSfenore RAM
B B  B CMRMnb
I  I  I  Rhodotd
aj itj  ij >.«	vhA
ifi m  in now IUK
Figure 3-21.  Fitted Regression Models Predicting Children's Blood-Lead Concentration as
              a Function of Area-Weighted Arithmetic Average Window Sill Dust-Lead
              Loading (Wipe Collection Method), for the Various Grantees in the HUD
              Grantees Evaluation and for the Rochester Lead-in-Dust Study

(Note: Venipuncture blood-lead data were exclusively used in each fitting except for Wisconsin, Milwaukee, and Vermont,
where fingerprick blood-lead data were exclusively used.)
       The regression models in Figures 3-20 and 3-21 were fitted to blood-lead concentration
data only under the venipuncture method for all but the three grantees (Wisconsin, Milwaukee
and Vermont) for which fingerstick sample results were predominant. For these three grantees,
only fingerstick blood-lead concentration data were used in the regressions.

       Note that the slopes of the fitted regression lines in Figures 3-20 and 3-21 are generally
similar in sign and magnitude (given expected ranges of variability) across the grantees and the
two other studies. This suggests that the relationships between blood-lead concentration and
household average dust-lead loading were relatively consistent across grantees.  In particular,
                                             184

-------
these relationships were similar to that observed for data from the Rochester study (i.e., the data
used to develop the empirical model presented in Chapter 4 of the §403 risk analysis). This
conclusion is important in that the data from the HUD Grantees evaluation reflect a much larger
geographical area than the Rochester study and represent several types of exposure conditions.

3.6.2  Evidence of the Impact of Housing Age/Condition
       on Blood-Lead Concentration

       The role that housing age plays in the increased likelihood of a resident child having an
elevated blood-lead concentration has been well-documented and is accepted by many experts in
residential lead exposure. Older housing is more likely to contain lead-based paint hi a
deteriorated condition, which contributes to lead in other environmental media within the
residence, especially those media that is most likely to come into direct contact with children. In
particular, the importance that the level of deterioration plays in the accessibility of lead-based
paint hazards implies that housing condition is an additional key factor in predicting blood-lead
concentration.

       Table 3-39 of the §403 risk analysis report summarized data from Phase 2 of NHANES
m to illustrate how geometric mean blood-lead concentration and the percentage of elevated
blood-lead concentrations (i.e., percentage exceeding a given threshold) for children are related
to housing age category.  For example, the percentage of children aged 1-5 years with blood-lead
concentration of at least 10 ug/dL increases from  1.6% for children living in post-1973 housing
to 8.6% for children living in pre-1946 housing, with a corresponding geometric mean increase
from 2.0 to 3.8 ug/dL. The Centers for Disease Control and Prevention (CDC) cited these same
results in their 1997 document, Screening Young Children for Lead Poisoning, to support their
conclusion that older housing (i.e., housing built prior to 1950) contained the greatest risk for
lead-based paint hazards.

       Figure 6-1 of the CDC's 1991 document, Preventing Lead Poisoning in Young Children
-A Statement by the Centers for Disease Control, presents results from the Cincinnati
Prospective Lead Study (Clark et al., 1985) to illustrate how the combination of housing age and
condition is related to children's blood-lead concentration and how this relationship changes with
the age of the child. This figure is duplicated in Figure 3-22. This figure shows that children's
blood-lead concentration tends to peak at 18-24 months, with the most rapid increase occurring
between 6-12 months. The highest blood-lead levels are associated with housing built prior to
World War n, as well as older housing (predominantly 19th century) that once contained
considerable lead-based paint but which later underwent rehabilitation. Within these groups of
.housing/children living in units in a deteriorated or dilapidated condition had consistently higher
geometric mean blood-lead concentrations through their first three years, with  this geometric
mean exceeding 20 ug/dL from about 12 to 24 months of age. CDC used the information
presented in Figure 3-22 to prepare a recommended screening schedule for testing children's
blood-lead levels.
                                          185

-------
                                                                             Deteriorated
                                                                       Sit-  $»fofact«T
                                                                             t»i#Wli
                                                                       fob,» NMic
          0369
12  13  1$  21  24  27  30   33 36  39  42
  Child's Age (in  Months)
Figure 3-22.  Geometric Mean Blood-Lead Concentration Versus Child Age, As Reported
              Within the Cincinnati Prospective Lead Study and Presented According to
              Housing Age and Condition

(Note: Duplicated from Figure 6-1 of CDC, 1991. Blood-lead concentrations for the same cohort of children
were measured over time.  Numbers in parentheses indicate numbers of children with blood-lead information at
18 months of age.)
                                          186

-------
4.0    DOSE-RESPONSE ASSESSMENT

       The objective of dose-response assessment was to characterize the relationship between
environmental-lead exposure and the resulting adverse health effects in young children.  The
foundation of this characterization was the relationship between environmental-lead levels and
blood-lead concentration. EPA's Integrated Exposure, Uptake, and Biokinetic (IEUBK) model
and an empirical model developed for this risk assessment (Sections 4.1 and 4.2 of the §403 risk
analysis report) were employed to make this characterization.

       Section 4.1 of this chapter documents an additional tool, obtained since the §403 risk
analysis report was published, for predicting blood-lead concentration as a function of
environmental-lead levels. This tool is a regression model developed from epidemiological data
collected from 12 studies and was suggested for use in the §403 risk analysis by some
commenters on the §403 proposed rule. As the U.S. Department of Housing and Urban
Development (HUD) sponsored the development of this model, it is referred to in this report as
the "HUD Model." The goal of this model was to "estimate the contribution of lead-
contaminated house dust and soil to children's blood-lead levels" (Lanphear et al., 1998).  This
goal is consistent with the objectives of the §403 risk assessment, and so the model merits
consideration for this analysis.  Section 4.1 includes documentation on the HUD model, key steps
that were taken in its development, and issues that are necessary to consider when interpreting
the results of the model fits.

       Section 4.2 of this chapter contains a revision of the Rochester Multimedia model,
introduced in Section 4.2.3 of the §403 risk analysis report, to allow the model to predict results
that are more comparable to the results of the performance characteristics analysis presented in
the preamble to the §403 proposed rule. The Rochester Multimedia model predicted a geometric
mean blood-lead concentration as a function of average dust-lead loadings in floors and window
sills, dripline soil-lead concentration, and a variable which indicates the presence of deteriorated
lead-based paint and a child with paint pica tendencies. In contrast, the performance
characteristics analysis in the preamble estimated risks associated with dust-lead loadings on
uncarpeted floors, dust-lead loadings on window sills, yardwide average soil-lead concentration,
and the percentage of painted components with deteriorated lead-based paint.  Because the
definitions of the data inputs were not always consistent between these two statistical
approaches, their findings were not comparable. Thus, the revised model presented in Section
4.2 uses the same types of data inputs as those used for the performance characteristics analysis.
Section 4.2 also documents multimedia models that omit one or more of the dust, soil, and paint
input variables to obtain predicted blood-lead concentration in instances where data for one or
more of these media were not available.

       Section 4.3 provides additional information regarding key assumptions made in the risk
characterization process: the "scaling" algorithm used to determine a post-intervention blood-
lead concentration distribution that is comparable to the baseline distribution, and the issue  of
                                          187

-------
adjusting for measurement error when deriving the empirical model used in the §403 risk
analysis.

4.1    HUD MODEL

       The HUD model (Lanphear et al., 199S) was developed by a team of researchers and
sponsored by HDD's Office of Lead Hazard Control. This modeling effort used data from 12
epidemiologic studies (hence its frequent reference as a "pooled analysis" model) to make
statistical inferences on the contribution of lead-contaminated house dust and residential soil to
children's blood-lead concentration.

       The HUD model predicts a geometric mean blood-lead concentration for children aged 6-
36 months (i.e., the age range of data considered from the 12 studies) as a function of exposure to
specific lead levels in dust, soil, paint, and water, and as a function of other important
demographic variables. Therefore, the model is used to estimate individual risks, or the risks
associated with a class of children determined by specified environmental-lead levels to which
they are exposed.  EPA addressed minimizing individual risks when establishing "levels of
concern" for lead in dust and soil within the §403 proposed rule. However, EPA was obliged to
consider population-based risks within a cost-benefit analysis to establish lead hazard standards
within the proposed rule.

4.1.1  Form of the HUD Model

       The HUD model takes the following form:

(1)   LnfPbB) =  1.496 + 0.183-Ln(DustLead) + 0.01398-Ln(WaterLead) + 0.02116-Ln(ExtLead) +
              0,005787-Ln(ExtLead)-ExtType +• 0.4802-Ln(ExtLead)'ExtLoc - 0.1336-ExtType •*• 0.5858-ExtLoc
              - 0.02199-Ln(MaxXRF) + 0.0381l-Ln(MaxXRF)-PaintCond - 0.0808-PaintCond + 0.02126-Age -
              0.001399-Age2 + 0.00007854-Age3 - 0.3932-Boston - 0.01167-Butte + 0.2027-Bcreek +
              0.2392-Cpgm + 0.5383-Csoil + 0.05717-Leadville +  0.1761-Magna - 0.04209-RochLong +
              0.07257-RochUD - 0.3712-Sandy + 0.1777-Midvale + 0.123-Race + 0.3175-SES1 + 0.2138-SES2
              + 0.1799-SES3 + 0.1691-SES4 - 0.03233-MouthQften - 0.2454-MouthRare - 0.1397-MouthSome
              + 0.002649-Ln(DustLead)-Age - 0.0003381-Ln(DustLead)-Age2 • 0.00001281-Ln(DustLead)-Age3
              + 0.2212-Ln(ExtLead)-MouthOften + 0.07892-Ln(ExtLead)-MouthRare +
              0.1663-Ln(ExtLead)-MouthSome + 0.5305-Ln(WaterLead)-SESl - 0.0136-Ln(WaterLead)-SES2 +
              0.1033-Ln(WaterUad)-SES3 - 0.09098-Ln(WaterLead)-SES4 + 0.01192-Age-Race -
              0.01023-Age-SESl + 0.003849-Age-SES2 + 0.00008468-Age-SES3 - 0.01679-Age-SES4 + error
where
       Ln(PbB) - log-transformed blood-lead concentration (ug/dL)
       Ln(DustLead) = log-transformed interior (wipe) floor dust-lead loading (ug/ft2), minus
              the mean of the log-transformed data used to develop the model (2.605 ug/ft2)
       Ln(WaterLead) = log-transformed water-lead concentration (ppb), minus the mean of the
              log-transformed data used to develop the model (0.785 ppb)
       Ln(ExtLead) = log-transformed exterior-lead concentration (ppm), minus the mean of the
              log-transformed data used to develop the model (6.232 ppm), where the exterior
                                            188

-------
             sample is either soil collected at the perimeter of the foundation, soil from the
             child's play area, or exterior dust
      ExtType = indicator of the type of exterior sample (1 = dust, 0 = soil)
      ExtLoc — indicator of whether exterior sample is represented by soil at the perimeter of
             the house's foundation (1 = exterior sample is not from perimeter soil, 0 =
             exterior sample is from perimeter soil)
      Ln(MaxXRF) = log-transformed maximum lead-paint measurement on interior surfaces
             (mg/cm2, as measured by XRF), minus the mean of the log-transformed data used
             to develop the model (0.921 mg/cm2)
      PaintCond = indicator of paint condition (1 = damaged, 0 = undamaged)
      Age = age of child (months) minus the mean age of children whose data were used to
             develop the model (16.3  months)
      Age2 - Age2 - (85.5 + 4.82-Age) (quadratic orthogonal polynomial)
      Age3 = Age3 - (-490.71 + 10.32-Age2 + 122.3-Age) (cubic orthogonal polynomial)
      Boston, Butte, Bcreek, Cpgm, Csoil, Leadville, Magna, RochLong, RochUD, Sandy, and
      Midvale are indicators that the data come from the particular study being represented (1 =
      data comes from the particular study, 0 = otherwise)
      Race = Race indicator (0 = white, 1 = other)
      SES1 = indicator of whether the pseudo-Hollingshead measure of socioeconomic status is
             equal to 1 (1 = yes, 0 = no)
      SES2 = indicator of whether the pseudo-Hollingshead measure of socioeconomic status is
             equal to 2 (1 = yes, 0 = no)
      SES3 = indicator of whether the pseudo-Hollingshead measure of socioeconomic status is
             equal to 3 (1 = yes, 0 = no)
      SES4 = indicator of whether the pseudo-Hollingshead measure of socioeconomic status is
             equal to 4 (1 = yes, 0 = no)
      MouthOften = indicator of whether mouthing behavior occurs often in the child (1 =
             often, 0 = otherwise)
      MouthRare = indicator of whether mouthing behavior occurs rarely in the child (1 =
             rarely, 0 = otherwise)
      MouthSome = indicator of whether mouthing behavior occurs sometimes in the child (1 =
             sometimes, 0 = otherwise)
      error = random error between the observed log-transformed blood-lead concentration and
             what is predicted by the model.

4.1.2 Development of  the HUD Model

      This section presents several issues on how the HUD model was developed that have a
direct impact on the predicted blood-lead concentration and how this prediction should be
interpreted. These issues include how studies were selected, how study effects were represented
in the model, and how data were handled or adjusted prior to or during  the model development
exercise.
                                         189

-------
Study Selection and Potential Selection Bias

       The HUD model was developed from environmental-lead, blood-lead, and demographic
data from 12 studies performed over a 15-year time frame (1982-1997). These studies
investigated the relationship between environmental-lead levels and children's blood-lead levels
in various locations and subpopulations. Five of the studies (representing 62% of the data used
to fit the model) were conducted in urban environments:

       •      Boston Longitudinal Study (Rabinowitz et al., 1985)
       •      Cincinnati Longitudinal Study (Bornschein et al., 1985b)
              Cincinnati Soil Study (Clark et al., 1991)
       •      Rochester Longitudinal Study (Lanphear et al., unpublished)
       •      Rochester Lead-in-Dust Study (Lanphear et al.,  1996a,b)

The remaining seven studies (representing 38% of the data used to fit the model) were conducted
in milling, mining, or smelter environments:

              Bingham Creek, Utah (1993)
       •      Butte, Montana (1990)
              Leadville, Colorado (1991)
              Magna, Utah (1994)
              Midvale, Utah (1989)
       •      Palmerton, Pennsylvania (1994)
              Sandy, Utah (1994)

According to Lanphear et al. (1998), these 12 studies were selected based on the following
criteria:

       •      The studies had well-defined sampling protocols for blood and environmental
              media (particularly dust, soil, and paint).
       •      The studies took measures of dust-lead levels, soil-lead levels, paint-lead content
              (via XRF), and paint condition.
       •      The original data were available and could be reanalyzed.
       •      Dust samples were collected via wipe techniques or by the Dust Vacuum Method
              (DVM)
       *      Dust samples were taken within three months of collecting blood samples from
              the resident child(ren) (to address seasonal variation in blood-lead concentration).
       •      Children were not selected on the basis of having a high blood-lead concentration.

In addition, only cross-sectional data were considered (i.e., data were not considered that could
reflect changes in environmental-lead levels over time).
                                          190


-------
       As a result of the inclusion criteria, data for at least seven other studies considered by
HUD were excluded from the model development process.  These studies and the primary
reasons for their exclusion (when specified within Lanphear et al., 1998) were
       •      UK Study (Davies et al., 1990) - dust collection method not wipe or DVM
       •      Boston and Baltimore segments of the EPA Urban Soil Lead Abatement
             Demonstration Study (USEPA, 1996a; Weitzman et al., 1993) - dust collection
             method not wipe or DVM
       •      Australian National Survey (1996) - lack of XRF paint-lead levels
             Baltimore R&M Study (USEPA, 1996c; USEPA, 1997c) - method for selecting
             children did not meet the criteria, and data were not available to the analysis
       •      Telluride, CO (1987) - reason for exclusion not given
       •      Trail, BC (1992) - reason for exclusion not given.

       Lanphear et al. (1998) indicates that the 12 studies were not chosen to represent the entire
nation or even communities like those in which the studies were conducted.  In fact, these studies
were conducted in communities with a recognized environmental-lead hazard, and any abatement
efforts within each study targeted those hazards. Furthermore, the study effects included in the
model were treated as fixed effects (i.e., they are the only studies of interest in the  model-
building process) rather than random effects (i.e., they are assumed to be a random sample of a
larger population of studies).  Thus, if the model is used to estimate risks to a broader population
of children than simply those within the 12 studies, additional information is needed to determine
the extent to which the pooled data used to develop the model are representative of the U.S.
housing stock.

Fixed vs. Random Study Effects and Interaction with the Study Effects

       As stated in the previous paragraph, the  study effect in the HUD model is a series of fixed
effects. If the study effect was assumed to be random instead of fixed (i.e., the studies can be
considered a random selection of all such residential-lead exposure studies), then study-to-study
variation would become a contributor to total variation in the prediction. Based on work with
previous models that incorporate a random study effect, the study-to-study component is typically
a major portion of total variability in the prediction. Thus, the variability associated with
predictions by the HUD model is likely underestimated.

       Additional underestimation in variability may result from the absence of interaction terms
in the model between study effects and other environmental exposure factors. This can
underestimate variability associated with inferences involving the environmental exposure
factors, including the principal inferences which involve dust-lead and exterior-lead levels.

Adjusting for Measurement Error in Environmental-Lead Predictor Variables

       The HUD model parameter estimates were determined from Simulation Extrapolation
(SIMEX) methods, which attempted to quantify the theoretical relationship between blood-lead

                                         191

-------
and "error-free" measures of environmental-lead. As a result, the parameter estimates were
adjusted to reflect measurement error present among the environmental-lead variables.  However,
the goal is to predict children's blood-lead concentrations as a function of wipe dust-lead
loadings and soil-lead concentrations as they would be measured in a risk assessment, not their
true, "error-free" (but unobservable) values. Carroll et al. (1995) states that for prediction
problems, adjusting for the effects of measurement error in predictor variables is rarely
necessary.  Adjusting for measurement error in these predictor variables tends to increase the
values of the slope parameters associated with these variables, which in turn can inflate predicted
blood-lead concentrations. Thus any predictions from the HUD model fits should be properly
labeled that values of the predictor variables are assumed to be "error-free" rather than measures
of environmental-lead levels taken from activities such as a risk assessment (as the predictor
variables hi the models used in the §403 risk analysis were assumed to represent).

Making Survey Variable Definitions Consistent Across Studies

      Because different survey designs in different studies can result in different definitions for
a common survey measure (e.g., SES, mouthing behavior, paint condition), which in nun can
introduce considerable complication when interpreting model predictions, certain survey
measures were redefined to make them more consistent across studies. Each redefinition
transforms the original data values to values generated from a domain that is consistent across the
studies.  However, such a transformation does not remove all study-to-study differences in these
values.  In particular, it does not consider factors that impact how the specific study
measurements were obtained and which differ from study to study, such as the use of different
survey instruments and different approaches to administering the instruments.

Converting DVM Dust-Lead Loadings to Wipe-Equivalents

      While it was desired to have floor dust-lead loading assuming wipe dust collection as a
predictor variable in the HUD model, some of the 12 studies used DVM methods to collect dust
samples. Rather than exclude data from these studies from consideration in the model
development effort, a procedure was derived to convert DVM dust-lead loadings reported in
these studies to wipe-equivalent loadings. This procedure used data from the Butte study to
develop the following conversion equations:

                                                         for carpeted floors,
                                                         for hard floors.
                         = 0.7727 + 0.9821-Log^DVM)
             Log^Wipe) = 0.7762 + 0.4839-Log,JDVM)
where "Wipe" and "DVM" indicate wipe dust-lead loadings and DVM dust-lead loadings,
respectively (Westat, 1998). Note that these same two equations were used to convert DVM
dust-lead loadings in each study, regardless of whether the relationship between DVM and wipe
dust-lead loadings differed among the studies.

       The method for deriving these conversion equations included a procedure to adjust for
measurement error in the DVM dust-lead loading measurements.  However, the purpose of the

                                          192

-------
conversion was to predict a measured wipe dust-lead loading based on a measured DVM dust-
lead loading, not the (unobservable) "true" DVM dust-lead loading. Thus, some bias may have
been introduced in this conversion process.

Interpreting the "Exterior-Lead" Predictor

       Certain households whose data were used hi the HUD model development effort did not
have soil-lead data for various reasons (e.g., no bare soil). In these instances, exterior dust-lead
concentration was generally measured instead. As a result, among the predictor variables in the
HUD model was an indicator variable that identified whether or not an exterior-lead
measurement was from dust or soil. This indicator allowed both the model intercept and the
slope factor associated with exterior-lead concentration to change  according to its value.
However, no other consideration was made for differences in sampling and analysis methods
between soil and exterior dust and the impact such differences can have on the reported lead
levels. In addition, no indication was given that differences in bioavailability between soil-lead
and exterior dust-lead were considered in developing the model.

       When a household had soil-lead data available, data from foundation perimeter  (i.e.,
dripline) soil were used when available; otherwise, play-area soil-lead data were used.  Like the
soil vs. exterior-dust issue in the previous paragraph, the HUD model includes an indicator
variable that identified whether or not soil-lead levels represent dripline soil. This indicator
allowed both the model intercept and the  slope factor associated with exterior-lead concentration
to change according to its value. However, certain study-to-study  differences in collecting soil
samples or obtaining a soil-lead measurement were not considered in model development, such
as the depth of soil sampling, soil surface type (e.g., covered vs. bare), chemical methods for the
digestion and analysis of soil samples, and soil compositing.

Handling Missing Water-Lead Measurements

       While the HUD model included water-lead concentration as a predictor variable, it was
necessary to impute values for this measurement during model development when data for a
given household were not available. Water-lead data were unavailable for all households in two
studies and up to 12% of study households in the other ten studies. In such instances, imputed
measurements were randomly generated from a lognormal distribution with geometric mean
equal to that observed from data for other study households (if data for other households were
available) or to the community-wide average (if data for other households were not available).

Handling Data Reported Below a Detection Limit

       When data values reflected measurements at or below some detection limit, the  HUD
model development effort replaced them with probability-based values between zero and the
detection limit, where probabilities were determined from a lognormal distribution associated
with data above the detection limit. According to Table 2.15 of Westat (1998), the incidence of
not-detected results was high with XRF paint-lead level, and to a lesser extent, with water-lead

                                          193

-------
concentration. In half of the studies, the percentage of not-detected paint-lead results ranged
from 20% to 86% (with the detection Limit being reported at either 0.1 or 0.7 mg/cm2). The
percentage of not-detected water-lead concentrations (i.e., results somewhere below 5 ppb)
exceeded 80% in two studies.  In contrast, none of the lead levels in blood, dust, or soil samples
were reported below a detection limit in 10 of the 12 studies, and lead levels in no more than 9%
of these samples were reported below a detection limit in the other two studies.

Selecting from Multiple Observations Within a Household

       When blood-lead concentration data were available for multiple children within a
household (as occurred in nine of the studies), only data for one child selected at random from
the household were considered in the HUD model development. If blood-lead data existed at
multiple time points for the same child (such as at 6,18, and 24 months of age in the Boston
Longitudinal study), those time points whose data met the initial inclusion criteria were identified
(e.g., lead interventions did not occur between the time points), and data for the time point
having dust-lead measurements taken more closely in time were selected for the model
development. Similarly, when environmental-lead measurements were repeatedly taken over
time for a given household, data for the time point closest to a blood-lead measurement were
used.

Handling Seasonalitv Effects

       The effect of seasonality on blood-lead concentrations was given some consideration in
the modeling effort (e.g., blood and dust samples must have been collected within three months
of each other for their data to be included). However, there is no effect of seasonality included hi
the final model.  It is unclear whether seasonality was determined not to be a significant effect
among the pooled data, or whether a seasonality term was intentionally left out of the model.

4.1.3  Interpreting Results of Fitting the HUD Model

       The previous section discussed issues concerning the pooled study data and development
of the HUD model that should be understood when using the model to estimate risks, as is done
in Section 5.1.1 and Appendix F. This section addresses the interpretation of results from fitting
the HUD model and, in particular, caveats associated with certain interpretations.

Individual risks vs. population-based risks

       As mentioned earlier, the HUD model estimates individual risks associated with lead
exposure to children aged 6-36 months.  While the §403 risk analysis included individual risks
analyses, which EPA used in efforts to establish levels of concern for lead in environmental
media, EPA was required to employ cost-benefit analysis to select §403 hazard standards.  The
cost-benefit analyses used population-based risks (i.e., risks posed by childhood lead exposure to
the nation as a whole) to estimate the benefit and cost associated with performing interventions
and other activities in response to §403 rules.

                                          194

-------
       Individual risks and population-based risks are generally not comparable. This must be
understood when attempting to compare the individual risks estimated by fitting the HUD model
at specified environmental-lead levels with the population-based risk estimates found in the §403
risk analysis.

Interpreting Model Parameter Estimates

       The prediction parameters in the HUD model are not independent.  For example, it is
known that soil-lead and dust-lead measures are correlated. Therefore, it is not appropriate to
interpret the parameter estimates in the HUD model (or in the models developed for the §403 risk
analysis) in isolation.  Using the parameter estimates to characterize a cause-and-effect
relationship that is attributable to a single parameter alone, such as measuring the extent of an
increase in blood-lead concentration associated with a given increase in dust-lead loading, is very
problematic.

       One example of how correlation among the predictor variables can influence the model
parameter estimates is seen with maximum XRF paint-lead measurement.  One would expect a
positive correlation between maximum XRF paint-lead measurement and blood-lead
concentration, and as a result, a positive slope parameter.  However, the estimated slope
parameter is negative (-0.022), although not significantly different from zero. The negative
estimate is likely due to confounding between paint-lead measurements and other predictor
variables.  The likelihood of confounding increases with the number of parameters in the model.

       Problems with interpreting model parameter estimates in isolation emphasizes the need to
consider total exposure (i.e., prediction based on considering the joint effect of all model
parameters) rather than exposure associated with a single environmental medium. In the  §403
situation, protectiveness needs to be judged by recognizing that hazard standards exist for dust,
soil, and paint, and that resulting actions from these multiple standards will determine the level
of protection, not just the actions associated with a single standard. For example, the level of
protection associated with a dust-lead loading standard of 5 ng/ft2, without consideration  of other
standards, may equal that associated with a joint set of standards that involve a higher dust-lead
loading standard.

Interpreting Results at Low Environmental-Lead Exposures

       The HUD model and the models developed for the EPA risk analysis are "log-log"
models. That is, they predict log-transformed blood-lead concentration as a linear function of
log-transformed environmental-lead levels. As the log transformation "stretches out" the lower
portion of the scale and contracts the upper portion of the scale, very low environmental-lead
levels and blood-lead concentrations have undue influence on inferences made from the models.
For example, the effect of increasing dust-lead loading from 1 to 10 jig/ft2 is equal to the effect of
increasing dust-lead loading from 10 to 100 ug/ft2.  Therefore, inferences at such low levels can
be overestimated and misleading.  Thus, any inferences at very low dust-lead loadings, such as 1
or 5 ug/ft2, should be made with caution.

                                          195

-------
4.1.4  Conclusions

       The following conclusions can be made on the HUD model and the comparison of risk
estimates originating from this model versus those originating from models used in the §403 risk
analysis:

       •     As the HUD model parameters associated with environmental-lead measurements
             in specific media have been adjusted for measurement error, the input parameters
             to this model are assumed "true" lead levels in these media. This can provide
             biased results when the model is used to predict blood-lead concentration
             associated with lead levels measured in current risk assessments. The Rochester
             multimedia model and the empirical model did not have such an adjustment
             incorporated.

       •     While the HUD model contains study effects, they are considered fixed effects
             and therefore allow the model to make predictions for only the group of children
             represented by the 12 studies.  Furthermore, the study effects impact only the
             intercept of the model, and any study-to-study differences that may be present in
             other model terms (such as in environmental-lead measurements) are not
             represented.

       •     The HUD model handles "exterior-lead measurements" (e.g., soil) differently than
             the §403 models; the impact of such difference has not been determined.
4-2    ALTERNATIVE MULTIMEDIA MODELS FOR PREDICTING

       BASED ON ENVIRONMENTAL-LEAD LEVELS

       As discussed in the introduction to this chapter, the Rochester multimedia model,
presented in Section 4.2.3 of the §403 risk analysis report, was developed using data from the
Rochester Lead-in-Dust study to explain children's blood-lead concentration as a function of
dust-lead loadings from floors (carpeted and uncarpeted) and window sills, dripline soil-lead
concentration, and an indicator variable on the presence of deteriorated lead-based paint and a
child with paint pica tendencies. This model was used in the risk characterization (Section 5.3 of
the §403 risk analysis report) to determine the probability that a child exposed to specific levels
of lead in paint, dust, and soil would have a blood-lead concentration at or above 10 ug/dL.  EPA
used these estimates of individual risk, as well as the findings of performance characteristics
analyses detailed in Section 6.1  of this report, in proposing levels of concern for lead in dust
(page 30318 in the §403 proposed rule).

       The §403 proposed rule  considered uncarpeted floors and a yard-wide average soil-lead
concentration when proposing dust and soil standards and levels of concern. The performance
characteristics analysis cited in the proposed rule considered these types of dust-lead and soil-

                                          196

-------
lead measures. However, in the Rochester multimedia model, the floor dust-lead loading
measure did not limit the type of floor surface to uncarpeted floors, the soil-lead measure
represented only dripline soil, and the paint/pica indicator variable was different from the paint
measure used in the performance characteristics analysis (the percentage of tested components in
the home with deteriorated lead-based paint). For these reasons, it was difficult to compare
estimates of individual risks based on this model to results obtained from the performance
characteristics analyses.  Thus, it was desired to fit an alternative multimedia model (cited as
"Model A" in this section) that replaced the floor dust-lead and soil-lead predictor variables used
in the Rochester multimedia model with uncarpeted floor dust-lead loading and yard-wide
average soil-lead concentration, respectively, and replaced the paint/pica indicator variable with a
measure of the percentage of tested components containing lead-based paint.

       While a household risk assessment for lead-based paint hazards is expected, at a
minimum, to characterize lead levels in floor dust and to identify the extent of deteriorated lead-
based paint, it is possible that some risk assessments may not measure lead levels in soil or
window sill dust. Therefore, to investigate how individual risks would be characterized in these
types of risk assessments, two alternative multimedia models were fitted that were reduced
versions of Model A.  One model excluded soil-lead concentration as a predictor variable
("Model B"), and the other model excluded both soil-lead concentration and window sill dust-
lead loading as predictor variables ("Model C").

       The three alternative multimedia models were fitted using data from the Rochester Lead-
in-Dust study, using the  same approach used to fit the Rochester multimedia model in the  §403
risk analysis. The models were log-linear in nature, where the dust-lead and soil-lead measures
were log-transformed, and the models predicted a log-transformed blood-lead concentration. For
example, Model A took  the following form:

            log(PbB) = po + p,*log(PbF) + p2*log(PbW) + p3*log(PbS) + p4*PbP

where PbB represents blood-lead concentration (ug/dL), PbF represents household average dust-
lead loading for uncarpeted floors (fig/ft2), PbW represents household average dust-lead loading
for window sills (ug/ft2), PbS represents yard-wide average soil-lead concentration (ug/g), and
PbP represents the larger (between the interior and exterior of the housing unit) of the
percentages of tested components containing deteriorated lead-based paint. As with the
Rochester multimedia model, ordinary least squares regression methods were used to fit the
models to the Rochester data.

       Table 4-1 presents the estimates of the model parameters for each of the three alternative
multimedia models. Note that the model fits were based on different numbers of housing units,
as eliminating certain predictor variables from the above model resulted in more housing units
that had all necessary data available for fitting the model. An investigation of model diagnostics
showed that the extent of collinearity among the predictor variables in these models was low.
Generally, the slope estimates associated with the paint variable were very low, and except for
                                           197
                                                                U.S. EPA Headquarters Library
                                                                       Mail code 3201
                                                                1200 Pennsylvania Avenue NW
                                                                   Washington DC 20460

-------
Table 4-1.     Parameter Estimates and Associated Standard Errors for the Three
                 Alternative Multimedia Models Fitted to Rochester Study Data to Predict
                 Log-Transformed  Blood-Lead Concentration1
Parameter
Po
p,
>
P3
p*
R2
0&«r
n
Predictor Variable
Intercept
log (PbF): Area-Weighted Arithmetic Mean
(Wipe) Dust-Lead Loading from Uncarpeted
Floors
log (PbW): Area-Weighted Arithmetic Mean
(Wipe) Dust-Lead Loading from Window
Sills
log (PbS): Yardwide Average Soil-Lead
Concentration2 (fine soil fraction)
PbP: The larger of the following two
percentages: % of interior tested surfaces
that contain deteriorated LBP, and % of
exterior tested surfaces that contain
deteriorated LBP3
Coefficient of Determination
Error
# data points included in model fitting
Parameter Estimate (Standard Error)
Model A
0.331
(0.263)
0.114
(0.049)
0.082
(0.037)
0.115
(0.040)
0.001
(0.002)
20.03%
0.561
177
Model B
0.899
(0.183)
0.130
(0.048)
0.101
(0.035)
—
0.002
(0.002)
15.62%
0.572
188
Model C
1.337
(0.122)
0.140
(0.043)
—
—
0.004
(0.002)
11.38%
0.580
196
"-* indicates that the variable is not included in the model as a predictor. The models are log-linear in nature.

' One housing unit (cid = 01689) had an uncarpeted floor dust-lead loading measurement of 18130.0/yg/ft2 (only one
uncarpeted floor wipe sample was collected in this unit).  This data value was omitted when fitting the above models as it
was highly influential and led to a noticeable reduction in the estimate of p,.
2  Yardwide soil-lead concentration at a given housing unit was calculated as the unweighted arithmetic average of dripline
and play area soil-lead concentrations. If one or the other is missing (but both are not missing), yardwide concentration was
set equal to the non-missing value. If both are missing, yardwide concentration is missing.
3  If one or the other of these two percentages is missing (but both are not missing), the value of this variable is set equal to
the non-missing value.  If both are missing, the value is missing.
                                                    198


-------
Model C, were not significantly different from zero at the 0.05 level.  See footnotes to this table
for additional details on the model fits.

       Other alternative multimedia models were considered when initially developing the
Rochester multimedia model. These models, and information used to evaluate these models in
selecting the final version used in the §403 risk analysis, were presented in Appendix G of the
§403 risk analysis report.
4.3    SUPPLEMENTAL INFORMATION ON MODEL-BASED
       APPROACHES IN THE S403 RISK ANALYSIS

       Some comments on the §403 proposed rule addressed issues concerning approaches taken
in the §403 risk analysis in which statistical models were used to predict a post-intervention
distribution of blood-lead concentration, and therefore, a means of assessing how health risks
associated with lead-based paint hazards would change as a result of implementing the §403 rule.
The types of models used in this analysis and the approach to characterize post-intervention
health risks were presented in Chapters 4 and 6 of the §403 risk analysis report. In addition,
technical details on these models and approaches were provided in appendices to the report.
However, certain comments on the §403 proposed rule indicated that this information may not
have been provided in sufficient detail or would have benefitted from additional clarity. In
particular, two issues in question involved how the §403 risk analysis estimated a post-
intervention blood-lead concentration distribution that was comparable to the baseline
distribution that was characterized by data from Phase 2 of NHANES m, and how the empirical
model (Section 4.2 of the §403 risk analysis report) account for measurement error issues
associated with the predictor variables. The following subsections provide additional
information on these two issues, specifically geared toward addressing the specific areas raised
by selected public comments.

4.3.1   The "Scaling"  Algorithm Used to Determine a Post-
       Intervention Blood-Lead Concentration Distribution

       In the §403 risk analysis, EPA used data from Phase 2 of NHANES m (collected from
1991-1994) as the basis for the baseline ("pre-§403 rule") characterization of children's blood-
lead concentration in the U.S. housing stock. As discussed in Section 3.5 of the §403 risk
analysis report, EPA took this approach because these data were considered the best available
data (as well as the most recent data) on blood-lead measures, the data consisted of actual blood-
lead measurements from a nationally-representative survey, and it was preferred (and considered
more defensible) to use such data to characterize the baseline distribution rather than data
generated from statistical prediction models. However, because the "post-§403 rule" time period
has not yet occurred, and it was desired to compare the blood-lead distribution in this time period
to the baseline blood-lead distribution, it was necessary to use statistical prediction models to
generate how the baseline distribution would change following interventions performed as a
result of the §403 rule.

                                         199

-------
       When estimating the blood-lead concentration and health effect endpoints used in the
 §403 risk analysis, it was assumed that the national distribution of blood-lead concentration was
 lognormally distributed. Initial investigations into the weighted NHANES ffi blood-lead
 concentration data used in the §403 risk analysis (Figure 5-3 of the §403 risk analysis report)
 suggested that this was a satisfactory assumption.  The lognormal distribution is characterized by
 the geometric mean and geometric standard deviation (GSD) of the data (or, equivalently, the
 exponentiated mean and exponentiated standard deviation of the log-transformed data).  Once the
 geometric mean and GSD were calculated from the weighted NHANES ffl data, the additional
 assumption of lognormality was used to obtain baseline estimates of the health effect and blood-
 lead concentration endpoints considered in the §403 risk analysis (e.g., probability that a child's
 blood-lead level was at or above 10 ug/dL).  These estimates were presented in Table 5-1 of the
 §403 risk analysis report.

       To estimate how the blood-lead distribution changed between the "pre-§403 rule" and
 "post-§403 rule" environments (based on a given set of candidate §403 standards and on
 assumed changes in environmental-lead levels resulting from implementing the §403 rule),
 model-based estimates of the blood-lead distribution were made for both environments.  For
 reasons explained in the §403 risk analysis report, EPA chose to characterize both the pre-§403
 and post-§403 blood-lead distributions twice, with the first characterization using the IEUBK
 model and the second characterization using the empirical model.  Each characterization
 involved fitting the given model to environmental-lead data separately for each home in the HUD
 National Survey and weighting each prediction appropriately to represent a given proportion of
 the nation's children.  Therefore, although the HUD National Survey homes do not represent a
 random sample of homes in the  national housing stock, and therefore, the set of predicted blood-
 lead concentrations do not themselves represent a random sample of blood-lead levels in the
 nation's children, the fact that each prediction is weighted appropriately to represent a given
 proportion of the nation's children allows the total set of predictions to be a good estimate of the
 national blood-lead distribution, in either a pre-§403 or post-§403 environment.

       Appendix E2 of the §403 risk analysis report discusses how model-predicted blood-lead
 concentrations generated for the HUD National Survey homes are used to estimate the geometric
 mean and GSD associated with the national blood-lead distribution. Recall that for a given HUD
 National Survey home, the model-predicted blood-lead level represents a geometric mean of
 children whose exposure is characterized by the environmental-lead levels in that home.
 Therefore, the estimated GSD of the national blood-lead distribution is characterized not only by
 the variability among the predicted blood-lead levels, but also by the assumed variability in
 blood-lead levels among individual children exposed to the same environmental-lead levels.

^.uux.. Under a given model (i.e., either the IEUBK or empirical model), the "scaling" algorithm
 (Appendix Fl of the risk analysis report) involves calculating the proportional change in the
 geometric mean and GSD of the model-predicted blood-lead distribution from the "pre-§403
 rule" to the "post-§403 rule" environment. Then, the same proportional change in both statistics
 was applied to the geometric mean and GSD of the baseline distribution determined from the
 NHANES ffl data:

                                           200


-------
The resulting geometric mean and GSD, along with an assumption that lognormality still holds,
characterized a blood-lead concentration that represented the post-§403 environment and that
was considered comparable to the baseline distribution.

       This type of algorithm was necessary due to the difficulties associated with comparing a
model-based post-§403 blood-lead distribution directly with a baseline distribution that was
characterized from observed (NHANES HI) data.  While the empirical model was calibrated so
that its estimate for the baseline national geometric mean blood-lead concentration for children
aged 1-2 years (as obtained within the approach taken in the §403 risk analysis) equaled the
NHANES DI Phase 2 estimate (although the predicted GSD was not similarly calibrated), such a
calibration was not possible for the BEUBK model. As a result, using the HUD National Survey
data as input, these models could not both predict the same national estimate of the geometric
mean blood-lead concentration as Phase 2 of NHANES HI. In addition, the empirical model
could not be developed based on data from a national survey that measured both blood-lead and
environmental-lead levels, which would have facilitated direct comparisons of the predicted
blood-lead distribution between pre-§403 and post-§403 environments. Therefore, the "post-
1403 rale" blood-lead distributions predicted by the two models had some inconsistency with the
baseline distribution estimated from the NHANES HI data, making direct comparisons
problematic.  For example, if a model underestimates the geometric mean, the benefits associated
with the §403 rale could be overestimated, while if a model overestimates the geometric mean,
this could result in estimates of negative benefits. Therefore, the "scaling" algorithm used in the
risk analysis used the models to predict the change in the geometric mean and GSD that occurs
from a pre- to post-§403 rule environment, then applied this same change to the baseline
distribution.

       Note that the scaling algorithm does not require that the two model-based blood-lead
distributions (pre-§403 and post-§403) be independent of each other. In  fact, the two
distributions are dependent, because the post-§403 environmental-lead levels, used to predict
post-§403 blood-lead levels, are dependent on the pre-§403 levels. Only the geometric mean and
GSD of these two distributions are necessary to characterize, and they are used simply to
estimate the proportional change in the geometric mean and GSD of the national blood-lead
distribution between pre-§403 and post-§403 conditions.

       While the geometric mean and GSD are scaled separately, one change is not necessarily
independent of the other. For example, if the pre-§403 geometric mean and GSD both have high
values, they are both likely to be reduced at a greater rate than at lower values. The approach
was kept as simple as possible while retaining scientific defensibility, in  order that it be easily
applied during risk characterization and in the economic analysis.

       In the peer review of the §403 risk analysis report, EPA specifically asked the peer
reviewers to comment on whether the scaling procedure was scientifically defensible in general,
                                          201

-------
and in particular, whether it was relevant in the situation where the environmental-lead data
(from the HUD National Survey) and the blood-lead data (from NHANES ID) were collected at
different periods in time. None of the peer reviewers specifically criticized the scaling algorithm.
Furthermore, the Science Advisory Board reviewing the §403 risk analysis stated that, in general,
the approach was scientifically defensible as presented, and specifically, the multi-step approach
was warranted due to the need to use various datasets of differing sources and representing
different time periods to make the characterization.

       See Section 6.4.4 below for an alternative approach to applying this scaling algorithm
where the probability of a child's blood-lead concentration exceeding 10 ug/dL is scaled rather
than the GSD.

4.3.2   Adjusting the Empirical Model Parameter
       Estimates to Reflect Measurement Error

       The approach to the measurement error adjustment, discussed in Section 4.2.4 of the §403
risk analysis report, attempted to correct for the fact that while the empirical model was
developed using data from  the Rochester Lead-in-Dust study, it was used to predict a (pre-403)
geometric mean blood-lead concentration assuming that the data input to the model originated
from the HUD National Survey. Because the Rochester study and the HUD National Survey
used different sampling schemes involving different collection devices and instruments, as well
as different analytical methods, and because the ranges of observed environmental-lead levels
differed between the two studies, it was necessary to adjust the model parameter estimates to
reflect these  differences prior to allowing the model to accept HUD National Survey
environmental-lead data as input to the prediction.

       Note that the measurement error adjustment made to the empirical model was not to
address the more-standard "errors in variables" issue (Carroll et al., 1995) which attempts to take
into account that a value input to the model represents a measurement subject to error, rather than
a "true" value. As discussed in Section 4.2.4 and Section G4.2 (Appendix G)  of the §403 risk
analysis report, the empirical model was not intended to be used in the risk analysis as a dose-
response model, which would have required the predictor variables to reflect actual exposures.
Instead, the model assumed that its input environmental-lead information reflected measurements
that would have been made as a result of a risk assessment within a home. Therefore, adjusting
the model for the fact that its inputs reflect measured rather than actual lead levels was
considered inappropriate for this analysis. This decision in the type of application that is
represented by the §403 risk analysis has been concurred upon in the published literature (e.g.,
Carroll and Galindo, 1999).

       When using the empirical model to predict a post-403 geometric mean blood-lead
concentration, some of the  HUD National Survey dust-lead and soil-lead data (i.e., those data for
homes that exceed the candidate 403 standards) were modified to reflect the impact of
performing interventions in response to the 403 rule on these measured data values (see Table
6-2 of the  §403 risk analysis report), then the model is fitted to the  modified data. These

                                          202

-------
modified data are still considered to be measured lead levels, rather than actual (or "true") lead
levels. However, the modified data values must first be transformed to represent measurements
that would have made under the methods used in the HUD National Survey. For example, the
assumed post-intervention floor wipe dust-lead loading of 40 ug/ft2 must be converted to a Blue
Nozzle vacuum-equivalent loading prior to using it as input to the empirical model.
                                          203

-------
This page intentionally blank.
            204

-------
5.0    RISK CHARACTERIZATION

       Chapter 5 of the §403 risk analysis report documented the final portion of the risk
assessment phase of the §403 risk analysis, in which the methods introduced in the earlier
chapters of the §403 risk analysis report were applied to characterize risks associated with current
(i.e., baseline) lead exposures for children aged 1-2 years. The baseline distribution of blood-
lead concentration in this population was characterized using data from Phase 2 of the Third
National Health and Nutritional Examination Survey (NHANES ID), conducted from 1991 to
1994. Alternative pre-§403 risk estimates were also calculated as a function of environmental-
lead levels by using data from the HUD National Survey as input to the EUBK and empirical
models. Both individual risk estimates (i.e., risks associated with specific environmental-lead
levels) and population-based risk estimates (i.e., average risks over the entire nation) were
presented. As mentioned in Section 2.0, the specific blood-lead concentration and health effect
endpoints used to measure the risks of lead exposure to children aged 1-2 years were
       •      Incidence of blood-lead concentration greater than or equal to 10 ug/dL
       •      Incidence of blood-lead concentration greater than or equal to 20 ug/dL
       •      Incidence of IQ score less than 70 in the population of U.S. children, which results
              from lead exposure
       •      Incidence of IQ score decrement (in the population of U.S. children) greater than
              or equal to 1 resulting from lead exposure
       •      Incidence of IQ score decrement (in the population of U.S. children) greater than
              or equal to 2 resulting from lead exposure
       •      Incidence of IQ score decrement (in the population of U.S. children) greater than
              or equal to 3 resulting from lead exposure
       •      Average IQ decrement within the population of U.S. children that results from
              lead exposure.

       The risk characterization included a sensitivity and uncertainty analysis where possible
alternatives to various approaches taken and assumptions made in the risk characterization were
identified and incorporated into the analysis, and the resulting impact on the risk estimates was
evaluated. This analysis resulted in a measure of the uncertainty associated with the risk
estimates due to methodological assumptions, thereby producing a range of estimates within
which the true risk may reasonably be expected to fall. Section S.I of this chapter contains the
following additional sensitivity and uncertainty analyses that were performed and documented
since the §403 risk analysis report was published:

       •      Calculate individual risks associated with specified lead levels in floor-dust and
              soil, as predicted by the HUD model introduced in Section 4.1 and the alternative
              multimedia models presented in Section 4.2.

       •      Calculate estimates assuming a 50% decline in the estimated geometric mean
              blood-lead concentration of children aged 1-2 years from the estimate generated
                                           205

-------
             from Phase 2 of NHANES m (in addition to the estimates associated with 10%,
             20%, and 30% declines that were presented in Section 5.4.3 of the §403 risk
             analysis report).

       *     Calculate model-based estimates of the pre-§403 blood-lead distribution under
             revised environmental-lead levels (from the HUD National Survey) input to the
             models, with the revisions representing the potential change in these levels that
             may have occurred since the survey was performed.

       •     Calculate baseline estimates of the IQ-related health effect endpoints assuming
             that specified non-zero thresholds exist in the relationship between blood-lead
             concentration and IQ.

5.1    RISK CHARACTERIZATION SENSITIVITY AND UNCERTAINTY ANALYSIS

       The following subsections present the results of additional sensitivity and uncertainty
analyses performed to  gauge the level of uncertainty in baseline risk estimates associated with
methodological assumptions. These results should be considered with those presented in the
sensitivity and uncertainty analyses in Section 5.4 of the §403 risk analysis report to characterize
overall uncertainty associated with the methods and assumptions taken in the risk assessment.

5.1.1  Estimates of Individual Risks from Applying the HUD Model

       In Section 5.3 of the §403 risk analysis report, the concept of individual risks was
introduced, and estimates of individual risks associated with lead exposure in children were
generated by fitting the IEUBK and Rochester multimedia models to specified environmental-
lead levels. Briefly, within the context of the §403 risk analysis, individual risks refer to the risks
associated with a young child's exposure to specified levels of environmental-lead. Once
environmental-lead levels were specified for each medium, the model-predicted blood-lead
concentration at these levels, along with the assumption that blood-lead concentrations have a
lognormal distribution with a specified variability, were used to estimate the percentage of
children exposed to the specified set of environmental-lead levels that would have elevated
blood-lead concentrations (i.e., at or above 10 |ag/dL).  Then, those sets of environmental-lead
levels associated with estimated elevated blood-lead percentages of 1%, 5%, and 10% were
identified and presented in Tables 5-5 through 5-7 and Figures 5-7 and 5-8 of Section 5.3 of the
§403 risk analysis report. The IEUBK model was used to identify soil-lead concentrations
associated with these elevated blood-lead percentages (at specified dust-lead loadings), while the
Rochester multimedia  model was used to identify (wipe) dust-lead loadings associated with these
elevated blood-lead percentages (at specified soil-lead concentrations). These results contributed
to the information which EPA used in proposing dust and soil levels  of concern in the §403
proposed rule. See Section 5.3 of the §403 risk analysis report for additional details.

       As discussed in Section 4.1 of this report, the HUD model was published shortly after the
§403 .risk analysis report was finalized, and some commenters to the  §403 proposed rule

                                          206

-------
suggested that it be considered within the §403 risk analysis. The HUD model can be used in the
same manner as the §403 risk analysis models to predict individual risks associated with
exposure to a specified set of environmental-lead levels. Therefore, this section presents
estimates of individual risks based on fitting the HUD model to specified environmental-lead
levels and compares these risk estimates with those based on the EEUBK model  (for soil) or the
Rochester multimedia model (for dust). Supporting summaries and discussion for comparing
HUD model results with those of the multimedia models developed for the §403 risk analysis are
presented in Appendix F.

Soil-Lead Concentrations

       When fitting the HUD model to evaluate individual risks associated with yard-wide
average soil-lead concentration, the model's soil-lead parameters were set to the following
values:

       •     ExtType = 0 (indicating that soil was sampled rather than exterior dust)
       •     ExtLoc = 0.5 (indicating that the total soil sampled was a rough average of drip-
             line and non-drip-line soil)

       Figure 5-1 plots estimates of the percentage of children's blood-lead concentrations at or
above 10 ug/dL as a function of soil-lead concentration, as predicted by the IEUBK model and
the HUD model. The left-most panel in Figure 5-1 corresponds to the results of IEUBK model
fits at specified dust-lead concentrations (100,200, and 500 ug/g) and is identical to Figure 5-7
in the §403 risk analysis report. The middle and right-most panels of Figure 5-1 correspond to
fits of the HUD model at specified dust-lead loadings (5,10,20,25,40,50,100 and 200 ug/ft2)
and differ according to the assumed geometric standard deviation (GSD) associated with the
blood-lead concentration distribution (GSD=1.6 for the middle panel; GSD=1.72 for the right-
most panel).  A GSD of 1.72 was estimated within the HUD model publication (Lanphear et al.,
1998).

       The IEUBK and HUD model fits portrayed in Figure 5-1 are not directly comparable as
the IEUBK model controls for dust-lead concentration while the HUD model controls for dust-
lead loading. However, the plots do suggest that the predicted patterns of change in blood-lead
concentration with soil-lead concentration differ considerably for the two models.

       Table 5-la, identical to Table 5-5 of the §403 risk analysis report, presents the soil-lead
concentrations for which the IEUBK model-predicted percentages of children having blood-lead
concentrations of at least 10 ug/dL equal 1%, 5%, and 10% assuming dust-lead concentrations of
100,200, and 500 ug/g. Table 5-lb  presents the soil-lead concentrations for which the HUD
model-predicted percentages of children having blood-lead concentrations of at least 10 ug/dL
equal 1%, 5%, and 10% assuming the dust-lead loadings considered in the HUD model panels of
Figure 5-1. Table 5-lb (HUD model) indicates that, for all dust-lead loadings considered, the
soil-lead concentrations estimated to maintain risks at 1% and 5% are less than 400 ug/g; even a
soil-lead concentration of less than 6 ug/g would not achieve these levels of protection if children

                                          207

-------
                                                   88

ggssssssss

       -IP/BTT oicaqd »« 't4iii«il»«i=>-'d
                                          r§
                                                                                       0*
                                                                                       "S  O
                                                   I
                                                                                   in

                                                                                    2
                                                                                    3
                           208

-------
Table 5-1 a.  Yard-Wide Average Soil-Lead Concentrations at Which the Percentage of
            Children Aged 1-2 Years With Blood-Lead Concentration At or Above 10
            A/g/dL is Estimated by the 1EUBK Model at 1, 5, or 10%, for Three Assumed
            Dust-Lead Concentrations (Table 5-5 in §403 risk analysis report).
Blgg^^
HHHiHiiiHI
100
200
500
155
35
Not achievable
'.;•:, . -
365
245
Not achievable
l&^MgSSSSiSSSSIiSS^BBS^SSH
515
395
25
Table 5-1 b.   Yard-Wide Average Soil-Lead Concentrations at Which the Percentage of
             Children Aged 1-2 Years With Blood-Lead Concentration At or Above 10
             //g/dL is Estimated by the HUD Model at 1, 5, or 10%, for Eight Assumed
             Dust-Lead Loadings and Two Assumed Geometric Standard Deviations

i^iwai^gtt^H

5
10
20
25
40
50
100
200
.'.. .. : • '. .."•
5
10
20
25
40
50
100
200
BBHflBBBiHHBBHBilBffff'''*

GSD
74.4
49.4
32.9
28.8
21.8
19.1
12.7
8.5
;'-;•: -:;h;;-: •'- GSD
45.9
30.5
20.3
17.8
13.5
11.8
7.9
5.2



186.2
123.7
82.3
72.1
54.7
47.9
31.9
21.2
_ t;72 1 ' ",::. • •
132.4
88.0
58.5
51.3
38.9
34.1
22.7
15.1
opi^usni
»M»W!PP«BHSralHB«HHKiSKPlBWS»


303.7
201.8
134.2
117,6
89.2
78.2
52.0
34.5

232.8
154.8
102.9
90.2
68.4
60.0
39.9
26.5
                                       209

-------
were exposed to dust-lead loadings of 40 ug/ft2 or more.  Based on Table 5-lb, if dust-lead
loadings are equal to 5 jig/ft2, a soil-lead concentration of approximately 300 ug/g (100 ug/g)
maintains the percentage of blood-lead concentrations greater than or equal to 10 ug/dL at 5% for
a GSD of 1.60 (1.72).

Floor Dust-Lead Loadings

       Figures 5-2 and 5-3 plot the estimated percentages of children having blood-lead
concentrations at or above 10 ug/dL as a function of floor dust-lead loadings as predicted by the
HUD model and the Rochester multimedia model assuming GSDs of 1.60 and 1.72, respectively,
on the blood-lead distribution. Soil-lead concentrations are assumed to be fixed at 100,400,
1200,2000 and 5000 ug/g and, for the Rochester multimedia model, window sill dust-lead
loadings are assumed to be fixed at 200 and 500 ug/ft2. In each figure, the left-most panel
contains estimates based on fitting the HUD model. The Rochester multimedia model panels for
GSD equal to 1.60 and soil-lead concentrations of 100 and 400 ug/g were presented in Figure 5-8
in the §403 risk analysis report.

       Tables 5-2 and 5-3 present the floor dust-lead loadings that are predicted by the HUD
model and Rochester multimedia model, respectively, to mamtaim the percentage of children
having blood-lead concentrations above or equal to 10 ug/dL at 1%, 5%, and 10% for specified
levels of soil-lead concentration, window sill dust-lead loading and GSD. Approximate 95%
upper confidence bounds, which account for the variability of parameter estimates from the
Rochester multimedia model, are also provided in Table 5-3.  See Appendix C2, Section 5.0 of
the §403 risk analysis report for the methodology used to compute these confidence bounds. The
rows of Table 5-3 corresponding to a GSD of 1.60 and soil-lead concentrations of 100 and 400
ug/g are identical to Table 5-6 in the §403 risk analysis report (after correcting an error in the
computation of the confidence bounds).

       Some key findings noted when comparing the individual risk estimates presented in
Appendix F between the HUD model and Rochester multimedia model include the following:

       • "      At very low floor dust-lead loadings (i.e., 1-5 ug/ft2), the HUD model and the
              Rochester multimedia model yield similar predictions for the geometric mean
              blood-lead concentration, which also results in similar predictions for the health-
              effect endpoints that are calculated directly from this geometric mean (e.g.,
              percentage of children with blood-lead concentration at or above a specified
              threshold; average IQ decrement resulting from lead exposure). However, due to
              the forms of these models and concerns involving the accuracy of very low dust-
              lead measurements, any conclusions made at such low dust-lead loadings must be
              made with caution.

       •      The predicted geometric mean blood-lead concentration under the HUD model
              ranges from 20% to nearly 60% higher than the prediction under the Rochester
              multimedia model as floor dust-lead loadings increase from 15 to 100 Mg/ft2 and

                                         210


-------
                          r§
                           §    •!
                              f Mi
S  S  S  S S
                    S 2
! I 1 i
1 ! i
i i
III!
ill
l. ' i
I ! i
11
I I
i i '' :
1 ' !
1 1 ! i
\ \\ \
\ I 1
'• \\ i
\
\ I \ '•
\ \ • \
* \ \ !
\ \ \ i
\ \ > '•
\ \N
» x \
^•>^-.



















• s


o o
i n
' 1 «^
1 1 ^
K- ! 1 <.
.§S§ii il
KS W* » |4
i i i 3jr

^ 1 ?
.g 1 | S|
" 1 1 ]}
g S
* § ^ ^

-7
^

   sssssass"
s  s  s  s  s  s  R
    -IP/Or* Ol*8
-------
         s


         § §1


         §


         s
s s
     ass
         § §1
           • i
          fl"
         a
          1
         -I §1
         •1
               i
                     lili
1111
 co o> o
 = c a
  l
                      «
 il
O O 3 =
Q.S 11.5
                     in
      212

-------
Table 5-2.   Floor Dust-Lead Loadings at Which the Percentage of Children Aged 1-2
            Years With Blood-Lead Concentration At or Above 10//g/dL is Estimated by
            the HUD Model at 1, 5, or 10%, for Five Assumed Soil-Lead Concentrations
            and Two Assumed Geometric Standard Deviations
t^
100
400
1200
2000
5000
. - . . - ..T^j, ::.:;;;
100
400
1200
2000
5000

-------
Table 5-3.   Floor Dust-Lead Loadings at Which the Percentage of Children Aged 1-2
             Years With Blood-Lead Concentration At or Above 10 figldL is Estimated by
             the Rochester Multimedia Model at 1, 5, or 10%, for Five Assumed Soil-
             Lead Concentrations, Two Assumed Window Sill Dust-Lead Loadings, and
             Two Assumed Geometric Standard Deviations (expanded version of Table 5-
             6 in §403 risk analysis report).
« - „ uSP = 1 .BU
100
400
1200
2000
5000

100
400
1200
2000
5000
200
500
200
500
200
500
200
500
200
500

200
500
200
500
200
500
200
500
200
500
0.05
0.02
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
i':' '.",- '...
0.00
o.oo
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.37
0.14
0.04
0.02
0.01
0.00
0.00
0.00
0.00
0.00
GSQ!t=1.7J
0.04
0.01
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.00
6.70
2.00
0.61
0.18
0.09
0.03
0.04
0.01
0.01
0.00
8;J"-f:- '• : '
1.11
0.33
0.10
0.03
0.02
0.00
0.01
0.00
0.00
0.00
22.00
8.93
2.81
1.12
0.59
0.22
0.28
0.10
0.07
0.02

4.90
1.95
0.64
0.24
0.12
0.04
0.06
0.02
0.01
0.00
89.08
26.62
8.13
2.43
1.22
0.36
0.50
0.15
0.10
0.03
327.73
92.01
20.32
9.24
4.86
1.99
2.47
0.96
0.69
0.25

21.87
6.54
2.00
0.60
0.30
0.09
0.12
0.04
0.03
0.01
67.96
24.54
7.04
3.00
1.59
0.61
0.77
0.29
0.20
0.07
1  The 95% upper confidence bounds here differ from those in Table 5-6 of the 5403 risk analysis report. The values here
are corrected for a mistake in the original computations.
                                         214

-------
             as soil-lead concentrations decrease from 2000 ppm to 10 ppm (assuming, for the
             Rochester multimedia model, that window sill dust-lead loadings are at their
             estimated national median level; Tables F-l and F-2 of Appendix F). Note that
             for a fixed value of the geometric standard deviation (GSD) for the blood-lead
             distribution, the average IQ decrement in the population that is associated with
             lead exposure is a multiple of the geometric mean (as calculated hi the §403 risk
             analysis). Therefore, similar differences in predictions between the two models
             would occur for average IP decrement.

             If the geometric standard deviation (GSD) associated with the blood-lead
             distribution is fixed, then as floor dust-lead loadings increase beyond 10 ug/ft2,
             the predicted percentage of children with blood-lead levels at or above 10 ug/dL
             increases at a much faster rate under the HUD model  (at a constant soil-lead
             level). For example, if window sill dust-lead loading is at its estimated national
             median and soil-lead concentration is below 2000 ppm, the predicted percentage
             under the HUD model is at a minimum twice as large as the prediction under the
             Rochester multimedia model. This difference in predictions gets even greater as
             the assumed soil-lead concentration gets lower. For example, at a GSD of 1.6, a
             floor dust-lead loading of 100 ug/ft2, and a soil-lead concentration of 10 ppm, the
             prediction is over 7 times higher for the HUD model compared to the Rochester
             multimedia model (13.1% versus 1.76%; Tables F-3 and F-4 of Appendix F).
5.1.2  Estimates of Individual Risks from Applying the
       Alternative Rochester Multimedia Model

       The approach to estimating individual risks that was discussed and applied (using the
IEUBK model and Rochester multimedia model) in Section 5.3 of the §403 risk analysis report
and was applied using the HUD model in Section 5.1.1 above was also applied using the
alternative Rochester multimedia model ("Model A") that was documented in Section 4.2 of this
report. Recall that this model uses 1) average dust-lead loadings on uncarpeted floors, 2) average
dust-lead loadings on window sills, 3) yardwide average dust-lead concentration, and 4) the
percentage of painted components containing deteriorated lead-based paint as predictor variables.
The blood-lead concentration predicted by this model is an estimate of the geometric mean of the
distribution of blood-lead concentration, which is assumed to be lognormal with geometric
standard deviation 1.6. The resulting distribution is then used to estimate the percentage of
children with blood-lead concentration at or above 10 fig/dL. This subsection documents the
results of estimating individual risks based on the alternative Rochester multimedia model.

       In this analysis, it is assumed that the given residential environment for which individual
risks will be estimated has no deteriorated lead-based paint  (as was done in Section 5.3 of the
§403 risk analysis report). Then, levels of two of the remaining three predictor variables are
fixed, and the level of the third variable is determined so that either 1%, 5%, or 10% of children
                                         215

-------
exposed to the combined lead levels given by the three predictor variables would be predicted to
have a blood-lead concentration at or above 10 ug/dL.

       Table 5-4a contains the estimated floor dust-lead loading at which the predicted
percentage of children with blood-lead concentration at or above 10 ug/dL is either 1%, 5%, or
10%, given that window sill dust-lead loading is at either 200 or 500 ug/ft2, and soil-lead
concentration is at either 100 or 400 ug/g. Table 5-4b contains the estimated window sill dust-
lead loading at which the predicted percentage of children with blood-lead concentration at or
above 10 ug/dL is either 1%, 5%, or 10%, given that floor dust-lead loading is at either 25 or 100
ug/ft2, and soil-lead concentration is at either 100 or 400 ug/g.  These two tables correspond to
Tables 5-6 and 5-7, respectively, in Section 5.3 of the §403 risk analysis report, where the
estimates were based on the original Rochester multimedia model. Finally, Table 5-4c contains
the estimated yard-wide average soil-lead concentration at which the predicted percentage of
children with blood-lead concentration at or above 10 ug/dL is either 1%, 5%, or 10%, given that
floor dust-lead loading is at either 25 or 100 ug/ft2, and window sill dust-lead loading is at either
200 or 500 ug/ft2. (A corresponding table for the soil-lead concentration does not exist in
Section 5.3 of the §403 risk analysis report as the soil-lead concentration predictor variable in the
original Rochester multimedia model assumed only dripline soil rather than a yard-wide average,
and the IEUBK model did not accept dust-lead loadings as input.)

       Effect on risk analysis:  If the target percentage of children with elevated blood-lead
concentration is 5%, the estimates in Table 5-4a (fourth column) are only slightly higher than the
corresponding estimates in Table 5-6 of the §403 risk analysis report, suggesting that at the fixed
levels of soil-lead concentration and window sill dust-lead loading, the corresponding floor dust-
lead loading is nearly the same for the two methods determined by the two forms of the
multimedia model. However, at a soil-lead concentration of 100 ug/g, the estimated floor dust-
lead loadings are considerably smaller (and below 50 ug/ft2)  than in Table.5-6 of the §403 risk
analysis report under 10% risk and considerably larger (but still below 1 ug/ft2) under a 1% risk.

       For window sills (Table 5-4b), estimated dust-lead loadings are reduced under the
alternative Rochester multimedia model compared to the estimates hi Table 5-7 of the §403 risk
analysis report at the 5% and 10% risk levels.  At the 5% risk level, the estimated window sill
dust-lead loadings were at 40 ug/ft2 or lower at the specified soil-lead and uncarpeted floor dust-
lead levels (Table 5-4b, fourth column).

       Yardwide average soil-lead concentration needed to be below 150 ug/g at each of the
specified levels of floor and window sill  dust-lead loadings to achieve risks of 10% or lower, and
at 32 ug/g or lower to achieve risks of 5% or lower (Table 5-4c). The estimated soil-lead
concentration increases as the two specified dust-lead loadings decrease, illustrating how the soil-
lead standard could become less stringent as the dust-lead loading  standards became more
stringent.
                                          216

-------
Table 5-4a.   Uncaroeted Floor Dust-Lead Loadings at Which the Percentage of Children
              Aged 1-2 Years With a Blood-Lead Concentration At or Above 10/ig/dL is
              Estimated by the Alternative Rochester Multimedia Model (A) at 1 %, 5%, or
              10%, Under Fixed Levels of Yardwide Average Soil-Lead Concentration and
              Window Sill Dust-Lead Loading
Yardwide Average
Soil-Lead Cone.
. U'g/gJ
100
400
Window SHI Dust-
Lead Loading
teg/ft2)
200
500
200
500
Uncarpeted Floor Dust-Lead Loading (//g/ft2) at Which the
Estimated % of Children With Blood-Lead Concentration
At or Above 10 f/gfdL is ...
1%
0.48
0.25
0.12
0.06
5%
7.9
4.1
1.9
1.0
10%
35
18
8.7
4.5
Note: The percentages of 1%, 5%, and 10% were determined assuming that the blood-lead distribution is
lognormal with geometric mean as predicted by the alternative Rochester multimedia model (A) and a
geometric standard deviation of 1.6.
Table 5-4b.   Window SHI Dust-Lead Loadings at Which the Percentage of Children Aged
              1-2 Years With a Blood-Lead Concentration At or Above 10/ig/dL is
              Estimated by the Alternative Rochester Multimedia Model (Al at 1 %,  5%, or
              10%, Under Fixed Levels of Yardwide Average Soil-Lead Concentration and
              Uncarpeted Floor Dust-Lead Loading
Yardwide Average
, Soil-Lead Cone.
(P9/3)
100
400
Uncarpeted Floor
1 Dust-Lead
Loading (//g/ft2)
25
100
25
100
Window Sill Dust-Lead Loading (pg/ft2) at Which the
Estimated % of Children With Blood-Lead Concentration
At or Above 10/sg/dl is ...
1%
0.78
0.11
0.11
0.02
5%
40
5.7
5.6
0.80
10%
320
46
45
6.5
Note: The percentages of 1%, 5%, and 10% were determined assuming that the blood-tead distribution is
lognormal with geometric mean as predicted by the alternative Rochester multimedia model (A) and a
geometric standard deviation of 1.6.
                                          217

-------
Table 5-4c.   Yardwide Average Soil-Lead Concentration at Which the Percentage of
              Children Aged 1-2 Years With a Blood-Lead Concentration At or Above 10
              //g/dL is Estimated by the Alternative Rochester Multimedia Model (A) at
              1%, 5%, or 10%, Under Fixed Levels of Dust-Lead Loadings for Uncarpeted
              Floors and Window Sills
Uncarpeted Floor
- Dust-Lead.
Loading (pg/ft2)
25
100
Window Sill Dust-
Lead Loading
dig/ft2)
"•"'•* 1" j 1^ * . J
*i-* •& *• •<
• > *
200
500
200
500
Yardwide Average Soil-Lead Concentration (frg/g) at
. Which the Estimated % of Children With Blood-Lead
Concentration At or Above 10 //g/dL is ...
'' -":i% :s ';-»'.
2.0
1.0
0.50
0.26
L >B%"~
32
17
8.0
4.2
"' fcjf . *
"- *10% ? - '
140
73
35
18
Note: The percentages of 1%, 5%, and 10% were determined assuming that the blood-lead distribution is
lognormal with geometric mean as predicted by the alternative Rochester multimedia model (A) and a
geometric standard deviation of 1.6.
5.1.3  Considering Potential Declines in Biood-Lead Concentration
       from NHANES III Phase 2 Measures

       The results of this subsection are an extension of the analysis in Section 5.4.3 of the §403
risk analysis report.  In that subsection, the geometric mean blood-lead concentration of 3.14
ug/dL for children aged 1-2 years, estimated from data collected in Phase 2 of NHANES ffl
(1991-1994), was assumed to be either 10%, 20%, or 30% lower, and the resulting impact on the
baseline risk estimates was investigated. This analysis was performed due to the likelihood of
continued decline in blood-lead concentrations in the U.S. population that has occurred in recent
years. It was desired to augment this analysis by considering an additional assumption on the
percentage decline since Phase 2 of NHANES m 50%.

       Table 5-5 presents the baseline estimates of the blood-lead concentration and health effect
endpoints for children aged 1-2 years, where each blood-lead concentration measurement in
Phase 2 of NHANES ffl was reduced by the same amount:  10%, 20%, 30%, or 50%. Thus, the
analysis assumed a constant percentage decline for the entire blood-lead concentration
distribution as characterized by Phase 2 of NHANES ffl. This table is an extension of the results
presented in Table 5-11 of the §403 risk analysis report and includes the baseline estimates
reported in the risk analysis for comparison purposes (i.e., where no reduction is assumed).

       Note that within NHANES  ffi, the estimated geometric mean blood-lead concentration
for children aged 1-2 years declined from 4.05 ug/dL in Phase 1 to 3.14 (ag/dL in Phase 2,
representing a 22.5% decline.  This is within the range of declines being considered in the
                                          218

-------
Table 5-5.   Sensitivity Analysis for the Estimated Baseline Number and Percentage of
             Children Aged 1-2 Years Having Specific Health Effect and Blood-Lead
             Concentration Endpoints, Assuming Various Percentage Declines in Blood-
             Lead Concentration Since Phase 2 of NHANES III
Health Effect and Blood-Lead
Concentration Endpoints
PbB* 20^g/dL
PbBi 10//g/dL
IQ score less than 70
IQ score decrement x 1
IQ score decrement s 2
IQ score decrement * 3
Average IQ score decrement
Geometric Mean (pg/dL)
: : Numbers (%) of ChOdren Aged 1-2 Years
Risk Analysis
Estimate Oablft 5-1
of the §403 risk
:- analysts report)
46,800
(0.588%)
458,000
(5.75%)
9,130
(0.115%)
3,060,000
(38.5%)
863,000
(10.8%)
294,000
(3.70%)
1.06
3.14
-;- ~DlAWAAM+.hMA nAjJma in tt
: - , • Percentage Decline m at
; Since NHAN1
10%
30,900
(0.388%)
340,000
(4.27%)
8,610
(0.108%)
2,640,000
(33.2%)
669,000
(8.40%)
213,000
(2.68%)
0.951
2.82
20%
18,900
(0.238%)
239.000
(3.00%)
8,160
(0.102%)
2,190,000
(27.6%)
493,000
(6.19%)
146,000
(1.83%)
0.845
2.51
ood-Lead Concentration
ES III Phase 2
30%
10,600
(0.133%)
156,000
(1.96%)
7,760
(0.098%)
1,740,000
(21.8%)
340,000
(4.27%)
91,900
(1.15%)
0.740
2.20
:- :"50%:fe
2,130
(0.0268%)
46,800
(0.588%)
7,140
(0.0897%)
863,000
(10.8%)
117,000
(1.47%)
25,200
(0.317%)
0.528
1.57
sensitivity analysis within Table 5-5. However, due to the NHANES HI survey design and how
this survey was performed, caution must be taken when interpreting observed differences in
results between the two phases of this survey.

       Effect on risk analysis: According to Table 5-5, if it were assumed that a 50% across-
the-board decline in blood-lead concentration (resulting in a national geometric mean blood-lead
concentration of 1.57 ug/dL), this would reduce, the estimated number of children whose blood-
lead concentration was at or above 20 ug/dL from 46,800 to 2,130, a decline of 95%, while the
estimated number at or above 10 ug/dL would be reduced by nearly 90%, to 46,800 children.
The 50% decline resulted in percentage declines of 12%, 86%, and 91% for numbers of children
with IQ score decrements of 1,2, or 3, respectively, as a result of lead exposure. The estimated
average IQ decrement in the population due to lead exposure is cut in half under this assumption
(from 1.06 to 0.53 points), matching the assumed 50% decline in blood-lead concentration
because the IQ/blood-lead concentration relationship is assumed to be linear across the entire
range of blood-lead concentration.  The effects of the lower assumed percentage declines (10%-
30%) were discussed  in Section 5.4.3 of the §403 risk analysis report.
                                         219

-------
       The results in Table 5-5 are based on the assumption that the blood-lead concentrations
for each child in the population have been reduced by the same percentage since Phase 2 of
NHANES ffl. In reality, different subgroups have achieved different rates of change over this
time. However, considering different percentage declines for different subgroups would be very
difficult, and the resulting estimates of the health effect and blood-lead concentration endpoints
would likely differ only slightly from that observed in Table 5-5.

5.1.4 Considering How Baseline Environmental-Lead Levels May Have
       Changed Since the HUD National Survey

       Although interim data from the NSLAH (Section 3.1) have recently been made available
to this risk analysis and have been summarized throughout this report, the fact that the public
could not have reviewed these summaries during the public comment period limits the extent to
which these data could be considered in the rulemaking. Therefore, for purposes of the
rulemaking, data from the HUD National Survey continue to be the only nationally-
representative data source on baseline environmental-lead levels in the nation's housing stock.
Nevertheless, it was desired to estimate how changes in these environmental-lead levels that have
occurred since the HUD National Survey was conducted would affect the baseline (i.e., pre-
1403) risk characterization. Therefore, a sensitivity analysis was performed where the HUD
National Survey data was adjusted to reflect possible change in the distribution over time. The
adjusted data would yield a surrogate distribution of baseline environmental-lead levels. Several
alternative adjustments would be considered, and risk estimates based on each set  of adjusted
data would be calculated.

       To help in determining appropriate adjustments to the HUD National Survey data, the
summaries presented in Section 3.2 of this document compared the distribution of dust-lead and
soil-lead data reported in the HUD National Survey with distributions from other studies
performed more recently, but typically in specific locations that may not necessarily be
nationally-representative. These summaries showed that the distributions were quite consistent
across studies, suggesting that the distributions based from the HUD National Survey data, even
after converting from Blue Nozzle dust-lead loadings to wipe-equivalent loadings, are likely
adequate for characterizing environmental-lead levels even up to ten years after the survey. In
fact, the HUD National Survey data distributions were often centered at lower lead levels than in
the other studies.  (A primary exception was household average dust-lead loadings, where data
from the interim NSLAH were considerably lower than in other studies,'including the HUD
National Survey. It is currently uncertain of the degree to which the observed distribution from
the interim NSLAH reflects actual declines in dust-lead levels since the HUD National Survey.)
Furthermore, the nature of the distribution appears to be affected by housing age, with higher
lead levels associated with older housing.

       For this sensitivity analysis, the following five alternatives for adjusting the HUD
National Survey data were made, in an effort to reflect more current environmental-lead levels in
households:
                                          220

-------
      •      Average dust-lead loading and dust-lead concentration reduced by 20%, (yard-
             wide) average soil-lead concentration reduced by 20%

      •      Average dust-lead loading/concentration reduced by 50%, average soil-lead
             concentration reduced by 50%

      •      Average dust-lead loading/concentration reduced by 50%, average soil-lead
             concentration reduced by 0%

      •      Average dust-lead loading/concentration reduced by 0%, average soil-lead
             concentration reduced by 50%

      •      Average dust-lead loading/concentration increased by 25%, average soil-lead
             concentration increased by 25%.

The dust-lead loading assumptions are assumed to be for both floors and window sills and are
made to the reported Blue Nozzle loadings (i.e., those estimates used as input to the empirical
model).  The same changes are assumed for Blue Nozzle floor dust-lead concentration, which is
used as input to the IEUBK model.

      Each of the above five sets of alternatives implies that the same percentage change would
be applied to data from each housing unit in the HUD National Survey. Thus, the resulting
national distribution of baseline environmental-lead levels would be a simple shift in the current
distribution used in the §403 risk analysis, with no change in the variability associated with the
distribution. Insufficient data exist to determine how a distribution's variability may have
changed, so it is assumed to remain unchanged. Within each set, different percentage changes
are considered for dust-lead loading and soil-lead concentration to allow for added flexibility in
how lead levels may have changed in different media.  Four of the five sets represent declines
over time, which are expected due to the increased prevalence of homes with no lead-based paint
hi the housing stock and the reduced likelihood of residual contamination associated with leaded
gasoline emissions. Nevertheless, one set representing an increase is considered, due to the
potential  for the new survey to include housing with generally higher levels than those houses
included in the original survey and the continued potential for deteriorated lead-based paint and
other lead sources to contaminate dust and soil.

      Note that in all five sets of adjustments, the assumed within-house geometric standard
deviation (GSD) remains equal to 1.6. Alternative values of this GSD assumption were
considered in the sensitivity analysis presented in Section 5.4.6 of the §403 risk analysis report.

      Table 5-6 presents the pre-§403 model-based estimates for the health effect and blood-
lead concentration endpoints, under each of the above five sets of data adjustments, as calculated
based on  blood-lead distribution generated from IEUBK and empirical model fits. For
comparison purposes, the table includes the estimates assuming no adjustments (i.e., as the data
were used in the §403 risk analysis) as reported in Table 5-2 of the §403 risk analysis report.

                                          221

-------
Table 5-6.   Sensitivity Analysis on How Changes in Household Average Baseline Dust-
             Lead Loadings/Concentrations and Soil-Lead Concentration Impact Pre-§403
             Estimates of Health Effect and Blood-Lead Concentration Endpoints for
             Children Aged 1-2 Years
Assumed Percentage Change in Average Dust-Lead-Loadings and Concentrations ,
(Both Floor and Window SOI) and in Yard-wide Average Soil-Lead Concentration s - •
Dust:
Soil:
. No change
No change
20%
decrease
20%
decrease
50%.
decrease
50%
decrease
50%
decrease
No change
•J— .J.-. nnt*
no cnange
50%
decrease
25%
increase
25%
- increase
Predicted Health Effect And Blood-Lead Concentration Endpoints {Based on Empirical Model)
PbB 220 {%)
PbB2lO(%)
IQ < 70 <%)
IQ decrement 2 1 (%)
IQ decrement 22 (%>
IQ decrement 23 (%)
Avg. IQ decrement
0.0278
1.54
0.0997
34.5
4.53
0.718
0.932
0.0212
1.28
0.0983
31.8
3.87
0.584
0.896
0.0117
0.849
0.0957
26.5
2.74
0.373
0.825
0.0187
1.17
0.0977
30.6
3.61
0.532
0.880
0.0176
1.13
0.0974
30.1
3.49
0.509
0.873
0.0364
1.85
0.101
37.2
5.27
0.877
0.969
Predicted Health Effect And Blood-Lead Concentration Endpoints (Based on IEUBK Model)
PbB 220 (%)
PbB2lO{%)
IQ < 70 (%)
IQ decrement 2! (%)
IQ decrement 22 (%)
IQ decrement 28 (%)
Avg. IQ decrement
2.24 .
12.4
0.146
50.4
19.9
8.95
1.40
1.39
9.33
0.131
45.1
15.8
6.46
1.24
0.427 .
4.60
0.110
34.6
8.97
2.90
0.978
0.957
7.28
0.121
40.3
12.8
4.92
1.12
1.44
9.70
0.132
46.4
16.4
6.72 ^
1.26
3.06
15.3
0.160
55.4
23.8
11.3
1.56
       Effect on risk analysis: The greatest total decline in baseline environmental-lead levels
being considered is the set containing 50% declines in both dust- and soil-lead levels (i.e., the
fourth column of Table 5-6).  Under the empirical model, Table 5-6 indicates that the most
sensitive endpoints to the 50% decline in both dust-lead and soil-lead are the incidence of IQ
decrement of at least 3 and the incidence of blood-lead concentration of at least 10 ug/dL, where
declines of 48% and 45%, respectively, were observed in these estimates relative to no decline in
environmental-lead levels. The empirical model-based estimates appear to be more sensitive to
changes in soil-lead concentration than in dust-lead concentration, as lower estimates were
observed when soil-lead concentrations declined by 50% (and no change was made to dust-lead
loadings) than when dust-lead loadings declined by 50% (and no change was made to soil-lead
concentrations). This is explained by the empirical model's larger slope estimate for soil-lead
                                          222

-------
concentration than for dust-lead loading in either floors or window sills (Table 4-3 of the §403
risk analysis report).

       When the IEUBK model is used to estimate the distribution of blood-lead concentration,
corresponding declines of 68% and 63% were observed for incidence of IQ score decrement at or
above 3 and for incidence of blood-lead concentration at or above 10 ug/dL, respectively, when
50% declines are assumed for both dust-lead and soil-lead concentration (Table 5-6). However,
hi this same scenario, the greatest decline (81%) among the endpoints is observed with the
incidence of blood-lead concentration at or above 20 ug/dL.  This is  a considerable decline
compared to the 24% decline for this endpoint observed under the empirical model.  Contrary to
the type of finding observed under the empirical model, the IEUBK model-based estimates
appear to be more sensitive to changes in dust-lead concentration than in soil-lead concentration,
as lower estimates were observed when dust-lead concentrations declined by 50% (and no change
was made to soil-lead concentrations) than when soil-lead concentrations declined by 50% (and
no change was made to dust-lead concentrations).

       Under both models, the last column in Table 5-6 shows that only modest increases in the
risk estimates were observed under the one adjustment assumption involving increases in
environmental-lead levels (i.e., 25% increases in both dust-lead and soil-lead levels).

5.1.5 Impact on the Estimated Incidence of IQ Point Decrement
       Assuming Certain Thresholds on the IQ/Blood-Lead Relationship

       As discussed in Chapter 4 of the §403 risk  analysis report, results of the meta-analysis
documented hi Schwartz (1994) indicate that an average IQ point loss of 0.257 is predicted for
every 1.0 ug/dL increase in blood-lead concentration, with no evidence of a threshold in this
relationship (i.e., non-zero blood-lead concentration below which the predicted IQ point loss is
zero). These results were used in the §403 risk analysis to characterize the IQ/blood-lead
relationship. Section 2.3  of this report provides additional justification for making these
assumptions.

       As discussed in Section 2.3, some researchers have suggested that a non-zero threshold
exists in the IQ/blood-lead relationship. While no consensus on a single threshold has been
adopted among those making this conclusion, and  such conclusions are occasionally made by
visual inspection of data rather than on statistical criteria, this sensitivity analysis considers the
impact of assuming non-zero blood-lead concentration thresholds on the baseline and model-
based pre-§403 risk estimates. A non-zero threshold will result in reduced estimates for health
effects measured by IQ decrement, as children with blood-lead concentrations below the
threshold will have an estimated IQ decrement of zero due to lead exposure.

       Estimates of the IQ decrement parameters under the following thresholds are presented in
this subsection: 1,2,  3, 5,8, and 10 ug/dL. hi addition, the estimates under an assumed
"threshold" of 0 ug/dL (i.e., those measured in the §403 risk  analysis and meaning that any
blood-lead level, regardless of how small, would have an adverse effect on a child's IQ score) are

                                          223

-------
presented for comparison purposes. The candidate threshold of 8 ug/dL has been suggested by
Rabinowitz et al. (1992), as discussed in Section 2.3. The candidate threshold of 10 (ig/dL was
selected as it represents the action level reported by the Centers for Disease Control and
Prevention (Section 2.5.1 of the §403 risk analysis report). It also is representative of the higher-
level thresholds reported by some early studies; thresholds any higher than 10 ug/dL would result
in extremely low (and likely very underestimated) risk estimates and have been discounted by
more recent studies. Levels of 1,2, 3, and 5 ug/dL represents possible candidates of a very low,
but positive, threshold.  Such thresholds would not be detectable by many studies in the literature
as they tend to fall below the range of observed data or the detection limit of the blood-lead
measurement procedure.

       The assumption of a positive threshold requires a minor modification to the method used
to predict IQ score decrement based on  blood-lead concentration.  An average IQ point loss of
0.257 continues to be predicted for every 1.0 ug/dL increase in blood-lead concentration, but
only above the assumed blood-lead concentration threshold value. Thus, if T represents the
threshold, then the predicted IQ score decrement at a blood-lead concentration of C would equal
0.257*(C-T) if C is greater than T, or zero if C is less than or equal to T.   While the
methodology used to obtain risk estimates remains the same as that documented in Appendices
El and £2 of the §403 risk analysis report, slight differences were required for calculating the
average and standard deviation of IQ decrement, as this measure was no longer assumed to be
lognormally distributed. See Appendix B for how these statistics are calculated assuming a non-
zero threshold.

       Table 5-7 presents the estimated percentages of children with IQ score decrements greater
than or equal to 1,2, or 3, and the average and standard deviation IQ point decrement under
assumptions of an IQ score decline of 0.257 points for every 1.0 ug/dL increase in blood-lead
concentration above the specified threshold. These estimates are presented assuming the baseline
blood-lead distribution (top section of the table), the pre-§403 distribution as generated by
BEUBK model fits (middle section of the table), and the pre-§403 distribution as  generated by the
empirical model fits (bottom section of the table) for children aged 1-2 years.

       Effect on risk analysis:  The magnitude of the assumed blood-lead concentration
threshold has a considerable impact on  the percentage of children affected by decrements in IQ
score.  As seen in Table 5-7, while the §403 risk analysis estimated an average IQ decrement of
1.06 points occurs due to lead exposure across the population of children aged 1-2 years, this
average declines by approximately 44% under a assumed threshold of 2 ug/dL (0.588 points) and
by 90% under a threshold of 8 ug/dL (0.103 points).  An estimated 38.5% of children aged 1-2
years were expected to experience an IQ score decrement of at least  1 if a threshold was not
assumed. This percentage is decreased by approximately 50% under a threshold of 2 ug/dL
(19.6%) and by 90% under a threshold of 8 ug/dL (3.5%). The percentage decline is decreased
in magnitude as the lower limit of IQ score decrement increases to 3, but it remains at least a
39% decline for a threshold of 2 ug/dL and 83% for a threshold of 8 ug/dL.
                                          224

-------
Table 5-7.    Sensitivity Analysis on the Assumed Blood-Lead Concentration Threshold on
               IQ Decrement and Its Impact on the Pre-§403 Estimates of IQ Decrement
               Endpoints for Children Aged 1-2 Years
Assumed
Threshold
: (pg/dU

0
1
2
3
5
8
10
% of Children Aged 1-2 Years with a Specified IQ
Decrement Due to Lead Exposure1
IQ Decrement 2
'"i:;;,U'L;;
10. Decrement 2
•';•" ":':';2 •..'
Baseline Estimates (Section 5.1
38.5
27.3
19.6
14.2
7.83
3.50
2.15
10.8
8.08
6.10
4.66
2.80
1.40
0.915
IQ Decrement 2 ;
•.' ' '-3 •
Average IQ
Decrement
(# points)2
. Standard
Deviation of IQ
Decrement?
.1 of §403 risk analysis report)
3.70
2.88
2.26
1.80
1.16
0.627
0.429
1.06
0.804
0.588
0.428
0.233
0.103
0.0638
0.895
0.891
0.860
0.802
0.666
0.494
0.408
Pre-§403 Estimates Based on IEUBK Model-Generated PbB Distribution
(Section 5.1.2 of 1403 risk analysis report)
0
1
2
3
5
8
10
50.4
39.3
30.8
24.4
15.7
8.58
5.96
19.9
16.0
13.0
10.6
7.27
4.31
3.13
8.95
7.42
6.19
5.20
3.73
2.35
1.76
1.40
1.15
0.921
0.738
0..483
0.273
0.194
1.35
1.35
1.33
1.28
1.15
0.964
0.854
Pre-1403 Estimates Based on Empirical Model-Generated PbB Distribution
(Section 5.1.2 of §403 risk analysis report)
0
1
2
3
5
8
10
34.5
20.4
12.0
7.14
2.61
0.652
0.278
4.53
2.76
1.71
1.07
0.443
0.130
0.0613
0.718
0.464
0.330
0.202
0.0926
0.0312
0.0158
0.932
0.675
0.442
0.271
0.0972
0.0224
0.00912
0.538
0.537
0.514
0.453
0.309
0.162
0.108
' A 0.257 IQ decrement is assumed for each 1.0//g/dL increase in PbB above the assumed threshold (see Section 4.4.1 of
the §403 risk analysis report).  Thus, the fallowing hold:
        •       P[IQ i  11 « P[PbB 2 (threshold + 3.9 jig/dL]
        •      P[IQ i  2] = PIPbB i (threshold + 7.
        •      PHQ *  31 = RPbB * (threshold +11.
2 Average and standard deviation of tQ decrement are calculated assuming no IQ decrement occurs below the assumed
threshold, and a  0.257 IQ decrement is assumed for each 1.0 (igldL increase in PbB above the threshold.

-------
       Similar patterns of decline were seen for the pre-§403 estimates generated under the
EUBK and empirical model-based blood-lead distributions, with the empirical model predicting
greater reductions for the larger thresholds. These model-based estimates were used in the
procedure to characterize changes from baseline that occur in a post-§403 environment, which is
addressed in Chapter 6.
                                          226

-------
6.0    ANALYSIS OF EXAMPLE OPTIONS FOR THE §403 STANDARDS

       Chapter 6 of the §403 risk analysis presented the methodology used to characterize
reductions to childhood health effect and blood-lead concentration endpoints expected to result
after interventions are conducted in response to the proposed §403 rule and applied this
methodology to a broad range of example options for standards. Assumptions were made on
post-intervention environmental-lead levels, which were applied to those HUD National Survey
housing units where a particular intervention was triggered as a result of having environmental-
lead levels that exceeded an example standard. Then, the EEUBK and empirical models were
used to generate the post-§403 blood-lead concentration distribution given post-§403
environmental-lead levels.  These results, combined with similar model-based estimates in the
pre-§403 environment presented in Chapter 5, were used to obtain a final post-§403 blood-lead
distribution which was comparable to the baseline distribution generated by data from Phase 2 of
NHANES m.  This procedure was detailed in Chapter 6 and Appendix Fl of the §403 risk
analysis report. This was the distribution upon which the health effects and blood-lead
concentration endpoints were estimated in the post-§403 environment.

       The risk management procedure in Chapter 6 of the §403 risk analysis report considered
example standards  for the following risk assessment measures:

       •      Average floor dust-lead loading
       •      Average window sill dust-lead loading
       •      Average soil-lead concentration
       •      Amount of deteriorated lead-based paint requiring paint maintenance
       •      Amount of deteriorated lead-based paint requiring paint abatement

Note that the lead-based paint standards considered in the risk management procedure differed
somewhat from the standards proposed in the §403 rule (see Chapter 1 of this report), as the rule
considered only a single tier rather than a two-tiered standard.

       Section 6.1  presents additional detail and results on the performance characteristics
analyses, a non-modeling data analysis procedure used by EPA to help establish levels of concern
within the §403 rule. Performance characteristics analyses cited in the §403 proposed rule are
detailed, and additional performance characteristics analyses performed after the proposed rule to
address public  comments and to finalize the rule are presented.

       Section 6.2 investigates the incidence of children with elevated blood-lead concentrations
in homes where no candidate standard is met or exceeded (i.e., children who would be "missed"
by a specified set of candidate standards).

       Since the §403 risk  analysis report was published, public comment resulted in an
additional investigation into the assumptions made in the risk management on average dust-lead
loading following an intervention involving dust cleaning (40 ug/ft2 on floors, 100 Mg/ft2 on
                                          227

-------
window sills). The results of this investigation are presented in Section 6.3. Based on this
investigation, the impact of alternative assumptions on post-intervention dust-lead loadings on
characterizing the reduction hi risk as a result of implementing §403 rules was evaluated through
a sensitivity analysis presented in Section 6.4.  Also included in Section 6.4 are sensitivity
analyses applied to baseline (pre-§403) data within Section 5.1 of this report to evaluate the
impact of potential changes to the HUD National Survey data and assumptions on non-zero
thresholds for the IQ/blood-lead relationship, where the analyses are implemented on data
representing the post-§403 environment.

6.1    PERFORMANCE CHARACTERISTICS ANALYSES

       The procedures defined and discussed in the §403 risk analysis report used statistical
modeling techniques to characterize risks of lead exposure to children in the nation's housing
stock and how these risks may be reduced as a result of interventions performed to reduce lead-
based paint hazards in the housing stock under the §403 rule. While using the findings of this
risk analysis to evaluate options for the standards specified in the §403 rule, EPA also wished to
base its evaluation partially on a non-modeling approach using data from field studies that
measured lead levels in both children's blood and in the same environmental media targeted by
the §403 rule. In particular, given the data reported in these studies, EPA was interested in
observing how often a specified set of candidate standards would "trigger" interventions hi
housing units within these studies and the extent to which these units contained a child with an
elevated blood-lead concentration (s 10 ng/dL). Such an investigation provided useful
information on the performance of a specified set of candidate standards without some of the
complexities associated with making conclusions from statistical modeling analyses.

       EPA employed performance characteristics analysis, sometimes referred to as
sensitivity/specificity analysis, as a non-modeling approach to evaluating candidate §403
standards. The underlying statistical principle of this approach involves conditional probabilities
and has been documented hi references such as Fleiss (1981, Section 1.2). This chapter presents
the findings of performance characteristics analyses applied to data from the Rochester Lead-in-
Dust study.  Applying data from this study was highly appropriate under the objective to evaluate
candidate lead standards in the §403 rulemaking.  The form of the study data used in this analysis
is discussed hi detail within Section 6.1.1. The methods used to perform this performance
characteristics analysis are presented in Section 6.1.2. Section 6.1.3 presents the results of
performance characteristics analysis presented in the preamble and which were used in the §403
rulemaking. Finally, Section 6.1.4 presents additional performance characteristics analyses
performed after the §403 proposed rule was published, where these analyses considered other
sets of standards (including the standards specified in the  §403 proposed rule) and other means of
handling data on amount of deteriorated paint within a household.

6.1.1  Data Used in The Performance Characteristics  Analysis

       The performance characteristics analysis was  applied to data from the recently-conducted
Rochester Lead-in-Dust study. A summary of objectives  and design information for this study is

                                           228


-------
found in Section 3.2.2.2 of the §403 risk analysis report. The Rochester study data were selected
for this analysis for the following reasons:

       •      The study reported information for all media for which §403 standards were
             proposed (e.g., dust-lead on floors and window sills, soil-lead, condition of lead-
             based paint).

       •      The study measured blood-lead concentration in 205 children aged 12-31 months
             who resided in the selected homes.

       *      The dust sampling methods used in this study included the wipe technique, from
             which dust-lead loadings were measured.

       •      For some homes, soil was sampled from multiple locations (i.e., dripline and play
             areas), allowing for yardwide average soil-lead concentration to be estimated.

       •      While homes and children were targeted for selection in this study, the selection
             process was more random and more representative of a general population than is
             the case with other lead exposure studies.

The primary concern with using data from the Rochester study in this analysis is the degree to
which the study may be considered representative of the nation as a whole.  The study selected a
targeted sample which was limited to a single geographic area.  The sample consisted of children
who had moderate exposure to lead in their home environment and did not necessarily include
children with very high or very low exposure to lead. In particular,

       •      22.9% of the  children in this study (47 children total) had blood-lead
             concentrations at or above 10 ng/dL, compared to the national estimate of 5.9%
             for children aged 1-2 years according to Phase 2 of the Third National Health and
             Nutrition Examination Survey (NHANES HI) (CDC, 1997).

       •      The geometric mean blood-lead concentration for the study children was 6.38
             |ng/dL with a geometric standard deviation (GSD) of 1.85. This compares with a
             geometric mean of 3.1 (ig/dL and GSD of 2.09 estimated for U.S. children aged 1-
             2 years according to Phase 2 of NHANES m (CDC, 1997).

       •      At least 84%  of the housing units included in this study were built prior to 1940,
             compared to the estimated 20% of the entire U.S. housing stock made within the
             §403 risk analysis (Table 3-5 of USEPA, 1998). There is a well-documented
             relationship between age of housing and presence of lead-based paint hazards.

       •      While geometric mean floor dust-lead loadings were comparable between the
             Rochester study and the HUD National Survey (after converting the data to wipe-
             equivalent loadings), whose results are considered nationally-representative,

                                          229

-------
              geometric mean estimates of window sill dust-lead loading and soil-lead
              concentration were higher for the Rochester study relative to HUD National
              Survey estimates (Section 3.2).

Despite these limitations, the Rochester study is considered one of the best resources of data for
characterizing the relationship between children's blood-lead concentration and residential
environmental-lead levels, and therefore, for evaluating national standards for lead in the nation's
housing stock.

       While data were available for 205 units in the Rochester study, somewhat fewer of these
units had values for all required data endpoints for this analysis. In particular, 177 units had data
reported on the amount of deteriorated lead-based paint, plus lead measurements for floor dust
(wipe), window sill dust (wipe), and soil (dripline and/or play area).  Of these units, 77 had soil-
lead data for both dripline and play areas, thereby allowing an average concentration across these
two areas to be calculated.

       For the analysis presented in the §403 proposed rule, the following five data endpoints
were calculated for each Rochester study housing unit:

       •       Area-weighted average uncarpeted floor wipe dust-lead loading (i.e., the measured
              loading for each sample was weighted by the area of the sample when averaged)

       •       Area-weighted average window sill wipe dust-lead loading

       •       Average of dripline and play area soil-lead concentrations

       •       The percentage of interior painted components tested in the study that contained
              lead-based paint (measurements at or above 1.0 mg/cm2) and some level of
              deterioration (paint condition listed as fair or poor)

       •       The percentage of exterior painted components tested in the study that contained
              lead-based paint and some level of deterioration.

Note that these endpoints are comparable to the standards included in the §403 proposed rule,
with the exception of the latter two paint-lead measurements.  While the proposed §403 standard
for the paint component is expressed as a square footage of deteriorated lead-based paint for
components with large surface areas (2 ft2 for interior surfaces, 10 ft2 for exterior surfaces) or as
the percentage of total painted surface area that is deteriorated for components with small surface
areas (10%), no indication on the amount of deteriorated lead-based paint on a given component
(either in square feet or as a percentage of the total surface area) was recorded in the Rochester
study.  Instead, each paint-lead measurement was associated with an indicator of the paint's
condition (good, fair, poor). Therefore, for this analysis, the amount of deteriorated lead-based
paint in a housing unit was taken to be the percentage of tested components in the housing unit
that contained lead-based paint along with some level of deterioration (i.e., condition of paint

                                          230

-------
either fair or poor). This result was assumed to he a good estimate of the total amount of lead-
based paint in the unit that was deteriorated.

6.1.2  Analysis Approach

       The performance characteristics analysis classified each housing unit in the Rochester
study according to two different criteria:

       1.     Whether or not the unit exceeded anv of the candidate standards for the various
              media being controlled.

       2.     Whether or not the unit contained a child with elevated blood-lead concentration
              (alOug/dL).

The first criterion represented whether a housing unit was "triggered" for any intervention by
exceeding at least one candidate standard, while the second represented whether the unit
contained a child requiring attention as a result of having an elevated blood-lead concentration.
The first criterion was determined by noting whether the value for at least one of the five
endpoints mentioned at the end of the previous section exceeded the standard associated with the
type of measurement represented by that endpoint.

       For a given set of candidate standards, the set of housing units in the Rochester study was
identified that had data for all of the above five endpoints. These units were classified according
to whether or not they achieved the above two criteria. These results are summarized in the
manner illustrated within the 2x2 frequency table in Table 6-1. From this information, the four
performance characteristics defined in Table 6-1 were then calculated: sensitivity, specificity,
positive predictive value (PPV), and negative predictive value (NPV). These characteristics
provide the necessary information for evaluating the sets of standards on their ability to target the
proper set of units for intervention.

       In this analysis, a "false positive" corresponds to triggering a housing unit for intervention
when it does not contain a child with an elevated blood-lead concentration, while a "false
negative" corresponds to not triggering a housing unit containing a child with an elevated blood-
lead concentration. Note that the proportion of false positives is equal to one minus the
specificity, while the proportion of false negatives  is equal to one minus the sensitivity.

       While information from all four performance characteristics are important for evaluating
the performance of a given set of standards, typically one or two characteristics are given more
weight than the others in the performance evaluation process. For example, in the preamble,
EPA evaluated candidate standards for dust-lead loading on uncarpeted floors and window sills
according to whether the performance characteristics analysis yielded a value of NPV from 95 to
99 percent under the given set of standards. This implied that no more than 5% of children living
in housing units with environmental-lead levels below the standards would have elevated blood-
lead concentrations (i.e., at or above 10 ug/dL).  More recent Agency inquiries have focused on

                                           231

-------
Table 6-1.    Definitions of Performance Characteristics Used to Evaluate How Various
              Combinations of Environmental-Lead Standards Classify Housing Units in the
              Rochester Lead-in-Dust Study
Blood-Lead Concentration
At or Above 10/ig/dL?
In the above table, the lette
or above 10^/g/dL who live
specified standards. Letters
equals a + b+c+d. From th
Performance
Characteristic
Sensitivity
(or True Positive Rate, or
1 - False Negative Rate)
Specificity
(or True Negative Rate, or
1 - False Positive Rate)
Positive Predictive Value
(PPV)
Negative Predictive Value
(NPV)

Yes
No
Any of the Standards Exceeded?
No
a
c
Yes
b
d
r 'b' represents the number of children which have a blood-lead concentration at
in a residence with environmental-lead levels that exceed at least one of the
> 'a', 'c', and 'd' represent similar counts. The total number of housing units
sse counts, the following performance characteristics are calculated:
Definition ;
Probability of a housing unit exceeding at least one
standard given that there is a resident child with an
elevated blood concentration (2 1 0 //g/dU
Probability of a housing unit not exceeding at least
one standard given that a resident child has a low
blood-lead concentration (< 10//g/dL).
Probability of a resident child having an elevated
blood-lead concentration <210 j/g/dL) given that the
housing unit exceeds at least one standard.
Probability of a resident child having a low blood-
lead concentration « 10 //g/dL) given that the
housing unit does not exceed at least one standard.
Calculation
b/fa + b)
c/(c+d>
b/
-------
      so
      40
   £  30

   |
   c
  -3
   "g  20
   o
   s
      10
            •  •.  x*
         \ . f    .. •
                   400         BOO        1.200        1.600
                           Dripline Soil-Lead Concentration (^g/g)
2.000
2,400
Figure 6-1.    Example of an Ideal Situation for Establishing Potential Dripline Soil-
              Lead Standards
                                        233

-------
        50
       40
       30
     v
     o
     o
    u
    •  30
    •o
     o
    i
        10
           S .  f
                                                                           •  1
                     400         800        1.200        1.600
                             Dripiine Soil-Lead Concentration
2.000
2,400
  Figure 6-2.   Example of a Situation Where the Negative Predictive Value and
               Sensitivity Equal 100%, but the Positive Predictive Value and Specificity
               are Less than 100%
       The performance characteristics analysis was repeated for different sets of standards. For
each analysis, the information within Table 6-1 was calculated, and those sets of standards that
maximized the desired performance criteria were identified.

       The different analyses presented in the subsequent sections of this chapter were
performed on different subsets of housing units in the Rochester study. The results are purely
descriptive in that they represent combinations of candidate standards that meet the specified
performance criteria when considering the housing units in the Rochester study and are not based
on any underlying probability model. Different results are possible if this analysis were to be
applied to data from different studies. In addition, only point estimates of the performance
characteristics are presented. The uncertainty in these estimates is primarily dependent on
sample size, and to a lesser degree on measurement error.
                                           234

-------
6.1.3  Results Cited in the §403 Proposed Rule

       The analysis presented in this section were cited in section B.l.d of Part IV of the
preamble. This section of the preamble contained a brief presentation of the information
presented in Section 6.1.2 above, then cited findings of analyses documented in a memorandum
dated 9/3/97 from Battelle (Ronald Menton and Warren Strauss) to EPA (Todd Holderman).
EPA requested that Battelle perform this analysis in an action item of a meeting between Battelle
and EPA on August 27,1997.  A copy of the cited memorandum is found in Appendix G.

       The analyses presented in Appendix G were performed on  data for the 77 housing units in
the Rochester study that had all necessary data for the analysis, including soil-lead concentrations
for both dripline and play areas.  As the §403 proposed rule was to contain a yardwide average
soil-lead standard, it was desired to consider only those housing units that had soil-lead data for
both locations. The considerable reduction in the number of Rochester study housing units
whose data were considered in this analysis (from 205 to 77 units) was due primarily to the fact
that play-area soil-lead concentration was measured for less than half of the study units.

       The combinations of candidate standards considered in this analysis were those requested
by EPA at the time, when EPA was actively considering candidate standards in the rulemaking.
These combinations included all 8x4x9x3=864 combinations of the following:

       •     uncaroeted floor dust-lead loading: 50,75, 100,125,150,175,200,400 ug/ft2
       •     window sill dust-lead loading:  100,300,500,800 ug/ft2
             average soil-lead concentration: 200,300,400,500,600,700,900,1000,1500
       •      maximum of percent of interior/exterior painted surfaces with deteriorated lead-
             based paint: 5,10,20%

Note that the type of endpoint that represented the paint-lead measurement in this analysis (i.e.,
the last bullet) differed from the type of paint-lead standard that EPA ultimately proposed in the
§403 proposed rule.

       The purpose of this analysis was to identify those sets of candidate standards (from the
864 combinations above) which, when applying the performance characteristics analysis under
those sets of standards, resulted in values of negative predictive value (as defined in Table 6-1
above) that met one of the following three criteria:

             NPV * 99%
             95%*NPV<99%
             90% s NPV < 95%.

The findings of this analysis are documented in Tables 1 and 2 of Appendix G.
                                          235

-------
       Twenty-one of the 77 housing units whose data were included in the analysis did not
exceed any of the candidate standards in at least one of the 864 combinations of candidate
standards. These housing units, with the values of the endpoints used to compare to the
candidate standards and children's blood-lead concentration, are listed in Table 6-2. This means
that the denominator of NPV (i.e., the number of housing units that do not exceed at least one of
the candidate standards being considered) never exceeded 21 across the 864 combinations.  For
some combinations, the denominator was as small as 2. Furthermore, all but two of the 21 units
in Table 6-2 contained children with blood-lead concentrations below 10 ug/dL. As a result, the
value of NPV was no lower than 84.6% across all 864 combinations of candidate standards. At
least one of the above three criteria for NPV was met for 808 (93.5%) of the combinations. Of
these 808 combinations, NPV equaled 100% for 690 of the combinations, equaled 95% for seven
combinations, and was at least 90% but below 95% for the remaining 111 combinations.

       All of the remaining 56 housing units in the analysis that are not represented in Table 6-2
exceeded either the soil-lead standard or one of the two paint standards (i.e., interior and/or
exterior) in each  of the 864 combinations of candidate standards. That is, each of these houses
had at least one of the following:

       •       average soil-lead concentration of at least 1500 ug/g
       •       at least 20% of painted surfaces with deteriorated lead-based paint in the interior
              and/or exterior.

Therefore, the 56 housing units not represented hi Table 6-2 were triggered in each of the 864
combinations of candidate standards, without regard to the floor or window sill standards.

       The results presented in this section led to the following conclusions stated in Part IV of
the preamble:

       "For uncarpeted floors, dust-lead loadings ranged from 50 ug/f? to 400 pg/f?
       depending on the dust-lead loading on interior window sills and the soil-lead
       concentration. For interior window sills, dust-lead loadings ranged from 100
       fig/ft* to 800 fig/ft2 depending on the dust-lead loading on uncarpeted floors and
       the soil-lead concentration. These ranges are significantly higher than the ranges
       yielded by the multimedia approach."

       "Soil-lead concentrations ranged from 200 ppm to l.SOOppm depending on dust-
       lead loadings on uncarpeted floors and interior window sills and the exceedance
       probability."

The ranges cited in the preamble were precisely the lower and upper ranges of the candidate
standards considered in this analysis. These findings reflect the very high values of the NPV
across the combinations of standards considered in this analysis.
                                          236

-------
Table 6-2.   Set of 21 Housing Units in the Rochester Study in Which No Standard Was
             Exceeded in at Least One of the 864 Combinations of Candidate Standards
Housing ID -
00034
00132
00302
00637
00874
00974
01047
01062
01195
01228
01930
01971
01991
02290
02411
02837
03174
03360
03527
05343
05498
*.; , / Statistics Compared to, _the Candidate Standards1,, £,."-" <
Roor Dust-
Lead Loading
to/ft2}
63.60
17.30
2.55
59.00
12.90
14.90
20.83
12.40
12.25
3.37
19.35
5.10
15.50
2.65
10.48
4.29
18.60
12.43
6.08
10.30
19.15
Window Sfll
Dust-Lead
Loading -
Uig/ft2)
349.9
90.6
70.7
74.9
293.7
45.6
372.3
87.1
32.2
16.2
118.8
41.9
398.9
74.1
178.5
2.8
235.6
702.0
148.8
75.7
66.0
": rtt'j.
Average SoB-
Lead Cone.
U/g/gJ -
438.5
268.0
124.5
950.0
102.9
51.1
574.3
447.5
830.5
419.0
773.4
506.0
104.0
465.0
828.5
458.5
625.5
912.0
539.5
552.0
1150.5
% of Interior
Components
with
Deteriorated
LBP
17
0
0
0
0
0
18
0
0
0
0
11
0
11
10
0
0
0
14
13
0
% of Exterior
Components
with
Deteriorated ,
LBP
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Blood-Lead
Cone. U/g/dL)
7.1
6.0
4.8
13.3
2.1
8.9
3.9
7.4
6.9
4.6
4.6
6.1
7.5
4.9
9.0
8.9
5.8
11.3
4.5
5.6
5.8
1 See Section 6.1.1 for the definitions of these statistics.
                                        237

-------
6.1.4  Results of Analysis on Specified Sets of Standards

       The analyses presented in Section 6.1.3 were performed prior to release of the §403
proposed rule and contributed to the information presented in the preamble. Since the proposed
rule was released, EPA has requested additional performance characteristics analyses be
performed on various combinations of candidate standards, to address various issues raised
within the public comments to the proposed rule and in support of preparing the final §403 rule.
This section presents the results of these additional performance characteristics analyses.
Additional performance characteristics analysis results are presented in Appendix J.

       As discussed hi Section 6.1.3, one of the limitations of the analyses presented in the
preamble was the relatively small number of housing units (77) in the Rochester study whose
data were used in the analyses.  This small number was primarily due to the lack of available
soil-lead concentrations from play areas and the desire to have soil-lead data for both dripline and
play areas in order to calculate a yardwide average. Thus, the additional analyses presented in
this section re-defined how the soil-lead measure was calculated (with different approaches taken
to this re-definition), thereby increasing the number of units whose data could be included in the
analysis.

       6.1.4.1 Analyses Performed on 41  Combinations of Candidate Standards, in Three
Iterations. The candidate standards that were considered in this analysis were the following:

       •     uncarpeted floor dust-lead loading:  5,10,20,25,40,50,100,200 jig/ft2
       *     window sill dust-lead loading: 250 ug/fi2
       •     vardwide average soil-lead concentration: 400, 1200, 2000, 5000 (ig/g
       •     amount of deteriorated lead-based paint: 2% of interior painted surfaces or 10% of
             exterior painted surfaces.

Thus, different candidate standards for floor dust-lead loading and soil-lead concentration were
considered, while only a single candidate standard was considered for window sills (i.e., that
specified in the §403 proposed rule) and deteriorated lead-based paint. This analysis considered
a total of 41 combinations of candidate standards, corresponding to the 8x4=32 combinations of
the above candidates, as well as the additional 9 combinations:

       •     only the paint standards (1 additional combination)
       •     only the paint and soil-lead concentration standards (4 additional combinations)
       •     only the paint, soil-lead concentration, and window sill dust-lead loading
             standards (4 additional combinations).

For each combination, the four performance characteristics were calculated and presented, as
well as the number of housing units that exceed at least one of the specified standards.

       Note that the above candidate paint standard (percentage of paint that is deteriorated lead-
based paint) is not expressed in the manner that the proposed paint standard in the §403 proposed

                                          238

-------
rule was expressed (amount of deteriorated lead-based paint, in square feet). As discussed in
Section 6.1.1 above, the Rochester study measured only lead content in paint plus an indicator of
paint condition, and therefore, did not measure the surface area containing deteriorated lead-
based paint. For the Rochester study data, the above paint standard triggered all units with
deteriorated lead-based paint present, as the lowest observed non-zero percentage of deteriorated
lead-based paint was 8% for interior surfaces and 14% for exterior surfaces.

       Three iterations of this analysis was performed, with each iteration involving data for a
different number of housing units:

       Iteration #1: Instead of requiring soil-lead concentrations be reported for both dripline
       and play areas, as was done within the analysis cited in Section 6.1.3 above, average soil-
       lead concentration was set equal to the reported concentration at one of these areas if no
       concentration is reported for the other area. This approach permitted data for 177 housing
       units to be used in the analysis.

       Iteration #2: After taking the approach in iteration #1, any units that did not have soil-
       lead concentration reported due to having no bare soil available from which to sample
       were assigned a soil-lead concentration of 0 ppm. This approach was taken as Title IV of
       TSCA restricts the §403 soil-lead hazard standard to bare soil and further assuming that
       any covered soil at these units would not pose a soil-lead hazard. This approach
       permitted data for 184 housing units to be used in the analysis.

       Iteration #3: After taking the approach in iterations #1 and #2, the 21 remaining units
       having missing data for at least one endpoint had an imputed value assigned to the
       endpoint(s) equal to the average value across units within the same year-built category
       (pre-1940,1940-1959,1960-1979, post-1979) and having the same indicator of whether
       or not lead-based paint is present in the unit. This method followed the same approach
       taken in the §403 risk analysis (Section 3.3.1.1 of the §403 risk analysis report) to impute
       data for housing units in the HUD National Survey. This approach permitted data for 205
       housing units to be used in the analysis.

The results of each iteration are now presented.

Iteration #1: Data for 177 Housing Units.

       Table 6-3 presents the results of the performance characteristics analyses performed on
data for 177 housing units (#1 above) under the 41 combinations of standards listed above. Note
from this table that the fixed paint standards (which were equivalent to finding any deteriorated
lead-based paint in the unit) triggered an intervention for nearly three-fourths of the 177 units.
These paint standards considered jointly with a soil-lead concentration standard of 400 ug/g
resulted in 100% sensitivity and negative predictive value regardless of the dust standards.
Sensitivity and negative predictive values of 100% were also met at a soil-lead concentration
standard of 1200 ug/g if the floor dust-lead loading standard was at  10 ug/ft2 and the window sill

                                            239

-------
a
£
•§
 e
§
 «
c
 o
€
£
 0)
 8i>
111
 §0 
       1

       I
       -5
       %
                                          e
                                            '
                                         l
                                                       co

                                                       CO
                                                       CM
                                                       CM
                                                           10

                                                           S
CM
co"
S
                                                                     CM

                                                                     CO
                                                    CM

                                                    CO
                                                                         CM
                  CM

                  CO
                                                                              CM
              CM

              CO
              *
              CM
              CM
              co
              *
              CM
                                                                                           10
                                                                                           CM
                                CO

                                CO
                                CO
                                CM
                                                                                                    CO
                                                                                                    CM
                                                                                                    CM
                                                                                    CD
                                                                                    CO
                                                                                    CM
                                                  I*.
                                                  CO
                                                  CM
              CO
              CM
              CO
              CM
                                                       00

                                                       IO
                                                       CO
                                       CO
                                       CM
                                       m
                                       CM
     N

     CM
w

CM
                                                                         O
                                                                         o
N


CM
                                                                              CM
                                                                              CM
                                                                                  8
                                                                                  CM
                                                                                  CM
                       8
                                                                                       CM
                                                                                                    8
                                                                                           CO


                                                                                           00
                                                                                                O>

                                                                                                3>
                                                                                                         CO
                                                                                                         CO
                                              S

                                              CO
                                              CO
                                              CO
                                                  o>

                                                  CD
                                                  CO
                                                  CO
                                                  CO
                  CO

                  CO
                  JO

                  CO
                  CO
                                  CO
                                  r*
                                  N

                                  CM
                                  CO
                                                       m
                                                           CO
                                                           CM
                                                           Ifl
                                                           CO
                                                                

                                                                                                         CO
                                                                                                         CO
                                                                                                                  CO
                                                                                                                  CO
                                                                                                                       CO
                                                                                                                       at
00
s
                                                                CO
                                                                00
                                                                CO
                                                                10
                                                                    00
                                                                    CO
                                                                    10
         CO
         00
                                                                         CO
                                                                         10
                                                                              CO
                                                                              CO
             CO
             10
                                                             CO
                                                             00
                                                                                  CO
                                                                                  10
                           CO
                           oo
                                                                  CO
                                                                  10
                       00
                       CO
                       oo

                       CO
                       to
                                                                                                0>
                                                                                                    eo
                                                                                                    co
                                                                                                              PS

                                                                                                              o>
                                                                                                                  CO
                                                                                                                       CO
                                                                                                         CO
                                                                     CM
                                                                         §
                                                                                      10
                                                                                      CM
                                                                                                     10
                                                                                                                  o
                                                                                                                  CM
                                                                ID
                                                                CM
                                                                    o
                                                                    10
                                                                    CM
         o
         10
         CM
                                                                             o
                                                                             10
                                                                             CM
                                                             o
                                                             10
                                                             CM
                           o
                           10
                           CM
                       to
                       CM
                       IO
                       CM
                                    s
                                    CM
         o
         10
         CM
         10
         CM
to
CM
                                                           O
                                                           O
                                                                    §

                                         o
                                         o
                                         CM
                                                                                                              CM
                                                                                                                  i

                                                           CM
                                                                CM
                                                                    CM
                                                                         CM
                                                                             CM
                                                                                       CM
                                                                                           CM
                                                                                                CM
                                                                                                    CM
                                                                                                         CM
                                                                                                              CM
                                                                                                                  CM
                                                                                                                       CM
                                                         240

-------
                                              co
q
CM

CM
                                                        CM
                                                        *t
                                                        CM
                                                             <0
                                                             CM
                                           CO
CO

CD
CM
CM
                                                                           CM
                                                     CO
                                                     CM
                              CM
                              CO
                              CM
                             CM
                             CO
                             CM
                                            CM
                                            CO
                                            CM
                                                                                               CM
          to

          CO
          CO
          CM
     IO

     CD
     CO
     CM
                                                                                                              CO
                                                                                                              CM
                                                      to
                                                      CO
                                                      CM
               ••
                       n
                       ft
                       o>
                       IO
                                              co
                                              to
                                                   CO
                                                   m
                                 CM
                                 PO
          CD
          s
          8
          3>
          CM
               *
               O
                                           CM
                                                                  CM
                                                
                                       co
                                       §
          CM

          CO
          CM

          CM
                                                      S
     CM

     00
     CM

     CM
     CO

     S
                                                                CO
                                                                CM
                                                                                            IO
                                                                                            CD
                                                                IO
                                                                CM
                                                                                                                        CO
SPECI
                                              CO
                                              CO
                                                   0>
                                                   co
                            
                                                 to
                                                 tM

                                                 CM
                                                 CO
                                  CD

                                  CO
                                  N

                                  CM
                                  CO
                                                                                                              co
                    CN

                    CD
                    CM
                                                                                            O

                                                                                            CM
                                                                IO
                                                                                                                        CO
 o
 u
m
(6
 (9
2
 a
                      3
0 -i •£
                                              CO
                                              IO
                                              o>
co
10
0>
CO
IO
ca
          en
          CM
                                           CO
                                                8
                                                CO
                        «t
                        co
                        oo
                        oo
                        to
                        en

                        0)
                        co
               o
               o>

               en
               co
O
co
cn
co
g
CO
en
                                                           CO
                                                           O)
                                                                                  o
                                                 fs
                                                 CD
                                                                                       CM
                                                                                            CO
                                                                                                 CO
                                              CM
                                                   CO
                                                        CO
                                                                  
-------

Aiioidioads
|


n
to


a
Set of Standard
to
CO
CO
CM
    CM
    00
    CO
    ID
   CM
    CO
   CO
   8
   IO
       CM
       CO
       to
       CM
       CO
       S
       CO
       8
       10
       CM
          CM
          CO
          10
          CM
           CO
          co
                                           8
          8
          CM
              to
              en
              00
              CO
              en
                                              8
              CM
                  CM
                  en
                  CM
                 CM
                     CM
                     CM
                     en
                     CM
                        00
                         CO
                        CO
                        CO
                        00
                        to
                        CM
                        CM
                            en
                            CO
                            CO
                            CM
                            CD
                            8
                            CO
                            en
                            IO
                            CM
                               CO
                               10
                               CO
                               IO
                               8
                               CM
                                    tsl
                                                                    ~ «
                                                                    «•§{;
                                                                               CD
                                                                               C
                                                                               O
                                                                              I
                                                                               CO
                                                                               s
                                                                               o
                                                                              1
                                                                    5 81
                                                                                    m
                                                                                    ui
                                                                                     5
                                                                                    £
                                    •
§
CD
f
S
5
•o

-------
dust-lead loading was at 250 ug/ft2, although the specificity declined considerably at these
standards. When the floor dust-lead loading standard was raised to 20 ug/ft2 in this situation,
both the sensitivity and negative predictive value remained above 95%. However, at soil-lead
standards of 1200 ng/g or higher, the 95% criterion for both sensitivity and negative predictive
value were no longer achieved once the floor dust-lead loading standard exceeded 20 ug/ft2.

       Among the 32 combinations of standards included in Table 6-3, the sum of the four
performance characteristics (i.e., the last column of the table) was maximized at 244.6% at a
floor dust-lead loading standard of 20 ng/ft2 and a soil-lead standard of either 1200 or 2000 ug/g.
(The paint and window sill standards were fixed in each combination.)

       The proposed §403 standards, assuming the different approach taken in this  analysis to
interpreting the paint standards, resulted in a 93% sensitivity (40 of 43 units containing an
elevated blood-lead child are triggered) and nearly a 92% negative predictive value  (see
shaded/bold row within Table 6-3). The sum of the four performance characteristics was
237.7%. Nearly 80% of the 177 units exceeded at least one of the proposed §403 standards.

Iteration #2:  Data for 184 Housing Units.

       Table 6-4 presents the same types of results as in Table 6-3, but it reflects analyses that
included data for seven additional housing units where soil-lead concentration was assumed to be
0 ug/g due to having no bare soil present for sampling (i.e., a total of 184 housing units).  Only
one of these seven additional units contained a child with an elevated blood-lead concentration.

       Slight reductions in the values of the performance characteristics were seen from Table
6-3 to Table 6-4 with the addition of these seven units. The one additional unit containing a
child with elevated blood-lead concentration did not exceed any of the paint, soil, or window sill
standards in the table and exceeded only floor dust-lead  loading standards below 50 jig/ft2.
However, as in Table 2-3, sensitivity and negative predictive values of 100% (and the
considerable declines in specificity) continued to occur at a soil-lead concentration standard of
1200 ug/g if the floor dust-lead loading standard was at  10 ug/ft2 and the window sill dust-lead
loading was at 250 ug/ft2.

       Despite the general declines in the values of the four performance characteristics from
Table 6-3, the largest observed value of the sum of these characteristics among the 32
combinations of standards (245.0) was slightly larger than in Table 6-3. This value was observed
for the same two combinations of standards for which the maximum occurred in Table 6-3: a
floor dust-lead loading standard of 20 ug/ft2 and a soil-lead standard of either 1200 or 2000 ug/g.

       The proposed §403 standards, assuming the different approach taken in this  analysis to
interpreting the paint standards, resulted in nearly a 91% sensitivity and nearly a 90% negative
predictive value, which were slight declines from Table  6-3 (see shaded/bold row within Table
6-4). The sum of the four performance characteristics was 233.2%.
                                           243

-------
 §
o
•o
 ID
 o
3
 «

 8
cc
 o
ill
«•
00
                       w
                       w
         :*.£i;^I-fe;
                       1

                       I
                       •5
                       s
                       
O
CO
                                                                 CO
                                                                 oo
                                                                 o
                                                                 (O
                                                CM
                                                                     §
                                                                     CO
                                                                     CM
                                                                     <0
                                                                     O
                                                                     o
                                                                     IO
                                                                     ao
                                                                     (O
                                                   CM
                                                                        CM
                                                       8
                                                                        ro
                                                                        CM
                                                                        03
                                                                        co
                                                       o
                                                       o
                                                       Tt
                                                                        IO
                                                                        oo
                                                                        CO
                                                       CM
           §
                                                                            00
*t

                                                           03
                                                                            CM
                                                                            ex
                                                           CM
    o
    o
                                                                                *
                                                                                O>
    o
    o
                                                               *t
                                                                                t
                                                                                en
                                                                                CO
                                                               CM
                                                                                    O
                                                                                    O
                                                                                    to
                                                                                    CM
                                                                                    O)
                                                                                    o
                                                                                    o
                                                                                    O)
                                                                                    IO
                                                                  CM
                                                                      CO I

                                                                      ro
                                                                      CM
                                                                                           CM
                                                                                           ro
                                                                                           co
                                                                                           CM
                              CM

                              CO
                              ro
                              CM
                       CM

                       CO
                       ro
                       CM
                                                                                        CO
                                                                                        CO
                          CO
                          00

                          CO
                          ro
                          S
                          n
                   co
                   CO

                   CO

                   S2
                   IO
                   ro
       co
       00

       CO

       S2
       IO
       ro
                                                                                               (O
                                                                                        CM
                                                                                           CM
                                                                                               CM
                                                                                                   cq
                                                                                                   r^
                                                                                                   CM
                                                                                           10
                                                                                               10
                                                                                                   to
                                                                                        en
                                                                                           IO
                                                                                           CM
                                                                                           IO
                                                                                           CO
                                                                              IO
                                                                              CM
                                                                              IO
                                                                              ro
                                  #
                                  q
                                  id
                                  CM

                                  IO
                                  ro
co

g
                                                                                               CO
d>

i
                                                                                           00
                                                                                               oo
                                                                                                   oo
                                                                                       
-------
                               a» m

                               H*
                                             CO
                                             CO
     CD

     CM
         CQ
                                                      CM
              10
              *
              CM
                                          0>

                                          id
                                          CO
                                          CM
     cn
     cn
     CM
     CM
     CM

     00
     CM
     CM
CM

00
CN
CN
CM

00
CM
CM
CM

CO
CM
CN
 CN

:;.«'

$•
                                                   CO
                                                                                                CO
                                                                                                CM
CO
rx
CO
CM
                                                                                         IO
                                                                                         *
                                                                                         CM
     cn
     id
     CO
     CM
     cn
     cn
     CM
     CM
en

oo
CO
in
CO
                                                 o»

                                                 CO
                                                 CO
                                                 m
                                                      o>
                                                      CO
                                          cn
                                          CD
                                          cn

                                          CM
                                          f
                                                               3
                                                                    O
                                                                    o
                                                   to
                                                   CO

                                                   IO
                                                   ^
                                                   cn
                                                   CO
                                                                             in
                                                                             00
              ^
              IO
              CO
                                                                                  *
                                                                                  IO
                                                                                  on
                                                                                  IO
                                                                                  to
                                                                                       IO
                   00

                   o

                   to
                   CO
                                                                                           to
                                                                                                en
                                                                                                en
                                                                                                CO
                                                                                                CO
                                                        en
                                                        »—
                                                        O)
                                                                                                     CO
                                          en
                                          CD
                                          cn

                                          CM
                                          CO
                                                                 #
                                      8


                                      fx^
                                      IX
                                             00
                                             CM
                                             CO
     00
     
                                                 IO
                                                 O)
                                                 CN
                                                      in
                                                      in
              cn

              CO
                   8
                                                   o
                                                   o
                            CD
                            CO
                            00
                            eo
CO

00
00

en
CO
CD

CO
CO

CO
CO
CD
CO
go

cn
CO
                                                                     .0}

                                                                     S
                                 CM
                                 CO
                                 o>
                                                                                                     CN
                                                                                        rx
                                                                                        en
                                                                                                         CO
                                               8
                                             CO
                                             cn
                                             to
                                                 O
                                                 eo
                                                 00
                                                      00

                                                      00
                                     CD

                                     CM
                                     CO

                                     CM
                                     10
                   CM
                   CO
                                                               O
                                                               r*
     CM

     CD
     cn
                            to
                            id
                            o>
                            m
                                                            CO
                                                            eo
              CO

              CD
                                                                                      CO
                                                                                           •
                                                                                           10
                   CO

                   CO
                                                                                                fx
                   en
                   en
                   cq

                   CN
                   00

                   CM
                   IO
                                                                                             CM
                                                                                             CO
                        CM

                        CD
                        cn
                 •s
                 •§
                                             to
                                                      IO
                                                      CN
                                                           CM
                       to
                                                                 o
                                                                 o
                                                                 CM
                                          8
                                                                     s
                                     to
                                     CM
                                                                                                         CM
                                             o
                                             IO
                                             (M
     IO
     CN
                                     IO
                                     CM
                                     IO
                                     CM
to
CM
O
to
CM
10
CM
to
CM
                                                                                      CM
     g
     CM
     to
     CM
S
CM
g
CM
IO
CM
to
CM
                                            i
     O
     O
     CN
                                                      CM
                                                           CM
                                              O
                                              O
                                              CM
                                              8
                                              CM
         8
                                                                                  CM
                                 8
                                 CM
                                                                                                    CM
                                                                                                         CM
                                                                                                              CM
                                                                                                                   O
                                                                                                                   CM
                                             N
                                                      CM
                                                           CM
                                                               CM
                                                                    CM
                                                                        CM
                                                                             CM
                                                                                  CM
                                                                                      CN
                                                                                           CN
                                                                                                CN
                                                                                                         CM
                                                                                                              CM
                                                                                                                  CM
                                                              245

-------

i

 . UI
ut >
> p ui



»5<
                 5*
                *s*
                         c «"


                           "
                     = -3 §
                     S   J-
                              00
                              01
                              CM
                              CO
                              m
                              to
                              CO
                              IO
                              OI
                                  CD
                                  OJ
                                  00
                                  04
                                  01
                                     CO
                                     CN
                                     01
                                        OI
                                        00
                                        01
                                        01
                                 8
                                        10
                                        OI
                                           00
                                           CO
                                           OI
                                           CO
                                           CO
                                           CD
                                           CO
                                           IO
                                    CM
                                           8
                                           OI
                                               CD
                                               OI
                                               m
                                               IO
                                              01
                                                  CD
                                                  OI
                                                  IO
                                                  a>
                                                  IO
                                                  in
                                                  01
                                                     CO
                                                     g
                                                     CO
                                                     CO
                                                     10
                                                     IO
                                                     01
                                                     01
                                                        w-

                                                        8
                                                        IO
                                                        00
                                                        o>
                                                        OI
                                                     w -

                                                     8
                                                            o>
                                                            i
                                                            in
                                                            OI
                                          246
                                                         Js?
                                                         •S s «

                                                         «•*
                                                         » c *
                                                         €.2 o
                                                          w U»



                                                          §11
                                                                S - o .2 o
                                                                0 S o ** ~K
                                                                •& at C <0 -°
                                                                            S


                                                                            3
                                                                            C
                                                                            o
                                                                            S "a


                                                                            il
                                                                            E (D
                                                                              O
                                                                              U

-------
Iteration #3: Data for 205 Housing Units.

       The third set of performance characteristics analyses was performed on data for all 205
housing units in the Rochester study.  The previous analyses involved data for fewer housing
units as some units did not have recorded data for the key endpoints used in the analyses to
compare to the various candidate standards. Therefore, this analysis replaced incidences of
missing data with data values that were imputed from information available from other study
units. It was assumed that these imputed values were accurate estimates of what would have
been reported for these units. This estimate for a housing unit could vary considerably from what
would have been reported, however, based on actual  conditions and behaviors in the household.

       As all 205 housing units had reported values for child's blood-lead concentration and for
the percentage of tested ulterior components containing deteriorated lead-based paint, no
imputation was necessary for these two endpoints. The other four endpoints had at least one
housing unit with missing data.  For each of these four endpoints, Table 6-5 contains the number
of housing units with missing data according to year-built category and whether or not the unit
contains lead-based paint, along with the imputed data value assigned to these units, which
equaled the average value across all units in that same category that had non-missing data. The
imputed data values depended on the  year-built category and lead-based paint indicator as these
two variables are typically important predictors of these values. This same approach was used in
the §403 risk analysis to impute environmental-lead data values for HUD National Survey units
having missing data (see Section 3.3.1.1 of USEPA,  1998).

       The data imputation process documented in Table 6-5 resulted in assigning imputed data
to 21 units:  19 built prior to 1940, one built from 1940-1959, and one built after 1979. A total of
eight average uncarpeted floor dust-lead loadings, nine average window sill dust-lead loadings,
six average soil-lead concentrations, and one percentage of deteriorated lead-based paint on
exterior surfaces were imputed.

       Table 6-6 presents estimates of the  four performance characteristics for the 41
combinations of standards, using reported and imputed data for 205 housing units in the
Rochester study. These estimates are very similar to those in Table 6-4 that were calculated from
data for 184 housing units. The same conclusions can be drawn from these results as were made
from the results in Tables 6-3 and 6-4. This implies that at the given combinations of candidate
standards considered in these analyses, the methods used in this section to estimate performance
characteristics were relatively robust across the different sets of data used in the analyses (i.e.,
177,184, or 205 units).

       As sensitivity and negative predictive value are the two performance characteristics of
most interest to Agency reviewers, the results for these two characteristics from Tables 6-3,6-4,
and 6-6 are summarized in Table 6-7.  This summary emphasizes the relative stability of the
estimates across the different approaches used to make the calculations.
                                           247

-------
Table 6-5.     Numbers of Housing Units with Missing Data for Four Endpoints and the
               Imputed Data Values Assigned to These Units in This Analysis
Year-Built
Category
Pre-1940
1940-1959
1960-1979
Post-1979
Lead-
, Based
Paint
Present?
-" "* K & \*\ "
Yes
No
Yes
No
-
Yes
No
Area-Weighted
Average Uncarpeted
- Ftoor Dust-Lead
Loading •
# Units
with
Missing
Data
6
1
0
O
0
1
0
Imputed
Value • >
tow/ft2),1;
160.2
(157)
13.3
(8)
-
-
-
91.3
(3)
-
Area-Weighted
" -Average Window
; Sill Dust-Lead
: Loading ,
# Units
with
'Missing
'Data
5
3
1
0
0
0
0
Imputed
Value
(re/ft2)?
633.2
(158)
95.2
(6)
569.0
(12)
-
-
—
-
Average Soil-Lead
Concentration
# Units
with
Missing
Date ,
5
1
0
0
0
0
0
Imputed
Value
(migf
1258
(158)
631.7
(8)
—
-
-
-
-
% of Exterior
. ' Components
. ' Containing s-
Deteriorated Lead-
Based-Paint
# Units
with
, tiKmtSm*
Missing
Date
1
0
0
0
0
0
0
Imputed" <
Value'
{%)?
25.2%
(162)
—
—
-
-
—
-
1 Number in parentheses equals the number of values (i.e., housing units) entering into calculation of the imputed value.
which is the average of these values.

                                             248


-------
t»
I
                           tn
                                  '"£
Sol^«:
c «- S 53,
   S  o
•81$
            CO
                            a

                          "I
                          .«
         'JI
                                                      a>
                                                           00
                                                           CO
                            to
                            CD
                            CO
                            CM
                                    CO

                                    CD
                                    CO
                                    CN
         CO

         CD
         CO
         CM
     CO
     id
     CO
     CN
                                               00
                                                                                  CN
                                                    CO
                                                                                       CN
IO
a>
to
CM
in
CM
CO
CM
CD


CM
CN
                                                                      CO
                                                                                                         CO
                                                                                                         CN
CO

CO
CO
CN
CO

CO
CO
CN
CO

CO
CO
CM
                   CO

                   CO
                   00
                                                      IO
                                                      (ft
                           £
                           00
                           CM
                                                           CM
CM
CD
CD

CD
S!
IO
CM
                                    CN

                                    CD
                                    CO
                                                                    to
                                                                    CN
CM

CD
CO

CO
CM

io
CM
                                                                             IO
                                                                             CN
                                                                             CM
                                               CN
                                               CN
                                                    CN
                                                    CN
                                                        CN
                                                        CN
     CN


     CM
                                                                  CD

                                                                  CO
                                                                                                         CO
                                                                                                         00
                                                                                                         CD
                                                                                                              to
                                                                                                              O>

                                                                                                              CM
                                                                                                                   IO
                                                                               CD

                                                                               CM
                                                                                                                       IO
                       S

                       CM
                                                                                       CO
                                                                                       CO
                                                      00
                                                               CO
                                                                    CO
                                                                         CO
                                                                                  to
                                                                                      to
                                                                                                CO
                                                      10
                                                      tv
                                                      a>
                                                      w
                        CD
                        CM
                                                               CN

                                                               CD
                                                                    CD
                                                                    CO
                                      CM

                                      CD
                                              CO
                                              CM
                                                                             00


                                                                             ^
                  CD
                  CM
                                                                                  00
              CO
              CN
CD
CN
                                                                                                CM
                                                                                                CO
                                                                                                CO
                                                                                                00
                                                                     M

                                                                     CO
                                                                     CD

                                                                     i
                                          N

                                          CO
                                          IO
                                                                                                         CO
                                     N

                                     CO
                                     CD
                        2

                        CO
                        CO
                        CN
                        CO
                        CO

                        1
                                                      00
                                                      CN
                                                      to
                                                           CM
                                                           CM
                                                               CO
                                                               1ft
                                                                CM
                                 CD
                                 id
                                                                    CN
                                                                         en
                                                                         to
                                                                         CN
                                              CO
                                              id
                                                                             CM
                  CO
                  id
              CO
              id
                                                   ct
                                                                                           CO
                                                                                           CM
     CD
                                                                                                CM
          oq
          CO
                                                                                                     
                                                                    CO
                                                                    CO
                                                                         CO
                                                                         CO
                                                                             00
                                                                             00
                                                                             CO
                                                  CO

                                                  CO
                                                  CO
                                                                                  CO
                       CO

                       00
                       00
                                                                                      00
                  CO

                  CO
                  CO
                       co

                       CO
                       en
          CD

          CO
          CO
                                                                                                         CO
                                                                                                         co
                                                                                                         to
                   to
                   CD
                                                                           CO
                                                                           CD
                   IO

                   CO
                                                                                   co
                                                                                   CD
                   IO

                   CO
                                                        CO
                                                        CD
                                    o
                                    o
                                    CN
                                                                         O
                                                                         O
                                                                             to
                                                   IO
                                                   CM
                                                                                           CM
                                                                                                                  CN
                                                               IA
                                                               CM
                                 to
                                 CM
                                         to
                                         CN
              g
              CN
                                                                                  CN
              O
              to
              CN
to
CM
IO
CM
o
10
CM
10
CM
10
CN
to
CN

                                                                                                         o
                                                                                                         o
                                                                                                         CM
                                                                                                              CM
                                                                                                                  CN
                                                                                    O
                                                                                    O
                                                                                    CN
                                                      CM
                                                           CM
                                                               CM
                                                                    CM
                                                                        CM
                                                                             CM
                                                                                 CM
                                                                                      CM
                                                                                           CM
                                                                                                CM
                                                                                                         CN
                                                                                                             CN
                                                                                                                  CM
                                                                                                                       CN
                                                        249

-------
                                        CM
                                                     in
                                                     CM
                                                         00
                                                         CM
                     in
                     en
                     CM
                     CM
         CD
         O
         co
         CM
                                                                      00
                                                                      CM
                                                                 00
                                                                 CM
                                                                 CM
                                                               00
                                                               CM
                                                               CM
                                      CM
                                     '
                                                                 CO
                                                                 CO
                                                                 CM
                                          CO
                                          CO
                                          CM
                                                                    CO
                                                                    It
                                                                    CM
                                                                                                    CO
                                                                                                    CM
                                  in
                                  en
                                  CM
                                  CM
2*

S>S
                                        a>
en
oo
co
CD
co
I
00
§
CO
                                                     en
                                                     CO

                                                     CO
                                                     CO
                              #
                              8
                              ^w*
                              in

                              in
                     *i*~
                     8

                     go
                     oo
         co
         oo
         o
         in
oo
92

co

CO
CO
                                                                          CO
                                                                          CO
                                                                          CO
                                                                 00
                                                                 CO
                                                                      00
                                                                      CO
                                                                CO
                                                                CO
                                                                    CM
                                                                   s
                                                                                       CO
                                                                                       O)
                                                                                       CO
              CO
              CM
              o>

              en
                                                                                           CO
                                                                                                    8
                                                           O
                                                           O
                                                                                               00
                                                                                               CO
                                                                                               CO
.   ill
£• >
               til
                                        CN
                                        in
                                        (0
                                        tn
                                            in
                                            CM
                                            CO
                                            CD
                                                 in
                                                 CM
                                                 co
                                                 i
                                                     in
                                                     CM
                                          CO
                                          in
                                          CM
                                                         5
                                                             CM
                                                             en
                                                   CN_

                                                   in
                                                   in
                                                                  CM
                          in
                          co
                          CM_

                          CM
                                                                      CO
                                                    in
                                                    co
                                                    N

                                                    CM
                                                    CO
                                                                          CO
                              in
                              CD
                              CM_

                              CM
                                                            2
                                                            CO
                                                            CO
                                                                                       u>
                                                   CM

                                                   CO
                                                   CD
                                                                                            n
                                                                    in
                                                                    ix1
                                                                    CM
                                                            CO
                                                            in
                                                            CM
                                                                                                        CM
                                                                                               |x
                                                                                               CD
                                                                                                        CO
                   n*rl
                   ** ia- ** . «
               CO
                                        CO
                                        CO
                                        CO
                                            CM
                                            CO
                                            CO
en
CN
CN

CO
CO
                                                     N
                                            CO
                                            CO
             CD
             en

             in
                                                             £
                                                             CO
                                                                  CO
                                                                  CM
                                                       CM

                                                       CM
                                                       CO
                                                       CO
                                               CM

                                               a
                                               00
                                               CO
                                                                               CM
                                  00
                                  CO
                                                                         :co
                                                                         !**'
                                               CO
                                               CM
                                               CM

                                               CO
                                               CO
                                  en
                                  CM
                                  CN

                                  CO
                                  CO
                                                                         q

                                                                         CM
                                                                         CO
                                                                         CO
                              CO
                              en
                                                                                           in
                                                                                         in

                                                                                         00
I

  *
CO
CD
 O
JO
 a
I
      to
                               CO

                               CO
                               in
                               •*
oq
in
en
co
«*
                           CO
                           in
                           en

                           CD
                           t
    en

    £
                              o
                              o
                                                CO
                     o
                     o
                                                    00
                                                         in
                                                   00

                                                   CM
             CO

             I
             CO
    CD
    en
    eo

    co
.co
 en
 oo
 co
co
co"
en
in
^t
oq
co
en
in
*
en
fx
o»
|x
«*
                                                                                           00
8
                                                                                               00
5-5 I 1
< 5 21
 ill1"

in

i
in
co
                                            ID
                                            00
                                                 00
                                                 co
                                                     CO
                                                     CO
CN
0»

8
                                                    CO
                                                    en
                                                             o>
                                                                  CO
                                                                  in
                                                                  in
                                                                  in
                                                             q
                                                             CD
                                                             CM
                                                             
-------
                             .,-5  S  i -.
(D

CO
CM

CO
CM
CM
00

O
CM
CM
     00

     6
     CM
     CM
     CO
     CD
     CM
     CM
                                                                 CM
                                                                 CM
                                                                 CM
          
in
oo
in
CM
                                                       in
oq
in
CM_

co
in
                                                                 CO
                                                                 CM
                                                                 CD
                                                                 CM
                        CO
                        CM
                                                                      CO
                                                                      CO
                                                                      CO
*
co
CM

CO
CO
                                                                           CO
00

CD
W

00
CO
                                                                                in
                                  in
                                  CM
                                       en
                                       00
                                                                                     00
                                                                                         CM
                                                                                         co
                                                  CO
                                                  en
                                                  CO
                                                       09
                                                       *
                                                       CM
          0)
          CO
                                                            00
          M

          CO
          CO
                                                                 CM
              2

              09
              CO
                        CO

                        CO
                        CM
                                                                      CO
                        CO
                        CM
                                                                           CO
                                                                                CM
                                                                                CO
                                                                                     CM
                                                                                     CO
                                  #

                                  in

                                  co
                                                  to
                                                       in
                                                       CO
               If)
               00
                                                                 in
               00

               CM
                   co
                   o>
                   00

                   CO
                    CO

                    CO
                    CO

                    CO
                   00
                   CO
                   o>

                   10
                                       00
               s*^
               I

               5
                                                  CO
                                                  o>
                                                  IO
                                                       (O
          0>
          10
                                                            <
                                                            O)
                                                                 iq
                                                                 CO
                                                                 co
                        IO
                        o>
                   CO
                   co
                                                                           CO
                                                                           CO
                                  CM
                                  CM
                                  O)

                                  S
                                                                                         CO
                                                                                         CO
                                                                                         CO
                                                       8
                                                       CM
              O
              o
                   IO
                        !O
                        CM
                                            in
                                                  m
                                                  CM
          IO
          CM
          in
          CM
              10
              CM
              ID
              CM
              to
              CM
                                                                               CS
                                                                                    CM
              in
              CM
                                             tn
                                                  in
                                                       O
                                                       O
                                                       O
                                                       10
                   §
                                                                 IO
                                                                      in
                        o
                        o
                                                                          in
                                                                                                   o o
                                             CM
                                                  CM
                                                       CM
                                                            CM
                                                                 CM
                                                                      CM
                                                                          CM
                                                                               CM
                                                                                    CM
                                                                                         CM
                                                               251
                                                                                                                    II

-------
Table 6-7.    Estimates of Sensitivity and Negative Predictive Value Presented in Tables
             6-3, 6-4, and 6-6
_. *•
.Sot off St&nddrds
%of -
Interior
Paint tlUlt
b :
D8fIt8Q9CI
LBP ",
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2
2':---
2
, %of
Exterior
Pasit tfist
- IS
Damaged
IBP
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
Soil-Lead;
* Cone*
(ppm)
-
400
400
400
400
400
400
400
400
400
400
1200
1200
1200
1200
1200
1200
1200
1200
1200
1200
2000
2000
2000
2000
2000
2000
Window
Sill Dust-
* Lead
Loading
frig/ft*}
-
-
250
250
250
250
250
250
250
250
250
-
250
250
250
250
250
250
250
250
250
-
250
250
250
250
250
Floor
Dust-Lead
Loading
fog/ft*).
-
-
-
200
100
50
40
25
20
10
5
»
-
200
100
50
40
25
20
10
5
-
--
200
100
50
40
SENSITIVITY*
(% of Housing Units with EBL
Children That Are At or Above
" At Least-One Standard}
Data for
177 units
(Table
6-3} x-
83.7%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
90.7%
93.0%
93.0%
93.0%
95.3%
95.3%
95.3%
97.7%
100%
100%
88.4%
90.7%
90.7%
90.7%
93.0%
93.0%
Data for
184 units
jjtaMe
:-6-4}-
81.8%
97.7%
97.7%
97.7%
97.7%
97.7%
100%
100%
100%
100%
100%
88.6%
90.9%
90.9%
90.9%
93.2%
95.5%
95.5%
97.7%
100%
100%
86.4%
88.6%
88.6%
88.6%
90.9%
93.2%
Data, for
205 units
OTabte
64)
81.3%
97.9%
97.9%
97.9%
97.9%
97.9%
100%
100%
100%
100%
100%
89.6%
91.7%
91.7%
91.7%
93.8%
95.8%
95.8%
97.9%
100%
100%
87.5%
89.6%
89.6%
89.6%
91.7%
93.8%
NEGATIVE PREDICTIVE VALUE
~ (% of Housing Units At or
Above No Standards That Do
" " Not Have EBL Children}
Data for
177 units
. (Table
/ 6-3f ,
84.4%
100%
100%
100%
100%
100%
100%
100%
100%
100%
100%
90.0%
91.7%
91.7%
91.7%
94.3%
94.1%
94.1%
96.7%
100%
100%
88.1%
89.2%
89.2%
89.2%
91.7%
91.4%
Data for
184 units
fTabte
- *4)f.
83.3%
96.2%
95.8%
95.8%
95.8%
95.8%
100%
100%
100%
100%
100%
88.4%
89.7%
89.7%
89.7%
92.1%
94.4%
94.4%
96.9%
100%
100%
86.7%
87.5%
87.5%
87.5%
89.7%
91.9%
Data for
205 units
(Table
6-6):
83.3%
96.4%
96.2%
96.2%
96.2%
96.0%
100%
100%
100%
100%
100%
89.1%
90.5%
90.5%
90.5%
92.5%
94.7%
94.7%
97.1%
100%
100%
88.0%
88.4%
88.4%
88.4%
90.2%
92.3%
                                         252

-------
                                    Table 6-7.  (cont.)
Set of Standards
. %ol
Ifitenor
Point thflt
is
Damaged
LBP
2
2
2
2
2
2
2
2
2
2
2
2
2
2
%of.
Exterior
Paint that
is .
Damaged
LBP
10
10
10
10
10
1O
10
10
10
10
10
10
10
10
Soil-Lead
Cone.
Ippm)
2000
2000
2000
2000
5000
5000
5000
5000
5000
5000
5000
5000
5000
5000
Window
SiD Dust-
Lead
Losding
digits
250
250
250
250
-
250
250
250
250
250
250
250
250
250
-• Floor
Dust-Lead
Loading
(PS/ft1)
25
20
10
5
-
-
200
100
50
40
25
20
10
5
SENSITIVITY
(% of Housing Units with EBL
Chadren That Are At or Above
At Least One Standard)
Data for
177 units
(Tabte
6-3)
93.0%
97.7%
100%
100%
86.0%
88.4%
88.4%
88.4%
90.7%
90.7%
90.7%
95.3%
100%
100%
Data for:
184 units
(Table
&4)
93.2%
97.7%
100%
100%
84.1%
86.4%
86.4%
86.4%
88.6%
90.9%
90.9%
95.5%
100%
100%
Data for
205 units
(Table
fr6)
93.8%
97.9%
100%
100%
83.3%
85.4%
85.4%
85.4%
87.5%
89.6%
89.6%
93.8%
100%
100%
NEC ATTVE PREDICTIVE VALUE
{% of Housing Unto At or -
-Above No Standards That Do .
Not Have EBL ChBdren)
Data for
177 units
CTabte
6-3J_
91.4%
96.7%
100%
100%
86.4%
87.2%
87.2%
87.2%
89.5%
89.2%
89.2%
93.8%
100%
100%
*
*Wx-v f *
-Data for
184 units
fTaUe
.-*JWI
91.9%
96.9%
100%
100%
85.1%
85.7%
85.7%
85.7%
87.8%
89.7%
89.7%
94.1%
100%
100%
Data .for
205 units:
(Tabte
6-6}
92.3%
97.1%
100%
100%
84.9%
84.8%
84.8%
84.8%
86.4%
88.1%
88.1%
91.9%
100%
100%
       6.1.4.2 Considering only Soil and Dust Standards. The analysis in the previous
subsection emphasized the difficulty in evaluating candidate paint standards using the Rochester
data, not only due to the fact that the Rochester study did not measure total area corresponding to
deteriorated lead-based paint, but also that most of the housing units with deteriorated lead-based
paint exceeded the candidate standards that were considered in that analysis. In the analysis
presented in this subsection, a paint standard was not considered. Instead, the performance
characteristics analysis considered only candidate standards for soil-lead, floor dust-lead, and
window sill dust-lead, and then investigated the percentage of painted surfaces that contained
deteriorated lead-based paint for those houses that did not exceed any of these three candidate
standards, in an effort to characterize the extent to which these houses would possibly exceed a
paint standard.  The candidate standards for dust and soil in this analysis were the same as in the
previous subsection:

       •       uncarpeted floor dust-lead loading: 5,10,20,25,40,50,100,200 ng/ft2
                                           253

-------
       •      window sill dust-lead loading: 250
       •      vardwide average soil-lead concentration: 400,1200,2000,5000 ug/g

The following 57 combinations of candidate standards were considered in this analysis:

       *      8x1x4=32 combinations of the candidate floor-dust, sill-dust, and soil standards
       •      4x1=4 combinations of only the candidate soil and sill-dust standards
       •      1x8=8 combinations of only the candidate floor-dust and sill-dust standards
       •      4 candidate soil standards without the others
       *      1 sill-lead standard without the others
       *      8 candidate floor-lead standards without the others.

The analysis was applied to data for housing units in the Rochester study having data that could
be compared to each of the standards included in the given combinaton. Average soil-lead
concentration for housing units equaled the average of the dripline and play area soil-lead
measures. Units having either dripline soil-lead data or play area soil-lead data, but not both, had
an average soil-lead concentration equal to the reported concentration at the area represented by
the available data. An average soil-lead concentration of 0 ppm was assigned to housing units
having no soil-lead data and no bare soil from which to sample.

       Table 6-8 contains the results of the performance characteristics analysis, with each row
of the table corresponding to one of the 57 combinations of candidate standards being
considered. The following are examples of how to interpret the findings within Table 6-8:

       •      Consider combinations of all three standards where the candidate soil-lead

              uncarpeted floor-dust standard of 50 (Jg/ft2, only one of the 44 homes containing
              children with elevated blood-lead concentration did not exceed any of these three
              standards and did not contain any deteriorated lead-based paint. (Two other
              homes with an elevated blood-lead child also do not exceed these dust or soil
              standards, but they do contain  some deteriorated lead-based paint) Therefore,
              under these standards, this particular unit would not be triggered for intervention,
              regardless of the paint standard, despite the unit containing a child with an
              elevated blood-lead concentration. However, if the uncarpeted floor-dust standard
              was lowered to 40 ug/ft2, the house would exceed this lower floor standard.

       •      Consider the combination involving only aJEloor dust-lead standard of 20 us/ft2
              and a window sill dust-lead standard of 250 us/ft2.  A total of 106 of the 188
              homes met or exceeded at least one of these two standards, including 36 of the 45
              homes with elevated blood-lead children. Of the 82 homes that did not meet or
              exceed either dust standard, 9 contained an elevated blood-lead child, of which 2
              had no deteriorated lead-based paint in either the interior or exterior. This means
              that if only dust and paint standards were considered, these two homes would not
              be triggered for any intervention, despite containing elevated blood-lead children.

                                           254

-------
•7
 e
i
€
 O
GC
 O
•ft
=s
»0
•i OT
w —
3 U
o i
z •
S «
*5
c »
51

ii
&«
« £
c S
< 2
   S-o
   c
j«
O (A
JH ^3
_g W
O -O
O  k
t  o
 O 3
^v5

 i »
 O 3
EC Q


GO
 o
2
 o
5
o

Al
       o
       o
       •o
        O
        o
elevated
       m
o>


q

A)
        ID
        Q.

       •o
        

     "3
     Q.
     .23

     •s
     CD
     O.
            •o
            03
             O
IT)
A  O
cn  (0
J  £:


.11
Q}  {D
O  £
CD  *-
CO O
   fc-
•o ^~
Q} OO

».E
*- ro
CO CD
C °-
° S

c .i
« 5
CD O)

— 'C


** .—

.2  «

*? •-
w it
Q RJ
S  ^
                    o^i
                          i
                             -
               S; j j   i 2
                       w

                   *TT " tO •«  **
                   •&iSil«
                                Hi HI
                                ;--•
                   a-.
                   »!•
                                   CA •
                                 n
                                            n
                                            n
                                            CM
                                           co 3
                                           •^ JK
                                           ^ CM
                                           5
                                           5 uj
                                                m
                                                o>
                                                3*
                                                C 10
                                                CO rsi
                                                *~ 00
                                                SO
                                                00
                                                     CO
                                              CD #
                                              CM Z
                                              0(6
                                              "
                                                     LO aj
                                                     5m
                                                     0 ^
                                                          10
                                                          CM,
                                                          00
                                                  CO
                                                  CM
                                                          CO
                                                              CM
                                                              CM
                                                       CM #
                                                              0 CO
                                                                   O)
                                                                   CO
                                                                   CM
                                                                   5#
                                                                   CO (0
                                                                        en
                                                                        
-------
5l*«-
Ji-f •'

51*1
  f<
*i:
                          CO
                              «D
                                   M
                                   CD
                                           CM
                                                   CM
                                                       CM
                                                       CM
                                                           CO
                                                               CO
                                                               CM
                                                               CO
                                                               CM
                                                                   CO
                 .8
                 A
                      CM
                          CM
                                              CM
                                                          CN
                                                               CM
                                                       CO
SJ/Sf:


!« w
                                  CD £
                                  a»
                                  2 w
                     2*
                     |q
                     2 «
                     s
                                      Us
                                 00 *

                                 E*
                                 CO CO
                                 o> s
                                                        «- e>
        m CO
        0> 00
                                   S*
                                   5 d>
                              
•* co
CO go
                              ^ IS
                              r-. en
                              Is
                                       IS
                                        CD 3
                                        sS
                                        S oo
                                        GO >£
                             is
                             is
                                              ? 10
                                              95 CD
                                              o> £
                                                                     00
                                                                   co £
     o


     >,

     :5!
I'll
                                i
    CO
    0)


    i
    IO
                              u>
                              CM
                              h. #

                              to
                              *"" CD
                              CO ^t
                                   (O
                            5s
                                          §
                                           CO
                              8
      I-.
                                                   CO
                                     10 jj?

                                     5 co
                                     10 10
                                     CM 5
                                                               CO

                                                               g
                                     IO
                                     CM
                                                                   00
                                                              s
                                   256

-------
    K£ i
    O Z|p«
  3 * *•'
  3*f

  S2£
                 8
                              n
                              N
                                      CM
                                      ID
                                                  CM
                                                      CM
                                                          CM
                 A.
§fii


Is 2?

ifi*
=> S J
           t
                          CO
                                  CM
                                      CM
                          at
                                  CD
  S|<£o>

 :|lH!
               IIP
               ' ! 
                                 r*» oo
                               co
                ^ to

                CM pj

                *~ CO
                                                        *
                                           CO
                             £ •*
                             CO o>
   >l  ° * 5 5
   II  2«a*
                                      V O
                                   n
                                     o 3
                                     10 *^
                                     r- CO

                                     ^ cn
                                                   to
                                                      r- UJ

                                                      ^ CO
                                               •- ^
                                               ^ CO
                                               •* £1
                                                                   CM W
                          o> S
CO ^




CO IO

o> £
                                          co
•^ ^
 SCO
 to

                                     rv ID
                                     W «
n *



CN oo
                                             P: co
                                             CO £l
                                                            eo
                                                                CM
                                                        co to
                                                        CO CM
If.I

%|a-ii

«1  *'

51  l;
                       CO
                           o
                           <5
      Si-.

                         CO
                         oo
                         S
                              §
                         §
    §*


    CO ^





    S
    r™

    CO
    0)
                                         10
                                                  S*
                                                  5 «
                                             S
                                                 S
                                                 8
                                               g
                                                          8
                                                          CM 10
                                                          * S

                                                         8
                                                              8
                                  257

-------
                     •


                                                CM
                    <*.
                     s
                     A
                                       CM
                                                 CM
                                                CM
                                                     (O
                                                                     CO
                                                                          CM
                                                          w
CO


                       CM 0)
                                  2*
                                  9J CM
                                  CM CO
                                       en
                                             «0
2 *
g>  oo
                                                 £ <°~
                                                 CO CO
          Ii
                                                                               10
                                                                                    "
                                                       5 jo
                                                       CO £»
                     ss
SD
                                     Ii

                                S
                                8
                                  « S
                                  ? (p
                                  CM 00
                                  co u>
- IO

CM 00
CO S
I?
g~~ p»
  IO
                                                            !*
                                                            ^ 0
                                                                            (0
                                                                    56
                                                                    S JM
                         •i  a
                                           35
                                                     CO 10
                                                 s*
                                                 i ">
                                                 to en
                                                 co ™
                                                      CO CO
                                                       oo co
                                                       co 5
                                                                            *
                     S co
                     CO CO
                     5 r-
                     2 "^
                     •* o>
                                S
'<*   «
                                     01
                                     CO
                                       co


                                       en
                                                          CO
          S
          ^-

          8
           00

           ri"
           o
                                                                          CO
                                                                               00
                                                                               IO
                                                                           S
1*-.
0« .I
*1-
                                                          o
                                                          IO
                                                            to
                                                            CM
                       ff
                                o
                                u>
                                (N
                                  s
                                            s
                                            CM
           o
           to
           CM
           O
           U)
           CM
                                                               s
                                                               CM

                                     O
                                     §
                                  §
                                           258
                                                  O
                                                  O
                                                  
-------
11  I :
s«l'*V
Is 21 li
sa*l5
     53
  i'l.
  i £
     1<
     5s
                 *•
                 .r»
                 A
                          
     •*: JO
                                       CO
                                                            (O
                                                                (O
o*

15'
W "I
         "
                                  co
                                  CD
                                       us
                                   SK
                                   CO
                                               ss

r- CO
a> jo
» Sp_
S?
? co
en co
CM*
0> ffl
   II I|l|(
                                CO
                                           S t
                                           !S t>
                                           m n
                                          - 9
                                          (^ CO
                                          co n
                                              U) ^
                                              »— CO
        S
           is!
              W
                     00 CM
                     §£
                              00 CM
                              co £
                              ȣ
            C "
            co *
            f- !?.

                                                         o>
                                                       §#•
                                                       ? en
                                                               ?s
                                                               CN IO
                                                               a £
       0^  gS  <
      £-»^£* s •'
                              r* ^
                                  0) in
                                                         S

                                               CO 50
                                                           10 co
                                                           CM to
                                                                    r- T^
Is

S« =
                          3
                              CD
                                  S
                         oS

                         S

                                                                    n
      15
      15-
              ii
             JLL
             \H
             !Q §.
                          8
                         8
                                       §
                                          8
                                                   8
                                      S
                                   259

-------
I


CO
(6
 ii
il
*i«i-
3
     "S51S.1

     *||3=|
         **  3

     — * _J 5 «S


     ? I III
        r*'w
il-.
           ^ ** Tt*

                   CO
                   g"~ 
                      2 10
                      CO ;>
                      3
                      0>
                      to
                      CM
                      O

                      8
                      IO
                          CM
                          IO
                          CM
                         1C

                         N
                              ID 3

                              c*
                              10 ri
                              « *
                          ^ to
                             m
                             CO
                              CM
                              CD
                             10 u>
                             CO d
84
48
                             o
                             IO
                             CM
                             8
                             S
72/184
                          260
                                     Il


                                     i!
                                     c E
                                     .S 3
                                     Q. »
                                     1!
                                     la
                                     il

                                     !!
                                     |l
                                     § S
                                     S*
                                     1?

                                     *s
                                     II

                                     p
                                     Is
                                     3-5
                                         5
                                              .
                                          !l
                                           s
                                   S-JB
                                   -o £

                                   °!


                                   IB

                                   II
                                   §•§

                                   !5
                                         *  f  8

                                         !  It
                                         i  s  s.
                                            \\
                                         I    I
                                              I I  I
                                              9 s  1

                                              I 1  I

                                              I -  I


                                              III


                                              HI
I |  f


« g  rf
** -fc.  GO
« s  2

I -I  I
So?
o <*>  -g

^ 1  S

     €
                                                   « 1
                                                     . *  I
                                                     111

                                                     ! !  !

                                                     ! I  ~
                                                     I ;>  5
                                                     £ S  §
                                                     > |  .£
                                                     2 i  £
                                                     •215

                                                     f  I  *
                                                 o  *

                                                 S  5
                                                 S  %


                                                 !  II
                                                     o
                                                     c
                                                S f

                                                S 2


                                                I 1
I I

I i
                                                i
I I
I §
                                      ii <11  i    ! f
                                              III  i
                                              *i?  I
                                                     ^
                                                E  *  g

                                                I  I  I

                                                I  I  !
                                      If «
                                      i! Ill
                                          5 g  I -e
                                          £ £ . > S

                                          i nil
                                          S o>0 > £•
                                        2  &
                                        ^ o £
                                                        .  g


                                                          i
                                                 S  • 2
                                                g
                                                1
                                               _
                                              o "5 o  o •

                                                s  s
                                      s
                                            -
                                          af
                                                  i .1 7 .8 5 .1 ^ If
                                                  I te CD S »= i OK)

                                                  Sf g|if|


                                                  i« aS *S S5

-------
       6.1.4.3 Analysis Involving Only Dust-Lead Standards and a Standard on the
Amount of Deteriorated Paint.  In some cases, a risk assessment may involve only dust
sampling (of floors and window sills) and a visual inspection of painted surfaces for
deterioration. That is, no testing of painted surface for lead within the paint would be done, and
no soil sampling would be done. In this setting, it was of interest to investigate the extent to
which candidate dust-lead loading standards, with standards on the maximum percentage of
surfaces with deteriorated paint, performed in the absence of soil standards, within a performance
characteristics analysis.  The combinations of standards considered in this analysis were the
following:

             uncarpeted floor dust-lead loading: 5,10,20,25,40,50,100 ug/ft2
       •     window sill dust-lead loading: 125,250 ug/ft2
       •     maximum amount of deteriorated paint on a tested surface: >5%, >15%.

The candidate paint standards were defined to coincide with the type of paint condition
measurement made in the Rochester study. The following  63 combinations of these candidate
standards  were considered in this analysis:

       •     7x2x2=28 combinations of the candidate floor-dust, sill-dust, and paint standards
       •     7x2= 14 combinations of only the candidate  floor-dust and sill-dust standards
       •     7x2=14 combinations of only the candidate  floor-dust and paint standards
       *     7 candidate floor-lead standards without the others.

       Table 6-9 contains the results of the performance characteristics analysis, with each row
of the table corresponding to one of the 63 combinations of candidate standards being
considered.  The following are examples of what can be concluded from Table 6-9:

       •     While, on their own, the higher candidate floor dust-lead standards trigger few
             units containing elevated blood-lead children, the number of these homes that are
             triggered with the addition of a deteriorated paint standard increases dramatically
             (e.g., from 10.6% to 70.2% at a floor dust-lead standard of 100 ug/ft2, if the  15%
             paint standard is added).

       •     The performance characteristics do not appear to increase substantially with an
             increase in the sill standard from 125 to 250 ug/ft2.

       If the risk assessment does, in fact, do paint testing  for lead, then the above standard for
paint can be re-defined to represent the maximum amount of deteriorated lead-based paint on a
tested surface. Table 6-10 contains the results of the performance characteristics analysis where
the paint standard is modified in this manner.
                                          261

-------
Table 6-9.   Results of Performance Characteristics Analysis Performed on Data for
             Housing Units in the Rochester Lead-in-Dust Study, for Specified Sets of
             Candidate Standards for Dust-Lead Loadings and Observed Amount of
             Damaged Paint on a Tested Surface

EBL - elevated blood-lead level fe10//g/dU
Sot of Csnciki&tO: Stsndvds.
f ^ < *
Uncarpeted
.Boor Dust-
Lad
* Loading
. (pgrft*)
100
50
40
25
20
10
5
100
SO
40
25
20
10
5
100
50
4O
25
20
10
5
100
50
40
25
20
10
Window
Sill Dust-
Lead
Loading r
big/ft*)
-
-
-
-
-
-
-
250
250
250
250
250
250
250
125
125
125
125
125
125
125
-
-
-
-
-
-
.Max.
Aim. Of
Damaged
Paint on a
/Tested
Surface
(%)',
••?'"' " "*•*
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
>15%
>15%
>15%
>15%
>15%
>15%
# Units
. At Or
Above
At Least
-One
Standard
/Total #
Units2 -
9/197
19/197
31/197
58/197
84/197
150/197
180/197
71/189
75/189
80/189
93/189
1O6/189
150/189
176/189
116/189
118/189
122/189
128/189
134/189
159/189
180/189
101/197
105/197
108/197
116/197
127/197
170/197
, Perfonit&nc6 C
*
Sensitivity
# {%} of Units
with EBL
Children That
Are At or'Above
At Least One
rJ^Stwidard* — :
5/47 110.6%)
9/47 (19.1%)
16/47 (34.0%)
26/47 (55.3%)
31/47 (66.0%)
44/47 (93.6%)
45/47 (95.7%)
25/45 (55.6%)
27/45 (60.0%)
30/45 (66.7%)
34/45 (75.6%)
36/45 (80.0%)
44/45 (97.8%)
44/45 (97.8%)
35/45 (77.8%)
36/45 (80.0%)
38/45 (84.4%)
39/45 (86.7%)
40/45 (88.9%)
45/45 (100%)
45/45 (100%)
33/47 (70.2%)
34/47 (72.3%)
35/47 (74.5%)
35/47 (74.5%)
38/47 (80.9%)
46/47 (97.9%)
. ,. Specificity
* ^ -
#(%) of Units
; withNoEBL
.Chadren That Are
• At or- Above No
s Standard*,
» - hi * '•< "
146/150 (97.3%)
140/150(93.3%)
135/150(90.0%)
118/150(78.7%)
97/150(64.7%)
44/1 50 (29.3%)
15/150(10.0%)
98/144(68.1%)
96/144 (66.7%)
94/144 (65.3%)
85/144 (59.0%)
74/144(51.4%)
38/144(26.4%)
12/144(8.3%)
63/144 (43.8%)
62/144(43.1%)
60/144(41.7%)
55/144 (38.2%)
50/144(34.7%)
30/144 (20.8%)
9/144 (6.3%)
82/150(54.7%)
79/150(52.7%)
77/150151.3%)
69/1 50 (46.0%)
61/150(40.7%)
26/150(17.3%)

tat JtCUH ISUCir
£PV
#{%} of Unto
At or Above At
Standard That
Have EBL
Children ^
5/9 (55.6%)
9/19 (47.4%)
16/31 (51.6%)
26/58 (44.8%)
31/84(36.9%)
44/150(29.3%)
45/180 (25.0%)
25/71 (35.2%)
27/75 (36.0%)
30/80 (37.5%)
34/93 (36.6%)
36/106(34.0%)
44/150(29.3%)
44/1 76 (25.0%)
35/116(30.2%)
36/118(30.5%)
38/122(31.1%)
39/128(30.5%)
40/134(29.9%)
45/159(28.3%)
45/180(25.0%)
33/101 (32.7%)
34/105 (32.4%)
35/108 (32.4%)
35/116(30.2%)
38/127(29.9%)
46/170(27.1%)
> V >*
i NPV
# {%) of Units At
or Above No
Standard That Do
Not Have EBL
. ChOdran*^ .
, H *^t s ?a- •
""--^-'
146/188 (77.7%)
140/178 (78.7%)
135/166(81.3%)
118/139(84.9%)
97/113(85.8%)
44/47 (93.6%)
15/17(88.2%)
98/118(83.1%)
96/114(84.2%)
94/109 (86.2%)
85/96 (88.5%)
74/83 (89.2%)
38/39 (97.4%)
12/13(92.3%)
63/73 (86.3%)
62/71 (87.3%)
60/67 (89.6%)
55/61 (90.2%)
50/55 (90.9%)
30/30 (100%)
9/9 (100%)
82/96 (85.4%)
79/92 (85.9%)
77/89 (86.5%)
69/81 (85.2%)
61/70(87.1%)
26/27 (96.3%)
                                        262

-------
Table 6-9. (cont.)
:' Set of Candidate Standards
Uncarpetad
Floor Dust-
Lead
' Loading
(ng/ft2)
5
100
50
40
25
20
10
5
100
50
40
25
20
10
5
100
50
40
25
20
10-
5 •
100
50
40
25
20
10
5
Window
SHI Dust-
Lead ;
1 nwii****
Loaning
<«im*>
-
-
-
-
-
-
-
-
250
250
250
250
250
250
250
125
125
125
125
125
125
125
250
250
250
250
250
250
250
Max!
Amt-Of
Damaged
Paint on a
Tested
Surface
c%r
>15%
>5%
>5%
>5%
>5%
>5%
>5%
>5%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>5%
>5%
>5%
>5%
>5%
>5%
>5%
# Units
-At Or
Above
At least
.' . One
Standard
/Totai#
Unto2
188/197
164/197
165/197
167/197
167/197
168/197
185/197
192/197
118/189
120/189
123/189
127/189
134/189
167/189
182/189
142/189
144/189
147/189
149/189
152/189
171/189
182/189
162/189
163/189
165/189
165/189
166/189
179/189
185/189

Sensitivity
#(%) of Units
whhEBL
Children That
Are At or Above
At Least One
•"•• StAIMadllP
47/47 (100%)
43/47191.5%)
44/47 (93.6%)
45/47 (95.7%)
45/47 (95.7%)
45/47 (95.7%)
47/47 (100%)
47/47 (100%)
36/45 (80.0%)
37/45 (82.2%)
38/45 (84.4%)
38/45 (84.4%)
40/45 (88.9%)
45/45(100%)
45/45(100%)
39/45 (86.7%)
40/45 (88.9%)
41/45(91.1%)
41/45(91.1%)
42/45 (93.3%)
45/45 (100%)
45/45(100%)
41/45(91.1%)
42/45 (93.3%)
43/45 (95.6%)
43/45 (95.6%)
43/45 (95.6%)
45/45 (100%)
45/45 (100%)

^•*uiii«Qii*>v \*nai€i\*tKn»tK
-------
                                                  Table 6-9.  (cent.)
:- Set of Candidate Standards
:UnC0fpeted
Roar Dust-
Lead
Loading
.teg/ft*)
100
50
40
25
20
10
5
Window
SOT Dust-
Lead
Loading
{figlfft
125
125
125
125
125
125
125
Max.,
Aim.OfV
•Damaged
Paint on a
; Tested
Surface
{%>'•'
>5%
>5%
>5%
>5%
>5%
>5%
>5%
9 Units
At Or
Above
; At Least
;;One
Stsnd&rd
/Total*
Units*
166/189
167/189
169/189
169/189
170/189
179/189
185/189
- Performance C
Sensitivity ,
#(%> of Units
wfflhlBUf'V
Ctiildr6nwjnat
Are At or Above
At Least One
r'StflHuoiu „-
42/45 (93.3%)
43/45 (95.6%)
44/45 (97.8%)
44/45 (97.8%)
44/45 (97.8%)
45/45 (100%)
45/45 (100%)
Specificity
#(%) of Units
with No EBL
Chadren That Are
At or Above No
St&ndsfd^
20/144(13.9%)
20/144(13.9%)
19/144(13.2%)
19/144(13.2%)
18/144(12.5%)
10/144(6.9%)
4/144 (2.8%)
». ••mjUiu-Si-rijte
Ml CtblOl H4MA
PPV
#(%) of Units
At or Above At;
•• Least One ' "~
Standard That-
HaveffiL:
Chadren5
42/166(25.3%)
43/167(25.7%)
44/169(26.0%)
44/169(26.0%)
44/170(25.9%)
45/179(25.1%)
45/185(24.3%)

NPV
# (%) of Units At
'- '» or -Above Ma
Standard Tnat Do
_/ Not Have EBL
Children'
20/23 (87.0%)
20/22 (90.9%)
19/20(95.0%)
19/20(95.0%)
18/19(94.7%)
10/10(100%)
4/4 (100%)
1 In the Rochester study, each measurement of lead in paint had the amount of damaged paint specified as * 15%" (poor condition) of the tested surface, with no indication of total damaged surface area.
1 Total number of units having available data that could be compared to all specified candidate standards.
9 Cell entries areinumber of homes at or above at least one standard that have EBL children)/ number of homes containing EBL children),
followed by the corresponding percentage (in parentheses).
4 Cell entries are (number of homes not at or above at least one standard that do not have EBL children)/(total number of homes not
containing EBL children), followed by the corresponding percentage (in parentheses).
* Cell entries are (number of homes at or above at least one standard that have EBL chik)ren)/(total number of homes at or above at least
one standard), followed by the corresponding percentage (in parentheses).
' Cell entries are (number of homes not at or above at least one standard that do not have EBL children)/(total number of homes not at or
above any standard), followed by the corresponding percentage (in parentheses).
                                                            264

-------
Table 6-10.  Results of Performance Characteristics Analysis Performed on Data for
             Housing Units in the Rochester Lead-in-Dust Study, for Specified Sets of
             Candidate Standards for Dust-Lead Loadings and Observed Amount of
             Damaged Lead-Based Paint on a Tested Surface

EBL = elevated blood-lead level (:>10//g/dU; LBP=Lead-Based Paint
Set of Candidate Standards
Uncarpeted
Floor Dust-
Lead
Loading
tt*g/f*l
100
50
40
25
20
10
5
100
50
40
25
20
10
5
100
50
40
25
20
10
5
100
50
40
25
20
10
Window
Sin Dust-
Lead ~
Loading
frig/ft*).
-
-
-
-
-
- •
-
250
250
250
250
250
250
250
125
125
125
125
125
125
125
-
-
-
-
-
-
Max.
Aim. Of
Damaged
LBPona
Tested:
Surface"'
:<%)'
-
-
-
-
-
-
-
-
-
_
-
-
-
• -
-
-
-
-
-
-
-
>15%
>15%
>15%
>15%
>15%
>15%
# Units
At Or.
Above
At Least
One
$t&nd8rd'
/Total*
: Units2
9/197
19/197
31/197
58/197
84/197
150/197
180/197
71/189
75/189
80/189
93/189
106/189
150/189
176/189
116/189
118/189
122/189
128/189
134/189
159/189
180/189
84/197
88/197
94/197
104/197
115/197
162/197
Performance Characteristics
Sensitivity
#<%) of Units
with EBL
CrfldrenThat
Aro At or Above
At Least One
' * Standard^
5/47 (10.6%)
9/47(19.1%)
16/47(34.0%)
26/47 (55.3%)
31/47 (66.0%)
44/47 (93.6%)
45/47 (95.7%)
25/45 (55.6%)
27/45 (60.0%)
30/45 (66.7%)
34/45 (75.6%)
36/45 (80.0%)
44/45 (97.8%)
44/45 (97.8%)
35/45 (77.8%)
36/45 (80.0%)
38/45 (84.4%)
39/45 (86.7%)
40/45 (88.9%)
45/45 (100%)
45/45 (100%)
27/47 (57.4%)
28/47 (59.6%)
31/47(66.0%)
33/47 (70.2%)
36/47 (76.6%)
44/47 (93.6%)
Soeeificitv
# (%) of Units
with No &L
Children That Are
At or Above No
Standard4
146/150(97.3%)
140/150(93.3%)
135/150(90.0%)
118/150(78.7%)
97/150(64.7%)
44/150(29.3%)
15/150(10.0%)
98/144(68.1%)
96/144166.7%)
94/144(65.3%)
85/144(59.0%)
74/144(51.4%)
38/144 (26.4%)
12/144(8.3%)
63/144 (43.8%)
62/144(43.1%)
60/144(41.7%)
55/144(38.2%)
50/144(34.7%)
30/144 (20.8%)
9/144 (6.3%)
93/150(62.0%)
90/150(60.0%)
87/150(58.0%)
79/150(52.7%)
71/150(47.3%)
32/150(21.3%)
PPV
* (%) of Units,,
At or Above At
Least One-
Standard; That ;
Have EBL
Chadren*
5/9 (55.6%)
9/1 9 (47.4%)
16/31 (51.6%)
26/58 (44.8%)
31/84(36.9%)
44/150(29.3%)
45/180(25.0%)
25/71 (35.2%)
27/75 (36.0%)
30/80 (37.5%)
34/93 (36.6%)
36/106 (34.0%)
44/150(29.3%)
44/176(25.0%)
35/116(30.2%)
36/118(30.5%)
38/122(31.1%)
39/128(30.5%)
40/134(29.9%)
45/159 (28.3%)
45/180(25.0%)
27/84(32.1%)
28/88(31.8%)
31/94(33.0%)
33/104(31.7%)
36/115(31.3%)
44/162(27.2%)
NPV
#{%} of Units At
• or Above No
Standard That Do
Not Have EBL
CtiikifSfii *«,
146/188(77.7%)
140/178 (78.7%)
135/166 (81.3%)
118/139(84.9%)
97/113(85.8%)
44/47 (93.6%)
15/17(88.2%)
98/118(83.1%)
96/114(84.2%)
94/109(86.2%)
85/96 (88.5%)
74/83 (89.2%)
38/39 (97.4%)
12/13 (92.3%)
63/73 (86.3%)
62/71 (87.3%)
60/67 (89.6%)
55/61 (90.2%)
50/55 (90.9%)
30/30(100%)
9/9 (100%)
93/113(82.3%)
90/109 (82.6%)
87/103 (84.5%)
79/93 (84,9%)
71/82(86.6%)
32/35(91.4%)
                                         265

-------
Table 6-10.  (cont.)
Set of Candidate Standards _ -
Uncarpeted
Floor Dust-
Lead
Loading
-torn1)
5
100
50
40
25
20
10
5
100
50
40
25
20
10
5
100
50
40
25
20
10
5
100
50
40
25
20
10
5
Window
SiD Dust-
Lead
Lending
(ra/frt
-
-
-
-
-
-
-
-
250
250
250
250
250
250
250
125
125
125
125
125
125
125
250
250
250
250
250
250
250
Max.
Ami. Of :
Damaged
LBPona
Tested.
'Surface
(%)'
>15%
>5%
>5%
>5%
>5%
>S%
>5%
>5%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>15%
>5%
>5%
>5%
>5%
>5%
>5%
>5%
# Units
At Or
Above
At Least
One
.St&ndara
/Total*
Units2
183/197
146/197
147/197
149/197
150/197
155/197
181/197
189/197
107/189
109/189
114/189
119/189
126/189
160/189
178/189
135/189
137/189
141/189
144/189
147/189
167/189
181/189
147/189
148/189
150/189
151/189
156/189
175/189
182/189

Sensitivity
# (%) of Units
withEBL
Children That:
Are At or Above
At Least One
Standard: ~
45/47 (95.7%)
39/47 (83.0%)
40/47 (85.1%)
41/47 (87.2%)
42/47 (89.4%)
44/47 (93.6%)
47/47 (100%)
47/47 (100%)
32/45(71.1%)
33/45 (73.3%)
36/45 (80.0%)
37/45 (82.2%)
39/45 (86.7%)
44/45 (97.8%)
44/45 (97.8%)
37/45 (82.2%)
38/45 (84.4%)
40/45 (88.9%)
41/45(91.1%)
42/45 (93.3%)
45/45 (100%)
45/45 (100%)
38/45 (84.4%)
39/45 (86.7%)
40/45 (88.9%)
41/45(91.1%)
43/45 (95.6%)
45/45 (100%)
45/45 (100%)
. . ^ . ^

Specificity
# (%) of Units
with No SL
Chfldren That Are
At or Above No
Standard?
12/150(8.0%)
43/1 50 (28.7%)
43/1 50 (28.7%)
42/150(28.0%)
42/150(28.0%)
39/150(26.0%)
16/150(10.7%)
8/150(5.3%)
69/144 (47.9%)
68/144(47.2%)
66/144(45.8%)
62/144(43.1%)
57/144(39.6%)
28/144(19.4%)
10/144(6.9%)
46/144(31.9%)
45/144(31.3%)
43/144(29.9%)
41/144 (28.5%)
39/144(27.1%)
22/144(15.3%)
8/144(5.6%)
35/144 (24.3%)
35/144(24.3%)
34/144(23.6%)
34/144(23.6%)
31/144(21.5%)
14/144(9.7%)
7/144(4.9%)
P£V
# (%) of Units
At or Above At
Least One
Standard That
HaveEBL
dhadren5
45/183(24.6%)
39/146 (26.7%)
40/147 (27.2%)
41/149 (27.5%)
42/150(28.0%)
44/155(28.4%)
47/181 (26.0%)
47/189 (24.9%)
32/107.(29.9%)
33/109(30.3%)
36/114(31.6%)
37/119(31.1%)
39/126(31.0%)
44/160(27.5%)
44/178(24.7%)
37/135(27.4%)
38/137(27.7%)
40/141 (28.4%)
41/144 (28.5%)
42/147 (28.6%)
45/167(26.9%)
45/181 (24.9%)
38/147 (25.9%)
39/148(26.4%)
40/150(26.7%)
41/151 (27.2%)
43/156(27.6%)
45/175(25.7%)
45/182(24.7%)
• <•
NPV
# (%) of Units At
or Above No
Standard That Do
Not HaveEBL
Children*
12/14(85.7%)
43/51 (84.3%)
43/50 (86.0%)
42/48 (87.5%)
42/47 (89.4%)
39/42 (92.9%)
16/16(100%)
8/8 (100%)
69/82(84.1%)
68/80 (85.0%)
66/75 (88.0%)
62/70 (88.6%)
57/63 (90.5%)
28/29 (96.6%)
10/11 (90.9%)
46/54 (85.2%)
45/52 (86.5%)
43/48 (89.6%)
41/45(91.1%)
39/42 (92.9%)
22/22 (100%)
8/8(100%)
35/42 (83.3%)
35/41 (85.4%)
34/39 (87.2%)
34/38 (89.5%)
31/33 (93.9%)
14/14(100%)
7/7(100%)
       266

-------
                                                 Table 6-10.  (cont.)
•••: Set of Candidate Standards
»""*•- f--£il "X.

ROOT Dust-
Lead "
Loading

100
50
4O
25
20
10
5
Window
Sffl Dust-
Lead
Loading
(i*g/ft*>*
12S
125
125
125
125
125
125
Max.:;
Amt. Of
Damaged
LBPona
tested-
Surface
(%)'
>5%
>5%
>5%
>5%
>5%
>5%
>5%
# Units
At Or
Above
At Least
One
Standard
/Total #
Units -
156/189
157/189
159/189
160/189
163/189
176/189
183/189
V " . , * '••<-.. -Performance C
f "* - • -t ' ~
Sensitivity
# (%) of Units
with EBL •'
Children That
Are At or Above:
At Least One
~ Standard^.
40/45 (88.9%)
41/45(91.1%!
42/45 (93.3%)
43/45 (95.6%)
44/45 (97.8%)
45/45 (100%)
45/45 (100%)
Specificity
#(%) of Units
with No EBL
Children That Are
At or Above Mo
Standard* :
28/144(19.4%)
28/144(19.4%)
27/144(18.8%)
27/144(18.8%)
25/144(17.4%)
13/144(9.0%)
6/144 (4.2%)

haracteristics
Bfv
-V.
# (%) of Units
At or Above At
Least One "
Standard That
Have EBL
Children5
40/156(25.6%)
41/157(26.1%)
42/159(26.4%)
43/160(26.9%)
44/163(27.0%)
45/176(25.6%)
45/183(24.6%)
-
- NPV
#(%)ofUnhsAt
or Above No
Standard That Do
Not Have EBL
Children*
28/33 (84.8%)
28/32 (87.5%)
27/30 (90.0%)
27/29(93.1%)
25/26 (96.2%)
13/13(100%)
6/6 (100%)
11n the Rochester study, each measurement of lead in paint had the amount of damaged paint specified as "<5%" (good condition), *5-
15%' (fair condition), or "> 15%' (poor condition) of the tested surface, with no indication of total damaged surface area.
2 Total number of units having available data that could be compared to all specified candidate standards.
* Cell entries arelnumber of homes at or above at least one standard that have EBL children)/ number of homes containing EBL children),
followed by the corresponding percentage (in parentheses).
* Cell entries are (number of homes not at or above at least one standard that do not have EBL children)/(total number of homes not
containing EBL children), followed by the corresponding percentage (in parentheses).
* Cell entries are (number of homes at or above at least one standard that have EBL chtldren)/(total number of homes at or above at least
one standard), followed by the corresponding percentage (in parentheses).
' Call entries are (number of homes not at or above at least one standard that do not have EBL children)/(total number of homes not at or
above any standard), followed by the corresponding percentage (in parentheses).
                                                           267

-------
6.2    INVESTIGATING INCIDENCE OF ELEVATED BLOOD-LEAD
       CONCENTRATION IN HOUSING UNITS MEETING ALL
       EXAMPLE OPTIONS FOR STANDARDS

       An alternative to the performance characteristics analysis approach (Section 6.1) to
evaluating a set of candidate standards is to use statistical modeling techniques to predict a
distribution of blood-lead concentration as a function of environmental-lead levels found in
homes which do not exceed any of the candidate standards, then estimate the percentage of
children residing in these homes that are expected to have elevated blood-lead levels (i.e., at or
above 10 ug/dL). It is desired to select a set of candidate standards so that the likelihood of
children with elevated blood-lead concentration residing in homes that do not exceed any of the
candidate standards would be very low.  This section presents a modeling approach to estimate
this likelihood, using the alternative Rochester multimedia model presented in Section 4.2 of this
report ("Model A" in Table 4-1), and applies this approach to data from the Rochester study.

       Recall from Section 4.2 that the reason for developing the alternative Rochester
multimedia model was to have the risk estimates from model-based analyses be more comparable
to the results of the performance characteristics analysis presented in the §403 proposed rule
(Section 6.1.3) and the results of the follow-up performance characteristics analyses (Section
6.1.4). In particular, both the performance characteristics analysis and the model-based approach
involving the alternative Rochester multimedia model use the following types of data as input
when characterizing risk:

       »      household average (wipe) dust-lead loading from uncarpeted floors
       •      household average (wipe) dust-lead loading from window sills
       *      yard-wide average soil-lead concentration
       *      the larger of the following two percentages: % of interior tested surfaces that
             contain deteriorated lead-based paint (LBP), and % of exterior tested surfaces that
             contain deteriorated LBP

In the model-based analysis approach presented below, the candidate standards were used to
identify a subset of homes in the Rochester study that were below all of the candidate standards,
calculate the average (across homes) of the above three measures of lead levels in dust and soil,
and fit the multimedia model to these average lead levels in order to predict a distribution of
blood-lead concentrations for children residing in these homes. For simplicity, this analysis
assumes that the homes do not contain deteriorated lead-based paint. Because the slope estimate
for the paint variable in the alternative Rochester multimedia model is nearly zero (Table 4-1 of
Section 4.2), making the assumption that no deteriorated lead-based paint exists in these homes
should have a very minor impact on the resulting risk estimates.

6.2.1  The Model-Based Approach

       This model-based approach had the following four steps:
                                         268

-------
       1.      For a given set of candidate standards for floor dust-lead loading, window sill
              dust-lead loading, and soil-lead concentration, identify those homes in the
              Rochester study that exceed none of the candidate standards in this set.

       2.      For each of the following three household measures, calculate the average across
              the homes identified in step #1: the household average floor dust-lead loadings,
              household average window sill dust-lead loading and for yard-wide average soil-
              lead concentration. These three averages are assumed to represent lead levels in
              housing represented by the Rochester study homes in step #1 (i.e., homes not
              exceeding any of the candidate dust and soil standards).

       3.      Use the three averages calculated in step #2 as input to the alternative Rochester
              multimedia model from Section 4.2 (assuming no .deteriorated lead-based paint
              exists in the units).

       4.      Assume that log-transformed blood-lead concentration for children residing in the
              homes identified in step #1 is normally distributed with mean equal to the
              predicted log-transformed blood-lead concentration that is output from the model
              fitting in step #3, and standard deviation equal to ln(1.6). (Recall that this
              assumption on variability was made throughout the §403 risk analysis.)  Using
              normal distribution theory, determine the percentage of children represented by
              this blood-lead distribution that have log-transformed blood-lead concentration or
              above log(10), or equivalently, that have blood-lead concentration at or above 10
              ug/dL.
6.2.2  Examples of Applying the Model-Based Approach

       To illustrate how the approach in Section 6.2.1 is applied to data from the Rochester
study, the following combinations of candidate dust-lead and soil-lead standards are considered:

       •      (uncarpeted) floor dust-lead loading: either 40 or 50 (ig/ft2
       •      window sill dust-lead loading:  250 ug/ft2
       •      yard-wide soil-lead concentration: 400 fig/g.

       When the candidate floor dust-lead loading standard is 40 ug/ft2. then the performance
characteristics analyses documented in Table 6-8 of Section 6.1 (i.e., the row of Table 6-8
corresponding to these three candidate standards) indicates that 39 of the 184 Rochester study
homes having measurements for dust-lead, soil-lead, and deteriorated lead-based paint do not
exceed any of the three candidate standards. Across these 39 homes, the following averages were
calculated from the Rochester study data:
              household average (uncarpeted) floor dust-lead loading: 12.7 ug/ft2
              household average window sill dust-lead loading: 87.0 ug/ft2

                                           269

-------
       *      yard-wide average soil-lead concentration: 125.3 ug/g.

When fitting the alternative Rochester multimedia model to these three averages (assuming no
deteriorated lead-based paint), the model predicts a geometric mean blood-lead concentration of
4.68 ug/dL. If the standard deviation of log-transformed data is assumed to be  1.6 and normal
distribution theory is applied as described above, then the estimated percentage of children with
blood-lead concentration at or above 10 ug/dL in homes that do not exceed any of the candidate
standards is 5.30%.  This matches closely with the estimate of 5.1 %, or 2 of these 39 homes in
the Rochester study dataset, which the performance characteristics analysis (Table 6-8) indicated
contained children with elevated blood-lead concentrations.

       If the candidate floor dust-lead loading standard is increased to 50 ug/ft2. then the number
of Rochester study homes having measurements for dust-lead, soil-lead, and deteriorated lead-
based paint and that do not exceed any of the three candidate standards increases by one home, to
40 total homes. Across these  40 homes, the following averages were calculated from the
Rochester study data:

       *      household average (uncarpeted) floor dust-lead loading: 13.4 ug/ft2
       •      household average window sill dust-lead loading: 85.6 fig/ft2
       «      yard-wide average soil-lead concentration: 122.2 ug/g.

The predicted geometric mean blood-lead concentration under these assumed dust-lead and soil-
lead levels (assuming no deteriorated lead-based paint) is 4.69 ug/dL. and the estimated
percentage of children with blood-lead concentration at or above 10 ug/dL is 5.34%. This is a
very slight increase from the estimate generated under the candidate floor dust-lead loading
standard of 40 ug/ft2. The performance characteristics analysis (Table 6-8) indicated that under
these candidate standards, 7.5% of homes not exceeding  any of the standards (i.e., 3 of these 40
homes in the Rochester study dataset) contained children with elevated blood-lead
concentrations.

       While these examples illustrate the estimation process, they also-show that the number of
homes in the given dataset whose lead levels fall below all specified candidate standards can be
quite small, especially when at least one of the candidate standards is set at the  low end of the
distribution of lead levels (i.e., most homes have data that fall above the candidate standard).
Therefore, as the set of candidate standards becomes more stringent, and as the size of the sample
from which the environmental-lead data originate becomes smaller as a result, the variability
associated with the estimated risk increases. Furthermore, as the set of candidate standards
becomes less stringent (i.e., as the standards increase), the group of homes not exceeding any of
the candidate standards is more likely to remain the same, and as a result, the estimated risk
eventually reaches a plateau.  This occurs in the above examples, as increasing  the candidate
floor dust-lead loading standard from 40 to 50 ug/ft2 does little, if any, to increase the estimated
risk beyond 5.3% under this approach and under the given set of data, assuming the candidate
standards for the other media  (window sill dust, soil) remain fixed.
                                           270

-------
       The Rochester study data were used in this analysis as the multimedia model was fitted
based on the Rochester data. If data from other studies were used instead, it would be necessary
to verify that the model parameter estimates adequately reflect the underlying variability in these
data in the same manner that they reflect variability in the Rochester study data.

       While the approach presented in this section is relatively easy to implement, it could be
modified even further in an attempt to achieve more accurate risk estimates.  Such a modification
could reduce the level of simplicity associated with applying the approach. For example, rather
than calculate average environmental-lead levels across all  homes and fit the model once to these
averages, a simulation approach could be applied in an attempt to more accurately represent the
entire distribution of environmental-lead levels in these homes and the resulting blood-lead
distribution associated with exposure across the entire distribution of environmental-lead levels.

6.3    REVIEW OF PUBLISHED INFORMATION ON POST-INTERVENTION
       DUST-LEAD LOADINGS

       This  section summarizes published information on lead loadings (amount of lead per unit
surface area) in dust samples collected by wipe techniques, as reported by earlier lead
intervention studies. This information is used to evaluate assumptions made on post-intervention
dust-lead loadings (40 ug/ft2 for floors, 100 ug/ft2 for window sills) within the §403 risk analysis.
Details to supplement the summaries in this section are presented in Appendix H.

       The following seven studies have been identified in which some type of paint or dust
intervention was performed, dust samples were collected using wipes or some other technique
(e.g., BRM vacuum) whose results could be converted to wipe-equivalent dust-lead loadings, and
post-intervention dust-lead loadings on floors and/or window sills were reported (references for
these studies are included in Appendix H):

       •     Baltimore Experimental Paint Abatement Studies
       •     Baltimore Follow-up Paint Abatement Study
       •     Baltimore Repair & Maintenance (R&M) Study
       •     Boston Interim Dust Intervention Study
       •     HUD Grantees Evaluation (data available through September  1997)
       •     Denver Comprehensive Abatement Performance (CAP) Study
       •     Jersey City Children's Lead Exposure and Reduction (CLEAR) Study

These studies employed a variety of intervention strategies, including single or repeated dust
cleanings and interim control or complete abatement of lead-based paint. Dust-lead loadings
were measured at varying intervals following intervention.  Post-intervention dust-lead loadings
were summarized for 19 groups of housing units across these seven studies.  These study groups
are defined hi Appendix H.

       For both floors and window sills, geometric mean and median dust-lead loadings were
observed below  the post-intervention assumptions established in the §403 risk analysis in a

                                         271

-------
majority of the study groups. However, this does not preclude results for individual housing
units from being above the assumed levels. Furthermore, the extent to which results for these
studies represent the nation's housing stock has not been determined. Results are now presented
separately for floors and window sills (with more detailed presentations found in Appendix H).

6.3.1  Post-Intervention Floor Dust-Lead Loadings

       Summaries of post-intervention floor (wipe) dust-lead loadings are presented in Table
6-11 according to housing group within each study. According to Table 6-11, all but two of the
19 study groups reported geometric mean or median floor dust-lead loadings at or below 41
ug/ft2 from 6 months to 6 years post-intervention.  The other two study groups were from the
Baltimore Experimental Paint Abatement Study, where pre-intervention geometric mean dust-
lead loadings were much greater (556 ng/ft2 and 1261 ng/ft2) than any other study group (at most
58.6 ug/ft2).  Eleven study groups reported geometric mean or median floor dust-lead loadings at
or below 21 ug/ft2 at follow-up periods ranging from 12 months to 2 years. Of these 11 groups,
four of the HUD Grantees study groups reported median floor dust-lead loadings at or below 10
ug/ft2 at 12 months post-intervention.  Median pre-intervention floor dust-lead loadings in these
four groups ranged from 9 to 26 ug/ft2.

       In the HUD Grantees evaluation, seven of the eight largest grantees have median floor
dust-lead loadings at or below 21 ug/ft2 at 12 months post-intervention, compared to a median of
14 ug/ft2 across all grantees.  Although pre-intervention floor dust-lead loadings were lower in
the HUD Grantees evaluation compared to other studies, these preliminary results suggest that
floor dust-lead loadings can be maintained at levels below 40 ug/ft2 for at least 12 months post-
intervention.

       Results from the Denver CAP study, the Baltimore Follow-up Paint Abatement study, the
Baltimore R&M study, the Boston Interim Dust Intervention study, and the Jersey City CLEAR
study suggest that  geometric mean floor dust-lead loadings of below 40 ug/ft2 can be observed
even beyond 12 months post-intervention and up to six years post-intervention, under the same
conditions experienced by the housing units in these studies.

6.3.2  Post-Intervention Window Sill Dust-Lead Loadings

       Summaries of post-intervention window sill wipe dust-lead loadings are presented in
Table 6-12 according to housing group. Post-intervention geometric means or medians range
from 24 jig/ft2 to 958 fig/ft2, which are considerably higher than the summaries for floors.
Eleven study groups had geometric mean or median post-intervention window sill dust-lead
loadings below 100 ug/ft2,6 groups  were at or below 51 ug/ft2, and 3 groups were at or below 41
ug/ft2.

       All but one of the HUD Grantees study groups (the Milwaukee grantee) had median
window sill dust-lead loadings below 100 ^g/ft2 at 12 months post-intervention.  As the
intervention strategy for homes in the HUD Grantees evaluation frequently included partial or

                                         272

-------
Table 6-11.   Summaries of Pre- and Post-Intervention Floor Wipe Dust-Lead Loadings for
                Housing Groups Within Seven Studies
Study
Baltimore
Experimental Paint
Abatement
Studies2
Baltimore Follow-
up Paint
Abatement Study2
Baltimore R&M
Study3
Boston Interim
Dust Intervention
Study2
HUD Grantees4
Denver CAP Study8
Jersey City CLEAR
Study
Study
Group
Study 1
Study 2
1 2-Month Follow-up
19-Month Follow-up
Previously-Abated Units
Units Slated for R&M
Intervention
Automatic Intervention
Randomized Intervention
All Grantees
Baltimore
Boston
Massachusetts
Milwaukee
Minnesota
Rhode Island
Vermont
Wisconsin
Abated Units
Intervention Group
Pro-Intervention Floor
Dust-Lead Loadings1
(pg/ft2)
1261
556
NA
NA
45.6
58,6
33.2
37.3
19
41
24 .
24
14
18
26
28
9
NA
22
Post-Intervention
Floor Dust-Lead Loadings1
Time Following
Intervention
(Months)
6-9
1.5 -3.5 Years
10-14
14-24
4 - 6 Years
24
6
6
12
12
12
12
12
12
12
12
12
2 Years
12
i Summary Value
(pg/ft*)
99
69
20
36
33.0
35.0
23.9
31.4
14
41
18
9
10
18
6
21
5
21.0
15
1 Values are geometric means except for the HUD Grantees studies, where values are medians. "NA" indicates not
available.
* Results are adjusted to reflect total dust-lead loadings by exponentiating the "bioavailable" dust-lead loadings as reported in
the study to the 1.1416 power.
3 Results for the Baltimore R&M Study are converted from BRM dust-lead loadings to wipe-equivalent loadings.
* Data collected through September, 1997
5 Results for the Denver CAP study are converted from CAP cyclone dust-lead loadings to wipe-equivalent loadings.
                                                  273

-------
Table 6-12.   Summaries of Pre- and Post-Intervention Window Sill Wipe Dust-Lead
                Loadings for Housing Groups Within Seven Studies
Study
Baltimore
Experimental Paint
Abatement
Studies2
Baltimore Follow-
up Paint
Abatement Study2
Baltimore R&M
Study3
Boston Interim
Dust Intervention
Study2
HUD Grantees4
Denver CAP
Study5
Jersey City CLEAR
Study
Study
Group
Study 1
Study 2
12-Month Follow-up
19-Month Follow-up
Previously-Abated Units
Units Slated for R&M
Intervention
Automatic Intervention
Randomized Intervention
All Grantees
Baltimore
Boston
Massachusetts
Milwaukee
Minnesota
Rhode Island
Vermont
Wisconsin
Abated Units
Intervention Group
Pro-Intervention SHI
Dust-Lead Loadings1
fe/g/ft2)
15215
2784
NA
NA
163.5
778.4
787
205
258
1191
174
328
264
266
314
147
150
NA
75
Post-Intervention
Sill Dust-Lead Loadings1
Time Following
Intervention
6-9
1.5 -3.5 Years
10-14
14-24
4 - 6 Years
24
6
6
12
12
12
12
12
12
12
12
12
2 Years
12
Summary Value
(TO/*2)
958
199
41
147
97.6
204.9
210
110
90
68
49
50
217
77
85
40
51
66.4
24
1 Values are geometric means except for the HUD Grantees studies, where values are medians.  "NA' indicates not
available.
2 Results are adjusted to reflect total dust-lead loadings by exponentiating the "bioavailable' dust-lead loadings as reported in
the study to the 1.1416 power.
3 Results for the Baltimore R&M Study are converted from BRM dust-tead loadings to wipe-equivalent loadings.
* Data collected through September, 1997
* Results for the Denver CAP study are converted from CAP cyclone dust-lead loadings to wipe-equivalent loadings.
                                                 274

-------
 complete window replacement, these results may not be representative of the outcomes of
 interventions prompted by the §403 rule.

       Geometric mean window sill dust-lead loadings were below 100 ^fi2 for up to two years
 post-intervention in the Baltimore Follow-up Paint Abatement study, Denver CAP study, and
 Jersey City CLEAR study. However, in the Baltimore R&M study, Baltimore Experimental
 Paint Abatement studies, and Boston Interim Dust Intervention study, geometric mean dust-lead
 loadings remain above 100 ug/ft2 over time.  In addition, the 19-month follow-up study group
 within the Baltimore Follow-up Paint Abatement study and study group #2 of the Baltimore
 Experimental Paint Abatement studies suggest that geometric mean dust-lead loadings can dip
, below 100 Mg/ft2 immediately after intervention, but then exceed this level after one year or so.

 6.4   SENSITIVITY AND UNCERTAINTY ANALYSES FOR RISK
       MANAGEMENT ANALYSES

       The following subsections present the results of additional sensitivity and uncertainty
 analyses performed to gauge the level of uncertainty in the post-§403 risk estimates (and the
 associated decline from baseline estimates) associated with methodological assumptions. These
 results should be considered with those presented in the sensitivity and uncertainty analyses in
 Section 6.4 of the §403 risk analysis report to characterize overall uncertainty associated with the
 methods and assumptions taken in the risk management.

 6.4.1 Considering How Baseline  Environmental-Lead Levels May
       Have Changed Since the HUD National Survey

       Section 5.1.4 of this report addressed the sensitivity of the pre-§403 model-based blood-
 lead distribution and the resulting health effects and blood-lead concentration endpoint estimates
 under the IEUBK and empirical models under different assumptions on how  the national
 distribution of baseline environmental-lead levels as estimated using HUD National Survey data
 may have changed since the time of the survey (1989-1990). The same five sets of adjustments
 (i.e., percentage changes) made to the average baseline dust-lead loadings, dust-lead
 concentrations, and soil-lead concentrations for each housing unit in the HUD National Survey
 were considered in this sensitivity analysis to observe the impact on post-§403 risk estimates
 under the following set of example options for standards:

       •      Average floor dust-lead loading =100 ug/ft2
       •      Average window sill dust-lead loading = 500 ug/ft2
       •      Average soil-lead concentration = 2,000 \igfg
       •      Amount of deteriorated lead-based paint requiring paint maintenance = 5 ft2
       •      Amount of deteriorated lead-based paint requiring paint abatement - 20 ft2

 This set of options was the primary set considered in the sensitivity analyses within Section 6.4
 of the §403 risk analysis report.
                                         275

-------
       Table 6-13 presents the post-§403 estimates for the health effect and blood-lead
concentration endpoints under both the IEUBK and empirical models, for each of the five sets of
adjustments to the post-§403 environmental-lead levels hi housing units within the HUD
National Survey and under the above assumption on example standards.  Also included in this
table are the percentage of homes exceeding the various example standards, which will be lower
than in the §403 risk analysis when declines in the appropriate environmental-lead levels are
considered and higher when increases are considered. The table also lists the baseline risk
estimates for comparison purposes.

       Effect on risk analysis:  Under the five sets of assumptions involving lower assumed
baseline environmental-lead levels, the percentage of houses that exceed at least one of the
example standards declined by at most about three percentage points (from 21.8% to 18.7%;
Table 6-13), or about three million homes.  The assumption that baseline environmental-lead
levels are 25% higher than assumed in the §403 risk analysis results in an increase in the
percentage of homes exceeding at least one standard from 21.8% to 24.1%, an increase of about
2.3 million homes (Table 6-13).

       As would be expected, Table 6-13 shows that all assumptions on baseline environmental-
lead levels result in post-§403 estimates of the predicted health effect and blood-lead
concentration endpoints that are lower than baseline (the last column of the table). However, as
the assumed baseline environmental-lead levels become lower in magnitude, the predicted post-
1403 risks actually increase, converging to the baseline estimates.  For example, as seen in Table
6-3, baseline lead levels that are 20% below what was assumed in the §403 risk analysis resulted
in an estimated percentage of children with blood-lead concentrations at or above 10 ug/dL of
4.85%, compared to the §403 risk analysis estimate of 4.70%. When baseline lead levels are
50% below the §403 risk analysis estimates, the estimate of this percentage increases to 5.10%.
Such a finding appears counter-intuitive when first reviewing the table. However, the alternative
assumptions being considered hi this sensitivity analysis are to baseline (i.e., pre-§403)
environmental-lead levels. As assumptions on these baseline levels move lower, fewer homes
are triggered by the §403 standards, and the post-§403 distribution of environmental-lead levels
becomes less removed from the baseline distribution. As a result, post-§403 estimates of
predicted health effects and blood-lead concentration are not as different from pre-§403
estimates. In contrast, as assumed baseline environmental-lead levels increase, more homes are
triggered by the §403 standards and, therefore, have their environmental-lead levels drop as  a
result of interventions, and lower post-§403 risk estimates relative to baseline are observed.

       As seen in Table 6-13, the effect that different assumptions on baseline environmental-
lead levels have on the risk estimates is considerably greater under the IEUBK model than the
empirical model. The percentage of children with blood-lead concentrations at or above 20
ug/dL more than triples under the IEUBK model approach when 50% declines hi both dust-lead
and soil-lead levels were assumed (from 0.054% to 0.166%), compared to a 16% increase under
the empirical model (from 0.406% to 0.469%).  Smaller percentage differences are observed for
the other endpoints for both models.
                                         276

-------
Table 6-13.   Sensitivity Analysis on How Changes in Household Average Baseline Dust-
              Lead Loadings/Concentrations and Soil-Lead Concentration Impact Post-
              8403 Estimates of Health Effect and Blood-Lead Concentration Endpoints for
              Children Aged 1-2 Years Under a Specified Set of Example Standards1
Assumed Percentage Change In Average Dust-Lead Loadings and Concentrations
(Both Floor and Window Sill) and in Yard-wide Average Soil-Lead Concentration
Dust:
•--:,. . Soil:
.LNo
change
l*to
change
20% ;
decrease
20% ,
decrease.
50%
decrease
', w^> ,''
decrease'
50%
decrease
No.
. change
No
change
. 50%
decrease
Percentage of Homes Exceeding Example Standards/Triggers
Floor Dust
Window Sill Dust
Soil
Interior Paint Maintenance
Exterior Paint
Maintenance
Interior Paint Abatement
Exterior Paint Abatement
Any Standard/Trigger
4.04
12.5
2.49
2.92
3.49
2.43
5.77
21.8
2.34
10.8
1.52
2.92
3.49
2.43
5.77
20.6
0.694
9.10
0.746
2.92
3.49
2.43
5.77
18.7
0.694
9.10
2.49
2.92
3.49
2.43
5.77
18.9
4.04
12.5
0.746
2.92
3.49
2.43
5.77
21.6
25%
increase
. 25%
increase
Baseline
Estimate
{from
Table 5-1
of tiie
§403 risk
analysis
report)

5.68
14.3
3.27
2.92
3.49
2.43
5.77
24.1

Predicted Hearth Effect And Blood-Lead Concentration Endpoints (Based on Empirical Model)
PbB *20 (%)
PbBi10(%)
IQ < 70 (%)
IQ decrement 2 1 (%)
IQ decrement *2 (%)
IQ decrement 23 (%)
Avg. IQ decrement
0.406
4.70
0.110
36.3
9.30
2.93
1.00
0.429
4.85
0.111
36.7
9.53
3.04
1.01
0.469
5.10
0.112
37.3
9.90
3.21
1.03
0.445
4.95
0.111
36.9
9.69
3.11
1.02
0.427
4.84
0.111
36.7
9.51
3.03
1.01
0.378
4.52
0.110
35.9
9.02
2.80
0.995
0.588 •
5.75
0.115
38.5
10.8
3.70
1.06
Predicted Health Effect And Blood-Lead Concentration Endpoints (Based on IEUBK Model)
PbB 220 (%)
PbB2lO(%)
IQ < 70 <%)
IQ decrement 2! (%)
IQ decrement ±2 (%)
IQ decrement a3 (%)
Avg. IQ decrement
0.0539
1.66
0.0984
28.3
4.31
0.858
0.848
0.117
2.48
0.102
31.0
5.77
1.37
0.894
0.166
2.98
0.104
32.7
6.65
1.71
0.924
0.121
2.55
0.102
31.6
5.94
1.42
0.904
0.0681
1.86
0.0992
28.8
4.67
0.983
0.857
0.0542
1.64
0.0982
27.7
4.22
0.847
0.839
0.588
5.75
0.115
38.5
10.8
3.70
1.06
1 Example dust and soil standards were set at: 100 //g/ft2 for floor dust-lead loading, 500 //g/ft2 for window sill
dust-lead loading, and 2,000 jt/g/g for soil-lead concentration. Paint maintenance is performed if more than 5
ft2, but less than 20 ft2 of deteriorated lead-based paint exists.  Paint abatement is performed if more than 20
ft2 of deteriorated lead-based paint exists.
                                            277

-------
6.4.2 Impact on the Estimated Incidence of IQ Point Decrement
       Assuming Certain Thresholds on the IQ/Blood-Lead Relationship

       The sensitivity of baseline and pre-§403 model-based estimates of IQ decrements greater
than 1,2, or 3, and of the average and standard deviation of the distribution of IQ point
decrements was addressed in Section 5.1.5 of this report for various assumptions of a non-zero
threshold of blood-lead concentration on the IQ/blood-lead relationship. The following
thresholds were considered: 1,2, 3,5, 8 and 10 ug/dL. In this section, post-§403 estimates of
these health effect endpoints are estimated (under the same set of options presented in Section
6.4.1, using both the DEUBK and empirical models) under these same alternative blood-lead
concentration thresholds. These estimates are presented in Table 6-14.

       Effect on risk analysis:  As was also seen in Table 5-7 of this report, Table 6-14 shows
that the post-§403 risk estimates decrease as the assumed blood-lead concentration threshold
increases (i.e., smaller percentages of children experience IQ score decrements under larger
threshold assumptions). The IEUBK model is more sensitive than the empirical model to the
threshold level.  For example, the probability of a child experiencing an IQ decrement of at least
1 point decreases by 63% under the IEUBK model (from 28.3% to 10.4%) when the threshold
increases from 0 to 2 ug/dL, compared to only a 52% decrease under the empirical model (from
36.3% to 17.6%).  As the assumed threshold increases, the likelihood of experiencing an IQ
decrement of at  least 1 point as a result of lead exposure decreases to very low values under both
models, and the average IQ score decrement in the population declines to small fractions of
points.

6.4.3  Considering  Alternative Assumptions  on Post-Intervention
       Dust-Lead Loadings

       In the risk management portion (Chapter 6) of the §403 risk analysis report, it was
necessary to make assumptions on predicted post-intervention lead levels when characterizing
the blood-lead concentration and health effect endpoints in a post-§403 environment. These
assumptions were documented in Table 6-2 of the §403 risk analysis report. Among these
assumptions were that dust cleaning activities impacted interior dust-lead loadings in the
following way:

       •      Post-intervention household average floor (wipe) dust-lead loadings equaled the
             minimum of 40 ug/tfand the pre-intervention value.

       •      Post-intervention household average window sill (wipe) dust-lead loadings
             equaled the minimum of 100 ug/ft* and the pre-intervention value.

A dust cleaning  was  assumed to be included among the interventions performed when either the
floor-dust, window sill-dust, soil, or interior paint abatement standards were exceeded within a
home. These two assumptions on post-intervention dust-lead loadings were made within the
§403 risk analysis based on data reported in EPA's Comprehensive Abatement Performance

                                         278

-------
Table 6-14.  Sensitivity Analysis on the Assumed Blood-Lead Concentration Threshold on
               IQ Decrement and Its Impact on the Post-§403 Estimates of IQ Decrement
               Endpoints for Children Aged 1-2 Years, Under a Specified Set of Example
               Standards1
Assumed
Threshold
fc/g/dLl
% of Children Aged 1-2 Years with a Specified IQ
Decrement' Due to Lead Exposure2
IQ Decrement as-,'
T . ,;':
IQ Decrement 2
;]:,:•• 2 '..
IQ Decrement 2
3 ' '
Average IQ
Decrement
{# points)3
Standard
Deviation of IQ
n»».^. ......
-------
study and in the Baltimore Experimental Paint Abatement study (see Section 6.1.2 of the §403
risk analysis report and Section H2.0 of Appendix H of this report).

       Tables 6-11 and 6-12 within Section 6.3 of this report presented additional information
on household average (wipe) dust-lead loading at pre- and post-intervention for floors and
window sills, respectively, from several recent lead intervention studies. This information, some
of which was received after the §403 risk analysis report was completed, suggests that it may be
common in some instances to observe household average post-intervention dust-lead loadings
below the assumptions made above, even from 12 months to six years post-intervention. These
findings prompted a sensitivity analysis to investigate how setting assumptions on post-
intervention household average dust-lead loadings to below the 40 ug/ft2 and 100 jug/ft2
specifications would impact the outcome of the risk management analyses.

       In this sensitivity analysis, two alternative assumptions on household average post-
intervention floor dust-lead loadings were made:  10 ug/ft2 and 25 fig/ft2. As the geometric mean
(12-month) post-intervention floor dust-lead loading in the HUD Grantees evaluation was 14
Hg/ft2 (Table 6-8) and was even lower for certain  grantees, an alternative of 10 ug/ft2 was
selected. The alternative of 25 ug/ft2 for floors was selected as it fell halfway between the
assumptions of 10 and 40 ug/ft2 and was within the range of expected variability in the
summaries for several of the studies in Section 6.3.1.

       Similarly, two alternative assumptions on household average post-intervention window
sill dust-lead loadings were made:  50 ug/ft2 and 75 ng/ft2. Evidence from Table 6-12 indicates
that average window sill dust-lead loadings following intervention could approach 50 ug/ft2 in
some instances, especially when floor dust-lead loadings are low. The alternative of 75 ug/ft2
was selected as it fell halfway between the assumptions of 50 and 100 ug/ft2, and it was similar
to the average levels observed by grantees within the HUD Grantees evaluation (although the
HUD Grantees evaluation included window replacement, which was not among the assumed
interventions in the §403 risk analysis).

       In the sensitivity analysis, if a given household's pre-intervention average floor dust-lead
loading fell below the given post-intervention assumption, its post-intervention household
average floor dust-lead loading was assumed to be equal to its pre-intervention average (as was
done in Chapter 6 of the §403 risk analysis report). Second, this sensitivity analysis considers
predictions made only by the empirical model, as the IEUBK model does not accept dust-lead
loading as input. Finally, the assumptions made in determining post-intervention soil-lead
concentrations (150 ug/g following soil removal) and amount of deteriorated lead-based paint
(none is present following paint intervention) remained the same as specified in Table 6-2 of the
§403 risk analysis report.

       Table 6-15 presents the estimated post-§403 health effect and blood-lead concentration
endpoints associated with the set of example options for standards specified in Section 6.4.1
above, for the alternative assumptions on post-intervention floor and window sill dust-lead
loadings specified above. Note that each alternative assumption is evaluated on its own (i.e., it is

                                          280


-------
     •S **
    IB*
                           »
                        IS-
                        n*
is
II
CO 10
".  •' I***"
                              CO ID
                              sg
«•*
is
250

id
V*

CD
.O
CO
                                    CO
                                    CO
                                    in
               10
               IX
               ui
                   IO
                IO

                CD
                CO
              CO
              6
          CO
          o
                                                     CO
                                    cn
                                    CO
               (O
                CM

                CO
                CO
              CM
              en
       0)
       00
       CM
   o
   o
            CO
               CO
               10
                0)
                id
                CO
              CO
              q
              oi
       CO

       CM
   10
   cn
   cn
                                    CM
                                       00
                                       
                 i3
                 
                                 a
                     TJ
                     a
                                                         ?
                                      281
                          -S? t
                          SI
                                                             «  TJ
               «  T> d>
               C *• -C
               t o •*
                    O 0)
                                                               I CL 5
                                                                    ^«
                                                                    o «?.E 2

-------
the only change from the §403 risk analysis assumptions). In addition, considering the high
correlation in dust-lead loadings between floors and window sills, the two lower alternatives (10
ug/ft2 for floors and 50 ug/ft2 for window sills) and the two higher alternatives (25 fig/ft2 for
floors and 75 fig/ft2 for window sills) are evaluated together. For comparison purposes, post-
intervention estimates under the §403 risk analysis (i.e., assuming 40 fig/ft2 for floors and 100
ug/ft2 for window sills) and the estimates generated under baseline (pre-§403) conditions (both
presented in Table 6-7 of the §403 risk analysis report) are also included in Table 6-15.

       Effect on risk analysis. Relative to the results reported hi the §403 risk analysis report
(column 2 of Table 6-15), the greatest deviation occurs with the most substantial change in the
assumptions, i.e., the assumptions of 10  ug/ft2for floors and 50 ug/ft2 for window sills (column 7
of Table 6-15).  Under this particular set of alternative assumptions, the percentage of the
nation's children aged  1-2 years that are  anticipated to have blood-lead concentration at or above
10 ug/dL following interventions conducted in response to the §403 rule (given the example
standards specified in the footnote to this table) is reduced from 4.70% to 4.53% (a 3.7% decline,
equivalent to approximately 13,700 children12). The corresponding reduction in the percentage
of children with blood-lead concentration at or above 20 ug/dL is from 0.406% to 0.380% (a
6.3% decline, equivalent to approximately 2,000 children).

       Under the assumptions of 25 jig/ft2 for floors and 75 ug/ft2 for window sills (column 8 of
Table 6-15),  the percentage of the nation's children aged 1-2 years that are anticipated to have
blood-lead concentration at or above 10 ug/dL is reduced from 4.70% to 4.64% (a 1.2% decline,
equivalent to approximately 4,800 children). The corresponding reduction in the percentage of
children with blood-lead concentration at or above 20 ug/dL is from 0.406% to 0.397% (a 2.3%
decline, equivalent to approximately 750 children).

       Generally, even lower percentage declines occur for the IQ endpoints compared to the
blood-lead concentration endpoints. The exception occurs with the percentage of children with
IQ decline of at least 3 points, where a 4.2% decline from the §403 risk analysis assumptions was
observed under assumptions of 10 ug/fi2 for floors and 50 Mg/ft2 for window sills.

       This sensitivity analysis indicates that while more housing units may achieve reductions
in average dust-lead levels  on floors and window sills following a dust cleaning if the assumed
post-intervention floor dust-lead loadings are lowered from those made hi the §403 risk analysis,
the corresponding reduction in the estimated blood-lead concentration and health effect endpoints
appears to be modest, especially when compared to the reduction observed from pre- to post-
§403 conditions.
       12 Assuming that 7.96 million children aged 1-2 years reside in the U.S. housing stock (Table 3-3S of the
§403 risk analysis report).

                                           282

-------
6.4.4  Characterizing the Post-Intervention Blood-Lead Distribution
       Based on Relative Change from Baseline in the Geometric
       Mean  and the Probability of a Child's Blood-Lead
       Concentration Exceeding 10/ig/dL

       As discussed in Section 4.3.1 above and in Appendix Fl of the §403 risk analysis report,
a "scaling algorithm" was used in the §403 risk analysis to characterize the distribution of blood-
lead concentration in the nation's children following interventions that would be performed as a
result of implementing the §403 rule (where the algorithm was applied under a specified set of
example options for the standards, using a specified blood-lead prediction model, and under
assumptions made on the changes in environmental-lead levels that result from the
interventions). This distribution is labeled the "post-§403" distribution. This approach
calculated the geometric mean (GM) and geometric standard deviation (GSD) of the post-§403
blood-lead distribution in the following manner:
                                                                                    ( 1)
                                                                                    (2)
where the subscripts indicate the blood-lead distribution which either the GM or the GSD
represents.  See Section 4.3.1 for additional information on this approach.

       One comment received on the §403 risk analysis was that because the blood-lead
concentration endpoints utilized in the risk analysis were exceedance probabilities (i.e., the
likelihood of a child's blood-lead concentration exceeding a specified value), it was more
important to accurately characterize the right tail of the post-§403 distribution compared to the
remainder of the distribution, especially at blood-lead levels beyond 10 ug/dL.  Therefore, a
variant of the scaling approach was considered that involved scaling the probability of a child's
blood-lead concentration exceeding 10 ug/dL rather than the GSD. If P10 was used to represent
this probability, then the alternative scaling algorithm would involve scaling the geometric mean
as in (I) above, but replacing (2) above with the following calculation:
                P10post-403 ~P 1 "baseline   0" ^model-based post-403 ' ** "model-based pre-403)
(3)
The resulting value is the estimate of the probability of a child's blood-lead concentration
exceeding 10 ug/dL in a post-§403 environment. It is calculated by multiplying the probability
as calculated in the baseline distribution by the relative change in the probability from the pre-
§403 to post-§403 environment as estimated from model-based blood-lead distributions. Then,
in order to calculate the other blood-lead concentration and health effect endpoints, the GSD of
the post-§403 distribution would be calculated by assuming that this distribution is lognormal.
Therefore,
                          = exp{(log(10) -
(4)
                                          283

-------
where O"1 denotes the inverse of the standard normal distribution function.

        Table 6-16 presents the estimated blood-lead concentration and health effect endpoints
that result when applying this alternative scaling algorithm, under both the IEUBK and empirical
models. The example options for standards that are assumed in this analysis are the same as
those considered in Section 6.4.1 above and are specified in a footnote to Table 6-16. For
comparison purposes, this table also contains the estimates under the original version of the
scaling approach that was utilized in the §403 risk analysis.
Table 6-16.   Estimated Post-§403 Health and Blood-Lead Concentration Endpoints Under
               the Original and Alternative Scaling Algorithms for Characterizing the Post-
               §403 Blood-Lead Distribution

Original scaling algorithm: Geometric mean and GSD are scaled.
Alternative scaling algorithm: Geometric mean and the probability of PbB exceeding 10^/g/dL are scaled.
Health Effect and Hood-Lead
: Concentration Endpoints
% of Children with PbB i 20 A/g/dL
% of Children with PbB 2 10//g/dL
%of Children with IQ < 70 due to
lead exposure
% of Children with IQ decrement 2 1
due to lead exposure
% of Children with IQ decrement 2 2
due to lead exposure
% of Children with IQ decrement 2 3
due to lead exposure
Avg. IQ decrement due to lead
exposure
Geometric Mean PbB (GSD)
Post-1403 Estimates Under the
Risk Management Analysis
(Original Scafing Algorithm)
IEUBK Model
0.0539
1.66
0.0984
28.3
4.31
0.858
0.848
2.74(1.84)
Empirical
Model
0.406
4.70
0.110
36.3
9.30
2.93
1.00
3.03 (2.04)
Post-1403 Estimates Under the
Alternative Scaling Algorithm
IEUBK Model;:
0.156
2.72
0.102
30.1
6.05
1.56
0.884
2.74 (1 .96)
- Empirical
Model 1 :
0.249
3.78
0.107
35.5
8.03
2.24
0.977
3.03(1.96)
Note: Example dust and soil standards were set at: 100 /rg/ft9 for floor dust-lead loading, 500 pg/ft* for window sill dust-
lead loading, and 2.000 ;ig/g for soil-lead concentration. Paint maintenance is performed if more than 5 ft*, but less than
20 ft1, of deteriorated lead-based paint exists. Paint abatement is performed if more than 20 ft2 of deteriorated lead-based
paint exists.  GSD = geometric standard deviation. PbB = blood-lead concentration
                                             284

-------
      Effect on risk analysis. As indicated in Table 6-16, when the probability of exceeding
10 ug/dL is scaled instead of the GSD, the estimated probability is approximately 64% higher
under the IEUBK model (1.66% to 2.72%), but nearly 20% lower under the empirical model
(4.70% to 3.78%). Note that under the alternative approach, estimates based on the IEUBK and
empirical models are more similar to each other than under the original scaling algorithm.  In the
alternative approach, the estimated post-§403 GSD is the same under both models:  1.96. Note
that there was no change in the manner in which the geometric mean blood-lead concentrations
were determined, and therefore, no change is noted between the two approaches.

      The above results indicate that the alternative scaling approach has a more significant
impact on the IEUBK model-based estimates compared to the empirical model-based estimates.
The impact of the approach on the empirical model-based estimates is a reduction in the risk
estimates due to a 4% reduction in the estimated GSD, while the impact on IEUBK model-based
estimates is an increase in the  risk estimates due to a 6.5% increase in the estimated GSD.
However, because the two approaches did not differ hi how the post-§403 geometric mean blood-
lead level was calculated, the empirical model estimates remain higher than the IEUBK model
estimates.

6.5   LEAD EXPOSURE ASSOCIATED WITH CARPETED  FLOOR-DUST

      While the §403 proposed rule included a proposed lead hazard standard for dust on
uncarpeted floors, EPA determined that sufficient technical data were not available to direct how
the rule should address lead-contaminated dust on carpeted floors. Based upon public comments
on the proposed rule, EPA is revisiting that determination. This section summarizes the key
findings of statistical analyses on dust-lead loading data for carpeted floors. The analysis had the
following three objectives:

       1.     Assess the need to have dust-lead on carpeted floors addressed by the §403 rule:

             a.     Characterize the relationship between floor dust-lead levels and blood-lead
                    concentration in young children and how this relationship differs for
                    carpeted and uncarpeted floors (with and without adjusting for the effects
                    of key demographic variables and for lead levels in other media in which
                  .  standards have been proposed in the §403 rule).

             b.     Determine the added value of including a carpet dust-lead standard given
                    the proposed §403 standards for soil, window sills and uncarpeted floors,
                    or expanding the definition of floors in the rule to include carpeted as well
                    as uncarpeted floors.

      2.     Identify appropriate candidates for carpeted floor dust-lead standards and, in
             particular, whether one candidate standard should correspond to 50 ug/ft2, the
             uncarpeted floor dust-lead standard from the §403 proposed rule.
                                         285

-------
       3.     Determine whether the wipe technique is acceptable for sampling dust from
             carpeted floors for evaluating the risk of lead exposure associated with carpet-
             dust, or whether alternative vacuum methods are more appropriate.

A more detailed presentation of the statistical analyses that address these three objectives is
found in Appendix I of this report.

       The carpet dust-lead measurement data used in this analysis originated from two lead
exposure studies:  the Rochester (NY) Lead-in-Dust study, and the pre-intervention, evaluation
phase of the HUD Lead-Based Paint Hazard Control Grant ("HUD Grantees") Program (data
collected through September, 1997).  Both studies were introduced in Section 3.3.1 of the §403
risk analysis report; additional details on these studies that are relevant to this analysis is
presented in Section 13.1 of Appendix I. The results of this analysis, along with relevant findings
documented in EPA's recent literature review report on lead exposure associated with carpets,
furniture, and air ducts (USEPA, 1997b), were used to address the above objectives.

       The summary of the analysis results now follows. It is formatted according to the above
three objectives. References to statistical significance are made at the 0.05 level. Unless
otherwise indicated, references to dust-lead loadings are assumed to be for samples collected
using wipe techniques. Section numbers within Appendix I are specified in parentheses where
additional information can be found.

Objective #1: Is there a need to have dust-lead on  carpeted floors addressed by the §403
rule?

       •      Using data collected in the 1997 American Housing Survey, EPA estimates that
             approximately 54 million housing units built prior to 1978 contain some wall-to-
             wall carpeting.  Of these units, wall-to-wall carpeting is found in a living room in
             approximately 47 million units and in a bedroom in approximately 46 million
             units (i.e., rooms in which children reside and play most frequently, and therefore,
             would be targeted in a risk assessment).

       •      While the §403 proposed rule indicates that lead from floor dust is an important
             exposure source for children, the proposed floor dust-lead loading standard was
             only relevant for uncarpeted floors. In homes with wall-to-wall carpeting, it is
             expected that floor-dust samples in certain rooms can come only from carpeted
             floors. While no guidance was given in the §403 proposed rule on a standard to
             which risk assessors should compare the results of lead analyses for carpet dust
             samples, EPA recognizes (and many commenters on the §403 proposed rule have
             noted) that some recommendation for a carpet dust-lead loading standard, based
             on using wipe collection techniques, is necessary.
                                          286

-------
 •      Because children come in frequent direct contact with carpeting when it is present
       in their homes, any lead that may be present in carpet dust is likely to be
       bioavailable to children.

 Objective #1a: Is there any association between carpeted floor dust-lead loadings
 and blood-lead concentration?

 •      For both carpeted and uncarpeted floors in the two studies, the correlation
       between household average floor (wipe) dust-lead loading and children's blood-
       lead concentration was positive and significantly different from zero. (Sections
       14.1.1.1 and 15.1.1.1 of Appendix I)

 •      No evidence was found in these analyses to suggest that wipe dust-lead loadings
       from uncarpeted floors are a better predictor of children's blood-lead
       concentration than wipe dust-lead loadings from carpeted floors. (Sections
       14.1.1.2,14.1.1.4,15.1.1.2 and 15.1.1.4 of Appendix I)

 •      No significant difference in the statistical relationship between average floor dust-
       lead loading and blood-lead concentration was found between homes with floor
       dust sampling conducted from mostly carpeted floors and homes with sampling
       from mostly uncarpeted floors. (Sections 14.1.1.3 and 15.1.1.3 of Appendix I)

 •      Mixed results were found when investigating whether the effect of average
       carpeted floor dust-lead loading on blood-lead concentration remained significant
       after adjusting for the effects of lead levels in soil, window sill dust, and
       uncarpeted floor dust (i.e., other environmental media addressed by the proposed
       §403 standards). The carpet dust-lead loading effect was no longer statistically
       significant after adjusting for these other effects when analyzing data from the
       Rochester study, while the effect remained statistically significant when analyzing
       data from the HUD Grantees program evaluation.  (Sections 14.1.2 and 15.1.2 of
       Appendix I)

' •      When interpreted as a whole, these findings provide a powerful argument for
       expanding the floor dust-lead standard in the §403 rule to include carpeted floors.

 Objective #1 b:  Is there any added benefit to adding a carpeted floor  dust-lead
 loading  standard to the proposed §403 standards  for lead in soil, window sill dust,
 and dust from uncarpeted floors, or to expanding the definition of floors in the rule
 to include carpeted floors?  (Sections 14.1.3 and 15.1.3 of Appendix I)

 •      The extent of any added benefit is dependent on the value of the carpet dust-lead
       loading standard and the particular criteria being considered in evaluating
       performance. Adding a new standard to a set of existing standards will not reduce
       sensitivity (i.e., the proportion  of homes with elevated blood-lead children that are

                                     287

-------
              triggered by the set of standards), but it also will not increase specificity (i.e., the
              proportion of homes with no elevated blood-lead children that are not triggered
              for an intervention by the standards).

       *       If the uncarpeted floor dust-lead loading standard of 50 us/ft2 that was proposed
              in the §403 proposed rule was extended to included carpeted floors as well, the
              resulting performance of the§403 proposed standards (based on the outcome of
              performance characteristics analysis) changed little, if any.  If a carpeted floor
              standard of 40 ug/ft2 was added to the §403 proposed standards, slight
              improvements in the performance characteristics were noticed.  These findings
              were observed regardless of whether or not uncarpeted floors were available to
              sample (i.e., whether or not the uncarpeted floor standard was considered).

       •       Analyses of the Rochester study data indicated that adding a carpet dust-lead
              loading standard of approximately 17 fig/ft2 to the proposed §403 standards
              considerably improved certain performance characteristics, particularly sensitivity,
              without a large decrease in specificity.

       •       Analysis of the HUD Grantees evaluation data indicated that adding a carpet dust-
              lead loading standard of approximately 5 ug/ft2 improved sensitivity and negative
              predictive value (NPV, equal to the proportion of homes not triggered for
              intervention by the standards that do not contain elevated blood-lead
              concentration), but was accompanied by a considerable decrease in specificity. If
              the proposed carpet dust-lead loading standard was increased to approximately J3
              ug/ft2. this loss of specificity relative to the gains in sensitivity and NPV was
              reduced.

       •       In general, these analyses concluded that expanding the proposed §403 floor dust-
              lead standard (of 50 ug/ft2) to encompass both carpeted and uncarpeted floors, or
              setting this standard slightly lower at 40 ug/ft2, would not lead to a large decrease
              in specificity, but it would tend  to result in only minor increases in sensitivity
              from what was observed when carpeted floor standards were not being
              considered.

Objective #2: If a carpeted floor standard is needed, what should it be?  Should  it  be
different from the proposed uncarpeted floor dust-lead loading standard of 50 //g/ft2?

       •       The findings listed above for Objective #lb suggest that it may provide an
              advantage to have a standard for carpeted floors that is lower than the standard for
              uncarpeted floors. (Sections 14.1.1.2,14.2.1,15.1.1.2 and  15.2.1 of Appendix I)

       •       Having a floor dust-lead loading standard of 40 to 50 ug/ft2 that is expanded to
              represent carpeted floors as well as uncarpeted floors would be at least as
              protective of children (in terms of the predicted blood-lead concentration at which

                                           288

-------
             95% of children exposed at the standard level would be expected to fall below)
             than if the standard represented only uncarpeted floors. (Sections 14.2.2 and 15.2.2
             of Appendix I)

       •      When the Rochester study data was used in a performance characteristics analysis
             that considered only standards for either carpeted or uncarpeted floors (Sections
             14.2.3 and 15.2.3 of Appendix I), a carpeted floor dust-lead loading standard in the
             range of 15 to 20 ug/ft2 maximized the total of the four performance
             characteristics.  In contrast, a standard of 50 fig/ft2 resulted in considerably lower
             performance when the standard was for carpeted floors versus uncarpeted floors.
             The level of sensitivity achieved by an uncarpeted floor dust-lead loading standard
             of 50 ug/ft2 was achieved for carpeted floor dust-lead loading standards below
             approximately 33 ug/ft2. However, the uncertainty associated with these estimates
             may suggest that these lower levels may not actually differ from a practical
             standpoint from the uncarpeted floor dust-lead loading standard in the §403 rule.

Objective #3; What dust sampling method should be used on carpeted floors? (Sections
14.3 and 15.3 of Appendix I)

       •      The HUD Guidelines (USHUD, 1995) support the use of wipe methods to sample
             carpet dust.  Participants in the §403 Dialogue Group meetings raised concerns
             that requiring widespread use of vacuum techniques for collecting dust samples in
             typical risk assessments would be impractical. Therefore, it would be preferable
             to allow wipe sampling as an option for collecting dust samples from carpets in a
             risk assessment unless wipe techniques were totally unacceptable.

       •      Different types of dust collection methods can collect different amounts of lead
             within a dust sample, especially when sampling from carpets where surface dust is
             easier to sample than dust that is deep within the carpet fibers. A laboratory study
             done in conjunction with the Rochester study (Emond et al., 1997) concluded that
             lead recovery from carpet dust was highest with the BRM vacuum (95.2%)
             compared to the wipe (24.4%) and the DVM vacuum (31.4%). For this reason,
             different dust collection methods for collecting carpet dust would require different
             lead standards to which to compare the results.

       •      When the wipe method is used on carpets, it tends to collect only dust on the
             carpet surface that can readily be removed by the method. This surface dust is
             also that which is most likely to come into direct contact with children (USEPA,
             1997b).

       •      Blood-lead concentration tends to be more highly associated with dust-lead
             loading than with dust-lead concentration in carpets.  (Only dust-lead loadings can
             be measured under wipe techniques, while loadings or concentrations can be
             measured under vacuum methods.) This contributes to the technical justification

                                          289

-------
that a carpet dust-lead standard would be better conveyed as a loading than as a
concentration.

Each of the three dust collection methods considered in the Rochester study
(BRM vacuum, DVM vacuum, wipe) collected carpet dust samples whose dust-
lead loadings were statistically associated with blood-lead concentration, with the
level of association being similar for each method.

On both carpeted and uncarpeted floors, dust-lead loading measurements from
different dust collection methods were significantly positively correlated.  This
suggests that using any of the three methods (including wipe) would portray the
extent of a carpet dust-lead hazard in a similar fashion.

As wipe sampling is currently the method of choice for uncarpeted floors and all
three methods have significant correlations with blood-lead concentration for
carpeted and uncarpeted floors, it is reasonable to develop a carpeted floor dust-
lead loading standard for the wipe sampling method. As this standard would not
apply to vacuum sampled dust-lead loadings, measurements for samples collected
using vacuum techniques could not be directly used in risk assessment via the
§403 rale without first being converted to wipe-equivalent loadings using methods
such as those documented in Section 4.3 of the §403 risk analysis report.
                             290

-------
7.0    REFERENCES

Abadin, H.G., Wheeler, IS., Jones, D.E., De Rosa, C.T., and MSPH (1997) "A Framework to
     Guide Public Health Assessment Decisions at Lead Sites." Journal of Clean Technology,
     Environmental Toxicology, and Occupational Medicine. 6(3):225-237.

Altmann, L., Weinsberg, F., Sveinsson, K., Lilienthal, Hi, Wiegand, H., and Winneke, G.  (1993)
     "Impairment of Long-Term Potentiation and Learning Following Chronic Lead Exposure."
     Toxicol. Lett. 66:105-112.

Angell, N.F., and Weiss, B. (1982) "Operant Behavior of Rats Exposed to Lead Before or After
     Weaning." Toxicology and Applied Pharmacology. 63:62-71.

Angle, CR., Manton, W.I., and Stanek, K.L. (1995) "Stable Isotope Identification of Lead
     Sources in Preschool Children:  the Omaha Study."  Journal of Toxicology - Clinical
     Toxicology.  33(6):657-662.

Angle, C. R., and Mclntire, M. S. (1979) "Environmental Lead and Children: The Omaha
     Study." Journal of Toxicology and Environmental Health. 5:855-870.

Annest, J.L., and Mahaffey, K.  (1984) "Blood Lead Levels for Persons Ages 6 Months - 74
     Years: United States, 1976-1980." National Center for Health Statistics, Centers for
     Disease Control and Prevention. Vital Health Stat Series No. 11 (233).  (PHS) 84-1683.
     August 1984.

Baghurst, P.A., McMichael, A.J., Wigg, N.R., Vimpani, G., Robertson, E.F., Roberts, R.J., and
     Tong, S. (1992) "Environmental Exposure to Lead and Children's Intelligence at the Age of
     Seven Years: The Port Pirie Cohort Study." New England Journal of Medicine. 327:1279-
     1284.

Banks, E.C., Ferretti, L.E., and Shucard, D.W. (1997) "Effects of Low Level Lead Exposure on
     Cognitive Function in Children: a Review of Behavioral, Neuropsychological and
     Biological Evidence." Neurotoxicology. 18(l):237-282.

Barltrop, D., and Meek, F. (1979) Effect of particle size on lead absorption from the gut.
     Archives of Environmental Health. 34: 280-285.

Barltrop, D., Strehlow, C.D., Thorton, I., and Webb, J.S. (1974) "Significance of High Soil Lead
     Concentrations for Childhood Lead Burdens." Environmental Health Perspectives. 7:75-
     82.
                                          291

-------
Bates, M., Malcolm, M., Wyatt, R., Galloway, Y., Gairett, N., Speir, T., and Read, D. (1995)
     "Lead in Children from Older Housing Areas in the Wellington Region." New Zealand
     Medical Journal, 108(1009):400-404.

Battelle (1996a) "Statistical Evaluation of the Relationship Between Blood-Lead and Dust-Lead
     Based on Data from the Rochester Lead-in-Dust Study." Final Report prepared by Battelle
     for the Office of Prevention, Pesticides, and Toxic Substances, U.S. Environmental
     Protection Agency. Contract No. 68-D2-0139. 22 February 1996.

Battelle (1996b) "Statistical Evaluation of the Relationship Between Blood-Lead and Dust-Lead
     Based on Pre-Intervention Data from the R&M Study." Final Report prepared by Battelle
     for the Office of Prevention, Pesticides, and Toxic Substances, U.S. Environmental
     Protection Agency. Contract No. 68-D2-0139.

Bellinger, D., Stiles, K.M., and Needleman, H. (1992) "Low-Level Lead Exposure, Intelligence,
     and Academic Achievement:  A Long-term Follow-up Study." Pediatrics. 90(6):855-861.

Bellinger, D., Sloman, J., Leviton, A., Rabinowitz, M., Needleman H.L., and Waternaux, C.
     (1991) "Low-Level Lead Exposure and Children's Cognitive Function in the Preschool
     Years."  Pediatrics. 87(2):219-227.

Bellinger, D.C., Leviton, A., Waternaux, C., Needleman, H., and Rabinowitz, M. (1987)
     "Longitudinal Analyses of Prenatal and Postnatal Lead Exposure and Early Cognitive
     Development."  New England Journal of Medicine. 316:1037-1043.

Bellinger, D.C., Leviton, A., Needleman, H.L., Waternaux, C., and Rabinowitz, M. (1986a)
     "Low-Level Lead Exposure and Infant Development in the First Year." Neurobehavioral
     Toxicology and Teratology. 8:151-161.

Bellinger, D.C., Leviton, A., Rabinowitz, M., Needleman, H., and Waternaux, C. (1986b)
     "Correlates of Low-Level Lead Exposure in Urban Children at 2 Years of Age."
     Pediatrics. 77:826-833.

Bellinger, D.C., Leviton, A., Waternaux, C., et al. (1985a) "A Longitudinal Study of the
     Developmental Toxicity of Low-Level Lead Exposure in the Prenatal and Early Postnatal
     Periods." In: Lekkas TD, ed. International Conference on Heavy Metals in the
     Environment, Athens, Greece, September, Vol 1. Edinburgh, United Kingdom: CEP
     Consultants, Ltd, 32-34.

Bellinger, D.C., Leviton, A., Waternaux, C., and Allred, E. (1985b) "Methodological Issues in
     Modeling the Relationship Between Low-Level Lead Exposure and Infant Development:
     Examples from the Boston Lead Study." Environmental Research. 38:119-129.
                                         292

-------
Bellinger, D.C. and Needleman, H.L. (1983) "Lead and the Relationship Between Maternal and
     Child Intelligence." Journal of Pediatrics. 102:523-527.

Bhatia, MS., Rai, S., Singbhal, P.K., Nigam, V.R., Bohra, N., and Malik, S.C. (1988) "Pica:
     Prevalence and Etiology." Indian Pediatr. 25:1165-1170.

Bornschein, R., Clark, S., Pan, W., and Succop, P. (1990) "Midvale Community Lead Study."
     University of Cincinnati.

Bornschein, R. L., Succop, P., Krafft, K., Clark, S., Peace, B., and Hammond, P. B. (1986)
     "Exterior Surface Dust Lead, Interior House Dust Lead and Childhood Lead Exposure in an
     Urban Environment." Conference in Trace Metals in Environmental Health, Columbia,
     MO.

Bornschein, R.L., Succop, P., Dietrich, K.N., Clark, C.S., Que Hee, S., and Hammond, P.B.
     (1985a) 'The Influence of Social and Environmental Factors on Dust Lead, Hand Lead, and
     Blood Lead Levels in Young Children." Environmental Research. 38:108-118.

Bornschein, R. L., Hammond, P. B., Dietrich, K. N., Succop, P., Krafft, K., Clark, S., Berger, O.,
     Pearson, D., and Que Hee, S. (1985b) 'The Cincinnati Prospective Study of Low-Level
     Lead Exposure and Its Effects on Child Development: Protocol and Status Report."
     Environmental Research, 38:4-18.

Brunekreef, B., Veenstra, S.J., Biersteker, K., and Boleij, J.S.M. (1981) 'The Arnhem Lead
     Study: Lead Uptake by 1- to 3- Year-Old Children Living in the Vicinity of a Secondary
     Lead Smelter in Amhem, Netherlands." Environmental Research. 25:441-448.

Burger, J., Carruth-Hinchey, C, Ondroff, J., and McMahon M.  (1998) "Effects of Lead on
     Behavior, Growth, and Survival of Hatchling Slider Turtles." Journal of Toxicology and
     Environmental Health. 55:495-502.

Burns, J.M., Baghurst, P.A., Sawyer, M.G., et al. (1999) "Lifetime Low-Level Exposure to
     Environmental Lead and Children's Emotional and Behavioral Development: The Port
     Pine Cohort Study." American Journal of Epidemiology.  149(8):740-749.

Bushnell, P.J. and Bowman, R.E. (1979a) "Persistence of Impaired Reversal Learning in Young
     Monkeys Exposed to Low Levels of Dietary Lead." /. Toxicology and Environmental
     Health.  5: 1015-1023.

Bushnell, P.J., and Bowman, R.E.  (1979b) "Effects of Chronic Lead mgestion on Social
     Development in Infant Rhesus Monkeys." Neurobehavioral Toxicology. 1:207-219.

Bushnell, P.J., Bowman, R.E., Allen, J.R., and Marlar, R.J.  (1977) "Scotopic Vision Deficits in
     Young Monkeys Exposed to Lead." Science 196:333-335.

                                         293

-------
Butte-Silver Bow Department of Health, and Department of Environmental Health, University of
     Cincinnati. (1991) "The Butte-Silver Bow Environmental Health Lead Study." Draft Final
     Report.

Calabrese, E., Barnes, R., Stanek, EJ. m, Pastides, H., Gilbert, C.E., Veneman, P., Wang, X.,
     Lasztity, A., and Kostecki, P.T. (1989) "How Much Soil Do Young Children Ingest: An
     Epidemiological Study."  Regulatory Toxicology and Pharmacology. 10:123-137.

Calabrese, E.J., and Stanek, EJ. (1993) "Soil Pica: Not a Rare Event." Journal of
     Environmental Sci. Health, A28(2):373-384.

Calabrese, E., Stanek, EJ., and Barnes, R. (1997)  "Soil Ingestion Rates in Children Identified
     by Parental Observation as Likely High Soil Ingesters." Journal of Soil Contamination.
     6(3):271-279.

Calabrese, E., Stanek, EJ., and Gilbert, C.E. (1991) "Evidence of Soil Pica Behavior and
     Quantification of Soil Ingested."  Human and Experimental Toxicology.  10:245-249.

Calabrese, E., Stanek, EJ., and Gilbert, C.E. (1993) "Lead Exposures in a Soil Pica Child."
     J. Environ. Sci. Health. A28 (2):353-362.

Carroll, R.J., and Galindo, C.D. (1999) "Measurement Error, Biases, and the Validation of
     Complex Models for Blood Lead Levels in Children." Environmental Health Perspectives.
     106 (Supp. 6): 1535-1539.

Carroll R.J., Ruppert, D., and Stefanski. L.A. (1995) Measurement Error in Nonlinear Models.
     London: Chapman & Hall.

CDC (1997) "Update: Blood-Lead Levels - United States, 1991-1994." Morbidity and Mortality
     Weekly Report. Centers for Disease Control and Prevention, Public Health Service, U.S.
     Department of Health and Human Services. 21 February 1997. 46(7): 141-146.

CDC (1991) "Preventing Lead Poisoning in Young Children: A Statement by the Centers for
     Disease Control."  Centers for Disease Control and Prevention, Public Health Service, U.S.
     Department of Health and Human Services. October 1991.

Chen, H., Ma, T., Hume, A. S., and Ho, I. K. (1998) "Developmental Lead Exposure Alters the
     Distribution of Protein Kinase C Activity in the Rat Hippocampus."  Biomedical and
     Environmental Sciences.  11:61-69.

Clark, C.S., Bomschein, R.L., Succop, P., Que Hee, S.S., Hammond, P.B., and Peace, B. (1985)
     "Condition and Type of Housing as an Indicator of Potential Environmental Lead Exposure
     and Pediatric Blood Lead Levels." Environmental Research. 38:46-53.

                                         294

-------
Clark, S., Bornschein, R.L., Succop, P., Roda, S., and Peace, B.  (1991) "Urban Lead Exposures
     of Children in Cincinnati, Ohio." Chemical Speciation and Bioavailability. 3:163-171.

Cohen, C.J., Bowers, G.N., and Lepow, MX. (1973) "Epidemiology of Lead Poisoning: A
     Comparison Between Urban and Rural Children." Journal of the American Medical
     Association. 226:1430-1433.

Cory-Slechta, D.A. (1997) "Relationships Between Pb-Induced Changes in Neurotransmitter
     System Function and Behavioral Toxicity." Neurotoxicology. 18(3): 673-688.

Cory-Slechta, D.A., Weiss, B., and Cox, C. (1985) "Performance and Exposure Indices of Rate
     Exposed to Low Concentrations of Lead." Toxicology and Applied Pharmacology.
     78:291-299.

Cory-Slechta, D.A. and Thompson, T. (1979) "Behavioral Toxicity of Chronic Post-weaning
     Lead Exposure in the Rat." Toxicology and Applied Pharmacology. 47:151-159.

Cotter-Howells, J. (1994) "Lead Phosphate Formation in Soils." Environmental Geochemistry
     and Health. 16(2): 86.

Cutler, M.G. (1977) "Effects of Exposure to Lead on Social Behavior in the Laboratory Mouse."
     Psychopharmacology. 52:279-282.

Davies, DJ.A, Thornton, I., Watt, J.M., Culbard, E.B., Harvey, P.G., Delves, H.T., Sherlock,
     J.C., Smart, G.A., Thomas, J.F.A., and Quinn, M.J. (1990) "Lead Intake and Blood Lead in
     Two-Year-Old U.K. Urban Children."  The Science of the Total Environment. 90:13-29.

Davis, A., Ruby, M.V. and Bergstrom, P.D. (1995) "Factors Controlling Lead Bioavailability in
     the Butte Mining District, Montana, USA." Environmental Geochemistry and Health.
     16(3-4): 147-157.

Davis, J.M., Otto, D.A., Weil, D.E., and Grant, L.D.  (1990) "The Comparative Developmental
     Neurotoxicity of Lead in Humans and Animals." Neurotoxicology and Teratology.  12:
     215-229.

Davis, S., Waller, P., Buschom, R., Ballou, J., and White, P. (1990) "Quantitative Estimates of
     Soil Digestion in Normal Children Between the Ages of 2 and 7 Years: Population-Based
     Estimate Using Aluminum, Silicon, and Titanium as Soil Tracer Elements." Archives of
     Environmental Health. 45:112-122.

Day, J.C.  (1993) Population Projections of the United States, by Age, Sex, Race, and Hispanic
     Origin: 2993 to 2050. U.S. Bureau of the Census, Current Population Reports P25-1104.
                                         295

-------
Dersimonian, R. and Laird, N. (1986) "Mela-Analysis in Clinical Trials." Controlled Clinical
     Trials. 7:177-188.

Diemel, J.A., Bnmekreef, B., Boleij, J.S., Biersteker, K. and Veenstra, SJ. (1981) "The Arnhem
     Lead Study H Indoor Pollution, and Indoor/Outdoor Relationships." Environmental
     Research. 25:449-456.

Dieter, M.P, Matthews, H.B., Jeffcoat, R.A. and Moseman, R.F. (1993) "Comparison of Lead
     Bioavailability in F344 Rats Fed Lead Acetate, Lead Oxide, Lead Sulfide or Lead Ore
     Concentrate from Skagway, Alaska." Journal of Toxicology and Environmental Health.
     39(1): 79-93.

Dietrich, K.N., Berger, O.G., Succop, P.A., Hammond, P.B., and Bornschein, R.L.. (1993) "The
     Developmental Consequences of Low To Moderate Prenatal and Postnatal Lead Exposure:
     Intellectual Attainment in the Cincinnati Lead Study Cohort Following School Entry."
     Neurotoxicology and Teratology. 15:37-44.

Duggan, M.J. and Inskip, M.J.  (1985) "Childhood Exposure to Lead in Surface Dust and Soil: A
     Community Health Problem."  Public Health Reviews.  13: 1-54.

Elhelu, M. A., Caldwell, D., and Hirpassa, W. (1995) "Lead in Inner-City Soil and Its Possible
     Contribution to Children's Blood Lead." Archives of Environmental Health.  50(2): 165-
     169.

Emond, M.J., Lanphear, B.P., Watts, A., Eberly, S., and Members of the Rochester Lead-in-Dust
     Study Group. (1997)  "Measurement Error and Its Impact on the Estimated Relationship
     Between Dust Lead and Children's Blood Lead."  Environmental Research 72:82-92

Emhart, C.B., Morrow-Tlucak, M., Worf, A.W., Super, D., and Drotar, D. (1989) "Low Level
     Lead Exposure in the Prenatal and Early Preschool Periods; Intelligence Prior to School
     Entry." Neurotoxicology and Teratology. 11:161-70.

Farfel, M.R. and Chisolm, J.J., Jr. (1991) "An Evaluation of Experimental Practices for
     Abatement of Residential Lead-Based Paint: Report on a Pilot Project." Environmental
     Research. 55:199-212.

Farfel, M.R. and Chisolm, J.J., Jr. (1990) "Health and Environmental Outcomes of Traditional
     and Modified Practices for Abatement of Residential Lead-Based Paint." American
     Journal of Public Health. 80(10): 1240-1245.

Fleiss, J.L.  (1981) Statistical Methods for Rates and Proportions. Second Edition.  New York:
     John Wiley & Sons. Section 1.2, pp. 4-9.
                                         296

-------
Fox, D.A., Campbell, M.L., and Blocker, Y.S. (1997) "Functional Alterations and Apoptotic
     Cell Death in the Retina Following Developmental or Adult Lead Exposure."
     Neurotoxicology 18(3):645-664.

Francek, M.A., Makimaa, B., Pan, V., and Hanko, J.H. (1994) "Small Town Lead Levels: A Case
     Study from the Homes of Pre-schoolers in Mt. Pleasant, Michigan." Environmental
     Research.  84(2): 159-166.

Freeman, G.B., Johnson, J.D., Killinger, J.M., Liao, S.C., Feder, P.I., Davis, A.O., Ruby, M.V.,
     Channey, R.L., Lovre, S.C., and Bergstrom, P.D. (1992) "Relative Bioavailability of Lead
     from Mining Waste Soil in Rats." Fund. Appl. ToxicoL  19:399-398.

Fulton, M., Raab, G., Thomson, G., Laxen, D., Hunter, R., and Hepburn, W. (1987) "Influence of
     Blood Lead on the Ability and Attainment of Children in Edinburgh."  Lancet. 1:1221-
     1226.

Galke, W. A., Hammer, D.I., and Lawrence, S. W. (1975) "Environmental Determinants of Lead
     Burdens in Children." In: International Conference on Heavy Metals in the Environment-
     Symposium Proceedings, T. C. Hutchinson, S. Epstein, A. L Page, J. VanLoon and T.
     Davey (eds.), Institute for Environmental Studies, Toronto, ON, Canada, 3:53-74.

Gallacher, J.E., Elwood, P.C., Phillips, K.M., Davies, B.E., and Jones, D.T. (1984) "Relation
     Between Pica and Blood Lead hi Areas of Differing Lead Exposure." Archives of Diseases
     in Childhood. 59(1): 4(M4.

Geist, C.R., Balko, S.W., Morgan, M.E., and Angiak, R. (1985) "Behavioral Effects Following
     Rehabilitation from Postnatal Exposure to Lead Acetate." Percept. Mot. Skills. 60:527-
     536.

Gilbert, S.G., and Rice, D.C. (1987) "Low-level Lifetime Lead Exposure Produces Behavioral
     Toxicity (Spatial Discrimination Reversal) in Adult Monkeys." Toxicology and Applied
     Pharmacology. 91:484-490.

Goldstein, G. (1992) "Developmental Neurobiology of Lead Toxicity." In Human Lead
     Exposure (Needleman, H., ed.). 125-135.

Goldstein, G. (1990) "Lead Poisoning and Brain Cell Function." Environ. Health Perspect.
     89:91-94.

Greene, T., Ernhart, C.B., and Boyd, T.A. (1992) "Contributions of Risk Factors to Elevated
     Blood and Dentine Lead Levels in Preschool Children."  The Science of the Total
     Environment. 115:239-260.
                                         297

-------
Harrison, G. (1987) "A Survey of the Lead Distribution in the Soil of Corpus Christi, Texas."
     Texas Journal of Science.  39(1): 15-22.

Harvey, P.O., Hamlin, M.W., Kumar, R., Morgan, G., Spurgeon, A., and Delves, H.T. (1988)
     "Relationship Between Blood Lead, Behaviour, Psychometric and Neuropsychological Test
     Performance in Young Children." British Journal of Developmental Psychology. 29:43-52.

Hastings, L., Cooper, G.P., Bomschein, R.L., and Michaelson, LA,  (1977) "Behavioral Effects
     of Low Lead Neonatal Lead Exposure." Pharmacology Biochemistry & Behavior. 7: 37-
     42.

Hatzakis, A., Kolevi, A., Katsouyanni, K., et al. (1987) "Psychometric Intelligence and
     Attentional Performance Deficits in Lead Exposed Children." International Conference on
     Heavy Metals in the Environment. Edinburgh, Scotland; CEP Consultants. 16:57-67.

Hatzakis, A., Kokkeni, A., Maranelias, C., Katsouyanni, K., Salaminios, F., Kalandidi, A., et al.
     (1989) "Psychometric Intelligence Deficits in Lead-exposed Children." In: Lead Exposure
     and Child Development (Smith MA., Grant, LD., Sora Al, eds.). London: Kluwer,
     1989:211-23.

Hawk, B.A., Schroeder, S.R., Robinson, G., et al. (1986) "Relation of Lead and Social Factors to
     IQ of Low-SES Children:  A Partial Replication." American Journal of Mental Deficiency.
     91:178-183.

Karam, H.S., and Beck, B.D. (1990) "Current Issues in Determining Acceptable Lead
     Concentrations in Soils." Comments on Toxicology. 3(6):509-529.

Kaufman, A.S. (1996) "IQ, Lead Level, and Inferences from Research Studies: Comments
     Addressing the Underlying Science Forming the Basis of HUD's Regulatory Impact
     Analysis." Submitted to the Environmental Health Committee of the Science Advisory
     Board, U.S. Environmental Protection Agency, in their technical review of the proposed
     TSCA Section 403 regulations (Identification of Dangerous Levels of Lead). August 1996.

Kenny, D.A. (1979) Correlation and Causality.  New York: John Wiley & Sons, Inc. Chapter 1,
     pp. 1-12.

Krueger, J. A. and Duguay, K. M. (1989) "Comparative Analysis of Lead in Maine Urban Soils."
     Bulletin of Environmental Contamination and Toxicology. 42:574-581.

Kuhlmann, A.C., McGlothan, J.L., and Guilarte, T.R. (1997) "Developmental Lead Exposure
     Causes Spatial Learning Deficits in Adult Rats."  Neuroscience Newsletters.  233:101-104.

Lacey, E.P. (1990) "Broadening the Perspective of Pica: Literature Review." Public Health
     Reports. 105(1 ):29-35.

                                         298


-------
Lanphear, B.P., Matte, T.D., Rogers, J., Clickner, R.P., Dietz, B., Bornschein, R.L., Succop, P.,
     Mahaffey, K.R., Dixon, S., Galke, W., Rabinowitz, M., Farfel, M., Rohde, C, Schwartz, J.,
     Ashley, P., and Jacobs, D. (1998) 'The Contribution of Lead-Contaminated House Dust
     and Residential Soil to Children's Blood Lead Levels:  A Pooled Analysis of 12
     Epidemiologic Studies."  Environmental Research (Section A). 79:51-68.

Lanphear, B.P., Weitzman, M., and Eberly, S. (1996a) "Racial Differences in Urban Children's
     Environmental Exposures to Lead." American Journal of Public Health. 86(10): 1460-1463.

Lanphear, B.P., Weitzman, M., Winter, N.L., Eberly, S., Yakir, B., Tanner, M., Emond, M., and
     Matte, T.D. (1996b) "Lead-Contaminated House Dust and Urban Children's Blood Lead
     Levels." American Journal of Public Health.  86(10): 1416-1421.

Lansdown, R., Yule, W., Urbanowicz, M.A., et al. (1986) "The Relationship Between Blood
     Lead Concentrations, Intelligence, Attainment and Behavior hi a School Population: The
     Second London Study."  International Archives of Occupational and Environmental
     Health. 57:225-235.

Laperche, V., Logan, T.J., Gaddam, P., and Traina, S.J. (1997) "Effect of Apatite Amendments
     on Plant Uptake of Lead from Contaminated Soil." Environmental Science and
     Technology. 31(10): 2745-2753.

Lasky, R.E., Maier, M.M., Snodgrass, E.B., Hecox, K.E., and Laughlin, N.K. (1995) "The
     Effects of Lead on Otoacoustic Emissions and Auditory Evoked Potentials in Monkeys."
     Neurotoxicology and Teratology 17(6):633-644.

Lilienthal, H., Kohler, K., Turfeld, M., and Winneke, G.  (1994) "Persistent Increases in Scotopic
     B-wave Amplitudes After Lead Exposure in Monkeys." Exp. Eye Res.  59:203-209.

Lilienthal, H., Winneke, G., and Ewert, T.  (1990) "Effects of Lead on Neurophysiological and
     Performance Measures: Animal and Human Data." Environmental Health Perspectives.
     89:21-25.

Lyngbye, T., Hansen, O.N., and Grandjean, P. (1990) "Predictors of Tooth Lead Level with
     Special Reference to Traffic: A Study of Lead-Exposure in Children." International
     Archives of Occupational and Environmental Health. 62(6):417-422.

Marais, M.L., and Wecker, W.E. (1998) "Correcting for Omitted Variables and Measurement
     Error Bias in Regression  with an  Application to the Effect of Lead on IQ." Journal of the
     American Statistical Association. 93:494-505.

McKinney, J. (1993) "Metals Bioavailability and Disposition Kinetics Research Needs
     Workshop, July 18-19, 1990."  Toxicological and Environmental Chemistry.  38: 1-71.
                                         299

-------
Mello, C.F., Kraemer, C.K., Filippin, A., Morsch, V.M., Rodrigues, A.L., Martins, A.F., and
     Rubin, M.A. (1998) "Effect of Lead Acetate on Neurobehavioral Development of Rats."
     Brazilian Journal of Medical and Biological Research. 31(7): 943-950.

Mielke, H. W. (1995) "Lead in New Orleans Soils: New Images of an Urban Environment."
     Environmental Geochemistry and Health. 16(3/4): 123-128.

Mielke, H. W. (1993) "Lead Dust Contaminated U.S.A. Communities: Comparison of Louisiana
     and Minnesota." Applied Geochemistry. Supp. Issue No. 2,257-261.

Mielke, H. W., Adams, J. E., Huff, B., Pepersack, J., Reagan, P. L., Stoppel, D., and Mielke, P.
     W. (1992) "Dust Control as a Means of Reducing Inner-City Childhood Pb Exposure." In:
     Trace Substances in Environmental Health - XXV. Supplement to Volume 14,
     Environmental Geochemistry and Health (B. D. Beck, ed.) 121-128.

Mielke, H. W., Adams, J. L., Reagan, P. L., and Mielke, P. W. (1989) "Soil-Dust Lead and
     Childhood Lead Exposure as a Function of City Size and Community Traffic Flow: The
     Case for Lead Abatement in Minnesota." In: Lead in Soil: Issues and Guidelines,
     Supplement to Volume 9 of Environmental Geochemistry and Health (B. E. Davis and B. G.
     Wixson, eds.). 253-271.

Mielke, H. W., Anderson, J.C., Berry, K. J., Mielke, P. W., Chaney, R. L., and Leech, M. (1983)
     "Lead Concentrations in Inner-City Soils as a Factor in the Child Lead Problem." American
     Journal of Public Health.  73(12): 1366-1369.

Nagymajtenyi, L., Desi, L, Schulz, H., and Papp, A. (1998) "Consequences of Lead Exposure of
     Rats During Pregnancy, Lactation, and Postweaning. A Combined Behavioural and
     Neurotoxicological Study." International Journal of Environmental Health Research.
     8(2): 121-135.

Needleman, H.L. (1998) "IQ, Lead Level, and Inferences from Research Studies." Comments to
     the Environmental Health Committee of the Science Advisory Board, U.S. Environmental
     Protection Agency, in their technical review of the proposed TSC A Section 403 regulations
     (Identification of Dangerous Levels of Lead). 2 September  1998.

Needleman, H.L., and Gatsonis, C.A. (1990) "Low-Level Lead Exposure and the IQ of
     Children." Journal of the American Medical Association. 263(5):673-678.

Otto, D.A., and Fox, D.A. (1993) "Auditory and  Visual Dysfunction Following Lead Exposure."
     Neurotoxicology. 14(2-3): 191-208.

Petit, T.L., Alfano, D.P., and LeBoutillier, J.C. (1983) "Early Lead Exposure and the
     Hippocampus: a Review and Recent Advances." Neurotoxicology. 4(l):79-94.
                                         300

-------
Pendleton, S. (personal communication to Brad Schultz, U.S. EPA).

Pocock, S.J., Smith, M., and Baghurst, P. (1994) "Environmental Lead and Children's
     Intelligence: A Systematic Review of the Epidemiological Evidence." BMJ. 309:1189-
     1197.

Que Hee, S., Peace, B., Clark, S., Boyle, J. R., Bomschein, R. L., and Hammond, P. B. (1985)
     "Evolution of Efficient Methods to Sample Lead Sources, Such as House Dust and Hand
     Dust, in the Homes of Children." Environmental Research. 38:77-95.

Rabinowitz, M.B., Wang, J.D., and Soong, W.T. (1992) "Apparent Threshold of Lead's Effect
     on Child Intelligence." Bulletin of Environmental Contamination and Toxicology. 48:688-
     695.

Rabinowitz, M., Leviton, A., Needleman, H., Bellinger, D., and Waternaux, C. (1985)
     "Environmental Correlates of Infant Blood Lead Levels in Boston." Environmental
     Research. 38:96-107.

Rice, D. C. (1996) "Behavioral Effects of Lead: Commonalities Between Experimental and
     Epidemiologic Data." Environmental Health Perspectives. 104 (Suppl. 2): 337-351.

Rice, D.C.  (1988) "Chronic Low-Lead Exposure Does Not Affect Simple Reaction Time."
     Neurotoxicology. 9:105-108.

Rice, D.C.  (1985) "Chronic Low-Lead Exposure from Birth Produces Deficits in Discrimination
     Reversal in Monkeys." Toxicology and Applied Pharmacology. 77:201-210.

Rice, D.C. and Gilbert, S.G, (1990a) "Sensitive Periods for Lead-Induced Behavioral
     Impairment (Nonspatial Discrimination Reversal) in Monkeys." Toxicology and Applied
     Pharmacology.  102:101-109.

Rice, D. C. and Gilbert, S.G. (1990b) "Lack of Sensitive Period for Lead-Induced Behavioral
     Impairment on a Spatial Delayed Alternation Task in Monkeys."  Toxicology and Applied
     Pharmacology.  103:364-373.

Rieuwerts, J.S., and Farago, M.E.  (1995) "Lead Contamination in Smelting and Mining
     Environments and Variations in Chemical Forms and Bioavailability." Chemical
     Speciation and Bioavailability. 7(4): 113-123.

Roberts, J.W., Han, W., and Ruby, M.G. (1996) "Project Summary: Evaluation of Dust
     Samplers for Bare Floors and Upholstery."  National Exposure Laboratory, Office of
     Research and Development, U.S. Environmental Protection Agency. EPA/600/R-96-001.
     January 1996.
                                        301

-------
Roels, H.A., Buchet, J.P., Lauwerys, R.R., Bruaux, P., Claeys-Thoreau, F., LaFontaine, A., and
     Verduyn, G. (1980) "Exposure to Lead by the Oral and the Pulmonary Routes of Children
     Living in the Vicinity of a Primary Lead Smelter." Environmental Research. 22:81-94.

Romieu, I., Carreon, T., Lopez, L., Palazuelos, E., Rios, C., Manuel, Y., and Hemandez-Avila,
     M. (1995)  "Environmental Urban Lead Exposure and Blood Lead Levels in Children of
     Mexico City." Environmental Health Perspectives.  103(11): 1036-1040.

Schmitt, M.D.C., Trippler, D.J., Wachtler, J.N. and Lund, G.V. (1988) "Soil Lead
     Concentrations in Residential Minnesota as Measured by ICP-AES." Water, Air, and Soil
     Pollution. 39:157-168.

Schroeder, S.R., Hawk, B., Otto, D., Mushak, P., and Hicks, R.E. (1985) "Separating the Effects
     of Lead and Social Factors on IQ." Environmental Research. 91:178-183.

Schwartz, J. (1994) "Low-Level Lead Exposure and Children's IQ: A Meta-Analysis and Search
     for a Threshold." Environmental Research. 65:42-55.

Schwartz, J. (1993) "Beyond LOEL's, p Values, and Vote Counting:  Methods for Looking at the
     Shapes and Strengths of Association." Neurotoxicology and Teratology. 14(2-3):237-246.

Sedman, R.M. (1989)  "The Development of Applied Action Levels for Soils Contact: A
     Scenario for the Exposure of Humans to Soils in a Residential Setting." Environmental
     Health Perspectives. 79:291-313.

Sedman, R.M., and Mahmood, RJ. (1994) "Soil Ingestion by Children and Adults Reconsidered
     Using the Results of Recent Tracer Studies." Journal of Air & Waste Management
     Association. 44:141-144.

Shellshear, I.D., Jordon, L.D., Hogan, D.J., and Shannon, F.T. (1975)  "Environmental Lead
     Exposure in Christchurch Children:  Soil Lead as a Potential Hazard." New Zealand Med.
     J. 81:382-386.

Sheppard, S.C. (1998) "Geophagy: Who Eats Soils and Where Do Possible Contaminants Go?"
     Environmental Geology (Berlin).  33(2-3): 109-114.

Silbergeld, E.K. (1991) "Aneurological Perspective on Lead Toxicity." In Human Lead
     Exposure (Needleman, H.L. ed.). Boca Raton, Florida: CRC Press, 89-103.

Silva, P.A., Hughes, P., Williams, S., and Faed, J.M. (1988) "Blood Lead, Intelligence, Reading
     Attainment, and Behaviour in Eleven Year Old Children in Dunedin, New Zealand."
     Journal of Child Psychology and Psychiatry. 29:43-52.
                                         302

-------
Simon, S.L. (1998) "Soil Ingestion by Humans:  A Review of History, Data, and Etiology with
     Application to Risk Assessment of Radioactively Contaminated Soils." Health Physics,
     74(6):647-672.

Stanek, E.J. m, and Calabrese, EJ. (1995) Daily Estimates of Soil Ingestion in Children.
     Environmental Health Perspectives. 103:276-285.

Stanek E.J., Calabrese, E.J., Mundt, K., Pekow, P., and Yeatts, K.B. (1998)  "Prevalence of Soil
     Mouthing/Ingestion Among Healthy Children Aged 1 to 6." Journal of Soil
     Contamination. 7(2):227-242.

Stark, A. D., Quah, R. F., Meigs, J. W., and DeLouise, E. R. (1982) "The Relationship of
     Environmental Lead to Blood-Lead Levels in Children." Environmental Research. 27:372-
     383.

Steuteville, W.D. (1990) "Interior Remediation of Residences:  Lead Abatement (A Case
     History of a Superfund Project and Public Involvement)." In: Procedures of the National
     Conference of Hazardous Wastes and Hazardous Materials. Silver Springs, MD:
     Hazardous Materials Research Control Institute.

Sutton, P. Athanasoulis, M., Ressel P., Guirguis, G., Haan, M., Schlag, R., and Goldman, L.
     (1995) "Lead Levels in the Household Environment of Children in Three High-Risk
     Communities in California." Environmental Research. 68:45-57.

Teichmean, J., Coltrin, D., Prouty, K., and Bir, W. A. (1993) "A Survey of Lead Contamination
     in Soil Along Interstate 880, Alameda County, California." Journal of the American
     Industrial Hygiene Association. 54(9): 557-559.

Tong, S.  (1998) "Lead Exposure and Cognitive Development: Persistence and a Dynamic
     Pattern." J. Pediatr. Child Health. 34:114-118.

Tong, S.  T. (1990) "Roadside Dusts and Soils Contamination in Cincinnati, Ohio."
     Environmental Management.  14(1): 107-114.

Tong, S., Baghurst, P., McMichael,  A., Sawyer, M., and Mudge, J. (1996) "Lifetime Exposure to
     Environmental Lead and Children's Intelligence at 11-13 Years: the Port Pirie Cohort
     Study." British Medical Journal. 312:1569-1575.

U.S. Census Bureau (1999) Current Housing Reports, Series HI50/97, American Housing
     Survey for the United States: 1997 (Table 2-1). U.S. Government Printing Office,
     Washington, DC 20222.
                                          303

-------
U.S. Department of Health and Human Services (1999) Toxicological Profile for Lead.  Update,
     1999. Atlanta: Agency for Toxic Substances and Disease Registry, Division of
     Toxicology/Toxicology Information Branch, U.S. Department of Health and Human
     Services.

U.S. Department of Health and Human Services (1992) Impact of Lead-Contaminated Soil on
     Public Health.  Atlanta: Agency for Toxic Substances and Disease Registry, Division of
     Toxicology/Toxicology Information Branch, U.S. Department of Health and Human
     Services. May 1,1992.

U.S. Department of Health and Human Services (1991) "Philadelphia Neighborhood Lead
     Study." Final Report. Atlanta: Agency for Toxic Substances and Disease Registry,
     Division of Toxicology/Toxicology Information Branch, U.S. Department of Health and
     Human Services.

U.S. Department of Housing and Urban Development (1998) "Evaluation of the HUD Lead-
     Based Paint Hazard Control Grant Program: Fifth Interim Report, Progress as of September
     1,1997." Prepared by the National Center for Lead Safe Housing and the University of
     Cincinnati Department of Environmental Health for the U.S. Department of Housing and
     Urban Development. March 1998.

U.S. Department of Housing and Urban Development (1995a) "The Relation of Lead-
     Contaminated House Dust and Blood Lead Levels Among Urban Children."  Volumes I
     and H. Final Report to the U.S. Department of Housing and Urban Development Grant
     from the Departments of Pediatrics, Biostatistics, and Environmental Medicine, The
     Rochester School of Medicine, Rochester, New York, and The National Center for Lead-
     Safe Housing, Columbia Maryland.

U.S. Department of Housing and Urban Development (1995b) "Guidelines for the Evaluation
     and Control of Lead-Based Paint Hazards in Housing." Office of Lead-Based Paint
     Abatement and Poisoning Prevention.

U.S. Department of Housing and Urban Development (1991) "The HUD Lead-Based Paint
     Abatement Demonstration (FHA)."  Final Report, Office of Policy Development and
     Research, U.S. Department of Housing and Urban Development.

U.S. Environmental Protection Agency (1998a) "Risk Analysis to Support Standards for Lead in
     Paint, Dust, and Soil."  Office of Pollution Prevention and Toxics, U.S. Environmental
     Protection Agency. EPA 747-R-97-006.

U.S. Environmental Protection Agency (1998b) "An SAB Report: Technical Review of the
     Proposed TSCA Section 403 Regulation (Identification of Dangerous Levels of Lead)."
     Environmental Health Committee, Science Advisory Board, U.S. Environmental Protection
     Agency. EPA-SAB-EHC-99-004, November 1998.

                                        304

-------
U.S. Environmental Protection Agency (1998c) "Analysis of Pathways of Residential Lead
     Exposure in Children." Office of Prevention Pesticides, and Toxic Substances. EPA 747-
     R-98-007. 27 April 1998 (peer review version).

U.S. Environmental Protection Agency (1997a)  "Lead Exposure Associated with Renovation
     and Remodeling Activities: Environmental Field Sampling Study.  Volume I:  Technical
     Report." Office of Pollution Prevention and Toxics, U.S. Environmental Protection
     Agency. EPA 747-R-96-007.

U.S. Environmental Protection Agency (1997b) "Summary and Assessment of Published
     Information on Determining Lead Exposures and Mitigating Lead Hazards Associated with
     Dust and Soil in Residential Carpets, Furniture, and Forced Air Ducts." Office of
     Prevention Pesticides, and Toxic Substances, U.S. Environmental Protection Agency. EPA
     747-S-97-001.

U.S. Environmental Protection Agency (1997c) "Lead-Based Paint Abatement and Repair and
     Maintenance Study in Baltimore: Finding Based on Two Years of Follow-Up." Office of
     Prevention Pesticides, and Toxic Substances, U.S. Environmental Protection Agency. EPA
     747-R-97-005.

U.S. Environmental Protection Agency (1996a) "Urban Soil Lead Abatement Demonstration
     Project, Volume I: Integrated Report." Office of Research and Development, U.S.
     Environmental Protection Agency. EPA 600-P-93-001aF, April 1996.

U.S. Environmental Protection Agency (1996b)  "Comprehensive Abatement Performance
     Study. Volume I: Summary Report." Office of Prevention, Pesticides, and Toxic
     Substances, U.S. Environmental Protection Agency. EPA 230-R-94-013a.

U. S. Environmental Protection Agency (1996c) "Lead-Based Paint Abatement and Repair and
     Maintenance Study in Baltimore: Pre-Intervention Findings." EPA 747-R-95-012, August
     1996.

U.S. Environmental Protection Agency (1995) "Report on the National Survey of Lead-Based
     Paint in Housing. Base Report." Office of Pollution Prevention and Toxics, U.S.
     Environmental Protection Agency. EPA747-R95-003.

U.S. Environmental Protection Agency (1994) "Guidance Manual for the Integrated Exposure
     Uptake Biokinetic Model for Lead in Children." Office of Emergency and Remedial
     Response, U.S.  Environmental Protection Agency. EPA/540/R-93/081, PB93-963510.

U.S. Environmental Protection Agency (1986) "Air Quality Criteria for Lead, Volume IV of IV."
     Research Triangle Park NC: Environmental Criteria and Assessment Office, U.S.
     Environmental Protection Agency. EPA-600/8-83/028dF. June  1986.
                                        305

-------
Viverette, L., Mielke, H.W., Brisco, M, Dixon, A., Schaefer, J., and Pierre, K. (1996)
     "Environmental Health in Minority and Other Underserved Populations: Benign Methods
     for Identifying Lead Hazards at Day Care Centres of New Orleans." Environmental
     Geochemistry and Health. 18(1): 41-45.

Walsh, T.J., and Tilson, H.A. (1984) "Neurobehavioral Toxicology of the Organoleads."
     Neurotoxicology. 5:67-86.

Wang, L., Xu S.E., Zhang, G.D, and Wang, W.Y. (1989) "Study of Lead Absorption and its
     Effect on Children's Development."  Biomedical and Environmental Sciences. 2:325-330.

Weitzman, M., Aschengrau, A., Bellinger, D., Jones, R., Hamlin, J.S., and Beiser, A. (1993)
     "Lead-Contaminated Soil Abatement and Urban Children's Blood Lead Levels." Journal of
     the American Medical Association. 269(13): 1647-1654.

Westat (1998) "A Report on the Analysis of the Blood Lead-Environmental Lead Relationship."
     Report for the U.S. Department of Housing and Urban Development. 7 August 1998.

Winneke, G., Lilienthal, H., and Kramer, U. (1996) "The Neurobehavioural Toxicology and
     Teratology of Lead." Arch Toxicol Suppl. 18:57-70.

Winneke, G., Brockhaus, A., Ewers, V., Kramer, V., and Neuf, M. (1990) "Results from the
     European Multicenter Study on Lead Neurotoxicity in Gulden: Implications for Risk
     Assessment."  Neurotoxicol Teratol 12:553-559.

Winneke, G., Beginn, U., Ewert, T., et al. (1985) "Comparing the Effects of Perinatal and Later
     Childhood Lead Exposure on Neurophysiological Outcome."  Environmental Research.
     38:155-167.

Yule, W., Lansdown, R., Millar, L, and Urbanowiez, M. (1981) "The Relationship Between
     Blood Lead Concentration, Intelligence, and Attainment in a School Population: a Pilot
     Study." DevMed Child Neuro. 23:567-576.
                                         306

-------
50272-101
         REPORT DOCUMENTATION
                   PAGE
1. REPORT NO.
   EPA 747-R-00-004
3. Recipient's Accession No.
 4. Title and Subtitle

 Risk Analysis to Support Standards for Lead in Paint, Dust, and Soil: Supplemental Report
                                                   5. Report Date
                                                     December 2000
                                                                                               6.
  7.  Author(s) Lordo, RA, Burgoon, DA, Ma, ZJ, Matthews, MC, Menkedick, JG, Naber, S, Nahhas,
              R, Neighbor, M, Niemuth, NA, Wood, BJ, Wooton, M
                                                   8.  Performing Organization
                                                      Rept. No.
  9.    Performing Organization Name and Address

       Battelle Memorial Institute
       505 King Avenue
       Columbus, Ohio 43201-2693
                                                   10. ProjecVTask/Work Unit No.
                                                        0476101-20
                                                   11. Contract(C) or Grant(G) No,

                                                   (C| 68-W-99-033
  12.  Sponsoring Organization Name and Address

       U.S. Environmental Protection Agency
       Office of Pollution Prevention and Toxics
       401 M Street, S.W.
       Washington, D.C.  20460
                                                   13.  Type of Report & Period
                                                       Covered
                                                       Technical Report
                                                   14.
  15.  Supplementary Notes

       Support staff involved in the risk analysis and the production of this report included Haisey Boyd, Ying-Liang Chou. and Kensuke
       Shirakawa.
  16.  Abstract (Limit 200 words)

  Title X of the Housing and Community Development Act, known as the Residential Lead-Based Paint Hazard Reduction Act of 1992,
  contains legislation designed to evaluate and reduce exposures to lead in paint, dust, and soil in the nation's housing.  As part of Title X,
  the Toxic Substances Control Act (TSCA) was amended to include Title IV, 'Lead Exposure Reduction'. Section 403 of TSCA requires
  EPA to define standards for lead in paint, dust, and soil. Federal, state, and local public health agencies, as well as private property
  owners and other private sector interests, will use these standards to determine in which homes actions should be taken to reduce or
  prevent the threat of childhood lead poisoning.

  This report is a supplement to EPA 747-R-97-006 (June 1998), which documented the outcome of a risk analysis that supported EPA's
  efforts to propose standards in response to Section 403. The additional literature reviews, data summaries, and data analyses presented
  in this report focus on selected topics raised  in comments on the risk analysis that were made by EPA's Science Advisory Board and the
  general public.  EPA has cited the findings of this report when preparing responses to these comments.
  17.  Document Analysis

       a.  Descriptors

          Blood-lead concentration, childhood health risks, adverse health effects, dust-lead loading, soil-lead concentration, lead hazards,
          lead-based paint, risk assessment, risk characterization, risk management

       b.  Identifiers/Open-Ended Terms

          abatement, intervention, hazard identification, exposure assessment, dose-response assessment, HUD National Survey,
          Rochester Lead-in-Dust Study, Evaluation of the HUD Lead-Based Paint Hazard Control Grant Program, NHANES III,
          bioavailability, modeling, performance characteristics analysis, sensitivity and uncertainty analysis. Section 403, Title X

       C.  COSATl Field/Group
18. Availability Statement
Release Unlimited
19. Security Class (This Report)
Unclassified
2O. Security Cfass (This Page)
Unclassified
21 . No. of Pa
340 (Vol.
ges
I); 232 (Vol. II)
22. Price
(See ANSI-239.18)
                                                        OPTIONAL FORM 272 (4-77)
                                                                  (Formerly NTIS-35)
                                                            Department of Commerce

-------

-------