xvEPA
unnea states
Environmental Protection
Agency
UTTICB or
Toxic Substances
Washington, DC 20460
DOU/ u-/a-
November 1980
Office of Toxic Substances
PCB Residue Levels
in Human Adipose Tissue,
A Statistical Evaluation
by Racial Grouping
-------
EPA 560/13-79-015
November, 1980
PCB Residue Levels in Human Adipose Tissue,
A Statistical Evaluation by Racial Grouping
by
Robert M. Lucas
Mitchell D. Erickson
Phil V. Piserchia
Stephen R. Williams
Research Triangle Institute
Research Triangle Park, North Carolina
27709
Contract No. 68-01-5848
Task Manager: .Cindy Stroup
Contract Project Officer: Joseph Carra
Design and Development Branch
Exposure Evaluation Division
Office of Pesticides and Toxic Substances
Washington, D.C. 20460
-------
DISCLAIMER NOTICE
This report was prepared under contract to an agency of the United
States Government. Neither the United States Government nor any of its
employees, contractors, subcontractors, or their employees, makes any
warranty, expressed or implied, nor assumes any legal liability or
responsibility for any third party's use or the results of such use of
any information, apparatus, product or process disclosed in this report,
nor represents that its use by such third party would not infringe
privately-owned rights.
Publication of the data in this document does not signify that the
contents necessarily reflect the joint or separate views and policies of
each sponsoring agency. Mention of trade names or commercial products
does not constitute endorsement or recommendation for use.
-------
ACKNOWLEDGEMENTS
The authors greatly appreciate the cooperation of Ms. Sandra Strassman-
Sundy of the U.S. Environmental Protection Agency, Washington, D.C., for
providing information on the sample design and data summaries; John D.
Tessari and Larry Griffin of Colorado State University, Fort Collins,
Colorado, for supplying information on the chemical analysis techniques;
Dr. R. G. Lewis and Wayne Sovocool and their colleagues of the Health
Effects Research Laboratory, U.S. Environmental Protection Agency,
Research Triangle Park, N.C.; and Dr. Henry Enos of the University of
Miami, Miami, Florida, for consultation on the chemical analysis. Also,
the authors greatly appreciate the helpful suggestions of Dr. Brenda Cox
in writing this report. Many thanks are also given to the secretarial
staff, Ms. Laurine Johnson, Ms. Pat Parker, Ms. Sarah Stumley,
Ms. Joanne Quate and Ms. Martha Clegg.
-(i)-
-------
TABLE OF CONTENTS
ACKNOWLEDGEMENTS i
LIST OF FIGURES iv
LIST OF TABLES v
1. INTRODUCTION 1
1.1. The National Human Monitoring Program 1
1.2. Study Methodology 2
2. CONCLUSIONS
2.1. Issue of Racial Differences 5
2.2. Statistical Design 6
2.3. Chemical Analysis 7
2.4. Recommendations • 7
2.4.1. Statistical Design 7
2.4.2. Chemical Analysis 8
3. STATISTICAL SURVEY DESIGN 11
3.1. Overview 11
3.2. Significance of Racial Differences Assuming
No Design Effect or Measurement Error 13
3.3. Comments on the Impact of Measurement Error 14
3.4. Impact of Sample Design 18
3.4.1. Restriction of Cities 18
3.4.2. Sampling Cadavers and Surgical Patients .... 20
3.4.3. Remarks on the Bias of Estimating
Racial Differences 23
4. EVALUATION OF CHEMICAL ANALYSIS AND DATA INTERPRETATION ... 24
4.1. Objective 24
4.2. Discussion 24
4.2.1. Qualitative Assessment 24
4.2.2. Quantitative Assessment 30
4.2.3. Potential Interference 32
4.2.4. Assessment of TLC Methodology 32
4.2.5. Correlation of Data Generated by
the Two Methods 34
-Cii)-
-------
TABLE OF CONTENTS (cont.)
Page
5. WEIGHTED ANALYSIS OF COMPUTER ACCESSIBLE DATA FILES 35
5.1. Introduction and Assumptions 35
5.2. Sample Weights 35
5.2.1. Calculatons of Weights 35
5.2.2. Sample Sizes used for Calculating Weights. . . 37
5.3. Racial Comparisons by Weighted Analysis 38
5.3.1. Overview 38
5.3.2. Statistical Method 39
5.4. Comments on Assumptions 48
REFERENCES 49
APPENDIX A: Letter from M. Aaronson A-l
APPENDIX B: Letter from John D. Tessari B-l
APPENDIX C: Interlaboratory PCB Analysis Results C-l
APPENDIX D: Breakdown of Census Regions and Divisions D-l
APPENDIX E: Survey and Site Quotas for Fiscal Years 1972-76 . . E-l
APPENDIX F: Guidelines and General Information About
Collecting Adipose Tissue for the National
Human Monitoring Program for Pesticides F-l
APPENDIX G: National Data Summaries for Fiscal Years
1972 to 1976 G-l
-(iii)-
-------
LIST OF TABLES
Table Page
3-1
3-2
3-3
3-4
3-5
5-1
5-2
5-3
Summary of Results for Testing the Null Hypothesis
that P -P ~ 0 for FY72-FY76
w n
Effect of Measurement Variance on the Significance
Level of Test Statistics
Effect of Measurement Variance on the Probability
of Detecting Real Differences
Probability of Detecting Real Differences for
Various Levels of Positive Bias
Probability of Detecting Real Differences for
Various Levels of Negative Bias
Model Coefficients (in Percent)
Tests for Differential Slopes — White Versus Nonwhite . .
Tests for Average Racial Grout) Differences
15
17
IP
?1
??
4?
45
45
-(v)-
-------
LIST OF FIGURES
Figure Page
3-1 Sampling Frame and Selection Procedures for Middle
Atlantic Census Division in FY73 12
4-1 Chromatogram of a Tissue Specimen on 4% SE30/6%
OV-210 Column 25
4-2 Chromatogram of a Tissue Specimen on 4% SE30/6%
OV-210 Column 26
4-3 Chromatogram of a Tissue Specimen on 4% SE30/6%
OV-210 Column 27
4-4 Chromatogram of Equivalent to 1.5 ppm Aroclor 1260
Standard 28
4-5 Electron Capture Gas Chromatograms of (A) Aroclor 1016
Standard and (B) PCB-residue 29
4-6 Computer Reconstructed Total Ion and Mass Chromatogram
of A Composite Human Adipose Tissue Extract 31
4-7 Chromatograms of Aroclors and Pesticide Standards
Illustrating Potential Interference 33
5-1 Racial Comparisons Over Time: Percent > 3 ppm PCB. ... 46
5-2 Racial Comparisons Over Time: Percent Positive
PCB Detections 47
-(iv)-
-------
1. INTRODUCTION
1.1. The National Human Monitoring Program
The National Human Monitoring Program (NHMP) was established in
1967 to monitor incidences of pesticide residues in the general United
States population and to assess changes and trends in these levels. The
program was initiated by the United States Public Health Service and was
transferred in 1970 to the newly created United States Environmental
Protection Agency.
The sample design of the National Human Adipose Tissue Survey, one
of the major ongoing programs operated by the NHMP, involves several
stages of selection. The conterminous 48 states were stratified into
several geographic regions. Within each stratum, cities with populations
greater than 25,000 were randomly selected for fiscal years 1972-76. In
fiscal year 1977, the first-stage units were changed to Standard
Metropolitan Statistical Areas (SMSA's). Hospitals were then selected
within each sample city or SMSA. Cooperating pathologists and medical
examiners supplied adipose tissue specimens.
Tissue specimens were analyzed by laboratories designated by the
NHMP. Chemical analysis was accomplished by thin layer chromatography
(TLC) until November 1974 when a gas chromatographic technique was
adopted. These techniques are discussed in more detail in Sections
4.2.4 and 4.2.1, respectively.
Polychlorinated biphenyls (PCB's) possess chemical characteristics
similar to those of organochlorine insecticides; hence, they also may be
detected by the same chemical analysis procedures used to analyze human
adipose tissue for pesticides. Because quantitating PCB's is more
difficult than quantitating many other substances, the PCB concentra-
tions in parts per million (ppm) were reported as falling in one of four
categories: not detected, trace (detected but <1 ppm), 1-3 ppm, and >3
ppm.
NHMP adipose tissue surveys indicate that measurable residue levels
of PCB's occur in a large percentage of the general population. Prelimi-
nary data summaries suggest that the higher levels of these chemicals
are more prevalent in the nonwhite population. The overall objective of
this report is to present the results of an evaluation of these apparent
racial differences.
-1-
-------
1.2. Study Methodology
The following activities were undertaken to evaluate estimates of
PCB residue levels in adipose tissue with regard to the statistical
meaning of apparent differences between racial groups:
1. Estimate and discuss the statistical impact of the survey
design employed by the NHMP on the PCB residue level estimates.
2. Evaluate the level of "reading" error in the interpretation of
the PCB chromatograms.
3. Investigate the implications of duplicity and possible biases
related to the use of autopsy and surgical materials.
4. Assuming various plausible levels of design and measurement
error effects, compute the statistical precision of PCB resi-
due level estimates and the probability that real residue
level differences between racial groups would be detected.
5. Statistically analyze the computerized data files.
It was concluded from a preliminary examination of a sample of the
chromatograms that any "reading" errors resulting from manually integrat-
ing the chromatograms would probably be small compared to other sources
of measurement error, i.e., in the analytical techniques used to measure
the chemical residues. The analytical techniques were investigated, and
lower bounds for the relative estimation precision of the PCB residue
concentrations were developed on the basis of previously published
results and experience. Different levels of precision in estimating the
PCB residue concentrations will also result in different levels of
precision in estimating the proportion (or percentage) of the population
falling in a given classification category (i.e., percentage >3 ppm).
The possible effects of measurement precision on the statistical signifi-
cance of differences in the proportions of racial groups are discussed
in Section 3.3.
In estimating the impact of the statistical survey design employed
by the NHMP, the discussion in Section 3.4 focuses on the highest PCB
residue level classification (>3 ppm). That is, the proportions of
nonwhites with greater than 3 ppm PCB residue levels are compared to the
proportions of whites having greater than that level. In this report,
these proportions are denoted by P and P , respectively. The discus-
sion is limited to this category for two reasons: (1) any health effect
-2-
-------
of PCB's would be most evident in the highest concentration category and
(2) one category is sufficient to illustrate the possible effect of the
statistical design and measurement error.
Estimates for the proportions of whites and nonwhites having PCB
residues greater than 3 ppm were calculated for the 1972-76 fiscal years
assuming no design effect or measurement error. Tables illustrate how
various levels of sample design effect and measurement error can affect
the statistical significances of these differences. Fiscal 1977 data on
the computer accessible data files were excluded because the first-stage
sampling units were changed from the previous years. This precluded
relating first-stage units in 1977 and other aspects of the sample
design with the previous years.
In the discussion of the survey design impact, an example is used
to illustrate the possible effects of the purposive exclusion of subpopu-
lations resulting from sampling only within cities with populations of
more than 25,000. The emphasis is on the bias potential for estimating
the difference P -P , where the bias is defined as the expected differ-
w n' r
ence minus the true difference. A similar discussion addresses the
final-stage sampling of cadavers and surgical patients and the methods
used in obtaining the tissue specimens.
The NHMP computer accessible data files were analyzed using a
statistical method especially adapted for multistage survey data. This
technique allows investigation of differences between racial groups
after adjusting for other factors such as sex, age, geographic region,
and fiscal year. The total variance of the estimates is approximated
including both the measurement and sampling components. The analysis
was based on the following two assumptions: (1) quantitation and sampl-
ing biases are similar for each racial group and (2) sampling within
sample cities produced a nearly simple random sample within each city.
-3-
-------
2. CONCLUSIONS
2.1. Issue of Racial Differences
The issue of whether or not racial differences exist in the dis-
tribution of PCB residue levels was examined for the category ">3 ppm"
and for percentage detected (percentage of population with a trace, 1-3
ppm, or >3 ppm).
A summary of the analysis of the data on the NHMP computer accessi-
ble data files labeled "PCB's" is given below. (The PCB label is put in
quotes because of the concerns noted below about quantitating PCB using
only a single isomer).
For percent "PCB" >3 ppm:
1. Over the fiscal years analyzed (1972-76), nonwhites averaged
7.15 percentage points higher than whites.
2. Over the fiscal years 1972-76, there was an increasing trend
in the percentage of people with "PCB" >3 ppm in their adipose
tissue. The rate of increase is estimated to be 1.73 percent-
age points per year. The percentages of individuals with >3
ppm ranged from 2.58 percent in 1972 to 7.34 percent, in 1976
for whites and 7.55 percent in 1972 to 16.67 percent in 1976
for nonwhites.
3. The trends over time (increase per year) for whites and non-
whites are 1.19 and 2.28 percentage points per year,
respectively.
For percent positive detections:
1. Over the fiscal years 1972-76, there was an increasing trend
in the percentage of people with some detectable "PCB" in
their adipose tissue. The rate of increase is estimated to be
2.99 percentage points per year and means ranged from 84.58
percent in 1972 to 96.54 percent in 1976.
2. The trends over time for whites and nonwhites are 2.80 and
3.18 percentage points per year, respectively.
These findings listed above may not be precise or accurate measure-
ments of PCB residues in the overall population because of potential
inadequacies in the sampling and chemical analysis. However, based on
the following assumptions: (1) quantitation and sampling biases are
similar for each racial group and (2) sampling within sample cities
produced a nearly simple random sample within each city, several signifi-
cant differences were detected. The percentage of nonwhites having
-5-
-------
greater than 3 ppm "PCB" is significantly higher than whites (probability
of no difference <.001). There was no evidence of a difference in the
percentage of whites and nonwhites with detected "PCB". The percentage
in both categories investigated demonstrated a significant increase over
time (probability of no time trend <.01). However, the apparent trend
may be due in part to the change in chemical analysis methodology during
the period studied. The trends appear to be the same for whites and
nonwhites.
The apparent differences between racial groups could neither be
confirmed nor denied because of the following two important reserva-
tions: (1) there are uncertainties about the accuracy and precision of
using a single isomer to quantitate aggregate PCB's (see Section 4), and
(2) the data appears to contain some inconsistencies and show greater
fluctuations than normally expected in surveys of this size (see Section
5). The first reservation admits the possibility of bias in estimating
differences between racial groups. The second is an indication of
possible measurement or sampling bias. Because of these reservations,
the data in their present form should not be used to make epidemio-
logical inferences about aggregate PCB residues.
2.2. Statistical Design
The geographic stratification used in the NHMP sample design appears
to have only a slight effect on the precision with which the proportions
of each racial group having PCB levels >3 ppm or percent detected are
estimated. This might be expected because of population mobility and
the widespread use of PCB's. The clustering of sample individuals by
cities is expected to decrease the precision of estimates, but this
component of the variance was not estimated separately. Further, the
effect on the variance of subsampling within the clusters cannot be
estimated with any assured degree of accuracy because this stage of
sampling was not conducted within a probability framework. The assump-
tion of simple random sampling within cities is required for any statis-
tical analysis of these data. The exclusion of some individuals in rural
areas and the use of both surgical patients and cadavers qualify statis-
tical inferences to the general U.S. population.
-6-
-------
2.3. • Chemical Analysis
The peak in the gas chromatograms labeled "PCB" has been tenta-
tively identified as a heptachlorobiphenyl isomer (Appendix A). It is
also possible that other halogenated organics may coelute, yielding
j-
false high values. The "PCB" values were obtained by measuring one
peak and correlating it with an Aroclor 1260 standard. This implies that
the rest of the Aroclor 1260 pattern is "buried" under the pesticides
and is present in a constant ratio to the measured peak. This procedure
does not follow the recommended protocol (USEPA 1974 Section 5A[1]) and
may not provide conclusive information on racial differences in PCB
residues. Some chemists knowledgeable in the area of human tissue
analysis doubt the reliability of the technique employed by the NHMP.
The TLC data (pre-November 1974) are estimated to have ±50 percent
precision (Appendix B). Current data are insufficient to confirm or
refute this estimate. As far as can be ascertained, some potential
interferences (halogenated aromatics) may yield false high values.
2.4. Recommendations
2.4.1. Statistical Design
The purposive exclusion from the sampling frame of hospitals in
cities with populations of less than 25,000 limits the inferences that
can be drawn to the general U.S. population. This exclusion results in
a substantial fraction of the total U. S. population having no chance of
being selected in the sample. The additional selection of sample hospi-
tals from cities of 2,500 to 25,000 persons would essentially eliminate
this inferential limitation. Special arrangements may have to be made
for hospitals in these smaller cities that do not have the medical
facilities to secure and store sample tissues. The technique of select-
ing alternate first-stage sample sites also needs improvement. Repeating
the process by which the original sites were selected would be satis-
factory (see Section 5.1).
The limitations resulting from the use of judgment or convenience
sampling are well known (Cochran 1977; Kish 1965). Hence, the method of
selecting hospitals (or pathologists and medical examiners) should be
conducted as nearly as possible within a probability framework. For
instance, sampling frames for hospitals could be constructed for each
"Personal Communications, G. W. Sovocool, U.S. Environmental Protection
Agency, Research Triangle Park, North Carolina, October 1979, and H.
Enos, University of Miami, Miami, Florida, October 1979.
-7-
-------
sample city. The hospitals could then be randomly selected with a known
probability. The judgment sampling of tissue specimens by cooperating
professionals (pathologists or medical examiners) should also be modi-
fied. Simple protocols for selecting patients or cadavers and tissue
specimens should be developed. Incorporating probability sampling
methods or explicit procedures (where probability sampling is not feas-
ible) would significantly improve the inferential value of the survey
data.
For the majority of patients and cadavers, the zip code of resi-
dence can be recorded and, hence, the approximate geographic location of
their residences can be ascertained. If the zip code of residence is
not available, then the zip code of the hospital could be substituted.
The anatomical site from which the tissue sample is taken should be
recorded. This may allow one to investigate and account for the differ-
ence in residue concentrations within the body. Variation of residues
within the body does not preclude residue estimation, but it necessi-
tates the use of a well-defined methodology for selecting or at least
recording anatomical sites.
The above comments should be accompanied by an awareness of special
problems that led to the purposive selection of certain hospitals and to
the exclusion of hospitals in cities with populations of 25,000 or less.
For example, long-term care facilities and mental hospitals in which
surgery and postmortem examinations are not usually performed were
excluded. The smaller cities were excluded because many of the hospi-
tals within these cities do not have health care facilities with labora-"
tory or pathology departments adequate for tissue sampling and storage.
Some of these hospitals routinely embalm cadavers before autopsy and
thus contaminate the adipose tissue.
2.4.2. Chemical Analysis
2.4.2.1 Evaluation of Current and Past Techniques
Data for PCB's obtained by gas chromatographic analysis in the NHMP
should not be used as a basis for epidemiological inferences unless one
of two validations is made as discussed below:
(1) The data collected over the past several years may prove
useful if the peak being quantitated can be confirmed as a
specific heptachlorobiphenyl isomer and shown to be free from
interference by all halogenated organics known to be in general
-------
tissue samples. The response of this isomer then could be
measured and the concentration calculated. This would result
in reporting a concentration (or range) of a given isomer that
is observed but would not assume the other isomers are present
when not observed.
(2) The second corrective measure would involve demonstrating
(with a statistically valid sample) that the PCB profiles in
the "unseen" part of the chromatogram are similar between the
racial groupings and, therefore, that the heptachlorobiphenyl
peak being measured is, in fact, indicative of "total PCB"
concentrations in adipose tissue.
Even with these precautions, the data should be assumed to have an error
of at least ±50 percent (see Section 4.2.2) unless better precision can
be demonstrated.
The TLC data should be further validated to check for potential
interference from such compounds as polychlorinated naphthalenes, ter-
phenyls, benzene, and other aromatics and polybrominated aromatics. If
used in the future, a selected portion (e.g., 10 percent) of the TLC
determination should be confirmed by an independent technique such as
GC/ECD or GC/MS.
2.4.2.2. Alternate Techniques for Future Work
The selection of analytical methods is a compromise between sensi-
tivity and selectivity. GC/ECD is not only highly selective for haloge-
nated compounds but also is one of the most sensitive techniques avail-
-12
able with detection limits approaching the picogram (10 g) level.
Several techniques are listed below in approximately increasing order of
complexity and information content.
1. The easiest modification of current procedures would be simple
confirmation of selected specimens (e.g., 10 percent) by an
independent technique such as GC/HECD GC/MS, TLC, or
perchlorination.
2. The method recommended in the EPA Pesticide Manual (USEPA 1974
Section 9 C) should be followed. The separation of PCB's from
pesticides will produce more reliable and accurate data and
will allow detection of early-eluting PCB peaks.
3. Modern chromatographic techniques should be considered. In
particular, glass capillary GC will permit resolution of many
more components and will significantly reduce chances of
coelutions. Modern ECD's can be temperature-programmed. This
increases the effective resolution of the chromatography.
Precise retention times (Hewlett-Packard claims ±0.002 minimum
precision) would also improve identification certainties.
-9-
-------
4. Detection of PCB's could be significantly improved with mass
spectrometric techniques. The "normal" electron impact MS
detection limit of about 1 ng would produce problems for the
detection of PCB's in NHMP tissue samples. This detection
limit may be improved 10-100 fold by using selected ion moni-
toring (SIM), which is available on all quadrupole and many
magnetic sector instruments. However, techniques such as
negative ion MS have achieved extremely good sensitivities—
better than ECD. If sufficient tissue specimen is available,
high resolution MS will provide precise mass measurements to
greatly aid in determining molecular formulae.
5. A combination of the above suggestions—cleanup by silicic
acid chromatography and analysis by glass capillary GC/MS—is
probably the most powerful routine PCB analytical system.
Despite its high capital costs, capillary GC/MS is considered
routine by many laboratories and is feasible for this program.
The above suggestions all move toward the ability to report indivi-
dual PCB isomers (i.e., "70 pg 2,2',3,3',4'4'5-heptachlorobiphenyl"
instead of "1.5 ng Aroclor 1260"). While this increases the burden of
calculation for both the chemist and the statistician, the extrapolation
to the health effects of PCB's is much more scientifically sound, parti-
cularly when it has been shown that toxicity and storage vary with
individual isomers (Matthews and Anderson 1976; Biocca et al. 1976;
McKinney 1976). Steps 1 and 2, above, along with a rigorous quality
assurance program, should be considered a minimum for future analyses
for PCB's in tissue.
-10-
-------
3. STATISTICAL SURVEY DESIGN
3.1. Overview
The statistical design used to collect data for the Adipose Tissue
Survey of the NHMP has several stages of sample selection. The contermi-
nous 48 states were stratified into several geographic regions. Sample
cities were selected from a list of eligible places (cities >25,000)
proportional to their population. The hospitals were purposively select-
ed according to type and size. Because sampling was conducted in select-
ed hospitals in the sample cities, the sampling areas may be considered
to be the union of the service area of the sample hospitals within each
city. The specimens of adipose tissue from cadavers and surgical
patients were obtained through the cooperation of pathologists and medi-
cal examiners. The cooperating professionals judgmentally selected the
specimens. Guidelines were given to aid in the selection of cadavers,
surgical patients, and anatomical sites (see Appendix F).
Certain sampling components differ between fiscal years 1970-72 and
1973-76. Before fiscal year 1973, the nation was stratified into the
four census regions for purposes of sample allocation. The number of
sampling sites within each region was assigned in proportion to the
region's population. Population sizes used to set age, sex, and race
quotas and for selection of cities were based on the 1960 census (Yobs
1971). Beginning in fiscal year 1973, the sample design was modified to
reflect the demographic distribution of the 1970 census. The number of
strata was increased from the four census regions to the nine census
divisions. The quotas were adjusted to reflect the demographic distri-
bution in each stratum (USEPA 1972b). Appendices D and E contain a
listing of the census regions, divisions, and states and an example of
the quotas assigned for fiscal years 1973-76. Figure 3.1 illustrates
the various stages of sampling for the Middle Atlantic census division
in fiscal year 1973. The sampling was conducted in an analogous manner
for each stratum and in each fiscal year.
Data summaries for fiscal years 1972-76 were provided by EPA (USEPA
1977b). These summaries presented frequencies and unweighted relative
frequencies for four residue concentration categories for each of several
racial groups. Included are tables for each stratum (census division or
region) and a national summary for each fiscal year. Appendix G con-
-11-
-------
37,152,813
16,607,245
unknown
183
Target Population — Total population of Middle
Atlantic Census Division.
Purposive exclusions of cities less than 25,000.
1st Stage Sampling Frame — Cities greater than 25,000.
Purposive exclusion of some.hospitals, pathologist
and medical examiners, selection of only
cadavers and surgical patients.
Sampled Population — Includes a subset of only
surgical patients and persons who died in FY73.
Subsampling of surgical patients and cadavers
by cooperating professionals.
Sample.
Figure 3.1 Sampling frame and selection proceudres for Middle
Atlantic Census Division in FY73.
Source: Based on information contained in references (USBC 1972,
USEPA 1972a and USEPA 1977b).
-12-
-------
tains the national summaries. In a majority of the census regions and
for each national summary, the proportion of nonwhites falling in the
high concentration categories is greater than the proportion of whites.
However, no statistical tests concerning the equality or inequality of
the proportion of whites and nonwhites having concentration of PCB's
greater than 3 ppm were performed by EPA on these data.
Selected statistical analyses (USEPA 1977c) using Statistical
Packages for the Social Sciences (SPSS) were supplied by EPA. These
analyses were run on data aggregated over several years and without
weighting the data. The results supported the apparent PCB residue
level differences between racial groups. However, because these analyses
assumed simple random sampling and did not account for sources of error
such as measurement variance, the statistical significance of these
findings is questionable. In the next section the impact of various
levels of measurement variance, sampling and measurement bias on the
apparent significance levels is discussed.
3.2. Significance of Racial Differences Assuming No Design Effect or
Measurement Error
The hypothesis of no difference between racial groups for the
category >3 ppm (P - P = 0) is tested using the national data summary
n w
in Appendix F for the fiscal years 1972-76. Because of the large sample
sizes, it is adequate to use the normal distribution to approximate the
significance level of the test statistic
P -P
_ n w
C [VAR (P -P )]^
where
n = Number of whites in the sample,
n = Number of nonwhites in the sample,
p _ Number whites in the sample with >3 ppm,
w n
w
p _ Number nonwhites in the sample with >3 ppm,
w n
n
and
-13-
-------
-v ~ P (1-P ) P (1-P )
Var[P - P ] = -2 2- + -S S_ •
w n n n
w n
(The above variance formula is calculated by conditioning on n and n ,
hence no covariance term appears in the expression.)
Under the assumptions of simple random sampling, no measurement
variance, and no measurement or sampling bias, the rightmost column of
Table 3-1 can be thought of as the probability of no difference between
the racial groups. Under these assumptions, fiscal years 1973 and 1976
exhibit what might be a "statistically significant" difference between
races (the proportion of nonwhites is "significantly" larger than that
of whites). In addition, there is a lesser indication that such a
difference might exist in fiscal years 1972, 1974, and 1975.
In the following sections, tables illustrate the effects of various
levels of measurement error, bias, and the sample design on the signifi-
cance levels in Table 3-1 and, hence, on the validity of the claims of
statistically significant differences.
3.3. Comments on the Impact of Measurement Error
It is difficult to relate the relative standard deviation of the
measurement techniques (see Appendix B) to the error in estimating the
proportion of the sample falling in a particular concentration range. A
rigorous derivation requires knowledge of or assumptions on the distri-
bution of the residue in the sampled population. Intuitively, it seems
that any uncertainty in classifying tissue specimens correctly would
increase the estimation variance. This is illustrated in the following
example.
Consider the population consisting of five elements (e1 , e?, e«,
e,, e }. Elements e , e«, and e_ have real value one and elements e,
and e have real value zero. However, assume that e~ and e^ are classi-
fied correctly 50 percent of the time and incorrectly 50 percent of the
time. Let P denote the proportion of the population with the value one.
Hence, in this example, P = 0.6. The mean and variance of the estimator
*\ /\.
of P for a sample of size two is calculated for P considering the case
above using first principles (Mood et al. 1974) and when no classifica-
tion error occurs (error-free case) using the formula for the mean of
-14-
-------
Table 3-1. Summary of Results for Testing the Null
Hypothesis that P - P = 0 for FY72-FY76
w u.
FY
72
73
74
75
76
n
w
1469
981
798
680
569
n
n
303
132
119
105
103
*w
.0530
.0428
.0438
.0956
.0738
A
Pn
.0825
.1667
.0924
.1619
.1942
/v >\
P -P
n w
.0295
.1238
.0486
.0663
.1204
y\
SE
.0169
.0331
.0275
.0377
.0405
Significance
t level*
c
1.74
3.74
1.77
1.76
2.97
.082
.000
.077
.078
.003
The information in this table is given in Appendix F, or calculated using the
following equations.
P _ Number whites in the sample with > 3 ppm
W ^
n
w
P _ Number nonwhites in the sample with > 3 ppm
n
n
SE = [Var(Pn - PJ]^
P - P
n w
c *•
[Var(P - P )]"*
L v n w
*
Probability of a value occurring greater in absolute value than t ,
calculated by the equation
_i, X2
* dx (Mood et al. 1974).
i- [ -L
J V~2T
-15-
-------
the hypergeometric probability distribution (Mood et al. 1974). In both
cases, the estimator P has an expected value of 0.6. Hence, P may be
estimated in an unbiased manner even in the classification error case.
In the error-free case the variance is 0.09 and for the classification
error case the variance is 0.1025. Even in the situation where the
measurement errors tend to be compensating, the variance of the esti-
mated proportion is increased, in this case by more than 10 percent.
With the summary data available (USEPA 1972a and USEPA 1977c), it
is not possible to estimate the reduction in precision of the estimated
difference due to the presence of measurement errors. Techniques are
available to obtain an overall estimate of the variance of the difference
estimator (Hansen et al. 1953). This would involve obtaining the esti-
mates of the proportions for each sample site from the NHMP data files.
These techniques are based on the assumption that the sample site esti-
mators are unbiased (as stated earlier, however, this requires assump-
tions about the data). An analysis technique discussed in Chapter 5
does approximate the overall variance of the proportions (or percentages).
Table 3-2 demonstrates the increase in the significance level for
various levels of increase in variance due to measurement error. The
relative standard error (y) is defined as
o o 3>
— /"CT 4- rt *\ */CT7
— (.or. T cr ) /or, ,
."N A
where SE is the standard error of the estimator P - P assuming no
w n 6
increase in variance due to measurement error, and a is the additional
' m
variance due to measurement error. Small increases do not greatly
affect the significance level; however, when the measurement variance
2
equals SE (y = 2), the significance level is drastically increased in
all cases. An increase of variance of this magnitude or greater due to
the presence of measurement errors is not beyond the realm of possi-
bility, considering the suspected measurement errors in the chemical
analysis.
Table 3-2 can be used to estimate the true confidence of the usual
A. A.
expression for a 95 percent confidence interval (0-1.96SE, Q+1.96SE)
when measurement errors exist. For example, the true confidence for
such an interval would be about 88 percent for a 25 percent increase in
the standard error due to measurement error and 81 percent for a 50
percent increase in the standard error.
-16-
-------
Table 3-2. Effect of Measurement Variance on the Significance
Level of Test Statistics
Relative SE
*
Y
1.00
1.01
1.05
1.10
1.25
1.50
1.75
2.00
3.00
True significance
level
for a"" =0.05
.0500
.0523
.0648
.0748
.1169
.1913
.2627
.3271
.5135
True significance
level
for a = 0.01
.0100
.0108
.0142
.0192
.0393
.0859
.1410
.1977
.3905
True significance
level
for a = 0.001
.0010
.0011
.0017
.0028
.0085
.0283
.0601
.1000
.2728
,v (SE2 +a2)^ . -
2
assuming no increase in variance due to measurement error, and a is
additional variance due to measurement error.
**
a is the probability of making a Type I error, that is, concluding
there is a difference when in fact no difference exists.
The entries in this table are calculated using the equation
P = 1 - I —— e "* x dx, (Mood et al. 1974)
V 20
where ZQ equals 1.96, 2.57, and 3.29 for a equal 0.05, 0.01, and 0.001,
respectively.
-17-
-------
Table 3-3 summarizes the effect of the increase in the standard
error due to the presence of measurement error on the probability of
detecting real differences between racial groups, if the probability of
making a Type I error is 0.05 (the probability of concluding there is a
difference between racial groups when in fact none exist). This is done
for various levels of the relative standard error y and the relative
difference 8/SE where 6 is the value of the true difference between P
W
and P . If the difference is as large as the standard error (0/SE = 1),
then the probability of detecting a difference of this size decreases
from 0.17 (for y = 1.00) to 0.0628 (for y = 3.0). If 0/SE = 2, the de-
crease is more dramatic, from 0.5160 to 0.1022.
3.4. Impact of the Sample Design
3.4.1. Restriction of Cities
The sample hospitals were selected from cities with populations
greater than 25,000. The purposive exclusion of smaller cities may
introduce bias in estimating proportions of interest. It is also pos-
sible that the bias could result in the appearance of differences be-
tween racial groups when no such differences exist.
Consider the following situation. The eligible cities constitute
approximately 42 percent (.42) and the excluded areas 58 percent (.58)
of the population. Let P and P denote the true proportions of the
sw sn
white and nonwhites in the sampled subpopulation, respectively. Let P
and P denote the true proportions in the unsampled subpopulation.
Hence, the proportions must satisfy the relationships
P = 0.42 P + 0.58 P
w sw uw
P = 0.42 P + 0.58 P .
n sn un
It is possible for P to equal P and for P and P to be quite dif-
w ^ n sw sn ^
ferent; that, is the subpopulation sampled is actually different from
the target population. The resulting bias in the estimation may give
misleading results. To demonstrate, let P =0.0428 and P = 0.1667.
sw sn
(See Table 3-1 for fiscal year 1973.) If P = P =0.10, then the above
w n.
equations can be solved for P and P yielding 0.1414 and 0.0517
U. W Ll.ll
respectively. Hence by sampling only a subset of the target population,
bias may be introduced into estimates.
-18-
-------
Table 3-3. Effect of Measurement Variance on the Probability of
Detecting Real Differences
e/sE**
Y*
1.00
1.05
1.10
1.25
1.50
1.75
2.00
3.00
*
Y
assu
addi
a*
.05
.05
.05
.05
.05
.05
.05
.05
(SE2
SE
ming no
tional
0.25
.0572
.0565
.0560
.0546
.0532
.0523
.0518
.0508
+a )"*
increase in
variance due
6/SE is the value of
the standard error.
' a is the
there is a
The
entries
probability
difference,
0.50
.0790
.0763
.0740
.0685
.0628
.0594
• .0572
.0532
1.00
.1700
.1584
.1488
.1259
.1022
.0882
.0790
.0628
1.50
.3231
.2980
.2795
.2244
.1700
.1374
.1165
.0790
2.00
.5160
.4781
.4437
.3596
.2659
.2079
.1700
.1022
SE is the standard error of the estimator
2
variance due to measurement error, and a
to measurement error.
the true
difference
of making a Type I
when in fact no such
3.00
.8508
.8152
.7785
.6700
.5160
.4031
.3231
.1700
A, /N.
P -P
n w
is
4.00
.9793
.9678
.9532
.8925
.7601
.6277
.5160
.2659
between P and P divided by
w n '
error, that is, concluding
difference exists.
in the table are calculated using the
equation
1.96 + 9/SE
P
= 1 -
-1.
Y
/
96 + 9/SE
Y
j'
/HT
2
"^ X dx
(Mood et al.
1974) .
-19-
-------
The standardized bias of the estimator of P -P is a function of
w n
the bias, P - P , and given by
o W o li
Bias _ -0.1234 , ,,
SE ~ 0.0331 " '
Tables 3-4 and 3-5 allow one to gauge the impact of this standardized
bias on the probability of concluding there is a difference between the
proportions for the racial groups. They indicate that between 85 and
97 percent of the time one would erroneously conclude that the racial
groups were different. Admittedly, this may be an extreme case. It is
recognized that hospitals do indeed serve outlying areas beyond the
sample city boundaries. However, in many states, particularly in the
Rocky Mountain region, a significant portion of the population is, in
fact, excluded from the sampled population because they live in rural
areas.
Table 3-5 must be evaluated in proper perspective. At first glance,
one may erroneously conclude that bias may even increase the likelihood
of detecting real differences. In some situations, bias may in fact
accentuate differences. In others, bias may conceal differences. In
either case, bias can be misleading to the investigator in both the
magnitude and direction of differences.
3.4.2. Sampling Cadavers and Surgical Patients
By restricting the sample to cadavers and surgical patients, one
limits the sampled population to elements not actually belonging to the
target population (cadavers) and to individuals that constitute a unique
subset of the entire population (surgical patients). Bias similar to
that discussed in the previous section may be involved if the propor-
tions for each race of the sampled population and unsampled populations
differ. Any inferences made about the general population are limited by
the assumption that the sampled and unsampled populations do not differ
substantially.
Another possible source of bias in using surgical specimens is the
chance for repeated observations on the same individual, which will
result in disproportionate sampling of persons who are prone to surgery.
Precautions should be taken to prevent this from happening, but adjust-
ments could be made in analysis to take this fact into account if it is
recorded.
-20-
-------
Table 3-4 . Probability of Detecting Real Differences for
Various Levels of Positive Bias
' **BIAS/SE
8/SE"
0.00
0.25
0.50
1.00
1.50
2.00
3.00
4.00
0.00
.0500
.0572
.0790
.1700
.3231
.5160
.8508
.9793
0.25
.0572
.0790
.1165
.2396
.4169
.6141
.9015
.9890
0.60
.0790
.1165
.1700
.3231
.5160
.7054
.9382
.9945
1.00
.1700
.2396
.3231
.5160
.7054
.8508
.9793
.9988
1.50
.3231
.4169
.5160
.7054
.8508
.9382
.9945
.9998
2.00
.5160
.6141
.7054
.8508
.9382
.9793
.9988
1.0000
3.00
.8508
.9015
.9382
.9793
.9945
.9988
1.0000
1.0000
4.00
.9793
.9890
.9945
.9988
.9998
1.0000
1.0000
1.0000
Standardized true difference.
Standardized bias in estimating the difference 6.
This table can also be used for the probability of detecting real
differences for 9 and bias both negative.
The entries in this table are calculated using the equation
1.96 + 6/SE + BIAS/SE
/2
—-— e"^ X dx (Mood et al. 1974)
-1.96 + 6/SE + BIAS/SE
-21-
-------
Table 3-5 . Probability of Detecting Real Differences for
Various Levels of Negative Bias
e/sE*
0.00
0.25
0.50
1.00
1.50
2.00
3.00
4.00
**BIAS/SE
0.00
.0500
.0572
.0790
.1700
.3231
.5160
.8508
.9793
-0.25
.0572
.0500
.0572
.1165
.2396
.4169
.7852
.9633
-0.50
.0790
.0572
.0500
.0790
.1700
.3231
.7054
.9382
-1.00
.1700
.1165
.0790
.0500
.0790
.1700
.5160
.8508
-1.50
.3231
.2396
. 1700
.0790
.0500
.0790
.3231
.7054
-2.00
.5160
.4169
.3231
.1700
.0790
.0500
.1700
.5160
-3.00
.8508
.7852
.7054
.5160
.3231
.1700
.0500
.1700
-4.00
.9793
.9633
.9382
.8508
.7054
.5160
.1700
.0500
Standardized true difference.
Standardized bias in estimating the difference 9.
This table can also be used for determining the probability of
detecting real differences for 6 negative and bias positive.
The entries in this table are calculated using the equation
1.96 + 9/SE + BIAS/SE
/In
1 " ' —— e"^ x dx (Mood et-al. 1974)
-1.96 + 6/SE + BIAS/SE
-22-
-------
When a method other than probability sampling is used to select
individuals, sampling bias in the resultant data becomes a real possi-
bility. The purposive elimination of particular subpopulations should
be avoided or, if unavoidable, the resulting limitations should be
clearly used to qualify the results. Biases such as those discussed
previously resulting from the purposive exclusion of small cities might
also arise with purposive exclusion of certain types of hospitals.
3.4.3. Remarks on the Bias in Estimating Racial Differences
There is no prior evidence that any of the possible sources of bias
discussed earlier affect the racial groups differently. If this is the
real situation, the biasing effects would tend to cancel out when differ-
ences are estimated. Hence, the biases of the estimated difference
between subpopulation proportions could be small. Also, bias introduced
by the method of chemical analysis may be similar for each racial group.
If true, this type of bias would also tend to cancel out when looking at
differences between groups.
-23-
-------
4. EVALUATION OF CHEMICAL ANALYSIS AND DATA INTERPRETATION
4.1. Objective
The objective of this section is to comment on the reliability of
the raw PCS data reported as part of the NHMP.
4.2. Discussion
4.2.1. Qualitative Assessment
Chromatograms of the original data were obtained through the cour-
tesy of Mr. John D. Tessari, Supervising Chemist and Laboratory Direc-
tor, Colorado Epidemiologic Pesticide Studies Center, Colorado State
University, Fort Collins, Colorado. A copy of the cover letter discuss-
ing the measurement calculation method is included as Appendix B.
Tissue specimens were extracted according to the modified Mills-
Olney-Gaither procedure (USEPA 1974 Section 5A[1]) and analyzed by gas
chromatography with electron capture detection (GC/ECD) on two columns,
as discussed in Mr. Tessari's letter. The pesticides and PCB's were not
separated using silicic acid chromatography as recommended (USEPA 1974
Section 9 C). Without this separation, some of the pesticides interfere
with the GC/ECD analysis of some of the isomers of PCB's. It should be
noted that the primary objective of this program was pesticide moni-
toring; the PCB analysis was appended with a directive that minimal
additional effort be expended. Three chromatograms are presented in
Figures 4-1, 4-2, and 4-3. Figure 4-1 represents a "high" PCB level
(>3 ppm) sample, Figure 4-2 a "medium" level (1-3 ppm), and Figure 4-3 a
"low" level (<1 ppm). Figure 4-4 shows the standard Aroclor 1260 chro-
matograms obtained under conditions similar to those in Figures 4-1,
4-2, and 4-3, on October 2, 1979, at Colorado State University. Compari-
son of Figures 4-1, 4-2 and 4-3 with the standard (Figure 4-4) indicates
that a clear Aroclor 1260 pattern is not discernible among the pesticide
peaks. Other PCB peaks (in addition to the later-eluting peak used in
quantitation) seem to be present and lend support to the PCB identifica-
tion, but there is no clear "fingerprint." This is reasonable because
the environmental fate, absorption and excretion dynamics, and metabolic
rate differences of the various PCB isomers dictate that a PCB pattern
in tissue will probably not be consistent with a given commercial mixture
as shown in Figure 4-5.
-24-
-------
I
N>
t_n
9 CHROMA1CWKAM
IDENTIFICATION
3M. or
Sent'ls:
Chror.vl. Hv y
Flail :,.•"i „£
Co?:" !• t, .. 40 .
Oil .
Sainp.1*
DJI6
Injecl '.'id.. .?•/[ -J
£luiio< IM-
9
Figure 4.1 Chromatogram of a tissue specimen on 4% SE30/6% OV-210 column, reported as
">3ppm PCB" (factor calculated as 18. along bottom)
See Appendix A for source
-------
HFTII. •• '*••
DENTif C.\IIOI»
C-i
I
Figure 4.2 Chromatogram of a tissue specimen on 4% of SE30/6% OV-210 column,
reported as "1-3 ppm PCB" (factor calculated along bottom
as 2.75). See Appendix A for source.
-------
I
NJ
IROMA1CXWAM
)EN1IF1CA"OM
OIL:.
A
•o
o y>-
lOENTIfCAIIOM
£td. or
Sample:
Chiomal. No—/^
flnil E». Vol
Concenual!-"—2^
OIL:
Sample
Date:
|n|ect. Vol..
CluNon lrac._
%
Figure 4.3 Chromatogram of a tissue specimen on 4% SE30/6% OV-210 column,
reported as "
-------
I
K3
OO
I
;tlROMATOGRAM
.IDENTIFICATION! . i
i •
- .:.— I
f I :•.. .
^i .1 . I :
Figure 4.4 Chromatogram of equivalent to 1.5 ppm Aroclor 1260 standard
Note comment on chromatogram. See Appendix A for source
-------
2 4
TIME, minutes
Figure 4.5 Electron capture gas chromatograms of (A) Aroclor 1016 standard
and (B) PCB-residue extracted from brain of rat fed on diet con-
taining Aroclor 1016 for one year. Reproduced from (Lewis 1977).
-29-
-------
GC/MS confirmation of a composite (Figure 4-6) and overwhelming
evidence from previous studies suggest that PCB's are in nearly all
human specimens. In a two-year intensive study, PCB's were confirmed in
•ff
322 tissues by GC/MS. In addition, recent evidence supplied by the
Office of Pesticides and Toxic Substances of the U.S. Environmental
Protecton Agency (OPTS-EPA) (Appendix A) confirms this peak as heptach-
lorobiphenyl. Although the evidence indicates that the peak being
measured is heptachlorobiphenyl, interferences may be present in some
specimens.
4.2.2. Quantitative Assessment
The single late-eluting peak was quantitated as discussed in Appen-
dix B. A comparison of five integration methods (USFDA) showed that
peak heights, disc integration, triangulation, peak height x width-at-
half height, and retention time x peak height were not significantly
different. Thus, the precision of integration is not a major issue in
this evaluation.
Assessing the quantitative accuracy and precision of the PCB values
reported is problematic (Appendix B). Disagreement exists among scien-
tists in the field as to the accuracy of the technique employed. Mr.
Tessari estimates precision of about ±50 percent. Furthermore, Larry
Griffin, also of Colorado State University, in notes on the standard
chromatograms (Figure 4-4), illustrates the calculations of the factors
(see Appendix B). His calculations show factors of 1.00 and 0.84 for
injections of the equivalent to 1.5 ppm Aroclor 1260 on the two GC
columns. This represents a 16 percent uncertainty in the GC analysis
alone. As he further noted, these factors "would have been reported as
> 3 ppm, therefore we do need to revise our factors."
Dr. R. G. Lewis and several of his coworkers at the Research
Triangle Park laboratory of the Environmental Protection Agency (US
EPA-RTP) were consulted about this problem. They indicated that a
precision of ±50 percent was the best to be expected in cases where the
pattern matches an Aroclor and where four or more peaks are quantitated.
When presented with the quantitative method used by NHMP, they felt it
would be a reasonable estimate to consider the precision at best ±100
percent.
Personal Communications, H. Enos, University of Miami, Miami, Florida,
October 1979.
-30-
-------
M-S7.0
5
u
i
100
M-40U
SCANNUMMM .
Figure 4.6 Computer Reconstructed Total ION and Mass Chromatograms of a
Composite Human Adipose Tissue Extract. Column: 45.7 m Scot
Column coated with SE-30. Programmed from 170-240°C at 2 C/
min (N.C. - not chlorinated). Reproduced from (Lewis 1977).
-31-
-------
The chromatograms and procedures were described over the telephone
to Dr. Henry Enos, an expert in this field, who also felt that a preci-
sion of ±100 percent would be a reasonable estimate. He noted that
although this procedure may be valid for Aroclor 1260, it does not
detect any of the lower PCB's that do not contain heptachlorobiphenyl
due to the "masking of those peaks by pesticides.
A further indication of accuracy and precision is presented in the
interlaboratory PCB analysis results supplied by R.G. Lewis in Appendix
C. Excluding two outliers, thirteen laboratories obtained a mean of 81
±68 percent on 10 (Jg/L of Aroclor 1254 in water. A more recent inter-
laboratory check with spiked fat reported 96 percent accuracy and ±33.8
percent precision. However, if the data are recalculated including both
the excluded "outlier" and the missed identification as "zero", the mean
recovery is 65 percent (accuracy) and ±96 percent precision. These
results are also provided in Appendix C. It should be noted that all
participating laboratories were aware this specimen was a check and
therefore should have devoted special attention to achieving their
"best".
Thus, estimates of ±50 percent precision are optimistic. Given the
above arguments of precision and that a clear Aroclor 1260 pattern was
not evident in the sample, it appears that PCB quantitation based solely
on the peak labeled as "PCB" by the methods used is not reasonable and
even "semiquantitation" is questionable.
4.2.3. Potential Interference
Many other compounds could coelute with PCB peaks giving erroneous-
ly high readings. Conversely, the early-eluting PCB isomers could and
do interfere with quantitation of the pesticides. This may be seen by
comparing retention times of some of the pesticides and the Aroclor 1260
standard in Figures 4.1 through 4.4. A more graphic representation is
shown in Figure 4.7 (USEPA 1974 Section 9E). These chromatograms were
obtained under instrumental conditions similar to those shown in Figures
4.1 through 4.4.
*
4.2.4. Assessment of TIC Methodology
Prior to November 1974, PCB values in NHMP were obtained by a thin
layer chromatographic (TLC) technique (USEPA 1977a and Mulhern et al.
1971). This semiquantitative technique is reported to have a precision
-32-
-------
Revised 11/1/72
Section 9, E
i
(-0
U)
i
4%SE-30/6%OV-210
Chroma to^ra-iis of three AROCLORS on column of
1% SE-30 / 6% OV-210. Column temp. 200°C.,
carrier flow 60 ml/min., ^H detector, electrom.
attenuation on an E-2 10 x 16; dotted line a
mixture of chlorinated pesticides, identity and
injection concentration given below:
1. Diazinon
2. Heptechlor -
3. Aldrin
li. Hept.Epox. -
5. p,p'-EDE -
6. Dieldrin
1.5 ng 7. o,p'-DUT — 0.2lj ng
- 0.03 8. p,p'-DDD — ,2U
9. P,P'-DDT — .30
10. Dilan — .75
- .09
- .0° 11. Methoxychlor .60
.12
AROCLOa 12 J 4
AROCLOR 1248
6 ng ini«ction
AROCIOR 1260
4i>9 i
Figure 4.7 Chromatograms of Aroclors and Pesticide standards
illustrating potential interferences.
Reproduced from (USEPA 1974 Section 9 E).
-------
of ±50 percent (Appendix B). This technique is advantageous in that all
tested PCB isomers or, more correctly, Aroclor mixtures have similar R,.
values (i.e., they elute to nearly the same position on the TLC plate),
so the PCB value reported is an integration of all isomers present.
This technique, however, lacks specificity for individual PCB isomers.
The analytical conditions used by the NHMP elute PCB's essentially
at the solvent front with Rf values ranging from 0.91 to 0.94 (Mulhern
et al. 1971). The chromatographic resolution is poor and other com-
pounds such as polychlorinated naphthalenes (PCN's), polybrominated
biphenyls (PBB's), polychlorinated terphenyls (PCT's) and nonpolar
pesticides may coelute. The best resolution on TLC is obtained at an R-
of about 0.2 to 0.8 (Stahl 1969). As an example, DDE elutes with the
PCB's and must be removed by oxidation prior to TLC analysis (Mulhern et
»u
al. 1971). That proper quality control procedures were followed,
including GC/MS confirmation of the pooled extracts, adds confidence to
the reported values.
4.2.5. Correlation of Data Generated by the Two Methods
Since the "PCB" values reported by the two chemical analytical
methods are used in parallel for statistical analysis, it is appropriate
to comment on the comparability of the data. The TLC method reportedly
detects all PCB's, while the GC method uses one isomer (of 209 total
possible) and extrapolates the rest. Evidence that the methods generate
comparable data is discussed by Mr. Tessari in Appendix B. However,
because no quantitative estimation of the relative bias of the two
methods was available, their relative equivalence is unknown. For the
purpose of discussion in this report, the differences in the data were
assumed to be negligible.
Personal Communication, J. Tessari, Colorado State University, Fort
Collins, Colorado, October 1979.
-34-
-------
5. WEIGHTED ANALYSIS OF COMPUTER ACCESSIBLE DATA FILES
5.1. Introduction and Assumptions
This chapter includes a technical discussion of the statistical
analysis of NHMP computer accessible data files. The following analysis
employs a regression technique especially adapted for multistage survey
data. This technique approximates the total variance of the estimates
including the measurement and sampling components and also allows adjust-
ments for other factors such as sex, age, census region, and fiscal
year. The analysis is based on the following two important assumptions:
(1) quantitation and sampling biases are similar for each racial group
and (2) sampling within sample cities produced a nearly simple random
sample within each city. The calculation of the sample weights was
based on assumption (2) and the significance levels of all tests of
hypotheses depend on the validity of both (1) and (2).
5.2. Sample Weights
Because some stages of selection involve nonprobability sampling,
the true probability of selection cannot be calculated. Even if one
assumes that the selection probabilities of a specimen within sample
cities are approximately equal, the sample design of the NHMP adipose
tissue network does not give equal "probabilities" of selection to all
elements in the sample. Because of this situation, a sample weight was
calculated for each observation that reflects its approximate probabi-
lity of selection. Including these weights in the analysis may reduce
bias in estimating means or proportions. In the following paragraphs,
the procedures used for computing approximate weights for the NHMP data
are described.
5.2.1 Calculation of Weights
For the purpose of calculating weights, two stages of selection
are considered. In the first stage, cities were selected within each
stratum with probability approximately proportional to population. In
the second stage, samples of cadavers and surgical patients were selected
in a nonprobabilistic manner from a hospital(s) or other facility located
in the sample cities. Equal probabilities of selection were assumed for
this stage.
The sample weights were calculated as the inverse product of the
probabilities of selection for each stage. This can be expressed alge-
braically as
-35-
-------
whij = f'hlv1" . .
where W, .. denotes the weight of the j-th specimen in the i-th city and
h-th stratum; P.. denotes the probability of selecting the i-th city in
the h-th stratum (for large cities selected with probability 1 this is
more accurately described as the expected number of times the city is
selected in the sample); and P. .. denotes the probability of selecting
the j-th specimen given the i-th city in the h-th stratum was selected.
To calculate W, . ., it is necessary to calculate P.. and P.., ..
The cities were selected independently within each stratum. For
each stratum, the cumulative total population of eligible cities was
divided by the number of cities to be selected in the h-th stratum.
This calculation gives the sample selection interval. A random number
is then selected between "1" and the value of the selection interval;
this gives the random start. The method then involves listing the
cities in a random order, calculating the cumulative totals and then
selecting cities by matching the cumulative totals to the random start
and integer multiples of the selection interval plus the random start
(USEPA 1973).
The above can be expressed algebraically in the following manner.
For each stratum, the selection interval is
:h = Vmh '
where I, , N, , and m, denote the selection interval, cumulative popula-
tion total, and number to be selected for the h-th stratum, respectively.
The cities selected in the sample are those for which the cumulative
totals match r, + K x I, for K = 0,...,m,-l, where r, is the random
h h h h
start for the h-th stratum.
This method assigns a probability of selection to each city equal
to the population of the city divided by the selection interval. This
can be expressed algebraically as
P, . = N, ./I, ,
hi hi h '
where N, . denotes the population of the i-th city in the h-th stratum
and P, . and I, are defined above.
hi h
-36-
-------
Under the assumption made above for random selection of samples
within a city, the calculation is straightforward. The probability of
selection is given by dividing the number of specimens selected in each
city by the population of the city. This can be expressed algebraically
as
Pj|hi = nhi/Nhi
for j = l,...,n, ., where n, . denotes the number of samples collected in
the i-th city in the h-th stratum and P. ., . and N, . are defined above.
I
The weight can be written as
W, . . = [N, ./I, x n, ./N, .J"1 ,
hij l hi h hi hij '
which simplifies to
W, . . = I,/n, . ' .
hij h' hi
Hence, the approximate sampling weight for the i-th city in the h-th
stratum is the same for all samples collected in the city and is calcu-
lated by dividing the selection interval for the h-th stratum by the
number of specimens collected in the city.
The method described above does yield selection probabilities
proportional to the population of the cities when certain precautions
are followed. Adequate methods for selecting alternate (substitute)
cities were not taken. Hence, the procedures followed in the study
alter the probability of selection for some cities, but the degree,
although probably minor, is undetermined. For the purpose of calculating
approximate weights, these special situations were treated the same as
all others.
5.2.2. Sample Sizes Used for Calculating Weights
As noted earlier, the weighted analysis was performed using only
fiscal years 1972-76. Fiscal year 1977 was excluded for the following
two reasons: (1) the initial data summaries received from EPA did not
include 1977 and (2) the sampling frames and primary sampling units
(PSU, the unit selected in the first stage of sampling) were changed in
-37-
-------
1977. In 1977 the PSU's were changed from cities with populations
greater than 25,000 to Standard Metropolitan Statistical Areas (SMSA's).
Not all data for fiscal years 1972-76 were used in the analysis.
Out of a total of 8,372 observations, only 5,880 were used. The exclu-
sion of 2,492 observations resulted for several reasons. Of these,
2,380 observations were volunteer contributions to the survey from
cities not even in the sample. These records were not members of the
survey design, and hence, their probability of selection cannot be
reasonably approximated. The other 112 observations were excluded for
one of several possible reasons; e.g., the percent extractable lipid may
have been to small or the confidence codes on the data file indicated
uncertainties about data quality. The precise rules were the following:
If the record indicated that a technical error had been made
for a particular residue amount, that residue amount was
considered missing.
If the confidence code for a particular residue amount was
blank, that residue amount was considered missing.
If the record indicated that there was less than 10 percent
lipid extractable material, all residue amounts were consider-
ed missing for that record.
5.3. Racial Comparisons by Weighted Analysis
5.3.1. Overview
This section presents results from weighted regression analyses
that compare whites with nonwhites over time (fiscal years 1972-1976)
adjusting for age (Age 1 < 14, 15 £ Age 2 £ 44, Age 3 >^ 45), census
region (CR1 = North East, CR2 = North Central, CR3 = South, CR4 = West),
and sex. Because the strata in fiscal year 1972 were census regions,
the census divisions within each census region were grouped together for
fiscal years 1973-76. Racial comparisons were performed for each of two
variables; they are
Percent of individuals with greater than 3 ppm "PCB's"
(percent >3 ppm)
Percent of individuals with detected "PCB's"
(percent positive detections (trace, 1-3 ppm and >3 ppm))
In summary, the analysis showed:
-38-
-------
For percent >3 ppm "PCB":
(i) Over the fiscal years 1972-76, there was an increasing trend
in the percentage of people with greater than 3 ppm PCB in
their adipose tissue. The rate of increase is estimated to be
1.73 percentage points per year, with a standard error of the
estimate of 0.54. A test of the hypothesis of no time trend
was rejected at the 1 percent significance (99 percent confi-
dence) level by the estimated trend and its standard error.
(ii) Over the fiscal years analyzed (1972-76), nonwhites averaged
7.15 percentage points higher than whites. The standard error
of the estimate is 1.41 and the racial difference is signifi-
cant at the 0.1 percent (99.9 percent confidence) level.
(iii) The hypothesis that there is a differential trend for whites
and nonwhites is not supported by the data. That is, the
white-nonwhite differential is reasonably constant over time.
For percent positive detections:
(i) There is an increasing trend (fiscal years 1972-76) of an
estimated 2.98 percentage points per year with a standard
error of 0.82. The estimated trend and its standard error
reject the hypothesis of no trend at the 0.1 percent (99.9
percent confidence) level.
(ii) Over time, differences between whites and nonwhites are not
significantly different from zero in either absolute level or
trend.
5.3.2. Statistical Method
Each individual for which a sampling weight could be calculated in
fiscal years 1972-76 was used in the analysis (except for quality-code
related exclusions). The analyses proceeded by defining an indicator
variable for each sample member. Specifically, the variables Y.. . , Y-.
were created for the i-th individual such that:
1 if the i-th sample member's adipose tissue had
>3 ppm PCB's and
Y
0 if <3 ppm PCB's,
1 if any PCB's detected and
Y =
1 0 if no PCB's detected.
-39-
-------
For each of these dependent (Y) variables, a linear model was
specified, which attempted to relate the Y variables to the sample
member's age, race, sex, census region, and year of tissue collection.
Specifically, Y was approximated by:
Y = M
2
I A.X.
. , 11
I CR £
= J J
.
J V;
(i)
where
X. =
i
1 if an individual is in age group i (i = 1,2),
0 otherwise.
1 if an individual is white,
0 otherwise.
XM =
1 if an individual is a male,
0 otherwise.
i. =
J
1 if an individual is in census region j (j = 1,2,3,4)
0 otherwise.
FY = Fiscal Year - 1970 (coding by subtracting 1970 from
fiscal year).
In the sample, the X.,..., FY are known and the coefficients
* L
(M,...,p.) must be estimated from the data and represent the "effects"
of an individual being a member of a particular age, race, sex combina-
tion living in a specific census region at a specific time.
In general, the model is simply a generalization of a covariance
model (with FY being the covariate) that allows a function linear in FY
to be fitted simultaneously for each combination of age, race, sex and
-40-
-------
census region. As an example, suppose two individuals are in the same
age group (age group 1, say), both are males, and both live in census
region 1 however, one is white and one is nonwhite. Then, for the white
individual:
Yw = M + A1 + Rw + SM + CRj + FY (PQ + PX + Pw + PM + P*) > (2)
and for the nonwhite individual:
Y^ = M + Aa + 0 + SM + CRj + FY (PQ + Pj + 0 + Pj, + P*) . (3)
Combining terms common to both individuals gives:
Y = M
*W "ill
Y = M
NW "ill
> Rw + C
. B1UFY
where MI and BI 1 are the sum of the terms common to both individuals
and (111) indexes the age group, sex, and census region of the indivi-
duals. Hence, the model is simply a formulation of a covariance model
that allows the estimation and testing of a differential FY slope (time
trend) by different subgroups of the population. Hence, by appropriate
manipulation of the slopes and intercepts of the fitted lines, differ-
ences between whites and nonwhites may be estimated and tested statis-
tically.
The parameters (the coefficients above) were estimated by using
weighted least squares, where an individual's weight was the sampling
weight discussed in Subsection 5.2. The variance-covariance matrix of
the estimated coefficients was computed by the use of Taylorized devia-
tions (Shah et al. 1978).
For the two variables analyzed (percent >3 ppm "PCB", percent
positive detections) the estimated coefficients are given in Table 5.1.
Following the notation developed in (4) and (5) above, for every
combination of age, sex and census region (denoted as i,j,k), an esti-
mated line can be produced for each race. They are of the form:
-41-
-------
JU
Table 5-1' Model Coefficients (Percent)
Percent >3 ppm "PCS"
Intercept
Terms
M
Al
A2
Rw
SM
CR1
CR2
CR3
-4.
3.
2.
-2.
1.
3.
9.
7.
68
15
68
79
20
40
38
78
Slope
Terms
PO 4-
p - ~ 2. •
Po ""1 •
PW -1-
PM °-
*
P! i-
*
Pf)
~ £, ,
*
Percent
Positive
Intercept
Terms
39
96
89
09
04
29
03
32
M
Al
A2
RW
SM
CR1
CR,
CR3
79
-5
1
2
6
-5
-6
-5
.84
.65
.96
.47
.00 -
.57
.06
.32
Detections
Slope
Terms
PO 3
PI "0
P2 -o
PW -°
PM -1
.<.
PI x
A
P2 l
->-
Po 0
.36
.87
.36
.38
.27
.22
.27
.95
*Based on data for fiscal years 1972-76 from NHMP computerized data
files, using methodology defined in section 5.2.
-42-
-------
For whites:
For nonwhites:
A /\ «%
Nw? ^ J J j ^ i T If T n k" *
Now, there are 3 age groups, 2 sexes and 4 census regions; therefore,
averaging over the 24 distinct lines in (6) and (7) gives:
Yw = M + ^ + (B + BW)FY , (8)
¥„, = M + B FY . (9)
where
" 324
Z Z Z
i=l j=l k=l
M = Z~ Z Z Z M (i, j, k)
3 2 4
B = 27- Z ^ Z B (i, j, k)
i=l =l k=l
For the data used in the analyses, (8) and (9) are:
For percent >3 ppra PCB,
YW = 2.99 - 2.79 + (2.28 - 1.09)FY ,
= 0.20 + 1.19 FY , (10)
A
Y^ = 2.29 + 2.28 FY . (11)
For percent Positive Detections,
-43-
-------
YW = 77.37 + 2.47 + (3.18 - 0.38)FY
= 79.8 + 2.80 FY , (12)
Y^ = 77.37 + 3.18 FY . (13)
To test for slope differences, it is sufficient to test &. = 0.
The tests are presented in Table 5-2.
Now assuming the slopes are not different for the two racial groups
(a hypothesis supported by the tests in Table 5-2), the difference
between the two levels may be estimated by averaging the yearly differ-
ence in the lines over the years. This is algebraically equivalent to
testing the difference in the lines at FY = 4 (1974). Table 5-3 gives
the results.
The tests given in Table 5-2 indicate that for both percent >3 ppm
"PCB" and percent positive detections, no differences exist between
whites and nonwhites with respect to their time trend (FY slope).
Therefore, a common trend is given as the average of the two race-
specific slopes. They are the following:
For percent >3 ppm PCB,
*. f\
PAVG = %(2.28 + 1.19) = 1.73, with SE(BAVG) = 0.54 ,
For percent Positive Detections,
••s ^
PAVG = 35(2.80 + 3.18) = 2.99, with SE(6AVG) = 0.82 .
Both of these common slopes are significantly greater than zero
(p < .01).
Figures 5-1 and 5-2 present the above estimates and analyses graphi-
cally. In addition to the trend lines discussed above, certain directly
adjusted white and nonwhite means are plotted. These are plotted merely
to present a sense of the data and are, roughly, what the race-specific
lines are predicting. To compute these means, the following procedure
was followed. First, weighted means were computed for each combination
-44-
-------
Table 5-2.' Tests for Differential Slopes—White Versus Nonwhite
Percent >3 ppm "PCB" Percent Positive
Detections
SE(fJw)
Z = PW/SE(PW)
Significance
-1.09
2.96
- .36
Not Significant
-.38
1.22
- .31
Not Significant
Based on data for fiscal years 1972-76 from NHMP computer accessible
data files, using methodology defined in Section 5.2.
Table 5-3. Tests for Average Racial Group Differences
Percent >3 ppm "PCB"
Detections
Percent Positive
Difference Between
^»
Lines (A) :White-Nonwhite
-7.15
.95
SE(A)
1.41
1.89
Z = A/SE(A)
Significance
-5.07
p < .001
0.50
Not Significant
Based on data for fiscal years 1972-76 from NHMP computer accessible
data files, using methodology defined in Section 5.2.
-45-
-------
Figure 5-1. Racial Comparisons Over Time: Percent > 3 ppm "PCB1
'% > 3 ppm "PCB" (Y)
20% -
15% -
10% -
5% -
I Nonwhite Trend Line (%)
Y = 2.99 + 2.28 FY
White Trend Line (%)
Y = 0.20 + 1.19 FY
1972
1973
1974
1975
1976 Years
LEGEND
O = Nonwhite estimated means adjusted for Age, Sex, and Census Region.
• = White estimated neans adjusted for Age, Sex, and Census Region.
FY = Years - 1970
STATISTICAL TESTS
1.
2.
Test
Slope Differences
White Versus Nonwhite
Average Distance Between
Estimate
-1.09
7.15
Standard Error
2.96
1.41
Lines
White and Nonwhite
3. Trend Summary
Average of White and
Nonwhite slopes
1.73
0.54
Significance
Not Significant
Highly Significant
(p < .001)
Highly Significant
(P < .01)
* Based on data for fiscal years 1972-76 from NHMP computer accessible data
files using methodology defined in Section 5.2.
-46-
-------
Figure 5-2. Racial Comparison Over Time: Percent Positive "PCS" Detections
% Positive
Detections ."PCB" (Y)
100% J
95%-
90%-
85%-
Average Distance Between
Lines (< 1%)
White Trend Line (%)
Y- = 79.8 + 2.80 FY
Overall Trend Line (%)
Y = 73.6 + 2.99 FY
Nonwhite Trend Line (%)
Y = 77.4 + 3.18 FY
(67.7%)
1972
1973
1974
1975
1976 Years
LEGEND
a =» Nonwhite estimated means adjusted by Age, Sex, and Census Region.
• =» White estimated means adjusted by Age, Sex, and Census Region.
FY = Years - 1970
STATISTICAL TESTS
Estimate Standard Error
1. Slope Differences
White Versus Honwhite
2. Average Distance Between
Lines
White and Nonwhite
3. Trend Summary
Average of White and
Nonwhite Slopes
- .38
.95
2.98
1.22
1.89
.82
Significance
Not Significant
Not Significant
Highly Significant
(P < • oi);
Based on data for fiscal years 1972-76 from NHMP computer accessible data
files using methodology defined in Section 5.2.
-47-
-------
of age, race, sex, census region, and fiscal year. There are potentially
240 (=3x2x2x4x5) such means. However, due to the absence in
the sample of young nonwhites in certain years in census region 4, only
234 means were computed. In essence, a five-dimensional table with one
entry per cell was produced (the weighted means). Next, for each race
and year, the simple average of the weighted averages was computed.
These averages are rough estimates of the means by race and year. It is
not difficult to show that the race-specific lines drawn in Figures 5-1
and 5-2 and these means estimate the same quantity if the assumed model
form (1) is correct. Finally, these means are plotted along with the
estimated lines in the two figures.
5.4. Comments on Assumptions
Several apparent inconsistencies in the data cast doubt on the
assumptions. The most striking is the very low value of percentage
detected for nonwhites in 1973. The 67.7 percent is more than three
standard deviations below the predicted mean value of 87.6 percent (see
Figure 5-2). Thus this low value is not likely to be due to random
variation in the data. The change in the chemical analysis methods in
1974 may contribute to the higher percentages for later fiscal years
than for early fiscal years, and the trend over time may be in part due
to the change in chemical analyses.
-48-
-------
References
Biocca M, Moore JA, Gupta BN, McKinney JD. 1976. Toxicology of selected
symmetrical hexachlorobiphenyl isomers: I. Biological responses in
chicks and mice. In: National Conference on Polychlorinated Biphenyls.
1975. Chicago IL. Ayer FA, ed: 67-72. EPA-510/6-75-004.
Cochran WG, 1977. Sampling Techniques. New York, NY: John Wiley and
Sons.
Hansen MH, Hurwitz WN, Madow WG. 1953. Sample Survey Methods and Theory,
Vol. 1 and 2. New York, NY: John Wiley and Sons.
Kish L. 1965. Survey Sampling. New York, NY: John Wiley and Sons.
Kutz FW, Yobs AR, Strassman SC. 1976. Organochlorine pesticide residues
in human adipose tissue. Society of Pharmacological and Environmental
Pathologists 4(1):17-19.
Lewis RG. 1977. Organic trace analysis—are we on target? Nevada
Section meetings, American Chemical Society. Las Vega, NV.
Matthews HB, Anderson M.'1976. PCB chlorination versus PCB distribution
and excretion. In: National Conference on Polychlorinated Biphenyls.
1975. Chicago, IL. Ayer FA, ed: 50-56. EPA-510/6-75-004.
-49-
-------
McKinney JD. 1976. Toxicology of selected symmetrical hexachlorobiphenyl
isomers: correlating biological effects with chemical structure. In:
National Conference on Polychlorinated Biphenyls. 1975. Chicago, IL.
Ayer FA: 73-75. EPA-510/6-75-004.
Mood AM, Grayhill FA, Boes DC. 1974. Introduction to the Theory of
Statistics. New York, NY: McGraw-Hill.
Mulhern BM, Cromartie E, Riechel WE, Belisle AM. 1971. JAOAC 54: 548-550.
Shah, BU, Holt MM, Folsom RE. 1978. Inference about regression models
from sample survey data. In: International Association of Sampling
Statisticians. New Delhi.
Stahl E, ed. 1969. Thin Layer Chromatography, a Laboratory Handbook.
New York, NY: Springer-Verlag, Inc.
USBC. 1972. U.S. Bureau of the Census. Census of population: 1970
general population characteristics, final report PC(1)-B1, United States
summary. Washington DC: U.S. Government Printing Office.
USEPA. 1972a. U.S. Environmental Protection Agency. Listing of cities
with populations greater than 25,000 for the Middle Atlantic Census
division, based on 1970 Census. Washington DC: Office of Pesticides
and Toxic Substances, U.S. Environmental Protection Agency.
-50-
-------
USEPA. 1972b. U.S. Environmental Protection Agency. National human
monitoring program--1972. Washington DC: Office of Pesticides and Toxic
Substances, U.S. Environmental Protection Agency.
USEPA. 1973. U.S. Environmental Protection Agency. Listing of eligible
cities with cumulative populations, selection intervals, and random
start numbers. Washington, DC: Office of Pesticides and Toxic
Substances, U.S. Environmental Protection Agency.
USEPA. 1974. U.S. Environmental Protection Agency. Analysis of
pesticide residue in human and environmental samples. Research Triangle
Park, NC: U.S. Environmental Protection Agency.
USEPA. 1977a. U.S. Environmental Protection Agency. Analysis of pesti-
cide residue in human and environmental samples, section 9D. Research
Triangle Park, NC. U.S. Environmental Protection Agency.
USEPA. 1977b. U.S. Environmental Protection Agency. Tables of
frequency of levels of polychlorinated biphenyls in human adipose tissue
by geographic area and race, wet weight basis for fiscal years 1972-76.
Washington, DC: Office of Pesticides and Toxic Substances, U.S.
Environmental Protection Agency.
USEPA. 1977c. U.S. Environmental Protection Agency. Computer printout:
computer analysis of PCB data by Statistical Packages for the Social
Sciences (SPSS). Washington, DC: Office of Pesticides and Toxic
Substances, U.S. Environmental Protection Agency.
-51-
-------
USFDA. U.S. Food and Drug Administration. , Pesticide Analytical Manual,
Vol. 1, Section 300.63. Washington, DC: U.S. Food and Drug
Administration.
Yobs AR. 1971. The national human monitoring program for pesticides.
Pesticides Monitoring J. 5(l):44-46. Reprinted by: U.S. Environmental
Protection Agency.
-52-
-------
APPENDIX A
Letter from M. Aaronson
(Reproduced with permission of S. Strassman-Sundy)
A-l
-------
January 7, 1980
Mrs. Sandra Strassman
Project Officer
National Human Monitoring Program
Field Studies Branch
Survey and Analysis Division (PS-793)
U.S. E.P.A.
401 "M" Street S.W.
Washington, D.C. 20460
Dear Sandy:
Enclosed please find a copy of the following items:
1) Reconstructed ion chromatogram (RIG) of Tissue Pool sample 79-A,6%
2) The mass spectrum of scan #266
3) The library search of scan #266
4) A comparison of the mass spectrum of scan #266 to that of
heptachlorobiphenyl
The RIC begins at pp-DDT (scan #51) and runs for twenty minutes. Scan
#266 is the peak utilized by the laboratory for the semi-quantitative
estimation of PCB's in tissue extracts. Further examination of the
peak represented by scan #266 tentatively identifies it as a heptachloro-
biphenyl. The actual heptachlorobiphenyl standard would be required
to produce a more definitive identification.
If you have any further questions or if I can be of more assistance do
not hesitate to call.
Sincerely
Michael J. Aaronson
A-2
-------
BIC
10/10/79 10:50:00
TISSUE pnm.
100.0-1
BIC
DATA: MS65 H263
CALI: 10579 113
I. 600 LABEL: » 0. 4.0 (HJAN: A 0. 1.0 BASE: U 20. 3
100
3:20
209
6:40
300
10:00
• 400
13:20
5i
16
-------
riASS SPECTRUM DATA: HS65 U266
10/IO/79 10:50:00 + 8:52 CALI: 10579 113
SAMPLE: TISSUE P001.
11263 TO »269 SUUI1ED - H253 TO «258 - H274 TO 11279 XI.CO
BASE H/E: 30-i
RIG:
••>•*• "fit »
J J.* v'*.
100.On
50.0-
3316.
254
73..
n/E so
100.0-1
100
120
HO
T*7
1. - I A f\*~w S\ 4 ft ^.-*J*
44 I 182 197 216 |
^9 1|9 |27 .III, 15711 ,i[ ,.,11.203 |.| I ,1 I 247
i I i i i "I i i > I rl'iTi I'l'l'l'i'i i i ri'i1 'I'l'i'i i i I'l'l'l'l i I'rl'J ri'i I'l'r'i'i'i r» I'l'i'i i i | i ri'l'i'i i i11
180
200
220
240
260
50.0-
239
2821
r-r-|-r-<
«0
315
359
TH"
20
:6'J
00
T"
40
-------
LIBRARY SEARCH
10/10/79 10:50:00 + 8:52
SAMPLE: TISSUE POOL /?
* 263 TO * 269 SUMMED -
DATA:
CALI:
MS65 * 26S
10579 * 3
BASE IVE: 394
RIC: 29727.
* 253 TO * 258 - * 274 TO * 279 XI.00
25409 SPECTRA IN LIBRARYN9'SEARCHED FOR MAXIMUM PURITY
54 MATCHED AT LEAST 3 OF THE 16 LARGEST PEAKS IN THE UNKNOUN
RANK IN NAME
1 20174 1.1'-BIPHENYL.2,2'.3.4.5,5',6-HEPTACHLORO-
2 9039 1.3,5-TRIAZINE.2,4,6-TRIS(TRICHLOROMETHYL:i-
3 23034 BENZENE,1,1'-THIOBIS-.TETRACHLORODERIV.
4 17181 FERROCENE.l,l',2,2'-TETRflCHLORO-
5 16534 1.1'-6IFH£NYL,2.3',4,4',5-PENTACHLORO-
RfiSK FORMULA
1 C12.H3.CL7
2 CS.H3.CL9
3 C12.HG.S.CL4
4 C1G.H5.CL4.FE
5 C12.H5.CL5
MASS INTEN
36
39
47
49
59
51
56
63
73 0
M.UT B.PK
392 394
PURITY
764
FIT
945
RFIT
795
f -^
75
77
62
84
91
93
94
97
99
1C3
103
112
1! !
117
119
125
126
127
131
132
133
134
135
135
142
143
144
145
U'3
147
143
157
160
17
19
13
10
5
8
12
18
13
44
60
43
51
30
16
19
2
153
231
63
14
4
23
21
35
34
210
129
133
119
47
187
80
61
98
84
37
66
429 398
322 324
322 324
324 326
338
309
215
185
568 445
935 327
780 261
743 199
4
75
78
543
95
41
49
231
215
196
552
343
275
265
15
19
7
171
54
55
49
43
73
274
103
67
A-5
-------
LIBRAIJY SEARCH
10/10/79 10:50:00 + 8:52
SAHPLE: TISSUE POOL
II 263 TO II 269 SUI1HED - » 253 TO it 258 - « 274 TO (I 279 XI.00
DATA: ttS65 tl 266
CAL1: 10579 it 3
BASE ll/E: 3S4
RIG: 29727.
1000 -i
SAHPLE
•
C12.H3.CL;
1000-1
H UT 392
B PK 394
RANK 1
IH 20174
PUR 764 .
1000-
0
mnn -
1UUU
H/E
\^
, ,1,1,. . ill,. ...n . .,n. ., ii,(|. ,ii, Ji!/ ilh .mi.
j ' ' ' ' ' ' ' ' ' j' I'-IBIPUEMYL 2 2* 3 45 51' G-HEPTACHLORO- ' '
II II
l| II . il| III . n, . Ill .|,
SAHPLE HINUS LIBRARY ' ' '
i i. . ... yi q.ll. i ,01,1 ,,|ji i i 1 lillh ll.ll . I.I.I. , . lll.l . ill, . |
• I i I • I i | i | i | i (— r— j'"-r | •!-'•>- !• > | • | i | < | • | -t 1 — •- j •• r-'T'-i 1 1 r" • 1 • | ' 1 ' 1
100 150 260 250 300
ill.
• • I i i i i
!
llh
1
• I • I • I • i - . • i • .
350
-------
APPENDIX B
Letter from John D. Tessari
(Reproduced with Permission)
B-l
-------
Colorado State University
College of Veterinary Medicine Fort Collins, Colorado
and Biomedical Sciences 80523
Institute of Rural Environmental Health
Colorado Epidemiologic Pesticide Studies Center
Spruce Hall
September 21, 1979
Dr. Mltchel Erickson
P.O. Box 12194
Drafysus Lab
Research Triangle Institute
RTF- North Carolina 27709
Dear Dr. Erickson:
Enclosed find the gas chromatographic charts you requested
with our estimate of the PCS content of those samples. We
randomly chose 38 of our most recently analyzed samples. Also
find enclosed a chart showing sample numbers and data.
The calculation method utilizes the isolated la"rge PCS peak
(Arochlor 1260) eluting 8-12 cm beyond p,p'-DDT on 1.5% OV-17/
1.95% OV-210 (6.0-6.9 relative retention time to aldrin) and on
4% SE-30/6% OV-210 (4.6-5,0 relative retention time to aldrin).
A comparison is made between the peak height of 40 pg of aldrin
(5 ul of an 8 pg/ul aldrin standard) and the peak height of this
PCB peak in 1 mg of sample (5 ul of a 200 ug/ul dilution of
sample) according to the following relationships
peak height (mm) of 40 pg aldrin
= FACTOR
peak height (mm) of PCB peak in mg of sample
The factor determines the amount of PCB's present, which is
reported to the EPA by code letter, according to the following:
FACTOR PCB's Present Code Letter
No peak present 0 PPM V
> 3.5 <1 PPM W
3.5-2.6 1-3 PPM Y
< 2.6 >3 PPM Z
This method is based on the TLC method of Mulhorn et al, (1)
which has a precision of - 50% when using Arochlor 1260 as the
reference standard. A series of representative adipose samples
were analyzed for PCB's by this method, and for all chlorinated
pesticides by the more accurate standard adipose tissue method
(2). A direct correlation was noted between the PCB result by
TLC and the height of the afore mentioned peak. Finally, a
relationship was established between the ratio of a specified
B-2
-------
-2-
amount of aldrin and the PCB peak height and concentration in
order to produce the factors. Thus, our method essentially relates
the relative peak height of the PCB peak to aldrin to obtain a PCB
concentration.
Due to the semi-quantitative nature of the method, certainly
no'better than the reference method which claims - 50% precision,
discrepancies between the two GLC columns occassionally arise.
For example, 1-3 PPM may be noted on one column and <1 PPM on
another. Judgement is applied to the individual situation to
determine which result to report. A factor of 3.5 (just in the
1-3 PPM range) from one column and a factor of 4.0 (<1 PPM) would
probably be reported as <1 PPM. A factor of 3.0 (1-3 PPM) and
a factor of 3.6 (just <1PPM) would probably be reported as 1-3
PPM. If the discrepancies are extreme (<1 PPM on one column, >3
PPM on the other), we attempt to resolve the difference. If no
errors can be found, an average is usually reported (1-3 PPM).
If you have any questions please contact us.
John D. Tessari
Supervising Chemist
Laboratory Director
(1) Analysis of Pesticide Residues in Human and Environmental
Samples, U.S. EPA, R.T.P. N.C.,,. 1974, Sect 9,D.
(2) Ibid., Section 5,A,C1.).
B-3
-------
factor
CPM
tter
facf
3-5
Jo. o
£7
n
r-7
*•/
3.2
22-7
r.s.
H-l
.2.3
c/
>3
>3
>3
>3
ut/'
y
IA/
(A/
(A/
W
\fj
Y
w/
Y
iV
IA/
uJ
z
T
~z.
-z,
•z.
/ 2-
3-6
Y.
5-0
za.
5.7
/•*?
A?
3-7
1-3
>3
>3
7-3
>3
Z.
Z.
(AJ
Y
(A/
Y
~v
Y
Z
•2.
Y
z.
u/
Y
/-J
Y
y
tA/
IA/
(A/
Y
Y
Y
~z_
Y
Y
z
Y
-z.
•R-A
-------
APPENDIX C
Interlaboratory PCS Analysis Results
(Reproduced with permission of R. G. Lewis)
C-l
-------
SUBJECT:
:sTAL TOXICOLOGY DIVISION
HEALTH EFFECTS RESEARCH LAEOSATCKY
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
DATE: August 17, 1977
Water Blind Sample No.
49
FROM: Chief, equality Assurance Section,
HSRL, ETD, ACS (MD-69) , RTP, NC 27711
TO: All Participating Laboratories
On June 3 a water round robin sample (No. 49) was mailed to 25
laboratories who had previously signified they wished to participate
in this interlaboratory excercise. Nineteen laboratories mailed in
their analytical data, and a brief summary report was mailed back to
all 25 laboratories on July 8 so' that the formulation would be avail-
able to all, coincidental with the receipt of the 6-m.cnth water SPRM
in acetone which v:as mailed June 28. •• .
GEKT^AL COMMENTS: •
We think that most of you will agree that this was.not the easiest
sarrcle possible nor was it intended to be.. We think, however, that
it was probably no more difficult than many routine samples; in fact, it
was probably somewhat less difficult than many monitoring samples of
unknown composition because of the spike with intact parent compounds.
Routine environmental samples are far more likely to contain-partially
altered compounds.
We were somewhat disappointed that five Labs failed to detect
both DOT metabolites, and seven Labs failed on oxychiordar.s. Five
Labs overlooked HCB, and to their credit all except three Labs identi-
fied the PC3 (Aroclor1254). Those three who overlooked this compound
shonld be deeply concerned because the chromatographic fingerprint
should have been unmistakable on just about any GC ccluirn cr;£ chess
to use. Similarly, the HCB peak should have bean most glaringly '
apparent. Two of the Labs who failed on Arcclcr 1254 thought they
saw all sorts of other things--dieldrin, Kepcr.e, and what have you.
G.C. column selection was undoubtedly one kay factor in the failure
to detect certain compounds in the presence cf the multitude of A::oclor
peaks. Hut another, and we think more cogent factor, was the failure to
apply a high degree of reasoning in the interpretation of che available
cliror.-.atogrciphic data. By way of illustration,- p,p'-DDT ar.d a major
peak of Aroclor 1254 nearly completely-superimpose "on the OV-17/OV-210
column operated at 2C.O"C. One laboratory had a peak i»: their sample .-•>;
extract which calculated an FJ
-------
-2-
this Aroclor peak had a base width of 19 nzn, a most clear indication
that something else besides the Aroclor was elutir.g frcra the sample
extract at this retention site. The prime suspect, of course, would •
be p,p'-DDT.
We could document a dozen more similar illustrations, but we
believe the point sho'-uld be clear that in the process of interpretation,
a number of factors.must be weighed, to a much greater extent than
simply measuring retentions.
QUANTITATION;
The most glaring irregularity in qxiantitation was experienced by
Lab Code 16 wherein the reported values were apparently in error by
an approximate factor of x 100. This laboratory was requested to review
their reported data but the values reported back were still far off.
The very best that could be done with HCB is about 85% recovery.
Therefore, no Lab should be expected to report a value in excess of
0.25 ppb. It was found in our Lab that the 4% SE-30/6% OV-210 was
most suitable for quantitating the HCB as this was the only column
producing baseline separation from the solvent peak.
Rejecting those outlier values designated by asterisks in Table 1,
the mean recovery values from all Labs, with respect'to formulation,
were as follows:
HCB 67%
Oxychlordane 83
p,p'-DDE — •— 85
p,p'-DDT 86
Aroclor 1254 81
Considering the relative difficulty of the sample, we think the
overall performance was not too bad. In terms of interlaboratory
precision, relative standard deviation values in the range of the
calculated values were quite acceptable. On less complex formulations
"Total Error" values in the under 50% range are generally considered .
satisfactory. Cn this particular sample, however, the calculated
values could not be considered too far out of line.
•
Our apologies to Lab Code 23 for overlooking their reported value
for p,p'-DDE. This correction is reflected in the attached corrected
summary (Table 1).
RELATIVE PSRFOr-yAIICE; .
Table 2 shows the relative overall performance of the 19 partici-
pating laboratories. The method for this computation may be found in
our Quality Control Manual, Chapter 2.
C-3
-------
-3- .
Tho3
-------
TABLE 1
INTF.RIJVGORATORY CUFCK bAMI'LK NO. 49, KATEP - SUX.".r.aY OF RESULTS
1J\{\
r:^fjf.
Forculatlon
8
16"
15
6 -.•
7
11
34
9
16
26
13
3 ' '
?3
1
24
25
5
12
10
>
Over. -ill Mean
Standard Deviation
Relative Std. Devlat., Z
Totbl Error, Z
PESTICIDES KITGUTED IN MICi
-------
•7 -• U'viTED STATES ENVIRONMENTAL PROTECTION AGENCY
DATE: January 29, 1980
SUBJECT: Fat Blind No. 70, Check Sample Report
FROM: Chief, Quality Assurance Section (MD-69) /? J?
ACB/ETD/HERL /L '"'
Research Triangle Park, NC 27711
TO: All Laboratories Receiving SPRM No. 70
On December 3, a fat blind sample was sent to the 22 laboratories
which had previously signified they wished to participate in the inter-
laboratory exercise. Eighteen laboratories returned reports of analy-
tical results. One Epidemiology Studies laboratory did not participate.
General Comments
This check sample study was designed to measure proficiency of a
laboratory in recognizing and quantitatively determining PCB contami-
nation in an adipose sample containing common organochlorine pesticide
residues. The.pesticides and fortification levels for SPRM No. 70 were
derived from national surveys of residues in human adipose tissue. This
exercise therefore represented a realistic analytical problem that might
be encountered in a residue laboratory.
Specific comments concerning submitted sample report sheets and
chromatograms may be found in individual reports.
Results and Discussion
Table 1 presents a summary of the standard pesticide reference
material (SPRM) formulation and analytical results. Laboratories above
the double line are Epidemiclogy/Human Monitoring contract laboratories.
This sample proved to be a difficult challenge for several of the
participants. Performance scores of 190 points or above (out of 200
possible) were obtained by only five laboratories. Eight of the eighteen
total respondents and four out of the eleven "epidemiology/human moni-
toring" laboratories failed to recognize the PCB residue pattern. Analysts
that did not recognize the PCB pattern often misidentified PCB peaks as
pesticides.
C-7
-------
The percent total error (TE) figures in Table 1 demonstrate a
generally unacceptable level of performance for this analysis. Five of
the nine TE figures were greater than 50% and therefore unacceptable
(for explanation of total error see EPA "Manual of Analytical. Quality
Control for Pesticides in Human and Environmental Media" section 2K).
Total error figures for only the EPA contract laboratories are as follows:
HCB, 53.1%; e-BHC, 16.4%; oxychlord^ns, 45.7%; t-nonachlor, 22.3%; Kept.
Epox., 43.5%; p.p'-DDE, 46.7%; p.p'-DDT, 30.1%; dieldrin, 42.7%; and PCB
1254, 120%.
GC column pairs used for analysis and confirmation were generally
OV-17/OV-210 and SE-30/OV-210. These are both good columns but gen-
erally do not make a good complementary pair for qualitative and quan-
titative confirmation. A better suggestion for most residues would be
either of the two mixed columns (SE-30/OV-210 or OV-17/OV-210) paired
with the. 5% OV-210. The apparent best selection for this sample No. 70
would have been OV-17/OV-210 paired with the OV-210. Pairing of either-
mixed phase with OV-210 unfortunately entails GC runs at 200 and 180°C
and necessitates either two GC's or change of column temperature and re-
run of sample extracts and standards.
Laboratories using silica columns on the 6% Florisil fraction to
effect a separation of PCB's and chlorinated pesticides appeared to
experience some difficulty with achieving proper fractionation. PCB's
sometimes eluted in the pesticide fraction causing a PCB peak to be
misidentified as a pesticide, i.e., o,p'-DDT.
At least two laboratory reports indicated some Florisil fractiona-
tion problem. These problems should be resolved as soon as possible.
Florisil that is too retentive could result from (1) improper activation
temperature, (2) improper percent of ethyl ether in pet. ether, and (3)
ethyl ether that does not contain the required 2% ethanol (read the fine
print analytical information on the can or bottle). Florisil that
appears insufficiently retentive or inactive might result from (1) or
(2) above, and (3) residual amounts of a polar solvent in the sample or
standard being placed on the column. Likely candidates here could be
acetonitrile from the sample partition cleanup (if the water-out steps
where not performed properly) or incomplete removal of benzene (or other
solvent more polar than hexane) from a standard solution placed on the
column. Other sources of Florisil problems are undoubtedly possible.
Relative Performance
Table 2 shows the overall performance ranking of the eighteen
participating laboratories. A possible score of 200 points is derived
from 100 points each for qualitative and quantitative analytical results.
Co
—O
-------
General Recommendations
Residue analysts should become familiar with PCS elution patterns
from their common GC columns. Chromatograms could be exhibited on a
wall.or filed in some other convenient manner for easy referral.;
The 5f; OV-210 column should he used routinely for identity con-
firmations. ' '
Laboratories should investigate the various schemes that have been
offered in the literature for separation of PCB's from other chlorinated
pesticide residues. Silicic acid column chromatography of the 6% Florisil
fraction to achieve this separation is being used by the EPA human milk
monitoring program.
CC: Dr. V. R. Hunt
Dr. F. W. Kutz
Ms. S. C. Strassman-Sundy
Ms. Madeline Dean
Dr. Hale Vandermer
Dr. C. W. Miller
Dr. Lee Leiserson
Dr. John Kliewer
Dr. W. F. Durham
C-9
-------
TABLE 1
INTERLABORATORV CHECK SAMPLE NO. 70, FAT-SUMMARY Of RESULTS
LAI!
cow:
4
5
Formulation
7
8
10
11
1?
14
24
25
26
51
1
6
9
13
16
38
52
Overall Mean
Standard Deviation
Relative Std. Devlat., X
Total Error, %
Ave. % Recovery
(*0utl ier)
PESTICIDES REPORTED IN PPM
IICU
0.061
0.031!
N 0
0.035
0.038
0.041
0.032
0.042
0 032
0.049
0.042
0.022
0.04
0.06
0.04!i
0.042
0.024
0.045
0.043
0.014*
0.040
0.009
23.0
65.0
66.7
G-WIC
0.25.
0.278
R E P 0 R
0.269
0.22
0.244
0.23
0.278
0.234
0.180
0.258
0.237
0.25
_
„
0.264
0.242
0.212
0.223
0.119*
0.24
0.03
11.0
24.8
96.0
OXYCIILOR-
tlANF
0.10
0.114
T
0.079
0.10
0.094 .
0.075
0.117
0.057
0.074
0.105
0.105
0.09
0.16
_
0.182*
0.083
0.122
0.058
0.036
0.092
0.030
32.6
68. 0
92.0
/•IVIM.'i-
NflN/lf 111 OR
0.15
0.156
0.175
0.16
0.154
0.12
0.148
0.151
0.130
0.152
0.13
0.15
0.11
0.12
0.208*
0.143
0.165
0.116
0.115
0.14
0.02
14.0
32.4
93.3
1IEPT.
FPOXIDE
0.081
0.092
0.091
0.105
0.095
0.086
0.096
0.062
0.080
0.090
0.059.
0.10
0.03
0.064
0.086
0.094
0.073
0.053
0.015*
0.080
0.02
25.1
51.1
98.8
P.P'-DDE
3.50
3.39
3.289
3.31
3.212
3.3
2.148
2.097
2.40
3.964
2.96
3.10
3.36
3.40
2.577
2.061
2.98
2.992
2.166
2.93
0.56
19.0
48.1
83.7
P.P'-DDT'
0.60
0.675
0.705
0.59
0.581
0.68
0.741
0.481
0.470
0.559
0.588
0.60
0.80
0.58
0.819
0.575
1.01*
0.503
0.403
0.61
0.12
18.9
39.9
101.7
niFIDRIN
0.13
0.162
0.149
0.133
0.081
0.13
0.132
0.082
-
0.152
0.148
0.13
0.13
0.12
OJ02
0.029
0.124
0.118
0.019*
0.12
0.03
27.7
58.8
92.3
AROCLOR
1254
1.00
1.99*
+
1.40
-
-
-
-
0.505
1.328
0.943
0.96
-
1.00
.
_
-
1.00
0.519
0.96
0.32
33.8
69.1
96.0
NON- SPIKE
O.P'-ODT = 0.057
O.P'-DDT Aldrin
n nfi? n ni4
Aldrin
rt 01 H
O.P'-DDT
n.ns?
Vn4RDT VnHP
O.P'-DDT Aid. Lindarifi
n 14 0.02 0.05
n
i
-------
TABLE 2
CHECK SAMPLE NO. 70, FAT-RELATIVE PERFORMANCE RANKING ''
LAB
CODE
51*
8*''
26*
38,
4*
52
25*
11*
7*
16
14*
13
9
12*
24*
6
10*
1
COMPOUNDS
MISSED
0
0
0
0
0
0
0
1
0
1
1
1
1
1
1
2
1
2
FALSE
IDENTIFICATION
0
0
0
0
0
0
2
0
1
0
0
0
0
1
1
0
2
3
IDENTIFICATION
SCORE
100
100
100
100
100
100 .
77.78
88.89
88.89
88.89
88.89
88.89
88.89
77.78
77.78
77.78
66.67
44.44
QUANTITATION
SCORE •'
95.57
92.89
91.17
90.37
90.24
73.80
94.03
81.30
80.99
79.14
77.68
77.12
75.95
80.62
78.38
73.33
83.21
69.10
TOTAL
SCORE
195.57
192.89
191.17
190.37
190.24
173.80
171.81
170.19
169.88
168.03
166.57
166.01
164.84
158.40
156.16
151.11
149.88
113.54
*Epidemiology/Human Monitoring Contract Laboratory
C-ll
-------
Recalculation of Fat Check Sample No. 70.
1) All quantisations including Lab #4 which was termed "outlier"
n 9
Mean 1.07
SD 46
RSD (%) 43
'Accuracy Mean 107%
Actual
2) All reporting laboratories with "zero" included for those who did
not report Aroclor 1254
n 18
Mean 0.65
SD 0.62
RSD (%) 96
Mean
Actual
A Mean ,_0,
Accuracy 65%
C-12
-------
APPENDIX D
Breakdown of Census Regions and Divisions by State
(Provided by EPA)
D-l
-------
Census Breakdowns of the United States
Region
North East
Division
New England
North Central
Middle Atlantic
East North Central
West North Central
South
South Atlantic
East South Central
West South Central
States
Connecticut
Maine
Massachusetts
New Hampshire
Rhode Island
Vermont
New Jersey
New York
Pennsylvania
Illinois
Indiana
Michigan
Ohio
Wisconsin
Iowa
Kansas
Minnesota
Missouri
Nebraska
North Dakota
South Dakota
Delaware
District of Columbj
.Florida
Georgia
Maryland
North Carolina
South Carolina
Virginia
West Virginia
Alabama
Kentucky
Mississippi
Tennessee
Arkansas
Louisiana
Oklahoma
Texas
D-2
-------
Census Breakdowns of the United States (Continued)
Region Division States
West Mountain Arizona
Colorado'
Idaho
Montana
Nevada
.New Mexico
Utah
Wyoming
Pacific . Alaska
California
; . Hawaii
Oregon
Washington
D-3
-------
APPENDIX E
Survey and Site Quotas for Fiscal Years 1972-1976
(Provided by EPA)
E-l
-------
Census Divisions
National Human Monitoring Program
Age, Race, Sex Distributions
New England (1)
Middle Atlantic (2)
East North Central (3)
West North Central (4)
South Atlantic (5)
Age Groups
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
Sex
M
4
5
_4
13
# Negroes = 1
4
5
_4
13
# Negroes = 3
4
5
^4
13
- # Negroes = 3
4
5
_4
13
// Negroes = 1
4
5
_4
13
// Negroes = 6
F
4
5
_5
14
3
6
_5
14
4
6
_4
14
4
5
_5
14
4
6
_4
14
Total
8
10
_9
27
7
11
_9
27
8
11
_8
27
8
10
_9
27
8
11
_8
27
E-2
-------
Census Divisions
National Human Monitoring Program
Age, Race, Sex Distributions
Age Groups
East S'outh Central (6)
0-14
15-44
45+
Total:
West South Central (7)
0-14
15-44
45+
Total:
Mountain (8)
0-14
15-44
45+
Total:
Pacific (9)
0-14
15-44
45+
Total:
Sex
M
4
5
_4
13
# Negroes = 5
4
6
_3
13
# Negroes = 4
4
6
_4
14
# Negroes = 1
4
6
_4
14
# Negroes = 2
F
4
6
_j4
14
4
6
_4
14
4
5
_4^
13
3
6
_4
13
Total
8
11
_8
27
8
12
_8
27
8
11
_8
27
7
12
_8
27
E-3
-------
National Human Monitoring Program
Collected by Census Division
Sex
Age Groups
M
Total
New England (1) -
4 Collection Sites - 4%
s
Middle Atlantic (2) -
14 Collection Sites - 19%
East North Central (3) -
14 Collection Sites - 19%
West North Central (4) -
5 Collection Sites - 7%
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
16
20
16
52
# Negroes
56
70
56
182
# Negroes
56
70
56
182
// Negroes
20
25
20
65
# Negroes
16
20
20
56
= 4
42
84
70
196
= 42
56
84
56
196
= 42
20
25
25
70
= 5
32
40
M
108
98
154
126
378
112
154
112
378
40
50
£5
135
E-4
-------
National Human Monitoring Program
Collected by Census Division
Sex
South Atlantic (5) -
11 Collection Sites - 15%
•*
East South Central (6) -
7 Collection Sites - 9%
West South Central (7) -
7 Collection Sites - 9%
Mountain (8) -
3 Collection Sites - 4%
Pacific (9) -
10 Collection Sites - 13%
Age Groups
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
0-14
15-44
45+
Total:
M
44
55
44
143
# Negroes
28
35
28
91
# Negroes
28
42
21
91
# Negroes
12
18
12
42
# Negroes
40
60
40
140
# Negroes
F
44
66
44
154
= 66
28
42
28
98
= 35
28
42
28
98
= 28
12
15
_12
39
= 3
30
60
40
130
= 20
Total
88
121
88
297
56
77
56
189
56
84
49
189
24
33
24 -
81
70
120
80
270
E-5
-------
National Human Monitoring Program
Collected by Census Division
Sex
Age Groups
M
Total
Percent
Summary:
0-14
15-44
45+
Total:
300
395
293
988
576
833
616
1037 2025
28.4
41.1
30.4
99.9
# Negroes = 245
Percent = 12%
Total Males = 49%
Total Females = 51%
Total Negroes = 12%
E-6
-------
APPENDIX F
EPA Guidelines and General Information About Collecting
Adipose Tissue for the National Human Monitoring Program
for Pesticides
F-l
-------
f
? £* \ • • . •
j UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON. D.C. 20460
Guidelines and General Information
About Collecting Adipose Tissue
For the National Human Monitoring Program for Pesticides
The National Human Monitoring Program for Pesticides is responsible
for determining, on a national basis, the incidences, levels and other
evidences of exposure to pesticides in the general population of the
United States. At present, the program collects and analyzes adipose
tissues for selected pesticides and their metabolites known to be stored
in'the lipid portion of these tissues. The results from the program are
ysed in evaluating various factors and conditions pertaining to human
health and effective pesticide regulation..
The adipose tissue for this program is secured through the coop- •
eration of participating pathologists and medical examiners located
throughout the continental United States. The tissue is obtained from
surgical specimens previously excised for pathological examination and
from postmortem examinations. The specimens are sent to the program
office in Washington, D. C., from which they are subsequently forwarded
to contract laboratories for chemical analysis. Periodic reports of the
laboratory results are sent to each participating pathologist for the
tissues which were submitted under his auspices. Summaries comparing
results with other regions of the country are also provided as they
become available.'
.F-2
-------
In order to develop valid information on a national basis, col-
lections must be made according to an experimental design which dictates
the number of samples required according to the demographic distribution
of the population in the appropriate census division. You should have a
copy of the annual quota of samples expected to be collected from your
location on a fiscal year basis. All collections should be made
according to this age/sex/race distribution. You should be able to.
^ • '
collect the number of samples required in each category. Since our
total sample is relatively small and the validity of the results depends
on a high response rate, your participation.is particularly important.
If you feel that you will be unable to collect the number of samples
required, please let us know.
Criteria for Selection of Patients tp_ be_ Sampled
Since the program objective is to reflect pesticide incidences and
levels in the general (man-on-the-street) population, a few suggestions
are listed here for your guidance:
• The highest priority should be given to satis-
fying the number and demographic distribution
of your annual quota. This quota should be
completed as soon after the start of the fiscal
year as possible.
'• Patients having known or suspected pesticide
poisoning should not be sampled. If you are
involved with a potential pesticide poisoning,
we would like to know about it. However,
samples should not be taken for the National
Human Monitoring Program for Pesticides.
F-3
-------
t Patients exhibiting cachexia or who have been
institutionalized for long periods' should not
be sampled for the national program.
Legal Considerations •
. The National Human Monitoring Program for Pesticides i.s both
interested and deeply concerned about the legal ramification of this
human research project. Since the program operates in about 40 states,-.
it is not feasible for us to handle the variety of local or state
interpretations from our location in Washington. Therefore, as a
matter of policy, the legal requirements, i.e., informed consent,
confidentiality, are matters for your consideration and resolution.
Collections for this program must be made in conformance with .the appli-
cable HEW guidelines on the protection of human subjects of biomedical
and behavioral research. We will, however, be pleased to assist you in
any way possible.
.We have completed several studies on these matters and do not
believe that they present major obstacles to your participation. In
most documents authorizing postmortem examinations, there is a clause '
granting the examining physician permission to remove tissues for
research purposes. We consider this project to be included in that
category. -In the case of specimens recovered from your surgical practice,
the use of a small amount of tissue from a previously excised specimen
certainly does not place the patient at risk in any way whatsoever.
As you will notice in our discussion of data needed for each patient
sampled, we do have several mechanisms to assure confidentiality. In
fact, the disclosure or release of certain data is protected by federal
statute. The fees paid to you by our program are solely intended to
remunerate you or your designee for professional services rendered.
F-4
-------
Collection of Surgical Adipose Tissue
Collect samples of adipose tissu'e from unfixed specimens which have
been surgically excised for therapeutic reasons. Take special care to
keep samples from different patients separate, correctly and securely
labeled, and avoid their contact with other chemicals, such as paraffin,
disinfectants, preservatives, or plastics.
At least five grams of good quality (subcutaneous, perirenal, or
^-mesenteric) adipose tissue should be collected; avoid fibrous or connective
tissue, 'i.e., omentum. Place the fat, without any fixatives or preservatives
•into the provided chemically-cleaned container; legibly complete and
attach the self-adhesive label in ball-point pen or pencil. The bottle
labels should be affixed before freezing. Store the specimens up-right
in a freezer at -4°F (-20°C) until shipment.
Collection of Postmortem Adipose Tissue
Adipose tissue samples must be obtained only from unembalmed
cadavers. The interval between death and the collection of tissue
should be as short .as possible and must not exceed 24 hours, assuming
refrigeration during the interval. Samples of adipose tissue must weig"h
at least five grams and should be placed in the supplied, chemically-
clean container with a completed label affixed. .Specimens should be.
stored at -4°F (-20°C) without any fixative or preservative until
shipment. Submit only good quality fat; do not submit omentum as it
contains too much connective tissue for satisfactory analysis.
F-5
-------
Adipose should be taken dry, and should not be rinsed before
placing in the provided containers. Many water supplies contain materials
which would interfere with chemical analysis.
Instruments should be well-rinsed with distilled water and dried
before taking the adipose sample.
Completion o_f_ the Patient Summary Report
A Patient Summary Report should.be completed for each patient from
whom a sample was taken. Special attention should be given to the . .
completeness of the data. All medical information submitted is protected
from disclosure or release by U.S.C'. 552, (b) (6); 45 CFR Part 5. First
and last initials, in that order, should be used instead of the complete
name to insure that confidentiality is maintained. The initials, along
with the data of birth, sex, and race, are used in this office to
compose.the AMA identification number. The patient's identification
number and/or the pathology department's accession number are for your.
information in referring back to the individual patient when you receive
the results of the pesticide analysis. .
Confirmed diagnosis should be detailed in the spaces provided.
Only the major ones should be supplied.
Other information required should be completed as accurately as
possible. The complete forms should be held and sent under the lid of
the insulated container when shipment is made.
F-6
-------
Packing and Shipping
Tighten all lids on the specimen bottles carefully. This is •
important since we are required to use special aluminum foil cap liners .
which make tightening a little difficult. Be certain that a completed
bottle label is firmly attached to each specimen bottle. Wrap each
bottle in gauze or paper to prevent breakage during shipment and to keep
the- label on' the container. Place the s'pecimen bottles in the insulated
'mailer and fill it with dry ice. If you have difficulty obtaining dry
'ice, please call us and we can arrange alternative methods of refriger-
ation for you. • .
A franked addressed label is on the reverse side of the address
card. This card is marked AIR MAIL - SPECIAL DELIVERY. (Do not send
Air Express, please). There is no cost to the sender because of the
franked label. All insulated mailers should have a PERISHABLE-PACKED IN
DRY ICE label visible from all sides on the outside..
Specimens should be mailed on a Monday or Tuesday of a week with ino
federal holidays. This assures that they will arrive before the end of
the work week on Friday. . .
Patient Summary Reports should be sent in the carton with.the
specimens when possible. They can be folded and placed on the top of
the.polyfoam lids.
F-7
-------
Only samples which meet our criteria and are handled according to
the guidelines can be accepted. Mo substitute containers will be
accepted. •
For Further Information
If you have any questions or comments, please contact us. Telephone
(collect): ,202/755-8060.
Sandra C. Strassman
Frederick W. Kutz, Ph.D.
National Human Monitoring Program
for Pesticides (WH-569)
F-8
-------
APPENDIX G
National Data Summaries for Fiscal Years 1972-1976
(Provided by EPA)
G-l
-------
o
FY 1972
\
FREQUENCY AND LEVELS OF POLYCHLORINATED BIPHENYLS IN HUMAN ADIPOSE TISSUE, BY GEOGRAPHIC AREA AND RACE: WET. WT. BASIS
Stratification
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
American Indian
Sample size
1778
1469
303
*
—
3
3
4098
3370
708
—
—
4
7
9
Percent not
Detected
10.24
9.12
15.18
0.00
66.66
26.06
25.16
30.22
0.00
28.57
44.44
Percent less
than 1 ppm
22.05
23.48
15.18
0.00
33.33
15.45
18.52
10.45
25.00
14.28
0.00
Percent
1-3 ppm
61.87
62.08
61. 38
66.66
0.00
50.63
51.57
46.46
50.00
14.28
55.55
Percent greater
than 3 ppm
5.85
5.30
8.25
33.33
0.00
7.86
6.73
12.86
25.00
42.65
0.00
File Type
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
Indicates no samples in this group.
-------
o
OJ
FY 1973
\
FREQUENCY AND LEVELS OF POLYCHLORINATED BIPHENYLS IN HUMAN ADIPOSE TISSUE, BY GEOGRAPHIC AREA AND RACE: WET. WT. BASIS
Stratification
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
American Indian
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
American Indian
Sample size
1120
918
132
2
1
3
1
1279
1102
169
2
2
3
1
Percent not
Detected
21.43
20.18
31.06
0.00
0.00
33.33
0.00
22.86
36.09
0.00
0.00
33.33
0.00
Percent less Percent Percent greater
than 1 ppm 1-3 ppm than 3 ppm File Type
41.61
31.16
5.80
DESIGN
45.25
14.39
100.00
0.00
33.33
0.00
30.27
37.87
0.00
0.00
33.33
100.00
4.28
16.66
0.00
100.00
0.00
0.00
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
MASTER
43.82
17.75
50.00
50.00
33.33
0.00
29.12
32.54
0.00
0.00
33.33
100.00
4.17
13.60
50.00
50.00
0.00
0.00
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
-------
FY 1974
•\
FREQUENCY AND LEVELS OF POLYCHLORINATED BIPHENYLS IN HUMAN ADIPOSE TISSUE, BY GEOGRAPHIC AREA AND RACE: WET. WT. BASIS
o
Stratification
NATIONAL
RACIAL GROUP
Caucasion
Negro
Mexican American
Puerto Rican
Oriental
Other
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
Sample size
926
798
119
1
1
7
1051
891
150
1
1
8
Percent not
Detected
9.29
8.27
14.28
0.00
0.00
42.85
9.04
8.30
12.00
0.00
0.00
37.50
Percent less
than 1 ppm
51.62
54.38
33.61
100.00
100.00
25.57
50.62
53.87
32.00
100.00
100.00
25.00
Percent
1-3 ppm
34.02
32.95
42.85
0.00
0.00
14.28
35.49
33.55
48.00
0.00
0.00
25.00
Percent greater
than 3 ppm File Type
5.08
4.38
9.24
0.00
0.00
14.28
4.85
4.26
8.00
0.00
0.00
12.50
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
-------
FY 1975
FREQUENCY AND LEVELS OF POLYCHLORINATED BIPHENYLS IN HUMAN ADIPOSE TISSUE, BY GEOGRAPHIC AREA AND RACE: WET. WT. BASIS
o
Cn
Stratification
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
Sample size
793
680
105
4
2
2
910
756
135
13
2
4
Percent not
Detected
5.80
5.88
4.76
25.00
0.00
0.00
5.82
5.95
4.44
15.38
0.00
0.00
Percent less
than 1 ppm
56.49
57.35
52.38
0.00
100.00
50.00
55.93
57.01
52.59
30.77
100.00
25.00
Percent
1-3 ppm
27.24
27.21
26.67
75.00
0.00
0.00
27.58
26.98
29.63
46.15
0.00
25.00
Percent greater
than 3 ppm File Type
10.47
9.56
16.19
0.00
0.00
50.00
10.66
10.05
13.33
7.69
0.00
50.00
DESIGN -
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
-------
FY 1976
\
FREQUENCY AND LEVELS OF POLYCHLORINATED BIPHENYLS IN HUMAN ADIPOSE TISSUE, BY GEOGRAPHIC AREA AND RACE: WET. WT. BASIS
o
Stratification
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
American Indian
NATIONAL
RACIAL GROUP
Caucasian
Negro
Mexican American
Puerto Rican
Oriental
Other
American Indian
Sample size
684
569
103
1
2
9
5
785
645
120
1
2
12
5
Percent not
Detected
2.03
1.76
1.94
0.00
0.00
22.22
0.00
1.78
55
67
0.00
0.00
16.67
0.00
Percent less
than 1 ppm
59.36
61.16
49.51
100.00
100.00
44.44
60.00
60.00
61.86
50.00
100.00
100.00
50.00
60.00
Percent
1-3 ppm
29.32
29.70
29.13
0.00
0.00
11.11
28.54
28.54
28.84
28.33
0.00
0.00
16.67
40.00
Percent greater
than 3 ppm File Type
9.29
7.38
19.42
0.00
0.00
22.22
9.68
9.68
7.75
0.00
0.00
0.00
16.67
0.00
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
DESIGN
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
MASTER
------- |