S3MIVOLATIItE
ORGANIC
COMPOUNDS
IN
THE
OEWERAL
U.
S.
POPULATION:
"
ATS
FY86
RESULTS
VOLUME
I
This
work
was
supported
by
the
U.
S.
Environmental
Protection
Agency
under
EPA
contract
numbers
68­
02­
4294,
68­
D8­
0115,
68­
DO­
0126,
68­
D2­
0139,
68­
02­
4252,
68­
02­
4293,
and
68­
D9­
0174
Prepared
for
Khoan
Dinh,
Work
Assignment
Manager
Technical
Programs
Branch
Chemical
Management
Division
Office
of
Pollution
Prevention
and
Toxics
U.
S.
Environmental
Protection
Agency
Washington,
DC
20460
July
1994
The
material
in
this
document
has
been
subject
to
Agency
technical
and
policy
review
and
approved
for
publication
as
an
EPA
report.
The
views
expressed
by
individual
authors,
however,
are
their
own
and
do
not
necessarily
reflect
those
of
the
U.
S.
Environmental
Protection
Agency.
Mention
of
trade
names,
products,
or
seryices
does
not
convey,
and
should
not
be
interpreted
as
conveying,
official
EPA
approval,
endorsement,
or
recommendation.
PREFACE
The
determination
of
the
levels
of
semivolatile
organic
compounds
in
the
general
population
of
the
United
States
described
in
this
report
was
achieved
through
cooperative
efforts
of
many
EPA
and
contract
support
staff.
EPA
staff
participating
in
the
program
included
principal
investigators
from
the
Technical
Programs
Branch
(
TPB)
of
the
Chemical
Management
Division
(
0)
of
the
Office
of
Pollution
Prevention
and
Toxics
(
OPPT).
Contract
support
to
OPPT
was
provided
by:

0
Battelle
under
EPA
Contract
Nos.
68­
02­
4294,68­
D8­
0115,
68­
DO­
0126,
and
68­
02­
0139.

Midwest
Research
Institute
(
MRI)
under
EPA
Contract
No.
68­
02­
4252.

0
Westat,
Inc.,
under
EPA
Contract
Nos.
68­
02­
4293and
68­
D9­
0174.

The
roles
and
responsibilities
of
each
of
these
organizations
and
key
individuals
participating
in
this
effort
are
presented
below.

Battelle
Battelle
was
responsible
for
developing
the
FY86
hTS
specimen
collection
program,
creating
and
maintaining
the
data
bases
on
the
Patient
Summary
Reports,
designing
the
specimen
compositing
plan
and
the
statistical
methodology
for
data
analysis,
conducting
the
statistical
analysis
to
develop
estimates
of
semivolatile.
re.
sidua1levels
in
the
general
U.
S.
population
based
on
demographic
factors,
and
producing
this
final
report.
Key
individuals
included:
Dr.
Robert
Lordo,
Dr.
John
Orban,
Mr.
Ying­
Liang
Chou,
Ms.
Pamela
Hartford,
and
Ms.
Tamara
Collins.

Midwest
Research
Institute
(
MRIL
MRI
was
responsible
for
the
coordination
of
the
collection
of
the
FY86
"
ATS
specimens,
preparation
of
the
"
ATS
composites
and
quality
control
(
QC)
samples,
conducting
the
HRGC/
MS
analysis
of
the
composites,
reporting
the
results,
and
contributing
to
this
final
report.
Key
individuals
included:
Dr.
John
Stanley,
Dr.
Stan
Spurlin,
Mr.
Jack
Balsinger,
Ms.
Hope
Green,
and
Ms.
Patti
Alm.

iii
Westat.
Inc.

Westat
was
responsible
for
creating
and­
maintainingthe
data
bases
fur
the
Analysis
Reports,
developing
and
executing
statistical
procedures
for
identifying
outliers
in
the
reported
concentrations,
and
writing
the
final
report
on
the
results
of
the
outlier
analysis.
Key
individuals
included:
Mr.
John
Rogers
and
Ms.
Helen
Powell.

EPA/
OPPT
EPA/
QPPT
was
responsible
for
oversight
in
the
development
of
the
study
plan,
managing
and
coordinating
the
conduct
of
the
overall
study,
and
reviewing,
editing
and
finalization
of
this
report.
Key
individuals
included:
Dr.
moan
Dinh,
Ms.
Janet
Remmers,
and
Mr.
John
Schwemberger
as
Work
Assignment
Managers
and
Dr.
Joseph
Breen,
Ms.
Edith
Sterrett,
Mr.
Gary
Grindstaff,
and
Mr.
Philip
Robinson
as
Project
Officers.

iv
TABLE
OF
CONTENTS
Volume
I
Pase
gxgcuTIvEsuMMARY
...................
.
0'
xvii
1.0
INTRODUCTION
...................
..
.
1­
1
1.1
Background
.................
..
.
1­
1
1.2
Objectives
.................
..
.
1­
3
1..
3
Report
Organization
..............
..
.
1­
3
2.0
NHATS
FY86
SAMPLg
DSSIGN
.............
..
.
2­
1
2.1
Sampling
Design
...............
.
.
I
.
2­
1
2.1.1
The
"
ATS
Stratification
Scheme
.
.
..
.
2­
3
2.1.2
MSA
Selection
............
..
.
2­
3
2.1.3
Specimen
Collection
Quotas
......
..
.
2­
6
2.2
Sample
Collection
Procedures
........
..
.
2­
9
2.3
Specimen
Collection
Summary
.........
..
2­
11
3.0
NEATS
FY86
COMPOSITE
DESIGN
...........
..
.
3­
1
3.1
Design
Goals
and
Compositing
Criteria
....
..
.
3­
1
3.2
Laboratory
Compositing
Procedures
......
..
.
3­
4
3.3
Summary
of
FY86
"
HATS
Composite
SampXes
...
..
.
3­
6
4.0
CHEMISTRY
.....................
..
.
4­
1
4.1
Analytical
Procedures
.............
.*
.
4­
1
4.1.1
Sample
Preparation
..........
..
­
4­
1
4.1.1.1
Extraction
.........
*.
.
4­
3
4.1.1.2
Lipid
Determination
.....
..
.
4­
3
4.1.1.3
Extract
Concentration
....
..
.
4­
3
4.1.2
Cleanup
Procedure
..........
..
.
4­
4
4.1.2.1
Gel
Permeation
Chromatography
..
.
4­
4
4.1.2.2
GPC
Eluent
Concentration
.
.
..
.
4­
5
4.1.2.3
Florisil
Column
Cleanup
...
..
.
4­
5
4.1.3
Analysis
Procedures
.........
..
.
4­
7
V
..
TABLE
OF
CONTENTS
.
(
cont.)

Volume
I
(
coat.)

4.1.4
Quantitation/
Data
Reduction
........

4.1.4.1
Qualitative
Identification
...
4.1.4.2
Quantitation
..........
4.1.4.3
Recovery
of
Surrogate
Standards
.
Qr1.4.4
Data
Qualifiers
.........
4.1.4.5
Estimating
the
Method
Limit
of
Detection
............

4.2
QA/
QC
for
Chemical
Analysis
...........

4.2.1
Demonstrating
Achievement
of
Instrument
performance
Requirements
..........
4.2.2
Calibration
.
forQuantitative
Semivolatile
Analysis
..........

4.2.2.1
Initial
Calibration
.......
4.2.2.2
Routine
Calibrations
......

4.2.3
Spiking
Solution
Preparation
.......

4.2.3.1
Native
Standard
Spiking
Solution
Paue
.
4­
8
.
4­
8
4­
14
4­
15
4.15.

4­
16
4­
17
4­
17'

4­
19
4­
19
4­
25
4­
25
.............
4­
25
4.2.3.2
Surrogate
Standard
Spiking
Solution
............
4.2.3.3
Internal
Standard
Spiking
Solution
............
4.2.3.4
Performance
Audit
Solutions
...

4.2.4
QC
Samples
................

4.2.4.1
Method
Blanks
..........
4.2.4.2
Control
Samples
.........
4.2.4.3
Spiked
Control
Samples
.....

4.3
Overall
Data
Quality
..............
4­
25
4­
28
4­
28
4­
28
4­
28
4­
30
4­
30
4­
31
5.0
DATAISSUES
......................
5­
1
5.1
Determining
Native
Compounds
to
Enclude
in
Statistical
Analysis
..............
5­
2
5.1.1
Detection
Status
of
the
Semivolatiles
...
5­
3
5.1.2
Data
Reporting
Unique
to
Dieldrin
and
p.
p.
DDE
..........
5­
8
vi
.
*

.
.
.
.
­.

a
.
.
.
.
.
.
­.
.
.

.
.
­.
­.

.
.
.
.
.
*

­.
Pase
.
5­
9
5­
10
5­
11
5­
14
5­
19
5­
22
5­
22
5­
30
5­
30
5­
30
5.­
30
5­
36
5­
36
5­
36
5­
45
5­
48
.
.
­
6­
1
­.
.
6­
2
.
.
.
6­
7
.
.
.
6­
7
.
.
.
6­
7
.
.
6­
10
.
.
6­
12
.
.
.
7­
1
.
.
.
7­
2
TABLE
OF
CONTENTS
(
coat.)

Volume
I
tcont.)

Page
7.2
Population
Estimates
from
Statistical
Modelling.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7­
8
7.3
Hypothesis
Testing
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7­
27
7.4
Outlier
Detection
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7­
29
7.5
Model
Validation
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
7­
32
8.0
COMPARISON
WITH
RESmjTS
FROM
PREVIOUS
SURVEYS
INTHENHATSPROGRAM..
­.
­..
­.
.
.
.
.
.8­
1
8.1
Cornparison
of
Design
and
Analytical
Procedures
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8­
2'

8.1.1
Comparison
of
Study
Designs
.
.
.
.
.
.
.
.
8­
2
8.1.2
Comparison
of
Analytical
Procedures
.
.
.
.
8­
6
8.2
LODs
and
Percent
Detection
Summaries
.
.
.
.
.
.
8­
10
8.3
Descriptive
Statistics
on
Measured
Concentrations
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
8­
15
8.3.1
Scatterplots
of
the
Sample
Concentrations
.
­.
.
.
.
.
.
.
.
.
.
.
8­
16
8.3.2
Unweighted
National
Averages
.
.
.
.
.
.
.
8­
17
8.3.3
Weighted
National
Averages
.
.
:
.
.
.
.
.
8­
18
8.4
Statistical
Comparison
of
National
Concentration
Estimates
.
.
.
.
.
e
.
.
8­
26
8.4.1
Semivolatile
Compounds
Included
in
Statistical
Comparison
.
.
.
.
.
.
.
.
.
.
8­
27
8.4.2
Fitting
the
Additive
Model
.
.
.
.
.
.
.
.
8­
28
8.4.2.1
National
Estimates
.
.
.
.
.
.
.
8­
29
8.4.2.2
Marginal
Estimates
.
.
.
.
.
.
.
8­
31

8.4.2.3
Likelihood
Ratio
Tests
:.
.
.
.
8­
34
.
8.4.2.4
Conclusions
.
.
.
.
.
.
.
.
.
.
.
8­
34
9.0
REFERENCES
.
.
­.
.
.
­.
­.
.
.
.
.
.
.
.
.
.
.
.
.
9­
1
Volume
I1
Appendix
A:
Listing
of
"
ATS
FY86
Composite
Data
%

viii
TABLE
OF
CONTENTS
(
cont.)

Volume
I1
(
cont.)

Listing
of
NHATS
FY86
QC
Data
For
Compounds
Detected
in
At
Least
50%
of
Composites
(
plus
octachlorobiphenyl)

Plots
of
Observed
VS.
Spiked
Concentrations
for
Spiked
Semivolatile
Compounds
Detected
in
At
Least
50%
of
Composites
(
plus
Octachlorobiphenyl)
.

Plots
of
Observed
Concentration
vs.
Batch
ID
for
Unspiked
Semivolatile
Compounds
Detected
in
At
Least
50%
of
Composites
Summary
of
QC
Data
for
the
FY86
"
ATS
Spiked
Semivolatile
Compounds
Not
on
the
Target
List
for
Statistical
Analysis
Characterizing
the
Distribution
of
"
HATS
FY86
Semivolatile
Compound
Goncentrations
(
inng/
g,
Unadjusted
for
Surrogate
Recoveries)
Based
on
All
50
Composite
Samples
Estimates
of
Average
Concentrations
with
95%
Confidence
Intervals
As
Determined
by
the
Additive
Model,
for
Selected
Subpopulations
in
the
FY86
"
HATS
Concentrations
(
ng/
g)
from
the
FY82,
FY84,
and
FY86
"
ATS,
Plotted
in
Chronological
Order
with
Age
Group
as
the
Plotting
Symbol
Surrogate­
Adjusted
Concentrations
(
ng/
g)
from
the
FY82,
FY84,
and
FY86
"
ATS,
Plotted
in
Chronological
Order
with
Age
Group
as
the
Plotting
Symbol
Arithmetic
Averages
(
and
Standard
Errors)
of
Extractable
Lipid
Concentrations
(
ng/
g)
for
Compounds
Analyzed
in
the
FY86
"
HATS
and
Also
Analyzed
in
the
FY82
and/
or
FY84
"
ATS
Comparisons
of
Predicted
Average
Concentrations
(
ng/
g)
by
Selected
Subpopulations
for
Selected
Semivolatiles
over
the
FY82,
FY84,
and
FY86
NHATS
Plots
of
Estimates
Average
Concentrations
(
ng/
g)
by
Census
Region
and
by
Age
Group,
Plus
and
Minus
Two
Standard
Errors,
Based
on
the
Additive
Model
for
the
FY82,
FY84,
and
FY86
"
ATS
I
LIST
OF
TABLES
Volume
I
Pase
Table
ES­
1
Semivolatile
Compounds
Detected
in
At
Least
44%
of
the
FY86
NHATS
Composite
Samples
.....
xx
Table
kS­
2
Estimates
of
National
Average
Concentrations
for
Selected
Semivolatiles,
with
95%
Confidence
Intervals,
from
the
FY86
MIATS
...
xxii
Table
ES­
3
Estimates
of
National
Average
Concentrations
for
Selected
Semivolatiles,
with
95%
Confidence
Intervals,
from
the
FY82,
FY84,
and
FY86
NHATS
.............
xxvi
Table
2­
1
Demographic
Categories
in
Which
Subquotas
Were
Established
for
Collecting
Adipose
Tissue
Specimens
............
2­
2
Table
2­
2
Sampling
Strata
Definitions
for
the
"
HATS
....
2­
4
Table
2­
3
Sample
MSAs
Selected
for
the
FY86
NHATS
.....
2­
7
Table
2­
4
FY86
Age
and
Sex
Subquotas,
and
the
Race
Subquota,
far
Each
NHATS
Collection
Site
Within
Each
Stratum
..............
2­
10
Table
2­
5
FY86
NHATS
Specimen
Collection
Summary
....
2­
13
Table
2­
6
FY86
NHATS
Speciment
Collection
Summary
by
Demographic
Subpopulation
.........
2­
14
Table
3­
1
Distribution
of
FY86
"
ATS
Composite
Samples
by
Census
Division
and
Age
Group
........
3­
7
Table
3­
2
Demographic
Makeup
of
FY86
NHATS
Composite
Samples
................
3­
8
Table
4­
1
Recommended
HRGC/
MS
Operating
Procedures
....
4­
8
Table
4­
2
Characteristic
Masses
and
Intensities
for
the
Qualitative
Identification
of
the
Semivolatile
Target
Analytes,
Chromatographic
Conditions,
and
Estimated
Limit
of
Detection
........
4­
9
Table
4­
3
DFTTP
Key
Masses
and
Abundance
Criteria
....
4­
18
Table
4­
4
Calibration
Solutions
for
the
6%
Florisil
Fraction
................
4­
20
X
­
­
LIST
OF
TABLES
(
cont.)

Volume
I
(
cont.)

Pase
___
4­
5
Calibration
Solutions
for
the
50%
d
Florisil
Fraction
...............
4­
23
ble
4­
6
Calibration
Solutions
for
PCB
Analysis
....
4­
24
­
le
4­
7
Proposed
QC
Spiking
Solutions
..........
4­
26
hle
4­
8
Spike
Levels
for
Surrogate
and
Internal
Standards
.....
.....
.
.
4­
29
#
e
4­
9
Quality
Control
Samples
Included
in
the
FY86
"
ATS
Analytical
Procedure
........
4­
30
sle
4­
10
Data
Quality
Objectives
for
the
FY86
NHATS,

h>
Along
With
Actual
Performance
.........
4­
32
>
le
5­
1
Percent
of
"
ATS
FY86
Composite
Samples
in
Each
Detection
Level
Cat.
egory
..
..
5­
4
&
le
5­
2
Matching
NHATS
FY86
Native
Compounds
with
Surrogate
Compounds
.
..
...
..
5­
1'
2
&
le
5­
3
Estimates
of
R
and
A
for
Surrogate
..
2:'
Compounds
....................
5­
15
.
e
..&
!
able
5­
4
Spiked
Target.
Compounds
for
the
FY86
NHATS,
$:­.
.
­
withSpiking
Levels
..............
5­
21
'
1
.

le
5­
5
Summary
of
(
Surrogate­
Adjusted)
QC
Data
for
the
FY86
"
HATS
Spiked
T'argetCompounds
5­
23
...%:
Table5­
6
Percent
Recoveries
for
Spiked
Target
Compounds,
as
Determined
'
fromTwo
'\
I.;.
1.,*..'
....
Calculation
Methods
.........
­..
­­
5­
29
C.
,'$*
kz&&
>
I
$
t~,$?..
#,

%
i.
i.',.
X..
,
Table
5­
7
Means
and
Standard
Errors
of
Surrogate
z?.$.
.:
:

Adjusted
Concentrations
(
ns
f
Unspiked
Target
Compounds
for
QC
Samples
(
by
Batch
and
Overall)
.................
5­
31
able
5­
8
Batch
Analysis
Results
on
Method
Blanks
and
Control
Samples
for
Compounds
Detected
in
At
Least
50%
of
Composites,
Where
the
Compound
Was
Detected
in
the
Method
Blank
...
5­
32
­

xi
j.
i
,
.
'.
,
..>
I
I,
.
I
LIST
OF
TABLES
(
cont.)

Pase
Table
5­
9
Regression
Models
Used
to
Analyze
NHATS
FY86
QC
Data
for
Spiked
Compounds
........
5­
34
Table
5­
10
Estimated
Batch
Recoveries
and
Average
Recovery
for
Spiked
Compounds
with
Percent
Detected
At
Least
50%
(
Adjusted
Data)
.....
5­
37
Table
5­
11
Tests
for
Significant
Differences
in
Batch
Slopes
Among
Selected
Batches
for
Spiked
Target
Compounds
..........
5­
39
Table
5­
12
Predicted
Concentrations
and
Coefficients
of
Variation
at
Each
Spike
Level
for
Spiked
Target
Compounds
Analyzed
by
the
Batch
Slopes
Model
..............
5­
40
Table
5­
13
Estimated
Batch
Background
Levels
and
Average
Background
Level
for
the
Two
Methods
of
Reporting
p,
p­
DDE
Concentrations,
as
Estimated
by
the
Batch
Intercepts
Model
....
5­
44
Table
5­
14
Results
of
Statistical
Analysis
of
QC
Data
on
Unspiked
Target
Compounds
.........
5­
46
Table
5­
15
Predicted
Concentrations
and
Coefficients
of
Variation
for
Unspiked
Target
Compounds
at
the
Control
Level
.............
5­
47
Table
6­
1
"
ATS
Analysis
Factors
and
Categories
......
6­
3
Table
7­
1
Descriptive
Statistics
of
NHATS
FY86
Semivolatile
Compound
Concentrations
Based
on
All
50
Composite
Samples
........
7­
3
Table
7­
2
Estimates
of
Average
Concentrations
for
Selected
Semivolatiles,
with
Standard
Errors
and
Approximate
95%
Confidence
Intervals,
According
to
Census
Region
from
"
ATS
FY86
Composite
Samples
...............
7­
10
Table
7­
3
Estimates
of
Average
Concentrations
for
Selected
Semivolatiles,
with
Standard
Errors
and
Approximate
95%
Confidence
Intervals,
According
to
Age
Group
from
NHATS
FY86
Composite
Samples
...............
7­
14
xii
BEST
COPY
AVAILABLE
LIST
OF
TABLES
(
cont.)

Volume
I
4cont.
L
Pase
'­
4
Estimates
of
Average
Concentrations
for
Selected
Semivolatiles,
with
Standard
Errors
and
Approximate
95%
Confidence
Intervals,
According
to
Race
Group
from
"
ATS
FY86
Composite
Samples
.............
7­
17
I­
5
Estimates
of
Average
Concentrations
for
Selected
Semivolatiles,
with
Standard
Errors
and
Approximate
95%
Confidence
Intervals,
According
to
Sex
Group
from
"
ATS
FY86Composite
Samples
.............
7­
20
­
6
Estimates
of
Average
Concentrations
for
and
Approximate
95%
Confidence
Intervals,
for
the
Nation
from
NHATS
FY86Composite
Samples
.............
7­
23
­
7
Chlorobiphenyl
Distribution
Across
the
Five
Target
PCB
Homologs
in
the
FY86
"
ATS
........
;
.......
7­
26
­
8
Significance
Levels
from
Hypothesis
Tests
for
Differences
Between
Demographic
Groups
for
"
ATS
FY86
Semivolatiles
.......
7­
28
­
9
Measured
Concentrations
with
High
Influence
on
Determining
the
Additive
Model
Fit
...
7­
34
Selected
Semivolatiles,
with
Standard
Errors
­
10
R­
Squared
Correlation
Between
Observed
Concentrations
and
Concentrations
Predicted
by
the
Additive
Model
for
NHATS
FY86
Semivolatiles
...............
7­
36
­
1
Number
of
Specimens
and
Composites
Within
the
FY82,
FY84,
and
FY86
"
ATS
According
to
MSA
..................
.
8­
3
­
2
Total
Number
of
Specimens
Included
in
Composite
Samples
Analyzed
in
the
FY82,
FY84,
and
FY86
"
ATS,
by
Subpopulation
and
Across
the
Entire
Study
..........
.
8­
4
­
3
Total
Number
of
Composite
Samples
Analyzed
in
the
FY82,
FY84,
and
FY86
"
ATS,
by
Subpopulation
and
Across
the
Entire
Study
.
8­
5
xiii
LIST
OF
TABLES
(
cont.)

Table
8­
4
Table
.8­
5
Table
8­
6
\

Table
8­
7
Table
8­
8
Table
8­
9
Table
8­
10
Table
8­
11
Appendix
A:
Volume
I
tcont.
1
Semivolatile
Compounds
Quantitated
Using
the
Same
Internal
Quantitation
Standards
(
IQS)
in
NHATS
FY84
and
FY86
.........

Semivolatile
Compounds
Quantitated
Using
Different
Internal
Quantitation
Standards
(
IQS)
in
NHATS
FY84
and
FY86
.........

Average
Lipid­
Adjusted
Limit
of
Detection
(
LOD,
ng/
g)
and
Percent
of
Composites
with
Detected
Concentrations,
for
Compounds
Analyzed
in
the
M86
NHATS
and
Also
Analyzed
in
the
FY82
and/
or
FY84
mTS
............

Weighted
National
Averages
of
Unadjusted
Concentrations
(
ng/
g)
and
Standard
Errors
for
Compounds
Analyzed
in
the
FY86
"
ATS
and
Also
Analyzed
in
the
FY82
and/
or
FY84
NHATS
.
.

Weighted
National
Averages
of
Surrogate­
Adjusted
Concentrations
(
ng/
g)
and
Standard
Errors
for
Compounds
Analyzed
in
the
FY86
NHATS
and
Also
Analyzed
in
the
FY82
and/
or
FY84
NKATS
............

Comparisons
of
Predicted
National
Average
Concentrations
(
ng/
g)
for
Selected
Semivolatiles
over
the
FY82,
FY84,
and
FY86
NHATS
..................

Chlorobiphenyl
Distribution
Across
the
Five
PCB
Homologs
Considered
for
Statistical
Analysis
in
the
FY86
NHATS
..........

Significance
Levels
from
Hypothesis
Tests
for
Differences
Between
Demographic
Groups
for
Selected
Semivolatiles
in
the
FY82,
FY84,
and
FY86
NHATS
.............

Volume
I1
Listing
of
NHATS
FY86
Composite
Data
Pase
.
8­
8
.
8­
9
8­
11
8­
20
8­
23
8­
30
8­
32
8­
35
xiv
­.
.....
­
­
­
LIST
OF
TABLES
(
cont.)

Volume
I1
(
cent.)

Listing
of
"
ATS
FY86
QC
Data
For
Compounds
Detected
in
At
Least
50%
of
Composites
(
plus
Octachlorobiphenyl1
Summary
of
QC
Data
for
the
FY86
NHATS
Spiked
Semivolatile
Compounds
Not
on
the
Target
List
for
Statistical
Analysis
Characterizing
the
Distribution
of
NHATS
FY86
Semivolatile
Compound
Concentrations
(
in
ng/
g,
Unadjusted
for
Surrogate
Recoveries)
Based
on
All
50
Composite
Samples
Arithmetic
Averages
(
and
Standard
Errors)
of
Extractable
Lipid
Concentrations
(
ng/
g)
for
Compounds
Analyzed
in
the
FY86
NHATS
and
Also
Analyzed
in
the
FY82
and/
or
FY84
NHATS
Comparisons
of
Predicted
Average
Concentrations
(
ng/
g)
by
Selected
Subpopulations
for
Selected
Semivolatiles
over
the
FY82,
FY84,
and
FY86
"
ATS
LIST
OF
FIGURES
Volume
I
Pase
Flot
Scheme
for
Analysis
of
S
mi
rolatile
Compounds
in
the
FY86
NHATS
.
.
.
.
.
.
.
.
.
.4­
2
Volume
11
Plots
of
Observed
vs.
Spiked
Concentrations
for
Spiked
Semivolatile
Compounds
Detected
in
At
Least
50%
of
Composites
(
plus
Octachlorobiphenyl).
Plots
of
Observed
Concentration
vs.
Batch
ID
for
Unspiked
Semivolatile
Compounds
Detected
in
At
Least
50%
of
Composites
Estimates
of
Average
Concentrations
with
95%
Confidence
Intervals
As
Determined
by
the
Additive
Model,
for
Selected
Subpopulations
in
the
FY86
NHATS
.
..
.
.
..
...
I,
I
LIST
OF
FIGURES
(
cat.)

Volume
I1
(
cont.)

Appendix
G
Concentrations
(
ng/
g)
from
the
FY82,
FY84,
and
EY86
"
ATS,
Plotted
in
Chronological
Order
with
Age
Group
as
the
Plotting
Symbol
Appendix
H
Surrogate­
Adjusted
Concentrations
(
ng/
g)
N82,
FY84,
and
FY86
"
ATS,
Plotted
in
Chronological
Order
with
Age
Group
as
t
Symbol
Appendix
K
Plots
of
Estimates
Average
Concentratio
Census
Region
and
by
Age
Group,
Plus
and
Min3.
s'
Two
Standard
Errors,
Based
on
the
Additive
the
FY82,
FY84,
and
FY86
"
ATS
mi
.
.­__.
­.
_
I_

.
.
I
RdMARY
National
Human
Monitoring
Program
(
NHMP),
operated
bited
States
Environmental
Protection
Agency's
Office
of
*
I
[
on
prevention
and
Toxics
(
USEPA/
OPPT),
is
a
national
monitor
the
human
body
burden
of
selected
chemicals.
$&
A
&
nal
Human
Adipose
Tissue
Survey
(
NHATS),
one
component
p,
was
performed
annually
to
collect
and
analyze
a
&
de
sample
of
adipose
tissue
specimens
from
autopsied
:
a
and
surgical
patients.
The
purpose
of
the
NHATS
was
to
kfy
and
quantify
the
prevalence
and
levels
of
selected
:
ais
in
human
adipose
tissue.
The
analysis
results
were
o
establish
an
exposure­
based
chemicals
list,
to
estimate
Lne
body
burden
levels
for
selected
chemicals,
and
to
!
F
terize
trends
in
these
levels
within
predefined
demographic
.
The
"
ATS
was
intended
to
fulfill
the
human
and
ronmental
monitoring
mandates
of
the
Toxic
Substances
Control
KC.

fandthe
Federal
Insecticide,'
Fungicide,
and
Rodenticide
Act,
$­­

The
EPA/
OPPT
earmarked
the
FY86
NHATS
tissue
repository
r
the
analysis
of
semivolatile
compounds
using
HRGC/
MS
methods.
&
NHATS
study
design
was
similar
to
those
used
in
the
FY82
i
FY84
"
HATS,
where
HRGC/
MS
analyses
of
semivolatile
compounds
F
?
re
also
performed.
This
report
presents
the
objectives,
ethodology,
and
results
of
the
FY86
"
HATS,
and
a
comparison
of
SUlts
with
the
FY82
and
FY84
"
ATS.

The
specific
objectives
of
the
FY86
NHATS
analysis
were
L~'
X*'
CF..
W
Determine
the
extent
to
which
semivolatile
organic:
ir'

,....
>,
.+
2'.
.
..?
compounds
are
present
in
human
adipose
tissue
samples,
I..
i..?
i
xvii
Estimate
the
average
concentrations
of
semivolatiles
in
the
adipose
tissue
of
humans
in
the
U.
S.
population
and
in
its
various
subpopulations,

Determine
if
any
key
demographic
factors
(
geographic
region,
age,
race,
and
sex)
are
associat'ed
with
the
average'concentrations
of
semivolatiles
in
human
adipose
tissue,
and
Compare
the
estimated
average
concentration
levels
of
semivolatiles
in
the
FY86
NHATS
with
estimates
from
the
FY82
and
FY84
NHATS,
when
similar
techniques
were
used
to
estimate
the
same
semivolatiles.

APPROACH
One
hundred
and
eleven
(
111)
qualitative
and
quantitative
semivolatile
organic
compounds
were
targeted
in
the
chemical
analysis
of
human
adipose
tissue
samples
in
the
FY86
NHATS.
For
compounds
with
sufficient
detection
percentages,
measured
concentration
data
were
statistically
analyzed
to
estimate
average
concentration
levels
in
the
U.
S.
population
and
to
determine
if
any
of
four
demographic
factors
of
interest
(
geographic
region,
age,
race,
and
sex)
were
associated
with
the
average
concentration
levels.
Statistical
analysis
was
also
used
to
compare
average
concentration.
levelsfound
in
the
FY82,
FY84,
and
FY86
NHATS
for
selected
compounds.
The
analytical
samples
in
the
FY82,
FY84,
and
FY86
NHATS
were
composites
of
individual
patient
specimens.
Compositing
criteria
were
established
to
achieve
the
study
objectives
of
estimating
and
comparing
average
concentrations
in
selected
subpopulations,
while
reducing
the
number
of
samples
to
analyze.
The
criteria
specified
that
composites
should
only
be
created
using
specimens
from
donors
in
the
same
age
group
and
from
the
same
U.
S.
Census
division.
This
ensured
makimum
precision
for
estimating
differences
in
body
burden
levels
among
populations
from
different
geographic
regions
and
age
groups.
A
total
of
50
composite
samples
were
analyzed
in
the
FY86
NHATS.
These
samples
were
prepared
from
671
individual
xviii
i
I,
i
1
lected
from
selected
metropolitan
statistical
areas
48
conterminous
United
States.

mivolatiles
targeted
for
analysis,
23
were
ds
ad
their
detection
percentages
among
the
FY86
e
samples
are
listed
in
Table
ES­
1.
For
rposes,
this
table
also
includes
the
detection
or
these
compounds
among
the
FY82
and
FY84
"
ATS
les.
Seventeen
(
17)
of
the
compounds
in
Table
ES­
1for
statistical
analysis
of
measured
concentration.
s
entration
estimates
for
the
five
PCB
homologs
Table
ES­
1
(
tetra­,
penta­,
hexa­,
hepta­,
and
octal
were
consolidated
to
characterize
overall
PCB
he
following
additional
PCB
parameters
were
tal
concentration
of
PCBs
(
sum
of
the
estimated
mologs.
However,
since
each
omitted
homolog
was
a1
average
concentrations
in
human
adipose
tissue
for
­­

­­

­­
­­
i
Table
135­
1.
Sedvolatile
Compounds
Detected
in
at
Least
44%
of
the
FY86
NIZATS
Composite
Samples
P
Compound
.

P'P­
DDE
PtP­
DDT
Heptachlor
epoxide
Beta­
BHC
Trans­
nonachlor
Oxychlordane
Dieldrin(')
CAS
Number
Pesticides
72­
55­
9
50­
29­
3
1024­
57­
3
319­
85­
7
39765­
80­
5
26880­
48­
8
60­
57­
1
Chlorobenzenes
Hexachlorobenzene
118­
74­
1
1,4­
Dichlorobenzene
_­
106­
46­
7
PAH8
Naphthalene
91­
20­
3
PCBS
Hexachlorobiphenyl
.
.
26601­
64­
9
Pentachlorobiphenyl
25429­
29­
2
Heptachlorobiphenyl
28655­
71­
2
.
I
Detection
Percentage
4
FY82
FY84
FY86
100%
96%
100%
68%
89%
96%
708
80%
94%
93%
89%
92%
57%
96%
92%
83%
78%
33%
39%
62%

719%
83%
98%
86%

42%
24%
84%

75%
98%
94%
73%
85%
88%
52%
84%
86%
Tetrachlorobiphenyl
26914­
33­
0
55%
41%
66%
Octachlorobiphenyl
31472­
83­
0
41%
18%
44%

Phthalate
Esters
Bis
(
2­
ethylhexyl)
phtha1ate(
2)
177­
81­
7
0%
78%
Di­
n­
butyl
phtha1atet2)
84­
74­
2
50%
100%
'
76%
Butyl
benzyl
phthalate(
2?
85­
68­
7
74%
62%
72%
96%
80%

72%

xxi
Table
ES­
2.
Estimates
of
National
Averag
Selected
Semivolatiles,
With
ce
Intervals,
frum
the
PY86
aTS
PIP­
DDT
PIP­
DDE
Beta­
BHC
Heptachlor
epoxide
Oxychlordane
Trans­
nonachlor
Dieldrin
l14­
DichlorobenZene
Hexachlorobenzene
Naphthalene
Tetrachlorobiphenyl
Pentachlorobiphenyl
Hexachlorobiphenyl
Heptachlorobiphenyl
Octachlorobiphenyl
Total
PCBs(')
Estimated
Pesticides
177.
2340.
157.
57.6
114.
130.
47.0
Chlorobenzene6
90.9
51.3
20.7
PCBa
56.4
135.
314.
125.
42.7
672.

Level
of
Chlorination(*)
58.3%

Other
(
Qualitative)

1­
Nonene
124.
Hexyl
acetate
123.
(
137­,
217.)
(
179?
3­,
2880.)
(
137­,
207.)
(
43­
2,
66.1)
(
33­
4,
129.)
(
43.6,
161.)
(
3
Of
63.1)

(
6Q.
2,
122.)
(
43­
3,
59.3)

(
15­
9,
25.4)

(
46.9,
65.9)
(
104­,
165.)
(
276.,
351.)
(
80.7,
169.)
(
19­
3,
66.1)

(
603­
742.)
(
51.2,
65.4)

(
20.6,
227.)
(
79.5,
166.)
over
t.
he
at
lea.
st
in
xxiii
the
17
compounds
included
in
the
statistical
analysis.
Approximate
95%
confidence
intervals
are
included
in
this
table
for
each
national
average.
Relative
standard
errors
of
these
estimates
ranged
from
5.9%
for
hexachlorobiphenyl
to
27.1%
for
octachlorobiphenyl.

Acre
Group
Effect8
The
effect
of
age
group
on
average
concentration
for
the
17
compounds
in
Table
ES­
2
was
statistically
significant
for
six
of
the
seven
pesticides
(
all
except
dieldrin),
five
PCB
homologs,
and
hexachlorobenzene.
In
each
case,
the
average
concentration
increased
with
the
age
of
the
donor.
Among
the
PCB
homologs,
the
average
concentration
for
the
45+
age
group
was
from
188%
(
pentachlorobiphenyl)
to
706%
(
heptachlorobiphenyl)
above
the
average
for
the
0­
14
age
group
(
an
increase
from
75.6
to
218
ng/
g
for
pentachlorobiphenyl,
and
from
26.9
to
217
ng/
g
for
heptachlorobiphenyl).
Similar
percent
increases
were
observed
with
the
pesticides.
For
example,
average
concentration
of
p,
p­
DDT
was
73
ng/
g
for
the
0­
14age
group
and
252
ng/
g
for
the
45+

age
group
­­
a
245%
increase.

Geocrraphic
Effects
Statistically
significant
differences
in
average
concentration
for
the
17
compounds
in
Table
ES­
2
were
observed
between
census
regions
for
p,
p­
DDT,
p,
p­
DDE,
heptachlor
epoxide,
hexachlorobenzene,
naphthalene,
and
three
of
the
five
PCB
homologs.
Average
concentration
of
p,
p­
DDT
and
the
PCBs
were
highest
in
the
northeast.
Heptachlor
epoxide
was
highest
in
the
south,
and
hexachlorobenzene
was
highest
in
the
west.
Similar
I
such
patterns
were
observed
in
the
FY82
and
FY84
NKATS.

;
I
Race
and
Sex
Grourm
The
differences
in
estimated
average
concentrations
between
race
groups
(
whitevs.
nonwhite)
and
between
sex
groups
xxiv
(
male
vs.
female)
were
not
statistically
significant
for
any
of
the
17
modeled
compounds.

Y
OF
THE
COMPARISON
WITH
FY82
AND
­
84
"
ATS
RESULTS
Fifty­
four
(
54)
of
the
111
semivolatiles
analyzed
in
the
86
"
ATS
were
also
analyzed
in
either
one
or
both
of
the
FY82
or
FY84
"
ATS.
Of
these
54
compounds,
twelve
were
detected
in
at
least
50%
of
the
samples
in
each
of
the
FY82,
EY84,
and
FY86
surveys.
Statistical
comparison
of
average
concentration
across
surveys
was
performed
on
ten
of
these
twelve
compounds
(
butyl
benzyl
phthalate
and
di­
n­
butylphthalate
were
excluded
from
statistical
analysis
based
on
FY86
QC
data
analysis
findings).
The
estimated
national
average
concentrations
within
each
survey
for
these
ten
compounds,
along
with
approximate
95%
confidence
intervals,
are
listed
in
Table
ES­
3.
Statistical
analysis
results
are
also
included
in
Table
ES­
3
to
identify
those
compounds
whose
results
for
FY82
and
FY84
differ
significantly
from
FY86.
For
the
four
PCB
homologs
considered
in
the
statistical
comparison,
the
FY86
average
concentration
was
from
48%
to
259%

higher
than
the
FY82
average
concentration.
The
differences
in
these
averages
for
tetra­,
penta­,
and
hexa­
chlorobiphenyl
were
statistically
significant
between
these
two
surveys.
The
observed
differences
in
average
concentration
for
PCB
homologs.
between
FY84
and
FY86
were
less
apparent;
the
only
statistically
significant
difference
was
a
58%
increase
from
FY84
to
FY86
in
average
concentration
for
hexachlorobiphenyl.
Total
PCBs
in
FY82
and
FY84
differed
significantly
from
FY86
results,
due
to
the
larger
national
average
noted
in
FY86.

Fewer
incidents
of
significant
differences
between
surveys
were
apparent
among
the
five
pesticides.
For
p,
p­
DDT
and
Plp­
DDE,
differences
of
43%
and
101%,
respectively,
between
the
FY84
and
FY86
average
concentrations
were
statistically
sigqificant.
Both
differences
were
increases
over
the
FY84
average.
Meanwhile,
the
only
pesticide
with
a
significant
xxv
_.
Table
ES­
3.
Estimates
of
National
Average
Concentrations
for
Selected
Semivolatiles,
With
95%
Confidence
Intervals,
from
the
FY82,
FY84,
and
FY86
"
ATS
Compound
Beta­
BHC
Trans­
nonachlor
Heptachlor
epoxide
Hexach1orobenzen.
e
Tetrachlorobiphenyl
Pentachlorobiphenyl
Hexachlorobiphenyl
:
i,
Estimated
,
95%
Avg.
Conc.
Confidence
"
ATS
(
ng/
g)
Interval
Pesticides
FY82
189.
(
125.,
253.)
FY84'
123:
(
102.,
145.)
FY86
177.
(
137.,
217.)

FY82
1840.
,
(
1130.,
2550.)
FY84
1150.
(
968.,
'
1330.)
FY86
2340.
(
1790.~
2880.1
FY82
291.
(
183.,
400.)
FY84
199.
(
150.,
248.)
FY86
157.
(
107.,
207.)

FY82
109.
[
53.0,
165.)
FY84
105..
I
94.4,
115.)
FY86
130.
(
99.6,
161.)

FY82
59.4
(
32.2,
86.5)
FY84
68.3
(
53.9,
82.6)
FY86
57.6
(
49.2,
66.11
Chlorobenzenes
FY82
118.
{
1.0,
256.)
FY84
42.9
(
31.9,
53.9)
FY86
51.3
(
43.3,
59.3)

PCBs
FY82
15.7
(
12.8,
18.6)
FY84
FY86
48.8
56.4
(
36.8,
60.8)
(
46.9,
65.9)

FY82
78.3
(
62.3,
94.4)
FY84
FY86
115.
135.
(
92.8,
137.)
(
104.,
165.)

FY82
FY84
FY86
176.
198.
314.
(
119.,
233.)
(
177.,
220.)
(
276.,
351.)
Diff.
From
FY86
12.1
­
53.4"

­
498.
­
1190.*

135.*
42.3
­
21.3
­
25.8
1.73
10.6
66.9
­
8.38
­
40.7"
­
7.60
­
56.2"
­
19.8
­
137.*
­
115.*
Estimated
95%
Diff.
Avg.
Conc.
Confidence
From
Compound
"
ATS
(
ng/
g)
Interval
'
FY86
PCBs
(
cost.)

Heptachlorobiphenyl
FY82
84.6
(
50.1,
119.)
­
40.5
.
FY84
­
129.
(
107.,
149.)
3.51
FY86
125:
(
80.7,
169.)

Total
PCBs(')
FY82
407.
(
337.
,
476.)
­
266.*
FY84
508.
(
469.
,
547.)
'­
164.*
FY86
672.
(
603.
,
742.)

Chlorination
Level@)
M82
59.3%
(
47.7,
71.0).
1.0
FY84
58.1%
(
53.1,
63.1)
­
0.2
FY86
58.3%
(
51.2,
65.4)

*
Significantly
different
from
zero
at
the
0.05
level.
('
1
Sum
of
concentrations
for
tetra­
to
octa­
chlorobiphenyl.
(
2)
Overall
chlorination
level
for
Pas,
defined
in
Section
6.2.1.2.

xxvii
I,

difference
in
average
concentration
between
the
Fy82
and
FY86
NHATS
was
beta­
BHC;
this
difference
was
a
46%
decrease
from
the
FY82
estimate.
When
interpreting
the
observed
differences
in
the
average
concentration
levels
between
the
FY86
 
ATS
and
both
the
FY82
and
FY84
NHATS,
it
is
important
to
consider
differences
in
analytical
approach.
For
example,
differences
in
the
internalquantitation
standards
used,
the
recovery
levels
observed,
the
analytical
laboratories,
and
improvements
made
in
the
analytical
method
over
time
all
may
contribute
substantially
to
observed
differences
between
surveys.
Additional
surveys
under
the
current
analytical
approach
(
HRGC/
MS
on
composite
samples)
covering
a
longer
time
period
are
needed
to
more
accurately
characterize
and
interpret
trends
in
average
concentration
levels
of
semivolatiles.
As
has
been
done
in
the
past,
the
designs
and
analysis
methods
for
these
surveys
should
be
established
to
meet
the
objective
of
comparing
results
across
surveys,
while
minimizing
any
nuisance
effects
contributing
to
the
comparisons.
1.0.
INTRODUCTION
1.1.
BACKGROUND
The
National
Human
Adipose
Tissue
Survey
(
NHATS)
has
been
the
main
operative
program
of
EPA's
National
Human
Monitoring
Program
(
NHMP).
The
"
HATS
program
has
collected
and
analyzed
human
adipose
tissue
samples
on
an
annual
basis
to
monitor
human
exposure
to
potentially
toxic
compounds.
The
"
MP/
NHATS
was
established
by
the'U.
S.
Public
Health
Service
in
1967
and
transferred
to
the
EPA
in
1970.
Since
1981,
the
EPA
Office
of
Pollution
Prevention
and
Toxics
(
EPA/
OPPT)
has
been
responsible
for
the
"
MP/"
ATS.
The
NwlTS
intended
to
fulfill
the
human
and
environmental
monitoring
mandates
of
the
Toxic
Substances
Control
Act
and
the
Federal
Insecticide,
Fungicide,
and
Rodenticide
Act,
as
amended.
Adipose
tissue
specimens
were
collected
annually
for
the
"
ATS
by
cooperating
pathologists
and
medical
examiners
during
routine
post­
mortem
examinations
or
elective
surgeries,
These
cooperators
were
selected
from
a
statistical
sample
of
Metropolitan
Statistical
Areas
(
MSAs)
within
the
48
conterminous
United
States.
Target
quotas
specifying
the
number
of
specimens
within
each
donor
age,
race,
and
sex
classification
were
established
for
each
collection
center.
Sampling
plans
were
designed
for
each
annual
survey
to
produce
statistically
unbiased
and
precise
estimates
of
the
levels
and
prevalence
of
compounds
in
the
U.
S.
population
and
in
various
demographic
subpopulations.
In
the
1970s
and
early
1980s,
the
"
HATS
program
characterized
the
prevalence
and
levels
of
19
organochlorine
pesticides
and
polychlorobiphenyls
(
PCBS)
in
individual
human
adipose
tissue
specimens,
using
packed
column
gas
chromatography/
electron
capture
detection
(
PGC/
ECD)
methods.
Recognizing
the
need
to
extend
the
capabilities
of
the
"
ATS
program,
the
EPA/
OPPT
initiated
a
series
of
programs
in
1984
to
expand
the
utility
of
the
tissue
repository.
In
order
to
expand
the
list
of
target
compounds
monitored
by
"
ATS
a
change
to
high­
resolution
1­
1
c
:;

ij
i
gas
chromatography/
mass
spectrometry
(
HRGC/
MS)
methods
was
made.
Individual
specimens
were
composited
prior
to
HRGC/
MS
analysis
to
optimize
the
amount
of
data
which
could
be
generated.
Analysis
on
composite
samples
rather
than
individual
patient
samples
necessitated
a
modified
statistical
analysis
approach
to
obtain
national
and
subpopulation
estimates
at
an
individual
level.
The
first
study
in
the
"
ATS
program
which
utilized
the
expanded
capabilities
of
the
HRGC/
MS
methodology
was
the
"
Broad
Scan
Analysis
Study"
(
Mack
and
Panebianco,
1986).
The
FY82
NHATS
specimen
repository
was
selected
for
this
study.
The
target
chemicals
considered
in
this
broad
scan
study
included
30
semivolatile
compounds,
17
volatile
organic
compounds,
and
11
polychlorinated
dibenzo­
para­
dioxins
(
PCDDs)
and
polychlorinated
dibenzo
furans
(
PCDFs).
The
broad
scan
study
demonstrated
that
13
of
these
semivolatile
compounds,
11
of
the
volatile
compounds,
and
nine
of
the
dioxins
and
furans
were
detected
in
at
least
half
of
the
composite
samples.
Estimated
average
levels
for
some
semivolatiles
increased
significantly
with
age,
while
the
South
and
Northeast
census
regions
tended
to
have
higher
levels
than
the
West
and
North
Central
regions.
The
FY84
"
ATS
specimen
repository
was
used
in
conducting
a
comparability
study
between
the
PGC/
ECD
and
HRGC/
MS
analytical
methods
(
Westat,
1990).
Paired
composite
samples
were
analyzed
using
both
methods.
A
total
of
58
semivolatile
compounds
were
analyzed
by
HRGC/
MS,
of
which
14
were
detected
in
at
least
50%
of
the
samples.
The
results
of
the
comparability
study
indicated
that
the
PGC/
ECD
method
was
generally
more
#

sensitive
than
the
HRGC/
MS
method
in
measuring
concentrations
for
a
variety
of
lipophilic
compounds,
with
the
opposite
holding
true
for
PCBs.
Method
comparability
issues
have
yet
to
be
resolved
for
many
of
the
target
semivolatile
compounds.
The
goal
of
the
study
performed
on
the
N  ATS
FY86
specimen
repository
was
to
estimate
baseline
body
burden
levels
of
semivolatile
organic
compounds,
and
to
characterize
trends
in
these
levels
within
predefined
demographic
groups
(
census
region,

1­
2
c
"
ba3Q.
t
region,
age,
race,
and
sex
classification)
are
concentration
levels
of
semivolatiles
the
NHATS
­­
­­­­­­­
...
.'
I/
I
I
age
group,
sex,
and
race).
HRGC/
MS
methods
were
employed
so
that
~
y86results
could
be
compared
to
FY82
and
FY84
results.
A
total
of
111
semivolatile
compounds
were
analyzed
in
the
FY86
"
ATS.

This
report
presents
the
results
of
the
FY86
"
ATS
analysis
on
semivolatiles,
along
with
the
comparison
to
results
from
the
FY82
and
FY84
"
ATS.

1.2.
OBJECTIVES
The
specific
objectives
of
the
FY86
"
ATS
and
analysis
rn
Determine
the
extent
to
which
semivolatile
organic
compounds
are
present
in
human
adipose
tissue
samples,

rn
Estimate
the
average
concentrations
of
semivolatiles
in
the
adipose
tissue
of
humans
in
the
U.
S.
population
and
in
its
various
subpopulations,

rn
Determine
if
any
key
demographic
factors
(
geographic
4
~

­­
J­­.=­­­­.
J
=­­=­­­­

associated
with
the
average
concentrations
of
semivolatiles
in
human
adipose
tissue,
and
Compare
the
prevalence
and
estimated
average
in
tiie
FY86
NHATS
with
that
from
the
FY82
and
FY84
"
ATS,
where
similar
sampling
and
analytical
techniques
were
used.

he
results
of
this
study
will
contribute
to
EPA's
knowledge
base
n
the
prevalence
and
concentration
levels
of
semivolatiles
in
an
adipose
tissue
samples.
Statistical
analysis
will
etermine
the
extent
to
which
concentrations
of
these
compounds
e
changing
over
a
six­
year
time
frame
in
the
1980s,
relative
to
alytical
effects
and
trends.

REPORT
ORGANIZATION
Volume
I
of
this
report
presents
the
methods,
results,
Conclusions
of
the
statistical
analysis
conducted
on
the
FY86
S
adipose
tissue
sample
data.
While
discussions
on
sample
n,
composite
design,
and
chemistry
methods
are
also
included
1­
3
in
this
report,
these
subjects
are
more
fully
addressed
in
other
references
(
see
Chapter
9).

Battelle
developed
the
sample
design
and
composite
design
for
the
FY86
NHATS.
The
sample
and
composite
designs
are
highlighted
in
Chapters
2
and
3,
respectively.
Chapter
4
discusses
the
chemistry
procedures
that
Midwest
Research
Institute
(
MRI)
used
to
analyze
the
FY86
composite
and
QC
samples.
Included
in
this
chapter
are
discussions
of
overall
data
quality,
analytical
procedures,
and
QA/
QC
procedures.
FY86
data
issues
and
other
pre­
statistical
analysis
results
are
presented
in
Chapter
5.
Detection
status
of
the
111
semivolatile
compounds
are
presented,
along
with
data
issues
found
to
be
unique
to
the
FY86
analysis
approach.
For
example,
methods
were
developed
in
this
effort
to
adjust
measured
concentrations
for
surrogate
recoveries
in
order
to
more
accurately
estimate
actual
sample
concentrations.
The
results
of
statistical
analysis
on
QC
sample
data
are
presented
in
Chapter
5;
these
results
characterize
measurement
error,
recoveries,
background
levels,
and
the
presence
of
batch
effects.
Chapter
6
contains
a
discussion
of
the
statistical
methodologies
used
by
Batteile
in
estimating
average
concentration
levels
ana.
associatedstandard
errors
for
target
compounds.
The
results
of
applying
these
statistical
methodologies
to
the
FY96
"
ATS
data
are
presented
in
Chapter
7.
Finally,
Chapter
8
presents
the
results
of
comparing
FY86
"
ATS
results
with
those
from
the
FY82
and
FY84
"
ATS
for
the
same
compounds.
Supporting
information
on
individual
sample,
data,
including
data
listings
and
plots,
data
summary
statistics,
QC
data
plots,
and
graphical
display
of
the
estimated
average
concentrations
with
associated
levels
of
uncertainty,
is
included
as
appendices.
These
appendices
constitute
Volume
I1
of
this
document.

1­
4
2.0
NHATS
FY86
SAMPLE
DESIGN
The
human
adipose
tissue
specimens
in
the
FY86
"
ATS
repository
were
collected
from
October,
1985,
through
September,
1986.
The
method
in
which
these
specimens
were
supplied
to
the
~.
J"
ATSprogram
follows
the
"
HATS
Sampling
Design.
In
each
year
of
the
NHATS
program,
cooperators
(
hospital
pathologists
or
medical
examiners)
collected
approximately
700­
1200
adipose
tissue
specimens.
Although
the
NHATS
target
population
is
the
general,
noninstitutionalized
U.
S.
population,
the
sampling
population
was
limited
to
cadavers
and
surgical
patients
due
to
the
invasive
nature
of
the
process
required
to
collect
the
adipose
specimens
from
living
persons.
Section
2.1
discusses
the
"
ATS
Sampling
Design
and
its
multistage
characteristics.
Methods
used
to
collect
specimens
are
discussed
in
Section
2.2.
Finally,
a
summary
of
the
types
of
specimens
collected
in
the
M86
"
ATS
is
presented
in
Section
.
I
SAMPLING
DESIGN
The
NHATS
program
used
a
multistaged
sampling
design
to
tain
adipose
tissue
specimens
from
autopsied
cadavers
and
rgical
patients
throughout
the
United
States.
The
NHATS
c
mpling
Design
consisted
of
three
components:

The
48
conterminous
states
were
stratified
into
distinct
geographical
areas.

A
sample
of
Metropolitan
Statistical
Areas
(
MSAs)
was
selected
within
the
strata.
The
probability
of
selecting
an
MSA
was
proportional
to
its
population
percentage
within
the
stratum.

One
or
more
cooperators
were
chosen
from
each
MSA
and
asked
to
supply
a
specified
quota
of
tissue
specimens
to
the
NHATS.
To
maintain
similarity
in
the
sampling
designs
across
fiscal
years,
the
same
MSAs
and
cooperators
were
retained
from
year
to
year
to
the
extent
possible.

2­
1
As
part
of
the
third
component
of
the
"
ATS
Sampling
Design,
the
manner
in
which
cooperators
selected
the
donors
and
tissue
specimens
was
nonprobabilistic,
but
followed
a
specific
set
of
criteria.
Quotas
and
subquotas
for
the
number
of
specimens
supplied
to
the
"
ATS
were
assigned
to
each
cooperator.
The
subquotas
determined
the
desired
number
of
specimens
coming
from
particular
combinations
of
donor
age
group,
race,
and
sex.
Demographic
categories
in
which
subquotas
were
defined
are
presented
in
Table
2­
1.
The
subquotas
were
proportional
to
the
1980
U.
S.
Census
population
counts
for
each
sampling
stratum.

Table
2­
1.
Demographic
Categories
in
Which
Subquotas
Were
Established
for
Collecting
Adipose
Tissue
Specimens
Age
group
0­
14years
15­
44years
45+
years
Race
group
Caucasian
Non­
Caucasian
Sex
group
Female
Male
I
Because
the
survey
required
some
divergence
from
strict
probabilistic
sampling,
the
validity
of
the
statistical
estimates
derived
from
the
data
depended
on
several
assumptions:

m
The
concentrations
of
toxic
substances
in
the
adipose
tissue
of
cadavers
and
surgical
patients
are
assumed
to
be
comparable
to
those
in
the
general
population.

W
The
levels
of
toxic
substances
in
urban
residents
are
approximately
the
same
as
in
rural
residents,
and
thus
the
selection
of
only
urban
hospitals
and
medical
examiners
(
i.
e.,
those
located
in
MSAs)
does
not
introduce
any
significant
bias
into
the
estimates
of
average
concentration
levels.

2­
2
rn
No
systematic
bias
is
introduced
by
the
fact
that
the
cooperators
are
not
randomly
selected
and
that
the
donors
and
specimens
are
nonprobabilistically
sampled
according
to
pre­
specified
quotas.

Further
discussion
of
the
three
components
of
the
NHATS
Sampling
Design
follow.

Prior
to
1985,
the
sampling
strata
from
which
MSAs
were
randomly
selected
were
the
nine
U.
S.
Census
divisions.
But
in
9
1985,
EPA
wanted
the
ability
to
obtain
estimates
of
average
\+

$
concentration
levels
in
each
of
the
ten
EPA
regions.
Thus,
beginning
with
the
FY85
NHATS,
the
sampling
strata
were
redefined
as
seventeen
geographic
areas
of
the
country,
resulting
from
the
&
&';$?:;;
,..,~
li..
Lt..­
ll
>$@
F:
i
ntmvaert
ion
of
t.
he
nine
census'
divisions
and
the
ten
EPA
regions.
Selecting
the
sample
under
this
new
stratification
scheme
made
it
possible
to
make
comparisons
with
previous
"
ATS
results
and
also
obtain
estimates
for
the
EPA
regions.
The
states,
census
divisions,
and
EPA
regions
that
define
the
seventeen
strata
are
shown
in
Table
2­
2.

Although
the
FY86
NHATS
sampling
design
specified
that
specimens
be
collected
across
the
seventeen
strata,
it
was
not
possible
to
create
composites
so
that
all
specimens
within
a
composite
came
from
the
same
stratum.
However,
the
Composite
Design
assured
that
each
composite
contained
specimens
orisinatinq
from
the
same
census
division
and
age
group.
This
ii
was
done
to
ensure
that
the
FY86
and
FY82
analysis
results
Could
be
compared.
Chapter
3
discusses
the
Composite
Design
in
greater
P
.
detail.

9:.­
isampling
.
f
plan.
Cooperators
were
recruited
from
each
selected
MSA
';
f
;,.
h
&
provide
tissue
samples
for
the
NHATS.
:

2
3
Table
2­
2.
Sampling
Strata
Definitions
for
the
"
ATS
Texas
10
West
North
7
Iowa
Central
Missouri
Nebraska
Kansas
11
West
North
a
North
Dakota
Central
South
Dakota
Table
2­
2.
(
cont.)

12
Mountain
8
Montana
Wyoming
Colorado
Utah
13
Mountain
9
Arizona
Nevada
14
Pacific
9
California
15
Mountain
10
Idaho
16
Pacific
10
Washington
Oregon
17
Mountain
6
New
Mexico
2­
5
.
Once
the
seventeen
sampling
strata
were
identified
for
the
FY85
"
ATS,
a
sample
of
MSAs
was
selected
using
a
controlled
selection
technique,
known
as
the
Keyfitz
technique
(
Kish
and
Scott,
1971).
This
sample
differed
from
those'MSAsselected
prior
to
the
FY85
"
HATS.
However,
the
Keyfitz
technique
maximized
the
probability
of
retaining
previously
selected
MSAS,
thus
allowing
to
continue
employing
existing
cooperators
(
Mack,
et.
al.,
1984).
The
MSA
sample
selected
in
FY85
served
as
the
base
"
ATS
sample
for
FY86
through
FY91.

The
FY86
"
HATS
sampling
design
contained
46
MSAs,
of
which
two
(
St.
Louis
and
Moline)
were
each
split
into
two
primary
sampling
units
to
reflect
areas
of
the
MSA
falling
into
different
sampling
strata.
All
but
one
of
the
MSAs
selected
for
the
FY85
"
HATS
were
used
in
the
FY86
NHATS;
the
omitted
MSA
(
Medford
OR)
was
replaced
(
Eugene
OR)
because
satisfactory
cooperators
could
not
be
found.
The
sample
MSAs
for
the
FY86
NHATS
are
listed
by
stratum
in
Table
2­
3.
Four
MSAs
(
LosAngeles,
Chicago,
Detroit,
and
New
York)
were
listed
as
double
collection
sites
because
their
populations
were
much
larger
than
other
MSAs
within
their
strata.
Strata
13,
15,
and
17
had
no
MSAs
selected
due
to
their
small
population
sizes.

2­
1­
3
SDecimen
Collection
Quotas
Pre­
assigned
quotas
determined
the
numbers
of
specimens
selected
within
each
sample
MSA.
In
addition,
demographic
subquotas
were
assigned
to
each
MSA
to
ensure
that
the
specimens
collected
were
representative
of
the
strata
with
respect
to
the
three
demographic
factors
in
Table
2­
1
(
age
group,
race
group,
and
sex
group).
The
subquota
assigned
to
each
MSA
was
determined
proportionally
represented,
but
the
subquota
did
not
specify
that
.
Table2­
3.
Sample
MSAa
Selected
for
the
FY86
"
ATS
1
New
England
2
Middle
Atlantic
3
Middle
Atlantic
4
South
Atlantic
5
South
Atlantic
6
East
South
Central
7
East
North
Central
8
West
North
Central
9
West
South
Central
10
West
North
Central,
1
Springfield,
MA
Boston,
MA
2
Albany,
NY
New
York,
Id)
Binghamton/
Elmira,
NY
Newark,
NJ
3
3
4
4
5
5
6
Philadelphia,
PA
Pittsburgh,
PA
Erie,
PA
Washington,
DC
Norfolk,
VA
Tampa,
FL
Greenville,
SC
Orlando,
FL
West
Palm
Beach/
Boca
Raton,
FL
Miami,
FL
Atlanta,
GA
Memphis,
TN(
2)
Birmingham,
AL
Lexington,
KY
Dayton,
OH
Detroit,
MI(')
Columbus,
OH
Cleveland,
OH
Akron,
OH
Chicago,
IL(')
Madison,
WI
Moline,
ILt2)

Rochester,
MN
El
Paso,
TX
Lubbock,
TX
Houston,
TX
San
Antonio,
TX
Dallas,
TX
Omaha,
NE
St.
Louis.
MO(~)
Table
2­
3.
(
cont.)

11
West
North
Central
12
Mountain
14
Pacific
16
Pacific
8
8
9
10
I
I
I
*
­"
*
h
(*)
Indicates
a
double
collection
site.
A
double
collection
site
is
an
MSA
whose
population
relative
to
its
stratum
is
so
large
that
its
proper
representation
in
the
sample
requires
it
to
be
selected
twice.

Indicates
a
split
MSA.
A
split
MSA
is
one
which
covers
more
than
one
stratum.
Only
the
portion
of
the
stratum
in
which
the
MSA
is
listed
is
represented
in
the
sample.

@)
Indicates
a
replacement
MSA.
A
replacement
MSA
is
an
MSA
that
was
not
selected
in
the
FY8S
probability
sample,
but
was
chosen
to
replace
an
FY8S
sample
MSA
for
which
a
satisfactory
cooperator
could
not
be
found.
Sioux
Falls,
SD
Salt
Lake
City,
UT
Denver,
CO
San
Francisco,
CA
Sacramento,
CA
LOS
Angeles,
a(')
Portland,
OR
Spokane,
WA
Eugene,
OR(
3)
Yakima,
WA
2­
8
represented
within
each
combination
of
age
group
and
sex,
The
subquotas
only
specified
the
total
nut&
er
of
Caucasian
and
non­
Caucasian
specimens
to
be
collected
from
each
MSA.
The
subquotas
for
the
seventeen
sampling
strata
for
the
FY86
sample
design
are
presented
in
Table
2­
4.
Each
MSA
had
a
quota
of
twenty­
seven
specimens,
except
for
the
four
MSAs
that
were
designated
as
double
collection
MSAs.
In
those
MSAs,
the
quotas
and
subquotas
were
doubled.
Cooperators
within
an
MSA
were
assigned
quotas
and
subquotas
appropriate
to
that.
MSA.
The
total
number
of
samples
specified
for
the
FY86
NHATS
was
1404.
This
was
based
on
the
quota
of
twenty­
seven
specimens
for
each
of
the
forty­
eight
MSAs,
plus
twenty­
seven
additional
specimens
for
each
of
th.
e
four
MSAs
designated
as
double
collection
MSAs.

2.2
SAMPLE
COLLECTION
PROCEDURES
NHATS
specimens
were
adipose
tissue
samples
excised
by
pathologists
and
medical
examiners
during
therapeutic
or
elective
surgery
or
during
postmortem
examinations.
If
the
specimen
was
Collected
postmortem,
the
tissue
was
obtained
from
an
unembalmed
cadaver
which
had
been
dead
for
less
than
twenty­
four
hours
and
had
been
kept
under
refrigeration
during
that
time.
The
death
should
have
been
caused
by
sudden
traumatic
injury,
such
as
cardiac
arrest,
car
accident,
or
gunshot
wound.
The
following
groups
were
excluded
from
specimen
collection:

institutionalized
individuals;

persons
known
to
be
occupationally
exposed
to
toxic
chemicals;

persons
who
died
of
pesticide
poisoning;
and
persons
suffering
from
cachexia.

2­
9
Table
2­
4.
FY86
Age
and
Sex
Subquotaa,
and
the
Race
Subquota,
for
Each
NHATS
Collection
Site
Within
Each
Stratum
New
England
1
2
3
3
6645
Middle
Atlantic
2
5
33
6645
I
Middle
'
Atlantic
I
3
3
3
3
6645
I
South
Atlantic
3
6
3
3
6744
South
Atlantic
4
6
3
3
6645
East
South
Central
East
North
Central
West
North
Central
West
South
Central
West
North
Central
West
North
Central
Mountain
Mountain
Pacific
Mountain
Pacific
Mountain
4
5
3
3
6645
5
4
3
3
6645
5
1
3
3
6645
3
6
6
4
4
I
I
7
2
3
3
6
6
4
5
'
8
2
3
3
6
6
4
5
a
2
3
3
9
4
3
3
6
6
4
5
9
7
3
3
6
7
4
4
10
1
4
4
6
6
3
4
10
2
3
3
7
7
3
4
6
7
4
3
For
each
stratum,
the
six
subquotas
across
age
and
sex
groups
add
to
27,
the
total
quota
for
each
selected
MSA
from
the
stratum.
The
non­
Caucasian
I.

subquota
represents
the
number
of
specimens
out
of
27
corresponding
to
non­
Caucasian
donors.

2­
10
These
guidelines
were
stipulated
so
that
the
levels
of
substances
detected
in
the
specimens
were
a
result
of
environmental
exposure.
..

Instructions
for
the
cooperators
stipulated
that
at
least
five
grams
of
tissue
be
obtained
from
each
donor.
In
addition,
the
cooperators
were
to
avoid
contamination
through
contact
with
disinfectants,
paraffins,
plastics,
preservatives,
and
solvents.
Cooperators
placed
the
collected
specimens
in
glass
jars
with
Teflon@
lids
and
stored
them
at
­
loo
to
­
200
C.
The
jars
were
packed
on
dry
ice
for
overnight
shipment
to
mI,
the
contractor
responsible
for
tissue
storage.
MRI
received
the
specimens
and
checked
them
for
adequacy
of
shipping
conditions
and
level
of
conformance
with
cooperator
quota.
MRI
determined
an
approximate
specimen
weight
and
transferred
the
specimens
to
storage
at
­
2OO
C.
Upon
examining
the
patient
summary
reports,
MRI
forwarded
the
reports
to
Battelle
for
processing.

2.3
SPECIMEN
COLLECTION
SUMMARY
In
the
FY86
NHATS,
cooperators
provided
739
specimens
in
31
of
the
sample
MSAs.
In
preliminary
review
of
the
specimens,
671
were
collected
in
accordance
with
the
quotas
and
subquotas.
These
specimens
were
labeled
"
Design1#
specimens.
The
remaining
specimens
were
labeled
lfSurplusff
specimens,
as
their
collection
as
considered
beyond
the
quotas
and
subquotas
requested.
The
process
of
labeling
specimens
as
Design
or
Surplus
dlowed
established
guidelines
(
Orban,.
et.
al.,
1988).
However,
PA
added
a
stipulation
that
the
collection
dates
of
Surplus
Pecimens
be
uniformly
distributed
throughout
the
fiscal
year.

SO,
it
was
necessary
to
modify
Surplus
specimen
assignment
from
e
preliminary
review,
as
one
composite
contained
mostly
low
ight
specimens.
Surplus
specimens
were
relabeled
as
Design
cimens
and
added
to
this
composite
in
order
for
the
composite
achieve
sufficient
tissue
mass.
Meanwhile,
the
same
number
of
n
specimens
from
another
amply­
representedcomposite
within
ame
census
division
were
relabeled
Surplus
specimens
and
2­
11
Or.!
g>@
q
­
4
1
removed
from
the
composite.
Thus
the
total
number
of
Surplus
specimens
collected
in
FY86
did
not
change
following
this
adjustment.
The
maximum
nueer­
ofspecimens
from
a
MSA
remained
I
at
the
original
quota
of
twenty­
seven
(
or
fifty­
four
from
a
double­
collection
MSA)

Table
2­
5
is
a
summary
of
the
collection
effort
for
the
FY86
"
ATS,
detailed
by
census
division.
In
FY86,
EPA
chose
not
to
make
estimates
for
EPA
regions.
Instead,
EPA
maintained
.
similarity
to
the
FY82
geographic
classifications
in
order
to
compare
FY86
results
to
FY82
results.
All
671
Design
specimens
were
placed
into
one
of
fifty
composites,
on
which
laboratory
analysis
was
performed.
Table
2­
6
shows
the
number
of
quota
specimens,
collected
specimens,
and
Design
specimens
in
each
of
the
four
demographic
subpopulations
(
census
region,
age
group,
sex,
and
race)
which
act
as
analysis
factors
in
the
linear
model.
Because
the
number
of
samples
in
the
chemical
analysis
was
not
large
enough
to
obtain
reliable
estimates
for
all
nine
census
divisions,
Battelle
combined
the
divisions
into
four
census
regions
for
the
FY82;
FY84,
and
FY86
model
analyses.

2­
12
i,

c
E"
4
0
aJ
a
0
3
#
al
0
0
a
­
48
0
0
4
m
3"

~

2­
13
Table
2­
6.
FY86
NHATS
Specimen
Collection
Summary
by
Demographic
Subpopulation
Census
Region
Age
Group
Sex
Group
Race
Group
Northeast
North
Central
270
405
124
265
123
248
Sijuth
459
255
205
West
270
95
95
Total
1404
739
671
0­
14
years
15­
44
years
45+
years
317
642
445
115
248
376
108
221
342
Total
1404
739
671
Male
681
354
315
Female
723
385
356
Total
1404
739
671
White
1179
564
529
Nonwhite
225
175
142
~~

k
I
Total
1
1404
I
739
I
671
,

3.0
NHATS
FY86
COMPOSITE
DESIGN
Battelle
assigned
the
671
Design
specimens
'
inthe
Fy86
"
ATS
tissue
repository
to
composite
samples
using
specific
composite
design
criteria
(
Orban,
et.
al.
1988).
The
necessity
for
cornpositing
samples
prior
to
chemical
analysis
was
to
ensure
that
at
least
twenty
grams
of
tissue
were
available
per
sample
to
meet
the
limit
of
detection
goals
for
the
target
compounds.
The
Composite
Design
resulted
in
constructing
50
composite
samples.

3.1
DESIGN
GOALS
COMPOSXTING
CRITERIA
listed
I
I
I
The
five
goals
of
the
FY86
"
ATS
Composite
Design,
in
order
of
importance,
were
to:

maintain
similarity
to
the
FY82
Composite
Design,

maintain
equal
weighing
of
specimens
within
the
composite
samples,

specify
additional
numbers
of
pure
sex
composite
samples
than
in
FY82,

control
the
MSA
effect,
and
provide
the
best
range
of
race
group
percentages
across
the
composite
samples.

Because
of
the
constraints
imposed
by
the
sampling
and
compositing
protocols
and
the
frequency
of
collection
nonresponse,
it
was
not
always
possible
to
meet
all
the
design
goals.
Each
of
the
above
goals
required
a
different
mix
of
individual
specimens
within
the
composite
samples.
Thus,
attempts
were
made
to
achieve
all
goals
across
the
design
to
the
extent
possible.
The
five
goals
are
discussed
in
detail
below.

.­­.
I­
Similaritv
to
the
FY82
Composite
Desisn
EPA
imposed
this
criterion
to
ensure
that
the
results
of
FY86
data
analysis
could
be
compared
with
FY82
results,
where
Ompositing
was
performed
and
the
same
semivolatile
compounds
ere
analyzed.
The
design
criterion
imposed
by
this
objective
is
3­
1
that
eaeh
composite
sample
had
to
be
constructed
from
individual
specimens
collected
from
exactly
one
census
division
and
exactly
one
age
group
category.
Thus
there
were
27
distinct
categories
within
which
composite
samples
were
formed.
Once
the
FY86
Composite
Design
was
established,
it
was
desired
to
compare
results
of
data
analysis
on
the
FY86
samples
with
the
results
obtained
from
the
HRGC/
MS
analysis
on
FY84
composite
samples.
The
FY84
Composite
Design
closely
paralleled
the
FY82
Composite
Design,
allowing
the
FY86
results
to
be
compared
with
the
FY84
results
as
well
as
the
FY82
results.
Of
primary
importanae,
the
FY84
design
stipulated
that
all
specimens
found
in
a
given
composite
originate
from
the
same
age
group
and
census
division.

2.
Euual
weiuhincr
of
mecimens
within
the
comnosite
samDles
This
criterion
is
primarily
for
ease
of
interpretation.
In
attempting
to
make
inferences
on
individual
specimen
concentrations,
it
is
far
easier
to
interpret
the
observed
composite
sample
concentrations
as
the
arithmetic
average
of
the
individual
specimen
concentrations.
Therefore,
this
design
goal
specified
that
each
individual
specimen
within
a
composite
sample
contribute
an
equal
amount
of
tissue
to
the
composite
sample.
This
specification
allows
the
lipid­
adjustedconcentration
of
the
composite
sample
to
be
interpreted
as
approximately
the
arithmetic
average
of
the
lipid­
adjustedindividual
specimen
concentrations,
with
equality
occurring
whenever
all
specimens
in
the
composite
sample
have
the
same
percentage
of
lipid
material.
In
the
FY86
analysis,
specimens
were
not
labeled
as
Surplus
as
a
result
of
specimen
weight,
nor
was
specimen
weight
used
to
determine
whether
the
specimen
would
be
included
in
a
composite
sample.
The
specimen
weights
were
evaluated
only
after
composites
were
defined
based
on
the
other
design
criteria.
Composites
with
insufficient
tissue
mass
 or
chemical
analysis
were
modified
if
practical
alternatives
were
available.
This
3­
2
policy
resulted
in
combining
two
initial
composites
and
modifying
an
additional
two
composites.
To
ensure
that
equal
weighing
of
specimens
within
the
composite
samples
was
maintained
throughout
the
analysis,
instructions
for
evaluating
individual
specimen
weights
were
based
on
the
ratio
of
the
maximum
weight
to
the
minimum
weight
of
all
specimens
within
the
composite
sample.
Any
low­
weight
specimens
causing
this
ratio
to
exceed
3.0
was
recommended
for
removal
from
the
composite.

3.
Construct
more
Dure
sex
comDosite
samnles
than
in
FF82
Pure
sex
composites
(
composites
containing
specimens
originating
from
either
all
male
patients
or
all
female
patients)
were
constructed
when
sufficient
numbers
of
specimens
were
available
within
a
particular
census
division/
age
group
category
and
more
than
one
composite
sample
was
allocated
to
this
category
by
the
design.
Pure
sex
composites
were
needed
to
achieve
more
precise
estimates
of
sex
effects
in
the
population.
This
design
strategy
was
in
contrast
to
the
FY82
Composite
Design,
which
provided
for
more
balanced
sex
composite
samples
(
samples
with
nearly
half
male
and
half
female
specimens).
Including
more
pure
sex
composites
in
the
FY86
design
intended
to
reduce
the
standard
errors
for
the
sex
group
estimates
from
that
observed
for
the
FY82
analysis
(
Draper
and
Smith,
1981,
pp.
52­
55).

4.
Control
the
MSA
effect
Controlling
the
number
of
MSAs
contributing
specimens
to
composite
samples
was
intended
to
reduce
the
effect
of
the
MSA
on
the
estimated
average
concentrations.
This
was
done
because
MSAs
are
regarded
as
being
major
sources
of
differences
in
observed
Concentrations
across
the
nation
due
to
their
varied
exposure
Scenarios
(
Panebianco,
1986).
To
avoid
confounding
the
MSA
L
effect
with
any
of
the
geographic
or
demographic
effects,
the
,
Composite
Design
stipulated:

CY
P
(
3
05
I,
"
h
I
I
4a.
to
keep
the
number
of
MSAs
represented
in
each
composite
sample
consistent
across
the.
design
(
targeted
at
2­
3
MSAS),
and
4b.
to
maintain
approximately
the
same
number
of
pure
sex
composite
samples
within
a
group
of
MSAs.

Criterion
4a
helped
to
ensure
a
constant
variance
of
measured
concentrations
across
the
sample
whenever
the
composite
sample
concentrations
are
averages
over
an
equal
number
of
MSAs.
Criterion
4b
was
intended
to
prevent
confounding
a
large
MSA
effect
with
the
sex
effect,

5.
Control
the
race
urouD
nercentaaes
across
the
cornnosite
samnles
The
benefits
for
constructing
pure
race
group
composite
samples
paralleled
the
benefits
for
constructing
pure
sex
composite
samples.
However,
achieving
this
design
goal
was
dependent'on
the
number
of
non­
Caucasian
specimens
collected
in
the
twenty­
seven
census
division/
age
group
categories
and
the
number
of
composite
samples
in
the
design.
At
least
one
pure
Caucasian
composite
sample
and
at
least
one
pure
non­
Caucasian
composite
sample
were
constructed
in
four
different
census
division/
age
group
categories.

3.2
LABORATORY
COMPOSITING
PROCEDURES
In
the
FY86
"
ATS
Composite
Design,
specimens
from
nine
census
divisions
and
three
age
groups
were
segregated
into
50
composites.
Battelle
provided
MRI
with
composite
sample
data
sheets
that
identified
the
specific
individual
specimens
to
be
included
in
each
composite
(
Appendix
A
of
Orban,
et.
al.,
1988).

A
composite
consisted
of
from
three
to
twenty­
four
specimens,
The
composite
sample
data
sheets
provided
sufficient
information
(
EPA
ID
number,
package
number,
sample
weight,
hospital
code,
etc.)
such
that
the
individual
specimens
could
be
cross­
checked
with
the
study
design.
The
data
sheets
were
used
as
work
sheets
to
record
actual
laboratory
compositing
procedures.

3.

..
Initially,
the
samples
were
grouped
into
composites,
and
any
samples
of
insufficient
weight
(<
1.0
g)
or
potentially
I
contaminated
samples
were
reported
by
MRI
to
the
EPA
Work
Assignment
Manager
(
WAM).
Such
samples
were
omitted
from
the
analysis­
The
weights
of
composites
included
in
laboratory
analysis
ranged
from
1.884g
to
22.514g,
with
three
composites
below
the
target
weight
of
20s.
The
composite
with
the
lowest
weight
consisted
of
only
three
samples
from
the
0­
14year
age
category.
The
other
two
composites
below
the
target
weight
had
insufficient
samples.
The
composite
samples
were
placed
on
dry
ice
during
the
compositing
procedure.
An
electronic
four­
place
balance
was
used
to
weigh
the
samples,
and
the
calibration
of
the
balance
was
checked
with
a
Class
P
set
of
weights
(
laboratory
grade,
tolerance
1/
25,000)
before
any
weighing
was
begun
and
once
during
the
sample
weighings.
To
weigh
the
samples,
a
clean
culture
tube
was
labeled
with
the
composite
number.
This
tube
was
placed
on
the
balance,
and
the
weight
was
tared.
A
sample
was
removed
from
the
composite
bag,
the
jar
opened,
and
a
portion
of
the
frozen
adipose
removed
with
a
clean
stainless
steel
spatula.
The
adipose
was
placed
in
the
culture
tube
and
the
weight
recorded
to
three
decimal
places
on
the
compositing
sheets.
Additional
adipose
was
added
if
necessary.
A
goal
of
210%
of
the
desired
weight
was
attempted
where
possible.
The
weights
of
the
individual
specimens
were
recorded
on
the
composite
data
sheets.
The
weight
of
the
culture,
beaker,
and
adipose
tissue
was
rezeroed,
and
the
next
sample
in
the
composite
was
weighed.

A
new
spatula
was
used
between
each
sample.
This
procedure
was
repeated
for
each
sample
in
the
composite.
When
the
composite
was
completed,
it
was
capped
and
stored
in
a
sample
freezer
at
­
loo
C.
All
data
on
the
actual
compositing
procedures
(
amount
.
dde
Id,
remaining
composite)
weight,
date
inventoried,
and
total
weight
recorded
on
the
data
sheets
provided
by
3­
5
Q*
i"<":
3
02;
2
If
t.
he
spec.
were
Battelfe.
MRI
submitted
all
data
sheets
in
a
separate
report
documenting
the
compositing
activity
(
MRI,
1988a).

^"

3.3
SUMMARY
OF
FY86
&
HATS
COMPOSXTB
SAMPLES
The
FY86
"
ATS
Composite
Design
resulted
in
const
50
composite
samples
using
671
individual
specimens
collected
from
31
MSAs.
Composite
samples
were
formed
from
specimens
.
collected
exclusively
from
the
same
census­
division/
age
group
category.
The
numbers
of
composites
within
each
of
these
categories
are
given
in
Table
3­
1.
Unlike
the
exclusivity
by
census
division
and
age
group,
the
composite
samples
had
specimen
percentages
within
sex
and
race
groups
which
iraried
across
the
design
depending
on
the
availability
of
specimens
within
specific
demographic
subpopulations.
Table
3­
2
shows
the
demographic
makeup
of
the
FY86
NHATS
composite
samples.
The
50
composite
samples
were­
randomlyassigned
to
five
laboratory
batches
of
ten
samples
each.
Within
each
batch,
the
ten
comDosite
samples
and
three
lipid­
based
QC
samples
were
­­­
placed
in
random
order
for
chemical
analysis.

3­
6
Table
3­
1.
Distribution
of
FY86
NHATS
Composite
Samples
by
Census
Division
and
Age
Group
Mountain
1
1
2
4
,.
Pacific
1
1
4
6
Total
10
16
I
24
50
,._....,
.
.
_..
,_

3­
7
ml­
o.
Cn0
.....
dDmmam
a3dDPa3lea300
dO.
OOO
....
.....
*
moo
eo000
em0
eo
0
rl
rlrl
a
E:
rd
rl
bl
d
w
3
mP*
mPmorn
OwaNlOmOm
........
rlrlorlrlrlrlrl
~
rn000000
........
waoooooo
dIn0
000
rl
i­
lrlrl
OOOdDOOOrn
OOOd
........
....
ooomooorn
OOOd
0
o*
0
m
mo
d(
drl
rl
rl
3­
8
~
omoooooooo
...........
a3omoooooooo
woa3a30
0
000
d
rl
d
ddd
womoo*
odow
..........
0
0
No~
oowoNoo
mom
omo*
d
dd
*
aOmoo
.....
dP000
CnPQ\
OO
dd
moo00
.....
a30000
mo
0
dd
A
c,
k
Od
za
'
2;
k
c,
c,
rnd
g8
3­
9
Ln
­
f
0000
....
om00
NO
ri
­
4
PIrnaOOOC
t
......
(
6
WmmOOO
at
wQ)
Q\
o
0
%
rid
A
8
U
3
2
k
OOrnOOOO
U
......
c
PrnOOOO
Ccm0
Pi
0
0
A
ma
­
0
­+
*
n
0
.*

E
(
6
x
*
I
Ln
d'

U
N
.*
0
2
A
B
rlcvmm
ri­
rcilmmmm
1
.
4.0
CHgMISTRY
The
50
composite
samples
in
the
FY86
"
ATS
were
prepared
by
MRI
in
the
analysis
laboratory
for
determination
of
semivolatile
compounds
using
high­
resolution
gas
chromatography/
mass
spectrometry
(
MGC/
MS).
The
performance
of
the
analysis
effort
was
demonstrated
through
recoveries
of
surrogate
compounds
and
internal
quantitation
standards
(
IQS),
as
well
as
through
analysis
on
20
QC
samples
(
method
blanks,
control
tissue
samples,
and
spiked
control
tissue
samples).
Section
4.1
discusses
the
various
steps
in
the
analytical
procedure,
including
how
results
are
quantified.
Section
4.2
presents
the
QA/
QC
methods
that
were
implemented.
The
presentation
of
the
results
for
analysis
of
QC
samples
is
primarily
relegated
to
Chapter
5.
Section
4.3
presents
data
quality
objectives
established
for
the
laboratory
analytical
method
and
the
extent
to
which
these
objectives
were
met.

4.1.
ANALYTICAL
PROCEDURES
The
analytical
procedures
performed
in
the
FY86
NHATS
included
the
extraction
and
cleanup
of
the
composite
tissue
samples
using
Gel
Permeation
Chromatography
(
GPC)
and
Florisil
column
fractionation,
the
analysis
by
HRGC/
MS,
and
the
quantitation
of
results.
A
flow
diagram
of
these
activities
is
found
in
Figure
4­
1.
Each
of
these
procedures
is
described
in
detail
below.

4.1.1.
Sanmle
Prevaration
The
preparation
of
the
cornposited
adipose
tissue
Specimens
for
determination
of
semivolatiles
required
a
multistep
Procedure.
The
stages
of
this
procedure
include
quantitative
extraction
and
cleanup
through
several
chromatographic
columns.
These
stages
are
described
below.

4­
1
II
Human
Adipose
Tissue,
20
g
..
.

I
1
Add
Stable
Isotope­
Labeled
Surrogate
1
Compounds
~

Ii
Extraction
=
Tissuemizer
1
Bulk
Lipid
Removal
Gel
Permeation
Chromatography
1
Florisil
Fractionation
200
mL
6%
ethyl
ether
/
hexane
300
mL
50%
ethyl
ether
/
hexane
HRGC/
MS
(
Scanning)
0.01
=
0.1
pg/
g
(
PCBs,
OCl
Pesticides,
etc.)

I
Quantitation
/
Data
Transfer
I
I
1
..

I
Figure
4­
1.
Flow
Scheme
for
Analysis
of
Semivolatile
Com~
pounds
in
the
FY86
"
ATS
4­
2
4.1.1.1.
Extraction.
After
the
compositing
stage
(
Chapter
3),

the
adipose
composites
were
stored
at
­
100~
in
50­
m~
culture
tubes
sealed
with
aluminum
foil.
To
begin
the
sample
extraction
procedure,
the
samples
were
allowed
to
come
to
room
temperature
and
then
fortified
with
200
pL
of
the
surrogate
spiking
solution,
Spiked
control
QC
samples
were
fortified
with
50
pL
and
200
pL
of
the
native
compound
spiking
solution
for
the
low­
and
high­
dose
samples,
respectively.
Ten
milliliters
of
methylene
chloride
was
added
and
the
sample
homogenized
for
1
min
with
a
Tekmar
Tissuemizer.
The
mixture
was
allowed
to
separate,
and
the
methylene
chloride
was
decanted
through
a
funnel
of
5
to
10
g
of
sodium
sulfate
into
a
200­
mL
volumetric
flask.
The
funnel
was
rinsed
with
10
mL
of
methylene
chloride
into
the
volumetric
flask.
The
homogenization
was
repeated
two
times
with
fresh
10mL
portions
of
methylene
chloride.
The
culture
tube
was
rinsed
with
additional
methylene
chloride
and
the
remaining
contents
of
the
tube
transferred
to
the
funnel.
Finally,
the
funnel
was
rinsed
with
additional
methylene
chloride
until
the
volumetric
flask
was
brought
up
to
volume
(
200mL).

4.1.1­
2.
Lipid
Determination.
At
this
point
the
flask
was
stoppered,
inverted
several
times
to
mix
the
extract,
and
1
mL
was
removed
with
a
disposable
pipet
and
placed
into
a
preweighed
(
mea,
suredto
0.0001
g)
1­
dram
glass
vial.
The
methylene
chloride
in
the
vial
was
reduced
under
nitrogen
until
an
oily
residue
remained.
The
weight
of
the
lipid
was
obtained
by
difference,
and
the
percent
lipid
for
the
composite
was
calculated
and
recorded.

4.1.1.3­
Extract
Concentration,
The
remaining
portion
of
the
extract
(
99
mL)
was
quantitatively
transferred,
with
a
30­
to
40mL
rinse,
to
a
500­
mL
Kuderna­
Danish
evaporator
equipped
with
a
20­
mL
receiver,
One
or
two
clean
boiling
chips
and
a
three­
ball
Snyder
column
were
added
to
the
flask.
The
Snyder
column
was
>

prewet
with
1
mL
of
methylene
chloride
and
the
volume
reduced
to
15
to
25
mL
over
a
steam
bath.
The
apparatus
was
removed
from
the
steam
bath
and
allowed
to
cool.
The
flask
and
joint
were
rinsed
with
5
mL
of
methylene
chloride
into
the
receiver.
The
extract
was
then
quantitatively
transferred
to
a
40­
mL
sample
vial
with
a
TFE­
lined
screw
cap,
adjusting
the
volume
to
approximately
40
mL
with
methylene
chloride.

4.1,2.
Cleanup
Procedure
4.1.2.1.
Gel
Permeation
Chromatography.
GPC
columns
were
packed
with
60
g
of
Bio­
Beads
SX­
3
that
had
been
allowed
to
swell
overnight
in
methylene
ch1oride:
cyclohexane
(
50:
SO).
The
beads
were
allowed
to
settle
to
form
a
uniform
packing.
Solvent,
methylene
ch1oride:
cyclohexane
(
50:
50),
was
pumped
through
the
column
at
a
flow
rate
oz
5
mL/
min.
After
air
had
been
displaced
from
the
column,
the
pressure
was
adjusted
to
5
to
15
psi.
The
GPC
column
was
then
calibrated
using
a
solution
of
approximately
1
mg/
mL
butyl
benzyl
phthalate,
1
mg/
mL
4­
nitrophenol
and
390
mg/
mL
extracted
bulk
human
lipid
in
methylene
chloride.
The
calibration
resulted
in
a
GPC
program
that
.
provided
135
mL
(
27minutes)
of
eluent
with
lipid
directed
to
a
discard
fraction,
followed
by
a'
225mL
(
45
minute)
collection
period.
This
was
the
chromatographic
pattern
established
from
the
elution
of
the
butyl
benzyl
phthalate
through
the
elution
of
4­
nitrophenol.
An
additional
wash
time
of
50
mL
(
10
minutes)
was
included
to
prevent
sample
carryover.
Prior
to
loading
the
GPC,'
thesample
collection
tubes
and
injector
port
were
rinsed
with
acetone,
methylene
chloride,
toluene,
and
hexane.
Syringes,
beakers,
and
filters
were
washed
with
soap
and
water,
rinsed
with
water,
deionized
water,
acetone,
methylene
chloride,
toluene,
and
hexane.
All
extracts
were
drawn
through
a
Millipore
stainless
steel
Swinney
filter
with
a
0.5­
pm
type
FH
membrane.
Sample
loops
were
rinsed
with
5
mL
of
methylene
ch1oride:
cyclohexane
(
50:
50)
and
loaded
with
2
mL
of
the
sample
extracted
followed
by
3
mL
of
solvent.
One
loop
between
each
composite
was
used
as
an
eluent
blank.
The
cleaned
extracts
were
collected
in
clean
4­
L
amber
solvent
bottles.

4.1,2.2.
GPC
Eluent
Concentration.
The
cleaned
extracts
from
the
combined
GPC
effluent
was
concentrated,
using
500­
or
1000­
a
Kuderna­
Danish
(
K­
I))
evaporators,
to
approximately
10
mL.
The
Snyder
column
was
prewet
with
methylene
chloride
and
a
new
boiling
chip
added
with
addition
of
eluent.
When
all
the
eluent
.
wasconcentrated
to
5
to
10
mL,
the
apparatus
was
allowed
to
cool.
If
the
extract
remained
highly
colored
or
viscous,
the
sample
was
quantitatively
loaded
onto
the
GPC
and
reprocessed
in
three
to
four
loops.
Then
the
extract
was
reconcentrated
and
transferred
to
Florisil
as
follows.
If
the
sample
extract
appeared
clean,
SO
mL
of
hexane
was
added.
The
Snyder
column
was
replaced
and
prewet
with
1
mL
of
hexane.
The
volume
was
reduced
to
10
mL
and
the
flask
and
lower
joint
rinsed
with
1
to
2
mL
of
hexane
into
the
concentrator
tube.
The
extract
was
then
concentrated
to
approximately
1
mL
under
a
gentle
stream
of
purified
nitrogen.

4.1.2.3,
Flotieil
Column
Cleanup,
A
25­
x
300­
mm
chromatographic
column
with
solvent
reservoir
and
TFE
stopcock
was
prepared
by
packing
the
bottom
with
a
small
wad
of
silanized
glass
wool
and
rinsing
with
20
mL
of
hexane.
A
100­
mL
aliquot
of
hexane
was
added
to
the
column.
The
precleaned
Florisil
was
allowed
to
COOL
in
a
desiccator,
and
12.5
grams
were
transferred
to
the
column.
When
the
Florisil
had
settled,
sufficient
anhydrous
sodium
sulfate
was
added
to
achieve
a
one­
half
inch
layer
on
top
of
the
Florisil.
The
hexane
was
drained
just
to
the
top
of
the
anhydrous
sodium
sulfate
layer.
The
extract
was
transferred
to
the
top
of
the
column.
The
extract
receptacle
was
rinsed
with
three
successive
2­
to
3­
mL
aliquots
of
hexane,
adding
the
rinses
to
the
column.
A
500­
mL
K­
D
flask
and
receiver
were
placed
under
the
column,
and
the
sample
was
drained
onto
the
column
until
the
4­
5
anhydrous
sodium
sulfate
was
nearly
exposed.
The
column
was
eluted
with
200
mL
of
6%
ethyl
ether
in
hexane
(
v/
v)
(
Fraction
1)
at
a
rate
of
about
5
mL/
min.
The
X­
D
flask
and
receiver
were
replaced
with
another
IC­
D
flask
and
receiver.
The
column
was
eluted
with
300
mb
of
50%
ethyl
ether
in
hexane
{
v/
v>
(
Fraction
2)

The
fractions
were
concentrated
to
approximately
10
mL
using
hexane
to
prewet
the
Snyder
column.
The
flask
and
lower
joint
were
rinsed
with
1
to
2
mL
of
hexane.
The
receiver
was
then
placed
under
a
gentle
stream
of
purified
nitrogen
and
the
volume
reduced
to
less
than
1
mL.
If
either
fraction
remained
highly
colored,
viscous,
or
turbid,
it
was
rediluted
in
methylene
chloride
and
loaded
again
on
the
GPC.
If
the
sample
appeared
clean,
the
sample
was
transferred
to
a
clean
precalibrated
reactivial,
The
receiver
was
rinsed
with
three
1­
mL
aliquots
of
hexane,
adding
the
rinse
to
the
reactivial.
The
volume
was
reduced
to
less
than
0.5
mL,
the
vials
sealed,
and
the
samples
refrigerated.
All
6%
fractions
were
reduced
to
200
pL
under
a
gentle
stream
of
purified
nitrogen.
The
6%
fractions
were
fortified
with
200
pl
of
an
internal
quantitation
standard
(
IQS)
solution
and
the
volume
returned
to
200
pL
under
a
gentle
stream
of
purified
nitrogen.
The
IQS
solution
included
naphthalene­
dg,
anthracene­
dlo,
and
benzo[
a]
anthracene­
d12.
An
aliquot
of
each
sample
was
transferred
to
an
autosampler
vial
and
submitted
for
HRGC/
MS
analysis.
The
50%
fractions
were
further
reduced
under
a
gentle
stream
of
purified
nitrogen.
The
50%
fractions
were
further
reduced
under
a
gentle
stream
of
purified
nitrogen.
A
white
precipitate
formed
in
some
samples.
The
volume
was
reduced
to
200,
400,
or
600
pL,
depending
upon
the
volume
of
precipitate.

An
aliquot
of
the
IQS
solution
equal
to
the
sample
volume
was
c
added,
and
then
the
samples
were
concentrated
to
the
same
volume
4­
6
...
.
1
they
each
had
prior
to
addition
of
the
IQS
solution.
sample.
wassubmitted
for
HRGC/
MS
analysis.
An
aliquot
of
4.1.3.
Analysis
Procedures
The
quality
assurance
program
plan
for
the
FY84
and
­
86
NHATS
analysis
of
composite
samples
(
Stanley
et.
al.,
1986)

describes
in
detail
the
analytical
methodology
for
the
HRGC/
MS
analysis
of
semivolatiles
in
the
FY86
NHATS.
Additional
information
related
to
the
method
can
also
be
found
in
USEPA
(
1986)

surveys
are
discussed
in
Chapter
8.

relevant
to
the
FY86
approach
are
included
below.
Specific
differences
in
the
methods
between
these
three
Sections
of
these
reports
At
the
beginning
of
each
day
that
analyses
were
per
formed,
the
analyst
verified
that
the
instrument
was
properly
calibrated
through
analysis
of
decafluorotripheylphosphine
(
DFTTP,
see
Section
4.2.1).
The
analyst
documented
whether
the
DFTTP
criteria
were
satisfied.
Prior
to
beginning
analysis,
a
hexane
blank
was
injected
to
document
system
cleanliness.
If
any
evidence
of
system
contamination
was
found,
then
another
hexane
blank
was
analyzed.
Two
microliters
(
determinedto
nearest
0.1
pL)
of
the
spiked
sample
extract
were
injected
into
the
HRGC/
MS
system
using
a
splitless
injection
technique.
The
syringe
was
carefully
cleaned
between
injections
by
the
following
procedure
to
prevent
carryover
of
contaminants:

I
Rinse
the
syringe
10
times
with
hexane;

I
Fill
the
syringe
with
toluene
and
sonicate
syringe
and
plunger
in
toluene
for
5
min
and
repeat
at
least
twice;

I
Rinse
the
syringe
10
times
with
hex\
ane.

After
applying
this
procedure,
the
syringe
was
ready
for
use.
Instrument
performance
was
monitored
by
examining
and
If
these
areas
recording
the
peak
areas
for
the
three
IQS.
decreased
to
less
than
50%
of
the
calibration
standard,
then
sample
analyses
were
stopped
until
the
problem
was
found
and
corrected.
The
recommended
HRGC/
MS
operating
conditions
for
the
semivolatile
organic
compounds
are
listed
in
Table
4­
1:

Table
4­
1.
Reconmended
FiRGC/
MS
Operating
Conditions
Column
temperature
column
Injector
temperature
HRGC/
MS
interface
Carrier
gas
Injector
technique
Electron
energy
Mass
range
6OoC
(
2
min)
then
10
°
C/
min
to
310
°
(
10
min)
25OOC
3OOOC
Helium
at
30
cm/
sec
2
pL,
splitless
with
a
45­
second
delay,
a
split
flow
of
30
mL/
min,
and
a
septum
purge
of
5
mL/
min
70
eV
(
nominal)
40­
550amu
4.1.4.
QuantitatiodData
Reduction
In
this
subsection,
the.
procedures
for
the
data
reduc
tion
are
outlined
for
the.
analysisof
data
from
the
HRGC/
MS
method
for
semivolatile
compounds.
The
data
for
each
sample
were
interpreted
with
computer­
assisted
quantitation
routines.
A
mass
spectral
library
and
quantitation
list
of
the
target
analytes
based
on
relative
retention
times
and
the
primary
characteristic
ion
were
used
to
search
each
data
file.

4.1.4.1.
Qualitative
Identification.
The
quantitation
routine
identified
positive
responses
based
on
the
primary
or
secondary
characteristic
ion
for
each
of
the
analytes.
Table
4­
2provides
a
list
of
these
analytes
(
native
compounds,
surrogates,
and
IQS),

along
with
the
primary
and
secondary
quantitation
ions
used
for
compound
characterization.
_.
_
I
4
Q)
al
8
0
cdk
tE
9
9
4
3
­

0
0
4
VI
0
0
0
0
_.

0
m
N
d
d
'
I
I
*
m
cy
cy
d
d
­

h
A
OD
OD
In
VI
Y
Y
VI
m
W
cd
cy
­­

A
h
W
0)
10
VI
Y
Y
c­
VI
m
10
N
cy
­

h
n
0
0
0
0
l­
l
d
Y
Y
VI
m
m
W
cy
cy
­

nn
E:
­
rl
k
.?
a
4­
9rl
PI
a,
­
4oi
n
­

01
­

C
r
C
C
­

C
P
I­

C
01
a
­

c
*
­­
r
*
a!
cv
­

W
l­
Y
0
ar
cy
­

h
0
0
d
Y
a
o\
cy
­

h
0
k
2
s
­
ri
I
cy
TT
Y
I
0
k
0
2
id
k
u
cu
rn
­
I
I
­

c
c
.
r
C
r
r
r
a
a
..
r
Q
U
­
­­
0
d
0
0
0
C
1
0
1
VI
c
jI
c
­
r
­
c
C
U
C
E
r
I­

­

c
c
U
0
c
­
r
I­
n
C
C
c
Fr
­

c
L
C
a
aC
­
ri
ti
1
cy
1
a
0
(
Y
m
­
4
>

9
%

­
4
3
k
0
PI
A
V
Id
k
+
I
al
JJ
1
5
*
.*
.­
el
.
a,
m
3
I
k
c!
x
u
a
2
­

T
I
1
rl
r
u)
a
,
VI
C
cI
OD
u)
v
03
0
m
­.

n
c­
OD
Y
(
Y
li
In
­.

n
0
0
li
V
0
d
ln
4­
12
I
m
a,
W
m
2
0
4
U
m
.
d
k
a,
U
u
E
4
9
w
0
3
i
3
9 
$!
.
d
U
Id
rl
a,
k
a,

9
U
8
m
al
k
8
k
m
a,

m
B
3
k
ld
a
E
­
I­!

m
al9
4­
13
 

rl
9
­
s
The
following
criteria
based
on
Table
4­
2
must
have
been
met
in
order
to
make
a.
qualitative
identification:

8
The
characteristic
masses
of
each
parameter
of
interest
must
maximize
in
the
game
scan
or
within
one
scan
of
each
other.

rn
The
retention
time
must
fall
within
kt10
seconds
of
the
retention
time
of
the
authentic
compound.

rn
The
relative
peak
heights
of
the
three
characteristic
masses
in
the
EICPs
must
fall
within
f30%
of
the
relative
intensities
of
these
masses
in
a
reference
mass
spectrum.
The
reference
mass
spectrum
can
be
obtained
from
a
standard
analyzed
in
the
GC/
MS
system
or
from
a
reference
library.

rn
The
response
for
each
of
the
Characteristic
ions
must
be
at
least
2,5
times
the
background
signal­
to­
noiseratio.

4.1.4.2.
Quantitation.
Data
were
quantitated
on
the
internal
standard
method.
IQS
were
paired
with
each
analyte
for
quantita
tion
purposes;
these
pairings
are
displayed
in
Table
4­
2.

Relative
response
factors
(
RRFs)
for
native
 
quantitativet1
semivolatile
compounds
were
calculated
from
the
data
obtained
during
the
analysis
of
calibration
solutions
using
the
following
formula:

RRF
=
Asm
c*
s
AIS
csm
(
4­
1)

where
Am
=
The
area
of
the
primary
quantitation
ion
for
the
analyte
in
question,
AIS
=
The
area
of
the
primary
quantitation
ion
for
the
labeled
IQS
paired
with
the
given
analyte,
Cm
=
Concentration
(
ng/
pL)
of
the
analyte,
and
CIS
=
Concentration
(
ng/
pL)
of
the
IQS.

Once
the
RRF
values
were
obtained,
the
lipid­
adjusted
concentration
of
a
semivolatile
analyte
within
an
adipose
tissue
sample
(
Cmpl,)
was
calculated
as
follows:

4­
14
where
RRF
was
determined
from
the
calibration,
=
The
area
of
the
primary
quantitation
ion
for
the
analyte
in
question
within
the
sample,
As
=
The
area
of
the
primary
quantitation
ion
for
the
labeled
IQS
paired
with
the
analyte,
=
The
amount
(
total
ng)
of
the
labeled
IQS
added
to
the
sample
prior
to
extraction,
WAT
=
Weight
(
9)
of
the
odginal
adipose
tissue
sample,
and
LC
=
Percent
extractable
lipid
from
the
sample.

4.1.4.3.
Recovery
of
Surrogate
Standards.
Recoveries
of
the
labelled
surrogate
standards
measured
in
the
final
extract
were
calculated
using
the
following
formula:

Ass
QIS
100%
%
Recovery
=
AIS
*
Qss
MFSS
(
4­
3)

where
&
and
Qs
are
defined
above,
ASS
=
Area
of
the
primary
quantitation
ion
determined
for
the
surrogate
standard,
Qss
=
Amount
(
ng)
of
the
surrogate
standard
added
to
the
sample
prior
to
extraction,
RRF
for
the
surrogate
standard
relative
to
its
IQS,
as
determined
from
the
initial
calibration.
and
RRFSS
=

4.1.4.4.
Data
Qualifiers.
Quantitative
data
were
classified
to
indicate
the
intensity
of
the
signal
response.
For
quantitative
compounds,
the
qualifiers
were
defined
as
follows:

a
Not
Detected
(
ND)
:
S/
N
ratio
less,
than
2.5.
Trace
{
TR):
S/
N
ratio
at
2.5
or
above,
but
less
than
10I)
Positive
Quantifiable
(
PQ):
S/
N
ratio
at
10
or
above.

The
semivolatile
compounds
described
as
llqualitativeanalytes'l
in
the
FY86
NHATS
were
not
quantitated
beyond
a
one­
significant­
figure
estimate.
A
"
positive
detect"
(
PD)
was
reported
for
analytes
that
met
'
thequalitative
criteria.

4.1.4.5.
Estimating
the
Method
Limit
of
Detection.
A
method
limit
of
detection
(
LOD)
was
estimated
for
a
given
sample
in
the
following
situations
for
a
specific
analyte:

I
no
response
was
noted
for
the
analyte;
a
response
was
noted
but
the
ion
ratios
were
incorrect;
I
a
response
was
noted
but
was
below
the
calibration
range;
or
I
the
reported
response
was
quantitated
as
a
trace
value.

If
no
response
was
noted,
the
LOD
was
reported
as
the
lower
end
of
the
established
calibration
range.
The
LOD
value
was
reported
as
total
ng/
injection
such
that
the
LOD
could
be
extrapolated
for
each
individual
sample.
For
samples
for
which
a
response
at
the
compound's
retention
time
was
noted
but
the
qualitative
criteria
for
ion
ratios
were
outside
an
acceptable
range,
the
estimated
LOD
was
calculated
as
the
response
of
the
interference,
and
the
concentration
value
was
regarded
as
not
detected
(
ND).
If
a
response
was
noted
at
the
correct
retention
time
and
met
the
qualitative
criteria
of
ion
ratio
agreement
for
identification,
but
the
calculated
response
was
below
the
calibration
curve,
then
the
value
was
identified
as
not
detected.
If
a
response
was
qualified
as
a
trace
value,
then
the
f
analyst
also
provided
an
estimated
LOD.
This
was
accomplished
by
using
the
observed
signal­
to­
noiseratio
on
either
side
of
the
response
or
the
lower
calibration
limit,
whichever
was
higher.
4.2.
QA/
ClC
FOR
CHEMICAL
ANALYSIS
4.2.1.
Demonstratinu
Achievement
of
Instrument
Performance
Reauirements
Achievement
of
the
instrument
performance
requirements
were
demonstrated
in
the
following
stages:

(
1)
MGC
Column
Performance
A
30­
m
HRGC
column,
DB­
5,
film
thickness
=
0.2
pm,
was
used
for
analysis
of
all
samples
and
standards
for
the
6%
fraction
extracts,
and
a
30­
m
DB­
1301,
film
thickness
=
0.2
pm,
was
used
for
all
50%
fraction
extracts.
The
column
performance
was
initially
demonstrated
using
a
Grob
hydrocarbon
mixture.
The
retention
times
should
be
within
*
30%
of
the
values
supplied
by
the
manufacturer
with
the
column
when
chromatographed
under
similar
conditions.
If
during
the
course
of
the
analysis
it
became
necessary
to
install
a
new
column,
this
column
was
verified
in
a
similar
manner.

(
2)
Tunina
and
Mass
Calibration.
The
mass
spectrometer
was
tuned
at
least
daily
to
yield
optimum
sensitivity
using
decafluorotripheylphosphine
(
DFTTP).
The
criteria
that
must
be
met
axe
listed
in
Table
4­
3.
Corrective
actions
were
implemented
whenever
the
resolving
power
did
not
meet
the
requirement.
Examples
of
these
corrective
actions
are
recalibrating
the
mass
Spectrometer,
changing
the
GC
column,
or
maintenance
of
the
instrument.
Corrective
actions
were
determined
by
consultation
between
the
analyst,
the
work
assignment
leaderts),
and
the
mass
spectrometry
facility
staff.

(
3)
RRF
Check
and
Instrument
Sensitivitv
Check.
As
part
of
the
initial
and
routine
instrument
performance
checks,
a
single
calibration
standard
Mas
analyzed
and
RRF
values
of
the
respective
analytes
were
compared
to
specific
internal
standards.
The
initial
and
routine
calibration
criteria
require
that
the
4­
17
Table
4­
3.
DFTTP
Key
Masses
and
Abundance
Criteria(')

51
68
69
I
70
127
8%­
82%
of
mass
198
~
2%
of
mass
69
11%­
91%
of
mass
198
~
2%
of
mass
69
32%­
59%
of
mass
198
198
base
peak,
100%
abundance
199
275
441
442
I
h
II
443
I
EPA
Method
1625
Revision
B:
Dilution
GC/
MS,
January
1985.

.,..
...
Sj
f
f
4%­
9%
of
mass
198
11%­
30%
of
mass
198
44%­
110%
of
mass
443
30%­
86%
of
mass
198
14%­
24%
of
mass
442
Semivolatile
Organic
Compounds
by
Isotope
f
1
precision
of
the
RRF
measurements
are
*
30%
for
the
target
analytes.
Sensitivity
of
the
MS
was
documented
through
the
responses
noted
for
the
first
calibration
standard
of
each
analysis
day.
The
method
requires
that
a
low
level
standard
be
analyzed
to
document
sufficient
instrumental
response
to
support
instrumental
detection
limits.
Routine
checks
on
the
instrumental
sensitivity
were
achieved
by
monitoring
the
response
for
the
IQS
from
injection
to
injection
and
documenting
the
responses
in
the
MS
log
book.
If
the
response
for
the
IQS
was
noted
to
drop
by
greater
than
50%
of
the
response
noted
in
the
previous
calibration
standard,
the
analyst
verified
instrumental
performance
through
the
analysis
of
an
additional
calibration
standard.
The
qualitative
analytes
in
the
FY86
"
ATS
were
identified
by
relative
retention
times
and
characteristic
mass
peaks.
These
met
the
same
qualitative
identification
factors
as
the
quantitative
targets
but
were
not
quantitated
beyond
a
onesignificant
figureestimate.
The
RRFs
for
the
compounds
were
not
a
required
factor
in
the
initial
calibration
and
daily
performance
checks.
A
'!
positive
detect"
(
PD)
was
reported
for
analytes
that
met
the
qualitative
criteria
in
Section
4.1.4.

4.2.2.
Calibration
for
Quantitative
Semivolatile
Analysis
4.2.2.1.
Initial
Calibration.
Initial
calibration
was
required
before
any
samples
were
analyzed,
or
when
any
routine
calibration
did
not
meet
the
required
criteria
for
the
consistency
of
RRFs
(~
30%
for
quantitative
targets
and
internal
standards).
An
initial
calibration
was
conducted
by
performing
the
following
steps:

(
1)
Tuning
and
calibrating
the
instrument
with
PFK
and
DFTTP.

4­
19
Table
4­
4.
Calibration
Solutions
for
the
6%
Flotieil
Fraction
Lindane
(
T­
BHC)
100
so
10
5
1
Mirex
100
so
10
5
1
Chlordane
100
50
10
5
1
Oxychlordane
100
50
10
5
1
Aldrin
100
50
10
5
1
CU­
BHC
100
50
10
5
1
I
A­
BHC
100
50
10
5
1
@­
BHC
100
50
10
5
1
Heptachlor
epoxide
Heptachlor
100
100
50
so
10
10
I
5
5
1
1
P,
P'­
DDT
100
50
10
5
1
0,
p'
­
DDT
100
so
10
5
1
p,
p'­
DDE
100
50
10
5
1
0,
p'
­
DDE
100
50
10
5
L
0,
p'
­
DDD
100
50
10
5
1
p,
p'
­
DDD
t­
Nonachlor
.
100
50
10
5
1
1,3­
Dichlorobenzene
100
50
10
5
1
1,4­
Dichlorobenzene
100
50
10
5
1
lI2­
Dichlorobenzene
100
50
10
5
1
1,2,4­
Trichlorobenzene
100
50
1
10
5
1
l12,3­
Trichlorobenzene
100
50
10
5
1
1,3,5­
Trichlorobenzene
100
50
10
5
1
1,2,3,4­
Tetrachlorobenzene
100
__
50
10
5
2
1,2,3,5­
Tetrachlorobenzene
100
50
10
5
1
1,2,4,5
Tetrachlorobenzene
100
50
10
5
l.

Pentachlorobenzene
100
I
50
10
5
1
Hexachlorobenzene
100
50
10
5
1
Naphthalene
100
50
10
5
1
Phenanthrene
100
50
10
5
1
4­
20
Tabla
4­
4.
(
Coat.)

Fluoranthene
.­
100
50
10
I
5
1
Chrysene
100
50
10
5
1
Benzotalpyrene
100
50
IO
5
1
Acenaphthylene
100
50
10
5
1
Acenaphthene
100
50
10
5
1
Fluorene
100
50
10
5
1
Biphenyl
I
100
50
10
5
1
1,2­
Dibromo­
3­
chloropropane
I
100
1
50
I
10
I
5
1
10
5
100
10
Octachlorostyrene
50
10
5
1
Tetrabromobiphenyl.
100
50
10
5
1
o­
Cymene
100
50
10
5
1
m­
Cymene
100
50
10
5
1
100
50
10
5
Pyrene
100
50
10
5
1
­­~

Hexachlorobutadiene
(
HCBD)
100
50
1
Hexachlorocyclopentadiene
50
5
1
100
p­
Cymene
1
4­
21
Analyzing
the
five
concentration
calibration
solutions
for
the
6%
fracton
eluates
listed
in
Table
4­
4.
The
low
concentration
solution,
CS5,
was
used
to
demonstrate
the
lower
limit
of
detection
provided
by
the
available
instrument.

Computing
the
RRFs
for
each
analyte
in
the
concentration
calibration
solution
using
the
criteria
for
positive
identification
of
semivolatile
analytes
and
the
computational
methods
given
in
Section
4.1.4.

Computing
the
means
and
their
respective
relative
standard
deviations
(
RSD,
expressed
as
a
percentage)
for
the
RRFs
for
each
analyte
in
the
standard.
The
RSD
was
calculated
as
the
standard
deviation
to
all
measurements
of
a
particular
RRF
value
divided
by
the
average
RRF
value
and
multiplied
by
100%.
These
samples
were
identified
in
the
individual
batch
reports.

Repeating
the
above
process
for
the
50%
Florisil
fraction
eluates
(
Table
4­
5)
and
PCB
calibration
solution
(
Table'
4­
6)
.

The
above
fractionation
was
based
on
the
previous
broad
scan
analysis
of
adipose
tissue.
In
the
case
of
pant­
itative
:
a
analytes
not
previously
determined,
comparisons
to
similar
Bcompounds
have
been
made
for
the
purpose
of
determining
in
which
t
:;

Florisil
fraction
the
analyte
was
most
likely
to
appear.
To
declare
an
acceptable
initial
calibration,
the
RSD
$
.3
for
the
response
factors
for
the
analysis
of
analytes
across
the
:
F
­$
­
4calibration'rangemust
have
been
less
than
*
30%.
If
this
3
r
criterion
held,
then
the
RRF
was
assumed
to
be
nonvariant
and
the
.$

average
RRF
could.
beused
for
calculating
a
RSD
value.
Alter'­
natively,
the
results
were
used
to
p1ot.
acalibration
curve
of
response
ratios,
A3/
AiS
versus
RRF.

An
acceptable
initial
calibration
also
required
the
traces
for
all
ions
used
for
quantitation
to
present
a
signal­
to
noise
(
S/
N)
ratio
of
at
least
2.5.
This
included
analytes
and
isotopically
labeled
standards.
Isotopic
ratios
must
have
been
within
+
30%
of
the
theoretical
'
values.

4­
22
Table
4­
5.
Calibration
Solutions
for
the
50%
Florisil
Fraction
Dimethyl
phthalate
100
50
I
10
I
5
1
Dibutyl
phthalate
100
50
10
5
1
Butylbenzyl
phthalate
100
so
10
5
1
Di­
n­
octylphthalate
100
50
10
5
1
Diethyl
phthalate
100
so
10
5
1
Di­
n­
butylphthalate
100
so
10
S
1
I
I
Tributyl
phthalate
100
50
10
5
1
Diethylhexylphthalate
(
DEHP)
100
so
10
S
1
Tributylphosphate
so0
250
SO
25
5
Triphenylphosphate
200
100
20
10
2
Tris(
2­
chloroethy1)
phosphate
500
250
50
25
5
Tributoxyethylphosphate
200
100
20
10
2
1
Tritolylphosphate
200
100
20
10
2
Tris(
dichloropropyl)
phosphate
500
250
so
2s
5
Dieldrin
500
250
50
25
5
Endrin
so0
250
50
25
5
Endrin
ketone
500
250
50
2s
5
Tris(
2,3­
dibromopropyl)­
phosphate
500
250
so
2s
5
2­
Phenylphenol
100
50
10
5
1
Trichloro­
o­
terphenyl
200
100
20
10
2
Tetrachloro­
o­
terphenyl
200
100
20
10
2
4­
Chloro­
o­
terphenyl
200
100
20
10
2
Pentachlorodiphenyl
ether
200
100
20
10
2
2­
Methoxy­
3­
methylpyrazine
200
100
20
10
2
Ethyl
hydrocinnamate
200
100
20
10
2
4­
23
Table
4­
6.
Caf;
ibration
Solutione
for
PCB
Analysis
Monochlorobiphenyl
100
50
10
5
1
Dichlorobiphenyl
100
50
10
5
1
Trichlorobiphenyl'
100
50
10
5
1
11
I
I
1
Nonachlorobiphenyl
100
I
20
2
I
I
I
I
I
Decachlorobiphenyl
500
I
250
50
25
5
4­
24
4.2.2.2.
Routine
Calibrationq.
Routine
calibrations
were
performed
at
the
beginning
of
every
day
before
actual
sample
analyses
were
performed
and
as
the
last
injection
of
every
day.
Routine
calibrations
involved
the
following
steps:

(
1)
Injecting
2
pL
of
the
concentration
calibration
solutipns
CS3
for
the
6%
fraction
as
the
initial
calibration
check
on
each
analysis
day
and
as
the
final
check
on
each
analysis
day.

(
2)
Computing
the
RRFs
for
each
analyte
in
the
concentration
calibration
solution
using
the
criteria
for
positive
identification
of
semivolatiles
given
in
Section
4.1.4.

To
declare
an
acceptable
routine
calibration,
the
measured
RRF
for
all
analytes
must
have
been
within
*
30%
of
the
mean
values
established
by
initial
calibration
of
the
calibration
concentraton
solutions.
Also,
isotopic
ratios
must
have
been
within
&
30%
of
the
theoretical
value
for
each
analyte
and
isotopically
labeled
standard.

4.2.3.
Snikinu
Solution
Prenaration
4.2.3.1.
Native
Standard
Spiking
Solution.
A
native
standard
spiking
solution
was
prepared
in
dichloromethane
from
the
individual
stock
standards.
This
solution
was
used
for
preparing
laboratory
spikes
of
adipose
tissue.
For
example,
if
the
anticipated
spike
level
is
0.10
pg/
g
in
a
20­
g
sample,
the
target
analyte
should
be
added
to
the
spiking
solution
to
achieve
a
final
concentration
of
10
pg/
mL.
The
specific
PCB
isomers
used
for
preparing
calibration
solutions
were
also
included
in
the
target
spiking
solution.
The
spiking
solution
and
proposed
levels
are
listed
in
Table
4­
7.

4.2.3.2.
Surrogate
Standard
Spiking
Solution.
A
mixed
surrogate
standard
spike
solution
was
prepared
in
dichloromethane
from
the
individual
stock 
standards.
The
surrogate
standard
spike
Table
.4­
7.
Proposed
QC
Spiking
Solutione
E,
E'
­
DDE
E,
E'
­
DDT
Dieldrin
Heptachlor
epoxide
­
t­
Nonachlor
Mirex
y­
Chlordane
Hexachlorobenzene
1,2,4,5­
Tetrachlorobenzene
1,4­
Dichlorobenzene
1,2,4­
Trichlorobenzene
Diethyl
phthalate
Butylbenzyl
phthalate
Triphenyl
phosphate
Tris(
dichloroethyl1phosphate
Benzo[
a]
pyrene
Phenanthrene
Chrysene
Hexachloro­
1,3­
butadiene
­
R­
Limonene
2­
Phenyl
phenol
Coumarin
­
o­
Cymene
2­
Indanone
DL­
Isoborneol
Ethyl
hydrocinnamate
Octamethylcyclotetrasiloxane
Monochlorobiphenyl
Dichlorobiphenyl
Trichlorobiphenyl
29.5
200
I
50
28.4
200
50
21.9
200
50
14.3
200
50
21.9
21.7
200
200
I
50
50
22.3
200
50
19.5
200
I
50
28.8
200
50
124
200
1
50
20.7
200
50
23.0
200
50
22.6
200
50
19.2
200
50
372
200
50
24.1
200
50
23.6
200
50
5.07
200
50
19.6
200
50
23.4
200
50
20.7
200
50
25.2
200
50
28.0
200
50
17.3
200
50
26.7
200
50
32'.
7
200
50
21.1
200
50
25.3
200
50
27.9
200
50
24.6
200
so
4­
26
(­
y(">:*]
3
Table
4­
7.
(
coat.)
.

Tetrachlorobiphenyl
I
56.2
200
50
Pentachlorobiphenyl
65.0
200
50
­

Hexachlorobiphenyl
52.6
200
50
Heptachlorobiphenyl
130
200
50
Octachlorobiphenyl
137
200
50
Nonachlorobiphenyl
154
200
50
Decachlorobiphenyl
96.1
200
50
('
1
Final
spike
level
is
based
on
ng
of
analyte/
g
of
adipose
(
20
g
sample).
The
actual
reported
value
would
be
based
on
ng
of
analyte/
g
of
extractable
lipid.

(
3
From
EPA
Method
680
list
except
for
the
nonachlorbiphenyl
which
is
not
included
in
Method
680.

4­
27
,,.
I
....­......,
solution
were
prepared
to
deliver
the
surrogates
at
the
amounts
specified
in
Table
4­
8
in
a
200­
pL
volume.
This
requires
that
the
stock
solution
contain
the
surrogates
at
concentrations
ranging
from
10
to
50
gg/
mL.

4.2.3.3.
Internal
Standard
Spiking
Solution.
The
internal
standard
spiking
stock
solution
concentrations
are
also
listed
in
Table
4­
8
for
each
of
the
deuterated
internal
standards.

4.2.3.4.
Performance
Audit
Solutions.
Included
among
the
samples
in
at
least
two
sample
batches
was
a
solution
provided
by
the
quality
control
coordinator
containing
known
amounts
of
specific
target
analytes
representing
each
major
compound
class
to
be
determined.
The
accuracy
of
measurements
for
performance
evaluation
samples
was
in
the
range
of
70­
130%.

1<
4.2.4.
pC
Samples
3G
Samples
included
for
QC
purposes
within
each
batch
of
4
i
composite
samples
are
summarized
in
Table
4­
9.
The
order
of
4
preparation
and
analysis
with
respect
to
the
FY86
"
HATS
composi
1
ites
was
specified
in
the
sample
design.
This
section
discusses
!
each
of
these
QC
sample
types.
Discussion
of
the
findings
and
conclusions
from
QC
sample
analyses
are
presented
in
Section
5.3.

4.2.4.1.
Method
Blanks.
One
method
blank
was
generated
within
each
batch
of
samples.
A
method
blank
was
generated
by
performing
all
steps
detailed
in
the
analytical
procedure
using
all
reagents,
standards,
equipment,
apparatus,
glassware,
and
solvents
that
Lwereused
for
a
sample
analysis,
but
not
adding
any
adipose
tissue.
The
method
blank
contained
the
same
amounts
of
labeled
surrogate
standards
that
were
added
to
samples
before
bulk
lipid
cleanup.
Protocol
dictated
that
if
the
levels
detected
in
the
method
blank
were
greater
than
10%
of
the
levels
seen
in
the
4­
28
Table
4­
8.
Spike
Levels
for
Surrogate
and
Internal
Standards(')

Surrogate
Compounds
l82,4­
Trichlorobenzene­
dg
I
3.428
Chrysene­
dlz
2.808
UC6­
i82#
4,5­
Tetrachlorobenzene
I
2.470
13C6­
Hexachlor~
benZene
1.932
'
k6­
4
­
Chlorobiphenyl
2.222
l3Cl2­
3,3
,4
4'­
Tetrachlorobiphenyl
I
4.016
'
3c1z­
28
2
8
38
3
,58
5
,6,6
­
0ctachlorobiphenyl
6.852
'
3C12­
Decachlorobiphenyl
12.20
Diethyl
phthalate­
3,4,5,6­
d)
2.252
Di­
n­
butyl
phthalate­
3,4,5
6­
40)
1.800
Lindane
1.672
Heptachlor
I3C
I1
2.030
Internal
Standards
Naphthalene­
4
II
1.901
Anthracene­
dlo
1.910
Benzofa]
anthracene­
d,
II
2*
102
Refer
to
EPA
Method
1625,
Revision
3­­
SemivolatileOrganic
Compounds
by
Isotope
Dilution
GC/
MS,
Federal
Register
1984,
49
(
2091,
pp.
184­
197.

Concentration
calculated
for
a
solution
of
200­
pL
final
volume.

O)
Were
not
reported
in
most
samples.

4­
29
..
.
__
_
i_
"..:,
:..
Table
4­
9.
Quality
Control
Samples
Included
in
the
FY86
"
ATs
Analytical
Procedure
Method
blank
One
per
batch
Spiked
control
adipose
Two
per
batch
(
two
tissue
sample
different
spike
levels)

Unspiked
control
One
per
batch
adipose
tissue
sample
Assess
laboratory
background
contribution.

Evaluate
method
performance
(
accuracy
and
precision)

Evaluate
method
performance
(
accuracy
and
nrecision)

7
g
8
i
tissue
samples,
then
the
solvents,
reagents,
spiking
solutions,
apparatus,
and
glassware
were
checked
to
locate
and
eliminate
the
source
of
contamination
before
any
further
samples
were
extracted
and
analyzed.

4.2.4.2.
Control
Samples.
Control
samples
were
prepared
from
a
bulk
sample
of
approximately
2
kg
of
human
adipose
tissue.
This
material
was
prepared
by
blending
the
tissue
with
methylene
chloride,
drying
the
extract
by
eluting
through
anhydrous
sodium
sulfate,
and
removing
the
methylene
chloride
using
rotoevaporation
at
elevated
temperatures
(
SOOC).
The
evaporation
process
was
extended
to
ensure
all
traces
of
the
extraction
solvent
have
been
removed.
The
resulting
oi,
lymatrix
(
lipid)
was
subdivided
into
20­
g
aliquots
which
were
analyzed
with
each
sample
batch.

4.2.4.3.
Spiked
Control
Samples.
Spiked
lipid
samples
were
prepared
by
using
a
portion
of
the
homogenized
lipid.
Sufficient
spiked
lipid
matrix
was
prepared
to
provide
a
minimum
of
two
spiked
samples
per
sample
batch:
one
sample
spiked
at
a
low
concentration
and
one
at
a
high
concentration.
Method
performance
was
addressed
in
this
study
by
calculating
recoveries
for
each
spiked
sample
as
follows:

4­
30
Recovery(%)
=
conc.
(
spikedsample)
­
conc.
(
controlsample)
*
Spike
level
(
4­
4)
This
method
to
calculating
percent
recovery
leads
to
a
test
of
ruggedness
of
che
method
with
respect
to
detecting
finite
differences
in
concentration.
Note
that
an
equally­
accepted
approach
to
calculating
percent
recovery
is
given
by
the
formula
conc.
(
spiked
sample)
Recovery
(%
)
=
conc.
(
control
sample)
+
spike
level
*
100%
(
4­
5)

Formula
(
4­
5)
can
lead
to
larger
percentages
than
formula
(
4­
4)

applied
in
this
study.
This
fact
should
be
considered
when
interpreting
observed
recovery
percentages
in
this
study.
Analytical
results
of
the
QC
samples
are
statistically
summarized
in
Chapter
5.
This
chapter
also
presents
conclusions
and
issues
resulting
from
the
QC
sample
analysis.

I
4.3
OVERALL
DATA
QUALITY
At
the
outset
of
the
analysis
effort
for
the
FY86
NHATS,
specific
data
quality
objectives
were
defined
for
the
quantita
tive
and
qualitative
analyses
of
the
target
semivolatile
com
pounds.
Data
quality
objectives
were
established
for
calibration
criteria
(
relative
response
factors
[
RRFs])
for
each
analyte
and
internal
standard,
internal
standard
respdnse
area,
and
method
performance
based
on
the
recoveries
of
labeled
surrogate
com
pounds
and
native
compounds
spiked
into
a
spiked
internal
QC
sample.
The
data
generated
with
respect
to
these
criteria
are
presented
within
this
report.
Further
details
were
provided
in
the
original
data
reports.
Table
4­
10
summarizes
the
performance
achieved
versus
the
specific
criteria
and
data
quality
objectives
for
the
analysis
of
the
FY86
NKATS
composites.

4­
31
Table
4­
10.
Data
Quality
ObjectLves
for
the
FY86
"
ATS,
Along
With
Actual
Performance
RRF
calibration
Labeled
surrogate
stan
dards
Spiked
internal
QC
Sam
ples
Internal
standard
re
sponse
areas
/
f30t
all
quantitative
analykes
40%­
160%

50%­
150%

505.­
150%
of
initial
daily
calibration
stan
dard
4­
32
>
90%
of
all
RRF
factors
within
DDQs.

>
84%
for
all
labeled
surrogate
spikes;
12%
of
the
deviation
due
to
50%
fraction
surrogates.

70%
of
all
measurements
within
criteria;
22%
of
all
deviations
due
to
50%
fraction
compounds.

>
90%
of
all
measurements
within
criteria.
­­­­­
5.0
DATA
ISSUES
The
NfIATS
FY86
sampling
effort
resulted
in
a
total
of
50
composites
of
adipose
tissue
specimens
for
chemical
analysis
(
see
Chapter
3).
In
the
analytic
laboratory,
these
50
composites
were
partitioned
into
five
groups,
or
batches,
of
ten
composites
each.
Each
batch
also
included
the
following
four
laboratory
QC
samples:

8
One
method
blank
I
Three
samples
prepared
from
a
homogeneous
bulk
lipid
extract;
two
of
these
samples
spiked
at
differing
levels
by
selected
native
compounds.

Thus,
the
"
ATS
FY86
chemical
analysis
was
performed
on
five
batches
each
containing
fourteen
analytical
samples,
for
a
total
of
70
analytical
samples.
Samples
within
a
batch
were
chemically
analyzed
as
a
group
under
similar
laboratory
conditions.
Prior
to
chemical
analysis,
all
non­
blank
analytical
samples
were
spiked
with
a
set
of
twelve
surrogate
compounds.
These
labelled
compounds
do
not
exist
in
the
natural
environment
and
were
selected
to
represent
the
native
compounds
of
interest.
Analysis
of
surrogate
recovery
data
was
performed
to
evaluate
method
performance
and
overall
recovery
levels.
This
chapter
addresses
a
series
of
preliminary
data
issues
which
include
a
summary
of
the
composite
data
and
statistical
analysis
on
the
QC
data.
The
information
gathered
from
this
preliminary
data
investigation
was
essential
for
the
statistical
analysis
and
interpretation
of
sample
results.
The
objectives
of
the
preliminary
data
analysis
included
the
following:

Identify
those
compounds
having
a
sufficiently
large
percentage
of
composite
samples
with
detected
results.
Results
for
these
compounds
will
likely
reflect
more
accurate
estimates
of
average
concentration
levels
and
variability.

...
I,

.
..
.__
I..._.
4...__...
,.,
"
.
~

.
i
*"..
.~
t.
,
.,
,,
I
_*

I
Identify
the
extent
that
systematic
errors
in
measured
concentrations
are
present
over
time
by
considering
surrogate
recovery
data.
If
necessary,
adjust
the
measured
concentrations
for
these
errors.

I
Characterize
method
performance
through
analysis
of
QC
sample
data,
identifying
sources
of
variability
and
the
extent
of
batch
effects
in
the
(
adjusted)
measured
concentrations.

Each
of
these
efforts
is
documented
in
separate
subsections
which
follow.

5.1
DETERMINING
NATIVE
COMPOUNDS
TO
INCLUDE
IN
STATISTICAL
ANALYSIS
A
total
of
111
semivolatile
compounds
were
considered
in.
theFY86
"
ATS.
These
compounds
fall
into
several
chemical
classes:

I
I
I
I
I
I
8
I
I
I
m
Pesticides
(
19
compounds)
Chlorobenzenes
(
11compounds)
Phthalate
esters
(
5
compounds)
Phosphate
triesters
(
5
compounds)
PAHs
(
9
compounds)
PCBs
(
10
compounds)
Other
quantitative
compounds
(
19
compounds)
Qualitative
pesticides
(
9
compounds)
Qualitative
chlorinated
aromatics
(
9
compounds)
Qualitative
PAHs
(
4
compounds)
Other
qualitative
compounds
(
11
compounds)

Section
5.1.1
identifies
the
compounds
analyzed
within
each
chemical
class
and
the
detection
percentages
for
each
compound
as
observed
within
the
"
HATS
FY86
composite
samples.
Statistical
analysis
was
performed
only
on
compounds
with
sufficiently
high
detection
percentages.
Section
5.1.2
discusses
unique
data
reporting
for
two
pesticides
which
have
been
historically
prevalent
in
the
"
HATS
program.

5­
2
I
­­
5.1­
1
Detection
Status
of
the
Semivolatiles
When
reporting
a
measured
concentration
for
a
given
semivolatile
compound
in
a
laboratory
sample,
the
NHATS
FY86
analytical
method
determined
whether
the
compound
was
successfully
detected
in
the
sample.
For
quantitative
compounds,
the
method
classified
each
result
into
one
of
three
possible
data
qualifier
categories,
indicating
the
intensity
of
the
signal
response:

m
Not
detected
­­
Result
is
less
than
2.5
times
the
signal­
to­
noiseratio.

Trace
Result
is
between
2.5
and
10
times
the
aignal
to­
noise
ratio.

m
Positive
uuantifiable
­­
Result
is
greater
than
10
times
the
signal­
to­
noiseratio.

If
a
result
was
categorized
as
trace
or
positive
quantifiable,
the
compound
was
considered
detected
in
the
sample.
For
qualitative
compounds,
only
detected
and
not
detected
results
wdre
reported.
Estimated
method
detection
limits
were
reported
when
not
detected
or
trace
results
occurred
for
a
sample.
When
a
compound
was
not
detected
in
a
sample,
it
was
assumed
that
the
sample's
true
compound
concentration
was
at
some
level
below
the
detection
limit.
For
the
statistical
analysis,
one
half
of
the
detection
limit
was
used
as
the
estimated
concentration
level
for
not
detected
samples.
Table
5­
1
reports
the
percentage
of
FY86
composite
samples
occurring
in
each
of
the
data
qualifier
categories
for
the
111
semivolatile
compounds.
The
percent
of
composite
samples
with
detected
results
are
also
reported.
Of
the
111
compounds,
23
were
detected
in
at
least
50%
of
the
50
composite
samples,
and
one
compound
nearly
met
the
50%

threshold
(
octachlorobiphenyl,
detected
in
44%
of
the
samples).
These
24
compounds
are
identified
as
target
compounds
for
5­
3
..
.
.
.
.
.
.
.
.
.
I
Table
5­
1.
Percent
of
"
ATS
FX86
Composite
Samples
in
Each
Detection
Level
Category
Compound
Number
and
Name
CAS
t
t.
Not
t
t
Pos.
Number
Detected
Detected
Trace
Quant.

*
1
2
*
3
*
3
4
5
6
*
7
8
9
10
11
*
12
*
13
*
14
15
16
60
*
60
61
62
17
*
18
19
20
21
22
23
24
25
26
*
27
*
41
42
43
44
45
46
47
48
49
P,
P­
DDT
0,
P­
DDT
P,
P­
DDE
{
M/
Z=
288)
P,
P­
DDE
(
M/
Z=
316)
0,
P­
DDE
0,
P­
DDD
ALPHA­
BHC
BETA­
BHC
DELTA­
BHC
GAMMA­
BHC(
LINDANE)
ALDRIN
HEPTACHLOR
HEPTACHLOR
EPOXIDE
OXYCHLORDANE
TRANS­
NONACHLOR
GAMMA­
CHLORDANE
MIREX
DIELDRIN
DIELDRIN
(
CORRECTED)
EXDRIN
ENDRIN
KETONE
PESTICIDES
50­
29­
3
789­
02­
6
72­
55­
9
72­
55­
9
3424­
82­
6
53­
19­
0
319­
84­
6
319­
85­
7
319­
86­
8
'
58­
89­
9
309­
00­
2
76­
44­
8
1024­
57­
3
26880­
48­
8
39765­
80­
5
57­
74­
9
2385­
85­
5
60­
57­
1
60­
57­
1
7221­
93­
4
CHLOROBENZENES
541­
73­
1
106­
46­
7
95­
50­
1
87­
61­
6
120­
82­
1
108­
70­
3
96
4
0
96
0
100
0
0
100
0
0
100
100
0
0
100
0
100
0
0
.
0
100
0
0
0
100
0
0
92
8
2
90
0
100
0
0
4
96
0
4
0
100
0
0
0
100
0
0
94
6
0
94
78
22
2
76
92
8
0
92
0
100
0
0
32
68
2
30
12
88
0
12
62
38
22
40
0
100
0
0
2
98
2
0
0
100
0
0
86
14
0
86
0
100
0
0
0
100
0
0
0
100
0
0
0
100
0
0
0
100
0
0
0
100
0
0
0
100
0
0
0
100
0
0
98
2
4
94
84
16
8
76
0
100
0
0
0
100
0
0
0
100
0
0
8
92
8
0
2
98
2
0
0
100
0
0
4
96
0
4
0
100
0
0
1,2,3,4­
TE­
CHLOROBENZENE
634­
66­
2
1,2,3,5­
TETRACHLOROBENZENE
634­
90­
2
182,4,5­
TETRACHLOROBENZENE
95­
44­
3
PENTACHLOROBENZENE
608­
93­
5
HEXACHLOROBENZENE
118­
74­
1
PAHS
NAPHTHALENE
91­
20­
3
ACENAPHTHALENE
208­
96­
8
ACENAPHTHENE
83­
32­
9
FLUORENE
86­
73­
7
PHENANTHRENE
85­
01­
8
FLUORANTHENE
206­
44­
0
PYRENE
129­
00­
0
CHRYSENE
218­
01­
9
BEN20
(
A)
PYRENE
50­
32­
8
5­
4
T8ble
5­
1.
(
coat.)

Compound
Number
CAS
%
%
Not
%
t
POS.
and
Name
Number
Detected
Detected
Trace
Quant.

50
MO 
CHLOROBIPl3RNYL
2732­
1818
0
100
0
0
51
DICWtOROBIPHENYL
25512­
42­
9
0
100
0
0
52
 
RICHLOROBIP 
YL
25323­
68­
6
30
70
2
28
*
53
TETRACHLOROBIPfIENYL
26914­
33­
0
66
34
0
66
*
54
PENTACHLOROBIPHENYL
25429­
29­
2
88
12
0
88
*
55
 
LOROBfP IENYL
26601­
64­
9
94
6
0
94
*
56
EII~
PTACHLOROBIPHENYL
28655­
71­
2
86
14
0
86
*
57
OCTAcfILoROBIPHENYt
31472­
83­
0
44
56
0
44
58
NONAUiLOROBIPHENYL
53742­
07­
7
26
74
0
26
59
DECACHWROBIPHENYL
2051­
24­
3
28
72
0
28
PHTWUJLTE
ESTERS
63
DIMETRYL
PIITHALATE
131­
11­
3
0
100
0
0
64
DIETHYL
PIiTHALATE
84­
66­
2
10
90
2
8
*
65
DI­
N­
BUTYL
PHTHALATE
84­
74­
2
76
24
6
70
*
66
BUTYLBENZYL
PHTHALATE
85­
68­
7
72
28
4
68
*
67
BIS
(
2­
ETIIYLHEXYL)
PHTHALATE
177­
81­
7
78
22
0
78
PIIOBPEULTE
TRIESTERS
68
TRXBUTYL
PHOSPHATE
126­
73­
8
0
100
0
0
69
TRIS
(
2­
CHLOROETHYL)
115­
96­
8PHOSPHATE
0
100
0
0
70
TRIS
(
2,3­
DIBROMOPROPYL)
126­
72­
7PHOSPHATE
0
100
0
0
71
TRIPHENYL
PHOSPHATE
115­
86­
6
4
96
0
4
72
TRITOLYL
PHOSPHATE
1330­
78­
5
2
98
2
0
OTHER
28
BIPHENYL
92­
52­
4
0
100
0
0
29
1,2­
DIBROMO­
3­
CKIX)
RO
96­
12­
8
0
100
0
0PROPANE
30
HEXACHLORO
BUTADIENE
87­
68­
3
0
100
0
0
31
HEXACHLORO
CYCLOPENTADIm
77­
47­
4
0
100
0
0
32
2,2
 ,
4
 ,
S­
TETRABROMO
0
100
0
BIPHENYL
527­
84­
4
80
20
76
*
33
0­
CYMENE
*
34
D­
LIMONENE
5898­
27­
5
96
4
94
35
D,
L­
ISOBORNEOL
124­
76­
5
0
100
0
36
1­
INDANOrn
a3­
33­
0
0
100
0
37
2­
1NDA 
E
615­
13­
4
0
­
100
0
38
BUTYLATED
HYDROXYTOL~
128­
37­
0
18
82
14
39
COUMARIN
91­
64­
5
0
100
0
*
40
OCTAMETHYL­
CYCLOTETRASILOXANB
556­
67­
2
72
28
4
68
73
ETHYL
HYDROCIN 
4L~
2021­
28­
5
2
.
98
2
0
74
2­
METHOXY­
3­
MEnCm;
P tRAZINE
2847­
30­
5
0
100
0
0
5­
5
Table
5­
1.
(
cont.)

CAS
%
t
Not
k
PQS.
and
Name
Compound
Number
Number
Detected
Detected
Trace
Quant.
I
­
i
4
dTHER
(
coat.)

75
2,2',
4,4',
5­
PENTACHLORO
DIBHENYL
ETHER
0
100
0
0
76
4­
CHLORO­
P­
TERPHENYL
0
100
0
0
77
TRICHLORO­
P­
TERPHENYL
0
*
100
0
0
78
2­
PHENYL
PHENOL
90­
43­
7
24
76
2
22
PESTICIDES
(
QUALITATIVE)
9)

85
ISOPHORONE
86
DICHLOROVOS
98
CHLORPYRIFOS
99
ISOPROPALIN
100
BUTACHLOR
101
NITROFEN
102
PERTHANE
­
106
DICOFOL
107
P.
P'
­
METHOXYCHLOR
78­
59­
1
16
84
62­
73­
7
2
98
2921­
88­
2
28
72
33820­
53­
0
10
90
23184­
66­
9
12
88
1836­
75­
5
8
92
72­
56­
0
0
100
115­
32­
2
6
94
72­
43­
5
0
100
CHLORINATgD
AROMATICS
(
QUALITATIVE)(
a)

88
2,4,6­
TRICHLOROANISOLE
89
2,4,6­
TRICHLOROPHENOL
90
2,4,5­
TRIcHLOROPHENOL
91
2,3,6­
TRICHLOROANISOLE
92
2,3,6­
TRICfaOROPHENOL
95
PENTACHLOROANISOLE
96
PENTACHLORONITROBENZENE
97
2,3,4­
TRICHLOROANISOLE
110
OCTACHLORONAPHTHALENE
105
BENZO
'(
A)
ANTHRACENE
108
BENZO
(
B)
FLUORANmENE
109
BENZO
(
K)
FLUORANTHENE:
111
DIBENZO
(
A,
H)
ANTHRACENE
87­
40­
1
100
88­
06­
2
100
95­
95­
4
100
50375­
10­
5
100
933­
75­
5
100
98
82­
68­
8
100
54135­
80­
7
96
2234­
13­
1
98
56­
55­
3
26
74
10
90
207­
08­
9
4
96
53­
70­
3
0
100
OTHER
(
QUALITATIVE)(
a)

*
79
1­
NO"
E
124­
11­
8
50
50
80
CUMENE
98­
82­
8
34
66
*
81
lt214­
TRIMETHYLBENZENE
95­
63­
6
62
38
*
82
HEXYL
ACETATE
142­
92­
7
82
18
83
1,3­
DIETHYLBENZENE
141­
93­
5
8
92
84
1,4­
DIETHYLBENZENE
105­
05­
5
0
100
87
QUINOLINE
91­
22­
S
8
92
93
DIBENZOFURAN
132­
64­
9
0
100
94
CHLORDANE
2
98
­

Compound
Number
CAS
%
0
Not
0
0
POS,
and
Name
Number
Detected
Detected
Trace
Quat.

OTHER
(
QUALITATIVE)
(
cont.
1
103
CHLOROBENZYLATE
510­
15­
6
0
100
­

104
BIS
(
2­
ET 
EXYL)
ADIPATE
103­
23­
1
10
90
L
Detected
in
at
least
44%
of
the
FY86
composite
samples.

(
a)
Qualitative
compounds
were
only
monitored
for
detection
versus
non­
detection.

5­
7
statistical
ahalysis
and
are
noted
with
asterisks
in
Table
5­
1,
Statistical
analysis
of
QC
and
composite
data
was
restricted
to
these
target
compounds.
For
the
other
87
compounds,
each
having
no
more
than
a
34%
detection
rate,
results
were
summarized
through
descriptive
statistics
only.

5.1.2
Data
ReDortincr
Uniuue
to
Dieldrin
and
P,
D­
DDE
For
two
pesticides
analyzed
in
the
"
ATS
FY86
program,
two
sets
of
measured
concentrations
were
obtained
from
different
protocols.
The
two
sets
of
results
for
these
compounds,
dieldrin
and
p,
p­
DDE,
were
each
treated
as
two
distinct
entities
in
data
analysis.
The
procedures
unique
to
these
compounds
to
obtain
measured
concentrations
are
discussed
in
this
subsection.
According
to
Table
5­
1,
dieldrin
had
only
a
12%
detection
rate
among
the
FY86
composite
samples.
In
Batches
1,

3,
4,
and
5,
the
reported
concentration
levels
for
29
samples
(
including
4
QC
samples)
were
below
the
lowest
calibration
standard.
According
to
the
QAPP
for
laboratory
analysis
(
MRI,
1988b),
if
the
calculated
laboratory
response
was
below
the
range
of
calibration
standards
while
satisfying
criteria
for
retention
time
and
ion
ratio
agreement,.
thevalue
was
to
be
identified
as
a
"
not
detected"
result.
While
this
approach
was
followed
for
the
initial
set
of
reported
dieldrin
results,
the
HRGC/
MS
results
indicated
that
dieldrin
was
indeed
present
in
some
samples
whose
measured
concentrations
were
below
the
calibration
standards.
Thus
the
data'qualifier
classification
of
dieldrin
data
was
redetermined
to
reflect
the
signal­
to­
noiseratio
that
would
have
been
applied
if
the
data
were
above
the
lowest
calibration
standard.
The
quantifiable
concentrations
for
these
samples
were
recalculated
using
the
signal­
to­
noiseratio
to
define
the
detection
limit.
This
second
classification
of
the
dieldrin
data
resulted
in
a
62%
detection
rate
among
the
composite
samples,
classifying
dieldrin
as
a
target
compound
for
statistical
analysis.
Thus
statistical
analysis
for
dieldrin
was
performed
only
on
the
recalculated
results.

5­
8
ce?
T')
Q
37
Historically,
the
compound
p,
p­
DDEhas
been
detected
in
a
majority
of
"
ATS
samples.
However,
in
the
FY86
analysis,
the
primary
quantitation
ion
used
to
calculate
the
p,
p­
DDE
concentrations
(
m/
z=
288)
was
saturated
at
the
mass
spectrometry
detector.
It
is
expected
that
using
an
ion
for
quantitation
at
or
near
saturation
would
result
in
an
underestimate
of
the
true
concentration.
To
help
remedy
this
situation,
a­
secondset
of
p,
p­
DDE
concentrations
was
calculated
based
on
a
lower
response
ion
(
m/
z=
316).
The
modified
p,
p­
DDEdata
were
obtained
based
on
recalculated
calibration
curve&.
Unless
interferences
were
present
under
the
lower
response
ion,
most
of
the
modified
data
were
higher
than
the
original
data
based
on
the
primary
quantitation
ion.
Although
the
modified
p,
p­
DDEdata
values
are
likely
more
accurate
estimates
of
the
true
sample
concentrations,
most
of
these
values
were
higher
than
the
highest
calibration
standard.
This
caveat
should
accompany
any
conclusions
made
on
the
reported
p,
p­
DDE
data
from
the
FY86
NHATS.

ADJUSTING
CONCEPJTRATION
DATA
FOR
SURROGATE
RECO­
RIES
5.2
Measured
compound
concentrations
in
MIATS
composite
samples
are
generally
contaminated
by
systematic
and
random
errors.
A
potential
source
of
systematic
error
in
the
NHATS
FY86
data
has
been
identified
by
the
recoveries
of
surrogate
compounds
spiked
into
the
composite
samples.
These
recoveries
were
much
This
type
of­
higher
in
­
86
compared
with
previous
surveys.
systematic
error
can
lead
to
the
conclusion
that
measured
concentrations
for
a
compound
are
increasing
over
time,
when
in
fact
the
true
concentration
has
remained
constant
during
the
period.
Statistical
methods
for
characterizing
trend
in
compound
concentrations
should
focus
on
how
the
true
concentration
changes
over
time
rather
than
how
the
average
measured
concentration
changes.
Dinh
(
1991)
has
developed
a
statistical
technique
to
estimate
true
concentration
levels
in
tkie
"
ATS.
This
technique
used
the
recoveries
of
surrogate
5­
9
@.
K?
i:
p
0QE3
I
I
/.

compounds
to
adjust
the
measured
concentration
data
of
native
compounds.
The
result
is
a
more
accurate
representation
of
the
true
concentration
of
native
compounds
over
time.
The
"
ATS
statistical
analyses
summarized
in
this
report,
including
trends
analyses,
were
conducted
on
FY82,
FY84,
and
FY86
data
that
were
first
adjusted
by
applying
this
technique.
A
discussion
of
this
technique
follows.

5.2.1
Data
Adiustment
Method
The
statistical
technique
developed
by
Dinh
(
1991)
for
adjusting
native
compound
concentrations
was
based
on
fitting
a
systematic
errors­
in­
variablesmodel
to
the
NHATS
data
(
see
Sections
5.2.1.1
and
5.2.1.2).
This
model
predicted
the
measupd
concentration
as
a
linear
function
of
the
unknown
true
concentration.
In
turn,
the
expected
value
of
the
unknown
true
concentration
given
the
measured
concentration
was
estimated
from
the
model
fit.
This
latter
result
was
considered
an
"
adjustmentgg
to
the
measured
concentration
and
provided
a
more
accurate
estimate
of
the
unknown
actual
concentration.
To
estimate
the
expected
value
of
an
unknown
true
concentration
in
a
composite
sample,
it
was
necessary
to
obtain
accurate
characterizations
of
recoveries
and
true
concentrations
for
the
native
compounds.
This
information
was
best
represented
by
analysis
results
on
surrogate
compounds.
As
part
of
the
daily
QC
procedure,
several
surrogate
compounds
were
injected
at
known
concentrations
into
each
"
ATS
composite
sample.
Surrogate
compounds
do
not
naturally
exist
in
composite
samples;
thus
the
actual
concentration
of
a
surrogate
compound
in
a
sample
is
known
to
equal
to
the
amount
spiked
into
the
sample.
As
a
result,
the
recovery
levels
for
surrogate
compounds
provided
information
on
overall
method
performance
and
accuracy.
While
recovery
data
were
available
for
native
compounds
as
well
as
surrogate
compounds,
only
recoveries
for
surrogate
compounds
were
used
to
adjust
the
measured
concentrations
of
native
compounds.
Native
compound
recoveries
were
excluded
for
the
following
reasons:

native
compound
recoveries
can
be
affected
by
contamination
and
interferences
of
unknown
magnitude,

m
native
compound
recoveries
were
available
only
for
the
15
spiked
QC
samples,
while
surrogate
recoveries
were
available
for
all
NHATS
samples.

Each
surrogate
compound
spiked
into
an
"
ATS
composite
sample
represented
a
class
of
one
or
more
native
compounds
of
interest.
The
surrogate
compounds
and
the
native
compounds
represented
by
each
surrogate
are
listed
in
Table
5­
2.
When
possible,
a
native
compound
was
linked
directly
to
its
surrogate
counterpart,
such
as
lindane
and
chrysene.
However,
most
native
compounds
did
not
have
a
direct
surrogate
counterpart
included
in
the
spiking.
These
compounds
were
associated
with
an
average
result
across
multiple
surrogates
in
the
relevant
chemical
group.
The
methods
used
to
adjust
the
measured
concentrations
of
composite
and­
QCsamples
are
now
discussed.

5.2.1.1
Composite
Data
Adjustment.
In
this
procedure,
the
measured
concentration
of
a
compound
is
assumed
to
be
linearly
related
to
the
actual
compound
concentration
in
a
composite
sample.
Let
C
be
the
number
of
"
ATS
composite
samples
analyzed,
let
Yibe
a
measured
concentration
of
a
compound
in
the
i&
"
ATS
composite
sample
(
ill,...,
C),
and
let
gi
be
the
compound's
>

unknown
true
concentration
in
the
sample.
Then
x
ri
1
Yi
=
Rpi
+
ei
,
(
5­
1)

where
R
is
the
unknown
recovery
of
the
compound
by
the
analytical
method,
and
eiis
random
error
having
mean
zero.
Assume
that
pi
and
ei
are
normally
distributed
and
are
uncorrelated
across
the
Composites.
Then
the
expectation
of
pi
given
Yi
is
given
by
5­
11
Table
5­
2.
Matching
NHATS
FY86
Native
Compounds
with
Surrogate
Compounds
Chrvsene­
d,,
I
~~­
Tr~
chl~
obenz
I
ene­

dq11
13C6
­
1,2,4,5­
Tetrachlorobenzene
1
1
13C6
­
Hexachlorobenzene
Mean
of
above
three
surrosates
13C6
­
4­
Chlorobiphenyl
13c12­
3,3',
4,4'­
Tetrachlorobiphenyl
2,2',
3,3',
5,5',
6,6'­
Octachlorobiphenyl
13CI2­
Decachlorobiphenyl
Mean
of
above
four
surrogates
Mean
of
tetra­
and
octa
chlorobiphenyl
surrogates
t
13C
­
Heptachlor
Lindane­
d6
'

Mean
of
above
two
surrogates
I
41­
49
20­
22
23­
25
27
17­
19,
26
50
53
57
59
51,
52,
58
54­
56
12
9
1­
8,
10­
11,
13­
16,
60­
62,
85­
86.
98­
102,
106­
107
Mean
of
all
ten
surrogates
above
28­
40,
63­
84,
97­
97,
103­
105,
108­
111
5­
12
11
where
Thus
the
true
concentration
pi
in
the
i*
composite
sample
(
i=
l,...,
C)
is
estimated
by
substituting
estimates
of
the
unknown
parameters
A,
R,
and
E(
Yi)
in
equation
(
5­
2).
The
arithmetic
mean
of
the
observed
Yiacross
the
50
­
composite
samples,
denoted
by
Y,
serves
as
an
estimate
for
E(
Yi)

in
equation
(
5­
2).
Estimates
for
A
and
R
were
obtained
by
fitting
the
regression
model
in
(
5­
1)
to
measured
concentrations
of
surrogate
compounds.
Let
pi
be
the
concentration
at
which
a
surrogate
compound
is
spiked
into
composite
sample
i,
and
let
Yi
be
the
resulting
measured
concentration
of
the
surrogate
compound
in
the
sample.
Because
pi
represents
a
true
concentration,
the
linear
regression
model
in
(
5­
1)
was
fit
20
the
composite
sample
data
to
obtain
a
least­
squaresestimate
(
R)
of
the
recovery
R
for
the
surrogate
compound.
The
"
r­
squared"
value
from
the
regression
(
the
regression
sum
ofA
squares
divided
by
the
total
sum
of
squares)
is
the
estimate
(
A)
for
the
adjustment
coefficient
A.
Substituting
these
parameter
estimates
in
equation
(
5­
2)
leads
to
an
estimate
of
the
actual
concentration
in
the
composite
sample:

pi
=
[
(
l­
A)
Y++
AYi]
/
8
(
5­
3)

Thus
for
a
given
composite
sample,
Formula
(
5­
3)
represents
an
adjustment
to
the
measured
concentration
for
a
given
semivolatile
compound.

5­
13
Table
5­
3
lists
the
estimates
of
R
and
A
for
all
compounds
in
the
FY82,
FY84,
and
FY86
NfiATS
for
semivolatiles.
For
FY86,
these
estimates
are
base'd
on
only
those
composite
samples
with
a
wet
weight
of
at
least
ten
grams.
This
table
shows
the
relatively
high
recoveries
in
FY86
for
most
surrogate
A
compounds
(
valuesof
R
greater
than
one)
compared
with
the
other
fiscal
years.
Meanwhile,
the
estimated
recoveries
were
similar
for
FY82
and
FY84.
Among
the
three
fiscal
years
in
Table
5­
3,
spiked
and
measured
concentrations
for
surrogate
compound
data
were
only
available
for
the
FY84
and
FY86
NiiATS.
Thus
only
for
the
FY84
and
FY86
NHATS
could
the
parameters
R
and
A
could
be
estimated
by
fitting
the
linear
regression
model
in
equation
(
5­
1).
In
contrast,
only
recovery
data
were
available
for
surrogate
compounds
in
the
FY82
"
ATS.
As
a
result,
an
estimate
of
R
for
a
given
surrogate
compound
in
the
FY82
NHATS
was
calculated'by
averaging
the
observed
sample
recoveries.
Because
an
estimate
of
A
could
not
be
determined
from
the
available
FY82
surrogate
data,
the
corresponding
estimates
of
A
from
the
FY84
data
were
applied
to
FY82.

5.2.1.2
QC
Data
Adjustment.
A
slight
modification
to
the
approach
in
5.2.1.1
was
needed
to
adjust
the
measured
concentration
of
a
native
compound
in
an
analytical
sample
when
a
portion
of
the
concentration
in
the
sample
was
known.
This
situation
occurred
when
the
sample
was
spiked
with
a
known
amount
of
the
compound.
For
example,
ten
of
the
FY86
NHATS
QC
samples
were
spiked
with
36
native
compounds
prior
to
analysis.
The
known
portion
of
the
concentration
must
be
considered
when
estimating
the
entire
actual
compound
concentration
in
the
sample.
Suppose
that
the
i*
QC
sample
was
spiked
with
a
native
compound
at
a
known
concentration
Si.
Let
the
unknown
concentration
of
the
native
compound
in
this
sample
be
pi
before
f
5­
14
1
Table
5­
3.
Estimates
of
R
and
A
for
Surrogate
Compounds
Trichloro
benzene
Tetrachloro
benzene
Hexachloro
benzene
Other
Chloro
benzenes(@

Chrysene
and
other
PAH
compounds
Monochloro
biphenyl
Tetrachloro
biphenyl
Penta­,
._­.
.:..,\
Hexa­,
and
Heptachlor0biphenvlm
Octachloro
biDhenvl
,
Pesticide
group(
4)

Chlorobenzene
group
1
0.5089
0.8697
0.2915
0.8697
0.6203
0.9100
0.4374
0.9301
0.4400
0.9301
0.7666
0.9290
0.5788
0.9716
0.5658
0.9716
0.9940
0.9413
0.5089
0.9514
0.4325
0.9514
0.7586
0.9315
..

PA8
group
7
0.5858
0.9805
0.6500
0.9805
.
1.0088
0.9743
PCBa
group
0.6223
­

'
0.6798
0.9552
0.6452
0.9552
0.5089
0.9696
0.6455
0.9696
1.2018
0.9154
0.4968
0.9662
0.6456
0.9662
I
0­
8975
5­
15
Table
5­
3.
(
cant.)

Diethyl
0.5764
0.9032
0.6313
0.9032
1.0369
phthlate
II
Di­
n­
butyl
0.5764
0.7850
0.4472
0.7850
phthalate
I1.0369
0.9340
I
Butyl
benzyl
0.5764
0.6145
0.4059
phthalate
N
Other
0.5764
0.8235
0.4948
0.8235
1.0369
0.9340
phthalates
w
F
Other
0.5764
0.9558
0.6637
0.9558
1.0369
0.9340
compounds
L
Notes
for
Table
5­
3
Grouping
of
compounds
without
direct
surrogate
counterparts
for
FY86
is
documented
in
Table
5­
2.

 )
Estimates
of
A
for
FY82
are
taken
from
FY84
estimates.

 )
Composite
samples
having
ten
or
more
grams
wet
weight
were
used
in
determining
estimates
for
R
and
A.

(
4)
Surrogates
for
pesticides
were
not
analyzed
in
­
82
or
FY84.
Estimates
for
these
two
years
are
based
on
the
linear
regression
in
(
5­
1)
where
Yi
and
pi
are
substituted
by
the
average
of
the
spiked
and
found
concentrations
across
all
surrogates.

( 
The
estimates
of
R
and
A
for
FY86
are
obtained
by
the
linear
regression
in
(
511
where
Yi
and
pi
are
substituted
by
the
average
of
the
found
and
spiked
concentrations,
respectively,
of
surrogate
heptachlor
and
lindane.

5­
16
Table
5­
3.
(
Coat.)

The
estimates
of
R
and
A
are
obtained
by
the
linear
regression
in
(
5­
1)
where
yi
and
/+
are
substituted
by
the
average
of
the
found
and
spiked
concentrations,
respectively,
of
surrogate
tri­,
tetra­,
and
hexa­
chlorobenzene.

0
The
estiwtes
of
R
and
A
are
obtained
by
the
linear
regression
in
(
5­
1)
where
yi
and
pi
are
substituted
by
the
average
of
the
found
and
spiked
concentrations,
respectively,
of
surrogate
tetra­
and
octa­
chlorobiphenyl.

The
estimates
of
R
and
A
are
obtained
by
the
linear
regression
in
(
5­
1)
where
Yiand
pi
are
substituted
by
the
average
of
the
found
and
spiked
concentrations,
respectively,
of
surrogate
mono­,
tetra­,
octa­,
and
decachlorobiphenyl

0
Surrogates
corresponding
to
phthalates
were
not
analyzed
in
FY82.
Surrogate
phthalate
data
in
FY86
were
not
analyzed
due
to
the
prevalence
of
missing
values.
Estimates
of
R
and
A
for
phthalates
in
FY82
and
FY86
are
based
on
the
linear
regression
in
(
5­
1)
where
Yi
and
pi
are
substituted
by
the
average
of
the
spiked
and
found
concentrations
across
all
surrogates.

(
lo)
Estimates
of
R
and
A
for
all
compounds
not
represented
on
other
rows
of
this
table
are
based
on
the
linear
regression
in
(
5­
1)
where
Yiand
pi
are
substituted
by
the
average
of
the
spiked
and
found
concentrations
across
all
surrogates.

5­
17
,
I
I
.
I
spiking
and
pi.
=
pi
+
Si
after
spiking.
Note
that
a
portion
of
the
unknown
concentration
p;
is
known.
Similar
to
equation
(
5­
l),
the
measured
concentration
Y;
of
the
i'
QC
sample
can
be
expressed
as
Y;
=
~
p;
+
ei
=
R(
pi
+
Si)
+
e,
,
(
5­
4)

As
in
equation
(
5­
21,
the
expectation
of
p:
given
Y<
is
given
by
where
A
and
R
are
as
in
equation
(
5­
2).
Thus
the
adjusted
measured
concentration
for
spiked
samples
is
given
by
t~
e
following
estimate
of>
E(&:
I
Y:)
:

(
5­
6)

where
B
is
an
estimate
of
the
background
concentration
(
discussed
A
A
in
the
following
paragraph),
and
A
and
R
are
as
in
formula
(
S­?).
A
The
last
two
columns
of
Table
5­
3
contain
the
estimates
A
and
R
that
were
substituted
in
equation
(
5­
6)
for
each
compound.
The
background
sample
concentration,
represented
by
B
in
equation
(
5­
6),
was
estimated
by
fitting
a
linear
regression
model.
This
model,
labeled
the
ful2
batch
effects
model
in
Section
5.3.2,
estimates
the
linear
relationship
between
the
spiked
concentration
and
the
measured
concentration
in
a
spiked
sample.
This
relationship
was
allowed
to
change
according
to
the
batch
in
which
the
sample
was
analyzed.
This
model
has
the
following
form:

yij
=
ai
+
Qi
S,,
+
eij
I
5­
18
where
Yij*
is
the
measured
concentration
for
the
jfhQC
sample
(
j=
1,2,3)
in
the
ia
batch
(
i=
l,...,
51,
Sj
is
the
spike
level
of
the
j*
QC
sample,
and
eijrepresents
random
error.
The
parameters
aiand
pi
(
i=
l,...,
5)
represent
batch
intercepts
and
slopes,
respectively.
These
parameters
were
estimated
by
fitting
the
model
to
the
QC
data.
The
average
of
the
estimates
for
the
five
batch
intercepts
cyi
(
i=
l,...,
5)
was
taken
as
the
value
of
 3
in
formula
(
5­
6).
Note
that
the
modification
presented
in
this
subsection
to
adjust
measured
concentrations
was
relevant
only
when
a
native
compound
was
spiked
into
the
given
sample.
No
modification
was
necessary
for
adjusting
measured
concentrations
for
unspiked
native
compounds
in
these
samples.

5.3
STATISTICAL
ANALYSIS
OF
QUALITY'CONTROL
DATA
The
statistical
analysis
of
quality
control
(
QC)
data
was
performed
to
meet
a
number
of
study
objectives
prior
to
composite
data
analysis.
These
objectives
include:

estimating
the
percent
recovery
of
the
analytical
method
for
spiked
compounds,

determining
if
any
significant
differences
exist
in
the
analytical
performance
among
the
five
batches,

characterizing
the
precision
of
the
analytical
method,

M
identifying
estimates
of
measurement
error
present
in
the
data
within
a
batch,

fl
establishing
the
relationship
in
spiked
compounds
between
the
precision
of
the
analytical
method
and
the
level
of
the
spiked
concentration,

identifying
anomalous
results
that
suggest
potential
problems
in
the
analytical
measurements
and
which
may
cause
removal
of
some
of
all
data
for
a
compound
in
further
statistical
analysis.

5­
19
i
Of
the
seventy
samples
analyzec
in
the
FY86
semivolatiles
study,
fifteen
were
QC
samples,
and
five
were
method
blanks.
Each
of
the
five
analysis
batches
contained
one
method
blank,
one
unspiked
control
sample,
and
two
spiked
samples
(
one
sample
spiked
at
a
lower
concentration
than
the
other).
The
QC
samples
were
prepared
from
a
homogenized
bulk
lipid
sample,
allowing
for
comparisons
in
method
quality
to
be
made
between
batches.
Within
a
batch,
the
three
lipid­
based
QC
samples
were
randomized
with
the
ten
composite
samples
in
determining
the
order
of
sample
testing.
The
randomization
ensured
that
no
systematic
trends
due
to
changes
in
laboratory
procedures
were
introduced
into
the
analysis
results.
The
method
blank
was
the
first
sample
analyzed
within
each
batch.
A
total
of
36
compounds
were
spiked
into
the
two
spiked
QC
samples
for
each
batch.
The
spiking
levels
and
compounds
were
determined
by
MRI
in
consultation
with
the
EPA/
OPPT
WAM.
Sixteen
of
these
compounds
were
identified
in
Section
5.1
as
target
compounds
for
statistical
analysis.
They
are
listed
in
Table
5­
4
with
their
spike
levels.
These
levels
were
multiplied
by
200
(
solutionswere
spiked
in
a
200
pL
aliquot),
then
divided
by
the
percent
lipid
weight
(
in
grams)
of
the
sample
to
obtain
spike
concentrations
(
ng/
g)
for'thesample.
QC
analysis
was
performed
on
these
spiked
target
compounds.
Eight
additional
compounds
were
identified
in
Section
5.1
as
target
compounds
for
statistical
analysis,
but
they
were
not
spiked
into
the
QC
samples.
These
compounds
were
identified
as
unspiked
target
compounds.
The
eight
unspiked
target
compounds
were:

R
Beta­
BHC
H
Bis
(
2­
Ethylhexyl)
phthalate
R
Oxychlordane
8
1­
Nonene
Naphthalene
H
1,2,4­
Trimethylbenzene
m
Di­
N­
Butylphthalate
H
Hexyl
acetate
5­
20
I
Table
5­
4.
Spiked
Target
Compounds
for
the
FY86
N#
ATS,
'

With
Spiking
Levels
Pesticides
1
P,
P­
DDT
5.28
21.1
3
P,
P­
DDE
7.38
29.5
12.
Heptachlor
Epoxide
3.58
14.3
14
Trans­
nonachlor
5.48
21.9
'

60
Dieldrin@)
5.48
21.9
Chlorobenzenes
18
1,4­
Dichlorobenzene
31.0
124.

27
Hexachlorobenzene
4.88
19.5
PCBs
53
Tetrachlorobiphenyl
14.1
56.2
54
Pentachlorobiphenyl
16.3
65.0
55
Hexachlorobiphenyl
13.2
52.6
56
Heptachlorobiphenyl
32.5
130.

57
Octachlorobiphenyl
34.3
137.

Phthalate
Esters
66
I
Butyl
Benzyl
Phthalate
I
5.65
II
22.6
Other
33
O­
cymene
7.00
I
28.0
34
D­
limonene
5.98
23.9
40
Octamethyl
Cyclotetrasiloxane
5.28
21­
1
(
1)
All
listed
compounds
except
octachlorobiphenyl
were
detected
in
at
least
50%
of
the
NHATS
FY86
composite
samples.
Octachlorobiphenyl
was
detected
in
44%
of
the
samples.
(
2)
Detected
in
>
50%
of
the
"
ATS
FY86
composite
samples
when
S/
N
calculation
is
used
(
see
Section
5.1.2).

Spike
level
(
ng/
g)
=
Spike
level
(
ns/
pL)
*
200
uL
Percent
lipid
weight
(
9)

Source:
Table
9
of
MRI
Batch
Reports
(
updated
8/
10/
90)

5­
21
QC
data
analysis
for
these
target
compounds
was
limited
to
identifying
effects
due
to
batch
and
to
QC
sample
type.
Thus
QC
analysis
was
performed
on
a
total
of
24
of
the
FY86
semivolatile
compounds.
If
a
compound
was
not
detected
in
a
QC
sample,
the
r
measured
concentration
was
computed
as
oneLhalf
of
the
detection
limit.
This
same
approach
was
used
in
the
statistical
analysis
of
the
composite
samples.

A
listing
of
the
QC
data,
both
unadjusted
and
adjusted
for
surrogate
recoveries,
is
found
in
Appendix
B.
All
QC
analysis
was
performed
on
data
adjusted
for
surrogate
recoveries.

5.3.1
Descrlntive
Summarv
of
QC
Data
5.3.1.1.
Spiked
Compounds.
.
Table
5­
5
contains
a
summary
of
the
QC
data
for
the
16
spiked
target
compounds.
The
data
are
corrected
for
surrogate
recoveries
as
discussed
in
Section
5.2.

Presented
for
each
target
compound
and
each
of
the
four
QC
sample
types
are
the
following
statistics:

m
the
number
of
samples
with
reported
results,

m
the
number
of
detected
results,

m
the
average
and
standard
deviation
of
the
observed
concentrations
(
ng/
g),

m
the
coefficient
of
variation
(%),
equal
to
the
standard
deviation
divided
by
the
average.

For
the
spiked
samples,
the
following
recovery
information
is
also
presented:

the
average
spike
level
(
ng/
g),

m
the
background
average
recovery
(%
I,
calculated
as
the
average
(
across
batches)
of
the
following
ratio:

5­
22
Q)

9
k
me
Om
rim0
ma
,
ow
mEI
W
If
..
II
..
It
..
mm
*
m
corn
(
d
wm
PN
rlco&
mI
ril
d
(
d
I
In
U
01
n
3
&

mmmm
mmmm
r(
rl
0
0
k
k
..
c
uc
ijc
uJZc3m
G3tn
G3bl
G3tnm
o
0­
d
rn
o
0­
4
a
o
0­
4
a
o
0­
4ZIVdX
ZEUCJX
ZUdX
CUCJX
.
In
I
Jn
E
n
CI
I
pc.
a
5­
23
c91.3112
eWul00?;
OWdcO
U­....
mul0000
mrl
PLzlnO
rluldcri
....
rlITOW
.
u)
mul
rlN
a
riult­
0
ooew
....
meum
ewe
dN00
04Wm
0CUoOm
OCVrltV
0ou)
m
....
....
000ulN
0 
m
0
amm
OrWW
rl
rlN
OOrdd
OdPI00
Oul*
Cz....
WOulOr
....
*
ma30
00*
rlcv
00000
ma0
Nul
rime
A
000P
ONri
1
...
OQIW
*
Or
ri
0
In
In
m 

O**
ul
mullnm
ululmln
rl
rl
0
0
k
k
a,

8
N
E:
Q)

h
­
8
a
k
a,
0
L)
d
dV
­
d
a,
­
8
­
4
n
1
*
.
L,
a,
rl
E­
c
5­
24
ouimo
 
0dV)
v)
....
omdco
dcwm
OQICVCQ
o*
ocu
....
o*
ocu
*
dW
ddm
OOIOII
dcoor
....
dWWU
CI
wm
Om
II
..
cnm
t­
lw
d
Od03
ON
om03
1
.
b
,
ow
ul
om0
I
...
cum
OOd
mrl
C­
03
W
cvmdcm
vlmmm
d
d
$.(
x
3
d
s4
(
4
mmmm
l­

i0
k
\

Q)

U
E
I
0
i
5­
25
h
0
cv
u).
m
u)

rnlnlnln
d
0
G
uc
G3tn
a
o
0­
4
EUJX
.
­..
..........
­,
.
mmmm
4
0
k
5­
26
Recovery
(
4i)
=
conc.
(
spiked
sample)
­
conc.
(
control
sample)
*
Spike
level
(
5­
8)

Table
5­
5
shows
that
the
higher
spike
level
for
p,
p­
DDE
was
approximately
ten
percent
of
the
average
background
level
given
by
the
control
sample.
The
laboratory
analysis
was
unable
to
estimate
recoveries
for
p,
p­
DDEdue
to
the
high
background
level
relative
to
the
spiking
levels.
As
a
result,
estimated
background­
adjusted
recoveries
(
BARS)
for
p,
p­
DDE
were
negative.
BARS
near
zero
were
observed
at
low
spike
levels
for
trans
nonachlor,
hexachlorobenzene,
and
D­
limonene,
all
as
a
result
of
high
background
levels.
BARs
of
less
than
50%
were
observed
for
o­
cymene,
Dlimonene
and
octamethyl­
cyclotetrasiloxane,
despite
spike
levels
generally
above
observed
background.
Thus
these
three
compounds
may
have
recovery
problems.
The
BAR
for
1,4­
Dichlorobenzenewas
less
than
60%,
reflecting
the
higher
volatility
in
this
compound
compared
to
the
other
target
compounds.
Except
for
hexachlorobiphenyl
(
which
had
low
recoveries),
the
BARs
for
PCBs
ranged
from
77
to
112
percent.
For
p,
p­
DDT,
heptachlor
epoxide,
hexachlorobenzene,
and
butyl
benzyl
phthalate,
the
BARs'ranged
from
64
to
122
percent.
The
"
BAR"
approach
to
calculating
percent
recoveries
given
in
equation
(
5­
8)
has
been
recommended
for
use
through
the
"
ATS
program.
However,
an
alternative
approach
to
calculating
percent
recoveries
does
not
place
as
much
emphasis
on
the
ability
to
detect
finite
differences
in
concentration.
This
approach
considers
the
formula
conc.
(
spiked
sample)
*
100%
Recovery
(%
1
=
conc.
(
control
sample)
+
spike
level
(
5­
9)

5­
27
Note
that
the
percentages
calculated
from
(
5­
9)
are
always
positive
and
are
equal
to
100%
when
the
observed
concentration
equals
the
sum
of
the
spike
level
and
the
control
sample
concentration
within
the
batch.
Table
5­
6
presents
the
percent
recoveries
under
both
approaches
(
5­
8)
and
(
5­
9)
for
the
spiked
target
compounds.
In
this
setting,
approach
(
5­
9)
generally
leads
to
improved
percent
recovery
values
over
approach
(
5­
8).
This
is
especially
apparent
with
p,
p­
DDE,
where
the
spike
levels
were
much
smaller
than
the
observed
levels
in
the
control
samples.
While
approach
(
5­
8)
has
been
recommended
for
the
"
ATS
program,
both
approaches
evaluate
method
performance
differently,
and
thus
both
sets
of
results
should
enter
into
performance
evaluation.
Coefficients
of
variation
were
widely
varied
among
the
samples
and
compounds
(
Table
5­
5).
Only
p,
p­
CDT,
heptachlor
epoxide,
hexachlorobenzene,
hexachlorobiphenyl,
and
heptachlorobiphenyl
had
coefficients
of
variation
which
were
at
25%
or
smaller
for
all
samples.
For
the
other
spiked
target
compounds,
the
variation
in
the
QC
results
at
a
given
spike
level
was
as
high
as
80%
of
the
observed
average
level
across
the
batches.
For
dieldrin,
butyl
benzyl
phthalate,
o­
cymene,
and
octamethyl­
cyclotetrasiloxane,
at
least
one
QC
sample
result
was
not
detected
at
the
low
spike
level.
Appendix
C
contains
plots
of
the
measured
concentrations
versus
the
spike
levels
for
all
study
compounds.
Although
some
plots
indicate
a
linear
increasing
relationship,
most
plots
show
highly
variable
results
among
the
batches
at
a
given
spike
level.
Several
of
the
plots
suggest
that
concentrations
were
higher
for
Batches
4
and
5
than
for
the
other
three
batches,
such
as
with
p,
p­
DDT,
p,
p­
DDE,
and
some
of
the
PCBs.
This
was
especially
evident
at
high
spike
levels.
Appendix
D
contains
summaries
like
those
in
Table
5­
5
for
spiked
compounds
not
on
the
target
list.

5­
28
Table
5­
6.
Percent
Recoveries
for
Spiked
Target
Compounds,
as
Determined
from
Two
Calculation
Methods
Hexachlorobiphenyl­
1.43
45.40
I
89.04
81.71
Heptachlorobiphenyl
90.56
81.49
97.43
88.14
Octachlorobiphenyl
111.94
101.36
109.92
101.17
Butyl
benzyl
phthalate
119.06
63.55
110.52
66.72
0­
cymene
2.63
17.26
15.81
20.35
D­
limonene
­
21.45
41.94
64.29
61.94
Octamethyl­
cyclotetrasiloxane
4.30
27.73
17.02
30.11
Two
methods
to
calculating
percent
recovery
on
the
surrogate­
adjusted
data:

Recovery
(%)
.­­
conc.
(
spiked
sample)
­
conc.
(
controlsample)
*
Spike
level
(
5­
8)

conc.
(
spikedsample)
­
Recovery(%)
=
conc.
(
controlsample)
+
spike
level
*
100%
(
5­
9)

5­
29
5.3.1.2
Unspiked
Compounds.
Table
5­
7
contains
descriptive
summaries
'
of
eight
target
compound
concentrations
that
were
not
spiked
in
the
control
samples.
The
descriptive
statistics
were
calculated
for
each
batch
and
across
all
batches.

5.3.1.3.
Method
Blanks.
Method
blanks
were
used
to
assess
laboratory
background
contribution
to
concentration
levels
within
the
composite
samples.
Eight
of
the
target
compounds
were
detected
in
the
method
blanks.
When
detectable
concentrations
were
measured
in
method
blanks,
the
results
are
presented
in
Table
5­
8.
Detection
in
the
method
blanks
suggests
a
potential
bias
in
the
reported
concentration
levels
within
the
affected
batches
for
the
given
compound.
The
method
blanks
for
Batches
1
and
5
had
detectable
levels
for
the
three
target
phthalates.
The
bis
(
2­
ethylhexyl)
phthalate
was
also
detected
in
the
method
blank
for
Batch
'
3.
The
method
blank
for
Batch
4
was
not
analyzed
for
phthalates.
In
most
cases,
the
method
blank
concentration
was
at
or
above
the
control
(
unspiked)
sample,
suggesting
laboratory
background
contribution
to
the
measured
concentration.

5.3.2
Statistical
APDroach
to
Analvzincr
the
QC
Data
To
address
the
statistical
objectives
presented
at
the
beginning
of
Section
5.3,
the
QC
data
were
statistically
analyzed
using
linear
models
fitted
to
the
surrogate­
adjusted
concentrations
for
each
compound.
A
linear
regression
model
was
applied
to
concentration
data
for
spiked
compounds.
This
model
included
effects
for
batch
and
spike
level.
A
similar
analysis
of
variance
application
determined
whether
batch
and
sample
type
effects
were
statistically
significant
on
concentrations
for
unspiked
compounds.
The
statistical
methods
and
results
are
described
in
this
subsection.

5.3.2.1
Spiked
Compounds.
Two
types
of
linear
regression
models
were
fit
to
the
QC
data
for
spiked
target
compounds.
One
model,

5­
30
h
h
h
h
n
h
m
*
a
*
m
rla
lnm
ma
mu,
dl
u,
m
Wcn
me
Or0
u,
d
rnN
dA
md
am
eo
ECV
0­
ma
do
..
..
..
..
..
..
AJJ
ma3
mrn
md
me
cnd
Om
id
dm
Od
m
Od
Nrn
a
N
d
mrl
d
W
v
Y
Y
v
W
I
h
A
A
n
n
h
I
a)*
*
m
am
rid
90
00
1
la
dOr
cnfl
u,
ln
ma
ma
00
..
..
..
..
..
..
1
dm
mP
00
WDrl
mu,
mm
4
Ncn
*
d
am
am
m
I
m
d
u,*
I
Y
W
Y
W
W
U
4
t
I
h
h
n
n
h
A
I
mrl
IDN
dd
CVa
u,*
00
I
md
mm
eu,
mu,
Q\
N
dOr
4*
..
..
..
..
..
..
1
OQ)
u,
rl
d0
Cvm
am
m*
I
d
dcd
d
u,
em
em
t
ri
rl
8
"
W
W
Y
Y
W
Y
I
I
n
n
n
h
n
h
m0
drc
*
nl
me
worn
a*
fi
mm
srm
Nu,
cvm
Wdc
cndc
z
Urn
..
..
..
..
..
..
cd
id
Wcn
cum
u,
m
old
drn
Fln
Q)

cp
NO
m
d
mu,
md
E
dd
d
d
dd
I
W
Y
W
Y
Y
v
3
2
I
c,
1
n
h
h
n
n
h
I
Qm
em
mm
u,
u,
ID*
m*
.
c
I
Qu,
0a
u,
a
m0
mm
coe
c,
IN
I
I
I
u,
N
Nd
W
m*

W
W
*
d
Y
m*
Nd
v
rn0
dd
W
2
a,
.4
I
I
m
I
n
h
CI
h
n
h
a,
..
..
..
..
..
..
rd
I
dd
*
O
cod
dce
db
me
I
*
d
rnN
ON
OCO
Od
i­
ld
1
..
..
dccv
Fd
ON
00
Id
00
*
O
..
..
..
..
d
I
eu,
NO
dm
mo
*
O
cnm
U
1
*
u,
d
d
cvd
drn
u,
d
C
a
I
d
W
v
W
Y
*
m
v
W
Eid
aal
U
E:
id
h
­
4
ri
cd
a,
s:
5
k
id2
E
Q)
U
m
PI
.
ar
a,
Qc
rlha,
Id
k
I­
I
4
E:
r­
i
su
L,
0
m
rd
a,
$*
urd
a,
k
E:
u
2
rl
u
or(
0
k
3
X
0
rd
3
ard
Id
Q)

0
a
d
c
.4
coc
a
A
c
3
A
"
Us
d
a
E
U
ma
k
u
a,
i
2
­
d
­
4
rda
0
z
a
a
$
2
3
U
03
5­
31
­."

Table
5­
8.
Batch
Analysis
Results
on
Method
Blanks
and
Control
Samples
for
Compounds
Detected
in
At
Least
50%
of
Compositee,
where
the
Compound
Was
Detected
in
the
­
Method
Blank
Phthalate
Estere
Di­
n­
butyl
phthalate
1
17901
44.9
5
17957
13.7
12.05
20.2
373.

67.8
Butyl
benzyl
phthalate
1
17901
29.1
5
17957
14.0
10.55
13.7
276.

102.

Bis
(
2­
ethylhexyl)
1
17901
205.
57.8
355.
phthalate
2
17915
581.
560.
104.

3
17929
288.
222.
130.

5
17957
15.4
348.
4.4
Other
D­
limonene
2
17915
27.9
85.4
32.7
5
17957
19.5
164.
11.9
Octamethyl­
3
1.7929
156.
20.3
768.
cyclotetrasiloxane
Other
(
sualitative)

1­
nonene
1
17901
600.
200.
300.

2
17915
1000.
600.
.
167.

~~

1,2,4­
3
17929
40.0
I
30.0
I
133.
trimethylbenzene
Hexyl
acetate
1
17901
40.0
.

3
17929
16000
~

Note:
Concentrations
are
unadiusted
for
surrogate
recoveries.

5­
32
known
as
the
batch
slopes
model,
provided
estimates
of
batch
recoveries
and
tests
for
equality
of
these
estimates
across
batches.
The
other
model,
called
the
batch
intercepts
model,
was
considered
when
spiked
sample
results
were
not
sufficiently
above
background
to
allow
for
batch
recovery
estimates
to
be
made.
The
batch
intercepts
model
provided
for
separate
background
levels
to
be
estimated
for
each
batch.
These
models
are
summarized
in
Table
5­
9
and
satisfactorily
characterize
the
FY86
QC
data
for
all
compounds.
The
full
batch
effects
model
introduced
in
Section
5.2.1
and
presented
in
(
5­
7)
was
also
considered
in
this
application.
The
full
batch
effects
model,
a
composite
of
the
batch
slopes
and
batch
intercepts
models,
contains
ten
parameters
which
represent
separate
slopes
and
intercepts
for
the
five
batches.
This
is
a
large
number
of
parameters
compared
with
the
number
of
data
points
(
151,
leading
to
overparametrization
problems.
When
either
constant
batch
backgrounds
or
constant
batch
slopes
cannot
be
assumed,
a
simple
linear
regression
model,
with
constant
background
and
slope
across
batches,
was
considered.
The
batch
slopes
model
tested
for
significant
differences
in
batch
recoveries
for
the
spiked
compound.
This
model
also
estimated
the
batch
recoveries
and
the
average
recovery
across
all
batches,
and
calculates
predicted
concentrations
at
each
spike
level.
The
average
recovery
was
tested
for
significant
difference
from
loo%,
thus
determining
the
accuracy
of
the
analytical
method.
The
estimated
intercept
term
was
interpreted
as
the
estimate
of
background
(
or
systematic
error)
across
all
batches.
Batch
effects
were
present
when
at
least
one
of
the
estimated
slopes
was
found
to
be
significantly
different
from
the
others.
According
to
the
descriptive
results
presented
earlier
in
this
section,
the
spike
levels
for
some
compounds
were
low
relative
to
background.
Thus
the
reported
concentrations
for
spiked
samples
were
at
the
background
level.
This
outcome
was
5­
33
a 

B0
a5
u
4
0,
m
3
0
d
U­
l
0
mk
4
d
a,
8
Q,
k
Id
Q,
k
Q,

9
k
a,

59
c3
@
2
+
6­

JI
w=
w
rl
a
9)
k
M8
a,

a0
x
rn
u
b,
a,
u
kal
U
c
H
r:
u
U
Id
p?

5­
34
observed
for
p,
p­
DDE.
A
batch
slopes
model
was
not
appropriate
in
this
situation,
as
batch
recoveries
cannot
be
estimated
from
the
observed
data.
Affected
compounds
were
analyzed
using
the
batch
intercepts
model
or
simple
linear
regression
model
to
note
overall
differences
among
batches.
The
statistical
analysis
of/
QCdata
established
that
significant
batch
effects
existed
in
the
data
for
virtually
all
spiked
target
compounds.
Specifically,
estimated
recoveries
for
Batches
4
and
5
tended
to
differ
from
the
first
three
batches.

As
a
result,
all
statistical
analyses
on
composite
samples
included
a
"
batch
classggeffect
(
Batches
1­
3
versus
4­
5).
Any
batch
effects
existing
beyond
the
"
batch
class"
effect
were
treated
as
random
effects.
The
"
ATS
additive
model
assumes
that
the
standard
deviation
of
the
measured
concentration
in
composite
samples
has
two
components:

a
component
associated
with
the
within­
batch
measurement
error,
estimated
by
the
mean­
squared
error
(
MSE)
from
the
batch
slopes
model,

a
random
component
associated
with
the
random­
batch
effects
within
each
batch
ggclasslg.

For
a
spiked
target
compound,
the
predicted
average
concentration
at
the
jthspiked
concentration
SCj
(
j
=
1,
2)
is
given
by
A
*
cj
=
&
+
&*
SCj
,
(
5­
10)

where
&
is
the
baseline
average
concentration
and
is
the
average
estimated
recovery
across
batches.
Note
that
is
the
least­
squares
estimate
of
the
parameter
QI
and
hVgis
the
average
of
the
least
squares
estimates
of
pi,
both
resulting
from
fitting
the
batch
slopes
model
in
Table
5­
9.
The
standard
deviation
of
is
computed
as
5­
35
(
5­
11)

where
MSE
is
the
mean­
squared
error
from
the
batch
slopes
model,
and
SD(
P)
is
the
sample
standard
deviation
of
the
estimated
batch
xecoveries.
Thus
the
standard
deviation
increases
with
the
concentration
of
the
sample;
however,
it
is
not
necessarily
proportional
to
the
concentration.
If
the
batch
slopes
model
indicated
that
a
significant
batch
effect
existed,
only
recoveries
from
Batches
1­
3
were
used
to
estimate
the
parametersaVgand
SD(
P).
Otherwise
all
five
batches
were
used.

5.3.2.2
Unspfked
Compouhds.
Although
batch
recoveries
could
not
be
estimated
for
the
eight
unspiked
target
compounds,
batch
effects
and
method
contamination
could
still
be
characterized
for
these
compounds.
A
two­
way
analysis
of
variance
approach
was
applied
to
these
compounds
containing
effects
representing
the
batch
and
the
sample
type
(
control,
low
spike,
high
spike).
The
batch
effect
provided
a
test
for
significant
differences
in
concentrations
between
batches.
The
effect
for
sample
type
allowed
for
tests
between
samples
containing
different
spiking
solutions.
This
latter
test
was
a
means
of
determining
the
presence
of
method
contamination.

5.3.3
Results
of
Statistical
Modellina
of
QC
Data
5.3.3.'
1
Spiked
Compounds.
The
results
of
fitting
the
batch
slopes
model
in
Table
5­
9
to
the
QC
data
for
spiked
target
compounds
are
summarized
in
Tables
5­
10
through
5­
12.
Table
5­
10
contains
the
estimated
batch
recoveries
for
each
spiked
target
compound,
as
well
as
the
estimated
average
recovery
across
all
batches.
Table
5­
11
reports
significance
levels
for
tests
of
equal
recoveries
among
sets
of
batches.
Table
5­
12
provides
information
on
observed
precision.

5­
36
4
d
F
m
I
I
$
I
I
I
I
u)
I
1
m
m
I
I
n
Y
rl
I
I
c
W
Q)
0
I
1
­
84
*
P
I
I
k
rl
I
I
E
2u
OD
I
I
m
P
I
I
%
cn
I
I
U
fd
m
a
a
I
Ia,
U
cv
0
I
i
I
m
I
i!
­
ri
m
w
co
I
I
rl
0
I
I
rl
8
I
r,

3
w
*
m
<
3\
rl
cv
VI
Ln
N
(
v
aop
oa
0
u)
4
*
N
u)
rl
1
ri
h
h
a,
F:
a
03
G
2
N
N
pc
II
F:
­
Pi
N
a,
\
R
%
E
0
k
v
k
0
0
ri
E­
c
w
%
c
E
;
rd
Id
1
k
cl
ua
8
a,
I6
X
E­
c
0
W
,

W
VI
m
W
e
m
P
VI
m
P
cv
h
.
0
cv
m
m
.
0
W
t
m
rl
W
al
rl
t3
Q\
w
rl
cn
rl
it
cn
*
w
a,

G
F:

­
Pi
r(
1
Q
a­
l
I
Table
5­
11.
Tests
for
Significant
Differences
in
Batch
Slopes
Among
Selected
Batches
for
Spiked
Target
Compounds
p,
p­
DDT
Heptachlor
epoxide
Trans­
nonachlor
Dieldrin
1,4­
Dichlorobenzene
He 
xachlorobenzene
Butyl
benzyl
phthalate
0­
cymene
D­
limonene
Octamethyl
cyclotetrasiloxane
Pesticide8
0.0003*
0.023*
0.101
0.0018*

Chlorobenzenes
0.0067*
0.031*

PCBs
0.389
0I
812
0.210
0.094
0.083
0.205
0.0016*
0.016*

0.023*
0.0032*

0.254
0.410
0.0018*
0.0006*
0.0001*
0.0018*
0.027*

0.071
0.0013*

0
*
0059*

0*
ooos*

0.0026*

0.066
0.077
(
I)
p,
p­
DDE
not
included
in
this
table
(
see
discussion)

Significance
occurs
at
the
0.05
level.

5­
39
&
lG
ox
3
­
4
U
(
d
d
d
(
d
U
E
­
4
B
3
m
I
d
m
4
u
(
d
A
k
w
0
xcy
n
4
rl
t
d
I
h
0
hm
3
+,
rn
0)
c,
0
­
4u
&
al
rl.
Qk
k
R
$
0
u
U
u
I
 
0)
&
Y
n
$
2
Q
v
F:
0­
l
e
­
4
E:@
(
d
a
k
oal
k
c11
a
 E
I
rl
­
4
(
d
fi
pc
Q)
rlJJ
­
d
tu
Pt
no
t
5­
40
For
all
but
p,
p­
DDE,
the
batch
slopes
model
provided
a
good
fit
to
the
surrogate­
adjusted
data.
The
estimate
of
average
recovery
for
p,
p­
DDE
was
outside
of
valid
ranges,
emphasizing
the
inappropriateness
of
estimating
batch
recoveries
for
this
compound.
Batch
recoveries
were
not
interpretable
for
p,
p­
DDE
due
to
large
differences
in
batch
intercepts.
Thus
no
estimated
batch
recoveries
were
reported
for
p,
p­
DDE
in
Table
5­
10.
For
the
other
compounds,
a
t­
test
was
performed
at
the
0.05
significance
level
to
determine
if
the
average
recovery
was
significantly
different
from
100%.
All
compounds
except
p,
p­
DDE
and
octachlorobiphenyl
had
average
recoveries
significantly
greater
than
100%.
For
twelve
of
the
compounds,
the
average
recovery
was
significantly
less
than
100%.
Five
compounds
had
average
recoveries
less
than
50%:
o­
cymene
(
18.4%),
octamethylcyclotetrasiloxane
(
29.5%),
dieldrin
(
46.2%),
D­
limonene
(
46.9%),
and
hexachlorobiphenyl
(
48.7%).
Two
compounds
had
average
recoveries
significantly
greater
than
100%:
hexachlorobenzene
(
110%)
and
p,
p­
DDT
(
124%).
Estimates
of
the
individual
batch
recoveries
from
the
batch
slopes
model
are
shown
in
the
remaining
columns
of
Table
5­
10.
Also
present
are
the
results
of
an
F­
test
to
determine
if
significant
differences
exist
among
the
batch
recoveries
at
the
0.05
significance
level.
This
test
determines
the
presence
of
batch
effects.
Significant
differences
among
the
five
batch
recoveries
were
observed
for
twelve
compounds.
For
virtually
all
of
these
compounds;
the
differences
seem
to
arise
from
the
large
recoveries
in
Batches
4
and
5
relative
to
the
first
three
batches.
For
p,
p­
DDT,
the
estimated
recoveries
in
Batches
4
and
5
average
a
65%
increase
over
the
first
three
batches.
Similar
results
are
observed
for
PCBs
and
other
pesticides.
F­
tests
on
linear
combinations
of
the
estimated
batch
recoveries
were
performed
to
determine
significant
differences
Y
among
these
recoveries.
The
significance
levels
for
the
test
of
equal
recoveries
among
the
five
batches
are
listed
in
Table
5­
10
for
each
spiked
target
compound
except
p,
p­
DDE,
where
batch
recoveries
could
not
be
accurately
estimated.
Because
of
the
apparent
difference
in
estimated
batch
recoveries
between
Batches
1­
3
and
Batches
4­
5,
Table
5­
10
also
contains
significance
levels
for
testing
differences
between
these
two
groups
of
batches,
as
well
as
among
the
first
three
batches
only.
For
eleven
of
the
fifteen
spiked
target
compounds
in
Table
5­
10,
the
estimated
recoveries
in
Batches
1­
3
differ
significantly
(
at
the
0.05
level)
from
the
estimated
recoveries
in
Batches
4­
5.
However,
only
three
of
these
compounds
have
significant
differences
in
estimated
recoveries
among
Batches
1­
3
only.
Thus
the
following
conclusions
can
be
made
from
Table
5­
10:

I
The
systematic
difference
in
recoveries
between
Batches
1­
3
and
Batches
4­
5
appears
real,

I
There
appear
to
be
no
additional
systematic
batch
effects
beyond
that
observed
in
Batches
1­
3
versus
Batches
4­
5.

The
first
conclusion
states
that
it
is
not
suitable
to
treat
all
batch
effects
as
random
as
was
done
in
the
FY87
analysis
of
dioxins
and
furans.
The
presence
of
a
systematic
batch
effect
indicates
that
some
batch
correction
is
necessary
when
analyzing
the
composite
data.
However,
any
additional
batch
effects
beyond
the
Batches
1­
3
versus
Batches
4­
5
effect
can
be
treated
as
random.
For
spiked
target
compounds,
Table
5­
12presents
the
predicted
average
concentration
and
estimated
coefficient
of
variation
(
CV)
for
each
compound
and
spike
level,
as
derived
by
the
batch
slopes
model.
These
results
were
used
to
characterize
the
precision
of
the
analytical
method.
Except
for
o­
cymene
and
octamethyl­
cyclotetrasiloxane(
which
had
very
low
recoveries),
all
predicted
concentrations
at
the
zero
spike
level
were
significantly
greater
than
zero
at
the
0.05
significance
level.

5­
42
This
is
consistent
with
the
fact
that
thi­
target
compounds
were
detected
in
nearly
all
of
the
QC
control
samples
(
Table
5­
5).

Whenever
the
batch
slopes
model
indicated
a
significant
batch
effect
present,
average
recoveries
from
only
the
first
three
batches
were
used
to
calculate
predicted
concentrations
and
CVs
for
the
compound.
This
reflects
the
assumption
that
the
primary
trend
in
batch
effects
is
due
to
Batches
4
and
5
having
higher
recoveries
compared
to
the
first
three
batches,
leading
to
biases
in
the
results
from
Batches
4
and
5.
From
Table
5­
12,
the
relative
precision
of
measured
concentrations
tends
to
be
better
for
pesticides
and
PCBs
compared
with
other
groups
of
compounds.
At
the
control
level,
the
CVs
for
pesticides
and
PCBs
range
from
7.6%
to
51.5%,
with
a
CV
of
71.9%
for
the
more
volatile
1,4­
Dichlorobenzene.
The
CVs
for
all
of
the
pesticides
and
PCBs
are
below
79%
in
the
spiked
samples.
Meanwhile,
except
for
D­
limonene
(
whose
CVs
rival
the
pesticides
and
PCBs),
the
CVs
for
phthalates
and
other
compounds
are
above
50%
for
control
and
spiked
samples.
Because
batch
recoveries
could
not
be
estimated
for
p,
p­
DDE
(
m/
z=
288
and
m/
z=
316)
based
on
the
observed
results
and
spike
levels,
the
batch
intercepts
model
was
fit
to
this
compound.
The
batch
intercepts
model
provides
for
background
levels
to
be
estimated
for
each
batch.
Thus
batch
effects
were
determined
by
testing
for
equality
of
the
batch
background
levels.
Table
5­
13
contains
the
results
of
fitting
the
batch
intercepts
model
to
p,
p­
DDE.
For
both
sets
of
p,
p­
DDE
results,
the
test
for
batch
effects
is
highly
significant.
As
apparent
in
the
QC
data
plots,
the
estimated
background
levels
for
Batches
4
and
5
are
over
twice
the
level
of
the
first
three
batches.
This
extreme
difference
in
background
levels
contributes
to
the
inability
to
estimate
batch
recoveries.
Thus
the
results
of
the
batch
intercepts
model
fitting
for
p,
p­
DDE
indicate
that
differences
between
the
two
"
batch
classes"
(
Batches
1­
3
versus
4­
51
are
highly
significant,
as
was
seen
for
most
of
the
spiked
target
compounds.

­
4
CP",
5­
43
6'
t
\
,
j
&
at
mo
40
CJm
em
d
8E
4
L)
ww
id
tnm
R
at2
Q)
drl
5
E:
­
rl
h
ul.
3
.
z
bl
­
3
id
A
uw
2
d)
JJ
i!
k
id
a
rnw
­
W
0
alwfd
a
m
5
m
al
rl
hh
+,
W
.

I
cow
a
0
Em
cod
­
4(
urn
u
al
H
11
Ul
NN
al
\\
 
a
x
E
k
YV
cd
ww
aa
aa
A
Y
5­
44
5.3.3.2
Unepiked
Cosapounds.
Results
of
statistical
analysis
of
unspiked
target
compound
concentrations
in
QC
samples
are
presented
in
Table
5­
14.
This
table
presents
significance
levels
for
differences
between
batches
and
between
sample
types.
Batch
effects
were
significant
at
the
0.05
level
for
oxychlordane
and
Di­
n­
butyl
phthalate.
Significant
batch
effects,
for
oxychlordane
are
attributed
to
the
large
number
of
not
detected
readings
in
Batch
1.
A
very
high
percentage
of
not
detected
readings
for
oxychlordane
in
Batch
1
is
also
present
among
the
composite
samples.
Since
the
frequency
of
not
detected
oxychlordane
 
readings
substantially
decreases
after
Batch
1,
the
Batch
1
oxychlordane
results
tend
to
be
suspect.
None
of
the
unspiked
target
compounds
showed
a
significant
effect
due
to
the
sample
type.
Thus
these
data
can
be
considered
as
control
sample
results
for
the
unspiked
compounds.
All
of
these
samples
are
used
to
determine
within­
batch
measurement
error.
Precision
was
estimated
for
the
unspiked
compounds
at
the
control
level
based
on
the
above
analysis
of
variance
model.
The
precision
summary
is
presented
in
Table
5­
15.
The
predicted
control
level
reflects
all
QC
samples,
as
it
was
determined
that
no
sample
type
effect
existed.
Because
data
exist
for
all
sample 
types
within
each
batch,
the
predicted
concentration
is
equal
to
the
average
concentration
across
the
15
QC
samples.
The
standard
deviation
of
the
predicted
concentration
is
equal
to
the
mean?
squared
error
estimated
by
the
model.
The
precision
summary
in
Table
5­
15
indicates
that
two
compounds
(
Bis
(
2­
ethylhexyl)
phthalate
and
hexyl
acetate)
have
CVs
above
100%.
These
compounds
have
one
extreme
observation
in
at
least
one
batch,
at
levels
up
to
four
times
the
value
of
the
other
results
within
the
batch.
Other
compounds
also
show
high
variability
in
the
data
within
each
batch,
especially
between
not
detected
results
and
detected
results.

5­
45
Table
5­
14.
Reeults
of
Statisthzal
Analysis
of
QC
Data
on
Unspiked
Target
Compounds
Beta­
BHC
I
0.626
I
0.643
Oxychlordane
I
0.009
0.634
PAHS
11
Naphthalene
I
0.545
I
0.698
Phthalate
Esters
Di­
n­
butyl
phthalate
0.050
0.073
.
Bis
(
2­
ethylhexyl)
phthalate
0.496
0.119
Other
(
qualitative)
Pesticides
1­
nonene
0.488
0.613
II
1,2,4­
Trimethylbenzene
I
0.144
1
0.199
Hexvl.
acetate
I
0.771
1
0­
342
n
,

5­
46
s­
li
Table
5­
15.
Predicted
Concentration8
and
Coefficients
of
­
Variationfor
Unspiked
Target
Compounde
at
the
Control
Level
Pesticides
Beta­
BHC
213.4
76.0
Oxychlordane
(
all
batches)
105.9
34.1
Oxychlordane
(
Batch
1
removed)
129.2
31.8
pm8
Naphthalene
1
9.731
I
44.9
Phthalate
Esters
Di­
N­
Butyl
phthalate
I
39.79
47.3
Bis
(
2­
ethylhexyl)
phthalate
309.1
126.0
0ther
(
qualitative1
1­
nonene
452.0
75.6
1,2,4­
trimethylbenzene
36.00
77.5
Hexyl
acetate
120.6
116­
1
Note:
These
statistics
reflect
results
for
all
QC
samples.

Coefficient of
variation
=
Sauare
root
of
mean­
suuared
error
.
 
h
Predicted
concentration
:
I
:<
.
5::.
.
9 .,

il
..
,

5­
47
5.3.4
Concluaions
The
following
summarizes
the
conclusions
and
findings
of
the
QC
data
analysis
(
courses
of
action
formulated­
fromthese
conclusions
are
underlined):

1.
Significant
batch
effects
appear
among
the
16
spiked
target
compound
concentrations.
The
primary
batch
effect
is
due
to
the
high
recovery
and
background
in
Batches
4
and
5
compared
to
the
other
three
batches.
Because
this
difference
between
"
batch
classes"
is
prevalent
in
nearly
all
of
the
spiked
target
compounds,
it
is
necessaw
to
include
an
effect
for
Batches
4­
5
versus
Batches
1­
3
in
the
model
used
to
analvze
the
comDosite
samoles.
Any
other
batch
effects
were
assumed
to
be
random
and
thus
were
not
considered
in
model
adjustments.

2.
The
difference
between
"
batch
classes"
was
not
as
significant
among
the
eight
unspiked
target'compounds.
However,
differences
in
control
level
concentrations
between
batches
were
noted
for
oxychlordane
and
di­
nbutyl
phthalate.
In
particular,
nearly
every
QC
and
composite
sample
indicated
a
not­
detected
result
for
oxychlordane
in
Batch
1.
As
a
result,
all
Batch
1
concentrations
for
oxvchlordane
will
be
deleted
Drior
to
comDosite'data
analvsis.

3.:
Seven
of
the
target
compounds
were
detected
among
the
five
method
blanks..
All
three
target
phthalates
were
included
among
these
seven
compounds.
In
particular,
bis
(
2­
ethylhexyl)
phthalate
was
detected
in
all
four
method
blanks
wh.
ichwere
analyzed
for
phthalates.
Dlimonene
and
octamethyl­
cyclotetrasiloxane,
also
detected
in
the
method
blanks,
were
among
those
compounds
with
relatively
low
recoveries.

4.
High
background
levels
relative
to
the
spiking
levels
were
observed
for
a
few
spiked
target
compounds.
In
particular,
the
spike
levels
for
p,
p­
DDE
were
no
more
than
10%
of
the
observed
background
level.
For
this
reason,
and
because
of
large
differences
in
background
level
among
the
batches,
batch
recoveries
could
not
be
estimated
for
p,
p­
DDE.
Other
compounds
with
high
background
levels
relative
to
spiking
levels
were
p,
p­
DDT,
heptachlor
epoxide,
trans­
nonachlor,
hexachlorobenzene,
and
D­
limonene.

5.
Estimated
average
recoveries
for
spiked
target
compounds
were
significantly
below
100%
for
all
but
p,
p­
DDTand
hexachlorobenzene,
where
they
were
significantly
above
100%.
0­
cymene,
D­
limonene,
%.
I
octamethyl­
cyclotetrasiloxane,
dieldrin,
and
hexachlorobiphenyl
had
average
recoveries
below
50%.
Most
estimated
batch
recoveries
for
all
compounds
were
less
than
100%
for
Batches
1­
3,
while
many
compounds
had
estimated
batch
recoveries
above
100%
for
Batches
4
and
5.

6.
Characterization
of
measurement
precision
for
spiked
target
compounds
indicated
that
better
precision
was
observed
for
pesticides
and
PCBs.
Precision
was
worse
for
phthalates
and
"
other"
compounds,
with
coefficients
of
variation
(
CVs)
exceeding
50%.
For
unspiked
target
compounds,
CVs
ranged
from
32
to
126
percent.

7.
Except
for
o­.
cymene
and
octamethyl­
cyclotetrasiloxane
(
which
had
very
low
recoveries),
all
predicted
concentrations
at
the
zero
spike
level
were
greater
than
zero
at
the
0.05
significance
level.

8.
The
relationship
between
measured
and
spiked
concentrations
for
spiked
target
compounds
was
generally
linear
over
the
range
of
spiked
concentrations,
but
the
variability
within
each
batch
was
high.

The
above
findings
in
the
QC
data
were'used
to
reevaluate
the
status
of
each
target
compound
prior
to
composite
data
aqalysis.
Several
compounds
had
recovery
and
cantamination
problems
as
summarized
above.
As
a
result
of
findinas
from
the
statistical
analysis
on
OC
data,
the
followina
comrsounds
have
been
removed
from
the
list
of
taraet
comrsounds
on
which
statistical
analvsis
of
comrsosite
data
is
rserformed:

Bis
(
2­
ethvlhexvl)
rshthalate
­

*+

3
w
detected
in
all
method
blanks
analyzed
for
this
&%

24
compound.
I:<
.+
v
low
precision
results.
&
+*

.&*
Di­
n­
butvl
Phthalate
detected
in
two
of
the
four
analyzed
method
blanks.

I
high
levels
of
not­
detected
results
among
the
composite
samples
in
Batches
3
and
5
make
these
batch
results
suspect.

5­
49
c
61(
2p38
Butvl
benzvl
phthalate
m
detected
in
two
of
the
four
method
blanks
analyzed
for
this
compound.

*
rn
the
low­
spiked
result
in
Batch
3
was
not
detected,
although
spiked
amounts
were
not
below
estimated
background.

1.2,4­
Trimethylbenzene
R
detected
in
the
method
blank
for
Batch
3.

rn
percent
detected
among
composite
samples
in
Batches
4
and
5
is
very
low
compared
to
the
other
three
batches,
making
these
batch
results
suspect.

0­
cvmene
rn
recoveries
extremely
low
for
all
spiked
QC
samples,
even
though
the
spiked
amounts
were
above
estimated
background.
All
results
for
spiked
sampled
failed
to
meet
DQOs.

rn
percent
detected
among
composite
samples
in
Batch
1
is
low
compared
to
the
other
batches.

D­
limonene
rn
detected
in
two
of
the
five
method
blanks.

­
I
R
recoveries
extremely
low
for
spiked
QC
compounds.

Octamethvl­
cvclotetrasiloxane
m
detected
in
the
method
blank
for
Batch
3.

m
recoveries
extremely
low
for
all
spiked
QC
samples,
even
though
the
spiked
amounts
were
above
estimated
background.
All
results
for
spiked
samples
failed
to
meet
DQOs.

M
percent
detected
among
composite
samples
in
Batch
1
is
low
compared
to
the
other
batches.
The
percentage
of
detected
results
increased
with
the
batch
ID
number.

A
total
of
17
compounds
remained
classified
as
target
compounds
for
statistical
analysis
following
analysis
of
the
QC
data.
However,
only
limited
analyses
were
performed
on
the
qualitative
compounds
hexyl
acetate
and
1­
nonene.

5­
50
6.0
STATISTICAL
METHODOLOGY
This
section
discusses
the
statistical
methodology
applied
in
the
FY86
NHATS
composite
sample
data
analysis.
The
statistical
analysis
of
FY86
"
ATS
data
had
three
objectives:

Estimate
average
concentration
levels
of
target
semivolatile
compounds
in
the
adipose
tissue
of
individuals
in
the
U.
S.
population
as
well
as
in
various
demographic
subpopulations.

8
Estimate
standard
errors
and
construct
confidence
intervals
for
these
average
levels.

H
Perform
statistical
hypothesis
tests
to
determine
if
average
concentration
levels
of
target
semivolatiles
in
the
U.
S.
population
differ
significantly
by
any
of
four
demographic
factors
(
geographic
region,
age
group,
race
group,
and
sex
group).

The
"
additive
model",
a
statistical
model
developed
to
estimate
average
concentration
levels
in
ind,
ividualspecimens
by
analyzing
NHATS
composite
data,
was
fit
to
the
FY86
data
to
address
each
of
the
above
objectives.
The
additive
model
involves
an
iterative
weighted
generalized
least
squares
method
to
estimate
model
parameters
representing
demographic
effects.
The
resulting
parameter
estimates
are
approximately
normally
distributed
for
large
samples.
This
approximate
normality
is
used
to
construct
confidence
intervals
and
hypothesis
tests.
Derivation
and
validation
of
the
additive
model
is
presented
in
Orban
and
Lordo
(
1989).
\

Section
6.1
briefly
presents
the
additive
model
and
its
necessary
modifications
in
analyzing
the
FY86
data.
The
methods
used
to
obtain
estimates
of
average
concentrations
for
target
compounds,
standard
errors
for
these
estimates,
and
hypothesis
tests
for
the
significance
of
demographic
effects
on
the
concentrations
are
presented
in
Section
6.2.

6­
1
6.3.
THE
24DDITIVE
MODEL
In
order
to
expand
the
"
ATS
to
address
a
broader
range
of
compounds,
it
was
necessary
to
develop'mass
spectrometry­
based
analytical
methods
that
provided
detailed
chemical
information
and
supported
method
specificity.
These
analytical
methods
required
larger
tissue
samples
than
the
available
samples
from
individual
patients.
As
a
result,
the
individual
adipose
tissue
specimens
were
composited
prior
to
chemical
analysis.
The
additive
model
was
developed
to
achieve
the
"
ATS
statistical
objectives
under
the
sample
compositing
scenario.
The
additive
model
was
used
to
analyze
the
FY87
mTS
dioxin
and
furan
concentrations
in
composite
samples
(
USEPA,
1991).
The
FY86
"
ATS
was
the
first
study
in
which
the
additive
model
was
applied
to
semivolatile
composite
data.
Orban
and
Lordo
(
1989)
have
shown
that
the
additive
model
has
the
following
attractive
features:
I
w
Under
very
general
assumptions,
the
additive
model
produces
asymptotically
unbiased
estimates
of
average
concentration
levels
in
the
population.

W
The
additive
model
establishes
a
more
tractable
relationship
between
the
distribution
of
analyte
concentrations
in.
individuals
and
the
distribution
of
measured
concentrations
from
the
composite
samples.

The
latter
feature
is
particularly
important
because
individual
specimens
are
collected,
but
the
chemical
analysis
is
performed
on
composite
samples.
Table
6­
1lists
the
categories
of
the
four
analysis
factors
of
interest
to
the
NHATS.
The
additive
model
assumes
that
the
four
analysis
factors
have
fixed
additive
effects
on
the
average
concentrations
in
specimens.
This
assumption
subdivides
the
population
into
48
alsubpopulationsll
defined
by
the
4x3x2x2=
48
unique
combinations
of
categories
for
the
four
factors.

6­
2
Table
6­
1.
"
ATS
Analysis
Factors
and
Categories
Census
region
Northeast
North
Central
South
West
4
Age
group
0­
14years
3
15­
44
years
45+
years
Race
group
Caucasian
2
Noncaucasian
­­
Male
2Sex
group
I
.
Female
I
Total
Number
of
Subpopulations
(
combinations
of
the
four
analysis
factors):
48
In
addition
to
the
four
analysis
factors,
there
are
three
ancillary
factors
that
have
random
effects
on
NHATS
data.
Two
of
these
factors
have
random
effects
on
the
actual
concentration
in
individual
specimens.
They
are:

effect
of
MSA
sampling
1
effect
of
sampling
individuals
within
MSAs
(
and
selecting
specimens
from
individual
donors)

The
third
has
a
random
effect
on
the
measured
composite
concentrations:

measurement
error
of
compound
concentrations
in
the
composite
samples.

A
fourth
ancillary
factor
applied
specifically
to
the
FY86
composite
data
is
the
fixed
effect
of
laboratory
batches
4
and
5
on
the
measured
composite
concentrations.
Analysis
of
FY86
6­
3
.
QC
sample
data
(
Section
5.3)
found
significant
differences
for
a
majority
of
target
compounds
in
the
measured
concentrations
for
Batches
4
and
5
versus
those
in
the
first
three
batches.
Thus
a
"
batch
class"
factor
has
been
included
in
the,
additivemodel
for
analysis
of
FY86
NHATS
semivolatile
data
on
composite
samples,
From
these
assumptions,
the
actual
concentration
CijkCm
in
a
.,
sbecimenfrom
the
ithdonor
in
MSA
j,
census
region
k,
age
group
e,
sex
m,
and
race
group
n,
is
represented
by
where
p
is
a
constant,

CRk
is
the
fixed
effect
of
census
region
k
(
k=
1,2,3,4),

At
is
the
fixed
effect
of
age
group
1!
(
1!=
1,2,3),

S,
is
the
fixed
effect
of
sex
group
m
(
m=
l,
2),

%
is
the
fixed
effect
of
race
group
n
(
n=
1,2)#

MSA,
is
the
random
effect
of
selecting
MSA
j
(
j=
1,2,...
I,

cij
is
the
random
effect
of
selecting
individual
i
in
MSA
j
.

To
uniquely
define
the
fixed
effect
parameters,
let
Thus
CR,,
A,,
S2,
and
R2
are
defined
as
a
linear
combination
of
other
effects,
leaving
eight
fixed
parameters
in
(
6­
1)
which
can
be
uniquely
estimated.
The
effect
MSAj
in
(
6­
1)
is
a
random
effect
due
to
the
selection
of
MSAs
prior
to
selecting
individual
specimens.
This
effect
is
assumed
to
have
mean
zero
and
variance
u:.
Meanwhile,
the
effect
eij
in
(
6­
1)
is
random
due
to
selecting
individuals
randomly
within
an
MSA.
The
distribution
of
eij
has
mean
zero
and
variance
a2Eand
is
independent
from
the
distribution
of
MS%.
Data
analysis
results
through
the
history
of
the
"
ATS
program
have
concluded
that
variation
in
specimen
concentrations
is
proportional
to
the
average
concentration
level.
This
finding
is
generally
true
in
most
environmental
monitoring
programs
where
chemical
concentrations
are
measured.
Thus
if
ps
is
the
average
concentration
level
in
subpopulation
s,
then
'
it
is
assumed
that
for
subpopulation
s
(
s=
l,...,
48),
there
exists
a
positive
number
b
such
that:

For
notational
simplicity
we
let
where
the
combination
of
indices
k,
C,
m,
and
n
define
subpopulation
s.

Equation
(
6­
1)
defines
the
model
for
the
actual
concentration
in
a
specimen
collected
in
the
FY86
NHATS.
However,
as
specimens
are
composited
prior
to
chemical
analysis,
measured
specimen
concentrations
Cijktm
are
not
observed.
Instead,
data
are
obtained
from
the
chemical
analysis
of
comDosite
samples.
Assuming
data
exist
for
C
composites,
and
letting
Yh
represent
the
measured
concentration
of
composite
h
(
h=
l,...,
C),
the
natural
additive
effects
of
cornpositing
imply
that
where
Cijs
is
the
actual
concentration
in
specimen
i
from
MSA
j
and
subpopulation
s,

Ch(
i,
j,
s)
is
equal
to
1
if
specimen
i
from
MSA
j
and
subpopulation
s
is
in
composite,
h,
and
is
equal
to
6­
5
.
..
..
..
.
..
....
.
.
....
.
.
..
..
.
.
­.__
­­­.
.
.
­.­.
..
.
zero
otherwise,

Mh
is
the
number
of
specimens
in
composite
h,

B4,
is
the
fixed
effect
of
analysis
in
Batches
4
and
5
on
the
Composite
concentration,

Ih
is
equal
to
1
if
composite
h
was
analyzed
in
Batches
4
or
5,
and
is
equal
to
zero
otherwise,
and
yfi
is
random
measurement
error
associated
with
composite
h,
assumed
to
have
mean
zero
and
variance
a;.

Because
Cijs
is
associated
with
demographic
effects
as
specified
in
equation
(
6­
11,
equation
(
6­
2)
relates
the
measured
composite
concentrations
with
the
demographic
effects
in
Table
6­
1.
Note
that
the
term
B,,
has
been
placed
in
the
model
in
(
6­
2)
as
a
result
of
the
QC
data
analysis
on
FY86
NHATS
data.
It
is
not
a
standard
term
in
the
additive
model
for
all
NHATS
applications.
The
statistical
analysis
performed
on
the
additive'model
in
(
6­
2)
will
be
explained
in
terms
of
matrix
notation.
Matrices
are
denoted
by
capital
letters.
Matrices
and
vectors
are
denoted
in
bold.
Let
be
the
9x1
vector
of
fixed
effects
from
equations
(
6­
1)
and
(
6­
2)
on
the
vector
of
composite
concentrations
y
=
(
Yl,
Y,,
...,
Yc)
I.

Fixed
effects
omitted
from
@
can
be
specified
as
a
linear
combination
of
the
effects
in
8.
Let
I.(
=
(
pI,*..,
p48)'
be
a
48x1
vector
containing
the
unknown
average
concentrations
from
the
48
subpopulations.
Then
~
c
is
calculated
as
p
=
X@
for
some
48x9
design
matrix
X.

If
the
QC
data
analysis
(
Section
5.3)
found
the
average
concentration
in
Batches
4­
5
to
be
significantly
different
from
that
for
the
first
three
batches,
the
matrix
X
is
constructed
so
that
p
will
depend
on
the
effect
B45.
In
this
situation,
two
average
concentrations
will
be
associated
with
each
+
subpopulation,
one
for
Batches
4­
5
and
one
for
Batches
1­
3,
This
is
due
to
potential
biases
attributed
to
the
results
in
Batches
4
and
5.

The
expected
value
of
the
composite
concentrations
y
is
given
by
where
2
is
a
Cx48
composite
design
matrix.
Thus,
according
to
the
additive
model,
both
the
actual
concentrations
of
the
individual
specimens
and
the
measured
concentrations
of
the
composite
samples
have
expected
values
that
are
linear
combinations
of
the
additive
effects
of
the
fixed
analysis
factors
in
8.
Orban
and
Lordo
(
1989)
also
show
that
the
variance­
covariance
matrix
of
y
{
denoted
by
Vy)
is
a
block
diagonal
matrix
that
depends
on
a:,
a:,
and
at.

6.2
STATISTICAL
ANALYSIS
OF
COMPOSITE
SAMPLES
This
section
describes
the
specific
methods
used
to
achieve
the
statistical
objectives.
The
estimation
methods
are
discussed
in
Section
6.2.1,
and
the
hypothesis
testing
procedures
are
presented
in
Section
6.2.2.
This
section
refers
to
terms
and
symbols
presented
in
Section
6.1.

6.2.1
Estimation
6.2.1.1
Estimating
Native
Compound
Levels.
The
specific
quantities
estimated
for
the
FY86
"
ATS
are
the
average
concentrations
in
the
adipose
tissue
of
the
U.
S.
population
and
*

the
average
concentrations
for
each
of
the
eleven
"
marginalt1
demographic
populations
defined
by
the
categories
listed
in
Table
6­
1.
These
estimates
were
calculated
in
three
steps:

6­
7
..
._...­..
­..
.­..
.___.
1.
The
additive
model
parameters
(
vector
(
s
in
Section
6.1)
were
estimated
using
a
method
called
iterative
weighted
,
generalized
least
squares
(
IWGLS).

2.
Estimates
of
average
concentration
levels
in
the
48
subpopulations
defined
by
the
four
analysis
factors
(
vectorp
in
Section
6.1)
were
calculated
from
the
parameter
estimates.

3.
National
and
marginal
population
estimates
were
obtained
by
taking
weighted
averages
of
the
appropriate
.
subpopulation
estimates
in
p.
Weights
were
proportional
to
the
population
counts
from
the
1980
U.
S.
Census.

To
obtain
asymptotically
unbiased
estimates
of
the
fixed
effects
in
B,
it
is
not
necessary
to
make
any
assumptions
about
the
form
of
the
distributions
of
the
random
effects
in
equation
(
6­
2).
If
the
variance­
covariance
matrix
Vy
of
the
vector
of
measured
composite
sample
concentrations
y
were
known,
the
method
of
generalized
least
squares
(
GLS)
produces
estimates
of
.
P
that
are
unbiased
and
have
minimum
variance
among
all
unbiased
estimates.
Furthermore,
if
the
errors
are
normally
distributed,
the
GLS
estimates
are
equivalent
to
the
maximum
likelihood
estimates.
The
GLS
estimate
of
@
is
given
by
5
=
(
DIV,­~
D)­~
DT,­~~
I
(
6­
4)

Awhere
D
is
defined
in
(
6­
3).
The
variance­
covariance
matrix
of
0
is
given
by
Unfortunately,
V,,
depends
on
three
unknown
variance
components
2
(
urn,
u2,,
and
0%)
from
(
6­
1)
and
(
6­
2),
as
well
as
on
the
vector
8.
Therefore,
Orban
and
Lordo
(
1989)
proposed
a
method
involving
iterative
weighting.
Thus
the
method
is
called
iterative
weighted
generalized
least
squares
(
IWGLS).
The
IWGLS
procedure
requires
starting
values
for
the
unknown
parameters.
These
starting
values
were
calculated
using
the
P3V
program
of
the
BMDPw
software
package.
This
program
uses
a
maximum
likelihood
procedure
in
fitting
a
mixed
model.
The
resulting
estimate
of
Vu
was
then
used
in
the
GLS
formula
to
produce
a
revised
estimate
of
#.
The
IWGLS
procedure
provided
continual
updating
of
the
estimates
for
Vu,
continuing
until
convergence
criteria
on
the
estimate
of
#
and
the
error
sum
or
squares
were
met.
Orban
and
Lordo
(
1989)
discuss
this
method
in
more
detail
and
describe
special
computer
programs
in
the
SAS"
System
for
implementing
IWGLS.
They
also
provide
formulas
for
calculating
the
stafldard
errors
of
the
estimates.
If
5
denotes
the
final
estimate
of
fl
from
the
IWGLS
procedure,
then
an
estimate
of
the
average
concentration
level
in
each
of
the
48
subpopulations
is
calculated
by
A
where
X
is
a
design
matrix.
The
variance­
covariancematrix
of
p
is
given
by
The
estimates
in
are
affected
whenever
batch
class
effects
are
present.

.
Weighted
averages
of
the
appropriate
subpopulation
A
concentrations
ps
are
calculated
to
estimate
fImargina1"
averages
for
the
categories
of
each
analysis
factor.
For
example,
if
the
set
of
12
of
the
48
subpopulations
found
in
the
Northeast
census
region
is
represented
by
NE,
then
the
estimated
average
concentration
in
the
Northeast
census
region
is
given
by
where
ws
is
the
proportion
of
total
population
in
the
Northeast
census
region
that
is
found
in
subpopulation
s
(
as
determined
by
­­
I,

1980
U.
S.
Census
figures).
Marginal
estimates
were
calculated
for
four
census
regions,
three
age
groups,
two
race
groups,
and
two
sex
groups.
The
U.
S.
population
estimate
was
calculated
in
the
same
way,
with
weights
corresponding
to
the
proportion
of
the
U.
S.
population
in
each
subpopulation.
Standard
errors
for
the
marginal
estimates
were
calculated
based
on
the
standard
errors
of
the
subpopulation
A
estimates
ps.
If
Var(
k)
indicates
the
estimated
variance
of
is,
then
the
standard
error
of
the
marginal
estimate
of
&
in
(
6­
5)

is
given
by
where
NE
and
ws
are
as
defined
in
(
6­
5).
An
approximate
95%

confidence
interval
for
each
estimate
was
calculated
by
adding
and
subtracting
two
times
the
standard
error
of
the
estimates.

6.2.1.2
Characterizing
PCB
Results.
Laboratory
analysis
in
the
FY86
"
ATS
measured
the
concentrations
of
each
of
the
ten
PCB
homologs
in
the
composite
samples.
These
concentration
estimates
were
integrated
to
characterize
the
nature
of
PCBs
detected
in
adipose
tissue.
If
pi
is
the
average
concentration
level
(
ng/
g)
of
the
ithPCB
homolog
(
only
p4
through
p8
were
estimated
in
the
statistical
analysis),
then
the
characterization
considered
the
.,

following
three
sets
of
information:

U
Total
PCB
concentration
(
ns/
cr)
the
sum
of
the
*­
estimated.
concentrations
for
each
homolog:

6­
10
­­

­­
w
Chlorobinhenvl
distribution
across
homoloss
(%
I
the
percentage
of
the
total
PCB
concentration
attributed
to
the
ith
homolog
(
ipl,...,
101
:

rn
Chlorination
level
(%
I
the
sum
of
the
chlorobiphenyl
distribution
percentages,
each
weighted
by
the
homolog's
chlorine
mass
fraction
(
ClMF):

These
PCB
parameters
were
estimated
by
substituting
estimates
of
the
homolog
concentrations
pi
in
the
hove
equations,
as
obtained
from
the
statistical
analysis
(
Section
6.2.1.1).
However,
statistical
analysis
was
performed
only
on
five
of
the
ten
PCB
homologs
(
tetra­
through
octa­
CB).
The
remaining
five
homologs
were
each
detected
in
no
more
than
30%
of
the
FY86
"
ATS
composite
samples,
Thus
in
estimating
the
above
PCB
parameters,
it
is
assumed
that
pi10
for
i=
1,2,3,9,10.
While
this
approach
may
lead
to
an
underestimate
of
total
PCB
concentration,
the
extent
of
underestimation
is
expected
to
be
very
low.
To
estimate
the
level
of
chlorination,
the
value
of
the
ClMF
is
0.4856
for
tetra­
CB,
0.5430
for
penta­
CB,
0.5893­
for
hexa­
CB,
0,6277
for
hepta­
CB,
and
0.6598
for
octa­
CB.
The
standard
errors
of
the
above
PCB
parameters
were
calculated
from
the
variability
estimates
in
the
average
concentration
levels
far
the
individual
PCB
homologs
(
Section
6.2.1.1).
If
pi
is
the
estimate
of
pi
as
obtained
from
the
statistical
analysis,
then
standard
error
estimates
are
given
as:

standard
error
of
total
PCB:
(
6­
10)

(
6­
11)
standard
error
of
chlorobiDhenv1
distribution
oercentases:
standard
error
of
level
of
chlorination:

(
6­
12)

Approximate
95%
confidence
bounds
for
the
PCB
parameters
were
taken
as
plus
and
minus
two
standard
errors.

6.2.2
Hwothea3.
s
Teatinq
Hypothesis
tests
were
performed
to
determine
if
average
concentration
levels
differ
significantly
by
any
of
the
geographic
or
demographic
factors.
The
specific
hypotheses
tested
were
Ha
:
CR1
=
CR2
=
CR,
=
CR,
=
0
HAGS:
A,
=
A2
=
A3
=
0
HSEX
 
s,=
s2=
o,
HRACE:
R,
=
R2
=
0
I
HB45
:
B4,
=
0
.

The
hypothesis
HCRtfor
example,
states
that
there
are
no
differences
in
average
concentration
levels
among
the
four
census
regions.
Each
hypothesis
was
two­
tailed;
that
is,
each
alternative
was
that
at
least
one
effect
was
nonzero
and
different
from
the
others.
In
order
to
test
these
hypotheses,
it
was
necessary
to
make
specific
distribution
assumptions
for
the
random
effects.
It
was
assumed
that
the
errors
associated
with
sampling
MSAs,

6­
12
sampling
individuals
within
MSAs,
and
measuring
concentrations
were
independent
and
normally
distributed.
The
additive
effect
of
compositing
specimens
suggests
that
the
normality
assumption
is
reasonable
because
specimen
sampling
errors
are
averaged
in
the
composite
sample.
Statistical
theory
states
that
averages
and
sums
are
approximately
normally
distributed.
Distributional
assumptions
were
tested
for
all
target
compounds
using
probability
plots
and
residual
analysis.
The
likelihood
ratio
method
was
used
to
test
the
above
hypotheses.
In
this
process,
the
additive
model
is
fit
to
the
observed
data
both
including
and
excluding
the
effects
to
be
tested.
According
to
asymptotic
theory,
the
log
of
the
ratio
of
the
likelihood
functions
from
these
two
fits
has
approximately
a
chi­
squared
distribution,
with
degrees
of
freedom
equal
to
the
number
of
independent
parameters
constrained
under
the
null
hypothesis.
Orban
and
Lordo
(
1989)
developed
programs
in
the
SAS@
System
to
perform
these
tests.

::

?.

6­
13
7.0
RESULTS
This
section
contains
the
results
of
the
statistical
analysis
of
the
FY86
"
ATS
for
sernivolatiles
in
human
adipose
tissue.
The
applied
statistical
methods
were
discussed
in
Chapter
6.
The
objectives
of
the
statistical
analysis
were
as
follows:

m
Estimate
average
concentration
levels
of
target
compounds
for
individuals
in
the
U.
S.
population
and
in
'
various
subpopulations;

m
Calculate
standard
errors
and
confidence
bounds
on
these
average
levels;

m
Perform
statistical
hypothesis
tests
to
determine
if
average
levels
differ
significantly
between
various
levels
of
demographic
factors
of
interest.

statistical
analysis
was
performed
on
data
obtained
from
laboratory
analysis
of
50
composite
samples.
The
composites
were
prepared
using
a
total
of
671
adipose
tissue
specimens
from
sampled
cadavers
and
surgical
patients.
Each
composite
contained
from
three
to
24
specimens,
with
an
average
of
13.4
specimens
per
composite.
The
specimens
within
each
sample
originated
from
a
common
census
division
and
age
group
but
may
have
differed
among
sex
and
race
groups.
Additional
information
on
sample
and
composite
design
is
presented
in
Chapters
2
and
3.

A
descriptive
summary
of
the
observed
concentrations
for
the
111
semivolatiles
is
provided
in
Section
7.1.
Statistical
analysis
was
performed
only
on
"
target"
semivolatiles
(
identified
in
chapter
5)
that
were
detected
in
a
majority
of
the
50
composite
samples
and
which
met
specific
data
quality
objectives.
Resulting
from
this
statistical
analysis,
estimates
of
average
concentrations
are
presented
in
Section
7.2,
along
with
standard
errors
and
confidence
bounds
on
these
estimates.

c
cection
7.3
presents
the
results
of
statistical
hypothesis
testing
to
identify
significant
effects
of
demographic
factors
on
sverage
concentration
levels.
Section
7.4
describes
the
outlier
7­
1
I
detectian
procedures
that
identified
potential
data
errors
to
be
corrected
prior
to
conducting
the
statistical
analysis.
Finally,
as
part
of
the
commitment
to
overall
data
quality
in
this
program,
procedures
were
implemented
to
demonstrate
the
validity
of
the
statistical
methodology
applied
to
the
FY86
NHATS
data.
The
results
of
this
data
validation
procedure
are
presented
in
Section
7.5.

Unless
otherwise
specified,
all
statistical
analyses
were
performed
on
composite
concentrations
adjusted
for
recoveries
of
surrogate
compounds.
This
adjustment,
discussed
in
Section
5.2,
corrected
for
systematic
error
identifiable
through
the
surrogate
recovery
data.

7.1
DESCRIPTIVE
STATISTICS
Prior
to
statistical
modelling
of
target
compour;
ds,
simple
descriptive
statistics
were
generated
on
the
measured
concentrations
for
all
111
semivolatiles
analyzed
in
the
NHATS
FY86
campaign.
These
statistics
summarized
the
laboratory
results
across
all
50
FY86
composite
samples
and
consisted
of
the
following:

1
arithmetic
average;
rn
standard
deviation;
rn
standard
error
of
the
average;
rn
percent
of
samples
with
detected
results
(
duplicated
from
Table
5­
11
;
rn
average
level
of
detection
(
LOD).

Table
7­
1presents
these
statistics
across
the
111
semivolatiles
for
measured
concentrations
adjusted
for
surrogate
recoveries,
as
well
as
on
the
unadjusted
concentrations.
A
compound
is
detected
within
a
composite
sample
if
the
result
is
classified
as
either
a
trace
or
positive
quantifiable
reading.
Prior
to
summarizing
the
data
for
a
given
compound,
the
measured
concentrations
for
all
samples
with
not­
detected
outcomes
were
replaced
by
one­
half
of
the
reported
LOD.
While
;?{

the
LOD
itself
was
not
adjusted
for
surrogate
recoveries,
the
7­
2
c
9­
t
(",
1
4
.......

wTw(
zw(
z(
z
a
mrla
morl
(
z
m
(
z*
m
wmwee
m*
mamw
.................
...
4oooooowocyool*
o*
a~
6wwm
m
om
d
dd
mm
QD
f9Wd
om0
*
0
rl0d
W'rnPrl
OOIOOcyrl....
....
.
w
...
........
w*
oomln*
w*
rl*
amool*
o*
QDwP
d
o*
d
wmea
~­
PW*
N
,
cy
rid
d
cym
m
mmm
orla
m
.........
....
.??
Y",??
O'la
*
YOrn*
o*
oowm**
lnolmmmad­
4odP*
a
0
mol
Q)
rlwo*
rlw*
wm
cy
wm
rl
rlrl
cycy
rl
olpcd
ow*
0
........?????........
(
z
mal­(
z512(
zmo(
z
Poooooooocyorlwcy(
zo~
PQDQD*
m
wnt
cy
d
drlm*

*
dol
PmoQDmPIOcyoloolaol
m
..
................
a0
mmrlocyddmdowoom
wm
rl
dddmdrlrlmmcycyrlflcy
*'
OD
8­
44
woooooocyo*
oo*
acyocycycyoN
ol
00
ol
mem
mrlw
drl
7­
3
0
ocywmcyQDrlmom
a.......
omodorlo
d
.......
mmm(
zw(
zu,
Q\

v)
(
zaolF)
cyw
b*
rlolw
wmu)
maa(
z
0
d
(
owwcyw
......
0000000
d
cy
mQ)
lnwmwwwmwom
.......
*
m*
m*
lnm
b
'*
­
Paam*
o*
rlQ)
oma
.......
mcIoDlnmmmm
(
z
rl
I­
m
a'cy
rl
c
OWOdOrld
rlrldrldrld
0~
00000
a
rl­
0
o\
PwoP*
00d
fn0000Q\
*
wwPPwwww
....
.
.
.
.
.
.
.
.
.
ooo*
NO
0
­
00
00
0
0
wri*
c­
Q\
www
m
*'
m
0
4)
m
0
00
.
.
.
.
.
.
.
.
.
0
mm
w
m
mm
m
*'
(
v
rlQ\
olmd\
wrlm
....
ml­
wl­
c­
yw~
y
(
I....

.
000111
cy00000000
....
m**
w
m
d
cIc
olODP.
o\
mqm.
.
.
.
.
.
.
.
.
oorlPwcyo4)
mm
dCr)(
Y
m*
cv00Q)
cl
U3rll­
i
..........

w4)
rlo
u~
oc­(
vrl~
cy~~
....
.
.
.
.
.
.
.
.
.
doom
mrlej(
vrloooQ\
rldrlrn
rll­
irldrlddd
rl
0004)
*
ooo4)
cyo*
o
m
00
7­
4
ow
*
a
..
a+*
...
my,
wmcv
0moD*
w..
...
WWu)
**
T*
md
.....
wo
omm
..
...
mPOmm
m*
wamdrlWVIt­
a
mu)
u)
cvcv
..
cvmw­
0
...
**
moa,*
m
m*
rl
waoo.....
mqm
..
mc­
mwo
wow
rirlwmrl
wmcv
0
rl
PO
..
...
ma
POWrlm
mm*
rid
rl
oowww
00
dPt­
P
wwoom
u)
m~
cv
(
cIU)
Pu)
PQI
u)
wwrlwcvw..............
 
oo~
oorlwooomoP(
v
cv
mm~~
rl
mwa
a
wp 
c­
wwrl
c­
p wmworl..............
OOOOocvmooomowm
cv
.............
 
*
mmmwrl*
mmmrlmwa
mu)
d
* 
cycv
W
w*
oo**
Pmcvc­
mPam.
.
. 
,
.
.
.
.
.
.
.
.
.
.
arlcvocvPrlddorloPP
liddddrlddrldrldm
oooooowoooQ)
oww
Wo\
rlP
mcyvr
0
m
m0rlVI....
cy
OWPlF
m
cn
l­
0CI)
Q)
VI.
.
.\
D
l­
cy*
rlCY
rl
rlCY
OD
7
OD"?**..
*
(
ycnoo
CY
rlmcn
m
OD
....
cy
0CYmrl*

CY
w
mODCYVI.
.
.
a)
cn
rJm*
cy
rl
rlCY
A
cn.
m
cn*
rlP
r,
....
a
8
m
CYorlm
CY
CYmm
Y
0
*
I
VI
QIODm0....
0
m
*
wCY
0
00o­
r
CY
o***
o
0wcnorlwODorl0
...
.
.
.
.
.
.
.
rlorloowo(
no
wmplrlrl
cnrl(
Ycy**
o(
Yo
.
.
.
.
.
.
.
.
.
rlrtrlo*
oDowo
om000
0
0VIwom\
orlo~
o
.
.
.
.
.
.
.
.
.
mm*
mcnwmrlm
rlrl
000
00
0
000000000
.
.
.
.
.
.
.
.
.
ulVIVIomVIolno
rl
rlrl
7­
6
cn
mr
0000000wCY
.
.
.
.
.
.
.
.
.
ooooooomo
00000*
0l­~
ooooowo
.
.
.
.
.
.
.(
n
.
cn
.
0000000CYrl
0
0
ooooorloom
.
.
.
.
.
.
.
.
.
ooooooo*
o
ooooorto
CYooooo*
omrl
......
ooooood~~
CY
0000000
0
ooooorlmom
,
.
.
.
.
.
.
.
.
VImVIVIVIVICYoVI
rl
0
000000000
.
.
,
.
.
.
.
.
.
oooooomoo
rlrlF4rlrtrl
drt
oooooCYo*
cy
(
vdW
l­
ro
mOmoPo~
a
wwmo
(
v
rl
....
.
.
...
.
.
.
.
.
.
*
hoomo
Nmlnrloo*
Orlo~
(
v
(
Y
(
v
m
8
­
ti
JJ
U
a
m
v
m
a
4
k
a
z
u
8
a
*
wm0
rlN
morlomod
**
In
00
0
JJ
....
.
.
.
.
.
.'
.
.
.
.
.
8k
oolno
**
wm00000rl0m
cv
(
v
3
m
k
0
.
'(
v.
w
m**
0
u
..
rl
JJ
f
00
00
9
cnm
(
vowomom
...........
9
(
vrlmmwmmowlnd
rfrlmm
rid
d
rl
d
u
9
­
rl
30000
000
m0000
ooooooooooa
.
I..
.
e.........

lnlnlnln
omlnmoooooom
6
(
v
ddd(
vdrl
0
u
fa
k
0
wo*
o
O*~(
v~
0000(
Y00
9(
vrl
mmwm
rl
rc
a,

9
x
.
Q
3
JJ
k
0
8
k
a
fa
6"
0
­
rl
u
fa
k
u
9
i
u
modified
measured
concentration
was
adjusted.
No
LOD
was
The
percentage
reported
for
detected
compounds
within
a
sample.
of
samples
in
each
of
three
qualifier
classifications
(
not
detected,
trace,
and
positive
quantifiable)
was
summarized
in
Table
5­
1
of
Chapter
5.

Appendix
E
contains
the
minimum,
median,
and
maximum
reported
concentrations
across
the
50
composite
samples
for
each
of
the
111
compounds.
These
statistics
are
based
on
concentrations
which
are
unadjusted
for
surrogate
recoveries.
The
descriptive
statistics
in
Table
7­
1are
based
on
simple
averages
of
the
measured
concentrations
within
the
50
composite
samples.
As
such
they
only
summarize
the
observed
data.
They
should
not
be
used
to
estimate
concentration
levels
within
the
population.
Statistical
analyses
were
implemented
to
obtain
population
average
estimates
for
seventeen
target
semivolatiles
meeting
specific
data
quality
objectives.
results
of
these
analyses
are
presented
in
the
following
sections.
The
7.2
POPULATION
ESTIMATES
FROM
STATISTICAL
MODELLING
The
statistical
modelling
techniques
presented
in
Chapter
6
were
used
to
determine'estimatesof
average
concentrations
for
selected
semivolatiles
within
subpopulations
as
well
as
for
the
entire
nation,
to
obtain
estimates
of
uncertainty
inherent
in
these
estimates,
and
to
identify
where
significant
differences
in
average
concentration
were
present
among
subpopulations.
These
techniques
centered
around
the
additive
model,
which
was
used
to
estimate
average
concentration
for
individuals
as
a
function
of
several
demographic
factors.
The
results
from
fitting
the
additive
model
to
the
NHATS
FY86
composite
data
are
presented
in
this
section.
Not
all
of
the
compounds
analyzed
in
the
FY86
NHATS
analysis
provided
sufficient
composite
concentration
data
to
warrant
a
meaningful
statistical
analysis.
Seventeen
of
the
111
compounds
were
identified
as
containing
a
sufficient
number
of
7­
a
p
?
I,:/
g­
4;
0
detected
samples
and
whose
analytical
measurements
were
deemed
accurate
in
reflecting
the
true
concentration
level.
Having
a
sufficient
number
of
composite
samples
with
detected
results
ensured
that
only
minimal
bias
was
generated
by
substituting
one­
half
of
the
detection
limit
for
the
measured
concentration
whenever
the
compound
was
not
detected
by
the
analytical
method.
Method
performance
was
determined
from
analysis
of
the
QC
data
(
Section
5.3),
which
indicated
the
presence
of
batch
effects
and
the
extent
that
anomalous
analytical
results
were
reported.
The
compounds
selected
for
statistical
modelling,
as
well
as
the
criteria
used
to
select
them,
were
identified
in
Chapter
5.

Fitting
the
additive
model
to
the
"
ATS
FY86
data
for
17
semivolatiles
resulted
in
average
concentration
estimates
for
the
entire
U.
S.
population,
as
well
as
'
Imarginalf1estimates
for
each
of
the
categories
defined
by
the
four
analysis
factors
presented
in
Table
6­
1
(
census
region,
age
group,
race
group,
and
sex
group).
The
formula
for
calculating
marginal
estimates
was
given
in
equation
(
6­
5)
of
Section
6.2.1.
The
estimates
are
presented
in
Table
7­
2
for
the
four
census
regions,
Table
7­
3
for
the
three
age
groups,
Table
7­
4
for
the
two
race
groups,
and
Table
7­
5
for
the
two
sex
groups.
Table
7­
6
presents
estimated
concentration
estimates
for
the
entire
nation.
The
estimates
are
asymptotically
unbiased
and
were
adjusted
for
the
presence
of
laboratory
batch
ef­
Eects(
Batches
1­
3
versus
4­
5)
and
for
population
percentages
based
on
the
1980
U.
S.
Census.
Accompanying
the
marginal
estimates
based
on
the
additive
model,
standard
errors
and
approximate
95%
confidence
intewals
of
these
estimates
are
displayed
in
Tables
7­
2
through
7­
6.
The
standard
errors
were
calculated
using
equation
(
6­
6)
of
Section
6.2.1
and
are
used
to
characterize
the
statistical
uncertainty
in
the
estimated
average
concentrations.
The
standard
errors
are
presented
in
both
absolute
and
relative
terms.
The
confidence
intervals
represent
the
marginal
estimate,
The
actual
plus
and
minus.
approxirnatelytwo
standard
errors.
,

2E
k
PP
Q
a
9
U
YYYV
WWYY
P
ml­
mrl
....
....

PP
4
4
0
....
4
....
U
e
Q
a
0
2
a
k
k
6
8
8
0
?
3
cda
9
a
U
3
x
ci
a7
m"
x
9
I
7­
10
..
r­
c$
3
....
dt
$
2
....

ri*
cu*
0
r
*
W
0
wmwm
*
9
w
cumcum
c
r(
cl
d
rl
l­
l
0
le
le
le
le
­
d
k
bl
uu
k
k
kuu
uu
d
$
l:
dm
kio
ale
UW
82
UW
3
s
559,
U
a
a
s
0
.
d
rl
k
leE?
a
U
0
d
4)
2
.
A
a
n
B
7­
11
nY)
Y)
Q)
one0
....
U'

a"!

..­.....
rl(
AODOD
(
ummto
(
umcyrl
YYYV
Y)
mrl(
u
....
....
....
­­
en
PO(
AV
......
FrlODOD
rlmrlrl
.....
...
0400
.&
ow&*

....
mw­
a
ocroo
.(
u
VYYY
Y
YYI­

Pmrl
oomul
OOOY)
....
rlrkrlrl
m(
DmY)
rl
(
u(
Amm....
....
m*
mP
me00
....
Q)(
Yoo....(
ucrmq
m(
Am(
A
OD
om*
rlhlrlrv
(
AODu)
P
....
....

c
0
4
r;
k
c
1;
I
4
3u
uu
L:
ua,.
$
2
31)
W
JW
IW
999,
L
kklrn
L
0004,
L
ZZWS
d
I
­
rJ
c
ff
I:
4
­
P
k
3
a
a
k4
0
5
m
0
m2
a
Ea
cd
V
U
0
U
r­
4c
a
a
7­
12
I
1
,

Ann­

....

­.
I.

0000
....
0000
0
0....
ONdW
Q)*
CIP
dd
t­
4
E
UJJ
8:
OW
al
U
(
d
4J
8
u
Id
rl
x
$
X
n
d
0
­
d
u
a
­
8i
k
0
!
v.
c3l
0
0
r(

8
8­
4
rl
(
d
u
..
W
rl
m
II
N
\
E
.
h
cv
h
cqy
7­
13
k"
0
k
k
m
co
1
a
9
u
m
h
k
0
?
?
8
a
a
ri
ri
x
k
0
*
0
a
r?
a,
a,

7­
14
(
YOU3
..>
..
l
WO
mal
wmu)
dmu)
mrll­(
Ymm
YUV
VYY
ddm
d(
mm
...
'
424
raUId
wmw
mrara
N
oow
...
...
rlwo
mrin
OWN
md(
w
dd
rl
4
­
4
?
­
3.
rl
ne

..­­.
..
...
mm.
om
.
rad(­...
...
mad
ow*
0Q)
Cl
dd4
dull­
mm
d
d
YYY
UYY
YWV
YYU
mmm
ml­
m
10
m
4.
owa
...
...
...
...
(
YIDd(
Wl­
l­(
Ym0
m
w'
(
Ydd
*
d
d(
Nd
(
v
qcud
..
...
NmN
mora
maa
amm
m
d(
w
m...
wal­
Ndd
dra
m
m
m
mLc
mtc
rl
rl
h
2
c
8
2aa
­
4­
rl
%
2
k
k
0
0
rl
4
Id
JJ
8
X
7­
15
ri
...
...
&&&
OOP
...
*
rip
.........
Pdm
mwriQOm
*
00Dm
dwm
r(
*
a
YYV
YYY
WYY
Q'C1u.
r
Q
3kOU
...
2
000
...
4.
000
rim
...
m
lorl
ewm
lo
0Q)
P
li
rl
FY
a
N
n
\
U
E
E
acorn
...
0aQ\
WWP
0
rnh
0
la
k
0
5
QU
a
6
m
cd
m
u8
V
h,
0
a
U
Ea
r(
rd
0
cd
rlu
u
x
X
8
d
x"

7­
16
c
I
A­­

h....
k"
0
k"
eo
........

..
..
41
..
..
rad
Ob
OQI
dal
mea
ant
QIra
rap.
OQI
mw
dd
ra
d­$
mw
a
Q.

..
..
QIW
..
Q..
..
Wd
!
w
00
WN
ad
Orl
u)­$
ma
dd
mul
mm
dul
2(
ve
dra
rld(
vra
0
a,
a,
a,
a,
a,
a,
al
a,
U
U
JJ
U
JJ
ua2u3
4gg
je
0
I
cda
a
JJ
a,
ci
d,
p9
7­
17
WO
eaw
om
dal
wm
r­
irl
YW
YY
mp.
..
md
dra
wolc
..
..
....
mm
Wal
mu,
Q1d
ea
me
YY
YY
YY
YY
VY
No\
rlm
ow
..
W*..
..
..
..
al0
OW
QDQI
dra
do
mW
ram
rl
rlm
rad
ram
dd
IzOQI
at­
ulo
md
­­
my,
..
WW
(
vm
..
I
?
c?
irm
drl
UV
vfi
u,(
v
..
(
vu,
vu,..
..
..
..
(
00
m@
4
(
vu)
NIO
rlm
90
rom..
00
mul
..
..
dry
mm
mrc
..
me
mv
rlm
rld
rlm
dd
a,
U
a,
U
cl
cl
x
x
c
c
2
2
a
a
­
d
­
4
3
3
k
k
0
0
rl
c
u
2
(
d
(
d
k
U
U
C
a,
E"
a
h
ci
a
8
c4
rt
a
L)

k"

7­
18
Ol
­
84
­
6
..
9
m0
c
ma
rl
CJV
­.
00
..
00
YY
or(
..
W(
vmlr
V..
rtd
W
Wcr
A
d
m
II
N
\
­
E
(
v.
..
m
..
QIW
OQI
dri
4
0
a,
&
u
b
rl
as!
a,
u3V
*
rl
E
43
$
zo
3
a,
0
P
8
0
!
lU
?
rt
­­

..
­­
­­
­­
­­
­­
­­
­­
­­
­
n
n­­
e
Ou)
­­­
e
(
YO
*.
**
**
eme
me
*­*.
*'
**
(
Ym
mu)
(
Y(
Y
md
m(
Y
rlli
(
Yd
(
Ym
..
.
mu)
..
­
m.
4...
.
.
..
..
.
.
..
?
4
ww
00
(
Yw
em
rle
am
rlli
dc'cz
li
lid
I
!\

mw
a
a"
1)
0
1)
*)
a
Q
a
Q)
Q)
Q)

ul
I
cc
a,

:
k
I
7­
20
­
c.

wul..
hh
..
lor­*
fi
cycy
ulco
mm
..
\
om
..
..
..
m*
..
..
dd
mu,
mrfmfi
ma\
(
YN
ulul
YY
YY
YV
aaw
al
3kO.
j;:
rlnl
ma
Om
r4W..
..
u)
m
..
m*
003
coe
dd
ri
Om
(
Yd
ulul.
I
..
00..
mm
mfi
wdma
mcn
..
..
* 
cy
dcy
mm
ulm(
Ym
wu)

a
a,
d
x
atd
al
:
3
dE
v)
22
2!
8!

d
CC2
x
2
8a
Jz71
a
a
a
4
­
4
k
39
8
0
k
0
rl
0
I
Ea
4
3
U0
E
tdU8
if
X
7­
21
......
...
,

i'

..

Z
4
.

WO
..
CYW
at­

PIk
..
ln
ad
U
VQ
dd
k
s"
0
._
.
Table
7­
6.
Estimates
of
Average
Concentrations(')
for
Selected
Semivolatiles,
With
Standard
Errors
and
Approximate
95%
Confidence
Intervals,
for
the
Nation
from
"
ATS
­
86
Composite
Samples
Pesticides
PrP­
DDT
177.
19.7
11.2
(
137.,
217.)

~,~­
DDE@)
2340.
270.
11.6
(
1792.,
2884.)

Beta­
BHC
157.
24.9
15.9
(
PO?.,
207.)

Heptachlor
epoxide
57.6
4.19
7.3
(
49.2,
66.1)

Oxychl~
rdane(~)
114.
7.52
6.6
(
98.4,
129.)

Trans­
nonachlor
130.
15.3
11.7
(
99.6,
161.)

Dieldrint4)
47.0
7.95
16.9
(
31.0,
63.1)

Chlorobenzenes
1,4­
Dichlorobenzene
90.9
15.2
16.7
(
60.2,
122.)

Hexachlorobenzene
51.3
3.97
7.7
(
43.3,
59.3)

PAHS
Naphthalene
1
20.7
2.37
11.4
1
(
15.9,
25.4)
PCBs
Tetrachlorobiphenyl
56.4
4.70
8.3
Pentachlorobiphenyl
135.
15.3
11.4
Hexachlorobiphenyl
314.
18.4
5.9
Xeptachlorobiphenyl
125.
I
21.9
17.5
~

(
46.9,
65.9)

~

(
104.,
165.)
I
276.,
351.)

Octachlorobiphenyl
42.7
11.6
27.1
(
80.7,
169.)

(
19.3,
66.1)

(
603.,
742.)

(
51.2,
65.4)
Total
PCBs@)
672.
I
34.6
5.2
Level
of
58.3%
3.54
6.1
Chlorination(@
<

Other
(
qualitative)

1­
Nonene
124.)
51.0
41.3
(
20.6,
227.)

Hexyl
acetate
123.
21.5
17.5
(
79.5,
166.)

7­
23
Table
7­
6.
(
cant.)

Notes
for
Table
7­
6:
0'
Data
adjusted
for
surrogate
recoveries
(
see
Section
5.2).
Estimates
are
based
on
1980
U.
S.
Census
figures.
(
2)
p,
p­
DDE
concentrations
use
the
following
response
ion:
m/
z=
316.
Data
results
from
Batch
1
not
included
in
calculations.
c4)
Corrected
(
see
Section
S­
1.2).
(
9
The
estimate
for
Total
PCBs
is
the
sum
of
the
estimated
averages
over
the
five
homologs
included
in
this
table
(
i.
e.,
homologs
detected
in
at
least
44%
of
the
"
ATS
M86
composite
samples).
(
6)
Estimated
percent
level
of
chlorination
is
calculated
as
follows:

8
where
=:
estimate
of
the
percent
of
total
PCBs
for
homolog
i,
and
Bi
=
mass
fraction
of
chlorine
for
homolog
i.
(
0nl.
y
the
five
PCB
homologs
included
in
the
table
are
considered
in
calculating
level
oE
chlorination.)

7­
24
number
of
standard
errors
in
the
confidence
interval
is
determined
by
the
Student­
t
distribution.
The
17
target
compounds
for
statistical
analysis
included
five
PCB
homologs
(
tetra­
through
octa­
chlorobiphenyl).
Using
the
average
estimates
for
these
five
homologs,
estimates
of
total
PCBs
and
level
of
chlorination
were
calculated
based
on
the
approach
documented
in
Section
6.2.1.2.
The
estimates
.
ofthese
two
PCB
parameters
are
also
included
in
Tables
7­
2
through
7­
6.

In
addition,
the
chlorobiphenyl
distribution
across
the
five
PCB
homologs,
corresponding
to
the
percentage
of
the
total
PCB
concentration
represented
within
each
homolog,
is
presented
in
Table
7­
7.
This
table
illustrates
that
the
penta­,
hexa­,
and
hepta­
chlorobiphenyls
represent
over
80%
of
the
national
average
PCB
concentration
across
the
five
homologs,
with
hexachlorobiphenyl
representing
47%
of
the
total.
As
will
be
seen
in
Chapter
8,
similar
distributions
were
observed
in
previous
"
ATS
campaigns.
Appendix
F
contains
plots
of
the
estimated
average
concentrations
and
their
associated
95%
confidence
intervals
for
the
17
target
compounds,
as
documented
in
Tables
7­
2
through
7­
6.
One
plot
exists
for
each
compound
and
contains
statistics
for
each
of
the
four
analysis
factors
and
the
entire
nation.
These
plots
illustrate
the
trends
observed
in
the
average
concentrations
across
the
subpopulations
and
the
variability
associated
with
these
trends.
Considerable
overlapping
of
the
confidence
intervals
indicate
that
while
average
concentrations
may
differ
between
subpopulations,
they
may
not
differ
statistically.
The
chlorobiphenyl
distributions
presented
in
Table
7­
7
are
also
plotted
in
Appendix
F.
Estimates
of
the
average
concentrations
in
the
population
categories
defined
by
the
four
demographic
factors
are
presented
in
Tables
7­
2
through
7­
6
even
if
the
effects
of
those
factors
were
not
found
to
be
statistically
significant.
through
hypothesis
testing.
For
example,
regional
estimates
of
average
~

concentration
for
Beta­
BHC
range
from
151
ng/
g
in
the
North
7­
25
I
,

i
Table
7­
7.
Chlorobiphenyl
Distribution
Across
the
Five
Target
PCB
Homologs
in
the
FY86
"
ATS
1
I
15.6%
North
Central
10.2%
25.5%
43.4%
5.2%

North
East
7.9%
20.2%
43.1%
21.3%
7.5%

South
Note:
Homologs
not
represented
in
this
table
were
detected
in
no
more
than
30%
of
the
NHATS
M86
composite
samples.
The
omitted
homologs
were
not
included
in
calculating
total
PCBs,
and
thus
the
percentages
in
a
given
row
add
to
100%.

7­
26
Central
census
region
to
177
ng/
g
in
thc
South
census
region.
However,
as
further
documented
in
Section
7.3,
this
difference
was
not
found'tobe
statistically
Significant.
Table
7­
6
indicates
that
the
standard
errors
of
the
national
estimates
among
the
17
semivolatiles
ranged
from
5.9
to
41.3
percent
of
the
estimates.
The
highest
relative
standard
error
was
observed
with
1­
Nonene,
which
is
a
qualitative
semivolatile
compound.
Among
the
four
analysis
factors,
higher
relative
standard
errors
were
generally
noted
among
subfactors
associated
with
fewer
composites,
such
as
the
West
census
region,
the
0­
14
year
age
group,
and
the
non­
Caucasian
race
group.
The
estimated
concentrations
for
most
of
the
17
semivolatile
compounds
appear
to
increase
with
age
group
according
to
Table
7­
3.
This
result
has
been
observed
in
data
analyses
on
other
"
ATS
datasets
(
e.
g.,
FY82
and
FY87).
Similar
trends
consistent
across
the
analyzed
compounds
are
not
as
apparent
among
census
regions,
race
groups,
and
sex
groups.
Statistical
conclusions
on
these
effects
are
based
on
the
hypothesis
tests
in
the
next
section.

7.3
HYPOTHESZS
TESTING
Statistical
hypothesis
tests
were
conducted
for
each
of
the
17
semivolatile
compounds
included
in
the
statistical
analysis
to
determine
if
there
are
statistically
significant
differences
in
average
concentrations
between
individuals
from
different
geographic
regions,
age
groups,
race
groups,
and
sex
groups.
The
tests
were
based
on
likelihood
ratio
tests
using
the
additive
model
analysis
and
were
described
in
Section
6.2.2.

Table
7­
8
lists
the
attained
significance
levels
for
the
tests
associated
with
the
four
analysis
factors.
In
addition,
a
test
was
performed
to
note
significance
of
the
effect
that
being
in
Batches
4
and
5
has
on
the
measured
concentration;
this
factor
was
significant
among
the
QC
sample
data.
The
attained
signifi.
cancelevel
is
t.
hesmallest
level
at
which
the
test
can
result
in
rejection
of
the
hypothesis
th
.
atno
differences
are
7­
27
1
Table
7­
8.
Significance
Levels
from
Hypothesis
Tests
for
Differences
Between
Demo
raphic
Groups
for
"
ATS
FY86
Sdvolatiles8)

1­
Nonene
0.702
0.751
0.764
0.695
Hexyl
acetate
0.301
0.826
0.672
0.445
Data
adjusted
for
surrogate
recoveries
(
see
Section
5.2).
(
2)
Likelihood
ratio
tests
based
on
the
x2
distribution.
(
3)
Likelihood
ratio
tests
based
on
the
x4'
distribution.
t4)
Likelihood
ratio
tests
based
on
the
x$
distribution.
(
5)
p,
p­
DDE
concentrations
use
the
following
response
ion:
m/
z=
316.
(
6)
Data
results
from
Batch
1
not
included.
Corrected
(
see
Section
5.1.21
!

*
Significant
at
the
0.05
level.
*.*
Significant
at
the
0.01
level.

7­
28
I
present
between
the
population
averages.
For
example,
the
differences
among
estimated
averages
of
Beta­
BHC
in
the
four
census
regions
could
only
be
considered
significant
at
the
0.947
(
94.7%)
level
of
significance,
while
the
differences
in
age
group
average
is
significant
at
the
0.015
(
1.5%)
level.
A
significance
level
of
less
than
0.05
(
5%)
is
generally
required
to
declare
statistical
significance.

An
apparent
conclusion
from
Table
7­
8
is
the
presence
of
significantly
different
estimated
average
concentrations
among
the
age
groups
for
pesticides,
hexachlorobenzene,
and
PCBs.
From
Table
7­
3,
the
older
age
group
(
45+
years)
had
the
highest
estimated
average
concentration
for
these
compounds,
and
the
youngest
age
group
(
0­
14
years)
had
the
lowest
estimate.
The
disparity
between
the
older
age
group
and
the
others
is
more
apparent
for
the
PCBs.
Statistical
significance
was
also
observed
among
census
regions
for
three
pesticides,
hexachlorobenzene,
naphthalene,
and
three
PCB
congeners.
Levels
of
p,
p­
DDT
and
hexachlorobenzene
were
highest
in
the
West
census
region,
while
for
some
PCB
congeners,
levels
were
lowest
in
the
West
census
region.
However,
a
consistent
trend
across
the
compounds
was
not
observed
with
census
region
as
was
observed
with
age
groups.
The
difference
in
estimated
average
concentration
between
Caucasian
and
non­
Caucasian
and
between
male
and
female
donors
were
not
statistically
significant
for
any
of
the
modelled
compounds.
The
effect
of
Batches
1­
3
versus
4­
5
on
the
measured
concentrations
in
composite
samples
was
also
not
significant
for
any
of
the
compounds.

7.4
OUTLIER
DETECTION
Prior
to
conducting
the
statistical
analysis
of
the
FY86
"
ATS
data,
outlier
detection
procedures
were
performed
to
identify
possible
data
entry
errors
and
errors
associated
with
the
analytical
method.
The
outlier
detection
process
was
performed
in
multiple
stages
by
Westat,
Battelle,
and
EPA.
MRI
reviewed
all
findings
of
this
process,
identified
a
list
of
changes
to
data
values
resulting
from
their
review,
and
notified
the
"
ATS
project
team
of
these
changes.
Battelle
corrected
the
database
according
to
MRI's
review
prior
to
performing
the
final
statistical
analysis.
Westat
performed
statistical
outlier
analysis
on
the
following
types
of
data:

I
measured
concentrations
of
native
analytes,

I
internal
quantitation
standard
recoveries,

I
LODs,
and
rn
percent
lipid
values
for
composite
and
QC
samples.

The
methods
and
findings
of
these
analyses
are
presented
in
Rogers
(
1991).
The
procedure
consisted
of
three
approaches:
logic
checks,
formal
outlier
identification
procedures,
and
informal
outlier
identification
procedures.
Logic
checks
were
performed
prior
to
database
completion,
to
identify
obvious
data
inconsistencies
or
coding
errors.
For
example,
by
printing
records
with
inconsistent
entries,
the
logic
check
procedure
would
reveal
records
having
recorded
concentrations
but
a
data
qualifier
of
"
not
detected".
The
formal
approach
to
outlier
identification
in
Rogers
(
1991)
assumed
that
the
concentrations
and
recovery
data
followed
a
lognormal
distribution,
and
the
percent
lipid
data
followed
a
normal
distribution.
A
mathematical
model
was
fit
to
the
data,
and
the
extreme
studentized
deviate
(
ESD)
test
was
applied
to
the
residuals
of
the
model.
This
test
considered
the
ratio
of
the
maximum
residual
to
the
standard
deviation
of
the
residuals.
Outliers
were
identified
if
this
ratio
exceeded
the
appropriate
critical
value
given
the
significance
level
(
1%
or
5%).
The
form
of
the
simple
linear
regression
models
varied.
amongthe
different
types
of
data
(
see
Table
2
in
Rogers
(
1991)).
Once
formal
outlier
identification
procedures
were
completed,
informal
identification
procedures
noted
any
7­
30
additional
data
which
may
be
in
question.
These
procedures
included
normality
tests
on
residuals
for
individual
compounds,
multivariate
tests
across
multiple
compounds
(
identifyingdata
points
which
do
not
conform
with
a
multivariate
normal
distribution),
boxplots
to
compare
measurements
of
different
types,
and
special
outlier
comparison
tests
for
the
LODs.
In
addition
to
the
approach
documented
in
Rogers
(
1991)
to
identify
outliers
among
native
compound
concentrations
in
composite
samples,
Battelle
identified
additional
potential
outliers
by
fitting
the
additive
model
(
Chapter6)
to
the
preliminary
FY86
semivolatile
data.
Residuals
exceeding
two
standard
deviations
from
zero
were
reported.
To
illustrate
patterns
due
to
analysis
order
and
batch,
time
series
plots
of
the
FY86
data
were
produced.
Any
outliers
and
questionable
data
points
were
highlighted
in
these
plots.
These
data
plots
and
listings
of
statistical
outliers
were
delivered
to
EPA
and
to
MRI
for
review.
A
total.
of
50
data
points
were
identified
as
outliers
from
the
procedures
in
Rogers
(
1991).
These
data
points
included
24
quantitative
concentrations,
6
qualitative
concentrations,
and
20
recoveries.
Of
these
points,
eight
were
changed
as
a
result
of
review
by
MRI.
The
findings
of
the
outlier
analysis
identified
unusually
low
surrogate
recoveries
for
two
samples,
implying
that
the
reported
concentrations
were
suspect
for
these
samples.
The
outlier
report
also
noted
that
recoveries
in
Batch
1
were
lower
than
in
later
batches,
apparently
due
to
changes
in
lab
procedures.
These
findings
supported
the
need
to
consider
effects
of
batch
in
statistical
analyses
and
to
correct
data
for
surrogate
recoveries.
Forty­
four
additional
data
points
were
identified
as
potential
statistical
outliers
as
a
result
of
fitting
the
additive
model
to
target
compound
data.
Review
of
these
data
by
MRI
resulted
in
changes
to
16
of
the
data
points.
Battelle
made
all
data
corrections
to
the
master
database
before
proceeding
with
the
statistical
analysis.

7­
31
(­
JC?
i;
1333
However,
as
a
result
of
the
data
review,
some
of
the
data
points
identified
in
the
outlier
detection
procedure
either
did
not
require
modification
or
remained
influential
after
modification.
Thus
these
data
points
contributed
to
increased
error
in
fitting
the
additive
model
and
to
inflated
variability
in
parameter
estimation
and
hypothesis
testing.
The
most
influential
data
points
are
documented
in
the
following
section.

7.5
MODEL
VALIDATION
As
part
of
the
commitment
to
overall
data
quality,
three
types
of
analyses
were
performed
to
evaluate
the
adequacy
of
the
additive
model
for
use
on
the
FY86
"
ATS
semivolatile
data
on
the
seventeen
target
compounds.
All
three
analyses
were
based
on
comparisons
of
the
observed
(
i.
e.,
measured)
and
predicted
concentrations
for
the
composite
samples.
Predicted
concentrations
were
calculated
using
the
IWGLS
method
applied
to
the
additive
model
(
Chapter
6).
Residuals,
which
were
also
used
in
the
model
validation
analysis,
were
calculated
by
taking
the
differences
between
the
observed
and
predicted
concentrations.
Model
validation
analyses
included:
a
residual
plots,

i
I
noma1
probability
plots,
and
a
R­
squared
analysis.
The
use
of
Shapiro­
Wilk
tests
for
normality
was
also
considered.
However,
in
this
application,
the
Shapiro­
Wilk
test
was
not
appropriate
because
the
data
were
correlated
and
variances
increased
with
increasing
concentrations.
In
several
of
the
target
compounds,
the
residual
plots
(
residuals
versus
predicted
concentration)
confirmed
the
model
assumption
that
the
variance
of
the
measured
concentrations
increases
with
the
average
concentration.
In
addition,
these
plots
showed
that
the
distribution
of
residuals
tended
to
be
symmetric
about
zero
across
all
predicted
concentrations.
For
some
compounds,
the
extent
to
which
residuals
were
symmetric
about
zero
was
less
evident
at
low
concentrations,
where
predicted
levels
tended
to
be
larger
than
the
observed
level.
This
finding
indicates
that
the
relationship
between
measured
concentration
and
the
model
predictors
may
not
be
as
linear
in
low
concentration
ranges
relative
to
larger
concentration
ranges.

Also,
the
low
concentration
range
can
include
a
substantial
number
of
measured
concentrations
at
or
below
the
detection
limit.
For
compounds
whose
non­
detect
percentage
approached
50%

(
such
as
octachlorobiphenyl,
l­
nonene,
dieldrin,
and
tetrachlorobiphenyl),
the
predicted
concentrations
in
areas
close
to
the
detection
limit
may
be
more
biased
in
portraying
the
true
concentration
than
predicted
concentrations
in
higher
detectable
ranges.
The
presence
of
unusually
high
or
low
data
points
also
contributed
to
an
overall
lack
of
fit
of
the
model
to
the
observed
data.
The
data
points
observed
to
be
among
the
most
"
influential1Ito
the
model
fitting
are
presented
in
Table
7­
9.
The
result
of
fitting
the
model
while
including
the
influential
data
points
is
either
an
underestimate
or
overestimate
by
the
fitted
model
in
certain
concentration
ranges.
Normal
probability
plots
for
most
target
compounds
resembled
a
linear
pattern,
supporting
the
normality
assumption
for
the
errors.
However,
the
linearity
assumption
for
some
compounds
did
not
hold
in
areas
of
extremely
large
or
small
concentrations.
This
is
explained
by
the
larger
variances
associated
with
these
concentrations,
and
by
the
presence
of
influential
data
points
with
large
positive
or
negative
residuals.
Table
7­
10
lists
the
R­
squared
correlations
between
the
observed
and
predicted
concentrations
calculated
for
each
target
compound.
R­
squared
can
be
interpreted
as
the
percent
of
the
total
variability
in
the
observed
concentrations
that
can
be
explained
by
the
additive
model.
The
correlations
range
from
12%
(
naphthalene)
to
65%
(
tetrachlorobiphenyl).
The
qualitative
compounds
have
low
R­
squared
values,
indicating
that
their
categorical
concentrations
are
not
highly
correlated
with
7­
33
e".
W<'$?>
S
Table
7­
9.
Measured
Concentrations
with
High
Influence
on
Detedning
the
Additive
Model
Fit
ACS8600270
1
ACS8600065
I
I
ACS8600163
1
ACS8600163
ACS8600314
ACS8600207
ACS8600350
ACS8600332
ACS8600225
ACS8600421
ACS8600127
ACS8600092
ACS8600092
I
ACS8600289
ACS8600289
.
p8p­
DDT
17922
1
1214
.
OxychIor'dane@)

17942
I
'
39.2
.
I
17946
I
306
Trans­
nonachlor
17946
510
Hexachlorobenzene
17986
123
17968
176
17948
192
Naphthalene
17965
66.9
17939
99.0
17924
70.5
Tetrachlorobiphenyl
17959
249
17909
217
Hexachlorobiphenyl
17909
I
1123
Heptachlorobiphenyl
17919
888
Octachlorobiphenyl
17919
322
7­
34
I
.
886
1
148.7
I
I
139
I
264
67.5
81.1
96.5
23.5
24­
6
I
26.3
1
124
146
I
493
I
376
1
142
.
Table
7­
9,
(
cont,)

1­
Nonene
ACS8600181
17960
72
8
I
261
Hexyl
acetate
ACS8600458
3.7938
459
45.8
ACS8600252
17926
369
74.2
ACS8600430
17921
729
233
p,
p­
DDE~)

ACS8600163
17946
10716
4062
ACS8600341
17958
11859
5599
Dieldrin(*)

ACS8600458
I
17925
I
212
I
58.8
ACS8600225­
p
17939
1
278
I
194
Data
adjusted
for
surrogate
recoveries
(
see
Section
5.2).
(
2)
Batch
1
results
not
included
in
statistical
analysis.
p,
p­
DDE
concentrations
use
the
following
response
ion:
m/
z=
316.
(
4)
Corrected
(
see
Section
5.1.2).

7­
35
Table
7­
10.
R­
Squared
Correlation
Between
Observed
Concentrations
and
Concentrations
Predicted
by
the
Additive
Model
for
"
ATS
FY86
Semivolatiledl)

Pesticides
PIP­
DDT
Beta­
BHC
Heptachlor
epoxide
Oxychlordane
Trans­
nonachlor
Dieldrin
8­­­
Chlorobenzmea
1,4­
Dichlorobenzene
Hexachlorobenzene
I
PAIls
INaphthalene
Hexachlorobiphenyl
Heptachlorobiphenyl
Octachlorobiphenyl
1­
Nonene
I
Hexyl
acetate
,
1
31
43
55
43
55
13
29
46
12
61
47
37
23
14
(
I)
R­
squared
is
the
square
of
the
Pearson
correlation
coefficient.
It
*

represents
the
percent
of
variability
in
the
data
that
is
explained
by
the
additive
model.
Data
adjusted
for
surrogate
recoveries
(
see
Section
5.2).

7­
36
predicted
values.
Note
that
these
R­
squared
values
are
not
as
high
as
seen
with
dioxins
and
furans
in
the
FY87
"
ATS
(
USEPq,
1991).
This
does
not
necessarily
imply,
however,
that
the
additive
model
is
an
inadequate
fit
to
the
semivolatile
compound
data.
Instead,
low
R­
squared
values
may
indicate
that
the
estimated
model
effects
are
small
relative
to
the
random
error
observed
in
the
measured
concentrations.
The
random
error
is
increased
by
the
presence
of
influential
observations
such
as
those
in
Table
7­
9.

7­
37
I
,

8.0
COMPARISON
WITH
RESULTS
FROM
PREVIOUS
SURVEYS
IN
THl3
NHATS
PROGRAM
The
FY86
"
ATS
is
one
of
three
surveys
in
the
"
ATS
program
to
use
HRGC/
MS
analytical
methods
in
measuring
the
prevalence
and
levels
of
semivolatile
organic
compounds
in
composited
adipose
tissue
samples.
Prior
to
the
FY86
survey,
the
FY82
and
FY84
surveys
also
performed
analysis
of
semivolatiles
on
composite
samples
using
HRGC/
MS
methods.
The
"
ATS
FY86
sampling
and
data
analysis
approach
was
designed
to
allow
valid
statistical
comparisons
to
be
made
between
the
FY86
results
and
the
results
from
these
two
surveys.
The
NHATS
FY82
Broadscan
Analysis
Study
(
Mack
and
Panebianco,
1986)
was
the
first
"
ATS
campaign
to
employ
the
HRGC/
MS
method
in
characterizing
an
expanded
chemicals
list.
The
objective
of
the
FY82
"
ATS
was
to
identify
and
characterize
additional
compounds
that
persist
in
human
adipose
tissue
but
could
not
be
measured
with
less
selective
analytical
techniques.
The
FY84
NHATS
was
designed
to
establish
the
comparability
of
the
HRGC/
MS
and
PGC/
ECD
analytical
methods
(
Westat,
1990).
The
FY84
"
ATS
revealed
that
issues
in
method
comparability
were
not
totally
resolved
for
many
of
the
target
semivolatile
compounds.
This
chapter
presents
comparison
of
the
FY86
"
ATS
results
with
the
results
from
the
NHATS
FY82
and
FY84
semivolatile
analyses.
There
are
several
differences
in
the
designs
and
analytical
procedures
used
in
these
three
surveys.
These
differences
are
documented
in
Section
8.1.
Only
the
semivolatile
compounds
analyzed
in
the
FY86
NHATS
and
in
at
least
one
of
the
FY82
and
FY84
NHATS
are
included
in
comparisons.
For
each
of
these
compounds
within
each
survey,
Section
8.2
presents
average
limits
of
detection
(
LODs)
and
the
percentages
of
detected
results
among
the
samples.
Statistical
procedures
were
used
to
compare
these
detection
percentages
across
surveys.
Section
8.3
presents
two
approaches
to
calculating
descriptive
statistics
in
summarizing
measured
concentration
data
within
each
of
the
three
surveys
at
the
national
level.
Finally,
statistical
comparisons
were
performed
on
only
"
hose
compounds
detected
in
at
least
50%

of
the
composite
samples
within
each
survey.
Section
8.4
presents
results
of
fitting
the
additive
model
to
these
compounds
within
each
survey.

8.1
COMPARISON
OF
DESIGN
AND
ANALYTICAL
PROCEDURES
8.1.1
CornParison
of
Studv
Desicms
Similar
sampling
designs
were
used
for
collecting
tissue
specimens
in
the
FY82,
FY84,
and
FY86
NHATS.
A
discussion
of
the
FY86
sampling
design
is
found
in
Chapter
2
of
this
report.
The
primary
difference
in
sampling
designs
between
these
three
'
surveys
is
the
method
of
stratification.
Prior
to
the
FY8S
NHATS,
MSAs
were
selected
from
strata
defined
by
the
nine
U.
S.
Census
divisions.
Beginning
with
the
FY85
"
HATS,
sampling
strata
were
redefined
to
be
the
seventeen
geographic
areas
that
resulted
from
the
intersection
of
the
Census
divisions
and
the
ten
EPA
regions
(
Table
2­
2).

A
controlled
selection
technique
(
Mack
et.
al.,
1984)
was
used
to
maximize
the
probability
of
retaining
MSAs
from
one
survey
design
to
another.
Table
8­
1displays
the
number
of
specimens
and
composites
associated
with
each
MSA
for
each
survey.
Except
for
double­
collection
MSAs,
no
MSA
contributed
more
than
the
quota
of
27
specimens
to
the
FY86
NHATS
design.
This
was
not
true
for
the
FY82
and
FY84
surveys,
where
as
many
as
72
specimens
originated
from
a
single­
collection
MSA.
Only
five
MSAs
sampled
in
the
FY82
and
FY84
NHATS
were
not
represented
in
the
FY86
NHATS,
while
only
four
MSAs
were
sampled
in
the
FY86
NHATS
but
not
in
the
other
two
surveys.
It
is
expected
that
differences
in
MSA
sampling
across
the
three
surveys
contribute
to
only
minor
differences
in
concentration
estimates.
For
each
census
region,
age
group,
sex
group,
and
race
group,
Tables
8­
2
and
8­
3
present
summaries
of
the
number
of
specimens
and
composites,
respectively,
originating
within
these
8­
2
Q6­
1
c';
131
i
1
Table
8­
1.
Number
of
Spechen8
and
Composites
Within
the
FY82,
FY84*
and
FY86
"
ATS
According
to
MSA
5200
Number
of
Number
of
Specimens
Composites(')

MSA
(
code
and
location)
FY82
FY84
FY86
FY82
FY84
FY86
800
AKRON,
OH
0
6
18
0
1
5
ATLANTA,
GA
0
0
27
0
0
8
10000
BIRMINGHAM,
AL
40
27
0
5
4
0
11200
BOSTON,
MA
0
0
25
0
0
4
16000
CHICAGO,
IL
17
37
45
8
6
6
16800
CLEVELAND,
OH
44
40
27
8
6
­
6
18400
COLUMBUS,
OH
0
0
14
0
0
3
19200
DALLAS­
FORTWORTH,
TX
38
26
27
4
4
3
19600
DAVENPORT­
ROCK
ISLAND­
MOLINE,
IA­
IL
12
9
0
5
2
0
20000
DAYTON,
OH
24
24
9
7
6
3
20800
DENVER­
BOULDER,
CO
10
10
10
2
3
3
21600
DETROIT,
MI
9
15
54
3
2,4
23350
ELMIRA,
NY
0
17
27
0
4
5
31600
GREEMIIUE­
SPARTANBURG,
SC
14
39
27
9
10
7
42800
LEXINGTON­
FAYETTE,
KY
45
38
27
5
4
4
44800
LOS
ANGELES­
LONG
BEACH,
CA
0
8
4
0
2
2
46000
LUBBOCK,
TX
35
12
0
4
4
0
47200
MpiDISON,
WI
40
29
27
8
6
4
49200
MEMPHIS,
TN­
AR­
MS
0
0
23
0
0
4
50000
MIAMI,
FL
26
16
27
9
8
8
56000
NEW
YORK,
NY­
NJ
76
0
25
6
0
5
57200
NORFOLK­
VABEACH­
PORTSMOUTH,
VA­
NC
72
43
27
10
9
8
59200
OMAHA,
NE­
IA
19
60
27
4
5
5
59600
ORLANDO,
FL
43
33
0
9
8
0
61600
PHILADELPHIA,
PA­
NJ
5
7
7
2
1
4
62800
PITTSBURGH,
PA
28
25
21
4
4
4
64400
PORTLAND,
OR­
WA
27
15
16
3
4
3
68200
ROCHESTER,
MN
41
29
27
4
4
5
69200
SACRAMENTO,
CA
4
0
2
1
0
2
71600
SALT
LAKE
CITY­
OGDEN,
UT
19
22
24
3
3
4
72400
SAN
ANTONIO,
TX
0
27
0
0
4
0
73600
SAN
mcIsco­
oAIcLAM),
CA
0
0
27
0
0
5
78400
SPOKANE,
WA
0
15
­
12
0
3
3
80000
SPRINGFIELD­
CHICOPEE­
HOLYOICG,
MA­
CT
56
37
18
3
4
4
82800
TAMPA­
STPETERSBURG,
FL
0
7
8
0
4
3
88400
WASHINGTON,
DC­
MD­
VA
19
16
12
8
6
5
­
Totals:
763
689
671
46
46
50
('
1
Column
entries
indicate
the
number
of
composites
having
at
least
one
specimen
from
the
given
MSA.
The
total
at
the
bottom
of
each
column
indicates
the
total
number
of
analyzed
composites
in
the
survey.
Since
specimens
within
a
composite
can
originate
from
more
than
one
MSA,
this
total
is
not
equal
to
the
sum
of
the
column
entries.

8­
3
Table­
8­
2.
Total
Number
of
Specheas
Included
in
Compusite
Samples
Analyzed
ip
the
FY82,
FY84,
and
FY86
NEATS,
by
Subpopulation
and
Across
the
­
tire
Study
Subpopulation
Northeast
North
Central
South
West
0­
14
years
15­
44
years
45+
years
a
Male
Female
White
Nonwhite
Total
#
of
Specimens
Number
of
Specimens
(%

FY82
FY84
Census
Region
166
(
22%)
86
(
12%)
206
(
27%)
249
(
36%)
331
(
43%)
284
(
41%)
60
(
8%)
70
(
10%)

Age
Group
178
(
23%)
142
(
21%)
312
(
41%)
2'
66
(
39%)
273
(
36%)
281
(
41%)

412
(
54%)
352
(
51%)
351
(
46%)
337
(
49%)

Race
632
(
83%)
579
{
84%)
131
(
17%)
110
(
16%)

763
689
of
Total)
1980
Census
FY86
123
(
18%)
248
(
37%)
205
(
31%)
95
(
14%)

108
(
16%)
221
(
33%)
342
(
51%)

315
(
47%)
356
{
53%)

526
(
785;)
145
(
22%)

671
%

26%
22%
33%
19%

23%
46%
31%

49%
51%

83%
17%
Table
8­
3.
Total
Number
of
Composite
Samples
Analyzed
in
the
FY82,
FY84,
and
FY86
"
ATS,
by
Subpopulation
and
Across
the
Entire
Survey
Subpopulation
Northeast
North
Central
South
West
0­
14
years
15­
44
years
45+
years
Mixedt3)
Male
only
Female
only
Mixed(
3)
White
only
Nonwhite
only
Number
of
Composites
(%

FY82
FY84
Census
Region(')

9
(
20%)
8
(
17%)
12
(
26%)
13
(
28%)
19
(
41%)
18
(
39%)
6
(
13%)
7
(
15%)

Age
Group(')

12
(
26%)
10
(
22%)
17
(
37%)
19
(
41%)
17
(
37%)
17
(
37%)

.35
29
6
(
55%)
8
(
47%)
5
(
45%)
9
(
53%)

Race@)

29
25
11
(
65%)
16
(
76%)
.
6
(
35%)
5
(
24%)

+

Total
#
of
Composites
46
46
of
Total)

FY86
9
(
18%)
16
(
32%)
15
(
30%)
10
(
20%)

10
(
20%)
16
(
32%)
24
(
48%)

18
14
(
44%)
18
(
56%)

29
16
(
76%)
5
(
24%)

50
1980
Census
%

26%
22%
33%
19%

23%
46%
31%

49%
51%

83%
17%

(
I)
All
specimens
within
a
given
composite
originated
from
the
same
census
region
and
age
group.

(*)
The
percentages
for
sex
and
race
groups
are
calculated
as
the
total
number
of
pure
composites
within
each
study
design.
For
example,
6
of
the
11
(
55%)
pure
sex
composites
in
the
FY82
study
design
were
composed
of
specimens
from
,

males
only.

0)
Composites
containing
specimens
from
both
sex
(
or
race)
groups.

8­
5
,
groups.
The
distributions
of
specimens
among
the
geographic
and
demographic
groups
were
relativ.
elysimilar
across
the
three
surveys.
The
FY86
survey
had
higher
percentages
of
specimens
from
the.
Westcensus
region
and
the
nonwhite
race
group:
two
groups
in
which
specimens
are
generally
less
procurable
than
other
groups.
The
FY82,
FY84,
and
FY86
NHATS
also
had
comparable
composite
designs
(
Table
8­
3).
One
of
the
design
criteria
for
compositing
FY84
and
FY86
specimens
was
to
maintain
similarity
to
the
FY82
design
(
see
Section
3.1).
However,
the
FY86
design
stipulated
more
pure
sex
composites
(
i.
e.,
all
male
or
all
female)
than
the
FY82
and
FY84
designs
in
order
to
more
accurately
estimate
differences
in
concentrations
among
the
sexes.
Sixty­
fourpercent
of
the
FY86
composites
were
pure
sex
composites,
compared
to
less
than
forty
percent
of
the
composites
in
the
FY82
and
FY84
surveys.
Overall,
the
percentages
of
composites
within
each
population
group
were
similar
across
the
three
surveys
and
with
the
1980
Census
percentages.

8.1.2
Comnarison
of
Analytical
Procedures
To
interpret
differences
in
estimated
concentrations
between
the
three
surveys,
it
is
necessary
to
consider
differences
in
their
analytical
methods.
While
some
major
differences
do
exist,
the
methods
were
otherwise
similar
between
the
three
surveys.
One
analytical
factor
having
a
large
potential
effect
on
data
comparability
between
the
three
surveys
is
the
type
and
number
of
internal
quantitation
standards
(
IQS)
and
how
these
standards
are
assigned
to
semivolatile
compounds.
Native
compound
concentrations
were
quantified
relative
to
the
IQS
findings.
Only
one
IQS
was
used
to
quantify
the
semivolatiles
in
FY82:
anthracene­
dlo.
The
FY84
and
FY86
surveys
included
three
IQS
for
quantification
of
semivolatiles:
anthracene­
dlo,
benzo(
a)
anthracene­
d12,
and
naphthalene­
d8.
In
addition
to
differences
caused
by
the
number
and
type
of
IQS
assigned
to
each
survey,
the
method
of
assigning
an
IQS
to
each
semivolatile
differed
between
the
FY84
and
FY86
NHATS.
Table
8­
4
lists
those
semivolatiles
analyzed
in
both
FY84
and
FY86
for
which
the
same
IQS'was
assigned
in
both
surveys.
Table
8­
5
lists
the
semivolatiles
with
differing
IQS
between
FY84
and
FY86.

Differing
IQS
assignments
between
surveys
must
be
considered
when
interpreting
differences
observed
in
results
from
one
survey
to
another.
Average
concentration
estimates
in
the
FY86
"
HATS
were
based
on
measured
concentrations
adjusted
for
surrogate
recoveries
(
Chapter
7).
The
adjusted
concentrations
are
more
likely
to
resemble
actual
concentrations
in
the
sample
than
unadjusted
measured
concentrations.
Thus
for
comparison
purposes,
it
was
necessary
to
obtain
average
concentration
estimates
in
the
FY82
and
FY84
surveys
based
on
surrogate­
adjusted
concentrations.
Like
the
IQS,
surrogate
compounds
were
matched
to
specific
semivolatile
compounds
within
each
survey
(
Table
5­
2)
for
adjustment
purposes.
However,
the
types
of
surrogate
compounds
included
in
each
survey
also
differed.
Thus
in
conducting
the
comparison,
it
is
noted
when
surrogate
compounds
differed
among
the
surveys.
Another
issue
to
consider
is
that
the
FY82
and
FY86
analyses
were
conducted
at
Midwest
Research
Institute,
while
the
FY84
analysis
was
performed
at
Colorado
State
University.
Thus
interlaboratory
variation
is
also
introduced
when
comparing
FY84
results
with
the
other
two
surveys.
Other
than
the
differences
noted
above,
the
techniques
in
the
analytical
methods
for
semivolatile
analyses
were
essentially
equivalent
between
the,
three
surveys.
The
flow
diagram
in
Figure
4­
1
(
Chapter
4)
illustrat.
esthe
order
of
activities
in
each
campaign.
Each
procedure
required
fortification
with
IQS
and
surrogate
compounds,
extraction,
\

removal
of
bulk
lipid,
separation,
cleanup,
and
quantification.
Extraction
was
achieved
with
methylene
chloride
using
a
Tekmar
a­
7
C,
fVfEf36
Table
8­
4.
Semivolatile
Compounds
Quantitated
Using
the
Same
Internal
Quantitation
Standards
(
IQS)
in
NEATS
FY84
and
FY86.

IQS:
Benzo
(
a)
anthracene­
du
PIP­
DDT
O,
p­
DDT
PIP­
DDE
O,
p­
DDD
TRANS­
NONACHLOR
MIREX
CHRYSENE
HEXACHLOROBIPHENYL
HEPTACHLOROBIPHENYL
OCTACHLOROBIPHENYL'
NONACHLOROBIPHENYL
DECACHLOROBIPHENYL
IQS:
Anthracene­
dlo
ALPHA­
BHC
BETA­
BHC
DELTA­
BHC
GAMMA­
BHC(
L1NDANE)
ALDRIN
HEPTACHLOR
HEPTACHLOR
EPOXIDE
OXYCHLORDANE
GAMMA­
CHLORDANE
PENTACHLOROBENZENE
HEXACHLOROBENZENE
ACENAPHTHALENE
FLUORENE
PHENANTHRENE
FLUORANTHENE
MONOCHLOROBIPHENYL
DICHLOROBIPHENYL
TRICHLOROBIPHENYL
TETRACHLOROBIPHENYL
IQS:
Nanhthalene­
4
1,2,3­
TRICHLOROBENZENE
1,2,4­
TRICHLOROBENZENE
1,3,5­
TRICHLOROBENZENE
NAPHTHALENE
Note:
Anthracene­
d10was
the
only
IQS
used
in
the
FY82
"
ATS.

8­
8
Table
8­
5,
Semivolatile
Compounds
Quantitated
Using
Different
Internal
Quantitation
Standards
(
IQS)
in
"
ATS
­
84
and
FY86,

o,
p­
DDE
1,2,3,4­
TETRACHLOROBENZENE
l,
2,3,5­
TETRACHLOROBENZENE
1,2,4,5­
TETRACHLOROBENZENE
I
ACENAPHTHENE
PENTACHLOROBIPHENYL
(')
Legend:
A
=
Anthracene­
dli
B
=
Benzo(
a)
anthracene­
d12
N
=
Naphthalene­
de
B
A
N
A
N
A
N
A
N
A
I
B
I
A
Note:
Anthracene­
dlowas
the
only
IQS
used
in
the
FY82
"
ATS.

8­
9
Tissuemizer
to
promote
thorough
extraction
of
lipids.
Extracts
were
filtered
through
anhydrous
sodium
sulfate.
Gel
permeation
chromatography
was
applied
to
separate
target
analytes
from
lipid
material.
Interference
separation
was
achieved
through
Florisil
column
fraction
procedures.

8.2.
LODa
AND
PERCENT
DETECTION
SI7MMARIES
A
total
of
54
quantitative
semivolatile
compounds
were
analyzed
in
the
FY86
"
ATS
and
also
analyzed
in
one
or
both
of
the
FY82
and
FY84
NHATS.
These
compounds
form
the
basis
of
the
,
descriptiveand
statistical
comparisons
in
measured
concentrations
of
target
compounds
across
the
three
surveys.
This
subsection
summarizes
the
LODs
and
the
percentages
of
detected
results
for
these
compounds
in
the
FY82,
FY84,
and
FY86
NHATS.
An
LOD
was
reported
for
a
compound
whenever
a
trace
or
not­
detected
reading
was
reported
for
the
sample.
These
LODs
(
ng/
g
lipid
weight)
are
averaged
and
presented
in
Table
8­
6
for
the
54
semivolatile
compounds.
The
LODs
were
not
adjusted
for
surrogate
recoveries
prior
to
averaging.
Table
8­
6
also
documents
the
percent
of
composite
samples
with
detected
readings
within
each
survey
for
the
54
compounds.
Only
compounds
with
at
least
50%
detected
readings
within
each
of
the
three
surveys
were
considered
for
further
statistical
comparisons.
For
most
compounds,
the
percentage
of
samples
with
Low
detected
results
was
consistent
across
the
surveys.
detection
percentages
were
reported
for
most
chlorobenzenes
(
with
the
exception
of
hexachlorobenzene),
phosphate
triesters,
and
PAHs,
while
some
pesticides
(
such
as
p,
p­
DDEand
beta­
BHC)
had
very
high
detection
percentages.
To
identify
those
compounds
in
Table
8­
6
where
significant
differences
were
present
(
at
the
0.05
level)
in
the
percent
detected
value
between
the
three
surveys,
a
chi­
square
test
for
homogeneity
was
used.
Among
pesticides,
significant
differences
in
the
percent
detected
value
were
present
for
p,
p­
n­
n­
hn­
nnh­
nn­
nn­
nn
Et
6
A
A­­­­­­~
C.
&&&&&&&&&&&
C&&&
C&&&
cv&&
C&*&&&
U
0000000000000000000
...................
000000000
.........
a"
wooooocu*
ooN(
Yo(
YQ)
qlocuo
000000004)
0
m
d
o
m
du,
Q\*
m
m
YVYYYVWWV
m
Y
Y
YYWYWYYYYYYYYYYYYY
Eu,
!
QfomcumQ)
......
u,
u)
0
rl
In
0
0'

&!
3
*
cu(
Ymmrld
d
ACIA~
CI~~
A
u
&
C.
LDC&&
C*
omooooou,
........
a"
aloInwooar(
YooQ)
oqlooooocu
&
am
Q)
m
Q)

ql
Y
YYYVVVYV
..

E
orlooooocu
........
mmmmmmmtn
rlcu
rlrlrlrlrlr(
rll4
h
n
n
h
&
&
&

a"
U
0
0
rl
0
0
&
rl
a.
tn
(

YE
Y
Y
Y
Y
Y
YY
*
cu
0
00
0
rl
m
v
tn
u,
0
m
rl
m
d
(
v
t.!

a
4
B
0
U
8­
11
ooooooq
a
a­
h­
h
&&&&&
U
oooqq
...
a"
OOW(
YOD
4~
1­
r
&
Y
W
E
!
i
9
P
hhAh
&&&&
0000
....
o*
OH
YYYY
PO
....
moO\
OD
dmmw
Ad
Ah
AhA
~
A~
AA
&&&&&
&&&&
JJ
o*
omo
om00
.....
....
a"
OCIOdO
OdOQ
OW
50
&
rl
*
Y
YYYYY
YVYY
E
9
9
P
haAA­
Ah­­­­­­
a
&&&&&&&
00000000
........
......
*
m(
Ywoooo
OD
YYYYYYYY
VYYYYYY
Q)
OCsWO
m
.......

AACI~
A­~
A
CSCI~
A­

A
&&&&&&&
cnr(~
moo(~~
0orlmalOD­
r
.......
........
md(
Y*
oo(
Y*
oocnd*
c­
w
(
Ye4
m*(
loDcna
YYYYYYYY
hh
h
Ahah­
A
­
ah
&&
&
&&&&&&&&
4
a0
0
tzmrom
JJ
WOO3
...
a
..
......
3
*
om
d*
0
(
v­
rNO(
Y
*
Lo*
*
d
(
vmrem
&
YYY
YYY
YY
Y
YYYYY
cs
Y
E
9
...
r
mmcn5
0mOD
3
drn
w
w
i!
MA&
mx
3
0
9(
E
0
U
w
3
PI
ul
0
E
8­
12
%
N
A
W
(

vE
rl
Qlme
m
...
mCYm
8
mmdc
­
d
u
0
a,

m
a
a,
m
Y
u)

dc
E
E:
E
4
n
.
4J
8
neu
Y
&&&
Qlwa
.
..
Ornu)

dcda
YYY
(
v
Q
E
am
...
mmdc
dc
*
o
d
a,
2
U
G::
a,
a
­
rl
u
U
ar
u
a,
a
8­
13
DDT,
dieldrin,
trans­
nonachlor,
heptachlor
epoxide,
and
mirex,
For
each
of
these
pesticides,
significance
was
primarily
the
result
of
low
detection
percentages
observed
in
the
FY82
survey.
For
p,
p­
DDT,
the
detection
percentage
increased
from
67.6%
in
FY82
to
96%
in
FY86.
This
increase
may
be
partially
explained
by
a
substantial
reduction
in
the
average
LOD
for
p,
p­
DDT
in
the
FY86
survey.
Percent
detection
also
increased
in
FY86
for
mirex,
from
below
15%
in
both
FY82
and
FY84
to
32%
in
FY86,
while
accompanied
by
a
gradual
reduction
in
the
average
LOL)
across
these
surveys.
Percent
detection
of
hexachlorobenzene
increased
across
the
FY82
to
FY86
surveys,
from
79.1%
to
98%.
These
differences
across
surveys
were
statistically
significant,
but
were
not
accompanied
by
corresponding
reductions
in
the
average
LOD.
The
average
percent
detection
declined
in
FY86
to
4%
for
triphenyl
phosphate
from
above
38%
in
the
other
two
surveys;
this
decline
was
statistically
significant.
Naphthalene
was
the
only
PAH
with
a
high
percent
detection
in
Fy86
(
84%),
leading
to
statistically
significant
differences
in
the
percentages
across
surveys.
Significant
differences
in
percent
detection
were
also
observed
for
diethyl
phthalate
(
where
the
average
percentage
dropped
substantially
from
the
FY82
value
of
47.6%),
di­
n­
butyl
phthalate
(
where
the
average
percentage
increased
from
50%
in
FY82
to
100%
in
FY84),
and
bis
(
2­
ethylhexyl)
phthalate
(
0%
in
FY84
to
78%
in
FY86).
However
for
di­
ethyl
phthalate,
the
decreasing
percentages
were
accompanied
by
decreases
in
the
J
average
LOD.
This
indicates
that
overall
measured
concentrations
have
decreased
across
the
surveys
for
this
compound,
despite
tential
Contaminations
in
the
phthalates
for
the
FY86
survey
as
ested
by
the
QC
data
analysis.
The
contamination
issue
was
ident
for
bis
(
2­
ethylhexyl)
phthalate,
which
was
detected
thod
blanks
in
the
FY86
analysis.
nificant
differences
in
percent
detection
across
lso
observed
in
the
higher­
order
PCB
homologs.

8­
14
0i,?
203
­
Average
percent
detection
was
low
in
FY82
compared
to
the
other
two
surveys
for
hexa­,
hepta­,
and
deca­
chlorobiphenyls,
leading
to
statistically
significant
differences
in
percent
detection
across
the
surveys.
However,
a
corresponding
reduction
in
the
average
LOD
from
FY82
to
FY84
did
not
hold
for
FY86.
In
fact,
the
average
detection
limit
in
FY86
for
these
homologs
exceeded
that
for
FY82
in
hexa­
and
hepta­
chlorobiphenyls.
This
result
appears
to
agree
with
other
findings
indicating
unusually
high
concentrations
for
these
homologs
in
FY86,
which
may
derive
from
analytical
sources
rather
than
environmental
sources.
Thus
while
average
percent
detection
in
FY86
remained
at
levels
consistent
with
earlier
surveys,
occasional
increases
were
observed
for
some
compounds.
However,
the
differences
in
analytical
methods
and
recoveries
observed
from
one
survey
to
another
imply
that
the
differences
may
be
the
result
of
analytical
rather
than
environmental
effects.

8.3
DESCRIPTIVE
STATISTICS
ON
MEASURED
CONCENTRATIONS
A
total
of
54
semivolatile
organic
compounds
were
analyzed
in
the
FY86
NHATS
and
in
at
least
one
of
the
FY82
and
FY84
"
ATS.
The
extent
to
which
statistical
comparison
of
measured
concentrations
was
appropriate
among
these
54
compounds
was
determined
by
initially
summarizing
the
analytical
results
within
each
survey
through
simple
descriptive
statistics.
Some
basic
differences
in
the
results
across
surveys
were
apparent
when
reviewing
summary
statistics.
The
summaries
also
assisted
in
interpreting
comparison
findings.
Initially,
scatterplots
were
produced
for
each
of
these
compounds
in
order
to
identify
any
large
differences
or
patternistic
behavior
in
the
measured
concentrations
between
and
within
the
three
surveys.
Then,
two
approaches
to
calculating
descriptive
statistics
were
applied
to
the
concentrations.
In
the
first
approach,
simple
arithmetic
averages
and
standard
errors
of
the
measured
concentrations
were
calculated.
While
these
statistics
summarize
the
measured
concentrations
across
8­
15
*
analytical
samples,
they
are
not
necessarily
good
estimates
of
the
national
average
concentration.
A
better
approximation
of
national
average
concentration
can
result
by,­
takingweighted
averages
of
the
observed
concentrations.
Thus
the
second
approach
was
to
partition
the
nation
into
subpopulations,
calculate
average
concentrations
within
each
subpopulation,
and
weight
each
average
by
the
1980
Census
population
percentage
for
its
respective
subpopulation.
The
second
approach
can
lead
to
a
improved
estimate
of
national
average
concentration
for
each
compound,
regardless
of
whether
further
statistical
analysis
was
warranted
on
the
compound
concentrations.
In
the
descriptive
summaries
from
both
approaches,
measured
concentrations
were
defined
as
the
total
mass
detected,
divided
by
the
sample
lipid
weight.
Whenever
a
compound
was
not
detected
within
a
sample,
measured
concentrations
were
taken
to
be
one­
half
of
the
LOD
(
as
was
done
in
the
statistical
analyses).
The
percent
detected
values
in
Table
8­
6
indicate
the
frequency
with
which
not­
detected
results
were
observed
within
each
compound.
The
descriptive
statistics
presented
in
the
following
subsections
were
calculated
on
measured
concentration
both
adjusted
and
unadjusted
for
surrogate
recoveries.

8.3.1.
Scatternlots
of
the
Sample
Concentrations
Prior
to
calculating
descriptive
statistics,
scatterplots
of
measured
concentrations
were
generated
for
all
compounds
detected
in
at
least
50%
of
the
FY86
samples
and
which
were
analyzed
in
the
FY82
"
ATS
and/
or
the
FY84
NHATS.
The
scatterplots
illustrate
any
general
differences
or
trends
in
the
concentrations
between
surveys
and
between
batches
within
surveys.
Plots
were
generated
for
concentrations
both
unadjusted
and
adjusted
for
surrogate
recoveries.
These
plots
are
located
in
Appendices
G
and
H,
respectively.
The
concentrations
are
plotted
as
a
function
of
the
analysis
date
in
these
scatterplots.
Therefore
any
trends
in
the
concentrations
over
time
or
batches
are
highlighted
in
these
8­
16
BP
tl;
2$
5
plots.
In
addition,
the
plotting
symbols
indicate
the
age
group
represented
by
the
result
(
1
=
0­
14years,
2
=
15­
44
years,
3
=

45+
years).
Results
in
Chapter
7
indicated
that
age
group
had
a
sign'ificant
effect
on
the
values
of
the
measured
concentrations.
These
plots
illustrate
the
large
extent
to
which
increasing
concentrations
were
associated
with
increasing
age
for
most
compounds.
Also,
the
unadjusted
concentrations
for
the
FY86
"
ATS
appeared
to
be
more
variable
than
in
the
previous
surveys,
excluding
the
effects
of
occasional
outliers.
This
is
especially
apparent
in
plots
of
PCBs
and
some
pesticides.
However,
variability
appears
to
be
more
consistent
across
surveys
when
considering
surrogate­
adjusted
concentrations.
The
plots
suggest
that
this
is
the
result
of
an
increase
in
variability
associated
with
the
surrogate­
adjusted
concentrations
across
all
surveys.
These
scatterplots
also
illustrate
apparent
trends
from
batch
to
batch
within
a
survey.
For
example,
unadjusted
concentrations
of
beta­
BHC
tend
to
decrease
in
later
batches
in
the
FY84
analysis.
The
difference
between
Batch
1
and
the
other
batches
in
FY86
oxychlordane
concentrations
is
also
evident
(
recall
that
Batch
1
data
were
qxcluded
from
statistical
analysis
for
oxychlordane1
.
The
primary
purpose
of
reviewing
scatterplots
prior
to
further
statistical
summaries
or
analyses
was
to
depict
any
obvious
differences
in
results
across
surveys.
Extreme
differences
in
the
values
of
the
concentrations
between
surveys
would
indicate
that
statistical
techniques
may
not
be
necessary
in
making
such
conclusions.
Extreme
differences
from
one
survey
to
another
were
not
apparent
for
these
compounds
based
on
the
scatterplots.

8.3.2.
Unweicrhted
National
Averacres
Appendix
I
presents
simple
arithmetic
averages
(
with
their
standard
errors)
of
the
measured
concentrations
among
the
54
compounds
for
each
of
the
three
surveys.
The
averages
were
8­
17
calculated
across
all
composite
samples
in
Table
8­
3
where
measured
concentrations
were
reported
for
the
given
compound.
Averages
were
calculated
for
two
endpoints:
on
measured
concentrations
adjusted
for
surrogate
recoveries
(
Table
1­
11,
and
on
unadjusted
concentrations
(
i.
e.,
the
recorded
concentrations)
(
Table
1­
21.
The
adjustment
for
surrogate
recoveries
was
performed
to
more
accurately
estimate
actual
concentrations
within
each
sample.
The
adjustment
was
described
in
Section
5.2.

With
some
exceptions,
concentrations
or
LODs
were
reported
for
all
composites
for
a
given
compound
analyzed
within
a
survey.
However
in
the
FY84
survey,
results
for
dieldrin,
endrin,
the
phthalate
esters,
and
the
phosphate
triesters
were
reported
in
only
13
of
the
46
composite
samples.
The
descriptive
statistics
in
Appendix
I
were
calculated
only
to
summarize
the
results
of
the
three
surveys.
Because
these
summaries
ignore
demographic
effects
which
were
determined
to
be
significantly
associated
with
measured
concentration,
the
descriptive
statistics
do
not
necessarily
estimate
national
average
concentrations
in
the
respective
surveys.
Such
estimates
were
obtained
from
statistical
modelling
techniques
for
a
limited
number
of
compounds.

8.3.3.
Weicrhted
National
Averacres
Estimates
of
the
national
average
concentration
estimates
were
obtained
in
this
study
through
statistical
modelling
procedures
rather
than
from
simple
descriptive
statistics
as
discussed
above.
However,
statistical
modelling
was
reserved
only
for
those
compounds
with
sufficiently
high
detection
percentages
within
each
survey.
Thus
an
approach
was
necessary
for
calculating
more
accurate
national
estimates
than
the
simple
descriptive
statistics,
regardless
of
detection
percentages.
To
do
this,
averages
of
composite
concentrations
were
calculated
within
each
of
the
three
age
groups
(
0­
14years,
15­
44
years,
45+
years)
and
were
weighted
by
the
population
proportions
within
each
group.
Age
group
was
selected
for
the
8­
18
weighting
criterion
because
its
effect
on
measured
concentrations
was
most
commonly
significant
across
the
demographic
groups
within
each
survey.
In
addition,
sufficient
numbers
of
sample
results
existed
to
provide
sufficient
accuracy
in
averages
within
each
age
group.
Calculating
the
weighted
national
averages
was
a
multistage
process.
First,
unweighted
arithmetic
averages
were
calculated
for
each
of
the
three
age
groups.
Then
each
age­
group
average
was
multiplied
by
the
population
proportion
in
that
age
group
(
based
on
the
1980
Census).
These
three
results
were
then
summed
to
obtain
the
final
estimate.
Tables
8­
7
and
8­
8
present
the
weighted
national
averages
for
the
54
compounds
analyzed
in
the
FY86
and
in
the
FY82
and/
or
FY84
NHATS.
The
results
in
Table
8­
7
are
based
on
the
actual
measured
concentrations,
while
the
results
in
Table
8­
8
are
calculated
from
concentrations
adjusted
for
surrogate
recoveries.
Results
from
these
two
tables
indicate
that
for
some
compounds,
the
values
of
descriptive
statistics
differ
greatly
between
surveys.
Some
of
these
differences
may
be
more
likely
due
to
differences
in
laboratory
methods
and
instrumentation
than
to
differences
rooted
in
environmental
effects.
For
example,
the
LODs
for
some
of
the
phthalate
esters
and
phosphate
triesters
were
found
to
average
much
higher
in
the
FY82
 
ATS
than
in
the
other
surveys
(
Table
8­
61,
leading
to
higher
average
measured
concentrations
among
the
FY82
composites
for
these
compounds.
The
largest
difference
in
average
concentration
occurred
with
triphenyl
phosphate,
where
the
FY82
weighted
average
was
two
orders
of
magnitude
higher
than
in
the
other
two
surveys.
Most
FY82
composite
samples
report
high
concentrations
for
this
compound
relative
to
the
other
surveys.
The
weighted
average
concentration
for
bis
(
2
ethylhexyl)
phthalate
also
increased
nearly
two
orders
of
magnitude
from
FY84
to
FY86,
primarily
due
to
the
presence
of
samples
with
detected
results
in
FY86
(
78%,
versus
no
detected
8­
19
rlInrtddrldQ)
..................
(
Y00000wm000m(
Y0ded0
EIo
'
Cpm
d
d
dd
(
Y
YYYYYYYYYYYYYYYYYI
4
*
mw
atma
40
w
oatw
atmat(
Yw
wwmm
..................
mmOewmmatwmmd(
Y(
Ymm(
Ym
o
m
Q3
me*
dPdd
(
Ym
d
rld
P4
*
E
YYYYYVYYYYYYYYYVYY
..................
Q3wc0c­
mooatatw(
YQ)(
Ymeatew
Q)
at
qrlm
(
Ydt­
dcye
ri
AA
h
AAA
A
m
ON
d
..
...
c­
0
d
mrarl
rl
N
Nu)
(
Y
N(
Yd
WQ
*
9
E
­.­
rl
Y
YYY
Y
d
oat
at
...
v)'
ul
Q)
at
Ob*
d
W
E
8­
20
hhnhn­
hh­
Q3dOwoatcym
c0.
a*
c­
a
c­
ewcn
.........
OOOOOOOO~

YYYYVYVYY
.........
mwmu)
u)
mv)
mat
cy
AAC.
hAACIA
c­
dl­
c­*
c­
c
0000000N
en
m
Nm
..
Oat
m
YY
YY
e4
Io
c00
..
4'
4
me
Io0
Q)
u)

E
d(

E
n­


...
v
D
rl
M
YYY
4
CJ
...
wrle
ewm
mCJ
w
2
........
oeorl00rl~

YVYYYYYY
rl
rlw
...
....
0
O(
YyQ\
Q)
rl
Y
Y
YUVY
m
0
m...
(
YOQ)
u)
mmm
u)
rl
a
#

a
H
8­
21
W
0
a
wE
4
(
d
Er,
0
k
o(

4
4)

*
9
E
E
0
k
W
nt
E
8­
22
0
MIOlnTT***
Q)
Ew
..
YYYYVYYYVYYYYYYVYY
Y
Y
v
w
Y
Y
YV
YY
Y
v
­.
vuv
Y
Y
drl
Q)
am
mmmPm
mo*
rlwID
..................
moOOPmwd,
mmml­
mQ)
d*
dmmdodmdmrlrl
mNoIDPrld
dm
rl
rl
rl
m
A
nnn
Ln
...
*

Y
YYV
..
mo
ul
.........
orlddoooom
YvYYYvYvY
CIm
0
In
m
ID
?
Im
mCId,
dmP(
Ymm.........
wmmmPCIPIDm
cr
........
0d00000~

YVYYYVVV
wv
dm
..
mm
om
0
'
rt
d
@
am
m
rl
8
U
8­
23
­­­­
anmdmrlmdm
m.
c­.
p..
w.
c­.
c­.
m.
w.
mooooooo
YYYYYYYU
mQ)
mwD(
Yr(
rl
*
d
aomwewmm
*
oom
......
moom
t
.....
wwmm**.
dmmd
dm
YYWYYUYU
n
w
­
m
..
.

&!

0
V
8­
24
8­
25
The
findings
in
this
subsection
imply
that
differences
between
the
surveys
may
not
be
environmental
in
nature
but
may
'
dataresults
in
Table
8­
7.
One
should
remember
that
no
general
conclusions
on
true
national
concentrations
should
be
made
from
these
tables
of
descriptive
statistics
unless
results
of
QC
data
analysis
and
statistical
modelling
agree
with
the
findings.

The
results
from
the
FY82,
FY84,
and
FY86
NHATS
for
semivolatiles
in
composite
samples
were
statistically
compared
by
fitting
the
additive
model
(
Chapter
6)
.
ondata
for
each
survey
8­
26
C'CB6%
2.
azj
separately,
calculating
marginal
estimates
and
standard
errors,
and
comparing
these
estimates
across
surveys
through
approximate
95%
confidence
intervale.
In
order
to
compensate
for
differences
in
recoveries
existing
across
the
three
surveys,
the
additive
model
was
fit
to
the
surrogate­
adjusted
concentrations
within
each
survey
(
see
Section
5.2
on
the
adjustment
method).
Previou slypublished
results
from
the
FY82
and
FY84
 
HATS
may
differ
from
those
presented
in
this
section
as
the
additive
model
and­
the
adjustment
for
surrogate
recoveries
were
not
previously
considered
in
these
surveys.
While
adjusting
for
surrogate
recoveries
attempted
to
remove
effects
of
differing
recoveries
across
surveys
and
to
better
estimate
actual
sample
concentrations,
other
differences
in
analytical
method
and
design
(
documented
in
Section
8.1)
may
contribute
greatly
toward
overall
differences
in
the
marginal
estimates
between
the
surveys.

8.4.1.
Semivolatile
ComDounds
Included
in
Statistical
Coxmarison
Statistical
comparisons
yield
useful
conclusions
only
when
sufficient
numbers
of
detectable
results
are
available
from
each
survey.
Specifically,
statistical
analyses
were
performed
on
only
those
compounds
detected
in
at
least
50%
of
the
composites
within
each
survey.
In
addition,
comparisons
were
made
only
on
compounds
which
were
not
removed
from
consideration
for
statistical
analysis
in
FY86
as
a
result
of
the
QC
data
analysis
(
Section
5.3);
thus
no
phthalates
were
considered
in
the
statistical
comparison.
Based
on
these
criteria,
the
compounds
considered
for
statistical
analysis
across
surveys
were
the
following:

p,
p­
DDT
p,
p­
DDE
Beta­
BHC
Trans­
nonachlor
Heptachlor
epoxide
Hexachlorobenzene
Tetrachlorobiphenyl
Pentachlorobiphenyl
1I
j
8
Hexachlorobiphenyl
m
Heptachlorabiphenyl.

Thus
the
statistical
comparisons
were
limited
to
only
ten
of
the
most
prevalent
pesticides,
PCB
homologs,
and
chlorobenzenes
found
in
the
"
ATS
over
the
years.
In
addition,
the
PCB
parameters
introduced
in
Section
6.2.1.2
(
total
PCB
concentration,
chlorobiphenyl
distribution
across
homologs,
and
chlorination
level)
were
estimated
for
FY82,
M84,
and
FY86
from
the
estimated
average
concentration
levels
for
five
PCB
homologs
resulting
from
fitting
the
additive
model,
The
additive
model
was
fitted
to
data
for
each
of
these
five
homologs
(
tetra­
through
octa­
CB)
since
these
homologs
had
high
detection
percentage8
in
FY86.
The
method
for
estimating
these
parameters
and
their
standard
errors
was
documented
in
Section
6.2.1.2.

8.4.2.
Fittinu
the
Additive
Mode&
The
method
for
fitting
the
additive
model,
as
well
as
the
form
of
the
model
itself,
was
essentially
similar
between
the
three
surveys.
The
primary
differences
in
the
model
fitting
approaches
across
surveys
were
as
follows:

m
The
FY86
model
fitting
included
an
effect
for
Batches
1­
3
versus
4­
5
(
Section
6.1).
This
effect
was
not
included
in
the
model
for
either
N82
or
FY84.

8
For
FY82
and
FY84,
the
errors
attributable
to
measurement
error
and
specimen
sampling
error
were
combined
into
one
error
term,
rather
than
individually
estimated
as
in
the
M86
analysis,
An
estimate
of
measurement
error
was
not
determined
for
these
two
surveys
because
FY82
QC
data
were
not
readily
available,
and
­
84
QC
data
were
not
statistically
analyzed.
Preliminary
analyses
indicated
that
measurement
error
from
the
FY86
QC
data
analysis
VJ~
S
not
appropriate
for
use
in
the
FY82
or
FY84
analyses.

8­
28
One
note
should
be
made
in
reporting
standard
errors
resulting
from
the
additive
model
fitting
to
the
FY84
"
ATS
data.
Large
absolute
error
attributable
to
MSA
sampling
was
observed
for
p,
p­
DDE,
beta­
BHC,
pentachlorobiphenyl,
and
hexachlorobiphenyl
in
this
survey.
When­
thiserror
was
included
in
the
formulas
for
calculating
standard
errors
in
the
marginal
estimates,
these
standard
errors
were
inflated
by
two
to
three
orders
of
magnitude
relative
to
the
marginal
estimates.
Because
these
errors
were
likely
not
an
accurate
portrayal
of
the
true
error,
the
MSA
error
was
not
considered
in
the
additive
model
fitting
in
this
survey.
Thus
the
calculated
standard
errors
may
/
J
be
somewhat
underestimated
for
these
four
compounds
in
FY84.

8.4.2.1.
National
E8thatee.
For
the
above
ten
semivolatiles
and
total
PCB
concentration,
Table
8­
9
presents
the
estimated
national
average
concentrations
(
and
standard
errors)
for
each
of
the
three
surveys,
based
on
fitting
the
additive
model
to
surrogate­
adjusted
concentrations
within
each
survey.
This
table
also
contains
the
estimated
overall
chlorination
percentage
for
PCBs
within
each
survey.
Along
with
these
estimates,
Table
8­
9
includes
the
estimated
difference
from
the
FY86
estimate
for
both
the
FY82
and
FY84
surveys
and
the
significance
level
for
testing
that
this
difference
differs
from
zero.
The
test
was
based
on
the
approximate
t­
statistic
of
the
form
(
i=
S2,
841,
where
NA,,
NASQ,
and
"
86
are
the
FY82,
FY84,
and
FY86
national
average
estimates
and
SEar
SEU,
and
SE86
are
their
standard
errors,
respectively.
Approximate
significance
levels
were
calculated
using
the
standard
normal
distribution.
More
exact
significance
levels
based
on
the
Student­
t
distribution
(
with
degrees
of
freedom
obtained
through
Satterthwaite's
approximation)
was
deemed
too
complex
to
use
in
this
application;

8­
29
rl
ul
cv
0
d
rl
rl
P
a
cv
0
cv
rl
0
.
cy
ul
0
00
0
u)
0
0
cv
a4
a
m
cv
co
ul
rl
0
0
0
4
0
V
0
0
0
0
0
V
0
V
0
U
d
n
rl
A
A
n
CI
n
Q)
CI
n
h
W
*
0
Iz
rl
cy
VI
Iz
n
cv
OD
A
m
cy
03
d
W
W
Iz
Q)
rl
*
ul
m
m
rl
v
Y
rl
cvim
Y
cv
Pa
v
Y
Y
cv
Y
m
d,

Y
u
V
Y
+
v
m
rl
a
VI
+*
0
a)
ro
0
Q)
+
VI
*
(
u
0
W
ul
0
m
+,
­
4
m
4­
4
*
an
rl
W
ul
rl
m
u)
0
u
n
VI
4
I
cy
I
r
rl
rl
I
7­
l
0
rl
0
m
rl
d
rl
d
d
W
cv
0
0
0
0
0
VI
0
Q)

a
P
cv
0
Ul
ul
rl
03
0
0
0
0
0
0
0
0
0
V
V
V
0
V
0
cv
bl
A
n
n
n
n
n
A
h
ri
n
'\
Q)

m
VI
0
m
cv
h
r
(
u
A
Q)

W
W
W
*
m
rl
d
d,
r
d
r
orcvm
m
d
Y
m
4
W
rl
m
Y
d
WE
Y
W
v
Y
Y
Y
Y
v
Y
rl
m
+
+
t
+
W
VI
m
r­
r
cv
0
W
cv
Q)
m
P
Y
W
4
rl
d
rl
0
W
m
0
W
n
1
*
I
cv
rl
I
d
VI
rl
d,
cv
d
I
A
n
A
n
A
n
n
A
n
cy'
P
h
A
a,
VI
w
0
0
VI
VI
d,
*
m
a0
(
u
d,
m
rn
cv
cv
cv
d
Y
Y
rl
d
cv
m
Y
Y
Y
Y
Y
Y
Y
Y
Y
Y
0
s
r
cy
P
0
'.
W
*
m
*
111
cv
m
r
m
VI
m
r
W
m
rl
cu
r
Q)

P
rl
cv
rl
d
ul
VI
rl
m
d
W
VI
A
h
h
A
A
A
h
n
A
rl
ul
h
h
A
VI
ul
rl
w
rl
0
d
4n.
e
VI
rl
rl
0
ul
cv
a
ul
,
­.
Y
rl
rl
rl
dv1
rl
ul
(
u
­
w
v
Y
Y
Y
Y
5
Y
Y
Y
Y
rl
C
0
m
Q)
d
VI
(
d
m
VI
m
an
Q)
a\
a0
0
cv
rl
ul
0
00
OD
rl
ul
cv
0
Q)

2
rl
rl
rl
rl
W
d,
rl
d
rl
VI
m
0
a
A
'.
A
n
n
h
n
n
A
9
h
A
h
d
d,
ul
n
0
ul
a0
W
0
url
m
d
Q)
m
rl
r
a,
P
d,
m
(
d
rn
m
m
VI
cy
rl
cv
rl
m
Y
Y
Y
v
Y
w
Y
w
w
w
Y
W
u
cc
0
d
[
z
m
W
W
cd
ul
d
d
a\
W
U3
a0
0)
al
0
al
m
a
P
*
[
z
al
­
4
f:
rl
rl
cv
rl
m
d
[
z
d
Q)
*
m
W
­
4
c
tn
­
4
W
E
$
W
m2
W
W
E
2
a
h
m
w
p:
8
X
ow
­
4
pc
E
X
E
w
pc
H
H
a
H
a8
w
m
H
m
u
0
0
3
U
0:
2
p:
v1
8
5
m
kU5
F3
9
­
8
?
9
8
s
sX
sB
s
u
3
a
w
0
id
B
3
3
u
X
a
WA
p:
­
404
?
?
A
9
W
2
!
3
3
E
f3
d
u5
a.
pc.
El
w
E
E
W
W
W
a
0
Q
a
a
m
X
w
pc
X
X
3
8­
30
these
significance
levels
are
well
approximated
by
the
standard
normal
distribution
with
the
sample
sizes
observed
in
each
survey.
Significant
differences
from
the
FY86
national
estimate
were
observed
at
the
0.05
level
for
both
the
FY82
and
FY84
surveys
(
Table
8­
9).
In
the
FY82
survey,
the
national
estimates
for
the
PCB
homologs
and
total
PCBs
were
lower
than
in
the
FY86
survey;
the
difference
was
highly
significant
for
tetra­,
penta­,
and
hexa­
chlorobiphenyls,
as
well
as
for
total
PCBs.
However,
except
for
tetrachlorobiphenyl,
different
IQS
were
used
between
the
FY82
and
FY86
surveys
for
the
PCB
homologs.
A
significant
difference
in
the
national
estimates
for
beta­
BHC
was
also
observed
between
FY82
and
FY86;
the
FY86
estimate
was
135
ng/
g
lower
than
the
FY82
estimate.
Both
surveys
used
the
same
IQS
for
quantitating
beta­
BHC.
In
the
FY84
survey,
the
national
estimate
for
only
one
of
the
analyzed
PCB
homologs
differed
significantly
from
the
N86
estimate.
The
115
ng/
g
increase
in
hexachlorobiphenyl
for
FY86
relative
to
FY84
was
highly
significant.
An
increase
of
164
ng/
g
in
total
PCBs
for
FY86
relative
to
FY84
was
also
highly
significant.
Increases
in
the
FY86
national
estimates
for
p,
p­
DDT
and
p,
p­
DDE
relative
to
the
FY84
estimates
were
also
significant
at
the
0.05
level.
All
three
of
these
compounds
were
quantitated
using
the
same
IQS
in
the
FY84
and
FY86
"
ATS.
Table
8­
10
presents
the
estimated
chlorobiphenyl
distribution
across
the
five
prevalent
PCB
homologs
for
the
FY82,

FY84,
and
FY86
surveys.
It
is
clear
that
the
dominance
of
hexachlorobiphenyl
observed
in
the
FY86
analysis
was
present
in
the
FY82
and
FY84
surveys
as
well.

8.4.2.2.
Marginal
Estimates.
Marginal
estimates
for
the
four
census
regions,
three
age
groups,
two
sex
groups,
and
two
race
groups
are
presented
(
with
their
standard
errors)
in
Tables
J­
3
through
J­
6
(
Appendix
J)
for
the
ten
analyzed
semivolatiles,
total
PCBs,
and
overall
chlorination
level
across
the
three
8­
31
h­.
1";
­,,,,,
,

Table
8­
10.
Chlorobiphanyl
Distribution
Across
the
Five
PCB
Homologs
Considered
for
Statistical
Analysis
in
the
FY86
IURATS
Chlorobiphenyl
distribution
for
homolog
i
(
i=
4,5,6,7,8)
is
calculated
as
follows:

average
concentration
estimate
for
Homolog
i
*
average
concentration
estimate
for
Total
PCB
where
"
Total
PCB"
is
the
sum
of
the
average
concentration
estimates
across
the
five
homologs
in
the
above
table.
Each
homolog
omitted
from
the
table
was
detected
in
no
more
than
30%
of
the
NHATS
FY86
composite
samples.

8­
32
surveys.
The
estimates
for
census
regions
and
age
groups
are
plotted
for
each
survey
in
Appendix
K
with
plus
and
minus
two
standard
error
bars.
The
tables
also
contain
estimates
of
the
difference
in
the
marginal
estimates
between
the
FY86
survey
and
each
previous
survey.
The
following
results
are
suggested
from
the
marginal
estimates
in
Tables
5­
3
through
J­
6
(
references
to
significant
differences
between
surveys
are
made
at
the
0.05
level
using
the
t­
test
described
above):

rn
Large
differences
in
the
estimates
of
PCB
homologs
and
of
total
PCB
concentration
were
evident
between
FY86
and
FY82
for
many
of
the
subpopulations.
These
differences,
often
several
times
larger
than
their
standard
errors,
were
generally
significant
for
the
northcentral
and
northeast
census
regions,
the
15­
44
and
45+
age
groups,
whites,
and
both
sexes.
In
each
case,
the
FY86
estimate
was
higher
than
the
FY82
estimate.
.
..

rn
Among
PCB
homologs,
significant
differences
in
the
marginal
estimates
between
FY86
and
FY84
were
primarily
relegated
to
hexachlorobiphenyl.
All
subpopulations
except
the
0­
14
age
group
observed
significant
differences
in
the
marginal
estimate
for
this
homolog
between
the
two
surveys.
For
total
PCBs,
significant
differences
between
surveys
were
observed
for
the
northcentral
and
northeast
census
regions,
the
45+
age
group,
both
race
groups,
and
females.
In
each
case,
the
FY86
estimate
was
higher
than
the
FYS4
estimate.

rn
Excluding
the
PCB
homologs,
few
significant
differences
in
marginal
estimates
were
observed
between
FY82
and
FY86
among
the
subpopulations.

rn
Excluding
the
PCB
homologs,
there
is
some
evidence
that
significant
differences
exist
in
marginal
estimates
for
p,
p­
DDTand
p,
p­
DDEbetween
the
FY86
and
FY84
surveys.
Differences
in
p,
p­
DDEwere
significant
across
all
age
groups,
sex
groups,
and
race
groups;
the
FY86
estimate
was
larger
than
the
FY84
estimate
in
each
instance.
For
p,
p­
DDT,
significant
differences
were
observed
for
the
45+
age
group,
northeast
census
region,
and
males
as
a
result
of
larger
marginal
estimates
in
the
FY86
survey.

8­
33
I
Thus
Tables
J­
3
through
5­
6
indicate
that
whenever
significant
differences
occurred
in
the
marginal
estimates
between
surveys,
higher
estimates
were
associated
with
the
FY86
survey.
In
FY82,
most
differences
occurred
with
PCBs;
these
differences
were
primarily
observed
for
the
two
highest
age
groups
and
the
northeast
and
northcentral
census
regions.
In
EY84,
most
+

differences
were
observed
for
hexachlorobiphenyl,
p,
p­
DDE,
and
p,
p­
DDT;
these
differences
tended
to
be
consistent
across
all
subpopulations.

8.4.2.3.
Likelihood
Ratio
Tests.
For
the
ten
compounds
analyzed
within
each
survey
using
the
additive
model,
statistical
hypothesis
tests
were
conducted
within
each
survey
to
determine
if
there
were
statistically
significant
differences
in
average
concentration
between
individuals
between
different
geographic
regions,
age
groups,
race
groups,
and
sex
groups.
Likelihood
ratio
principles
were
used
to
conduct
these
tests
(
Section
6.2.2).
For
the
FY86
survey,
these
tests
were
performed
in
Section
7.3.
Table
8­
11lists
the
significance
levels
obtained
from
performing
the
likelihood
ratio
tests
on
the
FY82,
FY84,
and
FY86
data.
These
results
show
a
relative
consistency
across
all
surveys.
No
significant'differences
were
noted
across
age
groups
or
sex
groups
in
either
survey.
Significant
effects
due
to
census
region
and
age
groups
were
observed
in
each
survey
for
most
of
the
PCB
homologs,
hexachlorobenzene,
and
pesticides.
Specifically,
the
importance
of
both
the
census
region
and
age
group
effects
on
the
concentration
values
remains
evident
in
the
FY86
NHATS
as
in
the
prior
surveys.

8.4.2.4.
Conclusions.
The
conclusions
of
statistical
analysis
on
surrogate­
adjusted
concentrations
for
semivolatile
organic
compounds
are
similar
between
the
three
surveys.
Age
group
and
census
region
appear
to
be
the
most
significant
demographic
effects
on
many
of
these
concentrations
within
each
survey.
,

Table
8­
11.
Significance
Levels
from
Hypothesis
Testa
for
Differences
Between
Demographic
Groups
for
Selected
Sdvolatiles
in
the
FY82,
FY84,
and
FY86
"
AT&)

Significance
Levels
Compound@)
FY82
FY84
.
FY86
Effect
of
Census
Region
P
P­
DDT
Pt
P­
DDE
BETA­
BHC
HEPTACHLOR
EPOXIDE
TRANS­
NONACHLOR
HEXACHLOROBENZENE
TETRACHLOROBIPHENYL
PENTACHLOROBIPHENYL
HEXACHLOROBIPHENYL
HEPTACHLOROBIPHENYL
eo.
001*
<
o.
001*
<
o
.001*
0.005*
eo.
001*
0.001*
0.
on*
0.141
0.947
0.442
<
o.
001*
0.031*
eo.
001*
eo.
001*
0.187
eo.
001*
eo.
001+
eo.
001:
>
O.
50
0.216
0.036
0.001*
eo.
001*
0*
009+
eo.
001*
eo.
001*
0.047*
0*
001*
0.408
0.140
Effect
of
Age
Group
PtP­
DDT
\

Pt
P­
DDE
BETA­
BHC
HEPTACHLOR
EPOXIDE
TRANS­
NONACHLOR
HEXACHLOROBENZENE
TETRACHLOROBIPHENYL
PENTACHLOROBIPHENYL
HEXACHLOROBIPHENYL
HEPTACHLOROBIPHENYL
>
O.
50
eo.
001+
eo.
001*
0.001*
>
O.
50
0.009*
0.001*
>
0.50
0.015*
0.117
eo.
001*
eo.
001:
0.022*
eo.
001*
eo.
001
co.
001*
eo.
001*
eo.
001*
0.057
eo.
001;
KO.
om*
<
o.
001*
<
0.001
eo.
001:
0.005*
>
O.
50
eo.
001
0.811
eo.
001*
0.001*

'
I
Effect
of
Sex
Group
PtP­
DDT
p,
p­
DDE
BETA­
BHC
HEPTACHLOR
EPOXIDE
TRANS­
NONACHLOR
HEXACHLOROBENZENE
TETRACHLOROBIPHENYL
PENTACHLOROBIPHENYL
HEXACHLOROBIPHENYL
HEPTACHLOROBIPHENYL
0.952
0.379
0.966
0.946
0.694
0.814
.
0.994
0.353
0.6'
23
0.534
0.551
0.565
0.771
0.233
0.321
0.974
0.971
0.777
0.379
0.543
0.260
0.675
0.617
0.549
0.562
0.381
0.693
0.203
0.243
0.490
Table
8­
11.
(
cont.)

Significance
Levels
Compound@)
~~
82­
84
FY86
Effect
of
Race
Group
PtP­
DDT
Pt
P­
DDE
BETA­
BHC
HEPTACHLOR
EPOXIDE
TRANS­
NONACHLOR
HEXACHLOROBENZENE
TETRACSLIOROBIPIIENYL
PENTACHLOROBIPHENYL
HEXACHLOROBIPHENYL
HEPTACHLOROBIPIIENYL
*
0.433
0.259
0.286
0.805
0.808
0.569
0.259
0.452
0.501
0.495
0.786
0.846
0.484
0.711
0.879
0.890
0.802
0.936
0.383
0.908
0.337
0.605
0.228
0
619
0.389
0.245
0.244
0.280
0.289
0.368
Significance
declared
at
the
0.05
level.

tl)
Data
adjusted
for
surrogate
recoveries
(
see
Section
5.2).
Likelihood
ratio
tests
are
based
on
the
chi­
square
distribution.

p,
p­
DDE
concentrations
for
FY86
use
m/
Z=
316
(
see
Section
5.1.2).

8­
36
Despite
the
similarities
between
surveys,
differences
in
estimated
subpopulation
concentrations
were
significant
for
some
PCB
homologs
and
pesticides
between
the
FY82/
FY84
surveys
and
the
FY86
survey.
In
most
cases,
these
differences
indicated
that
FY86
estimates
were
higher
than
in
the
previous
surveys.
These
results
are
contrary
to
the
downward
trends
concluded
in
previous
trends
analyses
(
Robinson,
et.
al.,
1990).
These
results
are
more
likely
due,
however,
to
analytical
effects
rather
than
environmental
effects.
Since
a
period
of
only
four
years
exist
between
the
collection
of
specimens
for
these
three
surveys,
it
is
unlikely
that
major
changes
in
the
actual
concentration
levels
in
human
adipose
tissue
will
be
observed
over
this
time
period
under
normal
exposure
conditions.
In
making
generalizations
across
the
surveys,
such
analytical
factors
as
differences
in
IQS
and
surrogate
compounds
between
surveys,
and
differences
in
design
factors,
must
also
be
considered
as
attributable
toward
observed
differences.
The
national
average
estimates
from
the
statistical
modelling
on
ten
semivolatiles
tend
to
agree
with
the
estimates
obtained
from
the
weighted
average
calculations
(
Section
8­
3­
21.

Thus
the
weighted
averages
in
Table
8­
8
may
provide
useful
estimates
in
national
average
concentrations
which
are
relatively
similar
to
what
would
be
achieved
through
statistical
modelling.

8­
37
­­

­­
9.0
~
FERENCES
Dinh,
K.
1991.
USEPA.
Guideline
for
adjusting
concentrations
in
"
ATS
data.
Draft
Final
Report.
Washington,
DC:
Office
of
pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.

Draper
NR,
and
Smith
H.
1981.
Applied
regression
analysis,
Second
Edition.
New
York:
John
Wiley
and
Sons.

Kish,
L,
and
Scott,
A.
1971.
Retaining
units
after
changing
strata
and
probabilities.
Journal
of
the
American
Statistical
Association,
66
(
335)
:
pp.
461­
470.

Mack
GA,
Leczynski
B,
Chu
A,
and
Mohadjer
L.
1984.
Battelle
Columbus
Division
and
Westat,
Inc.
Survey
design
for
the
national
adipose
tissue
survey.
Draft
Final
Report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
a
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Pfotection
Agency.
Contract
No.
68­
01­
6721.

Mack,
GA,
and
Panebianco,
DL.
1986.
Battelle
Columbus
Division.
Statistical
Analysis
of
the
M82
MIATS
Broad
Scan
Analysis
Data.
Draft
Final
Report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerlythe
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Document
No.
NHATS­
SS­
04.
Contract
No.
68­
02­
4243.

MRI.
f988a.
Analysis
of
adipose
tissue
for
semivolatile
analytes:
adipose
tissue
sample
compositing.
Interim
Report
#
1.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4252.

MRI.
1988b.
Quality
assurance
project
plan
for
WA­
28:
broad
scan
analysis
of
adipose
tissue
from
the
FY
1986
EPA
NEATSrepository
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4252.

MRI.
17
March
1989.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
FY
1986
NHATS
composites,
batch
1
interim
data
report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protecti.
onAgency.
Contract
No.
68­
02­
4252.

MRI.
19
May
1989.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
FY
1986
"
ATS
composites,
batch
2
interim
data
report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4252.
MRI.
21
July
1989.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
­­
FY
1986
NHATS
composites,
batch
3
interim
data
report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4252.

MRI.
21
July
1989.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
­­
FY
1986
NHATS
composites,
batch
4
interim
data
report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4252.

~
 21.
22
September
1989.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
­­
FY
1986
NHATS
composites,
batch
3
revised
tables
and
analysis
report
forms.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4252.

MRI.
25
September
1989.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
­­
FY
1986
NHATS
composites,
batch
5
interim
data
report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
NO,
68­
02­
4252.

MRI.
28
September
1989.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
­­
FY
1986
NHATS
composites,
~

batch
5
revised
tables
and
analysis
report
forms.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Bnvironmental
Protection
Agency.
Contract
No.
68­
02­
4252.

MRI.
13
July
1990.
Analysis
of
adipose
tissue
for
semivolatile
organic
compounds
­­
FY
1986
NHATS
composites,
revised
PCB
data
batches
1­
5.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
NO.
68­
02­
4252.

\
Orban
JE,
Leczynski
B,
Collins
TJ,
and
Sasso
NR.
1988.
Battelle
Columbus
Division.
FY86
NHATS
composite
design.
Final
Report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4294.

9­
2
f­?
fi
:(?
228
Orban
JE,
and
Lordo,
RA.
1989.
Battelle
Columbus
Division.
Statistical
methods
for
analyzing
NHATS
composite
sample
data
­­
evaluation
of
multiplicative
and
additive
model
methodologies.
Draft
Final
Report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4294.

Panebianco
DL.
1986.
Battelle
Columbus
Division.
A
review
of
hospital
nonresponse
and
its
effect
on
standard
errors
of
sample
estimates
in
NHATS.
Draft
Final
Report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
02­
4243.

Robinson,
PE,
Mack,
GA,
Remmers,
J,
Levy,
R,
and
Mohadjer,
L.
1990.
Trends
of
PCB,
hexachlorobenzene,
and
@­
benzene
hexachloride
levels
in
the
adipose
tissue
of
the
U.
S.
population.
Environmental
Research.
53:
pp.
175­
192.

Rogers,
J.
1991.
Westat,
Inc.
FY86
NHATS
Semi­
Volatile
Organic
Compounds:
Outlier
Analysis.
Final
Report.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
D9­
0174.

Stanley,
JS,
Balsinger,
J,
Mack,
GA,
and
Tessari,
JD.
1986.
Midwest
Research
Institute,
Battelle
Columbus
Division,
and
Colorado
State
University.
Comparability
study
of
analytical
methodology
for
TSCA
chemicals
in
human
adipose
tissue.
Quality
Assurance
Program
Plan.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerly
the
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contracts
No.
68­
02­
3938and
68­
02­
4243,

Westat,
1990.
NHATS
Comparability
Study,
Draft
3.0.
Washington,
DC:
Office
of
Pollution
Prevention
and
Toxics
(
formerlythe
Office
of
Toxic
Substances),
U.
S.
Environmental
Protection
Agency.
Contract
No.
68­
DO­
0174.

USEPA.
1986.
Broad
scan
analysis
of
the
FY82
NHATS
specimens.
Volume
111:
Semi­
volatile
organic
compounds.
EPA
Publication
NO.
EPA­
560/
5­
86­
037.

USEPA.
1991.
Chlorinated
dioxins
and
furans
in
the
general
U.
S.
populqtion:
NHATS
FY87
results.
EPA
Publication
No.
EPA­
S60/
5­
91­
003.

9­
3
­­
3.
Rccfpitnt's
Accession
NO.

PAGE
I
FPA
747­­
R
94­
OOi
I
4.
Title
8nd
SuMWe
5.
Report
Date
Semivolatile
Organic
Compounds
in
the­
Gener'al
U.
S.
Pbpulation:
NHATS
FY86
Results
­
Volume
I
6.

a
Petfonning
Organizatian
Rept.
N.,.

14.
12
spond~
.
Ownhtlon
Wmo
and
~
fim~~

Chemical
Management
Division
Office
of
Pollution
Prevention
and
Toxics
U.
S.
.
Environmental
Protection
Agency
Washinston,
DC
20460
1%
*
pPletY
Notes
Khoan
T.
Dinh,
Work
Assignment
Marlager
Technical
Programs
Branch
The
Environmental
Protection
Agency's
National
Human
Adipose
Tissue
Survey
(
NHATS)
was
performed
annually
to
quantify
the
levels
of
selected
chemicals
in
the
adipose
tissue
of
humans
in
the
U.
S.
population.
Specimens
collected
in
fiscal.
year
1986
were
earmarked
for
an
analysis
to
estimate
national
average
concentrations
of
111
semivolatile
compounds
in
adipose
tissue,
to
identify
differences
in
average
concentrations
among
subpopulatiuns,
and
to
compare
results
with
previous
surveys
in
the
"
ATS.
For
17
semivolatiles
detected
in
at
least
half
of
the
50
analytical
samples,
statistical
modeling
techniques
were
conducted
to
address
these
objectives.

Among
demographic
effects
on
average
concentration,
the
age
group
effect
was
most
often
statistically
significant,.
with
higher
concentrations
associated
with
higher
ages.
Among
PCB
homologs,
the
45+
year
age
group
had
from
108%
to
706%
higher
average
concentrajtions
than
$
he
0­
14
year
age
group,
with
similar
increases
observed
for
pesticides.
Geographic
effects
were
only
occasionally
significant,
and
no
significant
race
or
sex
effects
were
observed.

Statistically
significant
increases
in
concentration
(
generally
less
than
100%)
from
the
FY82
NHATS
were
observed
for
three
PCB
homologs,
while
increases
of
50%
to
100%
from
the
FY84
NHATS
were.
significant
for
p,
p­
DDT
and
p,
p­
DDE.
Mixed
findings
were
observed
for
other
semivolatiles
analyzed
in
all
three
surveys.
7
ooatmant
hslyJs
8.
oercrtpto~~

c.
COSATl
Field/
Group
B.
A~
aUabilftyatemant
19.
Security
Class
(
This
Rewort)
21.
No.
of
Pages
NTIS
1
22.
Price
I
1
See
Instructions
on
Reverse
OPTIONAL
FORM
2R
(
4­
7;
(
Formerty
NTIS­
35)
Department
of
Commerce
