&EPA NHATS COMPARABILITY STUDY
GC/ECO
-------
-J
o
EPA Contract No. 68-DO-0174
June 1992
NHATS COMPARABILITY STUDY
Exposure Evaluation Division
Office of Pollution Prevention and Toxics
U. S. Environmental Protection Agency
401 M Street, S.W.
Washington, DC 20460
77 West Jacks "':L 10..
Chicago, IL 60oo - ..an ' Uih Fl°or
-------
DISCLAIMER
This document has been reviewed and approved
for publication by the Office of Pollution
Prevention and Toxics and the Office of
Prevention, Pesticides, and Toxic Substances,
U.S. Environmental Protection Agency. The use
of trade names or commercial products does not
constitute Agency endorsement or recommenda-
tion for use.
-------
AUTHORS AND CONTRIBUTORS
This Comparability Study is one part of the National Human
Adipose Tissue Survey (NHATS) . The study was conducted through
the cooperative efforts of EPA and contract support staff. EPA
participation was from the Exposure Evaluation Division within the
Office of Pollution Prevention and Toxics (OPPT) . Within the
Exposure Evaluation Division, the Design and Development Branch
and the Field Studies Branch were responsible for the conduct of
the study and the preparation of the final report. Contract
support to OPPT included Westat, Battelle, Midwest Research
Institute (MRI), and the Institute of Rural Environmental Health
at Colorado State University (CSU).
Westatr Inc.
Performed data entry and data verification, performed the
statistical analysis, and wrote the final report on the compara-
bility study.
Key personnel included:
John Rogers Adam Chu
Institute of Rural Environmental Healthr CSU
Performed the chemical analyses on the samples using both the
GC/ECD and HRGC/MS methods.
Key personnel included:
John Tessari Sharon Chaffey
Michael Aaronson
111
-------
Midwest Research Institute
Collected and stored the tissue samples, prepared the
composite samples for analysis, and contributed portions of the
Quality Assurance Project Plan (QAPjP) .
Key personnel included:
John Stanley John Hosenfeld
Jack Balsinger
Battelle Columbus Division
Processed the patient summary reports (PSRs), developed the
composite design, and contributed the sections for the composite
design in the QAPjP.
Key personnel included:
Greg Mack Lesly Arnold
Tamara Collins
Exosure Evaluation Division
Conducted the National Human Adipose Tissue Survey on an
annual basis, managed tasks of contractors, and reviewed and
edited reports .
Key personnel included:
Cindy Stroup Mary Frankenberry
Phil Robinson John Schwemberger
Joe Breen Janet Remmers
IV
-------
TABLE OF CONTENTS
Page
EXECUTIVE SUMMARY xv
1 INTRODUCTION 1
2 SUMMARY AND CONCLUSIONS 5
3 SAMPLE COLLECTION, PREPARATION, AND ANALYSIS PROCEDURES ... 9
3.1 Collection and Storage of the Human Adipose Tissue
Specimens 10
3.2 Sample Design for the Comparability Study 11
3 .3 Sample Preparation 13
3.3.1 Preparing the composite samples and
batches 13
3.3.2 Method blanks samples 15
3.3.3 Quality control samples 15
3 .4 Chemical Analysis Methods 19
3.4.1 GC/ECD analysis procedures 20
3.4.2 HRGC/MS analysis procedures 21
3 .5 Data File Preparation 24
4 DESCRIPTION OF THE DATA 25
4 .1 Organization of the Data 25
4 .2 Preliminary Review of the Data 27
5 OVERVIEW OF THE STATISTICAL ANALYSIS 33
5 .1 Correction for Blanks 33
5.2 Outliers and the Use of Remarks 35
5.3 A Model for the Data 36
5.4 Basis for Analyzing the Log Transformed
Concentrations 40
6 ANALYSIS OF DETECTION LIMITS AND PERCENT RECOVERY 45
6.1 Comparison Detection Limits 45
6.2 Calculating Recovery 49
6.2.1 Calculating recovery using spiked
multisplit samples 49
6.2.2 Calculating recovery using quality control
samples 52
6.2.3 Comparison of HRGC/MS and GC/ECD recovery... 54
6.3 Comparison of Percent Detected 55
7 RELATIONSHIP BETWEEN THE GC/ECD AND HRGC/MS METHODS 61
-------
TABLE OF CONTENTS (continued)
Page
7.1 Modeling the Relationship Between the GC/ECD and
HRGC/MS Methods 61
7.2 Comparison of HRGC/MS and GC/ECD Measurements 69
7 .3 Comparisons for Each Compound 75
7 .4 Plots of HRGC/MS versus GC/ECD Measurements 81
8 COMPARISON OF GC/ECD AND HRGC/MS MEASUREMENTS ACROSS
YEARS 101
8 .1 Plots of Measurements Over Time 101
8.2 Assessment of the Results Ill
9 PCB MEASUREMENTS 115
9.1 Comparison of PCB Reporting and Measurement
Procedures 115
9.2 PCB Recovery Using the HRGC/MS and GC/ECD Methods... 116
9.3 Comparison of HRGC/MS and GC/ECD Paired
Measurements 118
10 ANALYSIS OF PRECISION AND COMPONENTS OF VARIANCE 125
10.1 Standard Deviation Versus Mean 125
10.1.1 Spiked multisplit samples 127
10.1.2 Paired samples 128
10.1.3 Results 129
10 .2 Components of Variance 132
10.2.1 Components of variance for the HRGC/MS
measurements 134
10.2.2 Components of variance for the GC/ECD
measurements 142
10.3 Summary and Comparison of HRGC/MS and GC/ECD
Precision 146
11 REFERENCES 151
APPENDIX A: SUMMARY DATA TABLES 153
APPENDIX B: CONVERSION FROM STANDARD DEVIATION OF LOG
TRANSFORMED DATA TO COEFFICIENT OF VARIATION 181
APPENDIX C: RECOVERY FROM MULTISPLIT SAMPLES 185
APPENDIX D: DESCRIPTION OF THE ANALYTICAL PROCEDURES 191
D. 1 Summary of the MOG-GC/ECD Procedure 193
D .2 Summary of the HRGC/MS Procedure 197
VI
-------
TABLE OF CONTENTS (continued)
Page
APPENDIX E: DISCUSSION OF THE VARIANCE COMPONENTS 203
vii
-------
LIST OF TABLES
Page
Table 1. Number of Samples for HRGC/MS and GC/ECD Chemical
Analysis by Sample Type and Batch 14
Table 2. Spiking Levels for the Multisplit Composite
Samples 16
Table 3. Concentrations of Target Compounds in GC/ECD
Porcine Fat QC Samples, in ug/g wet weight 18
Table 4. Spiking Levels of Surrogate Compounds in HRGC/MS
Samples 22
Table 5. Model for the GC/ECD and HRGC/MS Measurements
with an Explanation of Each Term 41
Table 6. Detection Limits (ug/g) Using the HRGC/MS and
GC/ECD Methods, for all Compounds Reported on the
GC/ECD Forms 46
Table 7. Average Recovery, with 95% Confidence Intervals,
for the GC/ECD Measurements on Spiked Multisplit
and Quality Control Samples 53
Table 8. Average Recovery, with 95% Confidence Intervals,
for the HRGC/MS Measurements on Spiked Multisplit
and Quality Control Samples 56
Table 9. Number of GC/ECD and HRGC/MS Measurements on
Paired Composite Samples and Percent Detected, by
Data Qualifier and Analytical Method 57
Table 10. Correlation Between Log Transformed HRGC/MS and
GC/ECD Measurements 71
Table 11. Summary of the Statistical Tests for Batch
Effects and Nonconstant Recovery 72
Table 12. Geometric Mean Ratio of GC/ECD to HRGC/MS
Measurements 73
Table 13. Summary of PCB Recovery Measurements 120
Table 14. Coded HRGC/MS versus Coded GC/ECD PCB
Measurements in Paired Samples 123
Table 15. Comparison of HRGC/MS and GC/ECD Measurements in
Multisplit Samples 124
viii
-------
LIST OF TABLES (continued)
Table 16.
Table 17.
Table 18.
Table 19.
Table 20.
Table A-l.
Table A-2.
Table A-3.
Table A-4.
Table A-5.
Table A-6.
Table A-7.
Table A-8.
Table A-9.
Table A-10.
Table A-ll.
Page
Variance Components for Log Transformed
Measurements for HRGC/MS Fraction 1 Surrogate
Compounds 136
Variance Components for Log Transformed
Measurements for HRGC/MS Fraction 2 Surrogate
Compounds 139
Variance Components for Log Transformed
Measurements for Fraction 1 Compounds in HRGC/MS
Spiked Dichloromethane Samples 141
Variance Components for Log Transformed Aldrin
Measurements in GC/ECD Samples 144
Summary of Variance Components for GC/ECD and
HRGC/MS Measurements 147
Summary of GC/MS Measurements on Method Blanks,
Nominal Concentration in ug/g 156
Summary of GC/MS Measurements on Dichloromethane
Spike Samples, as Percent Recovery 158
Summary of GC/MS Measurements on Paired Composite
Samples, ug/g
160
Summary of GC/MS Measurements on the High Level
Spiked Multisplit Composite Samples, ug/g 162
Summary of GC/MS Measurements on the Low Level
Spiked Multisplit Composite Samples, ug/g 164
Summary of GC/MS Measurements on the Mid Level
Spiked Multisplit Composite Samples, ug/g 166
Summary of GC/MS Measurements on Surrogate QA
Compounds in all Samples, Percent Recovery 168
Summary of GC/ECD Measurements on Method Blanks,
ug/g
169
Summary of GC/ECD Measurements on the Porcine Fat
Samples in Batches 1, 2, and 3, ug/g 170
Summary of GC/ECD Measurements on the Porcine Fat
Samples in Batches 4 through 10, ug/g 171
Summary of GC/ECD Measurements on the paired
Composite Samples, ug/g 172
IX
-------
LIST OF TABLES (continued)
Page
Table A-12. Summary of GC/ECD Measurements of Aldrin Recovery
in All Samples Except Extracts, ug/g 173
Table A-13. Summary of GC/ECD Measurements on the High Level
Spiked Multisplit Composite Samples, ug/g 174
Table A-14. Summary of GC/ECD Measurements on the Low Level
Spiked Multisplit Composite Samples, ug/g 175
Table A-15. Summary of GC/ECD Measurements on the Mid Level
Spiked Multisplit Composite Samples, ug/g 176
Table A-16. Summary of GC/ECD Measurements on Extracts of the
Unspiked Samples Associated with the High Level
Spiked Multisplit Composite Samples, ug/g 177
Table A-17. Summary of GC/ECD Measurements on Extracts of the
Unspiked Samples Associated with the Low Level
Spiked Multisplit Composite Samples, ug/g 178
Table A-18. Summary of GC/ECD Measurements on Extracts of the
Unspiked Samples Associated with the Mid Level
Spiked Multisplit Composite Samples, ug/g 179
Table B-l. Coefficient of Variation for the Untransformed
Data for Selected Values of s, the Standard
Deviation of the Log Transformed Data 184
Table C-l. Recovery for Spiked Compounds in Multisplit
Samples Analyzed Using the HRGC/MS Method 187
Table C-2. Recovery for Spiked Compounds in Multisplit
Samples Analyzed Using the GC/ECD Method 188
Table C-3. Recovery for Spiked Compounds in Multisplit
Samples Analyzed Using the GC/ECD Method, with
Unspiked Concentrations Measured in Extracts 189
Table E-l. Variance Components and Sources of Error 207
x
-------
LIST OF FIGURES
Page
Figure 1. Processing steps for the sample preparation and
analysis of composite samples 17
Figure 2. Subsets of the data with comparable measurements
by sample type and type of compound 26
Figure 3. Histogram of GC/ECD measurements of p,p'-DDE on
paired composite adipose tissue samples 31
Figure 4. Histogram of HRGC/MS measurements of p,p'-DDE on
paired composite adipose tissue samples 31
Figure 5. Analyses to be performed on subgroups of data
defined by sample type and type of compound 34
Figure 6. Average HRGC/MS and GC/ECD detection limits for
primary compounds 48
Figure 7. Percent of paired samples with detected
concentrations using the HRGC/MS and GC/ECD
methods 58
Figure 8 Average ratio of the GC/ECD and HRGC/MS
measurements for primary compounds, with 95%
confidence intervals 74
Figure 9. HRGC/MS versus GC/ECD concentration measurements
for p,p'-DDT in paired composite human adipose
tissue samples 83
Figure 10. HRGC/MS versus GC/ECD concentration measurements
for p,p'-DDE in paired composite human adipose
tissue samples 84
Figure 11. HRGC/MS versus GC/ECD concentration measurements
for beta-BHC in paired composite human adipose
tissue samples 85
Figure 12. HRGC/MS versus GC/ECD concentration measurements
for dieldrin in paired composite human adipose
tissue samples 86
Figure 13. HRGC/MS versus GC/ECD concentration measurements
for heptachlor epoxide in paired composite human
adipose tissue samples 87
Figure 14 . HRGC/MS versus GC/ECD concentration measurements
for oxychlordane in paired composite human
adipose tissue samples 88
xi
-------
LIST OF FIGURES(continued)
Page
Figure 15. HRGC/MS versus GC/ECD concentration measurements
for trans-nonachlor in paired composite human
adipose tissue samples 89
Figure 16. HRGC/MS versus GC/ECD concentration measurements
for uncorrected hexachlorobenzene in paired
composite human adipose tissue samples 90
Figure 17. HRGC/MS versus GC/ECD concentration measurements
for corrected hexachlorobenzene for recovery in
paired composite human adipose tissue samples 91
Figure 18. Transformed HRGC/MS versus GC/ECD p,p'-DDT
measurements used for statistical tests 92
Figure 19. Transformed HRGC/MS versus GC/ECD p,p'-DDE
measurements used for statistical tests 93
Figure 20. Transformed HRGC/MS versus GC/ECD beta-BHC
measurements used for statistical tests 94
Figure 21. Transformed HRGC/MS versus GC/ECD dieldrin
measurements used for statistical tests 95
Figure 22. Transformed HRGC/MS versus GC/ECD heptachlor
epoxide measurements used for statistical
tests 96
Figure 23. Transformed HRGC/MS versus GC/ECD oxychlordane
measurements used for statistical tests 97
Figure 24. Transformed HRGC/MS versus GC/ECD trans-
nonachlor measurements used for statistical
tests 98
Figure 25. Transformed HRGC/MS versus GC/ECD uncorrected
hexachlorobenzene measurements used for
statistical tests 99
Figure 26. Transformed HRGC/MS versus GC/ECD corrected
hexachlorobenzene for recovery measurements used
for statistical tests 100
Figure 27. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted p,p'-DDT concentrations for design
samples from 1970 through 1984 103
Figure 28. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted p,p'-DDE concentrations for design
samples from 1970 through 1984 104
XII
-------
LIST OF FIGURES(continued)
Page
Figure 29. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted beta-BHC concentrations for design
samples from 1970 through 1984 105
Figure 30. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted dieldrin concentrations for design
samples from 1970 through 1984 106
Figure 31. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted heptachlor epoxide concentrations for
design samples from 1970 through 1984 107
Figure 32. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted oxychlordane concentrations for design
samples from 1970 through 1984 108
Figure 33. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted trans-nonachlor concentrations for
design samples from 1970 through 1984 109
Figure 34. Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted corrected hexachlorobenzene
concentrations for design samples from 1970
through 1984 110
Figure 35,
Figure 36,
Figure 37,
Figure 38.
Figure 39.
Weighted average GC/ECD, HRGC/MS, and HRGC/MS
adjusted corrected hexachlorobenzene
concentrations for design samples from 1970
through 1984, with two outliers removed from the
calculation of the 1982 HRGC/MS and HRGC/MS
adjusted average
Histogram of HRGC/MS surrogate compound
recoveries in lipid samples
Histogram of HRGC/MS PCB measurements with
shading to indicate coded GC/ECD concentration
in paired samples
112
119
121
Slope of the linear relationship between the log
of the standard deviation and the log of the
mean for GC/ECD measurements on multisplit
spiked and paired samples
130
Slope of the linear relationship between the log
of the standard deviation and the log of the
mean for HRGC/MS measurements on multisplit
spiked and paired samples
131
xiii
-------
LIST OF FIGURES(continued)
Eage
xiv
-------
EXECUTIVE SUMMARY
The National Human Adipose Tissue Survey (NHATS) is a chemi-
cal monitoring program operated by EPA from 1970 through 1992.
The objectives of the NHATS program were: (1) to detect and quan-
tify the concentrations and prevalences of selected toxic sub-
stances in the adipose tissue of the general U.S. population; (2)
to measure trends in these concentrations over time; (3) to assess
the effects of regulatory actions; and (4) to provide baseline
body burden data for chemicals of interest to EPA.
The data for NHATS were generated by collecting and analyzing
adipose tissue specimens for selected toxic substances. The
adipose tissue specimens were obtained from cooperating hospitals
and medical examiners across the continental United States.
Initially, the NHATS program focused on organochlorine pesti-
cides and polychlorinated biphenyls (PCBs) using the Modified
Mills Olney Gaither (MOG) protocol and packed column gas chromato-
graph/electron capture detection (GC/ECD) method. However, in
1982 the program was expanded to include a wider range of
lipophilic compounds. In 1982 and 1984, the samples were analyzed
using a method based on a high resolution gas chromatograph/mass
spectrometer (HRGC/MS).
A comparability study was initiated to compare the measure-
ments from the two analytical methods for similar adipose tissue
samples. The determination of data comparability between the
methods is essential to allow valid assessments of the data when
combining results for GC/ECD and HRGC/MS for trend analysis and
baseline estimates. The specific objectives of the Comparability
Study were to: 1) characterize the detection levels, recovery, and
precision of the measurements, and to compare these for the two
methods; 2) assess whether there are significant differences
between the measurements from the two analytical methods; and 3)
xv
-------
describe the relationship between the HRGC/MS and GC/ECD measure-
ments and assess its usefulness for converting measurements from
one method to the other.
The study objectives were met by analyzing paired composite
samples using both the HRGC/MS and GC/ECD methods. Individual
FY84 specimens were combined to create larger composite samples to
reduce analysis costs and to provide adequate sample quantities to
meet the sensitivity requirements for both HRGC/MS and GC/ECD
analyses. The composite samples were grouped into batches for
processing and analysis. Each sample was split to create paired
samples, one for GC/ECD analysis and the other for HRGC/MS anal-
ysis. Method blanks, quality control, and multisplit samples were
analyzed, in addition to the paired samples, to assess the quality
of the data and to estimate the precision and recovery of the two
methods.
The HRGC/MS method can detect a wider range of compounds than
the GC/ECD, which is limited because of its specificity to com-
pounds with high electron capture cross-sections. Of the 20 com-
pounds measured by both methods, only the following nine compounds
were positively identified and quantified by both the GC/ECD and
HRGC/MS methods: p,p'-DDT; p,p'-DDE; beta-BHC; dieldrin, hepta-
chlor epoxide; oxychlordane; trans-nonachlor; hexachlorobenzene;
and PCBs.1 These are the same compounds that have been routinely
detected in previous GC/ECD efforts. These compounds are referred
to as the primary compounds for the comparability study. Because
the results for dieldrin were limited, the comparison of the two
analytical methods is based primarily on the measurements of:
p,p'-DDT; p,p'-DDE; beta-BHC; heptachlor epoxide; oxychlordane;
trans-nonachlor; hexachlorobenzene; and PCBs.
The GC/ECD method is more sensitive than the HRGC/MS method
in measuring all primary compounds except PCBs. Because the de-
1Total PCBs are counted as one compound here, however, for the HRGC/MS method,
the concentrations of individual PCB homologs are reported.
xvi
-------
tection limits vary among samples, the average detection limits
for each compound were used for comparison. The average GC/ECD
detection limits are consistently below 0.01 ug/g, compared to the
average HRGC/MS detection limits which range from 0.01 to 0.35
ug/g for different compounds.2 However, for PCBs, the GC/ECD
method is much less sensitive than the HRGC/MS method, with aver-
age detection levels of 0.43 ug/g and 0.01 ug/g respectively.3
Method recovery was determined from the analysis of samples
spiked with known amounts of target analytes. The recovery esti-
mates vary, depending on the compound. The GC/ECD recoveries for
spiked lipid material range from 52% to 109%. The HRGC/MS recov-
ery estimates are less precise and range from 26% to 62%, with the
exception of beta-BHC with an estimated recovery of 99%. These
general results are consistent with the ratio of the GC/ECD to
HRGC/MS measurements in the paired samples. The GC/ECD recovery
estimates are consistently greater than the HRGC/MS estimates for
all compounds except beta-BHC for which the estimates are similar.
For all compounds except dieldrin and PCBs, a strong linear
relationship existed between the HRGC/MS and GC/ECD measurements
in the paired samples, with the ratio of the HRGC/MS to GC/ECD
measurements being roughly constant. For some compounds, the
ratio of the HRGC/MS to GC/ECD measurements varied among analysis
batches and depended on the concentration in the samples. The
ratio of the average GC/ECD measurement to the average HRGC/MS
measurement in the paired samples ranged from 1.25 to 3.88 for
different compounds. For p,p'-DDT, p,p'-DDE, heptachlor epoxide,
oxychlordane, trans-nonachlor, and hexachlorobenzene, the ratios
are statistically significantly greater than 1.0.
For the statistical analysis, PCBs were analyzed separately
because the GC/ECD concentrations in the paired samples were
2The measurements are reported in micrograms per gram lipid weight, unless
otherwise noted.
3The BCD method provides data on total PCBs while the MS method is capable of
providing data on individual PCB compounds.
xvii
-------
reported on an interval scale4 rather than as a concentration,
which limited the ability to compare the HRGC/MS and GC/ECD mea-
surements. The results for the PCB measurements were similar to
other primary compounds in that the GC/ECD measurements tended to
be higher than the corresponding HRGC/MS measurements. The ratio
of the GC/ECD and HRGC/MS measurements could not be determined due
to the interval nature of the GC/ECD data.
The precision with which a concentration can be measured
depends on the analysis method and the concentration in the
sample, with the standard deviation of the measurements increasing
as the concentration increases. For the primary compounds other
than dieldrin the GC/ECD measurements were more precise than the
HRGC/MS measurements. For these compounds, the following rules of
thumb can be used to describe the precision: I) the coefficient of
variation of the HRGC/MS measurements was three times larger than
for the GC/ECD measurements; and 2) 95% of GC/ECD measurements
were within 22% of the actual concentrations and 95% of HRGC/MS
measurements were within 63% of the actual concentrations in the
samples.
The final objective of the Comparability Study was to assess
whether the relationship between the HRGC/MS and GC/ECD measure-
ments in the FY84 composite samples is useful for comparing
average concentrations measured by different methods in different
years. To make this assessment, the ratios of the average GC/ECD
to HRGC/MS measurements were used to adjust the 1982 HRGC/MS
measurements for comparison to the trends in the GC/ECD measure-
ments in neighboring years. Given the likely prediction errors,
the adjusted HRGC/MS averages from 1982 were consistent with the
trends in the GC/ECD averages prior to 1982 and in 1983. Use of
this adjustment method assumes that the relationship between the
GC/ECD and HRGC/MS measurements are unaffected by changes in the
Concentrations were reported as being in one of the intervals 0 to .33 ug/g,
.33 to 1 ug/g, 1 to 3 ug/g, or greater than 3 ug/g.
XVI11
-------
procedures over time, the laboratory used, or the concentration
levels in the samples.
An alternate procedure for comparing the HRGC/MS and GC/ECD
data is to correct all measurements for recovery. Because the
recovery depends on the sample matrix, this would require recovery
measurements on lipid samples for each year of data. An adequate
number of samples for estimation of recovery is recommended to
limit the possible error in the estimated recoveries and the
corresponding averages. When adequate recovery information is
available, correcting the HRGC/MS data for recovery may be prefer-
able to using the ratios from FY84 data because the recovery
correction requires making fewer assumptions.
The conclusions from the Comparability Study are:
The measurements from the HRGC/MS and GC/ECD methods
cannot be compared without accounting for differences
in recovery, precision, and detection limits.
• The GC/ECD method has generally more precise estimates,
higher recovery, and lower detection limits than the
HRGC/MS method when measuring the primary compounds.
The GC/ECD method was more precise than the HRGC/MS
method. The GC/ECD method has higher recovery than the
HRGC/MS method for all compounds except beta-BHC for
which the recovery estimates are similar. The GC/ECD
method has lower detection limits than the HRGC/MS
method for all compounds except PCBs.
• The HRGC/MS measurements are lower than the GC/ECD
measurements for all primary compounds. The differences
are statistically significant for some compounds.
• Compared to the GC/ECD method, the HRGC/MS method can be
used to study a wider range of target compounds and was
chosen in recent years in order to expand the list of
chemicals monitored by NHATS.
xix
-------
For comparison of measurements across methods and
years, the measurements can be adjusted using either
the ratio of the 6C/ECD to HRGC/MS measurements or by
correcting for recovery. A correction for recovery
requires making fewer assumptions.
• Using the ratios of the average GC/ECD and HRGC/MS
measurements in FY84 samples to convert from one method
to the other proved reasonable for the 1982 data.
However, this conversion method has limitations. The
relationship between .the GC/ECD and HRGC/MS measurements
may change with time, the laboratory used, or the con-
centration levels in the samples.
• A combination of theoretical arguments and data analysis
suggests that the comparability of the measurements from
the two methods can be improved by correcting for recov-
ery within each sample year and laboratory. Additional
analysis beyond the scope of this study is required to
evaluate the best method for making a recovery correc-
tion and how it compares to use of a ratio correction
based on the Comparability Study.
xx
-------
INTRODUCTION
The National Human Adipose Tissue Survey (NHATS) has been the
main operative program of the National Human Monitoring Program
(NHMP). NHMP is a chemical monitoring program designed to fulfill
the human and environmental monitoring mandates of both the Toxic
Substances Control Act (TSCA) and the Federal Insecticide,
Fungicide and Rodenticide Act (FIFRA), as amended. NHATS was
first established by the U. S. Public Health Service in 1967, and
was transferred to the U. S. Environmental Protection Agency (EPA)
in 1970. NHATS was conducted by EPA through 1992. The Agency now
is in the process of developing alternate human tissue monitoring
activities as part of the National Human Exposure Assessment
Survey (NHEXAS).
The data for NHATS were generated by collecting and analyzing
adipose tissue specimens for selected toxic substances. The
adipose tissue specimens were obtained from cooperating hospitals
and medical examiners within a statistically representative sample
of metropolitan statistical areas across the continental United
States. Cooperating pathologists and medical examiners collected
and sent adipose tissue specimens to EPA on a continuing basis
throughout each fiscal year. The pathologists and medical exam-
iners also supplied EPA with a limited amount of demographic,
occupational, and medical information for each specimen. This
information allows reporting of the residue levels by
subpopulations of interest, namely sex, race, age group, and
geographic region.
Historically, the NHATS program monitored human adipose
tissue for the presence and levels of 19 organochlorine pesticides
and polychlorinated biphenyl compounds (PCBs). Nine of these 20
compounds have been regularly detected in more than 90% of the
annual NHATS specimens (Robinson, Mack, Remmers, Mohadjer, 1990).
The NHATS program has documented a decrease in the concentrations
-------
of several of these compounds over time. NHATS data showed a
decrease in the percentage of samples with PCB levels above 3 ug/g
wet weight1 starting in 1978. This decrease began 2 years after
Congressional legislation in 1976 banned production and use of
PCBs.
The success in monitoring the effect of regulation and the
desire to broaden knowledge of the range of chemicals to which
humans are exposed led EPA to expand the list of chemicals moni-
tored by NHATS to include a wider range of lipophilic compounds.
The expansion of the chemical list led to a new chemical measure-
ment protocol in 1982. From 1970 through 1981 and in 1983, the
NHATS samples were analyzed by the Modified Mills Olney Gaither
(MOG) protocol using a packed column gas chromatograph with elec-
tron capture detection (GC/ECD) (Sherma and Beroza 1980) . In 1982
the samples were analyzed using a method that applies high resolu-
tion gas chromatography/mass spectrometry (HRGC/MS) for detection
of analytes (Stanley 1985). In 1984, samples were split and
analyzed using both methods for the Comparability Study.
Since a primary objective of the NHATS program has been to
estimate baseline levels and trends over time, possible changes in
the measurements associated with changes in the analytical tech-
nique are of interest. Therefore, this Comparability Study was
initiated to compare the measurements from the two analytical
methods for similar adipose tissue samples. Comparability of the
measurements from the two methods has focused on eight individual
organochlorine pesticides and one class of chemicals which have
been consistently detected in human adipose tissue with the GC/ECD
method. These compounds are p,p' DDT, p,p' DDE, beta-BHC, hepta-
chlor epoxide, oxychlordane, trans-nonachlor, hexachlorobenzene,
dieldrin, and PCBs. For the GC/ECD method, the hexachlorobenzene
measurements were reported as measured (referred to as uncorrected
1The measurements are reported in micrograms per gram lipid weight, unless
otherwise noted.
-------
hexachlorobenzene) and corrected for recovery based on historical
recovery estimates (referred to as corrected hexachlorobenzene).
The specific objectives of the Comparability Study were to:
• Describe and compare the detection level, recovery, and
precision of the measurements using the HRGC/MS and
GC/ECD analysis methods;
• Determine if there are significant statistical differ-
ences between the HRGC/MS and GC/ECD measurements on
paired samples; and
• Describe the relationship between the HRGC/MS and GC/ECD
measurements and evaluate its usefulness for converting
measurements from one method to the other.
These objectives have been met by analyzing split composite
adipose tissue samples collected in 1984 using both analytical
procedures. NHATS samples from 690 individuals were composited
into 46 composite tissue samples. These samples were split into
two portions, or paired samples, one for analysis using the GC/ECD
method and 1 for analysis using the HRGC/MS method. One of the
paired samples for GC/ECD analysis was lost during preparation,
leaving 45 paired samples for the Comparability Study. Three of
the 45 composite samples were larger than the others and are
called multisplit composite samples Each of these three composite
samples was split into 10 portions, five portions for GC/ECD anal-
ysis and five for the HRGC/MS analysis. Four of the five portions
for each method were spiked with the target analytes in order to
measure recovery. Additional quality control (QC) and blank
samples were prepared to bring the total number of samples for
each method to 80. This report summarizes the measurements,
presents the results of the statistical analyses used to determine
the recovery for each analytical method, and compares the perfor-
mance of the HRGC/MS and GC/ECD procedures.
The Comparability Study was limited to a comparison of the
data from the 1984 paired samples, the multisplit samples, associ-
ated quality control samples, and summary data from other years.
-------
This study did not attempt to analyze trends over time or provide
national or regional estimates.
The rest of this report describes of the sample collection,
analysis, and the results and conclusions of the Comparability
Study. The report is divided into the following chapters and
appendices:
Chapter 2 Summary and Conclusions;
Chapter 3 Sample Collection, Preparation, and Analysis
Procedures;
Chapter 4 Description of the Data;
Chapter 5 Overview of the Statistical Analysis;
Chapter 6 Analysis of Detection Limits and Percent Recovery;
Chapter 7 Relationship Between the GC/ECD and HRGC/MS
Methods;
Chapter 8 Comparison of GC/ECD and HRGC/MS Measurements
Across Years;
Chapter 9 PCB Measurements;
Chapter 10 Analysis of Precision and Components of Variance;
Chapter 11 References;
Appendix A Summary Data Tables;
Appendix B Conversion from Standard Deviation of Log
Transformed Data to Coefficient of Variation;
Appendix C Recovery from Multisplit Samples;
Appendix D Description of the Analytical Procedures; and
Appendix E Discussion of the Variance Components.
Reports that describe the NHATS program, the sample collec-
tion procedures, and the preparation and analysis procedures for
the tissue samples are listed in the references.
-------
2 SUMMARY AND CONCLUSIONS
The NHATS Comparability Study compared the performance of the
HRGC/MS and GC/ECD analytical methods for measuring pesticides and
PCBs in human adipose tissue samples. The results are based on
data from 45 paired composite FY84 NHATS samples and associated
multisplit spiked, quality control, and blank samples, and an
analysis of summary data from other years. The conclusions, based
on measurements of p,p'-DDT, p,p'-DDE, beta-BHC, dieldrin, hepta-
chlor epoxide, oxychlordane, trans-nonachlor, hexachlorobenzene,
and PCBs, are:
• Average detection limits for the GC/ECD method were
lower than for the HRGC/MS method, except for PCBs.
For the organochlorine pesticides, the average GC/ECD
detection limits are less than .01 ug/g. The corresponding
HRGC/MS detection limits range from .01 ug/g to .35 ug/g
for different compounds. For PCBs, the average GC/ECD
detection limit of .43 ug/g is much greater than the
average for the HRGC/MS method of .01 ug/g.
• Method Recovery in lipid samples was higher for the
GC/ECD method than for the HRGC/MS method for all
compounds except beta-BHC, for which the recovery
estimates were similar. These general results are
consistent with the ratio of the GC/ECD to HRGC/MS measure-
ments in the paired samples. The method recovery depended
on the sample matrix.
The GC/ECD recoveries ranged from 52% to 109%, except for
uncorrected hexachlorobenzene, with recovery from 33% to
61%. The HRGC/MS recoveries in the lipid samples ranged
from 26% to 62%, except for beta-BHC with a recovery esti-
mate of 99%. The ratio of the GC/ECD to the HRGC/MS recov-
ery ranged from 0.90 to 3.15 in the multisplit samples.
• The percentage of composite samples with detected
concentrations was greater for the GC/ECD method
than for the HRGC/MS method.
The primary compounds were detected in all GC/ECD samples.
The percent of HRGC/MS samples with detected concentrations
ranged from 82% to 98% for different compounds, with the
exception of dieldrin which was detected in 42% of the
samples tested. The percentage of samples with detected
concentrations depends on the method detection limit, the
recovery and the concentration in the samples.
-------
In paired samples, the measurements using the
GC/ECD method were greater than those using the
HRGC/MS method, with the ratio of the 6C/ECD to
HRGC/MS measurements being roughly constant. For
some compounds, the ratio depended on the concen-
tration in the samples or differed among batches.
When compared using geometric means, the GC/ECD measure-
ments were greater than the HRGC/MS measurements. The
ratio of the GC/ECD to HRGC/MS measurement ranged from 1.25
to 3.88. The ratio is statistically significantly greater
than 1.0 for p,p'-DDT, p,p'-DDE, heptachlor epoxide, trans-
nonachlor, and corrected hexachlorobenzene.
Differences in the ratio among batches were statistically
significant for four compounds: p,p'-DDE, beta-BHC,
oxychlordane, and trans-nonachlor. Changes in the ratio of
the GC/ECD to HRGC/MS measurements with changing concentra-
tions were statistically significant for three compounds:
p,p'-DDT and beta-BHC, and hexachlorobenzene. For these
compounds the ratio of the GC/ECD to HRGC/MS measurements
decreases as the concentration in the samples increases.
For PCBs, the HRGC/MS measurements were similar to
or lower than the corresponding GC/ECD measure-
ments .
The standard deviation of the HRGC/MS and GC/ECD
measurements increases as the concentration being
measured increases such that the coefficient of
variation of the measurements is constant. GC/ECD
measurements have coefficients of variation less
than that for the HRGC/MS measurements. For the
primary compounds other than dieldrin the GC/ECD
measurements were more precise than the HRGC/MS
measurements.
Approximate 95% prediction intervals for the GC/ECD and
HRGC/MS measurements of Fraction 1 compounds2 (all primary
compounds except dieldrin) are 22% and 63% respectively.
There are not enough measurements for Fraction 2 compounds
(including dieldrin) to reliably compare the measurement
precision for the HRGC/MS and the GC/ECD methods for
Fraction 2 compounds.
Given the likely prediction errors, the HRGC/MS
averages from 1982, adjusted by the GC/ECD to
HRGC/MS ratio found in the FY84 samples, were
consistent with the trends in the GC/ECD averages
in the years adjacent to 1982.
2The laboratory analysis procedures separate the target compounds into two
portions, called Fraction 1 and Fraction 2.
-------
For five compounds, the adjusted HRGC/MS average was close
to the GC/ECD trend. For one other compound, the adjusted
HRGC/MS average was close to the GC/ECD trend after
removing two outliers from the HRGC/MS data. For one
additional compound, the data and the ratio estimates were
too variable to attribute the observed differences in the
GC/ECD and HRGC/MS adjusted averages to the adjustment pro-
cedure. For one compound, no comparison was possible
because there were no data from 1982.
For comparison of measurements across methods and
years, the measurements can be adjusted using
either the ratio of the GC/ECD to HRGC/MS measure-
ments or by correcting for recovery.
Using the ratios of the average GC/ECD and HRGC/MS measure-
ments in FY84 samples to convert from one method to the
other proved reasonable for the 1982 data. However, this
conversion method has limitations. The relationship
between the GC/ECD and HRGC/MS measurements may change with
time, the laboratory used, or the concentration levels in
the samples. When recovery information, either average
recovery or within sample recovery estimates based on
surrogate compounds, is available, correcting the HRGC/MS
data for recovery may be preferable to using the ratios
from the Comparability Study. A combination of theoretical
arguments and data analysis suggests that the comparability
of the measurements from the two methods may be improved by
correcting for recovery within each sample year and labora-
tory. Evaluating the use of recovery estimates to adjust
the data across years was beyond the scope of this study.
-------
SAMPLE COLLECTION, PREPARATION, AND ANALYSIS
PROCEDURES
The sample and data processing steps for the Comparability
Study, from the initial tissue collection to the final analysis,
included:
(I) Collecting the individual adipose tissue samples;
(2) Preparing the sample design for the Comparability Study;
(3) Compositing the individual specimens into composite
adipose tissue samples;
(4) Preparing the quality control and blank samples;
(5) Extracting the lipid from the composite adipose tissue
samples;
(6) Dividing each lipid sample into a pair of samples, one
to be analyzed using the GC/ECD method and one to be
analyzed using the HRGC/MS method;
(7) Measuring the concentration of target and surrogate
compounds using the GC/ECD and HRGC/MS procedures;
(8) Entering the concentration measurements into a computer
file for statistical analysis; and
(9) Analyzing the data to determine the precision and
recovery of the GC/ECD and HRGC/MS methods and compare
the measurements in the paired composite samples.
Battelle Columbus Division participated in the sample collec-
tion design and prepared the sample design for the Comparability
Study. Midwest Research Institute (MRI) collected and stored the
tissue samples and prepared the composite samples and lipid
extracts. The Institute of Rural Environmental Health at Colorado
State University (CSU) performed the HRGC/MS and GC/ECD analysis
on the lipid extract samples. Westat prepared the data files and
performed the statistical analysis.
This chapter reviews the Comparability Study specimen collec-
tion procedures; the preparation of the composite, blank, and
-------
spiked samples; and the analytical procedures for the HRGC/MS and
GC/ECD methods. The data are summarized in chapter 4. Chapters 5
through 10 present the statistical analysis and results.
3.1 Collection and Storage of the Human Adipose Tissue
Specimens
As part of the ongoing NHATS program, individual adipose
tissue specimens are collected on a yearly basis from volunteer
hospitals and medical examiners in selected cities throughout the
continental United States. Specimens from the fiscal year 1984
collection were used to compare the GC/ECD and HRGC/MS analytical
procedures.
The cities in the NHATS survey were selected using probabil-
ity sampling. Within the selected cities, organizations
(hospitals and medical examiners) were recruited to participate in
the NHATS program. Each participating organization was assigned a
quota of individual tissue specimens to be collected over the year
from surgery patients and cadavers. While the non-random methods
of selecting participating organizations and of obtaining tissue
specimens within the hospitals may affect the interpretation of
national averages and comparisons between different years, the
specimen collection procedures do not affect the Comparability
Study results.
The individual tissue specimens were sent to MRI for storage
until further processing was performed. The composite samples
were prepared in the beginning of 1986. GC/ECD analyses were
performed in the first half of 1986. HRGC/MS samples in batch 1
were analyzed in April 1987; the remaining batches were analyzed
at the end of 1987 and the first half of 1988.
10
-------
3.2 Sample Design for the Comparability Study
The sample design specified which individual FY84 adipose
tissue specimens were to be combined into composite samples and
the order of preparation and analysis of the composites, method
blanks, and quality control samples.
The sample design specified four types of samples:
• Single-split composite samples, composite samples which
were split into two portions, one for GC/ECD analysis
and one for HRGC/MS analysis;
• Multisplit composite samples, composite samples which
were split into 10 portions, five portions for GC/ECD
analysis (four of which were spiked) and five for the
HRGC/MS analysis (four of which were spiked);
• Quality control samples to monitor recovery and preci-
sion; and
• Blank samples to identify and correct for possible
contamination of the samples.
The sample design specified 43 single-split composites, 3 multi-
split composites, 10 quality control samples, and 10 blank samples
for each method.
The criteria for choosing which specimens to composite were
based on demographic factors: geographic region, age group, race,
and sex. Composites were constructed from individual specimens
within the same age group and census division, allowing estimation
of mean concentrations for census division and age group. Within
a census division and age group, specimens were combined in a way
that permitted estimation of differences between sex and race
groups.
Composite samples were created in order to reduce the analyt-
ical costs and to achieve the desired sensitivity of the analyti-
cal methods. The sensitivity of the HRGC/MS and GC/ECD methods
depends on the size of the sample being analyzed. Individual
11
-------
NHATS tissue specimens are not usually large enough to provide
adequate sensitivity for both HRGC/MS and GC/ECD analyses.
Therefore, individual specimens were composited to create a larger
sample for chemical analysis. The HRGC/MS method required a
20-gram sample while the GC/ECD method required a 5-gram tissue
sample. Thus the target wet weight of the single-split composite
samples was 25 grams. Multisplit composite samples required 125
grams of tissue.
The samples were grouped into 10 batches for analysis. Each
batch was designed to have one quality control (QC) sample spiked
with a known concentration of the target compounds, and one blank
sample. The multisplit composites were analyzed in three differ-
ent batches. For each analytical method, the unspiked portion and
two of the spiked portions were analyzed in one batch. The
remaining two spiked portions were analyzed in two separate
batches.
There were two deviations from the sample design. First, one
sample for the GC/ECD analysis was lost during preparation,
leaving 45 paired samples for the Comparability Study.3 Second,
problems in the preparation of two composite samples from batch 8
and one composite sample from batch 9 resulted in the need to
reconstruct these composites. When these samples were recon-
structed from stored tissue specimens, an additional blank was
prepared for batches 8 and 9. As a result, batches 8 and 9 each
had two blank samples.
The term "paired" samples is used in this report to refer to
the single-split composite samples and the unspiked portions of
the multisplit composite samples. The paired samples were used to
compare the GC/ECD and HRGC/MS analysis methods.
3A 47th composite was prepared to replace the lost GC/ECD sample for making
national estimates and evaluating trends over time. This sample was not part
of the comparability study. In all, 697 individual specimens were combined
into 47 composite samples.
12
-------
A total of 80 samples was analyzed using each method.
However, the extracts from the unspiked multisplit GC/ECD samples
were analyzed three times, once in each batch containing a spiked
multisplit sample. Therefore, there are 2 additional analyses for
each of the 3 unspiked multisplit composites, giving 86 total
analyses for the GC/ECD method. Table 1 summarizes the number of
FY84 samples of different types.
3 . 3 Sample Preparation
Sample preparation is discussed below, including compositing
of the individual tissue specimens, extracting the lipid from the
tissue for analysis, and preparing blank and spiked QC samples. A
more complete description of the sample processing procedures is
presented in Appendix D.
3.3.1 Preparing the composite samples and batches
The sample design specified which individual tissue specimens
were to be combined into each composite sample and which compos-
ites were in each batch. The tissue weight contributed by each
individual specimen to a composite was made as consistent as
possible. A method blank sample and all composites within a batch
were prepared at the same time.
The target compounds to be measured are associated with the
lipid portion of the composite adipose tissue samples. The first
processing step extracts the lipid from the tissue. As part of
the extraction step, the weight of the lipid as a percentage of
the wet weight of the sample is determined. The lipid extracted
from the composite sample was split into two portions, one
reserved for the GC/ECD analysis and another reserved for the
HRGC/MS analysis.
13
-------
Table I, Number of Samples for HRGC/MS and GC/ECD Chemical
Analysis by Sample Type and Batch
Sample Type
Design Composite Samples
Single-split Paired
Unspiked Multisplit Paired
Paired Samples Subtotal
Unpaired Composites
Design Samples Subtotal
Spiked Multisplit Samples
Low level spike
Mid level spike
High level spike
Blank and QC Samples
QC samples
Blank
Regenerated blanks
HRGC/MS Samples Total
GC/ECD Extract Analyses
GC/ECD Samples and
Extract Analyses Total
Total
42
3
45
1
46
4
4
4
10
10
2
80
6
86
Batch
123456
253553
11 1
7 8 9 10
5554
1
21 1
2
2 1
111111
111111
888888
1 11
898998
1 1
1
1111
1111
1 1
8996
111
9 10 10 6
Note. Blank entries indicate that no samples with the indicated
sample type were analyzed in that batch.
14
-------
The three multisplit samples were prepared following the
procedure for the single-split samples, with the exception that
the multisplit samples had a total wet weight of 125 grams. One
25 gram portion of each multisplit composite was removed and
handled identically to the single-split samples. The remaining
lipid was sent to CSU for .spiking and splitting into eight spiked
multisplit samples, four for GC/ECD analysis and four for HRGC/MS
analysis. The spiking levels for the multisplit composite samples
are shown in Table 2. Figure 1 graphically summarizes the sample
processing steps.
3.3.2 Method blanks samples
A procedural blank sample, consisting of 100 milliliters of
methylene chloride, was prepared for each batch. The blank was
passed through all sample processing steps used for the single-
split samples. The measurements from these blanks were used to
determine if the laboratory background contributed to the target
analytes in the samples.
3.3.3 Quality control samples
The QC samples were prepared with known quantities of target
compounds. The QC samples are different for the GC/ECD and
HRGC/MS analyses. Reference porcine fat samples were obtained
from EPA's Las Vegas Laboratory (EPA/EMSL-LV) for the GC/ECD
analyses. The porcine samples in batches 1 to 3 had different
spiking levels than the samples in batches 4 through 10. The
concentration levels in these two groups of porcine fat samples
are provided in Table 3. Because the HRGC/MS method was new to
CSU and there was a limited source of standard reference adipose
material, a sample of spiked dichloromethane4 was prepared for each
batch analyzed using the HRGC/MS method. The dichloromethane
samples were spiked with 10 nanograms of all HRGC/MS target
compounds.
4Dichloromethane is a synonym for methylene chloride.
15
-------
Table 2. Spiking Levels for the Multisplit Composite Samples
Compound
Dieldrin
beta-BHC
Hexachlorobenzene
Heptachlor Epoxide
trans-Nonachlor
Oxychlordane
p , p ' -DDE
Low
.20
.10
.05
.10
.10
.10
1.00
Spikina levels
ug/g wet weight
Mid
.40
.20
.08
.20
.20
.15
3.00
High
.60
.40
.11
.30
.30
.20
5.00
16
-------
Physical Sample Diagram
Individual Adipose Tissue Samples
Processing Steps
Receive Adipose Samples at MRI
Composite
Sample
Spiked
Unspiked
I I
GC/ECD HRGC/MS
Create Composite Samples in Batches
25 gm. for single-, 125 gm for
multisplit Composites
I
Extract the Lipid from the Tissue
I
Calculate Percent Lipid
I
Divide the Lipid into 2 portions,
one for Spiked , one for Unspiked
Multisplit1 Samples
Divide Lipid for
HRGC/MS and GC/ECD Samples
Paired
Composite
Samples
Send Samples from MRI to CSU
I
Proceed with chemical Analysis
Send the Lipid from MRI to CSU
I
Spike the Lipid
I
3C
•
/BCD HRG<
Spiked
Multi-
Split
Composite
Samples
HRGC/MS and GC/ECD Samples
:/MS 1
"I Split the Lipid into 4 Spiked Samples
1
>K
Associate the Samples with Batches
1
1
4
Proceed with chemical Analysis
Figure 1. Processing steps for the sample preparation and analysis
of composite samples.
17
-------
Table 3. Concentrations of Target Compounds in GC/ECD Porcine
Fat QC Samples, in ug/g wet weight
Compound
Hexachlorobenzene
beta-BHC
Mirex
Oxychlordane
Heptachlor Epoxide
trans-Nonachlor
Dieldrin
p,p'-DDE
p,p'-DDT
PCB (Aroclor 1254)
Batches 1-3
.070
.120
.065
.060
.046
.080
.150
1.620
.230
-
Batches 4-10
.049
.300
.129
.112
.075
.119
.040
1.860
.175
1.000
18
-------
3 . 4 Chemical Analysis Methods
Although the specifics of the chemical analysis for the two
analysis methods differ, the general steps followed are:
(1) Spiking with surrogate standards for estimating
recovery;
(2) Removing the lipid and leaving the target compounds in a
solvent solution;
(3) Separating the target compounds into two fractions to be
analyzed separately;
(4) Spiking with standard solutions to aid in identification
and quantitation and for estimating recovery (for the
HRGC/MS method only) ; and
(5) Identification and quantification of the target
compounds .
Once the amounts of the target compounds in each sample have
been determined, the concentrations can be calculated by express-
ing the amounts as a proportion of the original sample wet weight
or lipid weight. The concentration in micrograms per gram (ug/g)
wet weight is :
. . Weight of the Target Compound (ug) . ,. . .
*wet weight 1)
where X is the measured concentration. The concentration in
micrograms per gram (ug/g) lipid weight is :
. _ Weight of the Target Compound (uq)
XLiPid weight (ug/g) - (Wet Weight (g) ) (percent Lipid/100)
_ Weight of the Target Compound (ug)
Lipid Weight (g) ( '
Unless otherwise stated, all concentrations in this report are
expressed on a lipid basis.
19
-------
The following sections discuss the analysis procedures,
performed at CSU, for the GC/ECD and HRGC/MS methods. Specifics
of each method are presented in Appendix D.
3.4.1 GC/ECD analysis procedures
For the FY84 samples, the GC/ECD method was used to quantify
the amount of 20 target compounds.5 The modified MOG procedure was
used for preparation of the samples submitted for GC/ECD analysis.
The lipid extracts for the GC/ECD analysis were spiked with one
microgram of aldrin. The bulk lipid material was removed by
partitioning the lipid sample between hexane and acetonitrile.
The target compounds were separated into two fractions using a
florisil chromatography column. The target compounds in the final
extract were identified and quantified using the packed column gas
chromatograph (GC) with electron capture detection (ECD) .
Concentrations of compounds were quantitated based on the
area of the peak representing the target compound and the aldrin
peak. The recovery of aldrin was calculated as a quality assur-
ance check on the entire process. Based on historical information
on the recovery at the partitioning step, the concentration of
several compounds were reported as measured (referred to as uncor-
rected) and corrected for recovery. Concentrations for hexa-
chlorobenzene and mirex were computed on both a corrected and an
uncorrected basis. Concentrations for p,p' DDT were reported on a
corrected and uncorrected basis for the porcine fat tissue samples
in batches 4 through 10.6
Total PCB concentrations were reported on the following
interval scale using letters to designate each interval: V = not
5£ighteen individual chemicals, one class of chemicals (PCBs), and aldrin
which was used to estimate recovery.
6 The values reported in the space provided on the form for hexachlorobenzene,
mirex and p,p' DDT are uncorrected unless the uncorrected values for these
compounds are written in at the end of the list, in which case the entry for
these compounds is the corrected value.
20
-------
detected, W = .33 to 1 ug/g wet weight, Y = 1 to 3 ug/g wet
weight, and Z = greater than 3 ug/g wet weight.
The following information was reported for each composite
sample and target compound:
• Compound name and code;
• "<" if the amount detected was less than the level of
quantification (LOQ) but greater than the level of
detection (LOD), in which case the amount reported was
the LOQ; and
• The concentration reported to 0.01 ug/g wet weight or a
letter code for the interval in which the PCB concentra-
tion lies.
3.4.2 HRGC/MS analysis procedures
In order to expand the list of chemicals that could be moni-
tored by NHATS, the standard NHATS method for detection and quan-
tification of chemical compounds was changed from GC/ECD method to
the HRGC/MS method. For the FY84 samples, the HRGC/MS method was
used to quantify the amount of 57 target compounds. The sample
was spiked with known amounts of the eleven surrogate compounds.
Table 4 identifies the surrogate compounds and the level added to
each sample. Gel permeation chromatography (GPC) was used to
separate target analytes from the lipid material. The target
compounds were separated into two fractions (referred to as
Fraction 1 and Fraction 2) using a Florisil chromatography cleanup
similar to the MOG method (the Fraction 2 extract was not analyzed
in batches 4 through 10) . Known amounts of the three internal
standards, anthracene-dlO, naphthalene-d8, and benzo-(a)-
anthracene-d!2, were added to each of the final extracts, prior to
analysis by high resolution gas chromatography/ mass spectrometry.
The identification and quantification was based on relative
retention times and response factors established during calibra-
tion. Specific compounds were identified by matching characteris-
21
-------
Table 4. Spiking Levels of Surrogate Compounds in HRGC/MS
Samples
Compound
Spike Level
ug
Chrysene-d12
1,2,4 -Trichlorobenzene-d3
13C6-1,2,4,5 - Tetrachlorobenzene
13C6- Hexachlorobenzene
13C6- 4 - Chlorobiphenyl
13C12-3,3 '4,4 - Tetrachlorobiphenyl
13C12-3,3 ,3,3 ,5,5 ,6,6 -
Octachlorobiphenyl
13C12 Decachlorobiphenyl
Diethyl Phthalate - 3,4,5,6 -d4
Di-N-Butyl Phthalate - 3,4,5,6 -d4
Butyl Benzyl Phthalate - 3,4,5,6 -d4
2
2
2
2
2
5
8
10
2
2
2
22
-------
tic spectra with reference material. One of the three internal
standards was designated as the appropriate internal standard for
each target analyte for purposes of identification and quantifica-
tion.
In order for an analyte to be identified, the following four
criteria had to be satisfied:
(1) The primary and secondary masses had to achieve their
maximum values within a. specified time span;
(2) The retention time of the primary and secondary mass
fragments relative to the designated internal standard
had to be within 10 seconds of the known relative reten-
tion time of the analyte;
(3) The relative abundances of the primary and secondary
masses all had to be within 20% of the relative abun-
dances in the reference spectrum of the analyte;
(4) The abundances of the primary and secondary masses all
had to exceed 2.5 times the background signal to noise
ratio.
Concentrations of analytes were computed from the calculated
amounts, the weight of the composite sample, and the percent lipid
in the composite sample. No concentrations were corrected for
recovery. The recoveries of the surrogate compounds were calcu-
lated as a check on method performance. A data qualifier was
determined based on the relative magnitude of the noise, peak
signal, and the quantity of each target compound in the lowest
calibration standard.
The following equation was used to calculate the lipid
adjusted concentration of each target compound in each sample:
As Ijs i i
X = 1000 Ais RRF Lipid Wt (3'3)
where:
X = Lipid adjusted concentration of the target compound (ug/g);
23
-------
As = Area of the primary characteristic ion response for the
compound being quantified;
Ij_s = Amount of internal standard added to the extract (ng) ;
Aj_s = Area of the primary characteristic ion response for the
corresponding internal standard;
RRF = Relative response factor, determined during calibration
of the instrument; and
Lipid Wt = Weight of the lipid in the adipose tissue sample
analyzed (g), calculated as the product of the wet
weight and the percent lipid divided by 100.
This equation is used later in the development of the statistical
model for the data.
The following information was recorded for each sample and
target analyte:
• Compound name;
• Data qualifier (either not-detected "ND", trace "TR", or
positively quantified "PQ");
• The calculated LOD for trace and not detected measure-
ments;
• The amount of each compound in nanograms (ng.) for
positively quantified and trace measurements; and
• Remarks to explain exceptional circumstances.
3.5 Data File Preparation
The measurements reported by the laboratory were entered into
computer files at EPA's National Computer Center (NCC) for use in
the statistical analysis. After data entry, the data were veri-
fied against the original reports and consistency checks were
performed to catch possible errors in the data entry or unusual
observations.
24
-------
4 DESCRIPTION OF THE DATA
This chapter describes the general characteristics of the
HRGC/MS and GC/ECD concentration measurements in the FY84 samples.
Numerical summaries are shown in Appendix A.
4.1 Organization of the Data
Measurements were made on four different types of samples:
(1) Method Blanks;
(2) QC standards: porcine fat samples in the GC/ECD analy-
sis, and spiked dichloromethane samples in the HRGC/MS
analysis;
(3) Spiked multisplit composite adipose tissue samples (at
three different spiking levels) and the associated
extract samples (for the GC/ECD analysis only); and
(4) Paired composite adipose tissue samples.
The compounds measured can be divided into two categories:
(1) Target compounds, those compounds of interest in the
NHATS program. These compounds have different levels in
different samples; and
(2) Surrogate compounds used for quality control and esti-
mating recovery. These compounds have similar levels in
all samples.
The different combinations of compounds and sample types
define groups of measurements which are directly comparable.
Figure 2 graphically portrays these groups of comparable measure-
ments. Tables A-l through A-18 in Appendix A summarize the
HRGC/MS and GC/ECD measurements within each group using the
percent detected, median, and, for the positively quantitated
observations, the mean, median and extremes.
25
-------
Type of Compounds
H
0)
r-\
I
Target Compounds
Primary Secondary
8 Compounds 11 GC/ECD Compound
49 HRGC/MS Compounds
Surrogate
Compounds
1 GC/ECD Compound
11 HRGC/MS Compounds
Method Blanks
12 Samples
QC Standards
10 Samples
Porcine Fat (GC/ECD)
Spiked Dichloromethane
(HRGC/MS)
Spiked Multisplit
Composite Samples
12 Samples
Paired Composite
Samples
45 Samples
Low
Mid
High
42 Single-split Composite
Samples
3 Unspiked Multisplit
Composite Samples
Surrogate
Compounds
(HRGC/MS)
Aldrin
(GC/ECD)
3 GC/ECD measurements were made on the final extract of
the unspiked multisplit paired composite samples
Figure 2. Subsets of the data with comparable measurements by
sample type and type of compound.
26
-------
The target compounds can be further divided into two groups:
• Primary compounds: compounds which are positively
quantified in numerous samples using both the GC/ECD and
HRGC/MS methods; and
• Secondary compounds which were either measured using
only one method or which were not positively quantified
using both methods.
Only nine compounds were positively quantified in the paired
samples using both the GC/ECD and HRGC/MS methods. These com-
pounds are: p,p'-DDT, p,p'-DDE, beta-BHC, dieldrin, heptachlor
epoxide, oxychlordane, trans-nonachlor, hexachlorobenzene and
PCBs. These compounds are referred to as the primary compounds
for the Comparability Study. In all cases except dieldrin and
PCBs, the number of positively quantified observations which can
be used to compare the two measurement methods is 37 or greater.
Dieldrin, in fraction 2, was only analyzed using the HRGC/MS
method in batches 1, 2, and 3, and therefore has few HRGC/MS
measurements. Only five samples had positively quantified
dieldrin concentrations reported for both the HRGC/MS and GC/ECD
methods. PCB measurements from the GC/ECD method were reported on
an interval scale, as described in Section 3.4.1, rather than as a
finite continuous value. The PCB results are discussed separately
in Chapter 9. Because the results for dieldrin were limited, the
comparison of the two analytical methods is based primarily on the
measurements of: p,p'-DDT; p,p'-DDE; beta-BHC; heptachlor epoxide;
oxychlordane; trans-nonachlor; hexachlorobenzene; and PCBs.
4 .2 Preliminary Review of the Data
The initial step in any data analysis is called exploratory
data analysis. This step usually involves plotting the data in
various ways to help identify important characteristics of the
measurements.
The percent lipid in the composite samples ranged from 41.5
to 99.3 percent, with an average of 77 percent, providing adequate
27
-------
lipid in each composite for measuring the concentrations of the
target compounds.
Initial plots of the data suggested that the distribution of
the data was skewed and that the standard deviation of the mea-
surements increased as the magnitude of the measurement increased.
There were no obvious outliers or extremely unusual observations
which might make subsequent preliminary work suspect. Plots using
the log of the measurements were consistent with the assumption
that the data can be described by a lognormal distribution. Plots
of the HRGC/MS surrogate compound measurements suggested that
there were systematic differences between batches.
Little laboratory background contamination was noted. There
were no quantifiable measurements in method blank samples using
the GC/ECD method. Positively quantified measurements in the
method blank samples were observed for 4 of 50 compounds7 using the
HRGC/MS method. With the exception of Di-n-Butyl Phthalate, the
measured quantities were close to the level of detection. Only
one measurement for a primary compound, p,p'-DDE, was positively
quantified.
From the tables in Appendix A, the coefficient of variation
can be calculated using the following equation:
**< • x. ^ • x. • / \ standard deviation , „ _
coefficient of variation (cv) = (4.1)
mean
The coefficient of variation is one measure of the variabil-
ity of the data. For many chemical measurements, the coefficient
of variation provides a stable measure of variability across a
range of concentrations. For samples with the same actual concen-
tration, such as spiked multisplit samples from the same composite
or QC samples, the coefficient of variation measures the precision
of the analytical technique. For the paired composite samples,
7p,p'-DDE, 1,2,4-trichlorobenzene, di-n-butyl phthalate, and di-n-octyl
phthalate
28
-------
the coefficient of variation measures the variability resulting
from both differences between composite samples and measurement
variation.
The coefficient of variation of the positively quantified
measurements was calculated for the compounds and groups of data
listed in Appendix A. It was found to be roughly similar for all
compounds within the two groups: paired samples and all other
samples. The average coefficient of variation across target
compounds measured in paired composite samples was 68% for GC/ECD
measurements and 72% for HRGC/MS measurements. For the remaining
samples, the average coefficient of variation across all compounds
was 11% for the GC/ECD method and 29% for the HRGC/MS method,
suggesting that the laboratory measurement error is greater for
the HRGC/MS method than for the GC/ECD method. The coefficients
of variation for measurements on the paired samples are substan-
tially greater than that for the spiked QC samples due to the
additional variation in the contaminant concentrations among
composites. Because the reported GC/ECD concentrations were
rounded to 0.01 ug/g wet weight, the variance estimates for the
GC/ECD method will tend to underestimate the true variance.
However, the effect of rounding on the results is expected to be
small.
The distribution of a set of concentration values which have
a coefficient of variation greater than 50% is often skewed to the
right. Investigation of the data confirms that, as a general
rule, the measurements on the paired samples are skewed to the
right. Measurements within a group of samples with the same
actual concentrations, such as the QC samples, have a smaller
coefficient of variation and a distribution which is roughly
symmetric.
Figures 3 and 4 show histograms of the p,p'-DDE measurements
on paired samples using the HRGC/MS and GC/ECD methods. Note that
the plots have different concentration scales and that the GC/ECD
29
-------
measurements are larger than the HRGC/MS measurements. Both
figures show a skewed distribution. Measurements on the composite
samples estimate the average of the concentrations in the individ-
ual samples which were composited. Since averaging decreases the
coefficient of variation, measurements in individual tissues
specimens can be expected to be more skewed than those from the
composites.
30
-------
10 +
8 -•
o
§ 6
cr
0)
4 -•
2 ••
0.00 4.50 9.00
p,p'-DDE BCD Concentration (ppm)
Figure 3. Histogram of GC/ECD measurements of p,p'-DDE on paired
composite adipose tissue samples.
15 r
Frequency
_*
O1 O
i • i
I I
0.00 1.25
p,p'-DDE MS Concentration (ppm)
2.50
Figure 4. Histogram of HRGC/MS measurements of p,p'-DDE on paired
composite adipose tissue samples.
31
-------
5 OVERVIEW OF THE STATISTICAL ANALYSIS
Different subsets of the data were used to achieve each of
the study objectives. The precision of the measurements was
determined from the QC samples, the spiked multisplit samples, and
the surrogate compounds which were common to all samples. The
recovery was estimated from the QC samples and the spiked multi-
split samples. The HRGC/MS and GC/ECD methods were compared
directly using the primary compounds in the paired samples.
Figure 5 summarizes graphically the different subsets of the data
and how the data in each subset were used in the Comparability
Study.
In order to achieve the objectives, some common procedures
for analyzing the data were established and a model for the data,
a mathematical description of the structure and important rela-
tionships in the data, was developed. This chapter discusses the
model for the data and assumptions behind the analysis.
5.1 Correction for Blanks
The method blanks consist of solvent samples that are pro-
cessed as single-split composite samples. If there is background
contamination during the sample processing, this contamination
will be observed in the blank samples. Assuming the background is
the same for all samples within a batch, the measured quantity in
the blank can be subtracted from the measured quantity in the
sample to correct for the contamination. Note however, when the
quantity in the blank is below the detection limit, the appropri-
ate contamination correction is some unknown concentration between
zero and the lowest quantifiable quantity.
33
-------
Surrogate
1 GC/ECD Compound
11 HRGC/MS Compounds
Precision
| (Components of |
| variance) , |
QA checks
n
•o
C
9
§
O
m
0
0
>l
**
CO CO
•S"S
3 3
><2 9-
M i" o
C co
002
O CJ "••»
CD w o
o oil
C5 X
O CO
f*H ^*
.. CO
>173
tO 3
M §
CU U
CD
0)
N
•H
i-i
(0
£
£
3
CO
CO
^.*
— o
(U a>
o xs
c o
(0
-H <
H a
c *
o > -
-H >1
w M-l H
-H O 0)
O >
m CO O
cu
0) M
O -P
a c
£ 0)
o o
o H
— 0)
CM
0)
N
-M
M
tO
3
CO
C ^i
O 4J »-l
•H G O
CO 0) >
-H O O
OHO
0)0)0)
H CU M
CU
0)
N
-H
H
(0
co
-H O
•-t O
-H 0)
(0 >
Cb-rl
£ -P
O (0
0),
a;1
c
o
•H
O
a)
a
•o
a>
c
-H
i
0) 4J
(1) C
XI tO
O 0)
4-> a
>i
CO J-J
0)
CO 0)
>irH
rH ft
tO £
C (0
<; co
-P CO
-H 0)
0)
4J
w
c
(0
fflS
i-(
*D Pi
CO
E a
10
co E
-H tO
4J CO
rH
15 «
•H3
•o co a
0) O 6
CO
O
T3 0> a
0) M 6
a>
M
3
c^
-H
O
•H £ WJ
&, O
-------
For the GC/ECD blank samples, no compounds were positively
quantified. In the HRGC/MS blank samples, four compounds were
positively quantified (p,p'-DDE, 1,2,4-Trichlorobenzene, Di-n-
Butyl Phthalate, and Di-n-Octyl Phthalate) of which only one,
p,p'-DDE, is used in the comparability analysis. The compound
p,p'-DDE was detected in only 1 of 12 blank samples, and then at a
level only 10% greater than the limit of detection.
It was decided to use all measurements without a blank cor-
rection rather than to correct a few measurements while ignoring
the unknown correction for most of the measurements. This deci-
sion is expected to have little effect on the statistical analy-
sis .
5.2 Outliers and the Use of Remarks
Outliers are observations which appear to be unusual compared
to the bulk of observations. A preliminary analysis of the data
indicated that there were no particularly unusual observations
which might significantly affect the statistical analyses. As a
result, no formal outlier analysis was performed before analyzing
the data. However, as different analyses are discussed in the
report, any values which might affect the statistical conclusions
are discussed.
The comment fields on the HRGC/MS data sheets sometimes
contained information on the quality of the reported quantities.
If a comment was supplied for a positively quantified measurement,
the comment generally indicated that the actual quantity was
judged to be either I) equal to or possibly less than or 2) equal
to or possibly greater than the reported measurement. Since the
size of any bias in these measurements could not be assessed, they
were used as reported.
35
-------
5.3 A Model for the Data
The model for the data is a mathematical description of the
relationships within the data which result from the process which
generated the data. Statistical procedures are used to estimate
parameters in the model and to check if the assumptions behind the
model are consistent with the data. The model is based on 1) the
sample design, i.e., the way in which the data are collected; 2)
the objectives of the study; and 3) the characteristics of the
data. The statistical analysis procedures are chosen to be
consistent with the model.
The model used to describe the NHATS data is developed below.
This particular model was selected only after data analysis deter-
mined that the model was consistent with the data.
The model assumes the measurement errors have a lognormal
distribution, and therefore that the standard deviation of the
measurement error is proportional to the magnitude of the measured
concentration. Environmental measurements and measurements made
on a scale from zero to infinity, such as chemical concentrations,
often have a skewed distribution such as a lognormal distribution.
A log transformation converts data which have a lognormal
distribution to data with a normal distribution. While the vari-
ance or standard deviation of the original measurements depends on
the concentration being measured, the variance of the log trans-
formed measurements is constant. Since many standard statistical
techniques are based on the assumption that the errors have a
normal distribution with constant variance, the log transformed
data are often easier to handle statistically than data in the
original scale.
36
-------
The equation for quantifying the HRGC/MS concentration X can
be used as a starting point to develop the model for the data8:
A I' 1 1
X = 1000 Ais RRF Lipid Wt (5>1)
Using the log of the measured concentration provides some
insight into an appropriate model for the log transformed data.
The equation for the log of the measurements is:
ln(X) = in + ln + In + In. (5.2)
where:
/ As \
Inl I depends on the actual concentration of the measured
IQQQ
compound. This term might also be written as:
where f (C) is a function relating the actual concen-
tration to the measured concentration and e is the
measurement error not associated with other terms .
depends on how much internal standard was added to the
sample and how much was detected by the equipment .
This term can be written as 8^ .
Infrrrj depends on the calibration of the equipment. This
term can be written as 6j .
ln[r~: — . , 1 depends on the characteristics of the sample, in
particular the measurements of wet weight and percent
lipid. This term can be written as 8S.
8This equation is the same as equation (3.3)
37
-------
After making the substitutions above, the model for one
HRGC/MS measurement might be written:
ln(X) = f (C) + 8j -f 8S + Si + e (5.4)
The equivalent model in the original concentration units
would be:
X = ef(C)e(63+ 5s + Si+ « (5.5)
Of these two models, equation (5.4), based on the log of the
measurements, is easier to fit using standard statistical tech-
niques than equation (5.5) because 1) the distribution of the
residuals is expected to be closer to normal; 2) the standard
deviation of the residuals are expected to be roughly constant;
and 3) the terms are additive.
Equation (5.4) provides an incomplete model in that it
ignores some characteristics of the data and the function f(C) is
not defined. Under ideal situations the HRGC/MS method would
measure the actual concentration in the sample, not counting some
variation in the measurements around the actual value. Because of
losses in the chemical preparation steps, the measured quantity is
usually less than the actual quantity of a compound in the sample.
Thus, the model for the measured concentration X might be:
X = CR (5.6)
where R is the percent recovery expressed as a fraction.
For spiked samples, an additional quantity of the compound
being measured is added to the sample, resulting in an increase in
concentration of the compound of S. With this addition, the model
for the log of the data might be:
ln(X) = ln(C+S) + ln(R). (5.7)
38
-------
The recovery may be different for the HRGC/MS and GC/ECD
measurements due to differences in the sample preparation. There
is also the possibility that the recovery depends on the charac-
teristics of the sample material. In other words, the recovery
may be different for QC samples, blank samples, and adipose tissue
samples .
In the final model for the relationship between the actual
and measured concentration, it is assumed that:
X = (C+S)P * R (5.8)
where p is added to cover the case in which the relationship
between the actual and measured concentrations is not linear. In
the log transformed units, equation (5.8) is:
ln(X) = pln(C+S) + ln(R). (5.9)
Because we have limited information on when the instruments
were calibrated, the effect associated with calibration, 8j, is
difficult to estimate. The calibration term may be different for
each internal standard. According to the Quality Assurance
Project Plan (QAP jP) for the comparability analysis (USEPA 1986) ,
the instrument's calibration was checked every day and calibrated
as necessary. Since batches were usually processed in one or two
days, the term for calibration effects is assumed to be confounded
with batch effects (i.e., cannot be estimated separately, based on
the data) .
After the addition of some terms for random effects associ-
ated with batch preparation, calibration, measurement of sample
characteristics, sample preparation and injection, and measurement
of the internal standard and target responses, the complete model
for the HRGC/MS and GC/ECD measurements is :
(XmCbsi) =Pmcln (CCS+SCS) +ln (R^) +6mb+6nibi+5Mbic+8s+5ms+5msi+emcbsi (5.10)
39
-------
The terms in the model are explained in Table 5 and
Appendix E .
Converted back to the original measurement units, equation
(5.10) becomes:
exp <+i+ic4-+i+^3i><5 . 11)
For the analysis of most of the data, the model shown in
Table 5 can be significantly simplified. When discussing specific
analyses in the following sections, the appropriate simplified
version of the model will be presented. The simplification
usually comes from combining terms which are confounded. In this
case, the combined term will be indicated by a change in the form
of the subscripts. When equations are provided in later chapters,
terms are defined only if they have not been previously defined.
5 . 4 Basis for Analyzing the Log Transformed
Concentrations
Transforming the data has implications for the model to use
in the analysis, the variance of the residuals (i.e., the esti-
mated magnitude of the measurement error) , the distribution of the
residuals, and the interpretation of the results. Assuming the
data have a lognormal distribution, the following statements can
be made:
• For the original measurements :
• The standard deviation of the measurement error is
linearly related to concentration (the coefficient
of variation of the original data is constant) ; and
• The measurement errors have a skewed distribution.
• For the log of the original data:
• The transformed data have a constant variance; and
• The transformed data will have a normal distribu-
tion.
40
-------
Table 5. Model for the GC/ECD and HRGC/MS Measurements with an
Explanation of Each Term
= l^cln(ccs+scs)+ln(Rmct)+5rnb+5inbi-t-5Mbic+8s+5ms+5msi+emcbsi
Where:
m = an index for the analytical method used to measure the
concentration, the two analytical methods are GC/ECD and
HRGC/MS (subscripts E and M respectively when specified
methods are used) .
c = an index for the compound being quantitated. Note that
some compounds were quantitated using only one analytical
method.
b = an index for the batch in which the samples were processed.
There were 10 batches of samples for each analytical
method .
s = an index for the sample being analyzed. There were a total
of 80 samples (not counting extracts for the GC/ECD
method) which are uniquely identified by the EPA ID
number, the batch in which the samples were analyzed and
the "Fldindic" number which indicates if the sample was a
spiked multisplit sample.
i = an index for the internal standard used to quantitate the
data. Three different internal standards were used to
quantitate compounds in the HRGC/MS analysis.
t = an index for the sample type. There are three different
sample types, Blank, QC samples, and Adipose tissue
samples .
= the measured concentration using analytical procedure
m for compound c measured in batch b, sample s, and quan-
titated using internal standard i.
Ccs = the actual concentration of compound c in sample s
before addition of any spike.
Scs = the increase in concentration of compound c due to a
spike being added to sample s.
Pmc = t^ie slope coefficient for the relationship between
ln(Xcs) and ln(Ccs + Scs) . This relationship may depend
on the analytical method m. In situations where there is
not enough data to estimate this term, it is assumed to be
1.0.
41
-------
Table 5. (Continued)
= a constant. If |3mc = 1.0 this constant can be
interpreted as the recovery for chemical c using
analytical method m on a sample with type t.
a random effect associated with batch b using
analytical method m, assumed to be normally distributed
with a mean zero and standard deviation of 0^,
depending only on m.
= a random effect associated with internal standard i in
batch b using HRGC/MS method, assumed to be normally
distributed with a mean zero and standard deviation of
^mbi- This term is confounded with 5^ in the GC/ECD
measurements.
= a random effect associated with calibration for
compound c quantitated by internal standard i in batch
b using HRGC/MS method, assumed to be normally
distributed with a mean zero and standard deviation of
°Mbic-
8S = a random effect associated with the measurement of wet
weight and percent lipid in sample s, assumed to be
normally distributed with a mean zero and standard
deviation os. This term is identical for both the
GC/ECD and HRGC/MS paired samples.
= a random effect associated with sample s using analyti-
cal method m, assumed to be normally distributed with a
mean zero and standard deviation O^, depending only on
m.
= a random effect associated with internal standard i in
sample s measured with the HRGC/MS method, assumed to
be normally distributed with a mean zero and standard
deviation O^i, depending only on m. This term is
confounded with 5^ in the GC/ECD measurements.
= a random measurement error for compound c measured
using analytical method m, in batch b, in sample s, and
quantitated using internal standard i, assumed to be
normally distributed with a mean zero and standard
deviation om.
42
-------
The characteristics of the data have been checked by looking
at the relationship between the measurement error and the concen-
tration. This relationship is discussed in Chapter 10. Although
the relationship between the standard deviation of the data to the
measured concentration is not very precise, it is consistent with
the use of a log transformation.
Use of a log transformation may be justified if the residuals
based on the logged data appear to have a normal distribution.
Several analyses were performed using both the original data and
the logged data. Based on histograms of the residuals, the resid-
uals from the log transformation are more normally distributed.
However, for many analyses, the differences are small.
We have assumed that the measurement error can be described
by a lognormal distribution for the following reasons:
• On theoretical grounds, the data can be expected to have
a distribution similar to a lognormal distribution;
• The residuals from the statistical analyses of the orig-
inal data have a skewed distribution. The residuals
from the statistical analyses of the log transformed
data have a roughly symmetric normal distribution; and
• The standard deviation of the residuals from the origi-
nal data increase roughly linearly with increasing
concentration.
When fitting models to the log transformed data, the esti-
mated error variance is for the transformed data. It may be
desirable to convert the variance in the log scale to a coeffi-
cient of variation in the original scale. The following formula
relates the variance of the log data, s2, to the coefficient of
variation of the original measurements, cv:
cv = Vexp(s2)-l (5.12)
For reference, Table B-l in Appendix B tabulates the coeffi-
cient of variation for selected values of s. In the discussion of
43
-------
measurement errors, the results are presented in terms of both the
variance of the log transformed values and the coefficient of
variation of the untransformed measurements.
44
-------
ANALYSIS OF DETECTION LIMITS AND PERCENT RECOVERY
This chapter describes and compares the detection limits, the
recoveries, and the proportion of samples with measured concentra-
tions above the detection limit (i.e., percent detected), for the
paired NHATS samples. The detection limits and recoveries are
discussed first. The percent detected is affected by both the
magnitude of the detection limit and the recovery and is therefore
discussed after these other two topics.
6. 1 Comparison Detection Limits
The measurement which is very unlikely to be exceeded in the
analysis of a blank sample is referred to as the detection limit
or limit of detection (LOD). The detection limit is defined
differently for the HRGC/MS and GC/ECD methods. The GC/ECD detec-
tion limit is based on historical experience with the method. The
detection limit using the HRGC/MS method is based on the lowest
calibration standard and the mass spectrometer signal to noise
ratio.
The lowest reported concentration for the GC/ECD method, the
limit of quantification (LOQ), has been established from years of
experience with the method. The limit of detection is defined as
LOQ/3. Measurements between the LOD and the LOQ are reported as
less than the LOQ using the symbol "<" on the report forms. The
label "Trace" was assigned in these cases in the data set. A
final concentration of zero was reported for samples with quanti-
tated concentrations less than the limit of detection. The aver-
age limit of detection for the GC/ECD target compounds is dis-
played in Table 6 in ug/g lipid weight. Because the GC/ECD detec-
tion limits are defined on a wet weight basis, the LOD based on
wet weight has been converted to an LOD based on lipid weight to
determine the average detection limit.
45
-------
Table 6. Detection Limits (ug/g) Using the HRGC/MS and GC/ECD
Methods, for all Compounds Reported on the GC/ECD Forms
Compound
p,p'-DDT
o,p'-DDT
p,p'-DDE
O,p'-DDE
p , p ' -ODD
O,p'-DDD
alpha-BHC
beta-BHC
gamma -BHC (Lindane)
delta-BHC
Aldrin
Dieldrin
Endrin
Heptachlor
Heptachlor Epoxide
PCB
Oxychlordane
Mirex
trans-Nonachlor
Uncorrected
Hexachlorobenzene
GC/ECD
Average LOD for
all samples
(ucr/q)a
.009
.009
.004
.009
.009
.009
.004
.009
.004
.004
.004
.004
.009
.004
.004
.433
.009
.043
.004
.004
HRGC/MS
Average reported
LOD
(ua/q)b
.049
.013
.352
.013
.013
.127
.020
.212
.019
.018
.013
.020
.041
.018
.016
.011
.016
.013
.014
.015
a Conversion from ug/g wet weight to lipid weight assumes
percent lipid equals 77%. The percent lipid in paired
samples ranged from 41.5% to 99.3%.
b Based on the LOD for trace and nondetect measurements.
46
-------
For the HRGC/MS method, the detection limit is the lowest
reported concentration and is based on the lowest calibration
standard used to calibrate the instrument and characteristics of
the signal and noise. For signals meeting the quality criteria
(see Section 3.4.2), the detection limit is the maximum of the
concentration based on 1) the lower calibration standard and 2)
2.5 times the noise level. For signals not meeting the quality
criteria, the LOD is the maximum of the concentration based on 1)
the lower calibration standard and 2) the observed signal. LODs
are reported for both nondetect and trace measurements. Because
the detection limit depends on several factors, it can vary
considerably between samples; however, it will always be equal to
or greater than the lipid adjusted concentration based on the
lower calibration standard. The average reported detection limits
for nondetect and trace HRGC/MS measurements in paired samples are
shown in Table 6. For p,p'-DDT, p,p'-DDE, and gamma-BHC, the
average HRGC/MS detection limits are based on five or fewer
reported detection limits, some of which were much larger than
that based on the lowest calibration limit.
For all compounds except PCBs and Mirex, the average GC/ECD
detection limit is less than the average HRGC/MS detection limit.
In addition, the average GC/ECD detection limits are less than the
minimum HRGC/MS detection limits based on the lowest calibration
standard. The average GC/ECD detection limits are smaller than
the average HRGC/MS detection limits by a factor of at least 10
for p,p'-DDE, beta-BHC, and o,p-DDD. Figure 6 compares the
HRGC/MS and GC/ECD detection limits for the primary compounds.
47
-------
•o
o
Q)
S
Q
U
W
\
o
O
XI
4J
U
O
A
a
to
•O
C
3
O
O
o
xaa-.d'd
-H
M
a
o
4-1
CO
4J
C
O
-H
4J
O
CP
nJ
M
(1)
vo
c
o
o e
a
(u
48
-------
6 . 2 Calculating Recovery
Due to losses in the sample processing steps, not all of the
target material in the original adipose tissue sample appears in
the final extract for measurement. As a result, the quantity of
compound measured by the GC/ECD or HRGC/MS equipment is less than
that in the original sample. The ratio of the quantity measured
to the quantity in the original sample is the recovery, usually
expressed as a percentage.
The percent recovery can be estimated using two approaches:
• Comparing measured concentrations with known concentra-
tions in quality control samples; or
• Determining changes in the measured concentration as a
result of spiking a tissue sample with a known amount of
compound, as in the multisplit samples.
Sections 6.2.1 through 6.2.3 discuss the calculation of
recovery using HRGC/MS and GC/ECD procedures for the QC samples
and spiked multisplit composite samples.
The model for the data, equation (5.10), allows for different
recoveries at different concentration levels. Because the QC
samples and multisplit samples were tested at three or fewer
concentrations, testing of the hypothesis to determine whether the
recovery is constant is not possible because there are insuffi-
cient degrees of freedom. Therefore, the calculation of recovery
assumes that the recovery is a constant, independent of the actual
concentration in the sample.
6.2.1 Calculating recovery using spiked multisplit
samples
For the multisplit samples, the average recovery and its
standard error are based on separate recovery estimates for the
three spiking levels. The calculations assume that the recovery
is the same for all concentrations. The recovery estimate at each
49
-------
spiking level is based on measurements in one unspiked multisplit
sample and four spiked multisplit samples.
Using equations for the measured concentration in the
unspiked and spiked portions of the multisplit samples (see equa-
tion 5.10), the following formula can be derived:
(s) (u)
. p lfi ,x
+ emcs (o.l)
Where:
= Measured concentration of compound c in the spiked
split of the sample;
= Measured concentration of compound c in the unspiked
split of the sample; and
emcs = random measurement error for the fraction R^, roughly
lognormally distributed.
Equation (6.1) can be used to calculate recovery in the
multisplit samples . With measurements on four spiked multisplit
samples four estimates of recovery for each spiking level can be
calculated. Because the HRGC/MS measurements in the unspiked
portions were sometimes below the detection limit or missing9, an
adjustment for the unspiked concentration required some judgment.
For the adjustments, the missing observations were replaced by
zero and the nondetect observations were replaced by the LOD/2 .
A weighted average of the estimates from the four spiked
sample portions was used to determine the recovery for each
spiking level. The weights for calculation of the weighted aver-
age are usually based on the measurement variance. Because of
possible differences between batches, the variance of the recovery
9No concentration or detection limit was provided for two trans-Nonachlor
measurements in the mid level unspiked samples. The footnote stated "compound
is present but cannot be quantitated." As a result of substituting zero for
the missing unspiked concentration, the calculated recovery will tend to
overestimate of the actual recovery.
50
-------
estimates based on spiked samples in the same batch as the
unspiked sample are likely to be less than that for spiked samples
in different batches. The measurement variances could be esti-
mated using the components of variance which are provided in
Chapter 10. However, assumptions must be made concerning the best
estimate of each variance component. For simplicity, the recovery
estimates from each spiked sample were given the same weight
(equivalent to assuming no batch effects) in order to calculate
the recovery at each spike level. An analysis showed that the
final results are affected very little by the weights chosen.
The recovery estimates across the three spiking levels are
independent and will have similar variances. Therefore, a confi-
dence interval for the mean recovery across all spiking levels was
calculated using a t-statistic. Because the t-statistic for each
compound had at most two degrees of freedom, a pooled estimate of
variance (pooled across all spiked compounds) was used to calcu-
late the confidence intervals for the recoveries. Calculation of
the t-statistic assumes that the recovery estimates have a normal
distribution. For the Comparability Study data, the bias intro-
duced by assuming normal errors was judged to be acceptably small,
relative to the standard error of the estimates.
In the GC/ECD analyses, portions of the final extract from
the unspiked samples were analyzed with each batch which included
a spiked sample. Although these measurements can be used in the
calculation of recovery, there was some concern that changes in
the extract samples over time as a result of storage might signif-
icantly affect the estimated recoveries. Therefore, the results
summarized in Section 6.2.3 are not based on the extract measure-
ments; however, these results are summarized in Appendix C. As
shown in Appendix C, the recoveries based on the extracts are
similar to those calculated without using the extracts, except
that the recovery estimates based on the extract have slightly
smaller variances.
51
-------
6.2.2 Calculating recovery using quality control
samples
For the quality control samples, average recovery and its
standard error were determined from the spiked QC samples in each
of the 10 independent batches. The calculations assume that the
recovery is the same for all concentrations . The recovery for
each sample is estimated from the measured quantity of each
compound and the known actual quantity in the sample.
The following equation was used to estimate recovery for the
quality control samples :
This equation can be derived from equation (6.1) where the concen-
tration in the unspiked portion is zero.
Confidence intervals on the average recoveries were based on
a t-statistic assuming normally distributed measurement errors.
However, if the measurements have a lognormal distribution, the
errors in equation (6.2) will also have a lognormal distribution.
For the Comparability Study data, the bias introduced by assuming
a normal distribution was small relative to the standard error of
the estimates. Equation (6.2) provides estimates which can be
directly compared to the results from the spiked multisplit
samples using equation (6.1). Equation (6.2) was used to estimate
recovery for the HRGC/MS analyses on dichloromethane spiked
samples and the GC/ECD analyses on porcine fat samples.
52
-------
Table 7. Average Recovery, with 95% Confidence Intervals, for
the GC/ECD Measurements on Spiked Multisplit and
Quality Control Samples
Compound
p,p'-DDT
p,p'-DDE
Beta-BHC
Dieldrin
Heptachlor Epoxide
Oxychlordane
Mi rex
trans-Nonachlor
Corrected
Hexachlorobenzene
Uncorrected
Hexachlorobenzene
PCBs
Spiked multi-
split samples3
b
82%±19%
89%±19%
90%±19%
83%±19%
73%±19%
b
73%±19%
77%±19%
53%±19%
b
Porcine fat
samples
batches 1-3 a
67%±14%
73%±14%
78%±14%
60%±14%
65%±14%
67%±14%
b
75%±14%
52%±14%
33%±14%
b
Porcine fat
samples
batches 4-10
98%±10%
109%±15%
98%±15%
104%±26%
97%±36%
96%±27%
106%±18%
98%±23%
99%±7%
61% c
78%±7% d
a 95% Confidence intervals are based on a pooled variance due to
the small number of measurements for each compound
b This compound was not spiked into the samples
c All measurements were identical
d See Chapter 9.
53
-------
6.2.3 Comparison of HRGC/MS and 6C/ECD recovery
Recovery estimates for GC/ECD measurements are shown in
Table 7. Different samples of porcine material, with different
spiking levels, were used for batches 1 to 3 and 4 through 10.
Therefore, recoveries for these two groups of samples are
presented separately. In general, recovery for the compounds
tested using the GC/ECD procedure was between 60% and 109%, with
the exception of uncorrected hexachlorobenzene which has a lower
recovery. The recovery based on the spiked multisplit samples was
between 73% and 90% for all compounds except uncorrected hexa-
chlorobenzene with a recovery of 53%. For batches 1 to 3, the
recovery was between 60% and 78% with the exception of hexachloro-
benzene with a recovery of 33% for the uncorrected measurements
and 53% for the corrected measurements. Recovery based on the
porcine adipose tissue samples from batches 4 through 10 were
between 96% and 109% with the exception of uncorrected hexachloro-
benzene and PCBs, with recoveries of 61% and 78% respectively.
For each compound except trans-Nonachlor, the GC/ECD recovery
estimates for the porcine fat samples from batches 1 to 3 were
less than for spiked multisplit samples which are in turn less
than that for the porcine samples in batches 4 through 10. This
pattern is statistically significant10. The differences between
the recovery in the three sets of samples may be due to differ-
ences in the sample material, deviations in the spiking solutions
from the nominal levels, or differences in the recovery in batches
1 to 3 versus in batches 4 through 10. In either case, the esti-
mated recovery depended on the set of samples being analyzed.
Recovery estimates for HRGC/MS measurements are shown in
Table 8. The average HRGC/MS recovery estimates for spiked
compounds in the dichloromethane samples measured were between 69%
10A two-way analysis of variance (assuming the errors in the recovery
estimates are independent) suggested that differences among compounds were not
significant and that differences among types of sample material were very
significant (p<.0001).
54
-------
and 81% for, with the exception of hexachlorobenzene with an esti-
mated recovery of 51%. The recovery estimates based on the
HRGC/MS spiked multisplit samples are very variable and not very
accurate. One compound, beta-BHC, contributes substantially to
the pooled error estimate. The precise estimate of recovery for
beta-BHC depended greatly on the weights used to average the
recovery in the four multisplit spiked portions. Removing beta-
BHC from the calculations gave a pooled error estimate of ±23%
(±28% for dieldrin). The recovery estimates for compounds other
than beta-BHC were between 26% and 56%.
The recoveries for all samples using the GC/ECD method were
similar to those from the spiked dichloromethane samples using the
HRGC/MS method; however, they were roughly twice those based on
the multisplit samples using the HRGC/MS method. The recovery
estimates differed depending on the sample matrix. Therefore, any
corrections for recovery should be based on recovery estimates
based on the same sample matrix. Considering only the recoveries
from lipid material, the GC/ECD recovery was greater than the
HRGC/MS recovery except for beta-BHC for which the recovery esti-
mates were similar.
6.3 Comparison of Percent Detected
The number of paired samples in which each compound was
detected, i.e., has a data qualifier of "Trace" or "Positively
Quantified", is shown in Tables A-3 and A-ll in Appendix A. The
nine primary compounds were detected using both the GC/ECD and
HRGC/MS methods. These compounds were used to compare the percent
detected for the two analysis methods. Table 9 shows the number
of measurements and percent detected for each method. Figure 7
shows the percentage of HRGC/MS and GC/ECD paired samples with
detectable quantities of each compound. The HRGC/MS dieldrin
measurements were determined in only 3 of the 10 batches, result-
ing in fewer HRGC/MS measurements than GC/ECD measurements for
55
-------
Table 8. Average Recovery, with 95% Confidence Intervals, for
the HRGC/MS Measurements on Spiked Multisplit and
Quality Control Samples
Compound
p,p'-DDT
p,p'-DDE
Beta-BHC c
Dieldrin d
Heptachlor Epoxide
Oxychlordane
trans-Nonachlor
Hexachlorobenzene
PCBs f
Spiked multisplit
samples a
b
26%±50%
99%±50%
37%±61%
50%±50%
42%±50%
56%±50% e
41%±50%
b
Dichloromethane spiked
samples
76%±13%
80%±21%
74%±15%
77%±75%
81%±9%
71%±6%
73%±8%
51%±5%
69%±6%
a
b
c
±95% Confidence intervals based on pooled variance
This compound was not spiked into the samples
The estimate of recovery for the lowest spike level of beta-
BHC in the multisplit samples is quite high, resulting in a
high estimate of recovery and contributing to the large
pooled variance.
Dieldrin, in Fraction 2, was analyzed in only 3 batches, thus
has fewer measurements and a larger confidence interval.
As a result of substituting zero for the missing unspiked
concentration, the calculated recovery will tend to overesti-
mate of the actual recovery.
See Chapter 9.
56
-------
CO
CD
e
(0
-P O
-H A
CO 4->
O 0)
0,2
OH
-H H
id nJ
O.C
co
(U 0)
e-H
0) 4-1
^4 -H
3H
CO (0
(0 3
Q) a
U) 4J
S 10
^ Q
O >i
C 0)
(0 4J
O
D CO
O -P
U -P CO
CD -r-l O T-
> CJ>T;
-H C C
j_j
CD TIT
a Q
oi
EH
a
si
M O
Ll *^
^ ^>
CM Q
Q
U D
W S3
PS
C*
Oj
T3
C
3
O
g
O
o
co ^j* co co
_y* Jug rjf* ^Lp ^tp
o\ in CM CM CM
00 CM 00 ^* 00
in CM in r- oo
o o o o o
CO ^* ^^ CO
dP dP dP dP dP
00000
O 0 0 0 0
o o o o o
O O CO O O
in m CM m m
^* ^* "* ^ ^*
cu
•o
-H
^
0
a
u
^
o
C rH
EH W -H ^3
Q Q MO
Q Q U -d (0
1 1 35 r-t 4J
- - CQ CO O<
a a i -H a)
CX CX «Q Q K
r-
ro
dP
CM
00
00
O
r-
co
dP
O
O
O
^r
rH
V
CD
C
-o
O
•H
A
U
X
o
5!
dP
m
CM
O
CM
dP
O
0
0
O
•9
^J1
^4
O
H
A
0
(0
C
o
z
1
CO
C
fO
M
-P
ro
cdrp
CM
00
CO
o
r-
co
dP
O
0
0
o
in
^*
0)
c
4)
N
C
0)
,Q
O
O
•H
o
(0
X
0)
r-
dP
00
rH
0
5
dP
0
0
0
5
^
CD
U
04
•d
0)
U
0)
0)
T5
4J
O
II
Q
S
<0
U
(0
EH
II
(£
EH
T3
0)
-H
H-l
-H
-P
C
(0
3
O
>*i
H
d)
»*
-H
•P
-H
CO
O
a
u
o
cu
57
-------
T5
O
A
JJ
8
W
^
CJ
O
£!
4->
Q)
CO
OS
re
CO
c
O
-H
c
0)
O
c
O
O
•o
0)
4-> •
o co
0) T>
J-> O
0) X5
73 4J
cu
x! e
S CJ
H
CO ^
0) U
-H O
t-o
(0 G
CO (0
TJ W
a) S
S-l \
-^ o
(0 O
35
«M
O Q)
X!
4-> -P
d
0) &>
o c
>-J -H
Q) CO
04 O
a)
S-l
OOOOOOOOOOO
58
-------
this compound. For PCBs the GC/ECD concentrations were coded into
concentration ranges. The coded concentrations were reported on
the analysis report forms.
The percentage of samples with detected quantities depends on
the detection limits, the recovery, and the concentration level in
the samples. Low concentrations are more likely to be detected
using methods with lower detection limits. The detection limits
apply to the concentration after recovery losses. Therefore,
given two methods with the same detection limit, the method with
lower recovery will have lower measured concentrations and possi-
bly more samples with concentrations below the detection limit.
Note that the recovery and detection limits make no difference to
the percent detected if all concentrations are high enough to be
positively quantitated in all samples.
In general, the analysis shows that the recovery for the
HRGC/MS method is similar to or lower than that for the GC/ECD
method and the detection limits for the HRGC/MS method are higher
than those for the GC/ECD method. Both of these factors would
indicate that there would generally be a higher percent detected
in the GC/ECD than the HRGC/MS paired samples. With the exception
of PCBs, this conclusion is consistent with the results shown in
Table 9. The GC/ECD detection limit for PCBs was much higher than
the HRGC/MS detection limit. However, the percent detected is
similar because the concentrations in most samples were high
enough that PCBs were detected in all GC/ECD samples.
59
-------
7 RELATIONSHIP BETWEEN THE GC/ECD AND HRGC/MS METHODS
This chapter presents the statistical procedures and results
comparing the GC/ECD and HRGC/MS measurements in the paired
samples. After a summary of the results, the relationship between
the GC/ECD and HRGC/MS measurements for each compound are dis-
cussed, followed by scatter plots of the measurements on the
paired samples.
7 .1 Modeling the Relationship Between the GC/ECD and
HRGC/MS Methods
Determining the relationship between the GC/ECD and HRGC/MS11
measurements is desirable in order to determine factors that
affect the relationship and to predict the measurements which
would be obtained using one analytical method based on the
observed measurements from the other method. The relationship
between the measurements can be expressed by a mathematical
formula such as:
BCD = f(MS) + error (7.1)
where f is a function to be determined. This equation can be
solved for the HRGC/MS measurements to give:
MS = g(ECD) + error (7.2)
where g is the inverse function of f.
Differences between the HRGC/MS and GC/ECD measurements may
be due to many factors, including differences in recovery
(extraction efficiency) and calibration. The following discussion
assumes that, due to the calibration process, the HRGC/MS and
GC/ECD measurements are unbiased estimates of the concentrations
11 HRGC/MS and GC/ECD are abbreviated as MS and BCD, respectively, in the
equations.
61
-------
in the final extract. Assuming the two portions of the split
sample have the same initial concentration, consistent differences
between HRGC/MS and GC/ECD measurements will be due to differences
in recovery. Therefore, the expected relationship between the
GC/ECD and HRGC/MS measurements is:
MS = R1 BCD (7.3)
where R' is the ratio of the recovery using the HRGC/MS method to
the recovery using the GC/ECD method. Note that R1 can also be
described as the ratio of the HRGC/MS to GC/ECD measurement.
According to equation (7,3), a doubling of either the GC/ECD or
HRGC/MS measurement should be accompanied by a doubling of the
other measurement.
Questions which might be asked about the applicability of
equation (7.3) as a model for the observed data are:
Is the ratio of the HRGC/MS to GC/ECD measurements
really constant?
• If the ratio is constant, is it different from 1.0? and
• Does the ratio of the HRGC/MS to GC/ECD measurements
depend on the batch?
Statistical analysis is used to determine if equation (7.3)
provides an adequate description of the data. For this analysis,
both the HRGC/MS and GC/ECD measurement errors are assumed to have
a lognormal distribution. The analysis of measurement error vari-
ances to support this assumption is discussed in Chapter 10. For
data with a lognormal distribution, 1) the measurement error
increases as the concentration increases such that the coefficient
of variation is constant, and 2) the log transformed data has
constant measurement error variance, independent of concentration.
To make the data consistent with the assumptions behind the
statistical analysis (i.e., that the error variance is constant),
the log transformed data are used. Taking the logarithm of the
62
-------
HRGC/MS and GC/ECD measurements in equation (7.3) gives the
following equation:
In (MS) = In(R') + In (BCD) (7.4)
The correlation of In (MS) and In (BCD) provides a descriptive
measure of the linear relationship between the HRGC/MS and GC/ECD
measurements. The correlation can be used to test if there is a
significant linear relationship between the HRGC/MS and GC/ECD
measurements.
The appropriateness of equation (7.4) for describing the data
can be tested by fitting additional terms to describe other
factors which might also affect the HRGC/MS-GC/ECD relationship.
Two additional terms were used: terms for differences between
batches and for nonconstant recovery ratio, i.e., a recovery ratio
which depends on the concentration in the sample. If these addi-
tional terms are statistically significant, there is evidence that
equation (7.3) and (7.4) do not adequately describe the HRGC/MS-
GC/ECD relationship. With these additional terms added, the model
which was fit to the data is shown below in equation (7.5) . This
model can also be obtained from the model (equation 5.10) by
equating the true concentrations in the equations for the HRGC/MS
and GC/ECD measurements and combining terms.
ln(XMcbs) = Rc + Pc * ln(XEcbs) + SbC + ecbs (7.5)
Where:
ln(XMcbs) = the HRGC/MS measurements for compound c in sample
s analyzed in batch b;
ln(XEcbs) = the GC/ECD measurements for compound c in sample s
analyzed in batch b;
Pc = a constant for each compound. This term will be 1.0 if
the recovery ratio for the two methods is constant, i.e.
independent of concentration. Testing if the recovery
63
-------
ratio is not constant is achieved by testing if J3C is
significantly different from 1.0;
Rc = a constant for each compound. If |3C = 1, this constant
can be interpreted as the log of the ratio of the
HRGC/MS recovery to the GC/ECD recovery;
&bc = a random effect associated with batch b and compound c,
assumed to be normally distributed with a mean zero and
standard deviation of cybc. This term combines the batch
effect terms in the models for both methods.
ecbs = a random error for compound c in sample s analyzed in
batch b, assumed to be normally distributed with a mean
zero and standard deviation o. This error is the
combined result of within sample errors and measurement
errors in both the HRGC/MS and GC/ECD measurements.
Equation (7.5) can be obtained from the model (equation (5.10)) by
equating the true concentration (Ccs) in equations for the HRGC/MS
and GC/ECD measurements and combining confounded terms.
Standard regression procedures can be used to estimate the
parameters Rc, |3C, and 8^ in equation (7.5). However, regression
procedures assume that only the dependent variable (HRGC/MS in
this case) has measurement error. When both the dependent and the
independent variable are measured with error, the regression esti-
mates of the parameters are biased. Fitting equation (7.5) to the
data using regression provides a functional equation. This equa-
tion is optimal in the sense that it minimizes the squared predic-
tion error for the data on which the equation is based. However,
the functional equation may provide particularly poor predictions
of HRGC/MS measurements from GC/ECD measurements when both mea-
surements have error and the equation is extrapolated beyond the
range of the original data.
Statistical models which account for measurement error in
both the independent and dependent variable are called structural
models. The slope for the functional model, estimated using
regression, is a biased estimate of the slope for the structural
64
-------
model. Under either one of the following conditions the bias is
minimized and the regression slope provides a good approximation
to the slope for both the functional and structural model:
• Measurement errors in the independent variable are small
relative to the range in the independent variable; or
• The slope of the relationship, |3C, is close to zero.
Although the slope estimate may be biased, a test of the
hypothesis that the slope is significantly different from zero is
relatively unaffected by errors in the independent variable.12
Therefore, regression results can be used to test if the slope for
either the functional or structural model is significantly differ-
ent from zero.
The slope in equation (7.5) is expected to be 1.0 if the
recovery ratio is constant. Therefore, in order to test if the
recovery ratio is constant, the slope from the structural model
must be compared to 1.0. Unfortunately, with errors in the inde-
pendent variable, regression, the primary tool for fitting equa-
tion (7.5), may perform poorly unless the slope is close to zero.
An alternate approach is to transform the data by rotating the
coordinate axes so that a statistical test comparing the regres-
sion slope to zero is used to test if the slope in the functional
model is different from 1.0.
Both the HRGC/MS and GC/ECD data from the Comparability Study
have measurement errors which cannot be ignored. As a result,
standard regression procedures must be modified to model the
structural relationship between the two measurements and to
satisfy the assumptions behind the regression procedures. To meet
these objectives, the following five steps were used to model the
relationship between the HRGC/MS and GC/ECD measurements:
12Errors in the independent variable will reduce the power of the test,
however the probability level for the test will be correct.
65
-------
(1) Use log transformed data, equalizing the measurement
error variance across concentrations within each
analysis method;
(2) Scale the data to make the measurement error for the
scaled GC/ECD and the scaled HRGC/MS data equal;
(3) Rotate the coordinate axes so that the regression slope
will be zero if the recovery ratio is constant;
(4) Use analysis of covariance on the transformed scaled and
rotated data to estimate parameters and test hypotheses;
and
(5) Transform the parameters back to the original units for
reporting and plotting.
The scaling of the data to equalize the HRGC/MS and GC/ECD
measurement errors is based on the analysis of variance components
discussed in Chapter 10. The measurement errors for the HRGC/MS
method were found to be roughly three times those for the GC/ECD
method. For estimating the structural models, a measurement error
ratio of 3 was used for all compounds.
The result of the scaling, transformation, and rotation is
equivalent to fitting the following equation to the data for each
compound:
ln(Y) = a + b * ln(Z) + ^ + error (7.6)
Where:
Y = the ratio of the HRGC/MS and GC/ECD measurements;
Z = the transformed product of the HRGC/MS and GC/ECD mea-
surements: MS'33 BCD3;
b = slope; if the slope b is equal to zero, the ratio of the
HRGC/MS and GC/ECD measurements is constant;
a = intercept; and
6b = a batch effect; if the batch effect is significant, then
there are differences in the ratio of the HRGC/MS and
GC/ECD measurements between batches.
66
-------
Equation (7.6) was fit to all positively quantified measure-
ments, ignoring pairs of measurements in which one member of the
pair was either a trace or nondetect measurement. Except for
dieldrin, both methods had positively quantifiable concentrations
for most (at least 37 out of 45) samples, thus any bias caused by
ignoring the below-detection results is expected to be small.
Because most of the trace and nondetect measurements were obtained
using the HRGC/MS method, the slope will tend to be biased toward
the low side. An analysis was performed to assess the affect of
trace and nondetect measurements. The results of the hypothesis
tests (significant versus nonsignificant) were unaffected by
replacing trace and nondetect observations by the corresponding
LOQ or LOD and including these samples into the model fit.
The full model for the comparability analysis (equation 7.6)
was fit to data for the following seven compounds: p,p'-DDT, p,p'-
DDE, beta-BHC, heptachlor epoxide, oxychlordane, trans-nonachlor,
and hexachlorobenzene. For hexachlorobenzene the relationship
between the HRGC/MS measurements and both the corrected and uncor-
rected GC/ECD measurements were modeled. Due to the small number
of paired observations for dieldrin, the results for dieldrin are
approximate and differences between batches could not be tested.
The results of fitting equation (7.6) to the data include
parameter estimates for the slope and tests of the hypotheses that
there are no batch effects and that the ratio of the HRGC/MS to
GC/ECD measurements (i.e., the recovery ratio) is constant. If
the regression slope is significantly different from zero, then
the hypothesis that the recovery ratio is constant is rejected.
The equation which fits the data has the form:
MS = R1 * ECD d (7.7)
where R1 and d are constants. The exponent d is determined by
transforming the slope in equation (7.6) back to the original
scale, involving an adjustment to correct for measurement error
67
-------
and to reverse the rotation, scaling, and log transformations.
Equation (7.7) for the HRGC/MS-GC/ECD relationship is referred to
as the "best fit" equation. The constant, R1, in equation (7.7),
depends on the measurement units. For the analysis, all concen-
trations are in micrograms per gram of extracted lipid (ug/g). If
different measurement units are used (for example, nanograms per
gram) the exponent in equation (7.7) remains the same, however the
constant must be adjusted for the new measurement units. No
adjustment is necessary if d = 1.0.
The scaling constant R1 is calculated so that the best fit
line passes through the average of the HRGC/MS and GC/ECD measure-
ments. The method for calculating the averages depends on assump-
tions about the significance of a batch effect. If there are
significant differences between measurements in different batches,
an estimate of R' is obtained by averaging the measurements within
a batch and then averaging the batch averages to obtain the
overall average, as in equation (7.8):
nv
• « S 10 m <7'8'
where m is the number of paired samples with measurements within
batch b.
If there are no differences between batches, then an estimate
of R' is obtained by averaging across all samples as in equation
(7.9) :
In(R') = *—l —^ fc (7.9)
where n is the number of paired samples with measurements.
68
-------
The two equations for estimating R' are equivalent if the
number of measurements in each batch is the same. Because (I) the
number of samples per batch vary from I to 5, (2) there are
statistically significant batch effects for some, if not all,
compounds, and (3) batch effects are significant based on the
components of variance, equation (7.8) was used in all cases to
estimate R1.
7.2 Comparison of HRGC/MS and GC/ECD Measurements
Statistical hypothesis tests were used to test the following
overall hypotheses: (1) there is no linear relationship between
the GC/ECD and the HRGC/MS measurements, (2) there are no differ-
ences between batches, (3) the recovery ratio is constant, and (4)
the average recovery ratio is 1.0. Each of these overall hypothe-
ses were tested at the 95 percent confidence level. There were
multiple chances to test each hypothesis, using data from each of
the primary compounds. To limit the probability of incorrectly
rejecting these overall hypotheses due to the use of hypothesis
tests on multiple individual compounds, the following procedure,
based on the Bonferroni inequality, was used. An overall hypothe-
sis was rejected at the 5 percent level if the probability level
for the hypothesis test using any one compound was less than 0.71
percent13. If the overall hypothesis was rejected at the 5 percent
level then hypothesis tests for each compound, testing at the 5
percent level, were used to identify individual compounds for
which the data were inconsistent with the hypothesis. Using this
procedure, the probability of rejecting an overall hypothesis
using all of the compounds is less than 5 percent.
For the first overall hypothesis, the correlations of the log
transformed HRGC/MS and GC/ECD measurements are shown in Table 10,
13There are 7 compounds with data which are both independent and numerous
enough to test the hypotheses (corrected and uncorrected Hexachlorobenzene are
considered to be one compound and dieldrin is not counted) . The conservative
formula for the Bonferroni limit would set alpha (i.e., 1 - the confidence
level) for the overall hypothesis test to 0.05/7 = .0071 or 0.71 percent.
69
-------
along with the results of a test of the hypothesis that the true
correlation is zero (i.e., no linear relationship). The signifi-
cance levels are only approximate, since possible batch effects
have not been considered. The overall hypothesis of no linear
relationship is rejected at the 5 percent level, indicating that
there is a statistically significant linear relationship between
the measurements from the HRGC/MS and the GC/ECD methods. As can
be seen from Table 10, the hypothesis of no linear relationship is
rejected at the 5 percent level for all compounds except dieldrin.
Significant correlations are indicated by using bold type.
For testing the second and third overall hypotheses, equation
(7.6) was fit to the GC/ECD and HRGC/MS data for the primary
compounds. Table 11 shows the results of the statistical tests to
answer the following questions:
• Were there consistent differences in the HRGC/MS-GC/ECD
relationship between batches (i.e., are there batch
effects)? and
• For samples within batches, does the ratio of the
HRGC/MS to GC/ECD measurement depend on the concentra-
tion (i.e., is there non-constant recovery)?
Both the overall hypotheses of no batch effects and constant
recovery are rejected, indicating that, at least for some com-
pounds, the recovery ratio differs among batches or that the
recovery depends on the concentration. As can be seen from Table
11, the hypothesis of no batch effects is rejected at the 5
percent level for p,p'-DDE, beta-BHC, oxychlordane, and trans-
nonachlor. The hypothesis of constant recovery is rejected at the
5 percent level for p,p'-DDE, beta-BHC, and hexachlorobenzene
(both corrected and uncorrected).
70
-------
Table 10. Correlation Between Log Transformed HRGC/MS and
GC/ECD Measurements
Compound
p,p'-DDT
p,p'-DDE
Beta-BHC
Dieldrin
Heptachlor
Epoxide
Oxychlordane
trans-
Nonachlor
Uncorrected
Hexachloro-
benzene
Corrected
Hexachloro-
benzene
Number of
paired
samples
39
41
39
5
37
37
41
37
37
Correlation21
0.86
0.84
0 .52
0.52
0.57
0.57
0.84
0.63
0.71
Significance test
for correlation
different from
zero
p < .0001
p < .0001
p = .0008
p = .68
p = .0002
p = .0002
p < .0001
p < .0001
p < .0001
aThe overall hypothesis of no linear relationship between the HRGC/MS and
the GC/ECD measurements was rejected at the 5 percent level using the
Bonferroni approach. Values in Bold text identify individual compounds
for which the hypothesis is rejected.
71
-------
Table 11. Summary of the Statistical Tests for Batch Effects and
Nonconstant Recovery
Compound
p,p'-DDT
p,p'-DDE
Beta-BHC
Dieldrin
Heptachlor
Epoxide
Oxychlordane
trans-
Nonachlor
Uncorrected
Hexachloro-
benzene
Corrected
Hexachloro-
benzene
Test for
Batch
Effects3
p = .37
p = .014
p = .0001
p = .13
p = .024
p = .0089
p = .13
p = .25
Test for a
nonconstant
HRGC/MS-
GC/ECD
measurement
ratio3
p = .16
p < .0001
p = .014
p = .99
p = .44
p = .065
p = .86
p < .0001
p = .0006
Comments
Batch effects are not
significant after removing
batch 3
Data in batches 1, 2, and 3
are significantly different
than the other batches
Not enough data to fit a
batch effect
There is one extreme
observation
Data in batches 1, and 2
are significantly different
than the other batches
3The overall hypotheses of no batch effects and constant recovery were re-
jected at the 5 percent level using the Bonferroni approach. Values in Bold
text identify individual compounds for which the hypothesis is rejected.
72
-------
On the assumption that the recovery ratio is constant (i.e.,
d = 1.0), the fourth overall hypothesis was tested using the ratio
of the GC/ECD to the HRGC/MS measurements in equation (7.8). The
results are presented in Table 12 and Figure 8. The confidence
intervals were calculated using a t-statistic and the standard
error of ln(R')/ assuming the batch means are independent.
Based on the confidence intervals, the overall hypothesis
that the recovery ratio is equal to 1.0 is rejected at the 5
percent level, indicating that, for at least some compounds,
recoveries for the HRGC/MS and the GC/ECD methods are different.
As can be seen from Table 12, the hypothesis that the recovery
ratio is equal to 1.0 is rejected at the 5 percent level for all
compounds except beta-BHC and dieldrin. For compounds for which
the test for nonconstant recovery ratio is significant, the calcu-
lated ratio (and confidence interval) represents an average ratio
for composite samples in the 1984 NHATS survey.
Table 12 also shows the ratio of the GC/ECD to HRGC/MS recov-
eries estimated from the multisplit spiked samples. Although this
ratio estimate is not very precise, the correlation of the recov-
ery ratio calculated from the multisplit samples and the ratio of
the GC/ECD to HRGC/MS measurements is statistically significant at
the 5 percent level.
The estimated GC/ECD to HRGC/MS ratios were greater than 1.0
for all compounds, indicating that the recovery for the GC/ECD
method is typically greater than for the HRGC/MS method. However,
the 95% confidence intervals for beta-BHC and dieldrin include
1.0. For all other compounds tested, the GC/ECD recovery was
significantly greater than that for the HRGC/MS method.
73
-------
Table 12. Geometric Mean Ratio of GC/ECD to HRGC/MS Measurements
Compound
p,p'-DDT
p,p'-DDEc
Beta-BHCc
Dieldrin
Heptachlor
Epoxide
Oxychlordane
trans-
Nonachlor
Uncorrected
Hexachloro-
benzene0
Corrected
Hexachloro-
benzenec
Geometric
Mean Ratio of
GC/ECD to
HRGC/MS
Measurements3
3.88
2. 67
1.25
2.63
2.76
2.04
2.31
1. 41
2.11
95%
Confidence
interval
3.38-4.47
1.99-3.59
0.78-2.02
0.50-14.0
2.13-3.57
1.52-2.75
1.90-2.80
1.19-1.68
1.85-2.41
Ratio of
GC/ECD to
HRGC/MS
Recoveries in
Multisplit
Samples
b
3.15
0.90
2.43
1.66
1.74
1.30
1.29
1.88
p-value
<.05
<.05
<.05
<.05
<.05
<.05
<.05
aThe overall hypothesis that the recovery ratio is 1.0 was rejected at the
5 percent level using the Bonferroni approach. Values in Bold text
identify individual compounds for which the hypothesis is rejected.
bp,p'-DDT was not spiked into the multisplit samples.
°The data suggests that the recovery ratio is not constant for these
compounds.
74
-------
100 Tr-
-
s o
d \ -H
CJ O -U
O
CD
•• I
0.1
Mean
•95% CI
!
!
1
H
Q
Q
1
—
p.
*.
a
i
u
a
Q
i
•
o<
*.
a
i
o
as
m
i
^
i
c
-rH
M
•o
1 — 1
CD
•H
a
1
M
o <„
^.
,c j
0 X
to 3
-P ^
0) W
X
1
0)
c
(0
TJ
M
O
rH
fi
0
>^
X
o
1
M
o
•H
A
0
(0
c
o
s
1
CO
c
M
4J
1
T3 1
fl) O
I 1 J^
0 0
d> -H
M .C
M 0
O (0
0 X
C 0)
D as
"O
CD 0)
C 4J
0) U
N 0)
C M
(V M
X! 0
U
1
1
O
M
O
rH
.C
O
(0
X
0>
0)
c
0)
N
c
0)
XI
1
Figure 8 Geometric mean ratio of the GC/ECD and HRGC/MS
measurements for primary compounds, with 95% confidence
intervals.
75
-------
7.3 Comparisons for Each Compound
The results for each compound are discussed below along with
an equation describing the relationship between the HRGC/MS and
the GC/ECD measurements assuming constant recovery ratio. If the
assumption of constant recovery ratio is rejected based on the
statistical tests, both the best fit equation and a simpler equa-
tion based on the assumption of constant recovery ratio are
presented. Although the best fit equation is likely to apply over
a wider range of concentrations, the simpler equation is easier to
use and describes the ratio of the geometric mean HRGC/MS and
geometric mean GC/ECD measurements for the 1984 NHATS survey.
p.p'-DDT
Neither the test for batch effects nor the test for noncon-
stant recovery ratio was significant. The equation which best
approximates the relationship between the HRGC/MS p,p'-DDT and
GC/ECD p,p'-DDT measurements in the 1984 NHATS survey is:
BCD
MS - Y~QQ = °-26 * ECD (7.10)
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were significant, indicating that the recoveries for the two
methods are different.
pfp'—DDE
The test for nonconstant recovery ratio was highly signifi-
cant (p < .0001) and the test for batch effects was significant (p
= .0136). The data for batch 3 are noticeably different from the
other batches. After removing the data for batch 3, the batch
effect is no longer significant and the test for nonconstant
recovery ratio is still highly significant (p < .0001). Because
the slope estimates are almost identical with and without batch 3,
76
-------
all of the data are used for the summary statistics presented
below.
The equation which best approximates the relationship between
the HRGC/MS p,p'-DDE and GC/ECD p,p'-DDE measurements is:
MS = 0.47 * BCD0-64 (7.11)
The following equation relates the HRGC/MS p,p'-DDE and
GC/ECD p,p'-DDE measurements in the 1984 NHATS survey, and can be
used for extrapolation if the recovery ratio can be assumed to be
constant :
MS - -=- = 0.37 * ECD (7.12)
Z . D /
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were significant, indicating that the recoveries for the two
methods are different .
Beta-BHC
The test for nonconstant recovery ratio was significant
(p = .0138) and the test for batch effects was highly significant
(p = .0001) . Although the batch differences follow a general
trend, with higher measurement ratios in batches 1, 2, and 3 than
in later batches, no observations or batches are obviously
unusual. Excluding the first three batches makes little differ-
ence in the results.
The equation which best approximates the relationship between
the HRGC/MS beta-BHC and GC/ECD beta-BHC measurements is :
MS = 0.38 * ECD0-64 (7.13)
The following equation relates the HRGC/MS beta-BHC and
GC/ECD beta-BHC measurements in the 1984 NHATS survey, and can be
77
-------
used for extrapolation if the recovery ratio can be assumed to be
constant:
MS = Y~^J = 0.80 * ECD (7.14)
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were not statistically significant.
Dieldrin
There were only five paired samples with positively quanti-
fied measurements for dieldrin. Although the full model could not
be fit to so few data points, the data were used to test for
nonconstant recovery ratio. No significant relationships were
found.
The equation which best approximates the relationship between
the HRGC/MS dieldrin and GC/ECD dieldrin measurements in the 1984
NHATS survey is:
MS = T = 0.38 * ECD (7.15)
/ . D-3
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were not statistically significant.
Heptachlor Epoxide
Neither the test for batch effects nor the test for noncon-
stant recovery ratio was significant. The lowest observation was
particularly influential on the regression fit. However, with
this observation removed, the conclusions are unchanged.
The equation which best approximates the relationship between
the HRGC/MS heptachlor epoxide and GC/ECD heptachlor epoxide
measurements in the 1984 NHATS survey is:
78
-------
MS = -r7 = 0.36 * BCD (7.16)
Z . /O
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were significant, indicating that the recoveries for the two
methods are different.
Qxychlordane
The test for nonconstant recovery ratio was not significant;
however, the recovery ratios for the GC/ECD and HRGC/MS oxychlor-
dane measurements differed significantly between batches (p =
.0243).
The equation which best approximates the relationship between
the HRGC/MS oxychlordane and GC/ECD oxychlordane measurements in
the 1984 NHATS survey is:
ECD
MS = 2(j£ - 0-49 * ECD (7.17)
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were significant, indicating that the recoveries for the two
methods are different.
trans-Nonachlor
The test for nonconstant recovery ratio was not significant,
however, the ratio of the GC/ECD and HRGC/MS recoveries varied
significantly between batches (p = .0089). The ratio of the
GC/ECD to HRGC/MS measurement was smaller in batches 1 and 2 than
for the other batches.
The equation which best approximates the relationship between
the HRGC/MS trans-nonachlor and GC/ECD trans-nonachlor measure-
ments in the 1984 NHATS survey is:
79
-------
ECD
MS = Y~3l = °'43 * ECD (7.18)
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were significant, indicating that the recoveries for the two
methods are different.
Uncorrected Hexachlorobenzene
The test for nonconstant recovery ratio of uncorrected hexa-
chlorobenzene was highly significant (p < .0001) and the test for
batch effects was not significant. Inspection of the plot of the
data and the residuals from the fit suggested that there were five
observations which might be judged to be unusual and were influen-
tial on the estimated slope. If these five observations were
removed from the model, the test for nonconstant recovery ratio is
still significant (p = .0332). The five unusual observations are
associated with the lowest three and highest two GC/ECD measure-
ments. The estimated slope depends on which observations are
removed from the data set. Because none of the observations are
clearly in error, the results summarized below are based on all
data points.14
The equation which best approximates the relationship between
the HRGC/MS hexachlorobenzene and GC/ECD uncorrected hexachloro-
benzene measurements is:
MS - 0.12 * ECD0-47 (7.19)
The following equation relates the HRGC/MS hexachlorobenzene
and GC/ECD uncorrected hexachlorobenzene measurement in the 1984
NHATS survey, and can be used for extrapolation if the recovery
ratio can be assumed to be constant:
14If the two most influential points are removed, the results change very
little and the conclusions do not change.
80
-------
ECD
MS = —£Y = 0-41 * ECD (7.20)
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were significant, indicating that the recoveries for the two
methods are different.
Corrected Hexachlorobenzene
The test for nonconstant recovery ratio for Hexachlorobenzene
was very significant (p = .0006) and the test for batch effects
was not significant. As with the uncorrected hexachlorobenzene,
several points are influential in determining the slope; however,
removing these points does not change the conclusion that there is
a nonconstant recovery ratio.
The equation which best approximates the relationship between
the HRGC/MS hexachlorobenzene and GC/ECD corrected hexachloroben-
zene measurements is:
MS = 0.13 * ECD0-56 (7.21)
The following equation relates the HRGC/MS hexachlorobenzene
and GC/ECD corrected hexachlorobenzene measurement in the paired
1984 NHATS samples, and can be used for extrapolation if the
recovery ratio can be assumed to be constant:
MS = j~ = 0.47 * ECD (7.22)
Differences between the geometric mean HRGC/MS and GC/ECD measure-
ments were significant, indicating that the recoveries for the two
methods are different. Note that the corrected GC/ECD hexachloro-
benzene measurements were adjusted by the laboratory to correct
for low recovery, increasing the ratio of the GC/ECD to HRGC/MS
measurements from 1.41 to 2.11.
81
-------
7.4 Plots of HRGC/MS versus 6C/ECD Measurements
Plots of the HRGC/MS versus GC/ECD concentration measurements
of primary compounds in paired composite human adipose tissue
samples are shown in Figures 9 through 17. Figures 18 through 26
show the same data after the transformation, scaling, and rotation
required for the statistical tests. Each figure shows:
• An open diamond (O), which indicates the paired HRGC/MS
and GC/ECD measurements used to compare the two methods;
• A dotted line (- -), which serves as a reference line
showing where the HRGC/MS and GC/ECD measurements are
equal;
• A dashed line (— —), which indicates the best fit
relationship between the HRGC/MS and GC/ECD measurements
under the assumption that the HRGC/MS measurements are
proportional to the GC/ECD measurements (i.e., the
recovery ratio is constant, independent of concentra-
tion) ; and
• A solid line ( ), which indicates the best fit
relationship between the HRGC/MS and GC/ECD measurements
using equation 7.7.
In addition, Figures 9 through 17 also show:
• A closed diamond (•), indicates paired HRGC/MS and
GC/ECD measurements where at least one of the measure-
ments was trace or not detected. The plotted value is
the LOD for non-detect measurements and the measured
amount for trace measurements.
82
-------
1
0.9
0.8 4-
0.7
0.6
HRGC/MS
Measurement 0.5
(ppm)
0.4
0.3
0.2
0.1 •-
1—I—I—I—I—I—I—I
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
GC/ECD Measurement (ppm)
Trace or non- O Paired data """Best fit for
detect for one for comparing paired data
or both methods methods
Reference line,
HRGC/MS equals
GC/ECD
'Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 9. HRGC/MS versus GC/ECD concentration measurements for
p,p'-DDT in paired composite human adipose tissue
samples.
83
-------
HRGC/MS
Measurement
3
2
1 • •
H 1 1 1 1 1 I 1 1
123456789
GC/ECD Measurement (ppm)
Trace or non-
detect for one
or both methods
""" Reference line,
HRGC/MS equals
GC/ECD
1 Paired data '
for comparing
methods
•Best fit for
paired data
•Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 10.
HRGC/MS versus GC/ECD concentration measurements for
p,p'-DDE in paired composite human adipose tissue
samples .
84
-------
HRGC/MS
Measurement
(ppm)
0.7
0.6
0.5
0.4
0.3
0.2
0.1 -• *
0 0.1 0.2 0.3 0.4 0.5 0.6
GC/ECD Measurement (ppm)
0.7
Trace or non- O Paired data ^^Best fit for
detect for one for comparing paired data
or both methods methods
Reference line,
HRGC/MS equals
GC/ECD
'Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 11.
HRGC/MS versus GC/ECD concentration measurements for
beta-BHC in paired composite human adipose tissue
samples .
85
-------
HRGC/MS
Measurement 0.1 +
(ppm)
I 1 1 1 1 1 1 1 1 1
0 0.020.040.060.08 0.1 0.120.140.160.18 0.2
GC/ECD Measurement (ppm)
Trace or non- O Paired data ™^Best fit for
detect for one for comparing paired data
or both methods methods
Reference line,
HRGC/MS equals
GC/ECD
'Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 12 .
HRGC/MS versus GC/ECD concentration measurements for
dieldrin in paired composite human adipose tissue
samples .
86
-------
HRGC/MS
Measurement
(ppm)
0.35
0.3
0.25
0.2
0.15 • •
0.1
0.05
0 0.05 0.1 0.15 0.2 0.25 0.3
GC/ECD Measurement (ppm)
\
0.35
Trace or non- O Paired data ^^"Best fit for
detect for one for comparing paired data
or both methods methods
Reference line,
HRGC/MS equals
GC/ECD
'Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 13.
HRGC/MS versus GC/ECD concentration measurements for
heptachlor epoxide in paired composite human adipose
tissue samples.
87
-------
0.2
0.18
0.16
0.14
0.12
HRGC/MS
Measurement 0.1 +
(ppm)
I—I—I—I—I—I—I—I—I—I
0 0.020.040.060.08 0.1 0.120.140.160.18 0.2
GC/ECD Measurement (ppm)
Trace or non-
detect for one
or both methods
™~ Reference line,
HRGC/MS equals
GC/ECD
Paired data
for comparing
methods
Best fit for
paired data
"•"• "Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 14.
HRGC/MS versus GC/ECD concentration measurements for
oxychlordane in paired composite human adipose tissue
samples .
88
-------
0.4
0.35
0.3
0.25
HRGC/MS
Measurement 0.2 +
(ppm)
0.15
0.1 • •
0.05
H 1 1 1 I 1 1 1
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
GC/ECD Measurement (ppm)
Trace or non- O Paired data "••••Best fit for
detect for one for comparing paired data
or both methods methods
Reference line,
HRGC/MS equals
GC/ECD
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 15.
HRGC/MS versus GC/ECD concentration measurements for
trans -nonachlor in paired composite human adipose
tissue samples.
89
-------
0.25
0.2
0.15
HRGC/MS
Measurement
(ppm)
0.1
0.05
0.05 0.1 0.15 0.2
GC/ECD Measurement (ppm)
0.25
Trace or non- O Paired data """Best fit for
detect for one for comparing paired data
or both methods methods
Reference line,
HRGC/MS equals
GC/ECD
"•"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 16.
HRGC/MS versus GC/ECD concentration measurements for
uncorrected hexachlorobenzene in paired composite human
adipose tissue samples.
90
-------
HRGC/MS
Measurement
(ppm)
0.35
0.3
0.25
0.2
0.15
0.1 • -
0.05
0.05 0.1 0.15 0.2 0.25 0.3 0.35
GC/ECD Measurement (ppm)
Trace or non- O Paired data ^^Best fit for
detect for one for comparing paired data
or both methods methods
Reference line,
HRGC/MS equals
GC/ECD
•"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 17.
HRGC/MS versus GC/ECD concentration measurements for
corrected hexachlorobenzene for recovery in paired
composite human adipose tissue samples.
91
-------
Ratio of
HRGC/MS to
GC/ECD
Measurement
0.1
0.01
111 nun i 11 linn i 11 nun i 11 mm i 11 nun
0.00001 0.0001
0.001
0.01
0.1
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33) (ECD**3)
Paired data
for comparing
methods
™ Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
'Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 18.
Transformed HRGC/MS versus GC/ECD p,p'-DDT measurements
used for statistical tests.
92
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
0.1
*
* o
0.01
0.1
10
100
1000
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33)(ECD**3)
O Paired data
for comparing
methods
~" •" Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 19.
Transformed HRGC/MS versus GC/ECD p,p'-DDE measurements
used for statistical tests.
93
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
1 • •
0.1
0.000001 0.00001 0.0001
0.001
0.01
0.1
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33)(ECD**3)
Paired data
for comparing
methods
"• Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
'Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 20.
Transformed HRGC/MS versus GC/ECD beta-BHC measurements
used for statistical tests.
94
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
0.1
1 I I I INI
1—I I I I III
0.00001
0.0001
0.001
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33) (ECD**3)
Paired data
for comparing
methods
™ Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 21.
Transformed HRGC/MS versus GC/ECD dieldrin measurements
used for statistical tests.
95
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
0.1
0.000001 0.00001
0.0001
0.001
0.01
0.1
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33) (ECD**3)
OPaired data
for comparing
methods
~ ~ Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 22.
Transformed HRGC/MS versus GC/ECD heptachlor epoxide
measurements used for statistical tests.
96
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
1 • •
0.1
I I IN Mil 1 I 11 Hill 1 I 11 Illli 1 I 11 Hill
0.000001
0.00001
0.0001
0.001
0.01
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33)(ECD**3)
paired data
for comparing
methods
— Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 23.
Transformed HRGC/MS versus GC/ECD oxychlordane
measurements used for statistical tests.
97
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
1 • •
•
o
•
o
0 0
o o <*>
Jk —^ ^%
^f ^7 ^Bk»
O
0.1 H—HHHHf—h++BW|—h+ttMH—M+Hftf—H-H+ttfl
0.000001 0.00001 0.0001 0.001 0.01 0.1
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33)(ECD**3)
O Paired data
for comparing
methods
•" ~ Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 24.
Transformed HRGC/MS versus GC/ECD trans-nonachlor
measurements used for statistical tests.
98
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
iiinil i 11inn i 11nuu
0.000000 0.000001 0.00001 0.0001 0.001 0.01
1
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33)(ECD**3)
OPaired data
for comparing
methods
~ ~ Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 25.
Transformed HRGC/MS versus GC/ECD uncorrected
hexachlorobenzene measurements used for statistical
tests.
99
-------
10
Ratio of
HRGC/MS to
GC/ECD
Measurement
0.1
i i HIM—i i lima i i iimil i friinu—M-HHHI
0.000001 0.00001 0.0001
0.001
0.01
0.1
Product of Scaled HRGC/MS and GC/ECD Measurements
(MS**.33)(ECD**3)
Paired data
for comparing
methods
~ Reference line,
HRGC/MS equals
GC/ECD
•Best fit for
paired data
"Best fit assuming
HRGC/MS proportional
to GC/ECD
Figure 26.
Transformed HRGC/MS versus GC/ECD corrected
hexachlorobenzene for recovery measurements used for
statistical tests.
100
-------
8 COMPARISON OF 6C/ECD AND HRGC/MS MEASUREMENTS ACROSS
YEARS
One objective of the Comparability Study is to assess whether
the relationship between the HRGC/MS and GC/ECD measurements in
1984 is useful for comparing trends in concentrations measured by
different methods in different years. The analysis in this
chapter uses the results from the comparability analysis to adjust
the 1982 HRGC/MS measurements for comparison to the GC/ECD mea-
surements using plots of the data.
An important consideration in the decision as to whether
measurements from different years and methods can be compared is
the extent to which any relationship between the GC/ECD and
HRGC/MS results from 1984 might apply to other years. The statis-
tical procedures discussed in Chapter 7 can be used to find the
"best" relationship to predict the 1984 HRGC/MS measurements from
the 1984 GC/ECD measurements. However, this relationship may not
be best for data collected in another year, data analyzed by
another laboratory, or for samples with different concentration
levels than those for which the equation was developed. These
issues are discussed at the end of this chapter.
8 . 1 Plots of Measurements Over Time
Figures 27 through 34 show the arithmetic average GC/ECD and
HRGC/MS measurements for the NHATS design samples15 from 1970
through 1984. The GC/ECD method was used in 1970 through 1981 and
in 1983 and 1984. The HRGC/MS method was used in 1982 and 1984.
For these two years/ both the HRGC/MS average and the adjusted
HRGC/MS average are shown. The averages in the figures are
weighted by age group and therefore approximate national averages.
15The design samples are those samples which are within the quota for the
hospital. Some hospitals collect more samples than required under their
quota.
101
-------
The adjusted HRGC/MS average approximates the arithmetic
average concentration which would have been obtained from the
GC/ECD method. It was calculated by multiplying the HRGC/MS
average for a compound by the corresponding geometric mean ratio
of the GC/ECD to HRGC/MS measurements in Table 12, referred to
here as the adjustment ratio. The calculation of the adjusted
HRGC/MS average assumes that the ratio of the GC/ECD to HRGC/MS
measurements observed in the ' FY84 samples is constant across
years. The adjusted HRGC/MS average for 1982 can be compared to
the trend in the GC/ECD data in the years around 1982 to assess if
this procedure for adjusting the HRGC/MS data provides a reason-
able approximation to the averages made using the GC/ECD method.
In calculating the average for the GC/ECD method, zero was
used for nondetect measurements and the approximate LOQ16 was used
for trace measurements. For the HRGC/MS average, the value LOD/2
was used for the nondetect measurements and the measured amount
for trace measurements. Differences in how the trace and non-
detect measurements were handled result in only small differences
between the GC/ECD and adjusted HRGC/MS averages. The adjusted
HRGC/MS averages for 1984 and corresponding GC/ECD averages for
1984 are not exactly equal due to a combination of 1) different
procedures for handling nondetect and trace measurements, 2)
exclusion from the averages of specimens from one hospital17, 3)
inclusion of measurements from the unpaired samples, and 4) use of
only paired samples to calculate the adjustment ratio. Note that
for dieldrin, only five paired measurements were used to calculate
the adjustment ratio. However, the GC/ECD average for dieldrin is
based on 42 measurements and the HRGC/MS average is based on 11
measurements.
16The procedures for recording the data have changed slightly over time. The
majority of trace measurements use the LOQ. Of the remaining cases, some use
the LOQ/2, some use a value higher than the LOQ.
17During the course of the Comparability Study, multiple specimens from one
hospital were suspected of coming from the same donor. It was decided to
exclude specimens from this hospital from the comparisons across years. For
the years in which composite samples were analyzed, 4 composite samples were
excluded from each of the 1982 and 1984 averages.
102
-------
0.4 -•
0.2 -•
0
IJ-— — «»»«ec
1970 1972 1974
1976 1978
Year
1980 1982 1984
•GC/ECD
•HRGC/MS --*- HRGC/MS
adjusted
Figure 27.
Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
p,p'-DDT concentrations for design samples from 1970
through 1984.
103
-------
8 T
7 ••
6 -•
a
04
. 5 -•
o
-H
-U
(0
s-l
4J
c
a>
o
c
o
u
4 -•
3 -•
2 -•
1 -•
0
1970 1972
1974 1976 1978
Year
1980 1982 1984
•GC/ECD
'HRGC/MS --
- HRGC/MS
adjusted
Figure 28.
Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
p,p'-DDE concentrations for design samples from 1970
through 1984.
104
-------
U T 1 1
1970 1972 1974
1 1
1976 1978
Year
1" — " HRGC/MS
1 1 r
1980 1982 1984
--^- HRGC/MS
adjusted
Figure 29.
Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
beta-BHC concentrations for design samples from 1970
through 1984.
105
-------
0.3
0.25
e
o.
a
G
O
0)
o
o
u
0.2 -•
0.15 -•
0.1 -•
0.05 -•
U T 1 1
1970 1972 1974
— I ' l —
1976 1978
Year
^^ HRGC/MS
1 1 1
1980 1982 1984
- - O- - HRGC/MS
adjusted
Figure 30. Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
dieldrin concentrations for design samples from 1970
through 1984.
106
-------
0.16
^ 0.06 +
c
o
CJ
0.04 +
0.02 -•
o -4
+
+
+
•+
1970 1972 1974 1976 1978 1980 1982 1984
Year
•GC/ECD
'HRGC/MS
- HRGC/MS
adjusted
Figure 31. Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
heptachlor epoxide concentrations for design samples
from 1970 through 1984.
107
-------
0.16 -r
0.14 -•
0.12 -•
Q.
a
. O.i 4
c
o
•H
$ 0.08
i-4
4-J
g 0.06
o
o
0.04
0.02 -•
U T 1 1
1970 1972 1974
1 1 —
1976 1978
Year
0 — -" HRGC/MS
1 1 r
1980 1982 1984
* - HRGC/MS
adjusted
Figure 32.
Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
oxychlordane concentrations for design samples from
1972 through 1984.
108
-------
0.2 -•
0.18 -•
0.16 -•
0, 0.14 -•
c 0.12
o
-H
£01 +
(u v • U.
g 0.08 -I-
o
c
0.06 -•
0.04 -•
0.02 -•
+
1 h
1970 1972 1974 1976 1978 1980 1982 1984
Year
•GC/ECD
HRGC/MS --
- HRGC/MS
adjusted
Figure 33. Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
trans-nonachlor concentrations for design samples from
1975 through 1984.
109
-------
0.18 j
0.16 -•
0.14 -•
0.12 -•
I
I
•S
-p
(0
_p
G
0)
o
c
o
o
0
0
0.1
.08
.06
0.04 -•
0.02 -•
_i , . 1 • 1 . , , , 1 1 • f
1970 1972 1974 1976 1978 1980 1982 1984
Year
•GC/ECD
HRGC/MS - -
- HRGC/MS
adjusted
Figure 34.
Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
corrected hexachlorobenzene concentrations for design
samples from 1974 through 1984.
110
-------
8 .2 Assessment of the Results
As can be seen from Figures 27 through 34, the 1982 adjusted
HRGC/MS average is closer to the trend in the GC/ECD averages than
is the unadjusted average for four compounds, p,p'-DDT, p,p'-DDE,
heptachlor epoxide, and trans-nonachlor. Although the unadjusted
beta-BHC HRGC/MS average approximates the trend in the GC/ECD
averages more closely than' the adjusted value, both the adjusted
and unadjusted beta-BHC values are close to the GC/ECD trend. For
oxychlordane there were no 1982 measurements for comparison.
For two compounds, dieldrin and corrected hexachlorobenzene,
the adjusted 1982 averages are considerably higher than the neigh-
boring GC/ECD averages. The 1982 hexachlorobenzene measurements
are known to have two very high observations, which might be
considered to be outliers. These values contribute to both the
high average concentration and to a high standard error of the
average. The difference between the HRGC/MS adjusted average and
the GC/ECD trend may be due to the presence of the outliers. With
the two outliers removed, the adjusted HRGC/MS average reasonably
follows the GC/ECD trend, as can be seen in Figure 35.
For dieldrin, the adjusted HRGC/MS average does not agree
well with the trend in the GC/ECD averages; however, the differ-
ences may be explained by variation in the estimates of the
HRGC/MS average and the adjustment ratio. The adjustment ratio
from Table 12 has a wide 95% confidence interval due to the small
number of paired measurements in 1984. In addition, the estimate
of the 1984 HRGC/MS average is imprecise due to the small number
of fraction 2 extracts analyzed using the HRGC/MS method.
Although the adjustment ratio results in poor agreement between
the HRGC/MS adjusted averages and the GC/ECD trend, there are
values within the 95% confidence intervals that would provide good
agreement. Additional paired samples would be needed to provide a
more precise ratio estimate with which to evaluate the adjustment
ratio for dieldrin.
Ill
-------
0.18 j
0.16 ••
0.14 -•
I 0.12 -•
0 0.1 -f
0.08 -•
c
0)
g 0.06 -4-
o
0.04 -•
0.02 -•
1970 1972 1974 1976 1978 1980 1982 1984
Year
•GC/ECD
HRGC/MS
~ HRGC/MS
adjusted
Figure 35.
Weighted average GC/ECD, HRGC/MS, and HRGC/MS adjusted
corrected hexachlorobenzene concentrations for design
samples from 1970 through 1984, with two outliers
removed from the calculation of the 1982 HRGC/MS and
HRGC/MS adjusted average.
112
-------
For all compounds tested, given the likely errors in estima-
tion of the yearly averages and the adjustment ratios, the proce-
dure of using the adjustment ratio from the 1984 data (Table 12)
and the average HRGC/MS concentration from 1982 to approximate the
1982 GC/ECD averages cannot be rejected based on the data. For
five compounds, the adjusted HRGC/MS average was close to the
GC/ECD trend. For one other compound, the adjusted HRGC/MS
average was close to the GC/ECD trend after removing two outliers
from the HRGC/MS data. For one additional compound, the data and
the ratio estimates were too variable to attribute the observed
differences in the GC/ECD and HRGC/MS adjusted averages to the
adjustment procedure. For one compound, oxychlordane, no 1982
measurements are available for comparison.
Extending this procedure to other years requires making the
assumption that the ratio of the GC/ECD measurements to the
HRGC/MS measurements is constant for all years. The relationship
between the GC/ECD and HRGC/MS measurements might be different
than in 1984 if the data were from another year, analyzed by
another laboratory, or for samples with different concentration
levels than those from which the relationship was developed. For
example, for the 1984 data, statistically significant differences
between batches were found for four compounds: p,p'-DDE, beta-BHC,
oxychlordane, and trans-nonachlor. If differences in the sample
processing among batches within a year can be significant, there
may also be differences in the sample processing between years
which would affect the ratio of the HRGC/MS to GC/ECD measure-
ments .
The comparability results for p,p'-DDE, beta-BHC, and hexa-
chlorobenzene suggest that in some circumstances the recovery may
not be constant and independent of concentration. In this situa-
tion, the ratio correction factor will depend on concentration.
Extrapolation to other concentration levels may provide quite
inaccurate results when the concentrations substantially change
from those observed in 1984.
113
-------
An alternate procedure for comparing the HRGC/MS and GC/ECD
data is to correct all measurements for recovery. This would
require good estimates of recovery. Measurements on different
sample matrices provide different estimates of recovery.
Therefore, care must be taken in selecting the samples on which
the recovery estimates are based. The recovery correction can be
done on a sample by sample basis, perhaps based on surrogate
compounds, or by adjusting all concentrations using a common
recovery value. If the same recovery ratio is used to adjust all
measurements for a selected compound, an error in the recovery
ratio will result in a similar error in the average or median.
The results from this Comparability Study can be used to determine
the number of samples required to achieve a desired precision in
the estimated recoveries and the corresponding averages.
114
-------
PCB MEASUREMENTS
This chapter summarizes the PCB measurements obtained using
the HRGC/MS and GC/ECD methods. The PCB recovery measurements are
discussed first, followed by a comparison of the PCB measurements
on paired NHATS samples using the HRGC/MS and GC/ECD methods.
Unlike the primary compounds discussed in Chapter 8, the proce-
dures for reporting PCBs using the HRGC/MS and GC/ECD methods were
quite different. Therefore, although the concentration levels can
be compared, it is not possible to determine a mathematical rela-
tionship between the GC/ECD and HRGC/MS measurements.
9.1 Comparison of PCB Reporting and Measurement
Procedures
PCB reporting procedures for the two analytical methods
differed considerably. In the GC/ECD method, the limit of quan-
tification for PCBs was 1 ug/g and the PCB concentrations, on a
wet weight basis, were reported on the following interval scale:
V = Not detected;
W = Detected, with a concentration between the LOD and the
LOQ, i.e., between .33 and 1 ug/g (Equivalent to a trace
measurement);
Y = Detected, with a concentration of 1 to 3 ug/g; and
Z = Detected with a concentration > 3 ug/g.
The porcine adipose tissue samples in batches 4 through 10,
used for quality control in the GC/ECD method, had known PCB
concentrations of I ug/g. Only for these GC/ECD samples were the
PCB concentrations reported in both the interval categories listed
above and in ug/g wet weight18.
18For these samples the measured concentration, rather than the limit of
quantification (LOQ), was reported, even though the concentrations were below
the LOQ.
115
-------
Although the GC/ECD method measures PCB concentrations as if
"PCB" is one compound, PCBs comprise many different compounds (or
congeners). PCBs can be grouped into 10 classes called homologs.
Each homolog class includes the chlorinated biphenyl compounds
with the same number of chlorine atoms (from 1 to 10) . In the
HRGC/MS method, the PCB concentration within each homolog class
was reported. The average detection limit for each homolog
concentration ranged from .01 to .03 ug/g lipid weight. Thus the
HRGC/MS measurements were much more sensitive than those from the
GC/ECD method for which the detection limit was 0.33 ug/g wet
weight.
The GC/ECD concentrations for the paired samples, reported on
the interval scale, could not be converted from a wet weight basis
to a lipid basis. Therefore, for the analysis in this chapter,
PCB concentrations from both methods are expressed as micrograms
per gram wet weight. Note that this is different than for the
discussion in Chapter 7, which used concentrations on an extract-
able lipid basis.
Interpretation of differences between the HRGC/MS and GC/ECD
measurements is complicated by differences in the reporting proce-
dures and significant differences in the data reduction procedures
for the two techniques. The GC/ECD analyses of PCB's is based on
comparison of a limited number of major peaks associated with the
PCB response. The HRGC/MS analysis of PCB's is generally based on
the area sums of the peaks for all congeners at each level of
chlorination.
9.2 PCB Recovery Using the HRGC/MS and GC/ECD Methods
Recovery measures the proportion of the PCBs in the sample
which were detected by the measurement method. The PCB recovery
can be estimated from measurements on samples which were spiked
with PCBs. These samples include the porcine tissue samples
(measured using the GC/ECD method) and the dichloromethane samples
116
-------
and all samples spiked with surrogate compounds (measured with the
HRGC/MS method). Because the multisplit samples were not spiked
with PCBs, these samples cannot be used to estimate PCB recovery.
For the GC/ECD method, only the porcine samples in batches 4
through 10 were spiked with PCBs. These porcine samples had known
PCB concentrations of 1 ug/g. The PCB recovery measurements in
these seven porcine samples were, in order: 68%, 75%, 75%, 78%,
80%, 80%, and 95%. The average recovery was 78%, with a 95%
confidence interval from 71% to 85%. The PCB recovery was lower
than for most primary compounds measured in the same porcine
samples (see Table 8).
For the HRGC/MS method, the PCB recovery can be determined
from measurements on the dichloromethane spiked samples and
measurements of the surrogate compounds in all samples. One
spiked dichloromethane sample was analyzed with each of the 10
batches. Because the HRGC/MS method provides a measurement for
each of the 10 PCB homologs, the total PCB concentration was
determined by adding the measurements for the homologs. The total
PCB concentration was used to calculate recovery19. The
dichloromethane samples were spiked with equal amounts of each
homolog. This even distribution of PCBs among homologs may not
reflect the distribution of PCBs found in naturally occurring
samples. In addition, recovery of PCBs from dichloromethane may
be different than that from lipid material. The PCB recoveries
for the 10 dichloromethane samples were, in order: 57%, 61%, 62%,
65%, 69%, 70%, 74%, 75%, 79%, and 80%. The average recovery was
69%, with a 95% confidence interval from 64% to 75%. This recov-
ery is similar to or higher than that for most other compounds
spiked into the dichloromethane samples (see Table 8).
19As can be seen in Table A-2 in the appendix, the PCB recovery in the
Dichloromethane spiked samples tended to increase as the number of chlorine
atoms increased, ranging from 51% to 83% recovery.
117
-------
Another means of estimating the PCB recovery using the
HRGC/MS method uses surrogate compounds which were added to each
sample. Four stable isotope labeled surrogate compounds, repre-
senting PCB homologs with 1, 4, 8, and 10 chlorine atoms, were
spiked into all HRGC/MS lipid samples at levels of 2, 5, 8, and 10
micrograms respectively. Because PCB recovery may be different in
the presence of lipid, only measurements on adipose tissue samples
were used to estimate PCB recovery. The recovery estimates are
for the sum of the four surrogate homologs. Figure 36 shows a
histogram of the PCB recovery measurements for lipid samples. The
average recovery for the surrogate PCBs in the 57 adipose tissue
samples was 62%, with a 95% confidence interval from 59% to 66%.
The PCB recovery estimates for the HRGC/MS and GC/ECD methods
are summarized in Table 13. The average recoveries of spiked PCBs
using the HRGC/MS and the GC/ECD methods ranged of 62% to 78%,
with the GC/ECD recovery slightly higher than the HRGC/MS
recovery.
9.3 Comparison of HRGC/MS and GC/ECD Paired Measurements
In order to compare the HRGC/MS and GC/ECD measurements, the
total HRGC/MS PCB concentration across all homologs was calcu-
lated. Figure 37 shows a histogram of the HRGC/MS PCB concentra-
tions for paired samples. The corresponding GC/ECD coded concen-
trations are shown using shading.
As can be seen from Figure 37, the HRGC/MS measurements are
skewed, with many low measurements and a few high measurements.
The sample with the largest measurement had a concentration of
1.29 ug/g, almost twice the next highest concentration in a paired
sample. The GC/ECD concentration for this sample was between 1.0
and 3.0 ug/g, in agreement with the HRGC/MS measurement. The
remaining GC/ECD measurements were between .33 and 1.0 ug/g and
the corresponding HRGC/MS measurements ranged from nondetect to
0.67 ug/g, with an average of 0.28 ug/g and a standard deviation
118
-------
CO
(0
to
T)
c
-H
w
cu
-H
»H
0)
o
o
•o
c
3
o
I
o
o
0)
JJ
(0
en
o
CO
co
2
o
0
«
05
o
E
(0
M
Cn
o
+J
CO
M 0)
c
^ o
a O
E
-------
Table 13. Summary of PCB Recovery Measurements
Analytical
method
GC/ECD
HRGC/MS
Sample type
Porcine samples
Dichloromethane
samples
Surrogate
compounds
Mean recovery
78%
69%
62%
95%
Confidence
interval
71% to 85%
64% to 75%
59% to 66%
120
-------
Number of
HRGC/MS
measurements
within the
concent rat ion
range
CD
CO M O
s O
* s **
Q e o
o o> •
J3 « e _
jj (o o e
o o> M a
ffl 6 4-i ft
0)
co *-< o
2 "3 •
iH
T? W
C 4-> O
(0 C jJ
Q)
Q S CO
O <1> co
,£ CO g
4J to o e
m e m ft
to
C .p 0)
to £
0)
S
Q
O
u
^ 0) 3
e .c w
ft 4-1 (0
4-1 ft OJ
C 6
0) O CO
M • X! W
(0 rH 4J S
i
(M CO
O -P
6 §
(0 g
M
(0
o
Cn
c
-rH
•o
CO CO
0)
X! -H
4-> ft
-H 6
s m
CO
w
4-> T3
C 0)
0) M
6-H
-------
of 0.24 ug/g. Two-thirds of the HRGC/MS measurements were below
0.33 ug/g, indicating that concentrations measured using the
HRGC/MS method are generally below those from the GC/ECD method.
A cross tabulation of the GC/ECD and HRGC/MS data in paired
samples, coded using the same interval categories as the GC/ECD
data,, is shown in Table 14. The GC/ECD and HRGC/MS measurements
are in the same concentrations intervals for 15 samples. In the
remaining 30 paired samples, the GC/ECD measurements are in the
range from 0.33 to 1.0 ug/g and the HRGC/MS measurements are
lower, less than 0.33 ug/g.
The PCB measurements on the multisplit samples can also be
used to compare the HRGC/MS and GC/ECD methods. Because the
spiking solution contained no PCBs, the PCB measurements on all
five multisplit samples for each of three composites were compar-
able. For these samples, all GC/ECD measurements were within the
0.33 to 1.0 ug/g range20. Table 15 shows the number of multisplit
samples with HRGC/MS measurements either below, in the same range
as or above the GC/ECD measurement. As can be seen from the
table, the HRGC/MS measurements are either below or similar to the
GC/ECD measurements. No HRGC/MS measurements fall into a concen-
tration category greater than that for the paired GC/ECD measure-
ment .
The data do not allow a determination of the ratio of the
HRGC/MS to the GC/ECD measurements due to the interval nature of
the reported data and the small range of the reported concentra-
tions. However, the results for the paired and multisplit samples
supports the conclusion that the ratio of the GC/ECD to HRGC/MS
measurements is greater than 1.0.
20For one unspiked multisplit sample, one of two extract PCB measurements had
a coded concentration between 1.0 and 3.0 ug/g, higher than the other two
measurements on the same extract. This measurement was ignored for this
analysis. Its inclusion would not change the result that the HRGC/MS
measurements are less than or similar to the GC/ECD measurements.
122
-------
Table 14. Coded HRGC/MS versus Coded GC/ECD PCB Measurements in
Paired Samples
Coded HRGC/MS
measurements
Coded GC/ECD measurements
Not Detected .33 to 1 ug/g 1 to 3 ug/g
Not Detected or less
than 0.33
.33 to 1 ug/g
1 to 3 ug/g
30
14
123
-------
Table 15. Comparison of HRGC/MS and GC/ECD Measurements in
Multisplit Samples
Composite3
A
B
C
Number of
HRGC/MS lower
than GC/ECD
5
1
lb
HRGC/MS measurements
HRGC/MS
Similar to
GC/ECD
0
4
3
HRGC/MS
greater than
GC/ECD
0
0
0
Note. All GC/ECD samples had coded PCB measurements between 0.33
and 1.0 ug/g.
aOther than distinguishing composites from which the multisplit
samples were prepared, the composite identifier has no meaning.
In one sample, five homologs could not be measured. Because the
HRGC/MS concentration based on the remaining five homologs may be
significantly lower than the actual PCB concentration, it has been
left out of the total for this cell.
124
-------
10 ANALYSIS OF PRECISION AND COMPONENTS OF VARIANCE
Precision refers to the variability of the measurements. The
variability may be measured in terms of the variance, standard
deviation, or coefficient of variation. An analysis of precision
attempts to quantify the variability of the measurements and
factors which affect that variability. The precision of the
measurements may be a function of concentration, the compound
being measured, and/or factors associated with the sample process-
ing steps, such as the batch in which a sample is analyzed.
The model in Chapter 5 was used as a basis for the analysis.
The variance components are estimated using only the positively
quantified measurements, unless otherwise indicated. Thus samples
with trace measurements and measurements below the detection limit
are not included in the analysis of variance components. Although
the decision to use only the positively quantified measurements
and the rounding of the reported GC/ECD concentrations will have
some affect on the variance estimates, the importance of these
factors is expected to be small.
10.1 Standard Deviation Versus Mean
Experience has shown that, for concentration data which
cannot be negative, the measurement error increases with the size
of the measurement. If the data have a lognormal distribution, as
assumed in Chapter 5, the standard deviation of the data will be
linearly related to the concentration as:
O = K C (10.1)
125
-------
where :
G = the standard deviation of the measurements;
K = the proportionality constant; and
C = concentration being measured.
For measurements on a group of samples with the same expected
concentration, equation (10.1) can be approximated by:
Sj = K Xj (10.2)
where :
Sj = the standard deviation of the measurements in group j;
Xj = the average of the measurements in group j .
Taking the log of both sides of equation (10.2) gives:
ln(S) = ln(K) + ln(X) (10.3)
The variance of ln(Sj) is roughly inversely proportional to
the degrees of freedom. Therefore, the following equation can be
fit using weighted regression, with the weights equal to the
degrees of freedom:
ln(Sj) = a + yln(xj) + e. (10.4)
For each compound, the slope, y, can be tested to determine if
equation (10.3) fits the data. The assumption that the data can
be described by a lognormal distribution is consistent with the
data if the confidence interval for y includes 1.0. An assumption
that the measurements have a constant variance is consistent with
the data if the confidence interval for y includes zero.
126
-------
Data from the spiked multisplit and paired samples were used
to estimate the slope in equation (10.4). These analyses are
described in the following sections.
10.1.1 Spiked multisplit samples
Four spiked multisplit samples were prepared at each of three
spiking levels. The mean 'and standard deviation of measurements
for the three spiking levels were used to estimate the parameters
in equation (10.4) . The mean and standard deviation were calcu-
lated using the following formulas:
-
Mean = Xj = - ; — (10.5)
nj
/ X"1 —
Standard Deviation = s^ = -\ / ^(xij~xj)2 (10.6)
nrl
where:
j = the spike level, 1 = lowest spike level, 3 = highest
spike level.
nj = the number of split samples at the j*-*1 spike level, in
this case nj = 4 for all spike levels unless some
measurements are missing; and
x^j = the measured concentration for the i*-*1 sample,
i = 1 to 4, for samples with the jtn spike level.
The estimates of standard deviation, Sj, each have three
degrees of freedom and thus are not very accurate. With only
three spiking levels, there is only one degree of freedom for
calculating confidence intervals for y. Under this condition, a
precise estimate of y is obtained only if the spacing between
spiking levels is large. Equation (10.4) was fit to the spiked
127
-------
multisplit data using all primary compounds except PCBs.21 The
results are presented in Section 10.1.3.
10.1.2 Paired samples
The procedure used for the multisplit samples was modified
for the paired samples because there are no paired samples with
the same expected concentration. Instead, the residuals and
predicted values from regression were used to estimate the mean
concentrations and standard deviations for application of equation
(10.4) .
The analysis assumed that the relationship between the GC/ECD
and HRGC/MS measurements followed the simple regression model:
xMcs " Ac + BC*XECS + error (10.7)
where:
XMCS = measurement for compound c in sample s using the
HRGC/MS method;
XECS = measurement for compound c in sample s using the GC/ECD
method;
Ac = intercept for the linear relationship between measure-
ment methods for measurements on compound c; and
Bc = slope for the linear relationship between measurement
methods for measurements on compound c.
The steps in the analysis were:
(1) Fit the model in equation (10.7) to all samples with
positively quantified values for each method, determine
the residuals and predicted values;
(2) Order the predicted values from smallest to largest;
(3) Divide the predicted values into four groups from
smallest to largest;
21 p,p'-DDT was not spiked in the multisplit samples, however, the
concentrations in the unspiked samples had a great enough range to fit
equation (10.4).
128
-------
(4) For each group, calculate the average predicted HRGC/MS
concentration, Xj, and the standard deviation of the
residuals, Sj/ and
(5) Use the four values of Xj and Sj to fit equation (10.4)
and estimate 7.
Equation (10.7) was used to estimate the relationship between
the standard deviation and the concentration for HRGC/MS measure-
ments. The following similar equation was used to estimate preci-
sion for the GC/ECD measurements:
XECS = Ac + BC*XMCS + error (10.8)
Although the results of these procedures are only approxi-
mate, they provide more degrees of freedom for estimating the
standard deviations and more data values for fitting equation
(10.4) .
10.1.3 Results
Plots of ln(Sj) versus ln(Xj) were prepared for all compounds
and groups of comparable measurements derived from the multisplit
and paired samples. From the plots, the following conclusions are
evident:
• The estimates of 7 for each compound are not very
precise; and
• There is a general increase in variability with increase
in concentration.
To estimate a more stable confidence interval for 7, the
calculations used a pooled variance across all compounds. To
provide a possibly more precise estimate of 7, a pooled slope esti-
mate was also calculated under the assumption that the slope 7 was
identical for all compounds.
129
-------
c => '
CO
j;
CO
2 3-
M
2-
*—
G
0
"•"I i
4-> 1 •
m
-H
1 0-
T3
M
(0 _n _
T5 -1
C
(0
4-)
w _2 -
2
O.
0
H
CO
4
I
£
C
C
^
(
c
1
1 '
-1 p
a c
i C
i 1
v
i (
i
r
4
1
a t
i 3
3 C
<
* 1
i o
1
4
1
j i
j
3
7
0
J (
1)
Q C.
r 1
i
t
5 <
H •*
>-4
3
H <
1) 1
H t
4
(
»i
1
>
1)
3
H
X 1
D
0.
tl
,
J
1) <
C
4
t
1
\
1
£ 1
H Q
0 V-i
J O
^ iH
o ^
- ^
5 S
>i 0)
« K
I
.
1
i i
>-i T3
0 CU
rH C
X! -H
0) O JD
C "J E
5 C 0
Sou
C 2
g CO 3
C
-------
C
(X5
W
Q)
o
•H
-p
(0
-H
a)
T!
•o
S-l
(0
T3
(0
JJ
(0
0)
a
o
•-I
CO
3-
2 -
-
0-
1 -
2 -
3-
4 -
•
i
4
E-
C
C
C
c
1
1
1
1
<
< t
) C
) C
l, C
^
l* C
1 '
1
B
i t
i i
] CJ 0) Q
ac -o c
OQ -H m
X T
(0 0 »•
i, -u a c
j; « w |
a 5
Q) C
i
r
i
4
) |
' 1 i
i r^
> s
s
> X
? ^
i K
' ,
P
•
S-
c
1^
X
Benzene
ns— Mnnar-
1 ,
1
i
< T
> a
< c
:
) £
5 E
: c
> c
r—
) i-
: <
i'
)
I
i
i
(
i
i
K
Pesticide residue
H
Paired samples
Multisplit spiked
<• samples, with 95%
confidence interval
Figure 39.
Slope of the linear relationship between the log of the
standard deviation and the log of the mean for HRGC/MS
measurements on multisplit spiked and paired samples.
131
-------
Figures 38 and 39 show confidence intervals for y based on the
multisplit spiked samples and point estimates of y based on the
paired samples. Because the measurements within each group of
paired samples are not true replicates, confidence intervals for
the slopes would be approximate and are therefore not reported.
Results for each primary compound and the pooled estimate for all
compounds are shown. Figure 38 shows estimates for GC/ECD
measurements. Figure 39 has estimates for HRGC/MS measurements.
On both figures, a dotted line marks a slope of 1.0 corresponding
to a lognormal distribution.
As can be seen from Figure 38 and 39, the results do not
contradict the assumption that the data have a lognormal distribu-
tion. Most confidence intervals include a slope of 1.0, consis-
tent with the lognormal distribution, and fewer confidence inter-
vals include a slope of zero, consistent with a normal distribu-
tion. The confidence interval for the pooled estimate is consis-
tent with the assumption that the errors have a lognormal distri-
bution. For subsequent analyses, the error components are assumed
to have a lognormal distribution, or equivalently, the log trans-
formed data are assumed to have a normal distribution.
10.2 Components of Variance
A components of variance analysis divides the variance of the
measurements into components which can be associated with specific
sample processing steps. The components add together. Thus, the
variance of the measurement error is the sum of the variance asso-
ciated with differences between batches and the variance associ-
ated with differences between measurements on samples within
batches. The calculation of variance components uses the log
transformed measurements from the QC samples, the spiked multi-
split samples, and the surrogate compounds.
132
-------
The model for the data, equation (5.10), includes components
of variance associated with:
• Differences between batches, a^-,;
• Differences between batches which depend on the internal
standard used to quantitate the measurement, 0^,^ (this
term is confounded with O^., in the GC/ECD data) ;
Differences between batches which depend on the internal
standard and compound being quant itated, CT^lc (this term
only applies to the HRGC/MS data) ;
Differences between samples within batches which are the
same for both the HRGC/MS and GC/ECD measurements, o|;
Differences between samples within batches which are
different for the HRGC/MS and GC/ECD measurements,
Differences between measurements quantitated using the
same internal standard within a sample, O^g^ (this term
is confounded with O in the GC/ECD data) ; and
• Unexplained measurement error, cr^.
Not all of these components can be estimated using every
subset of data. For most subsets of the data, the following two
variance components can be estimated, 1) a between-batch component
(°mb + ambi (+ °mbic for the HRGC/MS data)) and 2) a within-batch
component (o| + (J^s + ^msi + °m) • ^ne second of these components
is the sum of the individual components associated with samples,
internal standards within samples, and unexplained measurement
error.
There is no one optimal way to estimate the components of
variance. Calculation procedures based on different assumptions
about which estimation criteria to optimize give different
results. Estimation of the variance components was performed on
the computer using SAS PROC VARCOMP (SAS 1985) which provides four
calculation procedures. In order to show the range of estimates
which might be achieved using different estimation procedures,
133
-------
Tables 16 through 19 present results using three procedures
labeled MIVQUE, Type I, and REML22. Discussions of the results are
based on the restricted maximum likelihood (REML) procedure,
because this method provides information on the precision of the
variance component estimates which are useful for interpretation.
The discussion notes if the interpretation might change based on
results from the MIVQUE or Type I methods.
All of the variance components are estimated under the
assumption that the magnitude of the components are the same for
all compounds. To the extent that this is not true, the variance
components represent the average component across compounds. The
summary tables in Appendix A provide information on each compound
separately. In general, compounds within either Fraction 1 or
Fraction 2 appear to have similar variability. However Fraction 1
compounds appear to have less variability than Fraction 2 com-
pounds. Therefore, separate results are presented for Fraction 1
and Fraction 2 compounds.
10.2.1 Components of variance for the HRGC/MS
measurements
Eleven surrogate standards were injected into all but 3 of
the 80 samples analyzed using the HRGC/MS method. Eight of the 11
surrogate compounds were measured in Fraction I and three in
Fraction 2. Because Fraction 2 compounds were analyzed only in
three batches, less data are available for estimation of variance
components for Fraction 2 compounds. The spiked dichloromethane
and multisplit samples also provide data for estimation of the
overall measurement variance.
22The unrestricted maximum likelihood procedure was not selected due to the
longer computation times required and similarity to REML.
134
-------
Variance components for HRGC/MS Surrogate
compounds in Fraction 1
The model for determination of the components of variance for
Fraction 1 compounds is:
ln(XMcbsi> = ln% + ^i* c + **> + s1 + + £Mcbsi
where 8. = 6 + 6
S .
This model has components associated with batch, batch and
internal standard, batch, internal standard, and compound, samples
within batches, internal standards within samples, and unexplained
/ Sc \
error. The fixed effects for different compounds, Inrr - , are
VRMcty
removed before estimating the variance components.
The model assumes that (1) the measured concentrations in the
blank samples, dichloromethane spike samples, and composite
adipose tissue samples differ due to different recoveries in each
sample matrix and (2) after correcting for differences in recov-
ery, the variance components are the same for all subsets of
measurements. The differences among the sample matrices are
statistically significant (based on the output from the type I
components of variance analysis, p <.05) . The measured concentra-
tions in the blank samples are lower than in the other two
matrices .
The estimated components of variance are shown in Table 16.
All components except the batch component are significantly
different from zero. However, the component for batch effects
which differ by internal standard is significant. Thus there are
significant differences between batches. The differences depend
on the internal standard used to quantitate the measurements.
Visual inspection of the residuals indicates that the unexplained
variance is greater for some compounds than others. Therefore,
135
-------
CO
JJ
c
4-1
0)
.H
43
(0
TJ
j CO
S-H g
06 r-H 0)
*H i^i*
°S^
4J °
CO) (0
/iv O Q)
0 C
(0
0) M
M cj
3
T3 dP
0 in
0 CM
0 -H
ft i-3
2
C U
O p£
4J
(0
C |*
-H
4J Q)
CO ft
Q> >i
H
Q)
O
c M
-H D
fO >
s
•o
^^
1^
(0
1
Component to be esti
dP
CM
00
rH
0
HH
,_!
iH
O
•
vo
o
o
•
CM
O
O
%*
Batch
dP
m
rH
oo
rH
O
+|
*•"''
'a1
rH
0
m
i-H
0
•
in
0
o
-H
Batch and Internal
standard
dP
m
^*
o
0
4-|
*— • '
m
0
o
•
m
o
o
*
00
f— 4
o
o
•H
•O
e
Batch, Internal
standard and compoui
dP
00
iH
00
O
O
JH
CM
iH
O
00
O
0
•
rH
rH
O
CM CO
O
+
*$
Sample within batch
dp
CM
00
O
0
•t-1
^•*
CM
CM
O
*g«
CM
O
•
-------
these variance estimates represent an average across all com-
pounds .
For Fraction 1 compounds, the estimate of within-batch vari-
ance is .062.23 The estimate of between-batch variance is .030.24
Thus variation within batches (due to measurement error and
differences between samples) contributes more to overall measure-
ment error than differences between batches. Adding the compo-
nents together, the overall measurement error variance for the
surrogate compounds in Fraction 1 is .092. The corresponding
coefficient of variation of an untransformed measurement is 31%.
The residuals from the analysis of variance had several low
observations which might be designated outliers. To determine the
effect of the more extreme observations on the variance compo-
nents, the cases with the lowest six residuals were removed and
the components were recalculated. With the extreme outliers
removed there were no changes in the conclusions and the overall
measurement error variance estimate drops slightly to .083 with a
corresponding coefficient of variation for the untransformed
measurements of 29%. The results based on all measurements were
used in subsequent summaries.
Variance components for HRGC/MS Surrogate com-
pounds in Fraction 2
The model for determination of the components of variance is:
ln(XMcbsi) = l« + 5Mb' + We + SMS!' + ^Mcbsi (10.10)
23This is the sum of the REML variance estimates for the three within-batch
variance components, .012, 022, and .028.
24This is the sum of the REML variance estimates for the two between-batch
variance components, .011, .014, and 005.
137
-------
where
= 8S + ^MS + $Msi ' ttle variance associated with both
sample differences and differences in the within-sample
internal standard responses . This term is assumed to
have a variance of o| + o^s + GMSJ_; and
= $Mb + 6^1 , the variance associated with batch differ-
ences which are common to all measurements in a batch
and which are associated with internal standards . This
term is assumed to have a variance of O +
In this model, the sample and internal standard effects are
combined because all surrogate compounds in Fraction 2 were
quant itated on the same internal standard. The model assumes that
the average concentration in the blank samples/ dichloromethane
spike samples, and adipose tissue samples differ due to different
recoveries in the each sample matrix.
This model has components associated with batch, batch and
compound, samples within batches, and unexplained error. The
/ Sc \
fixed effects for different compounds, In _ , are removed before
estimating the variance components. The estimated variance compo-
nents are shown in Table 17 .
The component associated with differences between samples and
internal standards within samples is significantly greater than
zero. The variance components for sample and unexplained error in
Fraction 2 compounds are much larger than for Fraction 1 com-
pounds. For Fraction 2 compounds, the overall estimate of within-
batch variance is .749. The overall estimate of between-batch
variance is .006. Adding the components together, the overall
measurement error variance for the surrogate compounds in Fraction
2 is .754. The corresponding coefficient of variation of one un-
transformed measurement is 106%. Differences between measurements
in different batches is due mostly to differences between the
samples within batches and not differences between the batches.
138
-------
CO
4J
C
CO
CO
S-l
CO
cO
CO CO
S T3
•a 3
0) O
E Pi
o o
IW U
CO
C CO
M CO
EH CJi
O
CO
M
O CM
4-1
C
CO O
4-> -H
C 4J
CO O
C CO
O H
acu
o co
o s
\
0) O
o a
c os
m x
-H
M M
(0 O
> in
r-
rH
CO
CO
E-i
T3
S1 C
S-H s
iH
0 cox!
4J °
C CO CO
g 0 CO
ujj>
Ou
a) M
n o
D
T3 <#>
co m
O
0 +1
ci I-H
2
c w
O OH
~H
4J
CO
£ M
-H
4J CO
CO ft
EH
0
§ g
•r4 *~"^
M ^"
> s
T3
4J
c
0)
c
o
a
s
o
u
c«P
o^
vo
rH
•
-1*1
^•^
W3
0
o
•
r-
o
^D
o
-rH
if
+
^
r*
u
4J
CO
CQ
C
0)
a>
s
1 1
CO
CQ
0
o
oo
o
1
CM
00
0
1
o
.^
•rt
°l
T)
c
o
a
e
o
u
•o
c
CO
X!
0
CO
CQ
dp
r-
00
vo
00
•
-f-]
^^
0
00
^r
•
00
CM
00
CM
-H
m
+
m
f
CMW
H
CO
c
^1
CO
4J
C
-H
•o
C
CO TJ
i_l
CO CO
rHTJ
ac
6 cO
CO 4-)
CO CO
6P
CM
Co
00
rH
•
-f-|
* — '
c
-H
cO
rH
a
X
CO
c
D
e*P
0
O
rH
^
m
r-
00
r-
co
r-
rH
4->
O
EH
4J
C
0)
e
^
jj
CO
cO
co
£
i— 1
rH
(0 M
M O
CO M
> H
O 4>
139
-------
Variance components for HR6C/MS spiked
dichloromethane samples
One spiked dichloromethane sample was analyzed in each batch
using the HRGC/MS method. Because there is only one sample per
batch, the within-batch and between-batch components for the
dichloromethane samples cannot be estimated independently using
the dichloromethane data. The model for the data is:
ln(XMcbsi) = Inr-- + 5^+8 + 8^+^. + eMcbsi- (10.11)
where:
+ ^Ms"1" ^s a random effect associated with batch b
and sample s, assumed to have a variance of o^b + OMS+
o ; and
+ $tfoic + $Msi a random effect associated with
the internal standard used to quantitate the measure-
ments, assumed to have a variance of O^i + OMtdc+ °Msi-
£Mcbsi' = ^Mbic + eMcbsi a random effect associated with the
batch and compound and unexplained error, assumed to
have a variance of
The variances for Fraction 1 and Fraction 2 compounds differ
for the QC compounds. On the assumption that they may differ for
the dichloromethane samples also, Fraction 1 compounds are ana-
lyzed separately. This model has components associated with batch
and sample, internal standards, and unexplained error. The fixed
f sc \
effects for different compounds, In _. , were removed before
estimating the variance components. The estimated components of
variance are shown in Table 18.
140
-------
C
o
-H
-U
o
(0
S-l
o
CO
4J CO
C CD
CD H
e g.
CD S
H (0
3 W
to
(0 CD
CD C
T3 -P
CD CD
CO H
SS
H-H
EH Q
CnT3
O CD
o
4-1
CO
CO
-
C CJ
CD O
O
CO
CD 73
O C
C 3
(0 O
-a §•
u
CD
rH
£t
(0
73
I-H <"
|-5e
°tx!
^ CD C
g 0 CD
" (0 >t
Pn J"j
CD M
M O
p
73 dp
CD m
0 en
0 -H
M
o* d
"
C W
O OS
-H
(0
£ r"'
-H
4J CD
co O<
CD
o
.5 "
d o
m ^
> 2
•o
0)
(0
e
"jj
CO
CD
CD
X>
O
jj
C
CD
C
O
a
e
o
o
dP
CN
rH
00
00
0
•
HH
rH
rH
O
•
CN
O
O
rH
rH
O
%•
00
D
"1
CD
a
6
1C
CO
•o
C
(0
X!
0
(0
OQ
dP
VD
rH
^
0
•
il
rH
V^
O
•
r-
"SJ*
O
^
CN
O
-H
W
°
+
-iH
•o
(0
•o
C
(0
4J
CO
1 — I
(0
C
U
0)
4->
C
M
dP
CN
ro
O
0
•
.f-l
*-'
00
CN
O
•
ro
CN
o
o
^"
o
csS
+
0
-H
if
J^
o
^1
yi
CD
73
CD
C
-H
<0
i-H
a
X
CD
C
D
dp
O
O
rH
in
CTi
O
•
CN
r~
o
CN
r-
o
H
(0
O
EH
•P
C!
CD
e
CD
3
CO
(0
CD
£
rH
rH
(0 M
M O
CD V-l
^ i-1
O CD
141
-------
The overall measurement error variance for Fraction 1 com-
pounds in the dichloromethane spiked samples is .095. This corre-
sponds to a coefficient of variation for an untransformed measure-
ment of 32%. The component associated with internal standards is
significant, indicating that compounds within a sample quantitated
using the same internal standard will have correlated measurement
errors.
The recovery measurement for p,p'-DDE in batch 2 is 155%.
This measurement is unusually large compared to the recovery
measurements for p,p'-DDE in dichloromethane samples in other
batches. Because removing this measurement from the calculations
reduces the overall error variance by less than 2%, the results
for all data including this unusual value are reported.
Because Fraction 2 compounds were analyzed in only three
batches, relatively little data was available to estimate variance
components. Therefore, only the overall measurement error vari-
ance for the log transformed measurements is estimated for the
Fraction 2 compounds. As with the surrogate compounds, the vari-
ance for Fraction 2 compounds is greater than for Fraction 1
compounds. The overall measurement error variance for Fraction 2
compounds in the dichloromethane spiked samples is .396. This
corresponds to a coefficient of variation for an untransformed
measurement of 70%.
Variance components for Fraction 1 compounds in
the spiked multisplit samples.
Due to the sample design for the multisplit samples (three
composites, each with four splits analyzed in three batches),
estimated variance components would have few degrees of freedom
and not be very precise. Therefore, only the overall measurement
variance was calculated for the multisplit samples. The average
measurement error variance for the log transformed measurements on
primary Fraction 1 compounds in the spiked multisplit composite
142
-------
samples is .176. The corresponding coefficient of variation for
the untransformed concentration is 44%.
10.2.2 Components of variance for the GC/ECD
measurements
Aldrin was injected into all but 3 of the 86 samples analyzed
using the GC/ECD method. The recovery for the aldrin can be used
to estimate overall batch and sample effects. Porcine samples and
spiked multisplit samples analyzed in each batch can be used to
estimate the overall measurement variance.
Variance components for GC/ECD aldrin
measurements
Only one compound/ aldrin, was spiked into the GC/ECD samples
to estimate recovery. Because there is only one compound per
sample, only the within- and between-batch components of variance
can be estimated. Therefore, the model is:
ln(XEcbs) = ^ + b' + £Ecbs (10.12)
where S^1 is the between-batch component and EEcbs is the within-
batch component .
f sc "N
The fixed effects for different compounds, In p y , were
removed before estimating the variance components. The estimated
components of variance are shown in Table 19. The model fit to
the aldrin measurements assumes that the average concentration in
the blank samples, porcine fat samples, and adipose tissue samples
differs even though the spiking levels were the same. Although
the differences between these groups of samples are not statisti-
cally significant (p <.058), the differences were modeled because
they were close to significant and differences were found in the
HRGC/MS measurements.
143
-------
C
•H
TJ
0)
H
O
KH
co
C
(0
M CO
EH CD
O 6
t-3 (0
M
00
MH O
U
4JU
CO
0)
C C
O--H
ft
6 W
O 4J
U C
0)
0) S
O 0)
C M
fO 3
•H W
H (0
(0 0>
cy>
0>
rH
(0
EH
•o
rH- ®
W-H 6
m ft-p
0 $je
4-> °
S o a>
0 m >,
P^
Q> H
M U
TS <*=
O cn
0 +|
SH
^ trt
C U
O PS
1 \
-H
(0 ft
0) >t
EH
fl)
O
s g
-H X,
W ^
> S
CD
(0
E
4J
co
0>
d>
X)
O
4J
4J
C
Q)
C
O
ft
e
o
U
CN
VO
O
O
jj
il
ro
0
o
en
o
o
•
m
o
0
-H
-^-
•«
O
-p
(0
^fj
C
CO
0)
s
o
CO
00
00
00
o
o
, j
i!.
CN
CN
O
•
CN
CM
O
•
CN
CM
O
-rt
n
D
«
n
csE
M+«
o
_£-
0
4->
(0
42
C
•H
4J
-H
S
<#>
O
O
f^l
CM
O
•
m
CM
o
•
in
CM
o
rH
(0
4J
O
EH
144
-------
For the aldrin data, the between-batch variance component,
.003, is not significantly greater than zero. The estimate of
within-batch variance is .022. Adding the components together,
the overall measurement error variance for aldrin is .025. The
corresponding coefficient of variation of an untransformed mea-
surement is 16%.
Measurement error variance for the GC/ECD porcine
fat samples
The model for the QC samples is simpler than that for the
HRGC/MS measurements because only one internal standard was used.
As with the dichloromethane samples, with only one sample per
batch, the batch and sample components of variance cannot be esti-
mated independently. Therefore, only the overall measurement
error variance, pooled across spiked compounds, was estimated from
the porcine sample data. Because the porcine fat samples had
different sources and spiking levels for the first 3 batches than
the last 7 batches, the error estimates were made separately for
batches 1 to 3 and 4 through 10.
For the porcine fat samples from batches 1, 2, and 3, the
overall measurement error variance for the log transformed data is
.009. The corresponding coefficient of variation for an untrans-
formed measurement is 10%.
For the porcine fat samples from batches 4 through 10, the
overall measurement error variance for the log transformed data is
.008. The corresponding coefficient of variation for the untrans-
formed concentration is 9%.
Variance components for GC/ECD measurements on
Fraction 1 compounds in the spiked multisplit
samples
Because only one of the Fraction 2 compounds (dieldrin) had
positively quantified measurements, variance estimates from the
145
-------
GC/ECD spiked multisplit samples were calculated for Fraction 1
compounds only. On the assumption that the variances are similar
for all primary Fraction 1 compounds in the spiked multisplit
composite samples, the overall measurement error variance for the
log transformed data is .013. The corresponding coefficient of
variation for the untransformed concentration is 11%.
10.3 Summary and Comparison of BRGC/MS and GC/ECD
Precision
Based on the analysis of standard deviation versus concentra-
tion and plots of the data, the variability of the HRGC/MS and
GC/ECD measurements increases as the concentration being measured
increases. The decision to model the log concentration and assume
that the log transformed concentration had constant variance was
based on the relationship between the standard deviation and mean
concentration from spiked multisplit and paired samples, theoreti-
cal considerations, and observations on the residuals from analy-
ses using the log transformed data. Modeling of the log trans-
formed data results in a statistical model which is relatively
easy to fit to the data using standard statistical techniques.
Where possible, the overall measurement error variance was
divided into components of variance associated with sample pro-
cessing steps. Of particular interest is the within- and between-
batch estimates of variance. The within-batch component of vari-
ance is the variance of measurements on split samples within the
same batch. This includes errors associated with sample handling
within a batch or day, injection of internal standards, and
quantitation. The between-batch component of variance is the
portion of the measurement variance which is attributed to differ-
ences between batches. This includes errors associated with
calibration and possible changes in the preparation of different
solutions or equipment setup between batches or days. The
between-batch components of variance and overall measurement error
variance for the transformed data are summarized in Table 20. The
146
-------
Table 20. Summary of Variance Components for GC/ECD and HRGC/MS
Measurements
Analysis method and
subset of the data
HRGC/MS
HRGC/MS
HRGC/MS
HRGC/MS
HRGC/MS
GC/ECD
GC/ECD
GC/ECD
GC/ECD
Surrogate
compounds
Fraction 1
Surrogate
compounds
Fraction 2
Spiked
di chlor omet hane
samples
Fraction 1
Spiked
dichloromethane
samples
Fraction 2
Spiked multi-
split samples
Fraction 1
Aldrin
Porcine fat
samples Batches
1 to 3
Fraction 1
Porcine fat
samples Batches
4 to 10
Fraction 1
Spiked multi-
split samples
Fraction 1
Log transformed
measurements
N Between Overall
batch error
variance variance
620 .030
66 .006
428
27
82 Assumed
zero
73 .003
27
62
84 Assumed
zero
.092
.754
.095
.396
.176
.025
.009
.008
.013
Untransformed
measurements
Coef. of 95%
variation Pred.
Interval
31%
106%
32%
70%
44%
16%
10%
9%
11%
±61%
±208%
±62%
±137%
±86%
±31%
±19%
±17%
±22%
N is the number of measurements from which the variance components
are estimated.
The within-batch variance component can be calculated from the
difference between the between batch and overall variance,
The 95% prediction interval for an individual measurement is
calculated as 1.96 times the coefficient of variation.
147
-------
overall measurement variance is also expressed as a coefficient of
variation for an untransformed measurement.
The components of variance were estimated from different
subsets of the data according to how the samples were prepared.
The components based on different subsets may differ due to many
factors including 1) different compounds analyzed in the different
subsets, 2) different matrices in which the compounds are found,
and 3) different analytical procedures used to process the
samples.
The variances were calculated assuming that all compounds in
the subset of data being analyzed have the same variance. Review
of the data suggests that this assumption is not unreasonable but
that some differences may exist. Therefore, the components of
variance and overall variance represent averages across compounds.
Additional information about individual compounds can be obtained
from the tables in Appendix A.
As can be seen in Table 20, measurements for Fraction 2
compounds using the HRGC/MS procedures have greater variances than
those for Fraction 1 compounds. The three variance estimates for
Fraction 1 compounds are .092, .095, and .176. The estimates for
Fraction 2 compounds are .396 and .754.
Both the Fraction 1 and Fraction 2 compounds measured using
the HRGC/MS procedures have overall variance estimates greater
than those for the GC/ECD measurements. The overall variance
estimates for the GC/ECD compounds are all less than .03. The
HRGC/MS overall variance estimates range from .092 to .176 for
Fraction 1 compounds. As a rough rule of thumb, the variance of
the HRGC/MS measurements is nine times greater than that of the
GC/ECD measurements, resulting in a coefficient of variation three
times greater for the HRGC/MS measurements than the GC/ECD mea-
surements. Based on a weighted average of the variance estimates
148
-------
in Table 2025 approximate 95% prediction intervals for the
Fraction 1 GC/ECD and HRGC/MS measurements are 22% and 63% respec-
tively.
Systematic differences between batches are small relative to
the differences between measurements on samples in the same batch.
However, batch differences for the HRGC/MS measurements on the
surrogate compounds are statistically significant. Therefore, the
calculation of recovery and determination of the relationship
between the GC/ECD and HRGC/MS measurements (Chapters 6 and 7)
accounted for the presence of any batch effects.
The proposed model for the data hypothesizes sample and
internal standard effects, in addition to batch effects. The
statistical analysis indicates that both the sample and internal
standard effects are significant for the HRGC/MS measurements.
Although these effects can be ignored for this Comparability Study
and many uses of the data, comparison of measurements for differ-
ent compounds in the same sample and different samples in the same
batch must take these variance components into account.
25The average was calculated using the number of measurements, N, for the
weights.
149
-------
11 REFERENCES
Fuller W. 1987. Measurement Error Models. New York: Wiley, pp.
1-48.
Mack G. 1986. Battelle Columbus Division. Statistical Design and
Analysis For a Study to Determine Whether Changing
Analytical Methods Has a Significant Effect on Estimates of
Baseline Levels and Time Trends. Washington, DC: Office of
Pesticides and Toxic .Substances, U.S. Environmental
Protection Agency. Document No. NHATS-SS-03. Contract No.
68-02-4243.
Mack G., Panebianco D. 1986. Battelle Columbus Division.
Statistical Analysis of the FY82 NHATS Broad Scan Analysis
Data, Draft Final Report, August 1986. Washington, DC.
Office of Toxic Substances, Exposure Evaluation Division,
U.S. Environmental Protection Agency. Document No. NHATS-SS-
03. Contract No. 68-02-4243.
Mack G., Mohadjer L. 1985. Baseline Estimates and Time Trends for
Beta-benzene hexachloride, Hexachlorobenzene, and
Polychlorinated Biphenyls in Human Adipose Tissue 1970-1983.
Washington, DC: Exposure Evaluation Division, U.S.
Environmental Protection Agency. Document No. NHATS-SS-01.
EPA Pub. 560/5-85-025.
Remmers J. 1987. Letter from J. Rammers to J. Tessari, October 1,
1987. Washington, DC: Office of Pesticides and Toxic
Substances, U.S. Environmental Protection Agency.
Robinson, P. E., Mack, G. A., Remmers, J., Levy, R., Mohadjer, L.,
1990, "Trends of PCB, Hexachlorobenzene, and j3-Benzene
Hexachloride Levels in the Adipose Tissue of the U.S.
Population," Environmental Research 53, 175-192, Academic
Press, Inc.
SAS. 1985. SAS User's Guide: Statistics, Version 5. Gary, NC: SAS
Institute, Inc.
Sherma J., Beroza M. 1980. Analysis of Pesticides Residues in
Human and Environmental Samples: A Compilation of Methods
Selected for use in Pesticide Monitoring Programs.
Washington, DC: U.S. Environmental Protection Agency. EPA-
600/8-80-038, Section 5, A, (a), pp. 11-19 (1980).
Stanley J. 1986. MRI. Preparation of the FY 1984 NHATS Composite
Samples for the Method Comparability Study, Final Report.
MRI Report, June 17, 1989. Washington, DC: Office of
Pesticides and Toxic Substances, U.S. Environmental
Protection Agency. MRI Report, June 17, 1989. Contract No.
68-02-3938.
151
-------
Stanley J. 1985. Analytical Method for the Determination of
Semivolitile Organic Compounds in Human Adipose Tissue,
Draft Interim Report vl, Revision 3, December 20, 1985.
Washington, DC: U.S. Environmental Protection Agency.
Contract No. 68-02-3938, Work Assignment 8.
USEPA. 1986. U.S. Environmental Protection Agency. Comparability
Study of Analytical Methodology for TSCA Chemicals in Human
Adipose Tissue, Quality Assurance Program Plan. Washington,
DC: Office of Pesticides and Toxic Substances, U.S.
Environmental Protection .Agency.
USEPA. 1988. U.S. Environmental Protection Agency. Computer
printout from NCC: Historical averages for NHATS samples.
Oct. 13, 1988. Washington, DC: Office of Pesticides and
Toxic Substances, U.S. Environmental Protection Agency.
USEPA. 1990. U.S. Environmental Protection Agency. Computer
printout from NCC: Historical averages for NHATS samples
without hospital 313. Mar. 26, 1991. Washington, DC:
Office of Pesticides and Toxic Substances, U.S.
Environmental Protection Agency.
152
-------
APPENDIX A: SUMMARY DATA TABLES
153
-------
Tables A-l through A-18 summarize the measurements used in the
Comparability Study. The following guide is provided to help
locate the appropriate table:
Tables A-l to A-7 summarize the HRGC/MS measurements:
A-l Method blanks; page 156
A-2 Spiked dichloromethane samples; page 158
A-3 Paired composite samples; page 160
A-4 to A-6 Spiked multisplit samples; and page 162
A-7 Surrogate compounds in all samples. page 168
Tables A-8 to A-18 summarize the GC/ECD measurements:
A-8 Method blanks; page 169
A-9 and A-10 Porcine fat samples; page 170
A-ll Paired composite samples; page 172
A-12 Aldrin in all samples; page 173
A-13 to A-15 Spiked multisplit samples; and page 174
A-16 to A-18 Extracts of unspiked multisplit
samples. page 177
155
-------
.
^
3
C
-rl '
£
O
.,-{
(0
"
4->
0)
o
c
o
o
i-H
10
C
O
z
00
c
10
rH
pa
•o
0
x;
0)
2
c
o
n
c
i:
6
M
3
n
A
CO
8
as
X
M-l
O
^i
M
2
i
3
co
rH
|
V
1— 1
XI
10
EH
V.
4-
C
c
>1 ^
H ^
l5 "i
o-c
* 4
IH 'H
0 J
C
1C
£
c
i-H
,— 1
°
g
(H
IU
s
^J
01
o
M
V
at
£H
•Q ft
,*><*><*><*><*>,*',*>,#>*>
ooooooooooo
CMCMrHCNCNCMCSICMCXJCM
OC3OOOOOOOO
OOrHOOOOOOO
^^
Z
(^
f^
M
U O 0
Q O Q Q Cl Q I CD 1 1
i i i i i i 10 i
0
CM
o
o
M
a:
tJM
s
z§
ft z
*o cv
O 0
*> *>
o o
CN CM
O 0
0 0
a
Q
X
0
CU
cc ct
ss
X X
£$
r^ r^
Oi Q<
Mrvi
124
zzi
2^2
_,_._.
000
*><*»*>
o o o
CM CM H (0 g
M U IB
l"l ^1 *V
<*>
O O
CM CM
0 0
0 0
u
Z
U
U Z
Z H
&Q GQ
N O
55 CM
U O
CQ ^5
O 33
a: o
S|
u u
M H
CU I
m ro
CO CM
rH rH
Q Q
Z Z
CM CM
O O
,*> <*>
O O
CM CM
rH rH
O 0
O O
u a
z z
Cd Cd
N C-3
z z
la u
m m
o o
a: cc
S3
11
U U
EH EH
CO «3<
CM CM
rH rH
Q Q
Z Z
CM CM
o o
<*> <*>
0 0
<*> <*> <*> *> *.
0 0 0 0 O 0
CM CM CM CM CM CM
o o o o o o
o o o o o o
U U (d Cd
Cd J Z Z Z
z x a a a
U X X CC X
E-. < < a: < 2
X Z Z O Z O
•4! CJ O iJ X i-3
156
-------
TJ
3
fJ
-H
C
^^
•
rH
1
CD
rH
10
EH
M
4-
C
a
>i E
o i;
ffl
i ^
"_j
•4-J
O
(11
vv
A
M4
^1 1*
*Q Ml
*
O 4J **"]
C T
M QJ
Q> 6 2
•Q ® fi
3 3 rt
Q) ,S
s
.
X
1C
s
,
C
•H
•c
[ !
CO
c
(0
d
2
C
(0
^~j
•e
CD
s
c
10
ffl
a
s
•o
0)
4J
o
0>
0)
Q
g
.
?[
"
0
ts
10
s
rH
10
0
c
0)
pj
0
a o Q Q
2222
CM VO CM CM
O 0 0 0
<*>
O O O O
CM CM CM CM
O 0 0 0
O 0 0 0
to
T
h- 1
25 |J3
W JH
sc is
M X
03 Da
K O O
22 CO O H^
W ^* O 3C
a; a: 2 o
JH X O W
O4 O S Q
O Q Q
222
CM CM CM
O O O
O O O
CM CM CM
O O O
O O O
CO CO
CO r*H ^H
rH U |V|
J2 x 35
CP CU 04
Oa CO tt>
H O O
O O O
a: J j
X tf rtl
rj pr c^
M EH 2
(Ti f*l M
tH £-< 04
Q O
2 2
CM CM
O O
O O
CM CM
O O
O O
CO
CO r3
>H 2
2 W
Ed X
X pi
M OQ
CO O
II
O rtl
(4 &4
X &
EC EQ
X X
CM
cn
rH
rH
CO
O
rH
vo
0
CM
CM
rH
•
CM
cn
o
•
CM
QQQQQQQcn
2222222O
CM VD VD CM CO CM CM
o o o o o o o
dP
oooooooo
r 1
^^^(OCOCOCOO
oooooooo
oooooooco
H
CO CO EH
2j 2 *5 *C EH
rH
O
•
H A
« W 0
H? CN} 1
v^ ^^ c
EH W 1
O 'H rH
CQ ffl Q
Q Q Q
222
CM CO CM
O O O
O O O
ro co co
o o o
o o o
ta
Ed EH
EH i^
< X
X O4
p, CO
CO O
O X
Cu
HT *^j
M >*
rtl 3C O
3C E^ tf
fit U QJ
WOO
X O O
>H 1 H
^ %^ *-^
CQ CO CO
M M M
Oi 0£ Ct4
EH EH EH
Q Q Q
222
CD CM ro
o o o
o o o
CO CO CO
000
o o o
u
H
X
Oi
O EH Ed
Ol X I&4
O* X
ij CO pi
>H O CO
EH O< X
O 2 rH
D X 0
M M H
f*1 f^l p*l
EH EH EH
T3
•H
a
-r|
rH
4J
C
0
tH
0)
a
•d
c
10
CO
g
10
(71
o
CM
<4H
O
4J
Si
Cn
0)
4J
10
M
4J
C
0)
o
c
o
0
H
10
C
-H
g*;
O *r
c m
CO
CD
r* CM
4J 0
^
CO
C
10
n
M
o
~
j2
s
EH
a
CU
***^
^-^
a:
M W
+ 4-»
>
M M
CD 3
n n
g • CD
•0 w -d
10 0)
n co
•H CO (0
C jQ
•a o
CD -H CO
*J -P -H
U 10
(D M Q
4J 4J O
CD C rH1
•O 0)
U C
JJ C 10
COO)
CD U g
(J
U rH 0)
0 rH .C
ft. < H
157
-------
s
o
s
c
0)
o
0)
w
00
0)
ID
CO
a
co
0)
(C
JJ
I
H
O
r-\
U
CNJ
(V
3
EH
-H
(0
O
O.
O
M
<»
^•r^inmmr^minLnmmoovovomvomrHrHi-Hi-it-ii-HrO'S'rHCNCN^'inin
oo
iH CM <-> rH
dp
O
0
000002:0000000000000000000000000
ooooorrooooooooooooooooooooooooo
OOOOO^OOOOOOOOOOOOOOOOOOOOOOOOO
OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO
OOOOO_OOOOOOOOOOOOOOOOOOOOOOOOO
s
u u
z z
GQ U
a
I
CJ O CJ
u
Q
t-t
1
H
CO
u u u
a CQ
z z
u a
CO N
z z
a a
CQ 0)
Q Q
QQQQQQ
OQQQQQ
III! 1 l
I
o
O O Q 2
^ j 2 o
»BCtOtO SXOZ
ICQI 1 2 U U
aao.a
a o 0.0
10 I 10
.C « 4->
a a a jj -H
^ «. H ID 4)
Q* O 10 Q *O
osg
< X
EH O
(o j a a x
D>< S X O
§
s
o
§§
S X
o o
sss
z u
a: a: tx:
a s
o o
H EH
u u
H EH
I I
ac N a
U Z N U
H o m a iJ 2
u os o z >H w
EH O OS U X X
gg
§g
1
rO
1 1 ^^
TrincorO
•* ^J
• EH ft,
EH < < CC
s z z o
10 M
ens
.aa
-------
c
01
* a
0) £
33
i a;
a,
*"
co
0)
o
re
I
°«rH<
M 0) ^
Q rt)
"e n
g^
re *i
•a
0)
3
•H
JJ
C
O
CM
I
re
EH
re
u
-H
Ch
CM
CM
dP
dP dP CM
rH CO rH
CM
<*> <*>
dP dP dp dP dP dP dP
CO O O p> O U1 O
CM rH i-H CO t-H t-
OOOOOOOOOOOOOOOOOOdPfTdPdP
ooooooooooooooooooo."loo
OOOOOOOOOOOOOOOOOOCOrHCOCOCMCMOrH
OOOOOOOOOOOOOOOOOOOOOOOOOO
CM O O i-H rH CO CM
a
W EH
Six
X 04
O4 CO
J CO O
3 CO |4 CO CO
aa
ZiJXiaWZiaZZZ
HXZXSUXHHia
XZ(dO4OH X
X EH
H x
X O4
X rl
X X
H EH
U O
I O
o:ix:ZCjMEHZXCk.EHZUHQSU 1
XXOMtfUUUUUOGQMZMt-lM
O.USQEHEHO.XXOZQQUQOQ
EH CO I
CO ffl Q
a J x
EH X O4
< X O
x EH a;
Oi H Oj
WOO
o n: a;
X O O
O4 iJ iJ
X X
J U O
X I H
g^s
CQ CO CO
M M M
EH EH EH
X
04
co a
O EH U
X < EH
O4 X •<
04 X
iJ CO Oi
X O CO
X X O
a. x
04
EH
u
X
g
EH U rl
D X O
OQ 04 H
MUM
a; <£ a:
EH EH EH
0
c
JJ
3
4J
C
(1)
CO
0)
M
a
re
o
XI
o
0)
JJ
re
o
-rl
Q
Z
a
I 1 Q.
C *•
a
QJ
a)
I)
ti ^i
O D
•Q oo
•d S
0)
tr w
re
z -o
cr o
M 0)
3 4->
CO 0)
re -o
0} (£
3 EH
•l-i
•o -o
re c
•d*
Jg
rH C
O
CO
M TJ
re co
CO
«o a
§*
-H m
4J -rl
re
s§
c
o
81
rH £0
rH A
< EH
159
-------
ro
l
(0
EH
»
.
X
<0
V.
c
a
a
3
0;
(C
a
£
•c
-H
-In
C
(0
3
S
.
C
jj
•
T>
4J
CO
C
10
2
c
10
-rl
•o
0
vo
1*
•
{^
rH
O
vo
ao
0
•
en
Cn
0
en
vo
0
t-
°?
*Q*
CM
0
ao
rH
CM
T
m
•
vo
^*
ao
ao
CM
1-
^
CM
O
•
CO
rH
O
CT\
O
O
•
en
rH
0
O
OO
r-4
O
CO
rH
0
OO
,_(
o
•
CO
rH
O
10
01
I'M
o
M
IS
Z w
0
CM rH OJ in CO
CM rH CM O O
m ro o oo o\
000 O O
OOOOO
OOOOO
ooooo
co cn
rH O
rH CO
rH rH
O O
CM O
O O
oo cr\
CM rH
O O
CO CT>
CM rH
O O
VD CM CO
O CM O
VD vo ro
*£> rH rH
000
O
O
VO rH (O
OOO
VO CO CO
VO Csl rH
OOO
CO rH VO CO O CTl rH
WQr-QQQQaoQQQQ-J-'fl'VDQQQQQQQQQCNQQQQQQ
02«>222202222000222222Z220222222
VrHinrHrHCsICMrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrHrH
OOCOOOrHOCMOOOOOOOOOOOOOOOOOOOOOOO
** 5" m in in
«3-CN«3.^"«3<'3<*r«T
v^'^'^'fl'^'^'^j'vro^'i'trro^r
u
Q
2
O O O
QQQQDQOQXCQCD
Q Q O Q Q Q I O I I
JC « *J g
aaaaaaa^JrH g
^<»v»»rHQ)0)(0
aoaoaom.QT5O!
u
a
u
U Cd H
z z z
u u u
N N N
U U H Z Z Z
Z 2 2 H W W
QQ GQ rO PQ 03 CQ
N N N O O O
Z Z Z OS OS 05 U
U Ed U O O O Z
~ OOXXXNM
u
U
O O Q 2
jjgos
XXOZO
O U J
DC EH EH U C
SCU On >H 10
Ed td X M (0
-------
•o
a)
3
-H
4->
C
O
U
CO
I
a>
10
EH
CO
o
i-l
o
&u
P
3 3
5 §
«5 H-I
e
c
0)
SB
0» Q)
Q
CO p-
n in
in o
1-1 «•
o in
H o
VO CM
o o
CO O
c- vo
CO O
r- 10
co o
OrHf>'3
CM o o
co oo
in vo
rH O
ro o
in
H
o
CO
CM
in
rH
rH
CO
O
coinroro«)VDcoincoc^i-HrHOrH«i»<«' rHt-HCN^rrHincjicocM
rHrHrHi-HrHCOiHrHrHrHrHCMrHCM'trrHrH CMTCMrH^rrHCOrHin
OOOOOOOOOOOOOOOOO 000000000
ooooooooooooooooooooooooooo
ta
EH
CO
co w co
hP >J OT J CO CO
U
H
EH
EH
X W
CU EH
W
w
Ed EH
sg
3C Oj
a. w
w o ta
O X EH
X (Xi <
~ .-3 CO U
•J >H O H U
CQCwMOOCDOCQCQCO
OMCQCtGCOtfOOOn
X^ X O O EH O. X
Z X >i X X » iJ
X EH
Cn
ctiJijoiJooouz ni -.,._..__.
WHXdJOOXOXXX^SzXXffl lOEHCMSEHtd
>)OXOObEHi4EHi<—' C CD CO CO CO Ou
a
z
£
JJ
-H
3
U
Q •
Q 0)
a
M 4J
O C
iw TO
(0 D"
4->
C 4J
a; o
a)
M 4J
3 a
« ja
CM <0
0)
13 M
c a
JJ
t7>
•o c
C -H
(0 -U
-H CO
*J 4J -H
U <0
<» M Q
4J 4J O
0) C iJ
•O 0)
O C
JJ
COO)
fl) o 6
c
M H CD
-------
(71
T>
3
CO
0)
Q<
m
CO
0)
•H
n
0
C*
o
0
-H
rH
CO
J
•^
JJ
lM
S
•o
0)
-H
Q.
co
rH
0)
a
c
a
^ a
'OJ J;
tj "•
s «
03 S
O »JH
o* a
"r1
O **"
FJL4 T
C
3
O
rH
rH
K
o
c
£
A
.
x
£
c
s
•
JJ
c
CO
V
*
c
CO
•o
s
s
•rt
rtl
S
**
Q
O
J
0
vo
rH
OO
i-H
CTi
C3
•
rH
^*
i-H
CO
i-H
'
CO
"** §
A
*
CO
^H
o
OJ
^*
CM
CO
CO
Q
i-H
O
in
•
rH
«*
.
i-H
CO
*^
rH
tn
r^ Q D D
. " « ^
rH
OO CO rH
-^J ^J f^
^1 ^^ I
O O O
0 0
VD °
. CO
rH
ao CN
VO CT>
rH rH
q. f-.
00 «•
t- 0
CTl "rH
VO CO
OO CM
o\ r~
V H
03 CM
'
O^ f*"
Q ^* Q O Q O
o
vo
r-
o
*
vo
§Q
^
*
oooorooooo oooooooooooo
00000 000000
c
o
dp dp dp c/P dp
OOOOOC3OOOC3C3OOOOOOOOOOOOOOOOOOOO
o««OTj'^i«<^
-------
•rt
TJ
ID
•H
4J
O
O
.
^*
1
tt)
r-l
(0
EH
^,
0)
P
•H
-H
n
o
&4
.^
o
t,
^
Q
M-I n
O 4J
c
fl? €
•Q *
3 3
2 09
(0
0)
s
IK
C
d
ri
u
5
10
a
*
TJ
a
-r
i-
vj
c
1C
r
O
r-
r~
c
a
4-1
a
c
a
cu
Jrj
d
-r-
ti |
Hj
(C
6
(C
*w
C
^
2
A
.
c
-H
.
T3
4-1
WJ
c
(C
a;
*•
c
10
-H
TJ
rtl
UJ
c
(0
-H
TJ
(11
S
a
3
TJ
0
0>
0)
Q
Q
a:
EH
O
Oi
i
CO
CO
U
•g
0)
o
CO
I-l
o
CO
i-H
o
CO
rH
O
CO
rH
O
Q Q
Z Z
CO CO
o o
CM**
ro fMi-)ODrH f~ CO
CMCMt~-OOCOCMfO W "S*
OOOOOOT* CM O
iHCOOOOOOO V CM
coi-noooin in H
OOOOOi-l O O
COCTvr-OrHOOCM in rH
CMVOWCTlOOCM^' CO IO
OOOOOOIT) CO O
COCTlrHOVrHOOCM m •-!
cMr-ooorocM'S' co in
ooi-iooom co o
C^ t—i CT\ V^ ^ CM IO r-l
DQDr-OOOQ^HiH'VQQQCOQQinDQQ
ZZZorHOZooiDZZZcoZZoZZZ
rococo nmm mcoco cow romco
o o o ooo o o o oo o o o
ojQooooo°Sooooooooooo
rH i— 1 rH iH rH rH
-WCOTrOOO«J«CMCMOCM<>4CMOCMCMOCMCS)CM
oooooooooooooooooooo
OrHOTr-WVOCMCMCMOOOCMOOCMOOO
H
U EH
EH <
U < X
EH X Oi
< O. CO
i-5 co O
H X X S
EH EH Oi
CO CO OT W < X W ^
J J W »J CO CO EH i4 ft EH -~ iJ
COi-5ZZ>HZ>H>H>H EHCdMX-^MEH>HtdtaZCdZZZ HZXXUXMK1U >4HXXEHa;
ZtOaaXCUXXX <*4EHO-Hixa1oe;a;
oiHOoaonno x H x^ xoo
09000:00:0:0: a. ^i z x >H xx
(XOXXlJXljHqJrH >H|JDCQWU>-IIM
rqsCrfrtlO^CJCJOQMHX 1 rlCM 1 D—~
XOKEH«CEH<<4i4i-^o:UEH C >n "" C CD CO CO
O r-fEnZr^&iEHZOEKlQSEibQ 1 EHCO 1 MMrH
IH Qi U Ed Cd Ed O O Cd H Z ^"1 H n O rH M p£ Oi 0«
QEHE-"O)SXOZQQUOQQCDOQDEHE-iEH
Q
oo
0
0
CM
0
0
Ed
H
£
a.
CO
a.
i-]
>H
EH
(d
0
g
M
o:
EH
CM
CO
0
CM
CO
O
CM
CO
O
CM
CO
O
<£>
rH
O
CO
o
dp
0
in
rH
0
rH
Ed
i^
t1^
a-
CO
O
a.
i^
I
X
Ou
I-H
04
EH
Q
Z
vo
O
<*>
o
CM
O
0
s
X
Oi
CO
o
X
s
O
H
I-H
Oi
EH
Q
EH
O
O<
OS
EH
0
£i
M
*S •
3 TJ
C 0)
4->
(D n
x: 3
±J T~*
T)
CO <0
10
T5
TJ -H
a> a
C -H
-H -H
0) (D
T> W
10
n
-rl CO
TJ§
<» -H
4J 4J
U CO
(V M
4J 4J
0) C
T) 0)
O
4-> e
c o
a> o
o
-------
.
^^
Cn
CO
m
w
"I
(0
0>
4-)
-H
n •
K
O
o
^
**"!
a
n
•H
4J
rH
1
A
•o
0)
-H
a
CO
H
a>
Q)
"^
o
0)
A
c
o
n
C
asure
0)
CO
8
&
3G
o
•v.
&*
"
Js
i
CO
m
i
0)
.Q
.
^<
(C
ool
4-
c •
a c
S, E -r-
H 2 s
0) „
> 5
"(IT
•H s ^
w
0 -
01 « C
u -rl 1C
Q <<-< a:
(in "^ S
c c
IB (0
3 -H
a -c
1
rH .M
rH jj;
M —
« "S
g Q
2 °
T)
C . i
S g
a! *
Q
^ M
^-H Q
O 4} _^1
M ® 'K a:
Z » « 0
(Q T" 0^
S ^
10
Z
rH
10
O
V
r;
O
m f*
^* «
0 ~
rH
VO 0
rH CM
o r-
'
rH O\
CM 00
0 CO
rH CM
co m
o a\
* •
rH rH
CO CM
O 00
'
CO (H
0 Q CM Q
o Z co Z
O CM CM
O O O
dp *"
° Soo
OOOO
CM O «• O
EH EH H U
O O Q O
Q Q Q Q
a a a a
a o a o
r~ CM rH CO TT
^^ ^5 ^o r^ if)
VC rH O CM O
w> r- r- r~ rH
CM CO CO OO CO
rH O O O O
. ... .
o r» CM ^ o
tH CM rH 00 ^H
CM O O O O
... .
m CM <* ^ in
o r- in m *T
^* O O rH O
•
^* ^* o\ ao r*-
co r» m CM T
^r o o rH o
...
^* ^* O\ GO p-*
QQQCOQQQQr-inCMQQQDQQQQQtrQQQQQQ
.
CMCTlCM CMCMCMVO CMCMCMCMCMCMCMfMCM CMCMCMCMCMCM
OOO OOOO OOOOOOOOO O O O O O O
dp dP dP dP dP
ooooooooooooooooooooooooooo
1 _J 1 1 1
OC3OOOOOC3OOO^OOC3OOOOC5OC>OC3OO^
ooo^rooooT'T^rooooooooorj-oooooo
Ed H H
Z Z Z
Ed H U
N N N
U W W Z Z Z
ff\ 1*1 r*i fn m pn
N co N O O O
— a 2 2 2 c: a: a; ta
U Q UUUOOOZU
Z H CncQCQiJtJiJWZ
< X Oi H OOOXSXtOU
Q o o z a:o:a:uucjzco u
z a- .J < ooo<<i-i<;<:o:i
-------
•a
0)
c
C
O
O
in
i
XI
<0
EH
J>,
>
•H
-H
CO
O
CU
LJ
O
Cu
JJ
cp €
"i 2
3 3
Z n
co
s
cr
c
a
E
3
c
CO
-H
•a
2
c
CO
-H
•c
rtl
2
o
•a
0)
o
V
JJ
0)
Q
Q
Z
.
tJ
o
Cu
£2
(0
Z
CO
O
0)
o
Q Q Q
Z Z Z
CM VO CM
rH n i-H
OOO
dp <*> <#>
000
ooo
000
CO
u
Cu
H
1
a:
u o
u u a
Z co O
H JH O
^H a o
Cu U S
OOrH^rm^rHCMiH CMCM
OrHCMrHOOOrH rHi-l
Ln*rcDir>'*r\D o o
OO CM ^5 rH CM rH rH O CO f*")
OOOOOOOrH OO
Qoajo^voNio^
CMPOCOVCOOOO CMVO
OOOOOOOO OO
oocftor-CMvooon cno
lO l& ^ O 00 t-H r-\ rH ^ f**1
C^OrHrHOOOiH OO
CSJ ^* tH O d if) OO VD '•J1 r""1
LO ^O CO CO (^ r4 rH rH ^ r*"
OOrHrHOOOrH OO
00 ^ tH O O^ lO C** VD V O
QQ^VOrOCOCMi-HOrHQQQCTiOOQQQ
2S200rHiHOOO^H2SZOOSZS
CMCMO^ CMCMCNJ OOC^CM OOQOCslCSl
rHrHCSJ r-lr-li-l CNrHrH ^CNJrHrH
OC3C5 OOO OOO OOOO
dp dP dP dP dP
df>dpc*>OoOrt9^Pe¥>OdPdf>dPOclPdpdpdp
f~H rH rH ^^ rH t*H ^^
oooooooooooooooooo
oorov^vrorocMroooorocNooo
U
EH
*3f
^
Hg
EH EH
co co co H < a u
,-5 ^ CO J CO CO EH J CU EH
CO>H>HlJ>HlJlJt-3 U *? •< < W
COJZZ>HZ>H>H>H EH td i4 X ^~ i4 EH
(-3>niaiaZHZZZ rf EH < EH i-5 < <
xzxxwxatdw J < x x >H x x
2 W CU CU £C 04 *C *C X rtl »-3 EH C^ 1^ E^ Ai
UXMMCuMCuCuCu X < X td X CO
aCuCQOQMOHHM EH X Cu hJ X Cu O
cuHOOcaonoam x EH >H ^ a
Mmcdosooiooo cu x >j N >H .j a.
CQOOOCdOOiCiiOi Cu N Z X >H
OCC^JOJOOOZ J EH W EH EH ^
CtfOXXiJXiJJiJH >H iJ D CQ W U >H
OJOOXOXXXCt;ZX>HCQ 1 O EH
rqa<
O
'
0
0
H
EH
«;
a
Cu
CO
o
a
cu
*_lt
l_3
>1
OPROP
CH
s
o
H
Q
CO
M
a.
EH
Q
Z
rH
O
rH
<#>
O
O
o
[V]
EH
ss
Cu
CO
O
a
Cu
>H
a
EH
U
>H
X
o
EH
g
M
Of
EH
Q
Z
o
CM
O
<#>
O
(M
0
o
u
EH
OSPHA
a
CU
25
W
a
Cu
H
of
EH
Q
Z
ro
CM
rH
OP
O
0
O
u
SPHAT
O
Cu
>H
r3
O
EH
H
H
Q
Z
+
a
EH
O
CU
EH
+
O
Cu
M
0)
*3
c
0>
4J
CO
CO
•o
CD
C
-H
$
n
-H
T3
(I)
jj
O
0)
JJ
a>
jj
c
0)
o
M
0)
CU
n
j^J
c
2
g
0)
u
3
• CO
•d CD
« a:
3 EH
*n
T) T)
CO C
CO
•a
-H Q
az
-H
rH. C
0
V
M *O
CO (U
ra
n co
-H n
4-> -H
CO
H Q
0)
U C
C co
O Q)
o E
rH ()}
< H
165
-------
hn
w*
*s^
fn
w1
3
W
OT
fli
W
•H
a
e
(0
CO
(11
u*
4J
-H
i
M
a
i
to
VD
<
V
^H
X)
EH
m
4-1
C
>i a
H ?>
O
C
<0
s
«C4
4J
c
0)
u
(J
o
a
^H
•Q a
«H CO T
0 4J *
M «
0 ~
rH
s £
o "
rH
00 £
VO Q ^ Q Q Q
02^222
rH
rH rH f~ CO
rH rH OO OO
0 O 0 O
dP <*>
O e¥> O dp <*> <*>
0 0 0 O O 0
rH rH
O ^* *^ ^* ^* 4*
O O O O O O
^ o *r o o o
EH EH Ed Ed Q Q
§§§§§§
1 1 1 1 1 1
a a a a a a
a o a o a o
CO
rH
rH
•
VO
ao
o
*
CM
rH
O
•
r-
cy>
o
•
v
en
o
nr
Q en Q Q
2022
*
rH i-H rH
rH rH rH
O O O
dp
dp 0 <*> <*>
OOOO
rH
^* f""i ^9t ^*»
^T *mJ ^^ ^^
OOOO
o «a- o o
U
<
Q
Z
M
J
U CJ O
X O X X
£0 X (Q CD
1 0 1 1
10 1 10 <0
_C* *ft t 1 C
<*-i (Q +-' C
tt4J rH g
^J ft) fit (It
^1 W UJ *w
<0 £ -O O>
r-
«»•
CM
CO
CO
rH
"
en
*)•
o
•
<*
r~
rH
•
VD
m
rH
VO
Q Q m
Z Z rH
*
rH rH
rH rH
0 0
dp
dp dp 0
0 O 0
rH
^* ^J* ^?
O O O
o o *r
a
a
H
g
cu
a
a: a;
II
2 O U
M < <
«: cJ S
l*< C^ tH
a a< DJ
. n pji r_^
HH m |X4
•«: x x
CO T
ao in
rH CM
v m
CM in
rH rH
* *
ao in
CM ^
o o
• •
o en
vo en
rH rH
•
T fl*
vo en
rH rH
*r *r
vo en Q
rH rH 2
* *
rH
rH
0
dp dP
O O dp
000
rH rH
Of* _f|
tP ^"
O 0 O
*r v o
RDANE
ONACHLOR
HLORDANE
O 2 U
r3 1 1
X m 4 (C g
XLJ ffl
M ig
O 4J &
in
rH
O
•
in
rH
O
*
m
rH
0
•
m
rH
O
Q Q O
2 2 2
rH rH rH
rH rH rH
O O O
dp dP *!
0 ° N
tN
^* ^* CO
O 0 O
O O rH
H H
2 2
H U
N N
H Cd
m ca
RICHLORO
RICHLORO
EH EH
X f) •«•
CM CM
S rH rH
Q Q
Z 2
rH rH
rH rH
0 0
dP dP
O O
^* ^*
O 0
0 0
w
2
H
SI
W 2
2 W
U 03
N O
2 2
Cd O
CQ rH
RICHLORO
-TETRACH
EH «•
in ro
en CM
rH rH
Q Q
2 Z
rH rH
rH rH
0 0
dlP dP
0 0
^* ^*
O O
o o
Cd U
Z 2
U U
N N
2 2
Cd W
CD CQ
§§
ss
-TETRACH
-TETRACH
m m
^ ^
m «•
CM CM
rH rH
VO
VO
O
m
PO
O
•
*!•
rH
O
VO
m
o
•
CM
VO
O
CM
Q VO Q Q Q
20222
•
rH rH rH rH
rH i-l rH rH
O O O O
dP
dP O dP dP dP
O O O O O
rH
^* ^3 ^* ^* ^3*
O O O O O
o >» o o o
a
Z Ed
H 2
N U
Z N W
Cd 2 2
03 Ed U Cd
O O 63 A 2
OS O 2 >H H
o <£ a x x
J O ^ EH EH
X rH < X X
O X X 0< 04
< U EH < <
f~t *»• m s ^
C^ W» »»* ^« ^i
2 x a, Ed a
Mrj ^* f i r i
M-l "4 ^X VX
0« X 2 < <
Q O Q
Z 2 2
rH rH rH
rH rH rH
000
dP dP dP
O O O
^* ^* ^*
o o o
o o o
H u
2 2
Eg
W DC t*
OS o
*6g v^
DUD
1-3 X i-3
fci Oj Eu
166
-------
73
0)
C
-H
-P
O
U
VO
I
JQ
<0
EH
^
rH
0)
^
•H
•H
CO
O
cu
*H
O
fa
A
<4H 00
O 4J
C
co 1
•is
3 3
2 n
10
0)
2
K
4-
a
1
tt>
^
E
(C
a
•0
a
(^
4-
C
(C
C
I— 1
"
flj
t|
W
0)
c
10
s
c
c
Li
M
ft
W
in
-H
*W
•H
(Q
C
4-
fll
•u
G
.
£»*
s
A
.
C
-H
•
•a
CO
c
(0
0)
c
rd
•H
•a
S
C
(0
•o
2
(-V
1
•a
0)
o
0)
4J
0)
Q
g
OS
H
O
CU
0)
g
(0
2
•H
(0
O
Q)
x:
o
O Q
2 2
0 0
<#> dp
O O
«ST •*
0 0
O O
H
2
C3 Cd
2 co
U X
OS Cti
^ 1 l"p
Cu U
Q Q
2 2
0 0
dP dp
O O
q. ^.
0 0
o o
co
•J
>H CO
2 J
Cd X
X 2
Cu H
M X
CO CU
O M
A CQ
O O
•J a:
x o
CJ J
o x
2 O
0 H
X a
[- VO
rH rH
O rH
CM CO
rH CTl
0 0
co r-
0 rH
0 0
*f ^*
rH 0
0 rH
CO ^*
rH 0
0 rH
CO VO
rH ^"
0 0
0 0
dP dP
m o
rH CM
0 0
CO CM
CO
1^
3&
X Ed
2 X
Ed Cu
X H
Qj Q3
H O
CQ A
P
M EH
OS Ed
EH EH
O tfl
VO CO
i-H CM
o m
CO CTV
o o
CO VO
VO VO
O 0
in co
CM a*
rH rH
^* CO
m IH
rH CM
•er oo
in rH
rH CM
<*> <*>
O 0
O 0
rH rH
O 0
o o
<• •*
CO
r3 CO
Ed 2
X Cd
Cu X
H CU
CD H
O CQ
tf Q
X iJ
O X
< u
EH <
2 X
Cu X
co m
r- CM
rH O
CM in
in CM
rH O
en
o
0
CM m
VO CM
rH O
oq in
VO CM
rH O
CM
VO Q
rH 2
rH
rH
0
is
0 CO
0 O
T rH
CO
iJ CO
2 X
Ed 2
X Ed
CM X
H Cu
CQ M
O O
OS 0
o a:
rH 0
X .J
o x
CU EH
Ed U
X O
CO CM
CM CM
O O
rH rH
rH rH
O O
00 VO
o o
0 0
c- oo
rH rH
0 0
r~ o
rH (M
0 0
vo vo
0 rH
0 0
rH rH
rH rH
0 0
dP dP
o in
in r~
CM rH
0 0
CM CO
CO
J ,4
X X
Ed CQ
X X
Cu Cu
M M
0) CQ
O O
a: oc
o o
X X
o o
£3
O Ed
2 0
S
2
EH
a
Cu
*«^
^^
^Mt
cc
EH
a
Cu
M
Jj
C
c
CO
r*J
jJ
00
10
•o
a>
c
-H
UH
0)
•o
CQ
-H
•a
0
O
«
4J
0) g
4J
00 OS
3 EH
T"»
*O T3
(0 C
(0
•o
-H Q
0.2
-H
rH C
O
0)
W -O
CO 0)
00
00 (0
G «Q
O
-H 00
4-> -H
m
H Q
C J
0)
u e
C CO
o a>
u g
H 0)
rH X!
< EH
•d
0)
N
>1
rH
(0
ttj
J_)
C
00
10
3
4J
O •
CO 0)
H rH
4-) »Q
X 10
(1) 4->
x:
C 4->
O
-H C
4J -H
O
cO C
<4H O
I-J
0) 00
4J 4J
O
A C
O
-H 0)
x: M
3 CO
C 00
•H T3
00 3
0) O
"S i*
4J O
CO O
J3
CM
0)
x; c
4J O
-rl
C 4->
-H 0
CO
0) VI
VI <4-l
0)
3 0)
x:
(0 JJ
0)
rH »
Cu fl)
ts w
CO O
m
-------
M
0)
p;
CD
4->
C
0)
o
M
fli
w
04
n
0)
S-
(0
co
10
C
n
T)
C
Q
C
O
0
j,
r"H
\^
A\
-P
i
•H
-P
V)
IH
o
1*4
^
X)
ID
4J
C
i
isure
»«
s
tr
4-
C
5;
j-
J;
•"
re
a
z
T
a
-i-
M-
4-
Ojian
(C
f£
c
a
01
111
-p-
-H
•H
1C
(C
4-
c
(0
c
S~n
•o
CO
c
COCO CMOOCftOO
CM^invovo vovovoco
dPdPdPdPdP jojo*'*''
ooooo J^jToo
sssss <*<*ss
* 4e
OOOOO rHrHOO
OOOOO O O O O
oo oo co ao r- r~r~CMCM
^H|
Z
^^ 3C ^0
Ed 04 VI
N M T) VO
Z O 1 *
Ed O vo in
co ca cd i iJ ^ -
•oo ovo>Hi4inv
1 Ofi iJ ^Z X * ..
EdOEd XVOEdZVCO
Zi-H'ZrH'U -XEd -1
6Q X W >H rt 04 Z CO Id
NONZO.mM04 1 EH
Z*.
^J f*
3 d)
•Q ^
a>
c ex
Q) 04
n
co n
M 10
o<
n a>
-H 4J
c
rH 4)
(0 n
U V
-H M
s a
A 0)
O W
10
0)
JB n
4J 4J
C
o> a>
c g
-H 0)
•4J M
10 3
O «
-H 10
•§!
•H
JJ >
0) 4->
C g)
0 0
u a
10 &
•a il
10 S
M TJ
0) M
*w *O
•H C
rH 10
(0 4J
3 03
<
a
O
jj
(0
o^
o
^1
3
co
•0
0)
4->
^4
O
a
0)
M
•
#*•»
Q
EH
2"
*— •*
X^
ST
EH
5,
a
M
0)
4-J
co m
.C 3
.p "O
•a
n to
10
•0
•O -H
CD a
C -H
-H rH
VH
-§2
ro
m
•H CO
c
•o o
0> -H
J-> 4J
o 4-)
0) C
"O CD
U
4-> C
C 0
CD u
o
IH rH
CD rH
04 «C
168
-------
.
o^
{^
3
-A
f-l
CO
T(
O
5
&
C
O
00
JJ
i
Jl\
Measure
8
ta
o
o
M-4
O
M
2
5
S
co
•
CO
1
0)
i-H
•8
E-l
(0
Kj
4-
a c
C 'H
^ a a
> ?
'o re -o
j tt 4J
«S co
0 -
°« a c
nr] g
0 £ s
6" 3
c c
(t (0
3 -rl
51 T3
V
H C
S "
S| «
o s
4J -0
c 5
o o
H a>
2 ®
04 Q
.!l§
°sS
rH
||i g
S3«a
ji"
a>
g
a
H
u
c
r*
O
GQQQQQQQQQQQQ
ZZZZZZZZZZ2ZZ
ooooooooooooo
eMCMCMCMCMCMCMCMCMCMCMfMCM
OOOOOOOOOOOOO
ooooooooooooo
^^
U H
Z 2
W OS <
NO Q
Z 3 Z
US H
0 U J
o < —
OS Z O O O
•JzaaaaoacnKcaajH
aiQOQQOQIOl Itf
oni i i i i i 10 i ia ia Q
< C .C <0 g 4-) >4
x (o aa.Oia,aa,ai4J g ^-i w
S4->a,oa,oao(o.Q (71*0 o
Q Q Q Q Q
•7 7 *7 *7 ^
M X-t rf-t ^-4 A
dP <*P dP «*P dp
O O O O O
CM CM CM CM CM
O O O O O
O O O O O
U
Q
M
X
o
ftj
U H
2
06 OS rt3
ss§
s a o
z o o >3
H < < S X
05 EH M O 63
W 33 3C O S
Q
CC
E-i
a
Ou
2
H
£4
S
:he num
justed.
~-d
W tQ
"T>
•0 -H
0) O.
C -H
-H H
Q) (u
n
-H m
c
•O O
0) -H
4-> 4J
U (0
a> M
4J 4J
0) C
•0 g
4J C
C 0
0) U
0
® <
169
-------
•
O1
hn
U1
CO
r*
*•*
CM
rH
00
m
V
•^
w
JJ
<0
0
C
-H
00
0)
a
g
10
CO
10
h
0)
-H
O
M
O
Cu
0)
.C
c
o
00
0)
M
3
00
(0
33
8
a
^*?
rn
w
1
0)
-*
•H
t \
+*
•H
Of)
O
o
fa
XI
<4H CO
O 4J
M d)
£j O
Z oo
(0
rtl
S
tc
4-
c
C]
p
a
i-
(C
CC
a
S
T
(U
•H
M-
-H
C
(C
fl
c
H
It
M
0)
0
JJ
c
0)
o
M
0)
CU
01
-rl
-H
H
1
n
4J
n
0
.
X
jp
*
•
c
-H
.
•o
JJ
W
C
<0
c
•0
-H
•a
s
c
(0
-H
1
*•
•a
Q)
JJ
O
CD
JJ
0)
O
g
0£
EH
O
(D
CS
10
rH
ID
0
1
0)
O
o o o
co *r vo
000
o o o
CM CO VO
o o o
• • •
vo vo o
o o o
000
co r- o
CM CO VO
000
000
CM 'S1 VO
000
o o o
CM ^r vo
o o o
<*>*><*>
000
o o o
rH rH rH
o o o
0 O 0
CO CO CO
•o
a)
o
0)
M
o
0
u u
z z
td w a;
N N O
Z Z r3
H H 55
CQ m u
0 0 t£
ACHLORi
ACHLOR
ns-NOM
X X 10
Cd Cd M
rc s *j
0 ° 0
•"CM °
1-1 . ^
rH
° 2 °
^3* CTi
rH """I 0
1-1
m cr\ vo
rH in O
0 O 0
co C; co
in tJ °>
rH , O
1-1
0 g 0
in ~ a>
*H . <-3
rH
o 2 o
rHZ^.ZZZZO
rH
dp dp dP
OdPO<*>dP<*>dPO
OOOOOOOO
rH rH rH
OCOOCOCOCOCOO
OOOOOOOO
coocoooooco
o
Q Q Q Q Q Q 1 BQ
1 1 1 1 1 1 10 1
x io
aaaaaaa*j
z z
<*> dP
0 0
CO CO
O 0
0 0
^^
w
Z
g
H
i^
«*ir
O O
10 (0
i -r>
e H
IQ (D
0
o
rH
0
CO
O
0
rH
O
O
CT^
0
O
o\
o
0
o Z
dP
0 *>
O O
rH
O CO
O O
CO O
S Z
§ g
>j a.
U Q
M Z
O
CO
o
o
CO
0
o
o
0
o
CO
0
0
co
0
o
Q CO
Z o
dP
dp 0
o o
rH
CO O
o o
O CO
a
Q
M
^
o
cu
O* Qfj
TACHLO
TACHLO
Ol 04
u u
o
^1*
o
o
^*
0
o
o
0
o
^*
o
0
^*
0
0 0
0 rH
dP dP
0 0
0 0
rH rH
0 0
O CO
CO 0
w
z
2
CHLORD
EX
» CC
X M
8s
0
0
rH
dp
O
O
rH
O
CO
o
^,
•o
Q>
4J
O
IV
M
l-l
O
o
X
H
S
^
a
z
+
CC
E"1
o
cu
^
EH
O
2i
S
§ -d
C 0)
<1) 00
£ 3
•o
<0 10
T3
73 -H
at a
C -H
-H rH
CU 0)
t) M
<0
00
-H n
C
•a o
0) -H
JJ JJ
U <0
0) H
4J jj
0) C
cent d
conce
U rH
CD rH
Oi <
170
-------
.
Oi
1
3
w
0
rH
01
3
O
M
J3
1 1
4-*
*3"
tn
O
^3
4->
(0
cn
c
-H
n
di
w
rH
1*
(0
(O
4->
(0
h
S
-rl
O
r4
Cu
<1)
4J
Q
n
4J
(1)
M
rj
n
to
EH
>i
rH
fl>
>
-rH
4J
-H
n
o
CU
O
fa
>t
J3
M-I n
O JJ
H 0)
$ |
i 8
3 3
2 n
(0
/it
1
X
(C
n
-i-
c
a c
S -H
!r T1
a s
>H
3
K
n; t$
a; 4J
•rl
^
a c
-H (0
-r i
1 •*«
J—
C C
co *r
o o
0 0
co in
O £3
O O
co m
O 0
<#> <*>
0 0
O 0
-i *
^1 r^
^3 ^D
0 0
r** f-^
j— ^-»
9
0)
4J
O
0)
14
M
O
0
U U
2 Z
H U
N N
Z Z
U U
0 DQ
O O
CU CU
o o
rH rl
X X
o o
gg
&} M
SB SB
o o
CO VD
i-H CM
O O
0 rH
rH CM
rH O
t-H CM
O O
r- r-
H CO
rH CM
o o
CM CO
rH CM
0 0
CM co
rH CM
dp dP
O O
0 0
— J 1
Pn ^1
^3 ^3
o o
p"«- r^
r^ r^
a:
X
o
<
§H
=fg
n i
S a
4J a
0 ° 0
CO 2 ™
rH ^ CO
CM
0 0 0
«> S f-
rH » CM
rH
t- 0 00
0 rH rH
0 rH 0
rH £} CO
^ S W
rH ° CM
CM
° S °
f S *
rH ° CM
CM
° 2 o
t^aSQQQQ^
rHZ°.ZZZZCM
CM
dP dP dp
OdPOdPdPdPdPO
oooooooo
wJ _l m.1
T^ ^H ^1
^3 p* c* f^ F* r** r^ ^3
oooooooo
C^ ^D t^" CD ^5 Cj ^3 i^^
9
H)
4-1
O
(1)
M
M
O
U
O
rHHWWQQEO
QQOQQQCQK
Q Q Q Q Q Q 1 CQ
1 1 1 1 1 1 (0 1
£ (0
aaaaaaajj
aoaoaomxi
a a
z z
dP dP
O O
r^. f*^
n* ™
0 0
c? c?
^t.
U
<
2
H
•J
0 O
X X
CQ ffi
(0 10
g 4J
S -H
«fl)
w
WO
o
in
o
0
«S"
o
"
"S1
o
o
CM
"31
O
0
«•
o
o
v Q
o Z
dP
O dP
0 0
rH
Ol^
r*
o o
LO *~*
Z
H
CC Z
Q H
J cC
u a
H5»
Ai
Q U
Q
Z
dp
O
^^
0
c?
OS
s
0
<
H
Cu
w
X
o o
cn m
CD
V^
X Cu
000
CM cn m
rH 0 rH
000
cn oo CM
O O rH
co m o
rH O rH
O O O
r- r- r-
o ao co
rH O i-H
O O O
o en v
rH O rH
000
o en «»•
rH O i-H
dP dP dP
O O O
O O 0
i"H ^H ^H
Of^ C"i
Sj VB*
o o o
^^ ^^ t^
^^.
•o
0)
4J
o
U 0)
Z M
< -J
§ 5
o —
ij
XXX
O H U
SH OS OS
X rH M
OSS
.
Q
z
±
a
H
H-
O
CU
0?
H
-r
o
cu
M
3 T3
C 0)
4-)
0) n
£ 3
1 1 «n
^T3
n 10
"•0
•0 -rl
a) a
C -H
-rl rH
*H
0) (U
•3 M
CO
m
-rl (0
C
•d o
0) -H
4-> 4->
U 10
-------
•
«^
fc_
3
^
CO
CD
H
§•
G
•H
t \
4-1
(Q
O
U4
-
b
J3
IH oo
0 4J
(I) g
.Q 0)
E M
z-2
(0
$
tr
c
d
£
a
v.
5
it
a
z
•o
a
«»-
4-
c
«c
C
H
c
ffl
GU
a:
-rl
m
,-j
It
6
it
4J
Q
^
>cj
S
.
c
-H
Jsj
.
•o
4J
V)
C
10
0)
c
<0
-H
1
S
c
10
-H
•a
JE-
TS
i
o
0)
4J
y
§
C£
6-1
a
0)
g
(0
•-I
10
o
*
u
<&*•<»
rH CM OO
CM co co
CM CO VO
rH rH CM
O O 0
• • •
VO CO 0
co uo cr>
0 O 0
. .
00 VO CM
co in r-
O O rH
• • •
0 •«• I-
CO «3" 00
O O rH
O M1 rH
CO ^* OO
O O tH
dp dP dP
000
rH rH rH
O O O
o o o
355
"5
0)
4->
O
CD
M
O
O
P4 (d
Z Z
w w ac,
t& ta o
Z Z >J
F*1 M ^
0) CQ U
O O ^J co
00
CM CM CO
in o CM
o co o
• • •
rH "* 0\
o\ ^ oo
*•< . °
f-4
rH J^ CO
<^ SJ «*
CO ^J rH
CM
N CM ^
CM ™ rH
rH
CM rj a^
§ Q Q Q O ^ Q
°^ 2 S 55 55 rH 2
rH
dp dp dP
ooooooooo
rH rH rH
O*J-°«3'TrH O!
X H
0 S
CO
C
0
•H
4->
10
M
4J
C
0)
c
o
o
•a
0)
•a
o
0
a:
EH
dP
O
O
f—l
0
«9-
T
rH
CQ
O
•H
10
4J
o
H
*
^
Z
^h
OH
EH
O
Q*
^
EH
a
^
0)
C 0)
4J
-H
(1) Pi
c *^
•H rH
CD 0)
*O ^
10
00
-H m
a
•O o
4) -rl
-P 4J
u
-------
D>
-^
tn
3
W
03
. %
+j
o
m
M
4->
X
4J
Ot
i
M
0)
Idrin Recov
surements of A
«J
Ml
s
*-»
Q
U
a
^^
t >
g
\-S
f-i
o
S-
C1
5
5
fc
OT
N
r-l
1
<
ID
•H
A
10
EH
M
10
Ot
4-
C
a c
Sf a
t-
vo
dl°
10
<*>
r~
a\
dp
m
Ol
c*>
IO
CTl
dP
0
O
rH
O
O
ro
t~
Z
-i
3
<
•
/(PQ+TR+ND
number (PQ+TR)
:ed.
+^
C
c o
0) O
o
M H
0) r-1
04 <
173
-------
fTt
U*
~&i
3
00
0
1
rH
3
s
•o
9>
-H
a
to
rH
-
gf
" » ^
^j ^
« ^
O _
^ a
n ^
tl *
C
(t
2
c
rH
H
10
^
01
4J
0)
o
M
a>
a.
^^ lj
^
O 4-> ^J1
LJ Jii ^
1 1 ^
2 S J
2 °
.
X
£
c
-H
s
•
•o
w
C
10
2J
c
10
•H
•o
1'
C
10
-H
TJ
S3
T3
Q)
4J
0
Q)
(D
Q
Q
£
EH
a
(Xi
§
10
z
lemical 1
0
VO P- OO
rH VD P-
rH rH in
»* co in
VO O P*
o tH tr
• • •
CM 0 CM
CM CO V
000
•
co m co
CO CM CM
0 rH in
• • •
P- VO O
P- rH
X X 4J
m
OO
CO
CO
CO
•
CO
CM
0
•
CTt
in
CO
*
CTl
in
co
'
m Q
co z
dP
O dp
0 0
rH
0 5J-
0 O
V O
EH EH
Q Q
Q Q
a a
a o
CTl
^
o;
VO
OO
VO
CO
p-
•
CM
0
CTl
P*
™
CTl
DO
P-
°| Z Z Z
OO
dP
O dp dP dP
O O O O
rH
O ^* ^* ^
O 0 O 0
^" O O O
9 S S 9
Q Q Q Q
O £^ C3 C5
i i i i
a a a a
a o a o
m
^<
p-
CM
•91
VO
•
CO
^*
o
•
p-
p-
VD
*
,_)
VO
'
rH
Q vo
Z vo
dP
dP O
O O
rH
«• 0
O O
o «•
X U
m x
(0 1
a3
_i /tj
r*i w
Q Q
Z Z
dp dp
O O
«• V
0 0
o o
^•t
U
Z
rtj
M
^
^^
0 O
X X
m m
(0 (0
i *J
6 rH
(Q OJ
Oi-O
^
^*
00
00
VO
•
CM
00
o
•
o
CO
*
CM
O
00
'
CM
0 Q
co Z
dp
O dp
0 0
f^
o *r
O 0
•» o
[ELDRIN
iJDRIN
• ^ *•*
Q H
CM
in
in
CM
<3<
•
CO
m
o
•
m
00
*
rH
00
'
rH
Q co
Z *3*
dP
dP O
O O
rH
rr O
O O
O V
w
Q
M
X
o
(Xi
0£ 0£
SPTACHLO
SPTACHLO
X X
oo
CTl
CO
O
P-
CM
m
m
o
•
rH
CO
CO
"
P-
CM
CO
'
P-
CM Q
co Z
dP
O dP
0 0
rH
0 V
0 O
5J- O
H
55
rij
ICYCHLORD
[REX
0 S
•
a
z
oi
EH
+
O
CXi
0?
EH
+
Si
1 .
3 TJ
C 0)
-------
en
Cn
3
«J
/1\
"
V
JJ
o
to
JJ
a>
M
3
CO
n)
S
U
S
V— ^
U_l
o
^^
^
Itt
S
S
c
w
•
^9*
j
(p
3
EH
H
>
•H
•*~
"H
CO
O
CL,
M
A
IH co
O 4J
c
MJft
w
3 3
Z co
10
s
X
K
ai
c
a a
e ••-
a s
ji
K
(C T3
a jJ
S co
V
a c
-H K
4- (U
-H S
C C
re
Q
1-
iH
H
re oc
§H
£ o
a
0)
g
m
1-1
(0
o
"TJ
0)
u
O VO O
VO CTl V
O O CM
CO O O
^* VO CO
O O i-H
• • •
vc r~ cn
0 rH CM
O O O
. . .
in OO rH
0 0 CM
• • •
O O O
VO Ol rH
O O CM
• • •
000
VO Cft rH
O O CM
• • •
dP dP oV
O O O
o o o
rH iH rH
000
o o o
*, « «
•o
0)
o
0)
M
o
O
*•**
U U
WHO:
N N O
z z ^
a a s
(Q CO O
o o <
So o
>4 Z
X X 1
U U W
< < C
X X 10
MM Ll
1*1 *^
o ^
s: ^
CM
»3< 1~l
0 °
CM ^
CM
r- ro
rH CM
O CM
•
s S
CM ™
CM
S ^
* CM
V rH
CM Z ^
CM
dp dp
O dP O
o o o
i— 1 rH
0«*0
000
*-o^r
Q Q Q
Q O O
1 1 1
a a a
a o a
ao
CM
CM
O
OO
1-1
rH
CM
O
•
CO
a\
rH
•
CM
CJ)
•
CM
Z Z Z Z rH
*
<#>
O 0 0 O 0
*r ^r v ^o
o o o o o
oooo*.
o
Q Q Q CO X
Q Q Q 1 (0
1 1 1 (0 1
a >o
a, a a a jj
O CL O (0 ^3
O Q
Z Z
<#> dp
0 0
*•
0 0
o o
^^
U
2j
i
M
f>3
^^
O O
CQ ffl
i i
10 H
HS£
^
X O
Q
Z
dP
O
*•
o
0
X
§
n
^
Z
+
K
H-
O
~
2
H
O
M
0)
C 0)
JJ
(D n
JC _3
^ -o
co n)
TJ
•0-H
a) a
C -H
-H rH
(D 0)
*O M
10
CO
•rl CO
c
•O 0
<1) -H
4-> JJ
U 10
0) U
JJ JJ
0) C
•o a)
o
JJ C
C 0
0) O
o
M rH
&. <
175
-------
fc_
^_
3
0)
0)
H
|
CO
4)
4J
•H
03
O
e
0
o
^
"a
n
4J
3
*
•o
4)
•H
a
co
rH
4)
>
fl)
w
•o
s
4)
£
4J
C
O
oo
4J
§
M
3
n
10
S
8
H
o
^
*w
0
^
10
C
3
CO
.
m
iH
4)
rH
£
rH
4)
•H
*r
"O
CO
O
Ou
^
X>
<*H n
O 4J
c
IH 4)
4) £
•§ g
3 3
2 n
(0
s
*
c
a
!
tr
(C
a
X
_
a
-r
f-
4-
IC
5
C
H
s
10
•H
•a
2
C
10
-rH
•o
4)
•a
4)
O
4)
4)
Q
2
**
O
CU
0)
g
10
2
co
O rH
• •
O VO
W 00
O rH
O VO
O rH
* *
dp <#>
O 0
iH rH
O O
O O
*«
•o
4)
4->
O
4)
£
O
O
^•^
U U
2 Z
U U
N N
2 2
u a
m m
HLORO
HLORO
SS
X X
WH
w
X X
m
f--
^*
•
m
O*i
CO
•
CO ^ CT»
VO ° CO
^ ^ OO
• ^J •
n J" m
f" m °
CO ". 00
00000000
rH rH rH
OTOTT^rTO
00000000
^0^-0000^
o
EHMHUQQXU
OQDQOQDQX
a Q Q Q Q Q i m
1 1 1 1 1 1 (0 1
A fO
aaa <#>
0 0
«• ^<
o o
0 0
^
a
2
g
H
l^
n to
i i
10 10
g 4J
6 rH
m m
&•§
m
VO
m
m
r-
".
rH
^*
O
O
CM
m
o
CM
in
o
CM Q
in 2
•
<#>
0 <*P
O 0
0 TT
0 0
^0
2
M
ed 2
ss
U Q
n SS
Q U
r-
o
^*
•
03
CM
CO
t-
CO
O
0
c-
co
•
oo
f^.
CO
oo
Q r-
2 co
*
c¥>
dp O
0 0
„. o
0 0
O V
w
Q
H
X
o
a.
f»1
CHLOR
CHLOR
o< a.
X X
rH
r^
CM
VO
CM
CM
CM
CM
O
CO
CM
r-
CO
CM
CO Q
CM 2
•
<*>
o <#>
0 0
0 •»
0 0
»0
w
2
8
o
II
0 S
Q
2
+
o£
E-i
o
•— •
^^
S"
EH
o
IH
C 0)
-------
43
_j
•H
0)
A
•p
*
-H
S
Tl
\J
-P
(0
-H .
n
n
**!
i*
10
co
Tl
U
0)
•H
a
CO
c
D
V
JJ tJl
\
O S
m ••
4-> n
U CO
<0 H
ri a
X 10
U CO
C 4)
0 4J
n n
JJ O
suremen
it Comp
«8 rH
CJ> CU
Q 4J
O rH
v? 5
O T5
*o "*
r?CO
(0 rH
C (tl
i >
3 (U
CO r3
•
VO
rH
0)
H
10
EH
It
?1
0)
"j
*J
n
0
cu
-^
o
b
>,
.Q
O *J
C
Numbel
jasureme
s
a
^
C
£
a
^
5
fj
d;
S
n
\»
a;
-r-
**-
•r1
±.
c
1C
c
i— i
r™
1C
a:
O
4->
c
0)
U
0!
Cu
^
a
-H
*4-l
-H
,_j
It
3
0
ta
4J
(C
01
^
fl
S
C
.r
S
.
•a
4.
CO
c
(0
(V
*•
c
It
T
a
c
1C
•c
•o
(K
C
03
01
S
tt
0
a
g
rH r- in £
CM in VD co
0 0 rH CO N
' ' ' ' CO
r^ r^ VD in in
O O *}« VD CO
0 O 0 0 CM
. . . .
^* VD CO ^* ^
co m rH m f2
0 0 CM V .
CO
CTl rH rH CM £!
co in co VD ~
O 0 CM '9' ""
' ' ' ' CO
CTl rH rH CM Jj
OOCM'S'Z.ZZ
dP dP dp dP dp
oooooooo
rH rH rH rH rH
OOOOCOOCOCO
OOOOOOOO
cocococoocooo
^^
"O
0)
4J
o
u
M
0
o
~"
a a
a a K
N N O
2 Z rH-
w w a;
to o o
0 0 <
a5 EC 2
S S 1 Q Q Q Q Q
U O CO 1 1 1 1 1
«< < C
xx to a a a a a
•«r
^*
CM
o
oo
rH
CM
CO
0
•
0
rH
CM
in
0
CM
•
in
(3 C5 ^^
Z Z CM
*
dp
dP dp 0
O O O
rH
CO CO 0
O O O
O O CO
^ J^ „
C3 3C ^^
a i a
1 ro 1
Si rd
a a 4J
Q Q
2 2
dP dp
0 0
CO CO
0 0
0 0
^^
u
t-t
1^
*—
o o
X X
ID CD
1 1
ro ro
1 rH
(0 Q)
Oi-O
CO
o
rH
'
r~
r-
o
•
in
rH
0
•
•31
G\
O
CO
O
rH
"
CO
0 Q
rH Z
*
dp
0 dP
0 0
rH
0 CO
0 0
CO O
Z
H
A Z
ss
U Q
M 2
^^ *•«
Q H
Q
2
dp
o
CO
o
o
<£.
S
O
n
Cu
rH O
CO CO
CM rH
' '
«3" rH
in *!•
rH rH
Ol 0
CO CM
O 0
• •
CO CO
C^ VD
rH rH
co r-
O\ VD
rH rH
* "
co r-
c\ vo
rH rH
" *
dP dP
O 0
O 0
rH rH
O 0
o o
CO CO
u
a
H
g
Cu
u ta
2
cd <
38
53
!SB
QJ >H
DO jxj
a o
Q
z
dP
O
CO
o
o
1
2
•
S
Z
+
CC
E-i
+
O
Oi
"^
c?
EH
+
O
CU
*"^
number
:ed.
4-*
0) n
£ D
^j *n
•o
(D (0
ro
•o
t5 -H
o> a
C -H
•H rH
14-1
CD 0)
"O H
(0
n
•H CO
C
•o o
a> -H
4J 4J
O rfl
0) Vi
0) C
T3 0)
O
4-> C
C 0
(D O
O
M rH
cu <
177
-------
CD
42
*^
A
JJ
-H
^
Tl
W
Ml
Q>
4J
(0
•H •
0
05
00
**4
fit
U/
a
&
re
CO
•a
0)
-H
a
n
C
D
V
43 •
jj O*
\
O 3
n -
.u m
U CD
X re
(4 co
C 0)
O 4J
00 00
4J O
II
Q) U
3 *J
« -H
TO rH
S n
-H
r\
fjj g
••*. ^r*
O
CD *O
CD
•n x
O -H
a
gt
*J
i *
3 CD
CO h3
•
r^
rH
1
CD
rH
JO
EH
rH
0
S*
"^
H~
•H
n
O
Oi
0
.
43
«w n
O 4J
Si
i s
3 3
2 n
re
1
.
^
s
inj
c •
a c
E -H
a s
5
re -d
a 4->
S w
73
d) c
-r- re
M- 0)
-r S
4-
c c
0
•
CM
r- D
o 2
*
dP
O dp
O O
tH
O CO
O O
CO O
U
2
<
g
s
X X
U U
X OS
OS
^>
§
4-
K
EH
a
C
/^
EH
O
Oi
CD
f.
•o
C CD
CD 10
43 3
^ T?
co re
"•o
n -H
CD a
C -H
-fH rH
CD CD
re
m
•H m
•a o
CD -H
•U 4J
u re
CD M
*J 4J
CD C
TJ CD
O
4-> C
C 0
CD U
U
JH rH
CD rH
Cu <
178
-------
•o
•rH
2
0)
4J
4->
-rl
S
*rt
y
4->
(0
-H
o
CO
co
**«
CO
0)
'Tj
e
CO
•o
0)
-H
a
n
c
0)
4J C7^
"Ji
O S
to •-
U 0)
(0 rH
X (0
Ed co
C 0)
O 4J
n m
surement
it Compo
OS rH
a> a
Q 4->
O rH
^|
0
o -o
0)
l|_l y
n
V*M
>lW
i-l
ID rH
s S
3 0)
CO i-J
•
CO
rH
1
<
0)
rH
EH
rH
a>
>
"^j
j
CO
o
CU
g
Cu
>,
X}
IH n
O 4J
Number
jasuremer
s
n
4J
c
£
0
£
5
n
Q
2
_
a
£
"'
c
c
r-
«
^
a
o
4->
01
o
n
0)
0.
y
OJ
-rl
^
H
re
g
1C
4-1
It
01
^
R
C
2
•c.
4-1
CO
C
n
a
*
c
•c
a>
2
C
(C
•1-
T
d
/i1
;X rH
0 0 rH CO . rH
. . . ^
o r- r- r- cxi o
O O i-H rH VO O
O O O O rH O
. . . .
T CT> 10 rH 3 CO
CO IT C* rH " rH
O O rH *!< . rH
rH
v in CM P- S co
CO V CT> 0 "' rH
0 0 rH *T ~ r-l
• • • • . •
i— 1
T in CM f* ^ co
o o <-H -a- 2 ^ Z Z Z Z n Z
' ' ' ' rH
Q*P <(P oP t#P e^P tfp
OOOOdPOdPdPdPdPOdP
oooooooooooo
rH rH rH rH rH rH
oooocoococococooco
oooooooooooo
COPOCOCOOCOOOOOCOO
•c
0)
4->
o
0)
M
^4
o
^^
U U H
22 2
W H Cfl <
N N] O Q
2 2 HH 2
W W K H
CQ CO U rH
o o •< ~
OS OS 2 U U
OOOE-iE-iHHQQXUS
r3^2QQQQQDCQX«
XX 1 Q Q Q Q Q Q 1 CO 1
uoni i t i i IIQIM
< < C J3 10 E
xx « aao.aaa.a4J e
xx4jaoaodoioAD<
Q
2
dP
O
CO
0
o
1
(d
If)
•§
CM
O
rH
O
Ol
O
•
r-
o
o
•
•a*
O
O
rH
O
O
CO
a
f£
3
X
y
o
Q
2
dP
O
CO
0
o
X
Ed
2
•
Q
2
~r
EH
H-
O
cu
*•-
M?
EH
a
cu
number (
:ed.
+->
±) T~l
TJ
(0 (0
"•o
•O-H
o> a
C --H
-H rH
M-l
0) C
•6 4>
o
4J C
C O
0) U
o
U rH
CU rH
CU <
179
-------
APPENDIX B: CONVERSION FROM STANDARD DEVIATION OF LOG
TRANSFORMED DATA TO COEFFICIENT OF VARIATION
181
-------
When fitting models to the log transformed data, the estimated
error variance is for the transformed data. It may be desirable
to convert the variance in the log scale to a coefficient of
variation in the original scale. The following formula relates
the variance of the log data, s2, to the coefficient of variation
of the original measurements, cv:
cv = Vexp(s2)-l (B.I)
For reference, Table B-l tabulates the coefficient of
variation for selected values of s.
183
-------
Table B-l. Coefficient of Variation for the Untransformed Data for
Selected Values of s, the Standard Deviation of the Log
Transformed Data
s
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.00
.000
.100
.202
.307
.417
.533
.658
.795
.947
1.117
1.311
0.02
.020
.120
.223
.328
.439
.557
.685
.824
.979
1.154
1.353
0.04
.040
.141
.243
.350
.462
.582
.711
.854
1.012
1.191
1.396
0.06
.060
.161
.264
.372
.485
.607
.739
.884
1.046
1.230
1.441
0.08
.080
.181
.286
.394
.509
.632
.767
.915
1.081
1.270
1.487
The coefficient of variation is read at the
intersection of the "row" and "column"
defined such that the value in the left
column of the "row" and the top row in the
"column" add to s.
184
-------
APPENDIX C: RECOVERY FROM MULTISPLIT SAMPLES
185
-------
Table C-l. Recovery for Spiked Compounds in Multisplit Samples
Analyzed Using the HRGC/MS Method.
Compound
p,p'-DDE (a)
Beta-BHC
Dieldrin (b)
Heptachlor Epoxide
Oxychlordane (c)
trans-Nonachlor (a)
Hexachlorobenzene
(uncorrected)
Average
Recovery
±95%
Confidence
Interval
26±50%
99±50%
37±61%
50±50%
42±50%
56±50%
41±50%
Average
Recovery
by Spike
Level
23%
42%
14%
201%
32%
64%
43%
-
31%
46%
59%
46%
37%
76%
14%
48%
88%
31%
53%
32%
38%
Spike
Level
(ug/g)
1.00
3.00
5.00
0.10
0.20
0.40
0.20
0.40
0.60
0.10
0.20
0.30
0.10
0.15
0.20
0.10
0.20
0.30
0.05
0.08
0.11
Recovery
Based on
Samples
in Same
Batch
42%
45%
9%
322%
30%
196%
42%
-
62%
47%
45%
42%
42%
78%
23%
96%
104%
38%
56%
41%
35%
Recovery
Based on
Samples in
Different
Batches
5%
39%
19%
81%
34%
-68%
44%
-
—
44%
72%
49%
33%
73%
4%
-1%
72%
24%
49%
23%
41%
(a) Unspiked sample for mid spike level is missing (footnote says
"compound is present but cannot be quantitated" and no LOD was
provided), replaced by 0. As a result of substituting zero for
the missing unspiked concentration, the calculated recovery
will tend to overestimate of the actual recovery.
(b) Unspiked sample for low spike level is below the LOD, replaced
by LOD/2. Between batch recovery for low spike level based on
a single value.
(c) Unspiked sample for low spike level is below the LOD, replaced
by LOD/2.
187
-------
Table C-2. Recovery for Spiked Compounds in Multisplit Samples
Analyzed Using the GC/ECD Method.
Compound
p,p'-DDE
Beta-BHC
Dieldrin
Heptachlor Epoxide
Oxychlordane
trans -Nonachlor
Hexachlorobenzene
(corrected)
Hexachlorobenzene
(uncorrected)
Average
Recovery
±95%
Confidence
Interval
82±19%
89±19%
90±19%
83±19%
73±19%
73±19%
77±19%
53±19%
Average
Recovery
by Spike
Level
64%
98%
82%
85%
91%
92%
88%
95%
88%
75%
99%
76%
73%
83%
64%
45%
104%
69%
80%
100%
52%
55%
63%
41%
Spike
Level
(ug/g)
1.00
3.00
5.00
0.10
0.20
0.40
0.20
0.40
0.60
0.10
0.20
0.30
0.10
0.15
0.20
0.10
0.20
0.30
0.05
0.08
0.11
0.05
0.08
0.11
Recovery
Based on
Samples
in Same
Batch
49%
97%
74%
75%
93%
99%
98%
95%
91%
80%
103%
75%
80%
80%
63%
65%
100%
75%
60%
106%
36%
56%
69%
32%
Recovery
Based on
Samples in
Different
Batches
80%
100%
90%
95%
90%
85%
78%
95%
85%
70%
95%
77%
65%
87%
65%
25%
107%
63%
100%
94%
68%
60%
56%
50%
188
-------
Table C-3. Recovery for Spiked Compounds in Multisplit Samples
Analyzed Using the GC/ECD Method, with Unspiked
Concentrations Measured in Extracts
Compound
p,p'-DDE
Beta-BHC
Dieldrin
Heptachlor Epoxide
Oxychlordane
trans-Nonachlor
Hexachlorobenzene
(corrected)
Hexachlorobenzene
(uncorrected)
Average
Recovery
±95%
Confidence
Interval
85±14%
89±14%
90±14%
84±14%
73±14%
81±14%
82±14%
51±14%
Average
Recovery
by Spike
Level
75%
94%
84%
85%
91%
91%
88%
94%
89%
80%
96%
76%
75%
80%
65%
65%
103%
77%
83%
94%
69%
55%
63%
36%
Spike
Level
(ug/g)
1.00
3.00
5.00
0.10
0.20
0.40
0.20
0.40
0.60
0.10
0.20
0.30
0.10
0.15
0.20
0.10
0.20
0.30
0.05
0.08
0.11
0.05
0.08
0.11
Recovery
Based on
Samples
in Same
Batch
49%
97%
74%
75%
93%
99%
98%
95%
91%
80%
103%
75%
80%
80%
63%
65%
100%
75%
65%
100%
75%
50%
69%
32%
Recovery
Based on
Samples in
Different
Batches
101%
92%
95%
95%
90%
84%
78%
94%
87%
80%
90%
77%
70%
80%
68%
65%
105%
78%
100%
88%
64%
60%
56%
41%
189
-------
190
-------
APPENDIX D: DESCRIPTION OF THE ANALYTICAL PROCEDURES
191
-------
D. 1 Summary of the MOG-GC/ECD Procedure
The modified Mills Olney Gaither (MOG) packed column gas
chromatography/electron capture detector (PGC/ECD) procedure was
used for a number of years as the standard NHATS method for
detecting and quantifying chlorinated pesticides and PCBs in
samples of adipose tissue. The MOG-PGC/ECD procedure has five
major components:
(1) Extraction;
(2) Partitioning;
(3) Florisil clean-up;
(4) Dilution; and
(5) Identification and quantitation.
For the 1984 Comparability Study, the normal extraction step
was modified. Extraction of composite samples was carried out by
MRI. The extracts were split, with one portion reserved for the
MOG-PGC/ECD analysis and another portion reserved for the HRGC/MS
analysis. The MOG-PGC/ECD procedure in the 1984 Comparability
Study is described below. This description is for single-split
composite samples, the multisplit composite samples, the extracts
of the multisplit composite samples, and the regenerated composite
samples and their method blanks.
In the extraction step, a 25 gram composite was placed in a
culture tube. Next, 10 milliliters (mL) of methylene chloride
were added to the culture tube, and the entire contents of the
tube were blended in a Tekmar tissuemizer. The blended mixture
was allowed to separate and the methylene chloride layer was
transferred to a filter funnel containing glass wool and sodium
sulfate. The mixture resulting after eluting through the filter
funnel was collected in a 100 mL flask.
193
-------
Another 10 mL of methylene chloride were added to the
sediment, and the process of blending, separating, transferring to
funnel, and storing in flask was repeated. The addition of 10 mL
of methylene chloride were repeated one or two more times.
The culture tube was rinsed with methylene chloride, the
rinse mixture added to the funnel, the funnel rinsed, and the
final output from the funnel was added to the 100 mL flask. A
sufficient amount of methylene chloride was added to the flask to
bring the volume of the mixture in the flask to 100 mL.
For some samples, particles remained in the mixture in the
flask. These samples were refiltrated, and the final volume for
these samples was 200 mL.
The percentage of the composite sample that was lipid tissue
was determined as follows. A glass vial was weighed to the
nearest 0.0001 gram. One mL of the final mixture was transferred
to the vial. The volume of the mixture in the vial was reduced
until only an oily residue was left. The vial was again weighed
to the nearest 0.0001 gram, and the percentage of the composite
sample that was lipid material was calculated.
Of the remaining volume for the sample, approximately 20% was
designated for the MOG-PGC/ECD procedure, with the remainder
designated for the HRGC/MS procedure. The appropriate volumes
were concentrated to remove most of the methylene chloride. These
concentrated volumes were transferred to vials for shipment to
CSU.
After receipt at CSU, 1 microgram of the internal standard,
aldrin, was added or "spiked" to the sample designated for the
MOG-PGC/ECD method. The sample was then ready for the next step
in the procedure.
194
-------
In the partitioning step, the sample was transferred to a 125
mL separatory funnel by rinsing with hexane. Next, hexane and
acetonitrile were mixed together in a separate container. About
30 mL of the hexane/acetonitrile mixture were added to the
separatory funnel. The funnel was shaken for 2 minutes, and
layers were allowed to separate. The lower layer was then drained
into a 1-liter separator to which 550 mL of a 2% sodium sulfate
solution and 100 mL of hexane had previously been added. The
process of adding 30 mL of hexane/acetonitrile mixture to the
funnel, shaking the funnel for 2 minutes, allowing layers to
separate, and draining the lower layer into the 1-liter separator
was repeated three more times.
Next, the 1-liter separator was inverted, and layers were
allowed to separate. The lower layer was discarded. The contents
of the 1-liter separator were washed twice with 100 mL of 2%
sodium sulfate solution. After separation of layers in the
1-liter separator, the lowest layer was discarded. Excess water
was drained. The remaining contents of the 1-liter separator were
transferred to a 500 mL boiling flask. The contents of this flask
were reduced to 3 to 5 mL through rapid evaporation.
The third step was Florisil cleanup. A reservoir column was
prepared with one half to 1 inch of sodium sulfate on the bottom,
4 inches of Florisil in the middle, and an inch and a half of
sodium sulfate on the top. A solution of 6% diethyl ether in
hexane and a solution of 15% diethyl ether in hexane were
prepared.
Approximately 100 mL of hexane was added to the column and
allowed to flow through the column. This "rinse" hexane was
discarded. A 500 mL boiling flask labeled "6%" was placed under
the column to collect the output of the column. Then the 3 to 5
mL from the partitioning step was added to the column. Next, 200
mL of the 6% solution mix were added to the column.
195
-------
At this point, a 500 mL boiling flask labeled "15%" was
prepared with 0.2 micrograms of aldrin. When the 6% solution mix
reached the sodium sulfate at the top of the column, the 6%
receiver boiling flask was replaced with the 15% boiling flask.
Two hundred mL of the 15 percent solution mix were added to the
column, and the column was allowed to drain into the receiver
flask labeled 15%.
The contents of both the 6% receiver flask and the 15%
receiver flask were reduced to 3 to 5 mL through rapid
evaporation.
In the dilution step, the contents of the 6% and the 15%
flasks were transferred to separate centrifuge tubes. A volume of
hexane was added to each centrifuge tube in accordance with an
established algorithm.
In the identification and guantitation step, a 5-microliter
sample from each centrifuge tube was injected separately into a
gas chromatograph. Compounds were identified by their retention
time relative to the chromatographic "peak" representing aldrin.
Concentrations of compounds were quantitated based on the area of
the peak representing the compound.
The limits of quantification and the limits of detection for
the MOG-PGC/ECD method have been established from years of
experience with the method. For samples for which quantitated
concentrations were less than the limit of detection, a final
concentration of zero was reported. For samples for which the
quantitated concentration was between the limit of detection and
the limit of quantification, the limit of quantification was
reported as the final concentration of the sample. The label
"Trace" was assigned in this case. For all other samples, the
quantitated concentration was reported as the final concentration.
196
-------
The recovery of aldrin was calculated as a quality assurance
step on the entire process. However, this recovery was not used
to correct the compound concentrations. Concentrations for
hexachlorobenzene and, in some cases, mirex and p,p'-DDT were
computed on both a corrected and an uncorrected basis. The
correction factors were based on historical information on the
recovery of these three compounds at the partitioning step. The
concentrations of the other chemicals were not corrected for
recovery.
D . 2 Summary of the HRGC/MS Procedure
In order to expand the list of chemicals that could be
monitored by the NHATS, the standard NHATS method for detection
and quantification of chemical compounds was changed from MOG-
PGC/ECD to High Resolution Gas Chromatography/Mass Spectrometry
(HRGC/MS). The HRGC/MS procedure has five major components:
(1) Extraction;
(2) Gel permeation chromatography;
(3) Florisil clean-up;
(4) Addition of internal standards; and
(5) Identification and quantification.
The HRGC/MS procedure used in the 1984 Comparability Study is
described below. The description is for single-split composite
samples. The same basic procedure, with appropriate
modifications, was followed for the method blanks, the quality
control samples, the multisplit composite samples, and the
regenerated composite samples and their method blanks.
The HRGC/MS procedure began with the extraction step
described for the MOG-PGC/ECD procedure. A volume designated for
the HRGC/MS procedure was placed in a vial and shipped to CSU.
197
-------
After receipt at CSU, the sample volume was spiked with a
known amount of the 11 surrogate compounds. The extraction step
was continued as follows.
Twenty mL of methylene chloride were added to the sample, and
the sample and the methylene chloride were transferred to a
culture tube. The rest of the extraction step followed the same
process as described in the extraction step for the MOG-PGC/ECD
procedure. Ten mL of methylene chloride were added to the culture
tube, and the entire contents of the tube were blended in a Tekmar
tissuemizer. The blended mixture was allowed to separate into a
sediment and a methylene chloride layer. The methylene chloride
layer was transferred to a filter funnel containing glass wool and
sodium sulfate. The mixture resulting from the filter funnel was
stored in a 100-mL flask.
Another 10 mL of methylene chloride were added to the
sediment, and the process of blending, separating, transferring to
funnel, and storing in flask was repeated. The addition of 10 mL
of methylene chloride and the subsequent steps were repeated 1 or
2 more times.
The culture tube was rinsed with methylene chloride, the
rinse mixture added to the funnel, the funnel rinsed, and the
final output from the funnel was added to the 100-mL flask. A
sufficient amount of methylene chloride was added to the flask to
bring the volume of the mixture in the flask to 100 mL.
The mixture in the flask was transferred to a 250-mL
evaporator. The volume in the evaporator was reduced to a volume
of 10 to 20 mL. The contents of the evaporator were transferred
to a vial. A sufficient amount of methylene chloride was added to
the vial so that the vial contained about 0.25 grams of lipid
material for each mL of mixture. The mixture in the vial was the
final extract, and this extract was passed to the next step.
198
-------
The next step in the procedure was gel permeation
chromatography. The purpose of this step was to separate the
target analytes from the lipid material. A column of Bio-Beads
SX-3 in dichloromethane was prepared.
The sample extracts were filtered to remove particles that
might interfere with the flow of the column. Then the extracts
were transferred to the column in successive aliquots. The output
of the column was collected in amber bottles sized 1 to 4 liters.
The contents of the amber bottles were transferred to an
evaporator, and reduced to a volume of about 10 mL. Next, 50 mL
of hexane were added to the evaporator. Then the volume of the
evaporator was reduced, first to 10 mL and then to 1 mL.
If the 1 mL volume was unusually colored or viscous, 2 to 5
mL of methylene chloride were added, and the resulting volume of 3
to 5 mL was injected into the column. The output of the column
was collected in amber bottles, placed in the evaporator, and
again reduced to 1 mL.
A Florisil column was prepared by placing glass wool at the
bottom of the column, adding 100 mL of hexane to the column, then
adding 12.5 grams of Florisil, and finally placing a one-half inch
layer of sodium sulfate on top of the Florisil. The hexane was
drained so that the top of the hexane was level with the top of
the sodium sulfate.
The 1 mL mixture from the previous step was transferred to
the top of the column. A 500 mL flask was placed underneath the
column. The column was drained until the sodium sulfate layer was
almost exposed. At that point, 200 mL of a 6% solution of ethyl
ether in hexane were added to the column at a rate of
approximately 5 mL per minute. When the sodium sulfate layer was
almost exposed, the flask underneath the column was replaced with
a second 500 mL flask, and 200 mL of a 15% solution of ethyl ether
199
-------
in hexane were added to the column. When the sodium sulfate layer
was almost exposed, 200 mL of a 50% solution of ethyl ether in
hexane were added to the column, without any change in the flask
under the column.
After the column drained into the second collector flask, the
output of the column was contained in two separate 500 mL flasks.
The contents of the first collector flask are referred to as the
"first fraction", and the contents of the second flask are
referred to as the "second fraction". The first and second
fractions were concentrated separately, first to 5 to 10 mL, then
to less than 1 mL, and then to less than one-half mL. The final
volume of each fraction was stored in two separate vials with
Teflon-lined screw tops. The vials were stored at 4 degrees
centigrade.
The volumes for the two fractions were each reduced to 0.2
mL. Known amounts of the three internal standards, anthracene-
dlO, naphthalene-d8, and benzo (a) anthracene-d!2, were added to
each of the fractions. A 1- to 2- microliter aliquot of each
fraction was injected into the gas chromatograph/mass
spectrometer.
The final step was identification and quantification. There
were 57 target analytes for the HRGC/MS analysis. Associated with
each of the 57 were a primary mass fragment and two secondary mass
fragments. In addition, one of the three internal standards was
designated as the appropriate internal standard for each target
analyte for purposes of identification and quantification.
In order for an analyte to be identified, the following four
criteria had to be satisfied:
(1) The primary and secondary masses had to achieve their
maximum values within a specified time span;
(2) The retention time of the primary and secondary mass
fragments relative to the designated internal standard
200
-------
had to be within 10 seconds of the known relative
retention time of the analyte;
(3) The relative abundances of the primary and secondary
masses all had to be within 20% of the relative
abundances in the reference spectrum of the analyte;
(4) The abundances of the primary and secondary masses all
had to exceed three times the background signal to noise
ratio.
Quantitation was carried out as follows. The
chromatograph/mass spectrometer was calibrated. In calibration,
known amounts of the target analyte and a known amount of the
internal standard were injected into the instrument. The
calibration was done with five different amounts of the target
analyte. A relative response factor was calculated for each
analyte from the calibration data for use in quantitation. The
lower calibration limit was also used in quantitation, as
described below.
For a composite sample, the amount of the target analyte
present was calculated from the abundance of the primary mass of
the analyte, the abundance of the primary mass of the internal
standard, the amount of the internal standard, and the relative
response factor. This calculated amount was given a label of
"Positive Quantifiable" if the abundances of the primary and
secondary masses all exceeded 10 times the background signal to
noise ratio and the lower calibration limit was exceeded. A label
of "Trace" was to be assigned to an amount for which 1) the
primary and secondary masses were all above three times the signal
to noise ratio, 2) one of the primary or secondary masses was
between 3 and 10 times the background signal to noise ratio, and
3) the lower calibration limit was exceeded.
For samples for which 1) at least one of the mass abundances
was less than three times the signal to noise ratio and 2) the
lower calibration limit was exceeded, a label of "Not Detected"
was assigned. The detection limit was the maximum of the lower
201
-------
calibration limit and three times the signal to noise ratio. For
samples for which there was no response or the response was less
than the lower calibration limit, a label of "Not Detected" was
assigned. The detection limit was the lower calibration limit.
For samples for which a response above the lower calibration limit
was observed, but the identification criteria were not met, a
remark was made indicating an interference was present. The
detection limit was quantitated based on the response observed.
Cases not covered above were handled by inclusion of written
remarks indicating what was observed.
Concentrations of analytes were computed from the calculated
amounts, the weight of the composite sample, and the percent lipid
in the composite sample. No concentrations were corrected for
recovery. The recoveries of the surrogate compounds were
calculated as a check on method performance.
202
-------
APPENDIX E: DISCUSSION OF THE VARIANCE COMPONENTS
203
-------
204
-------
In order to provide a more intuitive understanding of the
variance components, this appendix discusses the various terms in
the model in the context of the lab procedures which might have
contributed to each term. This presentation is not meant to imply
that the possible causes of error are limited to those discussed
here or that the those discussed here are more important than
other sources of error.
The discussion below assumes the following simplified
procedure for the HRGC/MS measurement method:
The analysis procedures, laboratory equipment, and solutions
are prepared for each batch, including the internal standards
spiking solution with three internal standards. The HRGC/MS
equipment is calibrated using a set of six dilutions of a
standard calibration solution. From the calibration data, a
relative response factor (RRF) is calculated for each
compound. The RRF relates the target compound response to
the internal standard response.
The sample preparation involves measuring the wet weight and
percent lipid followed by several processing steps which move
the target compounds from the tissue sample to the final
extract and adjust the volume of the final extract to 200 uL.
A measured amount of the internal standard solution is spiked
into the final extract before a portion of the final extract
is injected into the HRGC/MS.
The primary output from the HRGC/MS is a response trace with
multiple peaks. The area below the internal standard peak,
the area below the peak for the target compound, the wet
weight and percent lipid, and the relative response factor
are used to quantitate the amount of each target compound
(see equation (3.3)).
Measurement errors can result from these processing steps due
to small variations in: (1) the measurement of weights and volumes
or the area under a mass spectrometer peak (due partly to
uncertainty in assessing the background response); (2) the
rounding in calculation results; and (3) the equipment setup and
sample processing steps (possibly due to age of the reagents, the
equipment condition, or the time spent on the processing of each
205
-------
sample). Large variations due to measuring, processing, or
calculation errors will also contribute to the variance
components.
The model assumes that the measurements for each compound (on
a log scale) vary around a constant equal to the product of the
true concentration and the method recovery, Rmct^ which are
constant for each compound.
liUXjnobsi) =ln(CcsRmct) + error (E.I)
Variation of the method recoveries from year to year or lab to lab
are not considered. Variation or measurement error around this
expected value can be separated into the components shown in the
model. Table E-l provides a description of each component and an
example of processing steps which might contribute to each
component.
206
------- |