EPA
September 2009
REVIEW OF THE UNIVERSITY OF MICHIGAN
DIOXIN EXPOSURE STUDY
National Center for Environmental Assessment
Office of Research and Development
U.S. Environmental protection Agency
Washington, DC 20460

-------
Preferred citation:
U.S. EPA (Environmental Protection Agency). 2009. Review of the University of Michigan
Dioxin Exposure Study. September 30, 2009. National Center for Environmental Assessment,
Washington, DC. Available from http://www.epa.gov/ncea.
Page ii

-------
DISCLAIMER
This document is itself a review and as such has not been peer-reviewed. EPA will
consider any significant technical comments it receives. Mention of trade names or commercial
products does not constitute endorsement or recommendation for use.
ABSTRACT
This document presents EPA's review of the University of Michigan Dioxin Exposure
Study (UMDES). This is one of the largest studies of dioxin exposure ever conducted and is an
important contribution to the understanding of levels of dioxin in human blood, house dust, and
soils and the factors associated with dioxin exposure.
The design of the study was well-suited for meeting the objective of identifying patterns
of serum dioxin, furan and PCB levels among adults living in the Midland-Saginaw region and
provided the basis for making comparisons with a reference region in Jackson-Calhoun counties,
100 miles to the south. The relatively large number of participants and population based
statistical sampling design are important strengths of this study. The data collected support
reliable estimates of the distributions of dioxin concentrations in blood, soil and dust.
The UMDES design was only partially successful in meeting the objective of evaluating
factors associated with serum dioxin concentrations. The associations between the observed
serum levels and a variety of demographic factors (age, body mass index, where one lives, work
history, etc.) were estimated; however, estimates of associations with other factors, particularly
soil and dust concentrations, may be problematic. This was because subpopulations likely to be
subject to high exposures (e.g., gardeners, fishers) were not specifically included in the sample
design. Of particular note was that the UMDES study focused on exposure to individuals aged
18 years or older. The results are, therefore, not directly relevant to children, a sensitive
subpopulation that is often the focus of studies of environmental exposure especially with regard
to possible exposure through contact with soil and dust. Without increased focus on
subpopulations likely to have high exposures, the study does not provide data that readily
support analysis of the impacts of sources that may increase total exposure a relatively small
amount relative to diet or other large sources of background exposures.
For risk-based decision-making, EPA's focus is typically on highly exposed and/or sensitive
subpopulations, in addition to the general population. The UMDES did not target such
subpopulations and coverage of groups of interest for risk-based decision-making is limited.
Thus, the lack of emphasis on sampling of subpopulations likely to be most affected — such as
people living on properties with very high soil levels and people consuming large amounts of
possibly contaminated fish and game — is a significant drawback.
Page 1

-------
CONTENTS
EXECUTIVE SUMMARY	5
Study Description	5
Comments on Study Design	6
Comments on Study findings	7
Implications for EPA	9
Conclusions	9
SECTION 1: INTRODUCTION	10
SECTION 2: DESCRIPTION 01 STUDY	12
2.1	University of Michigan Dioxin Exposure Study (UMDES)	12
2.2	Goals and Objectives	12
2.3	Summary Description of Study	12
2.4	Personnel	16
2.5	Sample collection	16
2.6	Data Management	17
2.7	Stati sti cal Analy si s	17
2.8	Quality Assurance and Quality Control	18
2.9	Data Confidentiality and Human Subjects	18
SECTION 3: EVALUATION 01 STUDY	19
3.1	Limitations of the Scope of the Study	19
3.2	Sample Collection, Analysis and Quality Assurance	19
3.3	Clarity of Study Goals and Objectives	20
3.4	Stati sti cal Analy si s	21
3.5	Study Design	23
SECTION 4: EVALUATION OF MAJOR CONCLUSIONS	28
4.1	Nonserum correlations Involving Soil and Dust	28
4.2	Serum correlations with Soil and Dust	30
4.3	Serum correlations with Other Factors	34
SECTION 5: UNANSWERED QUESTIONS ABOUT EXPOSURE	41
5.1	Can Pharmacokinetic Modeling help to Inform Study Results?	41
5.2	What about Children and Exposure to Contaminated Soils and Dust?	42
5.3	Were all Populations of Interest Adequately Represented in the Study? 	42
5.4	Is it Reasonable that the Correlations Between Soil and Blood were weak?	43
SECTION 6: RELEVANCE TO EPA	44
SECTION 7: CONCLUSIONS	45
SECTION 8: REFERENCES	48
Page 2

-------
APPENDIX A: EVALUATION OF THE IMPACT OF SOIL CONCENTRATIONS
ON BODY BURDENS 01 INDIVIDUALS	52
APPENDIX B: EVALUATION FOT HE BLLOD SERUM DATA FROM
SUBJECTS LIVING ON PROPERTIES WITH MAXIMUM SOIL TEQ
VALUES GREATER THAN 1000 PPT	70
Page 3

-------
AUTHORS, REVIEWERS, AND ACKNOWLEDGEMENTS
The National Center for Environmental Assessment (NCEA) within the Office of Research and
Development (ORD) of the U.S. Environmental Protection Agency prepared this evaluation.
Authors (in alphabetical order):
Jeffrey B. Frithsen, USEPA, ORD, NCEA
Henry Kahn, USEPA, ORD, NCEA
Matthew Lorber, USEPA, ORD, NCEA
John Schaum, USEPA, ORD, NCEA
U.S. EPA Reviewers:
Marlene Burg, USEPA, OSWER
David Bussard, USEPA, ORD, NCEA
Wendy Carney, USEPA, Region 5
Milt Clark, USEPA, Region 5
Kacee Deener, USEPA, ORD, NCEA
Steve Ells, USEPA, OSWER
Mary Logan, USEPA, Region 5
Mario Mangino, USEPA, Region 5
Greg Rudloff, USEPA, Region 5
Paul White, USEPA, ORD, NCEA
Acknowledgements:
The development of this project benefited by conversations with Glenn Rice, USEPA, NCEA.
We also acknowledge the assistance of David Garabrant and A1 Franzblau from the University of
Michigan who freely shared their time to discuss the University of Michigan Dioxin Exposure
Study with us.
Page 4

-------
EXECUTIVE SUMMARY
This document presents EPA's review of the University of Michigan Dioxin Exposure
Study (UMDES). This is one of the largest studies of dioxin exposure ever conducted and is an
important contribution to the understanding of levels of dioxin in human blood, house dust, and
soils and factors associated with dioxin exposure. In recognition of the significance of this
study, U.S. Environmental Protection Agency Administrator, Lisa P. Jackson, committed the
Agency to conducting this review as a component of the EPA's Science Plan for Activities
Related to Dioxins in the Environment, dated May 26, 2009 (www.epa.gov/dioxin/scienceplan).
The objectives of EPA's review were to comment on the design, implementation and
results of the UMDES and the relevance of the study's findings to EPA's regulatory mission.
The EPA review comments are summarized below and provided in detail in the attached report.
Study Description:
The University of Michigan (UM) conducted a field study of dioxin and dioxin-like
compounds in the blood serum of the human population living in the area around Midland,
Michigan. This is referred to as the "University of Michigan Dioxin Exposure Study" or
UMDES. Financial support for the study was provided by the Dow Chemical Company through
an unrestricted grant to the University. A large staff at the University of Michigan, and
collaborators at several other academic institutions, has worked on the study for more than five
years. All primary data collection was completed in 2005 and to date 20 journal articles and
abstracts have been published or accepted for publication in scientific journals with many
additional manuscripts under development. More than one hundred presentations based on the
data have been made at scientific conferences and meetings. UM has constructed a
comprehensive web site that provides ready access to the study protocol, publications,
presentations, and review comments and responses (http://www.sph.umich.edu/dioxin/index.htmi).
The objective of this study was described as follows: "to describe the pattern of serum
dioxin, furan and PCB levels among adults and to understand the factors that explain variation in
serum dioxin, furan and PCB levels." (Garabrant et al., 2005).
The data collection for the study was designed as a two-stage, area probability, household
sample of the adult population living in five geographic regions in the area in and around
Midland, Michigan. Four of the five geographic areas sampled are in the counties surrounding
Midland and are thought to be most affected by releases from Dow. The fifth area was located in
two counties 100 miles south of Midland and was intended to serve as a background or reference
area. A total of 946 participants provided serum samples (695 in the four Midland study areas
and 251 in the background area). Additionally, soil and household dust samples were collected
from the residences of most of these participants. An extensive questionnaire was used to collect
demographic and life style information from all participants.
Page 5

-------
Comments on Study Design:
The UMDES study focused on exposure to individuals aged 18 years or older. The results
are, therefore, not directly relevant to children, a sensitive subpopulation that is often the focus of
studies of environmental exposure especially with regard to possible exposure through contact
with soil and dust. Additionally, other subpopulations likely to be subject to higher than typical
exposures (e.g., gardeners, fishers, individuals living on more highly contaminated soils) were
not specifically included in the sample design. Information was collected regarding such
exposures, but limited sample sizes may limit the strength of study findings. The study design
does not provide data that readily support analysis of the impacts of sources that may increase
total exposure a relatively small amount relative to diet or other large sources of background
exposures.
The first objective of the study was to identify patterns of serum dioxin, furan and PCB
levels among adults, and EPA concludes that the design of the study was well-suited for meeting
this objective. The relatively large number of participants and population based statistical
sampling design are important strengths of this study. The data collected support reliable
estimates of the distributions of dioxin concentrations in blood, soil and dust, reported as means,
medians, lower and upper percentiles.
The second objective of the study was to evaluate factors associated with serum dioxin
and EPA believes that the UMDES design may have been only partially successful in meeting
this objective. The associations between the observed serum levels and a variety of demographic
factors (age, body mass index, where one lives, work history, etc.) were estimated successfully.
However, estimates of associations with other factors, particularly soil and dust concentrations
may be problematic because contributions from typical background dietary exposures are usually
much greater compared to typical contributions from soil and dust. The UMDES researchers
attempted to address this by including in the sample two areas that would be more likely to have
higher soil concentrations. These were an incinerator plume area and a 100 year floodplain area
downstream of the plant. Further, they used a very high sampling rate in these areas (17% of the
total population in the floodplain and 30% in the plume area). The final overall sample included
a total of 23 properties out of 766 with maximum soil dioxin levels over 1,000 ppt TEQ
(TEQ=Toxicity Equivalent1). The relatively low median soil levels in the floodplain (11 ppt
TEQ), and suggestion of higher levels from other studies (ATS, 2009), support the concern that
this representation of subjects from high soil properties was limited. A related design issue is
how well the study represented individuals with specific behaviors (such as gardening,
consuming local fish or game, or raising animals for local consumption) that could lead to
elevated dioxin exposures. No design elements were used to ensure representation of these
activities in the sample. However, the UMDES sample did identify two subjects with elevated
1 Dioxin concentrations and exposures are presented in terms of toxic equivalents (TEQs). TEQs allow
concentrations of dioxin mixtures to be expressed as a single value computed by multiplying each congener
concentration by a toxicity weight (toxic equivalency factor or TEF) and summing across congeners. TEFs are
expressed as a fraction equal to or less than 1 with 1 corresponding to the most toxic dioxin congener, 2,3,7,8-
tetrachlorodibenzo-p-dioxin (2,3,7,8-TCDD).
Page 6

-------
serum levels who engaged in high soil contact activities (one gardener and one farmer) and the
UMDES researchers have highlighted these findings.
The UMDES provided some unanticipated but useful findings. These involved the
identification of several unexpected dioxin exposure scenarios. First was the discovery of soils
containing elevated levels of dioxin outside the floodplain due to movement of soils from the
floodplain to residences for fill or other purposes. Second was the discovery of elevated serum
levels that appear to be associated with the activities of a ceramic clay hobbyist. Third was the
discovery that polychlorinated biphenyls (PCBs) were the dominant contributors to high TEQ
levels in soil at one residence (linked to past paint use) and in dust at nine residences (where one
was linked to a carpet pad and others had unknown sources). In most other samples, dioxins and
furans were the dominant contributors to TEQ levels.
A Quality Assurance Project Plan was drafted which, although not complete in all areas,
addressed the key elements of quality assurance procedures. However, the publications to date
have not included a report on quality assurance performance measures.
Comments on Study Findings:
EPA concludes that the following findings are supported by the UMDES data and analyses:
•	Soils from properties in the Midland/Saginaw study areas contain higher concentrations
of dioxin than soils in the reference area of Jackson/Calhoun.
•	Higher concentrations of dioxins were found in the serum of residents living within the
four study areas comprising the Midland/Saginaw region as compared to the reference
Jackson Calhoun area.
•	Dusts from households in the Midland Plume area contain higher dioxin levels than dusts
from households in the other study areas.
•	Living in the Midland/Saginaw study areas during the years 1960-1979 was associated
with elevated serum dioxins.
•	Working at Dow during the years 1940-1959 was associated with elevated serum dioxin
levels. This does not appear to be a finding for TEQ, but it does appear to be a finding
for 2,3,7,8-TCDD.
The most highlighted finding of this study is that age is positively associated with serum
dioxin levels. EPA generally agrees that the analyses done on UMDES data support this finding.
Indeed this is an expected result as other studies of U.S. populations show increase in average
levels of serum dioxins and TEQs with age. However, while there is a tendency for blood levels
to increase with age, many older people in the UMDES study had lower blood levels than many
younger people. Age may have the strongest association with serum levels of all variables
studied, but substantial variation apparent in the data indicated this to be a weak to moderate
positive relationship. The value of R2, a measure of the variation in the dependent variable
(serum level) explained by variation in the explanatory variable (age) was not reported for serum
level versus age alone. For serum TEQs versus an aggregate of nine demographic variables
(including age) the R2 was reported to be 0.396 (Garabrant et al., 2009b). This indicates that the
nine demographic variables in the aggregate account for 39.6% of the variation in the observed
Page 7

-------
serum levels and that R2 for age by itself would be smaller. Thus, the statements that imply that
age is a strong predictor of serum levels appear overstated.
The relationship between blood serum dioxin and soil dioxin was less clear. Generally, weak
relationships between various measures of soil and serum dioxin levels were noted but the
overall conclusions in the more recent documents imply that there is not a meaningful
relationship between these two key factors. EPA believes this finding is supported by the
statistical analysis of the data presented in Garabrant et al (2009b). This is not an unexpected
finding, given that dietary exposures explain over 90% of total exposures to the general
population; direct soil impacts explain only about 1% of total exposures. A pharmacokinetic
modeling exercise done as part of this evaluation showed that at the 95% soil concentration
found in the Floodplain soils, 223 ppt TEQ, adult body burdens would only increase by 3 ppt
TEQ or less (Appendix A). The impact to an adult body burden was more significant at a soil
concentration of 1000 ppt TEQ, increasing body burden by about 11 ppt TEQ, which was a little
less than half the median level in Midland Saginaw area of 27.3 ppt TEQ. However as noted
above, the study design would be expected to result in limited coverage of high dioxin-
concentration soils which would make evaluation of this relationship problematic. EPA also
emphasizes that this finding should not be stated without the caveats about the study limitations.
This is especially important with regard to possible effects due to soil exposure which are
unlikely to occur unless soil concentrations are high.
UMDES found little association between household dust dioxin concentrations and serum
dioxin levels. EPA notes the lack of strong or consistent findings regarding dust in the regression
relationships. However, the design issues noted above for soils would also apply to dusts. Also,
it is noted that two key documents, the 2006 report (University of Michigan, 2006) and the 2009
journal article (Garabrant, 2009b), have slightly different results. The 2006 report suggested that
a relationship might exist between PCB 1 18 levels in serum and household dust, whereas
Garabrant et al. (2009b) suggest that the relationship is with PCB 126.
A variety of findings addressed serum dioxin/fish relationships including: where the fish
came from (store bought or from impacted water bodies), what kind of fish, whether the
individuals recreationally fished, and so on. The overall comment that there is a relationship
between consumption of fish and serum dioxin levels is generally supported by the findings that
were outlined in the primary publication describing the statistical analysis of the data (Garabrant
et al, 2009b). While the association was noted, the statistical relationship was weak. The
UMDES reported somewhat mixed results with regard to eating fish from the contaminated
waters (Tittabawassee River, Saginaw River, and Saginaw Bay). This is based on the statement
in UM (2006) that people who eat fish from the contaminated water bodies have higher levels
of dioxins in their blood compared to people who don't eat fish from these areas, while Garabrant
et al (2009b) stated that there was no positive association between consumption of fish from
these areas and serum dioxins.
Certain recreational activities were found to be associated with dioxin serum levels. This was
supported by a positive association between recreational activities and dioxin body burden that
appeared in the 2006 report (UM, 2006). However, like the conclusion noted above on
Page 8

-------
consuming fish from the impacted water bodies, this conclusion was not supported in the 2009
literature publication on predictors of serum levels.
Implications for EPA:
The UMDES study focused on exposure and included no health status information on those
included in the study. Therefore, the UMDES data do not support analysis of the association
between observed blood levels and possible health effects.
The direct relevance of the UMDES study to EPA's mission is uncertain. The UMDES is an
overall population study. For risk-based decision-making, EPA's focus is typically on highly
exposed and/or sensitive subpopulations, in addition to the general population. The UMDES did
not target such subpopulations and coverage of groups of interest for risk-based decision-making
is limited. Thus, the lack of emphasis on sampling of subpopulations likely to be most affected —
such as people living on properties with very high soil levels and people consuming large
amounts of possibly contaminated fish and game — is a significant drawback.
EPA's soil dioxin remediation goal of 1000 ppt TEQ is largely based on a scenario of
childhood soil ingestion (i.e., based on a scenario of exposure in ages 1 to 30 years old with
about two thirds of the dose occurring prior to age 18). With regard to the remediation goal, it is
possible to examine the results for the subpopulation of subjects in the UMDES sample living on
properties with soil concentrations greater than 1000 ppt. Interpretation of the results would be
limited, however, by the absence of children and the likely sparse coverage by design of this
subpopulation in the study.
Dioxin presents a significant concern for risk assessors and policymakers because the
background exposure to dioxin is already at a level considered to be of concern (EPA, 2003).
Food consumption explains 95% of total exposure. Any incremental exposure over the dietary
background would be expected to increase the risk over the range of concern.
Conclusions:
The UMDES has produced a credible and valuable source of data on dioxin levels in adults.
To the credit of the UM researchers, these data have been, and continue to be, used to support
extensive analyses. EPA believes that the data will support additional analysis that may further
clarify relationships between blood serum measurements and key factors, like soil
concentrations. Specifically, EPA believes that further study of specific subpopulations in the
UMDES sample - such as those exposed to high soil concentrations, high fish and game
consumers, and those with high blood serum levels - may provide the basis for additional
informative insights. For example, preliminary analysis of data from properties with maximum
soil concentrations greater than 1000 ppt suggests that the subjects living on these properties
have a distinctly different pattern of serum dioxin levels (see Appendix B).
Page 9

-------
SECTION 1: INTRODUCTION
This document presents EPA's review of the University of Michigan Dioxin Exposure
Study (UMDES). This is one of the largest studies of dioxin exposure ever conducted and is an
important contribution to the understanding of levels of dioxin in human blood, domestic dust,
and environmental soils, and factors associated with dioxin exposure leading to increased
concentrations in blood. In recognition of the significance of this study, U.S. Environmental
Protection Agency Administrator, Lisa P. Jackson, committed the Agency to conducting this
review as a component of the EPA's Science Plan for Activities Related to Dioxins in the
Environment, dated May 26, 2009 (www.epa.gov/dioxin/scienceplan).
The objectives of EPA's review were to comment on the design, implementation and
results of the UMDES and the relevance of the study's findings to EPA's regulatory mission. In
conducting this evaluation , EPA reviewed the study protocol (Garabrant et al., 2005) and
August 2006 report (UM, 2006) published on the UMDES web site
(www. sph.umich/edu/dioxin/index.html), and peer reviewed publications and in press papers,
focusing on those emphasizing the primary findings (Garabrant et al., 2009a, 2009b; Hedgeman
et al., 2009; Demond et al., 2008) Manuscripts in preparation, and other materials that became
available after about August 1, 2009, have not been considered in the preparation of this report.
EPA does acknowledge receipt of the presentations made by UMDES researchers at the August
Dioxin 2009 conference in Beijing, China.
To gather information and ideas to assist with our review, we conducted a number of
interviews with individuals who are knowledgeable about this study. These individuals
included:
•	Region V Staff (Mary Logan, Mario Mangino, Milt Clark, Wendy Carney) - meeting in
Chicago, IL May 14, 2009
•	Tracey Easthope (Ecology Center) and Ted Schettler (Science and Environmental
Network) - phone interview July 20, 2009
•	John Kern (Kern Statistical Services) - phone interview July 28, 2009
•	Linda Dykema (MI Department of Health) - phone interview July 30, 2009
•	Deborah MacKenzie-Taylor (MI Department of Environmental Quality) - phone
interview July 30, 2009, with follow-up questions on September 9, 2009.
•	Linda Birnbaum (NIEHS, chair of the UMDES Science Advisory Panel) - phone
interview August 6, 2009
•	David Garabrant and Alfred Franzblau (University of Michigan) - meeting in Alexandria,
VA, August 10, 2009, with a follow-up data request on September 11, 2009
•	David Kleinbaum, (Emory University, former member of the UMDES Science Advisory
Panel) - phone interview August 18, 2009
Several of those contacted submitted written comments to us. Key points of comments
included:
Page 10

-------
•	The study purpose and hypotheses have not been clearly stated. This was an issue mostly
near the beginning of the study when it was not made clear why it was being undertaken
and the intended use of the results.
•	There was not a focus on the "high end" of the potentially exposed population: children,
those who consume locally produced foods (including fish and terrestrial animal food
products) and game, and those who live on the most contaminated soil. Some suggested
that these high end subpopulations, not including children, could have been better
targeted in a stratified, or otherwise different, survey design.
•	There was a common concern relating to the regression model used by the UMDES
researchers to evaluate the data. Including a large number of possible explanatory
variables in a single model can confound the results since many of the variables are
known to have little or no effects on dioxin exposure. An alternate approach proposed
was to establish a "base" model which included variables known to influence or be
related to dioxin exposure, such as age, gender, dietary factors, body mass index, and so
on, and to this model add one (or maybe two related) parameter to see the influence of
this individual parameter. This model was termed "base + 1". With this approach, one
can better evaluate and compare these individual factors.
•	The soils on the sampled properties were not representative of the high end and given
what was sampled, may not have been studied properly to date. Regarding
representativeness, the study by Dow termed the Middle Tittabawassee River remedial
investigation (ATS, 2009, also discussed in Section 3) identified many more areas and a
higher percentage among areas sampled, of higher soil concentration. Regarding
evaluations to date: a) UMDES used the most contaminated sample on the site and
should have tested the average soil concentration, b) soil concentrations and years on
properties could have been combined to obtain a new parameter similar to "pack years"
that are used to evaluate the impacts of cigarette smoking, and c) the soil concentrations
found suggest that a bimodal trend exists and that an attempt to separately evaluate the
high portion of this data set instead of combining all the data may have been more
informative.
•	There was a concern that this study could be used to determine clean-up levels at this site
and possibly other sites nationally.
The concerns described above parallel many of EPA's concerns and are addressed in the
following discussions.
The EPA understands that the UMDES is not finished and is likely to continue for many
years. Future reports will present new findings and may modify earlier ones. Thus our review
should be regarded as a review of the findings to date with the understanding that these could
change in the future.
Page 11

-------
SECTION 2: DESCRIPTION OF STUDY
2.1	University of Michigan Dioxin Exposure Study (UMDES)
The University of Michigan (UM) conducted a field study of dioxin and dioxin-like
compounds in the blood serum of the human population living in the area around Midland,
Michigan. This is referred to as the "University of Michigan Dioxin Exposure Study" or
UMDES. Financial support for the study was provided by the Dow Chemical Company through
an unrestricted grant to the University. A large staff at the University of Michigan has worked
on the study for more than five years. All data collection was completed in 2004, and to date 20
journal articles or abstracts have been published or accepted for publication in scientific journals
with many additional manuscripts under development. More than one hundred presentations
based on the data have been made at scientific conferences and meetings. UM has constructed a
comprehensive web site that provides ready access to the study protocol, publications,
presentations, and review comments and responses (http://www.sph.umich.edu/dioxin/index.htmi).
2.2	Goals and Objectives
The motivation for this study has been described (with small variations among the
articles and presentations) as: "The University of Michigan Dioxin Exposure Study (UMDES)
was undertaken in response to concerns among the population of Midland and Saginaw Counties
that dioxin-like compounds from the Dow Chemical Company facilities in Midland have
contaminated areas of the City of Midland and sediments in the Tittabawassee River
Floodplain."
The objective of the study is described in the protocol (Garabrant et al., 2005) as "to
describe the pattern of serum dioxin, furan and PCB levels among adults and to understand the
factors that explain variation in serum dioxin, furan and PCB levels."
2.3	Summary Description of Study
The design of the UMDES is a population-based two stage area probability household
sample survey designed to select subjects from five regions in the state of Michigan. These five
areas are discussed below (see Figure 1).
The Floodplain Area (FP): This area includes the 100-year Federal Emergency
Management Agency (FEMA) floodplain of the Tittabawassee River as well as respondents who
reported flooding of their home by the Tittabawassee River. The FP area was assumed to be the
most impacted by contaminated sediments of the Tittabawassee River since flooding would
transport dioxins onto surface soils of residential properties. Prior sampling has documented
contamination of Tittabawassee sediments. Dow Chemical owns most waterfront property along
the first six miles on the Tittabawassee River downstream from their facility. EPA was informed
that no sampling occurred along these six miles (Garabrant, pers. comm.. 2009), although there
Page 12

-------
are residential properties within these first six miles (McKenzie, pers. comm., 2009). Given their
location, these areas might be more impacted than locations further downstream from the Dow
facilities.
Bay
County
Dow plant
100-year floodplain
| Near floodplain
Plume boundary
Other Midland,
Saginaw,
Bay Counties
Saginaw
County
i i i i i i i i i
0 3.5 7 14 Kilometers
Midland
County
Saginaw Bay
Figure 1. Map of Midland, Saginaw, and Bay counties, Michigan, showing the Dow Plant and the 100-year
floodplain of the Tittabawassee River.
The near-floodplain of the Tittabawassee River (NFP): The NFP area was assumed to be
less impacted by flooding compared to the Floodplain area. Still, proximity to the river could
influence soil concentrations and it might be a factor in recreational use of the river.
The plume area in the City of Midland downwind from the historic incineration activities
of the Dow plant (MP): The MP area was added to the study later during the planning and
implementation of the study due to comment received by the UMDES research team during
public meetings and consultations. Concern was expressed that individuals living in this area
might be impacted from historical emissions from the Dow incinerator. The area was delineated
by UMDES researchers who used an EPA air deposition model (the Industrial Source Complex
Short Term (ISCST3) model) to identify an area where predicted soil concentrations would equal
or exceed 75 ppt TEQ based on modeled depositions from the Dow incinerator. Later on, the
2 Dioxin concentrations and exposures are presented in terms of toxic equivalents (TEQs). TEQs allow
concentrations of dioxin mixtures to be expressed as a single value computed by multiplying each congener
Page 13

-------
UMDES researchers verified their definition of the area with the use of a more recently
developed EPA air deposition model (AERMOD).
Elsewhere in Midland and Saginaw counties and parts of Bay County outside the
floodplain, near-floodplain, and plume areas (MS). There was no explicit expectation of
impacts in this area, and as noted below, the sampling weights were very low in this area; and
Jackson and Calhoun counties as a control or referent population (JC). This area is
located more than 100 miles away from the Dow facilities in Midland. The JC area was chosen
as an area that was not impacted by activities of the Dow facilities, nor was otherwise impacted
by specific known sources of dioxin. Results from soil and serum testing in the JC area have
generally shown this was an accurate assessment.
Prior to selecting individuals within the five areas to participate in the study, a power
calculation was conducted to determine a target number of individuals needed within each area
to give the study adequate statistical power. Details of this power calculation have not been
published, but the result was a target of 175 people (based on information provided in an EPA
meeting with Garabrant and Franzblau on August 10, 2009) with the exception of the MP area.
As noted, the decision to include this area came after other areas had been delineated. Decisions
regarding study sample size and other design details had already been made, and interviews were
already underway in other study areas when the MP area was added to the study. The final
number of participants selected exceeded 175 within the four areas. As discussed below, while
the number of people selected in the MP area was much less than this target population, the area
in general was small and the sample weights demonstrate that the MP area was oversampled in
comparison to other areas.
Study participants were selected using a 2-stage design. The first stage was a random
selection of census blocks within the region. Census blocks are delineated based on population
density. Generally all census blocks contain the same, or a similar, number of household units.
Therefore, a census block in a densely populated region would be smaller in geographic size
compared to a census block in an area of lower population density. Also, there would be a larger
number of census blocks in areas of higher population density compared to areas of lower
population density. With these characteristics, and the design to sample census blocks randomly,
the study sampled at a higher rate in areas of high population density. Once a census block was
selected, "clusters" of housing units within the blocks were identified by the UMDES researchers
using a "drive-around" procedure. The second stage involved the random selection of clusters
within the census block. The intent was to enlist each household within a cluster, assuming that
the household was "eligible" and willing to participate. An "eligible" household was defined as
one with at least one eligible subject residing. Eligible subjects were individuals who were at
least 18 years old and had lived in their homes for at least 5 years. Repeated attempts were made
to contact each household. UMDES researchers claim a participation rate of eligible households
to be about 80%. If there was more than one eligible individual within a household willing to
participate, only one individual was selected and that selection was made randomly.
concentration by a toxicity weight (toxic equivalency factor or TEF) and summing across congeners. TEFs are
expressed as a fraction equal to or less than 1 with 1 corresponding to the most toxic dioxin congener, 2,3,7,8-
tetrachlorodibenzo-p-dioxin (2,3,7,8-TCDD).
Page 14

-------
Interviews (lasting about one hour) were given to eligible individuals prior to blood
sampling. Not all individuals who were interviewed consented to provide blood samples. A
total of 1324 individuals deemed eligible were interviewed, and blood samples were obtained
from 946 of these. In addition to a serum sample from the eligible individual within the selected
household, there was also the intent to collect soil samples as well as dust samples representing
that household. However, various issues precluded obtaining soil and dust samples at each
residence, including whether the eligible individual rented or owned, whether the owner (living
in the house or not) wanted soil and/or dust sampling, and who owns the dust in the house. In
any case, there were a smaller number of households supplying soil and dust samples as
compared to individuals supplying a serum sample. Vacuum house dust samples were obtained
from 764 homes, and soil samples were obtained from 766 households. A summary of the final
numbers and types of samples are shown in Table 1. Included in this table are the sample
weights for the serum samples. Since individuals were randomly selected within each region,
these weights can be understood as follows: each person's sample represents the sample weight
number of similarly defined people within the region. For example, the sample weight of 6 in
the FP region means that each person's serum results represent 6 other eligible individuals within
that region.
Table 1. Final numbers of study samples and sample weights*
Description
Flood-
plain
Near
Flood-
plain
Midland
Plume
Other
Midland/
Saginaw
Jackson/
Calhoun
Total
Blood
samples
243
205
43
204
251
946
Household
dust
samples
205
161
32
168
198
764
Soil
samples
203
164
32
173
194
766
Blood,
Dust, and
Soil
195
156
30
167
183
731
Serum
Sample
Weights
6
6
3-4
300
600
NA
* The sample numbers are from UM (2006) and sample weights from verbal information
provided at the August 10, 2009 EPA meeting with Garabrant and Franzblau.
Data collection for the main study was completed in 2004-2005. The chemical analyses
of soil, blood, and dust, and the extensive questionnaire has resulted in a rich data base. The
primary way these data have been analyzed is via a linear regression model that expressed blood
serum levels as a function of multiple independent variables. A stepwise selection procedure was
Page 15

-------
used to identify which independent variables were statistically significant predictors of serum
dioxin concentration. This methodology is summarized in Garabrant et al. (2009a).
The first report was published in 2006 (UM, 2006), and the first of ongoing public
dissemination of materials at scientific conferences was also in 2006 at the annual dioxin
international conference in Oslo, Norway. To date 20 journal articles have been published or
accepted for publication in scientific journals with many additional manuscripts under
development.
2.4	Personnel
The six principle investigators for UMDES are David Garabrant (Professor,
Environmental Health Sciences), Alfred Franzblau (Professor, Environmental Health Sciences),
Brenda Gillespie (Assistant Professor of Biostatisties), Peter Adriaens, (Professor and Director
Fundamental and Applied Microbiology for the Environment), James Lepkowski (Associate
Professor Department of Biostati sties) and Avery Demond (Associate Professor, Civil and
Environmental Engineering). Numerous other faculty members and graduate students have been
involved in this study. UM formed a Scientific Advisory Board for this study consisting of
Linda Birnbaum (formerly Environmental Protection Agency now National Institute of
Environmental Health Sciences), Paolo Boffetta (International Agency for Research on Cancer),
Ronald Hites (Indiana University), and David Kleinbaum (Emory University).
2.5	Sample Collection
Detailed descriptions of the soil sampling protocol were provided in Appendix 8 of the
Study Protocol (Garabrant et al., 2005). Soil samples were collected at up to 7 locations at each
property: along the house perimeter, contact areas (i.e. gardens) and near the river for properties
in the flood plain. At each location 3 cores were collected inside a 3 foot diameter ring. Each
core had two vertical layers, a surface layer, 0-1 inch, and a lower layer, 1-6 inches, consistent
with other dioxins studies. The reasons for sampling the surface layer are twofold: the surface
layer best represents what individuals are exposed to, and dioxins are typically deposited onto
the surfaces of soil from sources including air sources resulting in the highest soil concentrations.
The samplers allowed for direct sample collection in the tube, sealing of the tube, and
minimization of cross-contamination between samples. The removal of vegetation from the
surface with means of a scissors is typical, and analysis of the vegetation provided a potentially
useful data point for further analysis. All sealed sample cores were stored on ice (4° C) before
transport to the University of Michigan Environmental and Water Resources Engineering
(EWRE) laboratories.
Detailed descriptions of the dust sampling protocol were provided in Appendix 7 of the
Study Protocol (Garabrant et al., 2005). The household vacuum dust samples were taken from
two sampling locations that present the highest potential for human contact with household dust
and dirt. A High Volume Small Surface Sampler (HVS3) was used to collect the sample. The
dust sampling technicians attempted to collect a minimum of 10 grams of total dust in order to
Page 16

-------
yield an analytical detection limit of 1 part per trillion (ppt). If the amount of dust collected from
the initial sampling area within each location was not sufficient, secondary areas were marked
and sampled as needed. The total surface areas of all of the sampling areas that make up each
sampling location were recorded, as well as the surface types from which the sample was taken.
By measuring the sampling areas, results can also be provided in terms of "loadings", which are
mass of dioxin per surface area. Loadings are often used in place of, or in addition to,
concentrations in exposure studies. Samples were transported on ice to a dedicated 4° C cooler
until delivery to Alta Analytical Laboratory for analysis.
Detailed descriptions of the blood sampling protocol were provided in Appendix 6 of the
Study Protocol (Garabrant et al., 2005). Each participant was asked to give an 80 mL sample of
blood. Blood samples were collected and handled by a mobile phlebotomy service. Blood was
allowed to clot, centrifuged and the serum decanted. Serum was frozen at -20°C and shipped on
dry ice to the analytic laboratory.
2.6	Data Management
Data management details are described in Garabrant et al., 2005. All data were stored as
Microsoft Access datasets (separately for soil, dust, and blood) and were also converted to SAS
datasets. All data were range-checked, and variable crosschecking performed as appropriate.
Data were merged on the participant ID number, which appears in all records. Participant names
are only available in interviewer and sample tracking databases, and are not included with the
study data available for statistical analysis.
2.7	Statistical Analysis
The fundamental basis of the UMDES is a statistical survey of the populations of five
geographic areas in the Midland and Saginaw region. The design used to collect the data for this
survey is referred to as a "two-stage area probability household sample". Data collected in this
manner are referred to as observational which means that a sample is drawn and characteristics
of the sample are measured, i.e., observed. The data collected have been used by the University
of Michigan team to support a wide range of statistical analyses which have been documented in
numerous publications and presentations. The basic approach used by the UMDES to analyze the
data was estimation of the distribution of serum dioxin levels in the adult population and
multiple regression analysis to examine the relationship of the serum levels to a number of
possible explanatory factors. This is consistent with the summary statement in the study protocol
document: "This is the protocol for a study of dioxin, furan and coplanar PCB exposure among
the population of Michigan to describe the pattern of serum dioxin, furan and PCB levels among
adults and to understand the factors that explain variation in serum dioxin, furan and PCB
levels." (Garabrant et al., 2005). Detailed descriptions of the sample design, sample size
determination and plans for analyzing the data are provided in the protocol document.
The primary overall goal, "...to describe the pattern of dioxin, furan and PCB levels among
adults..", in operational statistical terms, becomes one of using the observed data to estimate the
Page 17

-------
distribution of blood serum concentrations in the adult population of the geographic areas
included in the study. The two-stage probability sample design yields data that support the
estimation of the distribution of serum levels over all five areas and comparisons of the
distributions among the five areas.
The phrase "... and to understand the factors that explain the variation...." expresses a second
main goal to collect data that support investigation of the association of serum levels with a
number of explanatory variables. The basic approach, which the UMDES has employed, is to
use multiple regression analysis to estimate and examine associations between a dependent
variable, such as blood serum levels, and a number of explanatory variables that are
hypothesized to have an effect on or influence the dependent variable.
The statistical analysis plan provided in Section 4.7 of the protocol document describes in
some detail appropriate statistical methods for using the UMDES data to estimate population
distributions of blood serum levels and using multiple regression to evaluate and estimate
relationships between the observed blood levels and explanatory variables. The methodology
employed is commonly used for the analysis of large biomedical, sociological and environmental
data sets and the description provided in the protocol is sound and thorough.
2.8	Quality Assurance and Quality Control
The Quality Assurance Project Plan (QAPP) is presented in Appendix 9 of the Study
Protocol (Garabrant et al., 2005). The QAPP covers project management; data generation and
acquisition; assessment and oversight; and data validation and usability activities.
2.9	Data Confidentiality and Human Subjects
This study was performed in compliance with University of Michigan policies and
procedures governing the use of human subjects in research (Garabrant et al., 2005). The
Consent forms are presented in Appendix 11 and the certificate of confidentially is presented in
Appendix of 13 of the Study Protocol (Garabrant et al., 2005).
Page 18

-------
SECTION 3: EVALUATION OF STUDY
3.1	Limitations of the Scope of the Study
The primary limitation of this exposure study is it was limited to individuals aged 18
years or older. Thus the study did not include children, a particularly sensitive subpopulation
especially with regard to the possible exposure through contact with soil and dust
In addition the study has two design related limitations. First, this is an observational
study. Observational studies provide the basis for examining relationships or associations
between the variables. However, it is difficult, if not impossible, to establish reliable cause and
effect relationships on the basis of observational studies. This is distinct from experimental
studies where the values that variables assume and the assignment of those values to subjects are
under the control of those conducting the experiment. The UMDES has produced credible
estimates of the distribution of serum levels in the adult populations studied which is an
important contribution. The concomitant environmental variables associated with each subject
such as dioxin levels in soil and dust, fish and game consumed and so on are only observed. As
a consequence, the full range of these variables or sampling rates within specified categories
(e.g., high or low levels) for these variables are not guaranteed to be represented in the data.
Second, the study does not specifically include as part of the design, focus on subpopulations
who engaged in activities suspected to be associated with significant chance for elevated
exposure (e.g., consumption of local fish and wild game; contact with highly contaminated soils;
consumption of food grown in FP-NFP areas).
Although these limitations have been fully acknowledged by UMDES investigators, they
may not have always been sufficiently emphasized in the presentation of results and findings.
3.2	Sample Collection, Analysis and Quality Assurance
Sample collection, handling and analysis was well described, appropriate and generally
acceptable.
This QAPP was developed using the 1998 EPA Guidance for Quality Assurance Project
Plans, EPA QA/G5. The UMDES QAPP addresses the critical elements identified in the
guidance but does not provide the necessary detail that would allow one to evaluate and or
validate the results. The publications to date have not included a report on quality assurance
performance measures. Don Patterson (CDC) made a presentation to the UMDES Scientific
Advisory Board on October 20, 2005 describing the results of a quality control exercise
involving blinded serum samples. His lab and Alta (now Vista) analyzed 20 split serum samples
for dioxins, furans, and PCBs (D/F/Ps). The presentation indicated that the two labs had very
close agreement on these measurements.
Page 19

-------
3.3 Clarity of Study Goals and Objectives
The UMDES literature EPA reviewed has had slight variations in the wording of the
goals, purposes, and objectives of the study. One statement implied that the relationship between
soil and dust and blood levels was the primary objective of the study, with other relationships
(eating fish, etc) also examined. Others stated that the primary objective was to determine what
factors best explained blood levels found, with soil and dust among the list of factors. Consistent
among all statements was the intent to correlate various factors to blood serum concentrations.
The following is an example of the statement of purpose and/or objectives as they were similarly
portrayed in the primary journal articles EPA reviewed for this evaluation:
The UMDES is a hypothesis-driven study designed to answer important questions
about human exposure to dioxins in the environment of Midland, where the Dow
Chemical Company has operatedfor > 100 years, and in neighboring Saginaw,
Michigan. In addition, the UMDES includes a referent population from an area
of Michigan in which there are no unusual sources of dioxin exposure andfrom
which inferences regarding the general Michigan population can be derived. A
central goal of the study is to determine which factors explain variation in serum
dioxin levels and to quantify how much variation each factor explains (Garabrant
et al, 2009a).
This paper also describes the "hypothesis" as "whether the contamination of the
environment by dioxins from the Dow Chemical Company's operation in Midland, MI is
associated with increased body burdens of dioxins among some residents of the surrounding area
of Midland, Saginaw and southwestern Bay counties." This is the only location in any of the
publications where EPA could find statements regarding the description of the study being
"hypotheses-driven". Additional hypotheses have been delineated in a PowerPoint response to
comments from Dr. John Kern, a statistician consultant, as:
•	Soils and household dust in the Midland/Saginaw area are contaminated
with DLCs (clioxin like compounds,) having congener profiles that
correspond to the congener profiles of Dow 's historic contamination.
•	Living on contaminated soil is associated with increased serum DLCs.
•	Living in a house with contaminated household dust is associated with
increased serum DLCs.
•	Consumption offish from contaminated water bodies is associated with
increased serum DLCs.
•	Consumption of game from the contaminated areas is associated with
increased serum DLCs.
Page 20

-------
• Consumption of eggs, milk, dairy products, fruits, and vegetables raised
on contaminated property is associated with increased serum DLCs.
(Garabrant 2009c)
This PowerPoint response then goes on to state, "The UMDES is a confirmatory observational
study, the goal of which is to test the above hypotheses."
EPA concludes that the most appropriate objective statement, given the design of the
study, is provided in the study protocol (Garabrant et al., 2005): "This is the protocol for a study
of dioxin, furan and coplanar PCB exposure among the population of Michigan to describe the
pattern of serum dioxin, furan and PCB levels among adults and to understand the factors that
explain variation in serum dioxin, furan and PCB levels." The reason EPA favors this phrasing
over others in the literature is that it is the only one that clearly states that a goal of the study is to
"describe the pattern of serum dioxin, furan and PCB levels among adults." All others have
focused only on the second of the two objectives noted here, "...and to understand the factors that
explain variation in serum dioxin, furan and PCB levels." EPA interprets the phrase "describe
the pattern of serum dioxin	"to mean that the goal was to estimate the distribution of serum
dioxin levels in the overall adult population living in the geographical areas studied. This is the
primary goal supported by the data collected and detailed plans provided in the protocol. EPA
interprets the phrase "... and to understand the factors that explain the variation...." to mean that
a second main goal was to investigate the association of serum levels with a number of variables
through multiple regression analysis.
3.4 Statistical Analysis
The UMDES is an observational study based on sound principles of statistical sampling
design. The two-stage area probability household sample design is an appropriate design for this
study. Accordingly, the data collected support the estimation of the overall distribution of blood
levels in the populations of the areas sampled in the study and comparisons of the distributions
among the areas sampled. The data also support estimation of the associations between blood
levels and the various demographic and environmental factors measured in the study. The
multiple regression methodology employed by UMDES to estimate these relationships and
investigate associations is a standard general approach for the analysis of this kind of data. The
multiple regression methodology involves modeling a dependant variable as a linear combination
of a number of explanatory variables. In the UMDES study, the dependant variable is blood
serum dioxin levels and the explanatory variables are several demographic and environmental
factors. The blood serum measurement for each of the subjects in the study is associated with a
set of explanatory variables and the methodology provides the mechanism for estimating
associations between the dependant variable and the explanatory variables. The UMDES team
has demonstrated considerable knowledge and skill in analyzing the data.
Interpretation of the results of the analysis is, however, critically important. The data do
not support the determination of causal relationships between blood levels and demographic and
environmental factors. The presentation of the results does not always make clear that
estimation of associations between blood levels and other factors is not the same as establishing
causal relationships. For example, the following statements from UMDES (2006) could be
Page 21

-------
easily misinterpreted to assert that causal relationships have been established: "Older age is by
far the most important factor related to higher levels of dioxins in people's blood. Eating fish
from the Tittabawassee River, Saginaw River, and Saginaw Bay also leads to higher levels of
dioxins in blood." That is, the phrase "also leads to higher levels" implies that the study has
found that age and eating fish are direct causes of higher blood levels.
In response to the critical review by statistician consultant Dr. John Kern (2009), a
presentation by Dr. Garabrant (2009c) refers to the UMDES as a "confirmatory observational"
study. The UMDES protocol discusses model building to account for various demographic and
environmental factors in general terms but does not identify specific models as would be
expected in a confirmatory study. In fact, the UMDES team appears to have conducted a
substantial amount of exploratory analysis in order to arrive at the results presented. This is a
reasonable approach for analysis of observational data. In particular, presentation of results in
Garabrant et al. (2009b) identifies an aggregate set of variables referred to as "demographic
factors" which explain 39.63% of the observed variation in serum levels [Garabrant et al.
(2009b), Table 1], The aggregate set of variables in "demographic factors" is a compilation of a
number of explanatory variables: age, age2, sex, BMI, BMI loss in the past 12 months, breast-
feeding, number of pregnancies, race, smoking, and interaction terms. This finding, i .e., that a
set of nine variables plus interaction terms are required to explain about 40% of the variation
appears to be the result of considerable exploratory analysis. Interpretation of this finding is
problematic because the effect on any one of the variables is difficult to discern from the
summary listing of linear regression results in Table 2 of Garabrant et al. (2009b). In the
interests of clarification, it would likely be helpful to examine the data separately for each
possible explanatory variable in a descriptive manner. The plot of blood serum versus age in
Figure 1 of Garabrant et al. (2009b) is an illustration of such an analysis. The plot shows a
tendency for blood levels to increase with age but, from an overall perspective, the relationship is
weak. Additionally, there is considerable variability apparent in the plot and it is clear that many
older subjects have lower blood serum levels than younger subjects. The basis for the statement:
"Age was strongly associated with serum TEQ and with all serum congener concentrations." in
Garabrant et al. (2009b) is apparently the statistically significant p-values for the age variable in
the first line of Table 2 of Garabrant et al. (2009b). The R2 value for age, i.e., percent of
variation in blood serum level explained by age is not provided but would appear to be
substantially less than the 39.63% for all the demographic variables combined. The plot of the
data in Figure 1 may be more informative with regard to the relationship. As stated elsewhere in
this review, EPA believes that the importance of age as an explanatory factor for blood serum
levels is overstated. As a general matter, the conclusions drawn by the UMDES appear to be
overly dependent on regression statistics such as p-values and R2 values. This approach, while
technically valid, is not consistent with guidance provided in the literature. For example,
Cummings and Rivara (2003) state: "Much regression output serves little purpose in medical
research publication; this usually includes the intercept coefficient, R2, log likelihood, standard
errors, and P values. Estimates of variance explained (such as R2, correlation coefficients, and
standardized regression coefficients (sometimes called effect size) are not useful measures of
causal associations or agreement and should not be presented as the main results of an analysis.
These measures depend not only on the size of any biological effect of an exposure, but also on
the distribution of the exposure in the population. "
Page 22

-------
3.5 Study Design
As noted earlier, EPA concluded that the study objectives were best described in the
protocol document (Garabrant et al, 2005) which indicated that the primary objective was to
describe the pattern of serum dioxin levels among adults. In this regard the study design was
adequate. The population based study design resulted in credible estimates of the distribution of
serum levels in adults in the MS study areas. The sample from the JC reference area produced a
distribution of adult serum levels that was shown to be consistent with the national NHANES
survey (Hedgeman et al., 2009). The four survey areas identified in the MS area appear to be
appropriately selected to represent the areas most likely impacted by activities of the Dow
facilities. The analytical matrices evaluated in this study also appear appropriate. Blood is the
best indicator of long term exposure to the dioxin-like compounds and soil and dust are also long
term reservoirs for these compounds. Also, soil and dust were the matrices of concern with
regard to deposition of contaminated floodplain sediments. The diets of the population in the
five study areas were not sampled. However, diet is the primary source of dioxin exposures for
the general population, and dioxin levels in food are declining. Even if diets were sampled in the
UMDES, the data would not represent long term dietary exposure. Even representing current
dietary levels would require collecting multiple meal samples over a long term for each
participant, significantly increasing the cost of the study. UM chose instead to assess dietary
issues via a questionnaire which was a practical choice.
The second part of the objective statement in the protocol document (Garabrant et al,
2005) was to understand the factors that explain variation in serum dioxin, furan and PCB levels.
EPA concludes that the study design may have been sufficient to meet this objective for some
factors, but may not have been sufficient for other factors, as will now be discussed.
With regard to general population exposures, the UMDES identified the key demographic
factors that have been known to be associated with dioxin blood serum levels: age, sex, body
mass index (BMI), breast-feeding, and so on. The NHANES studies on dioxin in blood have
always found a relationship with age (higher serum concentrations tend to be associated with
older individuals), while others (Lorber, 2002) have also studied temporal trends in dioxin
exposure and concluded that higher levels of dioxins in the environment (and food) in the middle
decades of the 20th century led to higher body burdens in current older individuals. Lakind et al
(2001) and others have studied the decline in the mother's body burden of dioxins during the
course of breast-feeding. Higher consumption of animal food products might lead to higher
body weights (and higher BMI) and subsequent higher dioxin body burdens.
However, the population-based statistical sampling design of the UMDES survey does
not guarantee that the "high end" of the population within the studied areas was adequately
represented. The discussion below evaluates how well the UMDES design addressed high end
exposures related to soil and activities such as fishing, hunting and gardening.
To fully evaluate the relationship between soil dioxin and serum dioxin, the study should
represent properties containing high end soil contamination levels. UM recognized this issue and
took some steps to address it. This involved identifying two areas that contained soils most
Page 23

-------
likely impacted by Dow activities (the FP and the plume area) and oversampling these areas.
The success of this approach is difficult to judge but several discussion points are presented
below for consideration.
There have been other soil surveys which suggest (but do not prove) that the UMDES did
not identify the highest, or enough, properties with high dioxin soil levels. A study referred to as
Dow's Middle Tittabawassee River (MTR) remedial investigation measured dioxin levels in
MTR flood plain soils (ATS, 2009). This MTR investigation targeted areas most likely to have
impacts from flooding and the deposition of river sediments contaminated with dioxins. It is
expected that this subarea is part of the 100-year floodplain of the UMDES Floodplain study
area. The two studies had soil sampling methodology differences (depth of sampling to describe
the uppermost "surface" sample, for example). EPA has not conducted a detailed evaluation of
the MTR study. However, visual inspection of satellite photographs and maps included in the
study are informative. First, people living in the Tittabawassee FP generally reside on higher
ground which is less often flooded than lower areas. Second, it appears that the majority of soil
samples of the MTR study were collected in the river or near it in low-lying areas that were
generally distant from residences. The maps also show that the higher concentrations were
generally found in these low areas of little or no population near the river. Many of these
samples were over 1000 ppt and some over 10,000 ppt.
However, the MTR also included several samples that were classified as having been
obtained from "residential" parcels of land. The maps did show these samples. The MTR did
not have a separate analysis of these samples, but all MTR samples are available in spreadsheet
format. This spreadsheet was obtained by the Michigan Department of Environmental Quality
(MDEQ). They separated out these residential samples from the full MTR dataset, and
compared trends from this subset with findings of the UMDES (from an email dated 4/23/2008
from D. Mackenzie-Taylor to several individuals, supplied to EPA by Deborah MacKenzie-
Taylor, MDEQ):
The UMDES estimates that 7% of the 203 properties with soil samples in the
floodplain group (about 14 properties) have any soil samples exceeding 1,000 ppt
dioxin andfur an total toxic equivalency (TEQ). In contrast, the MTR residential
top interval soil sample set had 47-67% (27properties with only nearest to the
river locations and 38properties with all locations out of 57 residential
properties) of the residential properties with any top interval location over 1,000
ppt estimated TEQ (ETEQ). In addition, the means and medians of the two data
sets are very divergent. The mean values of the UMDES data nearest to the river
samples are 237-285ppt TEQ, while the MTR mean value nearest to the river is
1,300 ppt ETEQ. The medians of the UMDES data nearest to the river samples
are 11-12.7 ppt TEQ, compared to the MTR median nearest to the river samples
of 750 ppt ETEQ. The median of the topmost interval MTR residential samples is
780 ppt ETEQ.
Visually inspecting the MTR maps, EPA also noted that they appear to show a much
higher percentage of samples over 1000 ppt TEQ than 7%. Again, the designs of the UMDES
and the MTR are very different, with different objectives and goals. Also, the UMDES
Page 24

-------
researchers have rebutted these comments with an analysis of their own which showed that,
when looking at analogous subsets of the data, the two studies yielded similar percentages of
properties over 90 and 1000 ppt TEQ (May 7, 2008 letter from Garabrant and Franzblau to
MacKenzi e-T ay 1 or).
Another effort which sought to find residential properties that may have elevated dioxin
concentrations was from a program conducted by the US EPA Region 5, who has investigated 20
"exposure units" along the Tittabawassee and Saginaw Rivers. Perhaps the most impacted of
these units was at the confluence of the two rivers, on Riverside Boulevard. Eleven properties
were sampled and elevated dioxin concentrations were found on all properties, many between
1000 and 10,000 ppt TEQ and some exceeding 10,000 ppt TEQ. Another exposure unit was
West Michigan Park, which sampled some park area and 10 residential properties. Soil dioxin
concentrations between 1000 and 10,000 ppt TEQ were observed throughout this exposure unit.
Additional points to consider on how well high end soils are discussed below.
The median level of dioxins found in the surface soil samples was 11 ppt TEQ in the FP, 58
ppt TEQ in the Plume area and 4 ppt TEQ in JC (Demond et al, 2008). This suggests that the
plume area was clearly elevated over JC, but much less elevation was seen for FP relative to JC.
Half of the FP properties included in the study have soils less than the median value of 11 ppt
TEQ, a relatively low value compared to the concern levels of 90 and 1000 ppt TEQ.
The maximum soil level found by UMDES in the FP was 7,258 ppt TEQ (Demond et al,
2008). ATS (2009) shows multiple soil samples over 10,000 ppt TEQ, although again the
location of these elevations in low-lying unpopulated areas must be considered. Also, Garabrant
and Franzblau (May 7, 2008 letter to MacKenzie-Taylor) stated that the greater sampling depth
used in the MTR study (1.1 feet vs. 6 inches in MDES) would contribute to the higher levels.
The population-based, probability sampling design of the UMDES identified what is
expected to be a reasonable result: that 7% (with appropriate error bounds) of properties on the
flood plain contain soil concentrations above 1000 ppt TEQ. Given they sampled 203 properties
with each having a sample weight of 6, one can surmise there were 1200 properties in all, 84
(plus or minus) of which had soil concentrations above 1000 ppt TEQ. They only sampled 14 of
them, leaving perhaps about 70 residences not sampled, and given that their sampling scheme
was population-based, they would have trended towards sampling locations of higher population
density. If it is possible to generalize based on a visual inspection of maps in the MTR study,
more densely populated areas are located in higher elevations, thus likely to be the less often
flooded areas with lower concentrations.
EPA believes that further evaluation of the UMDES data may provide additional insight
with regard to the impact that soil concentration can have on blood serum concentration. This is
because other factors (primarily diet) are likely to overwhelm any contributions to serum from
soil at low levels, making it very difficult to clearly identify relationships. For example, a
comparison of the serum levels of the individuals living on soils over 1000 ppt TEQ to those
living on background soils may be informative. In response to a request from EPA, UM
identified the study participants living on soils over 1000 ppt TEQ in the age versus serum plot
from Garabrant et al., 2009b (forwarded by email from Garabrant to Frithsen on September 11,
Page 25

-------
2009). This plot and further analysis of these data are presented in Appendix B. The plot shows
that most of these participants had serum concentrations above the average levels for their age
group. Appendix B also presents a separate serum versus age regression for these 23 individuals.
Considering the need for further data analyses and the uncertainties about whether the
high end soils were adequately represented in the study, it is difficult to reach a definitive
conclusion about the adequacy of the design to meet the second objective with regard to soil
dioxin concentrations.
The potential for high end exposure is not only related to living on properties with high
dioxin soil concentrations, but also to behaviors that lead to more exposures, such as fishing,
gardening, consuming local game, raising animals used for home (or local) food consumption,
and others. A fish consumption survey for the region has been conducted (MCDH, 2007), but it
is unclear whether any of the "high end" patterns identified in this survey were adequately
captured in the UMDES survey. The UMDES study design did not include any features aimed at
ensuring that individuals engaged in these activities would be adequately represented. Also, the
possibility exists that there simply are few high consumers of local fish or other individuals with
high end behaviors that could be studied. It is noteworthy that UMDES researchers did identify
one farmer who raised cattle for home consumption and one home gardener on a property with
elevated dioxins, and both showed elevations in blood serum dioxins (Garabrant et al., 2009b).
It should also be noted that the random population based survey design provided some
unanticipated but useful findings. These involved the identification of several unexpected dioxin
exposure scenarios. One was the discovery of elevated soils outside the floodplain due to
movement of soils from the floodplain to residences for fill or other purposes. Second was the
discovery of elevated serum levels that appear to be associated with the activities of a ceramic
clay hobbyist. Third was the discovery of several incidents of elevated soil and dust levels due
to the use of PCB containing products.
In summary, EPA concludes that the UMDES study design was well suited for meeting
the objective of describing the distribution of dioxins in the blood in the identified populations.
On the other hand, EPA believes that the UMDES design may only have been partially
successful in meeting the second objective, which was to identify factors associated with serum
dioxin and to quantify their degree of influence. Key demographic factors that are known to be
associated with dioxin body burden in the general population, including primarily age but also
BMI, whether or not a woman breast-fed, and other factors, were identified in the UMDES.
However, it is not clear whether individuals at the "high end" of the population, whose behaviors
or residence location may lead to body burdens elevated over the general population, were
adequately characterized. These include, for example, individuals living on properties with high
dioxin soil concentrations, individuals fishing and consuming fish from the impacted water
bodies, farmers and gardeners who produce a portion of their food on their own property, and so
on. Because of the importance of soil in this region, discussions above focused mostly on this
factor. Evidence was presented that there were a fair number of residences, perhaps about 70
residences in the FP, that contained dioxins above 1000 ppt TEQ that were not sampled. Given
evidence presented above, one can at least hypothesize that UMDES under sampled the most
heavily contaminated properties. This would have limited their ability to see relationships
Page 26

-------
between dioxins in soil and dioxins in blood serum. Still, data supplied by UMDES researchers
for the individuals living on properties with soil concentrations above 1000 ppt TEQ did appear
to suggest that these individuals had serum levels elevated over the central values for their ages
(see Appendix B). To date, the UMDES researchers have not studied this specific
subpopulation or similar subpopulations that may exist within the UMDES data set. Finally, one
must consider the possibility that key subpopulations, such as high consumers of local fish, game
or farm animals, simply could be very few in number or may not exist within the study areas,
which would certainly limit the ability of the UMDES researchers to estimate relationships with
these subpopulations.
Page 27

-------
SECTION 4: EVALUATION OF MAJOR CONCLUSIONS
The major conclusions and findings from the UMDES were spread out in the various
reports and journal articles. In this section, these conclusions/findings are first described in
general terms, but then quotes from the key literature articles are provided so that the exact
phrases developed by the UMDES researchers are identified. Then, a comment is provided on
EPA's judgment on the statements made within these publications.
4.1 Nonserum Correlations Involving Soil and Dust
Soil Versus Study Area: Soil from properties in the Midland/Saginaw study areas contain
higher concentrations of dioxin than soils in the reference area, the Jackson/Calhoun area.
Specific statements supporting this finding include:
Soil from properties in the Floodplain, Near Floodplain, Midland Plume, and
Other Midland Saginaw is more contaminated with dioxins than soil in
Jackson/Calhoun (UM, 2006).
More properties in Midland Saginaw than in Jackson/Calhoun have soil TEQ
levels at or above 90 parts per trillion, an important level set by the State of
Michigan (UM, 2006).
Levels of dioxins in soil in Jackson/Calhoun are similar to levels in soil across
Michigan based on a small number of soil samples collected by the Michigan
Department of Environmental Quality (UM, 2006).
Soils in the Floodplain and Near Floodplain show patterns of dioxins suggestive
of Dow's historical discharges. These patterns are not seen in Jackson/Calhoun
(UM, 2006).
On the basis of previous sampling campaigns, it was expected that the plume and
floodplain areas had higher levels than background; this study indicates that the
difference is statistically significant and is not based on convenience or targeted
sampled. The higher levels in the near floodplain and other Midland Saginaw
samples were unanticipated and suggest either the movement of contaminated
soils to those areas or the presence of additional sources other than the
Tittabawassee river or the Dow incinerator.
The mean TEQdfp-1998 of the HP (sic, house perimeter) 0-1 in. soil samples from
Jackson and Calhoun Counties of 8.3 pg/g is similar to the mean value computed
for the residential properties in the Denver Front range study of 8.8 pg/g TEQdfp-
1998 suggesting that Jackson and Calhoun Counties are a suitable background
comparison group (Demondet a/., 2008).
Page 28

-------
This study employed a probabilistic sampling design that allowed for a statistical
determination as to whether the soil concentrations in the target areas were
above background. On the basis of this analysis, it was determined that the
geometric mean soil concentrations in all four target areas in Midland, Saginaw,
and Bay Counties, not just the Tittabawassee River floodplain and the incinerator
plume in the City of Midland, were elevated relative to background. In addition,
the probabilities of a soil sample having a concentration above the 75th and 95th
percentiles of background were statistically greater in these areas (Demand et a/.,
2008).
EPA COMMENT: This conclusion is supported by the data generated in UMDES. The medians
of dioxin TEQ in the surface inch of soil in JC is 4 ppt, while for the FP it was 11 ppt and for the
MP it was 58 ppt. It was also supported by the percentages of properties above the MDEQ soil
criteria of 90 ppt TEQ: 36.5 % in floodplain 35.8% in plume, 9.7% in near floodplain, 1.7% in
other Midland/Saginaw and 0.3% in Jackson/Calhoun (Demond et al., 2008). The finding
relating to the pattern of dioxins in soil is supported by analysis showing that the congeners that
contribute most to soil TEQs in the FP and NF were 23478-PeCDF, 2,3,7,8-TCDF, 2,3,7,8-
TCDD, 1,2,3,7,8-PeCDD and 1,2,3,4,7,8-HxCDF (Demond et al., 2008). This pattern is similar
to the pattern found by the Michigan Department of Environmental Quality in Tittabawassee
River sediment which is believed to be the result of Dow's historic discharges into the river
(UM, 2006). The congeners that contribute most to soil TEQs in the Jackson/Calhoun were PCB
126, 1,2,3,7,8-PeCDD, 1,2,3,4,6,7,8 HpCDD and 2,3,4,7,8-PeCDF (Demond et al., 2008).
House Dust versus Study Area: Dusts from households in the Midland Plume area contain
higher dioxin levels than dusts from households in the other study areas. Statements in the key
documents which address this relationship include:
Household dust from properties in the Midland Plume is more contaminated than
household dust from properties in any other region (UM, 2006).
Living in the Midland Plume region (compared to living in the Jackson/Calhoun)
was associated with higher 2,3,7,8-TCDD concentrations (Knutson et al., 2007).
EPA COMMENT: This is supported by a comparison of median levels in household dust in ppt
TEQ: Midland Plume — 35, Other Midland/Saginaw - 19, Floodplain - 17, Jackson/Calhoun -
14, Near Floodplain - 12 (UMICH, 2006). Also, it is supported by the percent of homes with
dust levels > 35 ppt TEQ: Midland Plume - 50, Other Midland/Saginaw - 21, Floodplain - 27,
Jackson/Calhoun - 26, Near Floodplain - 11 (UM, 2006). Although Knutson et al. (2007) report
a positive correlation between dust levels in Midland Plume versus JC for 2,3,7,8-TCDD, they
report a negative correlation for TEQs.
Soil versus House Dust: Levels of dioxins in household dust are generally higher than levels of
dioxins in soil around the house. Statements in the key documents which address this
relationship include:
Page 29

-------
Levels of dioxins in household dust are higher than levels of dioxins in soil in
most of the study regions, suggesting that there are sources other than soil for
dioxins in household dust (UM, 2006).
For both the TEQ and 2,3,7,8-TCDD models, soil concentration from around the
house was the most important predictor in house dust concentrations - it
explained the most variation in the models. Higher soil concentrations were
associated with higher household dust (Knutson et a/., 2007).
EPA COMMENT: This is supported by presented linear regression analysis showing that soil
around the house was found to be the most important predictor of house dust levels (Knutson et
al., 2007). However, Franzblau et al. (2009) suggests other factors can also be important based
on an investigation of homes with dust TEQ levels more than 2.5 standard deviations over the
mean. Out of the 20 homes that met this criterion, only two appeared related to outdoor soils
(based on profile comparisons). UM (2006) notes that commercial products can release dioxins
indoors, such as wood preservatives and old electrical appliances. Franzblau et al. (2009)
concluded that PCBs in an old carpet pad were the source in one house with high levels of PCBs
in the dust. Although not noted in the UMDES reports, other possible indoor sources include
cooking, wood combustion, track-in of soils from locations outside the residence and blow-in of
dust which may be enriched in contaminants relative to bulk soils.
4.2 Serum Correlations with Soil and Dust
Serum versus Soil: Many statements appear in the various documents relating to this key
relationship. Generally, weak relationships between various measures of soil and serum dioxin
levels were noted but the overall conclusions in the more recent documents imply that there is
not a meaningful relationship between these two key factors. Statements in the key documents
which address this relationship include:
People who have higher levels of dioxins in their soil have a higher TEQ (total
dioxin-like activity) and higher levels of some specific dioxins in their serum (UM,
2006).
In some cases there is a direct relationship between higher levels of dioxins in soil
and higher levels of dioxins in people's serum. This relationship is small and
applies to some, but not all of the specific dioxins (UM, 2006).
The region in which people live, soil contamination, and household dust
contamination combined account for about one percent of the variability in levels
of 2,3,7,8-TCDD, 1,2,3,7,8-PeCDD, PCB-118, and PCB-126 in people's serum.
For the other types of dioxins and the TEQ, region, soil and household dust
contamination combined account for less than 0.2 percent of the variability in
levels of dioxins in people's serum (UM, 2006).
Page 30

-------
Four of the seven specific types of dioxins examined showed no relationship
between levels of dioxins in soil and levels of dioxins in serum. People who have
higher levels of 2,3,4,7,8-PeCDF, 1,2,3,7,8-PeCDD, 1,2,3,6,7,8-HxCDD, and
PCB-156 in their soil do not have higher levels of these dioxins in their serum
(UM, 2006).
Three of the seven specific types of dioxins plus the TEQ showed some
relationship between levels of dioxins in soil and levels of dioxins in serum.
People who have higher TEQ levels and higher levels of 2,3,7,8-TCDD, PCB-118,
and PCB-126 in some of their soil samples have higher levels of these dioxins in
their serum (UM, 2006).
We looked at the highest level of dioxins in soil out of all the soil samples taken
on each property. This value, the highest soil value measured on the property, is
related to the TEQ level in people's serum. For a person living on a property with
a soil TEQ of 1,000 parts per trillion, the serum TEQ is about 0.7parts per
trillion higher than that for a person living on a property with a soil TEQ offour
parts per trillion (which is the median soil TEQ in Jackson/Calhoun) (UM, 2006).
We also looked at levels of dioxins in soil from vegetable andflower gardens.
People whose gardens have higher levels of 2,3,7,8-TCDD or higher levels of
PCB-118 have higher levels of these dioxins in their serum. For a person living
on a property with a garden soil 2,3,7,8-TCDD level of 22 parts per trillion
(which is the median level of soil from homes in the Midland Plume), the serum
2,3,7,8-TCDD level is 0.7 parts per trillion higher than that for a person living on
a property with garden soil at the median level of Jackson/Calhoun (0.1 parts per
trillion). People who have higher levels of PCB-126 and PCB-118 in soil from
around the house have slightly higher levels of these two specific dioxins in their
serum. For example, a person whose soil around the house has a PCB-118 level
of1000 parts per trillion has a less than 1 percent increase (about 18 parts per
trillion) of this chemical in their serum compared to someone whose soil has a
lower level of this chemical (UM, 2006).
Soil and household dust dioxin content explained only a small part of the
variation in serum dioxin levels: 0.5% for TCDD, 1% for PCB-126, and < 0.01%
for the other congeners (Garabrant et al., 2009b).
A principal focus of our study was to assess activities that might involve contact
with contaminated soils, river sediments, and household dust. These included
living on contaminated soil, living with contaminated household dust, and
pursuing activities in the contaminated water bodies and floodplain (boating,
swimming, picnicking, hiking, etc.). We found little evidence in the general
population that these activities were associated with increased serum dioxins
(Garabrant et al., 2009b).
Page 31

-------
The TCDD concentration in garden soil was statistically significantly associated
with higher serum TCDD. This association was due to a single observation that
was highly influential (Garabrant et al., 2009b).
The house perimeter top 1 in. (2.5 cm) soil PCB-126 concentration was
associated with serum PCB-126 (parameter estimate = 1.001 ppt PCB-126 in
serum/ppt PCB-126 in soil; p = 0.0002). This association was due to two
influential observations. When we excluded these, we found no statistically
significant relationship (Garabrant et al., 2009b).
With the exception of results described above, no soil variable was statistically
significantly associated with serum dioxins (Garabrant et al., 2009b).
There are statistically significant positive associations between serum dioxin
concentrations and soil dioxin concentrations for TEQ, 2,3,4,7,8-PCDF, 2,3,7,8-
TCDD and PCB-126 (Chang et al., 2007).
The variation in serum dioxin levels explained by the soil parameter did not
exceed 1.3% for any soil parameter (Chang et al., 2007).
In most instances, soil concentrations had little effect on the serum levels for
TEQ, 2,3,7,8-TCDD, 2,3,4,7,8-PCDF, or PCB-126. The one exception to this was
the relationship between 2,3,7,8-TCDD in serum and garden soil (soil contact
zone), for which serum TCDD increased by approximately 50% as soil TCDD
increased by 22 ppt (Chang et al., 2007).
Soil and household dust were not important contributors to serum TEQ, PCDDs,
PCDFs, or PCBs (Garabrant, 2008).
EPA COMMENT: EPA offers several comments on this finding:
This overall general finding is supported by the statistical analysis that is presented in
Garabrant et al (2009b). When measures of household dust and soil concentrations were
combined, these explained 0% of the variation for TEQ, 1,2,3,7,8-PCDD, 1,2,3,6,7,8-PCDD, and
2,3,4,7,8-PCDF. It explained only 0.51% of the variation for 2,3,7,8-TCDD and 0.96% of the
variation for PCB 126.
This general finding pertains to the full adult populations sampled in the 5 areas of the
study, or as noted in a quote from Garabrant et al (2009b), "We found little evidence in the
general population that these activities were associated with increased serum dioxins."
However, as discussed earlier, the study may not have adequately represented high end soil
levels. Therefore it is uncertain if this general statement applies to individuals living on these
properties containing the highest levels of dioxin. Even more importantly, EPA believes that
making these statements without also noting that it does not pertain to children is misleading.
Page 32

-------
The messages have not been consistent among the documents EPA gathered for this
review. The 2006 report did have general conclusions, along with some statistical results, that
suggest that relationships exist. Like in other cases, PCB 118 along with PCB 126 were
identified as congeners for which relationships exist in the 2006 report, but only PCB 126 was
identified in later documents. The poster by Chang et al. (2007) did note weak associations
between certain congeners in soil and serum concentrations. Interestingly, TEQ and 2,3,4,7,8-
PCDF, noted as having associations in the poster were also noted as explaining 0% variation in
the Garabrant et al. (2009b) journal article.
EPA is not convinced that the statistical analysis and results in Garabrant et al. (2009b)
provide the final word on the UMDES data base. A statistician, Dr. John Kern, contracted by the
Michigan Department of Environmental Quality, has argued that the overall statistical model
employed by the UMDES research team is not the best model for use at this site. Rather, he
proposes a model which includes a "base" set of parameters already known to be related to
dioxin body burden (age, diet, etc), and to this base, add only one or two additional parameters in
independent analyses to see the influence of these other parameters on body burdens. He calls
this approach the "base + 1" approach and EPA sees potential merit in that approach. EPA asked
for and received additional data regarding serum levels of individuals living on properties with
soil over 1000 ppt TEQ., These data were discussed in Section 3 and in more detail in Appendix
B. These data suggest that these individuals had serum levels elevated over the central values for
their ages The UMDES team has not studied this specific subpopulation or others that may exist
that could be characterized as "high end" subpopulations. With further study, EPA believes that
additional relationships could be unearthed within the UMDES data base.
An independent pharmacokinetic modeling exercise was conducted by EPA as part of the
evaluation of the UMDES study (details are provided in Appendix A and further comments in
Section 5). The purpose of the exercise was to model the difference between a population
having only background exposures, dominated by food and having soil-related exposures based
on a soil concentration of 11 ppt TEQ, and another with the same background exposures except
the soil-related exposures were based on soils at 223 ppt TEQ. This level corresponds to the 95th
percentile of surface soils at residences in the FP. Those exposed to the higher soil had a body
burden that was only 3 ppt higher than those exposed to background soils. This suggests that
perhaps 95% of all adult individuals in the FP would have serum level increases less than 3 ppt
TEQ due to soil ingestion. This would be a small increase for most individuals considering that
the median serum level was 23.2 ppt TEQ. Therefore, it is not unexpected and consistent with
this external modeling exercise that UMDES would find little influence of soil dioxins on their
measured dioxin body burdens.
The same PK analysis described above showed much more impact to a child. A child of
5 sampled in the early 2000s would have a 6 ppt TEQ difference if living on soils at 223 ppt
TEQ: their body burden is modeled to be 17 ppt TEQ at background soils but 23 ppt TEQ at 223
ppt TEQ soils. The modeling also showed a difference of nearly 19 ppt if living on soils at 1000
ppt TEQ: 17 versus 36 ppt TEQ.
Soils and sediments are recognized reservoirs for dioxins that result in ongoing impacts
to food chains. While the direct pathway of soil ingestion may alone explain only a small part of
Page 33

-------
body burden, certainly soil and sediment dioxins are the sources for animal concentrations of
dioxins, and, as found worldwide in the study of exposure to dioxin, there is a direct correlation
with animal food consumption and dioxin body burdens.
Serum versus House Dust: Little association was found between household dust concentrations
of dioxins and serum. Statements in the key documents which address this relationship include:
People who have higher levels of dioxins in their household dust have higher
levels of one of the specific dioxins (PCB-118) in their serum (UM, 2006).
Soil and household dust dioxin content explained only a small part of the
variation in serum dioxin levels: 0.5%for TCDD, 1%for PCS-126, and < 0.01%
for the other congeners (Garabrant et a/.. 2009b).
We found no statistically significant associations between household dust dioxins
and serum dioxins (Garabrant et a/.. 2009b).
The lack of evidence that contaminated household dust was associated with serum
dioxin concentrations is also reassuring in that even though soil contamination
contributes to household dust contamination, it does not contribute appreciably to
the body burden of dioxins (Garabrant et a/., 2009b).
EPA COMMENT: Here is yet another example where the two key documents, the 2006 report
(UM, 2006) and the 2009 journal article (Garabrant, 2009b), have slightly different results. The
2006 report suggested that a relationship might exist between PCB 1 18 levels in serum and
household dust, whereas Garabrant et al. (2009b) suggest that the relationship is with PCB 126.
Generally, the findings on the influence of household dust and serum concentrations, and
between household dust and outdoor soil, appear supported by the analyses presented in the
journal articles. In Garabrant et al. (2009b), combining soil and dust together showed a possible
relationship only for PCB 126, with that congener in dust explaining 0.96% of the variation in
serum PCB 126. The only other congener which showed a relationship was with 2,3,7,8-TCDD,
which explained 0.51% of the variation. However, this finding appeared to have been driven by
a single household showing a high concentration of 2,3,7,8-TCDD in garden soil. Without this
data point, the relationship was not statistically significant.
4.3 Serum Correlations with Other Factors
Serum versus Study Area: Higher concentrations of dioxins were found in the serum of
residents living within the four study areas comprising the MS region as compared to the
reference JC area. The following statements have been made supporting this finding:
People who live in some regions ofMidland/Saginaw have higher levels of some
dioxins in their serum than do people in Jackson/Calhoun (UM, 2006).
Page 34

-------
The median serum TEQdfp-2oos varied by 6.4 ppt among the five populations, was
significantly elevated in the floodplain compared with the referent population
(23.2 ppt us. 18.5 ppt, p >0.01), and was lowest in the Midland plume population
(16.8ppt). Maximum TEQdfp-2oos concentrations varied by more than 130ppt
across the five populations, with the most elevated serum concentration in the
floodplain (211 ppt) and the near floodplain (154 ppt) populations, and the lowest
maximum serum concentration in the Midland plume (78.5 ppt) (Hedgeman et al.,
2009).
EPA COMMENT: This conclusion is supported primarily by analysis provided in Hedgeman et
al, 2009. This publication provides key percentiles for the full distributions in the two regions.
The analysis showed that the distributions are fairly similar up until the 50th percentile (the
median), but that the MS concentrations (combining all results from the 4 study areas comprising
the M7 region) then were higher than the JC region. Specifically, the medians for the two
populations were 20.7 and 18.5 pg/g TEQ lipid weight (ppt lwt) for the MS and JC populations,
respectively. The 75th, 95th, and maximum concentrations were 32.3, 62.9, and 211 ppt TEQ
lwt for MS and 25.3, 46.5, and 109 ppt TEQ lwt for JC.
Serum versus Living in the MS Region: Living in the MS region during the middle decades of
the 20th century was associated with elevated serum dioxins. Statements in the key documents
which address this relationship include:
People who lived in Midland Saginaw between 1940 and 1959 have higher levels
of 2,3,7,8-1'( 1)1) in their serum. We believe this is suggestive of Dow's operations
during that time period. Other types of dioxins and TEQ do not have this kind of
relationship. Other time periods do not have this relationship (UM, 2006).
Residence in Midland and Saginaw counties was examined in three different
historic periods, 1940-1959, 1960-1979, and 1980 2005, with the duration of
residence during each period handled as a continuous variable. Residence in the
area during 1960-1979 was positively associated with TEQ, TCDD, and
1,2,3,7,8-PeCDD, but not with other congeners... .Neither residence in 1940-1959
nor residence in 1980 2005 was associated with any congener (Garabrant et al.,
2009b).
After adjustment for all other factors in the model, including historical residence
in Midland or Saginaw counties as described above, living currently in the
floodplain or near the floodplain of the Tittabawassee River or in the plume area
downwind of the Dow facility was not associated with either TEQ or any of the
PCDD or PCDF congeners. We found a significant positive association for serum
PCB-126 and living currently in the near floodplain compared with living in
Jackson/Calhoun counties. However, we are not aware that PCB-126 was
manufactured or used at Dow (Garabrant et al., 2009b).
Page 3 5

-------
EPA COMMENT: This is an instance where the two primary reports we reviewed have a
different message altogether. It is noted that the statement attributed to the 2006 report (UM,
2006) was not included as a primary finding - all other findings from the 2006 report originated
on page 2 of the summary whereas the quote above was from page 10 in the body of the report.
Based on other information, we believe that the later findings are more supported and we cannot
understand this disparity in the two documents. New work that was presented at the Dioxin 2009
conference (Hong et al., 2009; Garabrant et al., 2009d) concluded that the number of years living
in the MS area between 1960-79 resulted in elevations in certain dioxin measures, and that this
effect is most pronounced with 2,3,7,8-TCDD. The UMDES researchers hypothesize that this is
due to emissions from the Dow incinerator during those years. In Garabrant et al. (2009b), this
factor was part of a category called "residence factors" (defined as, "Number of years in MS in
1960-1979", but also including "reside in Near Floodplain"), and it was shown that these factors
explained only about 0.6% of TEQ, but 3.4% of 2,3,7,8-TCDD variation.
Serum versus Age: A primary finding from this study is that age is positively associated with
serum dioxin levels. The following statements have been made supporting this finding:
The most important factor related to levels of dioxins in people's serum is age.
We found that older people have higher levels of dioxins in their serum. This
effect was found in both Midland Saginaw and Jackson/Calhoun and has been
found in other studies for people across the United States (UM, 2006).
The largest part of the variation (31-44%) was explained by what we have labeled
demographic factors: age, age2, sex, BMI, BMI loss in the past 12 months, breast-
feeding, number of pregnancies, race, smoking, and interaction terms. (The
amount of variation associated with age alone is not provided.) (Garabrant et al.,
2009b)
Age was strongly associated with serum TEQ and with all serum congener
concentrations. In addition, age2 was negatively associated with serum TEQ,
1,2,3,6,7,8-HxCDD, and 2,3,4,7,8-PCDF, indicating that the relationship between
age and serum concentrations was slightly less than exponential for these dioxins
(Garabrant et al, 2009b).
TEQ increased more dramatically among females than among males (Various
statements similar to this TEQ statement were made on a congener-specific basis)
(Garabrant et al, 2009b).
EPA COMMENT: This finding is supported by the analysis conducted to date. However, while
age may be the most important variable relating to serum levels, plots of serum level versus age
show substantial scatter or variability, yielding a significant, but weak positive correlation. The
R2 for age alone was not reported. However, the lumped demographic factors have an R2 of
0.396 for serum TEQs (Garabrant et al ., 2009b), so age by itself would be lower. Thus, the
statements that age is a strong predictor of serum levels appear overstated.
Page 36

-------
Serum versus Fish: Aside from the demographic factors, perhaps the factor otherwise most
correlated with serum measurements was consumption of fish. Various statements of findings
addressed issues including: where the fish came from (store bought or from the impacted water
bodies), what kind of fish, whether the individuals recreationally fished, and so on. Specific
statements relating to the relationship of serum levels with fish consumption and/or recreational
fishing activities, include:
Eating fish, no matter whether it is sport-caught, store-bought, or from a
restaurant, is related to higher levels of dioxins in people's serum. This applies to
fish both from the contaminated area in Midland Saginaw andfrom outside the
contaminated area (UM, 2006).
People who eat fish from the Tittabawassee River, Saginaw River, and Saginaw
Bay have higher levels of dioxins in their serum than people who do not eat fish
from these areas. Most of these people live in Midland Saginaw. For example, 65
percent ofpeople who live in the Floodplain have eaten fish from the
Tittabawassee River, Saginaw River, or Saginaw Bay at some point during their
lives, compared to only 8 percent in Jackson/Calhoun (UM, 2006).
Fishing and fish consumption explained 0.5 3% of the variation in serum dioxin
concentrations (Garabrant et al., 2009b).
Consumption of fish after 1980, regardless of their source (store-bought,
restaurant, or sport caught), was positively associated with TEQ, 111)1), and
PCB-126. We found no positive associations between consumption offish from
the contaminated areas (Tittabawassee River, Saginaw River, or Saginaw Bay)
and serum PCDD or PCDF levels. There were equivocal results for eating fish
from the Saginaw River or Saginaw Bay in the past 5 years (positive association
for consumption less than once per month, but a negative association for
consumption of more than once per month). We found positive associations for
serum 1,2,3,6,7,8-HxCDD and 2,3,4,7,8-PeCDF and consumption of walleye or
perch from other sources (other water bodies, restaurants, and grocery stores).
All other variables related to consumption offish from the contaminated water
bodies were either significantly negatively associated with serum dioxins or not
associated. Overall, consumption of fish was positively associated with increased
serum levels of a number of dioxins, but the findings were not related to fish from
the contaminated areas, with the exception of PCB-126 (Garabrant et al., 2009b).
Fishing in the Saginaw River and Saginaw Bay after 1980 was positively
associated with serum TEQ, 1,2,3,7,8-PeCDD, and 2,3,4,7,8-PeCDF, and fishing
in the Tittabawassee River between 1960 and 1979 was positively associated with
1,2,3,6,7,8-HxCDD (Garabrant et al., 2009b).
The overall pattern offindings in this population suggest little contribution from
exposures in Midland and Saginaw counties during the last 25 years to serum
Page 37

-------
dioxins from either contaminated soil, household dust, consumption of fish and
game from contaminated areas, or recreational habits (Garabrant etal., 2009b)..
Our model suggests that fishing in the Saginaw River or Bay at least once per
month since 1980 multiplies the serum 2,3,4,7,8-PeCDF concentration by a factor
of 1.3 (an increase of 1.6ppt compared with the median level of 5.4 ppt in
Jackson and Calhoun counties). In contrast, consumption of fish from the
contaminated areas was not associated with increased serum 2,3,4,7,8-PeCDF. It
is possible that a larger sample of people who consumed contaminatedfish
regularly might have shown an association (Garabrant et a/., 2009b).
EPA COMMENT: The general comment that there is a relationship between consumption of
fish and serum dioxin levels is generally supported by the several findings that were outlined in
the primary publication detailing the statistical correlation analysis of the data set (Garabrant et
al, 2009b). While the correlation was noted, the statistical relationship was weak as overall
fishing and fish consumption appeared to explain only about 1% of the variation in serum TEQ
levels, and less than 3% for any of the individual congeners (Table 1 in Garabrant et al, 2009b).
There appeared to be some slight differences and inconsistencies between the 2006 report (UM,
2006) and the 2009 journal article (Garabrant et al., 2009b) which are summarized here. The
2006 report (UM, 2006) included the conclusion: "People who eat fish from the Tittabawassee
River, Saginaw River, and Saginaw Bay have higher levels of dioxins in their serum on average
than people who do not eat fish from these areas." However, the 2009 publication states, "We
found no positive associations between consumption of fish from the contaminated areas
(Tittabawassee River, Saginaw River, or Saginaw Bay) and serum PCDD of PCDF levels"
(Garabrant et al, 2009b, p. 822). Also, the 2006 report seemed to promote the notion of a clear
relationship between fish consumption and serum levels, while the later EHP report was more
equivocal about this, as they stated, "The overall pattern of findings in this population suggest
little contribution from exposures in Midland and Saginaw counties during the last 25 years to
serum dioxins from either contaminated soil, household dust, consumption of fish and game
from contaminated areas, or recreational habits."
Serum versus Recreational Activities: There was evidence that certain recreational activities
may be related to dioxin serum levels. Statements in the key documents which address this
relationship include:
People who do recreational activities in the Tittabawassee River, Saginaw River,
and Saginaw Bay have higher levels of dioxins in their serum than people who do
not do recreational activities in these areas (UM, 2006).
A principal focus of our study was to assess activities that might involve contact
with contaminated soils, river sediments, and household dust. These included
living on contaminated soil, living with contaminated household dust, and
pursuing activities in the contaminated water bodies and floodplain (boating,
swimming, picnicking, hiking, etc.). We found little evidence in the general
population that these activities were associated with increased serum dioxins.
Participating in water activities on the Tittabawassee River more than once per
Page 3 8

-------
month in 1980-2005 was associated with increased TCDD, but less frequent
activities and activities before this period were not. We found no relationship
between these activities and any other dioxin in any historic era (Garabrant et a/.,
2009b).
The overall pattern offindings in this population suggest little contribution from
exposures in Midland and Saginaw counties during the last 25 years to serum
dioxins from either contaminated soil, household dust, consumption of fish and
game from contaminated areas, or recreational habits (Garabrant et a/., 2009b).
EPA COMMENT: This was supported by a positive association between recreational activities
and dioxin body burden that appeared in the 2006 report (UM, 2006). However, like the
conclusion noted above on consuming fish from the impacted water bodies, this conclusion was
not supported in the 2009 literature publication on predictors of serum. A category of "Water
activities factors" was included in Table 1 in Garabrant et al (2009b). This table provided the
percentages of variation for several factors for explaining serum dioxin, and the percentages of
variation for water activity factors ranged from 0.00 to 0.77 for TEQ and several congeners. To
at least show some positive correlation may have been enough to have motivated the UMDES
researchers to list the finding the in 2006 report, but the conclusions in the later journal article
did not echo that sentiment.
Serum versus Working at Dow: A relationship was found between working at Dow during the
middle decades of the 20th century and serum dioxin levels. Statements in the key documents
which address this relationship include:
People who worked at the Dow plant from 1940 to 1959 have higher levels of
dioxins in their serum than people who did not work at Dow during that period
(UM, 2006).
Even though 6% of the subjects reported ever having worked at Dow, we found
little evidence that such work was associated with increased serum dioxin
levels... Work at Dow in 1940-1959 was positively associated with TCDD;
however, working at Dow in 1980-2005 was inversely associated with
1,2,3,6,7,8-HxCDD and 2,3,4,7,8-PeCDF. None of the other variables that
explored work at Dow, work with chlorophenol derivatives, or likely occupational
exposure to dioxins was associated with increased serum dioxins (Garabrant et
al, 2009b).
EPA COMMENT: Interestingly, this does not appear to be a finding for TEQ, but it does appear
to be a finding for 2,3,7,8-TCDD. In Garabrant et al. (2009b), it is noted that this factor
explained only 0.18% of the variation in TEQ serum levels, but 1.82% of the variation in 2,3,7,8-
TCDD. The UMDES researchers have recently conducted analyses (Garabrant et al. 2009d) that
show that impacts from the Dow facilities, including to workers and importantly also to nearby
residents, may have had the most impact during the middle decades of the 20th century (1940 to
1959 for workers, and 1960 - 1979 for residents), but not the later decades (1980 - 2005 for any
subset of individuals), and furthermore, that this impact is seen most with 2,3,7,8-TCDD. It is
Page 39

-------
interesting to note that, once again, a slight change of tone from the 2006 report the 2009 journal
article is present, even though both are based on the same analysis.
Page 40

-------
SECTION 5: UNANSWERED QUESTIONS ABOUT EXPOSURE
5.1 Can Pharmacokinetic Modeling Help to Inform Study Results?
Simple pharmacokinetic (PK) modeling was used to study the potential impact from
living on soil with elevations in dioxin. This exercise is described in detail in Appendix 1. The
basic structure of the exercise was to simulate the lifetime accumulation of dioxins in a 20 year-
old, a 40 year-old, and a 60 year-old who might have been sampled for dioxin-like compounds in
2007. Doing these simulations entailed constructing their lifetime exposures, considering the
years they were living (the 40 year-old lived from 1968 to 2007, for example) as well as how
exposures would change over their specific lifetimes (children have a higher body-weight based
exposure due to elevated soil ingestion rates, etc). In this exercise, there were, two 20 year-olds,
two 40 year-olds, and two 60 year-olds, in two distinct populations. In one, the 20, 40, and 60
year-old had all background exposures including soil-related exposures (ingestion plus dermal
contact) at background soil levels. This level, as ascertained by the Dioxin Reassessment (EPA,
2003a), was assumed to be 11 pg/g TEQ (dioxins, furans, and PCBs: D/F/P). In the other, the
three individuals had all of the background exposures and the soil-related exposures were based
on a soil concentration of 223 pg/g TEQ (D/F/P), which was the 95% concentration found in
properties measured on the flood plain. The principal output from this exercise was the body
burden of the six individuals in simulated year 2007. Available body burden data to compare
this with include age-related NHANES survey results from 2001-2002 and 2003-2004, as well as
data from the study area. The principal findings from this exercise are summarized below.
The medians for the age ranges in the measured data from JC appear quite comparable to the
analogous (not exactly the same) age ranges and metrics from the NHANES survey (geometric
mean from NHANES 2001-2002 and median from the NHANES 2003-2004). For example, the
median concentrations in the 45-59 and >60 age ranges in JC of 20.8 and 31.3 pg/g TEQ,
respectively, compare well to geometric means of 20.7 and 33.7 pg/g TEQ from the same two
age ranges in NHANES 2001-2002. Interestingly, the lower median from Jackson/Calhoun of
7.8 pg/g TEQ for the 18-29 age range compares better to the NHANES 2003-2004 results (7.1
and 8.9 pg/g TEQ medians for 12-19 and 20-39 ages, respectively) than the NHANES 2001-
2002 results (12.5 geometric means for 20-29 year olds).
The modeled concentrations for the 20, 40, and 60 year-old with background exposures
including background soils are certainly within the range of the Jackson/Calhoun and NHANES
data. The concentrations for the 20, 40, and 60 year-old were 13.6, 16.4, and 19.7 ppt TEQ lipid
weight. The span of concentrations, from 13.6 to 19.7 or about a 6 ppt span, is similar to the
span of about 8 ppt TEQ for both NHANES surveys up until age 60. However, for both
NHANES data sets, the concentration for ages >60 then jump somewhat, to 33.7 pg/g TEQ for
NHANES 2001-2002 and 26.9 for NHANES 2003-2004. It is unclear whether the model would
simulate this high a body burden for individuals older than 60, but that remains untested.
Assuming the above finding suggests credibility with the model for characterizing the
varying body burdens of individuals living in the US in background settings, one can then use
the model to study the impact to body burdens from exposures higher or lower than the
Page 41

-------
background scenario crafted. The exposure examined here was to elevated soil. The modeled
difference between body burdens of individuals exposed to background only compared to
background plus elevated soils, at 223 ppt TEQ, appears to be only about 3 ppt TEQ lipid
weight. This difference was consistent for all three ages.
This analysis is preliminary with several uncertainties, which were identified in the
appendix. Still, given the uncertainties, limitations, assumptions, and so on, this at least begins
to provide supportive evidence that exposures to elevated soils, by themselves (i.e., no secondary
impacts such as elevated fruit/vegetables, elevated fish or terrestrial food animal products) may
not provide that great an impact to body burdens of adults of all ages who may be surveyed
during the 2000s.
This same modeling structure can now be used to study other issues that may be of
concern, such as potential exposures to children, potential exposures to soils even higher than
223 ppt TEQ, and other scenarios.
5.2 What About Children and Exposure to Contaminated Soils and Dust?
It was stated in the UMDES study documentation that children were not measured
because the volume of blood required for laboratory analysis was evaluated as being too high for
them. However, of all populations, children might be deemed the most exposed to dioxins in
soils.
The pharmacokinetic modeling structure described above and in the appendix can
provide an initial sense of what the body burden difference might be for a child living on
elevated soils versus background soils. As part of the simulation of lifetime accumulation of
dioxins in the PK exercise, children aged 1-6 were assumed to ingest 100 mg/day of soil. This
was reduced to 50 mg/day for all later ages. For the fully background scenario, the soil
concentration was 11 ppt TEQ, while for the background plus elevated soil, this was increased to
223 ppt. The difference in the two 20 year-olds when they turned 5 was examined; this is
because the 20 year-old turned 5 in 1993, and based on the structure of the historical intake
model, background exposures to dioxin leveled off at about 1980, so this prediction of 5 year-old
body burdens in 1993 would be comparable to simulating an individual being born in 2002 and
turning 5 in 2007. The modeling suggests about a 6 ppt TEQ difference in the body burdens of
the two 5 year-olds in 1993: it is 23.4 ppt TEQ for the child living on elevated soil and 17.4 for
the child living on background soils. Besides noting the 6 ppt TEQ difference, one might also
characterize the difference at about 33% from living on soils at about 223 ppt TEQ; i.e., an
increase from 17 to 23 ppt TEQ is about a 33% increase.
5.3 Were all Populations of Interest Adequately Represented in the Study? What About
Highly Exposed Individuals and Areas with Highly Elevated Soil Concentrations?
This issue was discussed in detail in Section 3.1. Basically, it was concluded that
although UMDES took steps to increase the likelihood that impacted soils would be represented
Page 42

-------
in the survey, the success of this effort is unclear. Data presented in ATS (2009) suggest that
high soil levels may be more prevalent than found in UMDES. Section 3.1 also concludes that
UMDES did not include any special features to ensure representation of activities related to high
end exposures such as farming, fishing, hunting and gardening. The study did end up including
one farmer who consumed his own beef and one gardener with elevated soils. In both these
cases, the individual was found to have elevated serum dioxin concentrations (Garabrant et al.,
2009b).
5.4 Is it Reasonable that the Correlations Between Soil and Blood Were Weak?
Yes, this is a reasonable expectation, given the arguments presented above. As noted,
95% of the properties on the floodplain have topsoil concentrations at 223 ppt TEQ and lower,
and yet the PK exercise suggested living on soils this elevated compared to living on background
soils would only raise body burden TEQ by about 3 ppt TEQ. Without the use of rigorous
survey statistics, and given the range of serum levels found in the UMDES survey, it is at least
intuitive to surmise that a weak association between soil and blood would be expected when over
95% of the individuals that could be sampled might have body burden impacts at 3 ppt TEQ and
less. Indeed, had a strong correlation been found (given the range of soil concentrations found in
the FP and other areas), EPA's own evaluations (here and in EPA, 2003, for example) suggesting
a small impact from soil exposure compared to all other exposures, particularly food exposures,
would need to be reassessed.
EPA cautions that the findings in the UMDES are not general findings, for at least
two important reasons: they do not address children's exposures, and they may not address
living on properties with high dioxin soil concentrations. EPA asked UMDES investigators and
received data for properties with soil concentrations exceeding 1000 ppt TEQ. Twenty-three
properties were identified. Inspection of those data suggests that blood serum dioxin levels
tended to be elevated for subjects living on those properties (See Appendix B).
Page 43

-------
SECTION 6: RELEVANCE TO EPA
EPA acknowledges that many stakeholders are concerned about the implications of this
study for risk assessment, particularly with regards to dioxin levels in soils. Current EPA soil
dioxin remediation goals are largely based on a scenario of childhood soil ingestion. The
Superfund-supported benchmark of 1000 ppt TEQ, for example, is based on a scenario of
exposure for ages 1 to 30 years old with about two-thirds of the dose estimated to occur prior to
age 18. No children (<18 years) were sampled in the study, so no insights as to the potential
impacts to children as measured by body burdens can be gained from the UMDES.
The ability of the UMDES to accurately capture "high end" scenarios was questioned
(see discussions in Section 3). Conclusions in the UMDES study about the relationship between
soil concentrations and serum concentrations pertain to the general adult population. The study
design did not guarantee inclusion of the properties with the highest soil concentrations, the
fishers with the highest consumption rate of recreationally caught fish, or those with highest
exposure from gardening in soils with elevated dioxin concentrations. New data supplied to
EPA on individuals living on properties at concentrations above 1000 ppt TEQ may suggest a
relationship between high dioxin soil concentrations and body burdens.
Regulatory benchmarks such as dioxin remediation goals or fish advisories are based on
high end exposure scenarios. They are based on factors that have uncertainties and variabilities,
such as soil and fish ingestion rates, years of exposure, cancer potency factors, and so on. None
of these features are addressed in the UMDES study. Any discussion on the usefulness of the
UMDES results on regulatory activities must first recognize this fact.
Perhaps the most interesting findings from the UMDES, with regard to this discussion,
are the findings that some individuals have elevated serum dioxins that can be associated with
specific activities that are linked to higher exposures. One individual in the FP who had elevated
dioxin levels in his blood (Franzblau et al, 2009b) raised cattle for consumption (for himself,
family, and friends). Interestingly, the elevations were in the furan congeners and these
congeners are elevated in FP soils. Therefore, the pathway of soil to cattle to humans was likely
in this case. In another instance, a gardener with elevated dioxin serum levels (Garabrant et al.,
2009b) also had elevated garden soil (Franzblau et al, 2009c). EPA suggests that these
individuals support the contention that contact with soils, directly or via the food chain, may be
associated with elevated blood dioxin concentrations.
Finally, dioxin presents a challenge for risk assessors and policymakers because the
background exposure to dioxin is already at a level considered to be of concern. This was a
finding of EPA's draft Dioxin Reassessment (EPA, 2003), and another finding was that animal
food consumption explained 95% of the exposure and that resulting risk level. Therefore, any
incremental exposure over the dietary background will increase the risk over the range of
concern.
Page 44

-------
SECTION 7: CONCLUSIONS
The most important limitations to the scope of this study are that it addresses exposure
only (i.e. no health status or outcomes data were collected) and participants were limited to
individuals aged 18 years or older. A design limitation is that as an observational study it
provides the basis for examining associations between the variables, but cannot be relied on for
determining cause and effect relationships.
EPA concludes that the UMDES study design was well-suited for meeting its first
objective of characterizing distributions and identifying patterns of exposure. The relatively large
number of participants and the population-based statistical sampling design are important
strengths of this study. They provide reliable estimates of the distributions of dioxin
concentrations in blood, soil and dust, and the study provides several measures of those
distributions (i.e. means, medians, lower and upper percentiles).
EPA believes that the UMDES design may only have been partially successful in meeting
the second objective, which was to evaluate factors associated with serum dioxin. EPA
concludes that the UMDES identified the associations between serum levels and a variety of
demographic factors (age, sex, body mass index, where one lives, work history). However, EPA
is uncertain that the UMDES evaluated the full impact on serum levels of soils or dust, or of
various activities, including dietary choices. Determining the incremental effect on serum levels
of dioxin exposures from soil or dust is problematic because soil and dust contributions to total
exposure are generally small compared to background dietary exposures. Pharmacokinetic
modeling done as part of this review indicates that direct ingestion of soil and dust would not be
expected to contribute significantly to adult serum levels except at relatively high soil and dust
concentrations (i.e. over 1000 ppt TEQ).
The UMDES researchers recognized this need and took steps to address it. They targeted
sampling areas that most likely had soil impacts from the Dow facility (an area impacted by the
incinerator plume and the 100 year floodplain downstream of the plant). Further, they used a
very high sampling rate in these areas (17% of the population in the floodplain and 30% of the
population in the plume area) and ultimately included 23 properties with dioxin levels over 1,000
ppt TEQ. Judging whether this is an adequate representation is difficult. The relatively low
median soil levels in the floodplain (11 ppt TEQ) and suggestion of higher levels from other
studies (ATS, 2009) support the concern that it may not be adequate.
A related design issue is how well the data represented behaviors occurring on
contaminated soils that could lead to elevated dioxin exposures such as gardening, consuming
local fish and game or raising animals for local consumption. No design elements were used to
ensure representation of these activities. However, two such scenarios were identified in the
study (one gardener and one farmer) and the UMDES researchers have highlighted these
findings.
The UMDES data provided some unanticipated but useful findings. These involved the
identification of several unexpected dioxin exposure scenarios. First, was the discovery of soils
Page 45

-------
containing elevated levels of dioxin outside the floodplain due to movement of soils from the
floodplain to residences for fill or other purposes. Second was the discovery of elevated serum
levels that appear to be associated with the activities of a ceramic clay hobbyist. Third was the
discovery that polychlorinated biphenyls (PCBs) were the dominant contributors to high TEQ
levels in soil at one residence (linked to past paint use), and in dust at nine residences (where one
was linked to a carpet pad and others had unknown sources.
A Quality Assurance Project Plan was drafted which, although not complete in all areas,
addressed the key elements of quality assurance procedures. However, the publications to date
have not included a report on quality assurance performance measures.
The points discussed above represent EPA's conclusions regarding the study design.
EPA's primary comments on the study findings are summarized below.
EPA concludes that the following findings are well supported by the UMDES data and
analyses:
•	Soils from properties in the Midland/Saginaw study areas contain higher concentrations
of dioxin than soils in the reference area, the Jackson/Calhoun area.
•	Higher concentrations of dioxins were found in the serum of residents living within the
four study areas comprising the Midland/Saginaw region as compared to the reference
Jackson/Calhoun area.
•	Dusts from households in the Midland Plume area contain higher dioxin levels than dusts
from households in the other study areas.
•	Living in the Midland/Saginaw study areas during the years 1960-1979 was associated
with elevated serum dioxins.
•	Working at Dow during the years 1940-1959 was associated with elevated serum dioxin
levels. This does not appear to be a finding for TEQ, but it does appear to be a finding
for 2,3,7,8-TCDD.
The most highlighted finding of this study is that age is positively associated with serum
dioxin levels. EPA agrees that the analysis done on UMDES data support this finding.
However, while age may have the strongest correlation with serum levels of all variables studied,
plots of serum level versus age show substantial scatter or variability, yielding a significant, but
weak to moderate positive correlation. The R2 for age alone was not reported. However, the
correlation between serum TEQs and an aggregate of nine demographic factors, including age,
was reported to have an R2 of 0.396 (Garabrant et al., 2009b), so age by itself would be lower.
Thus, the assertion that age is a strong predictor of serum levels appear overstated.
Many findings appear in the various documents relating to the key relationship between
serum dioxin and soil dioxin. Generally, weak relationships between various measures of soil
and serum dioxin levels were noted but the overall conclusions in the more recent documents
imply that there is not a meaningful relationship between these two key factors. EPA believes
this general finding of a weak association between soil and dioxin serum dioxin levels is
supported by the statistical analysis that is presented in Garabrant et al (2009b). This may be
associated with the large relative contribution from diet, which would be expected to overwhelm
Page 46

-------
contributions from soils at background concentrations. The study design may not have fully
represented the soils with high concentrations of dioxin making it difficult to fully evaluate this
relationship. EPA also emphasizes that it is important that this finding not be stated alone
without the caveats about the study limitations.
Similarly, UMDES found little association between household dust concentrations of dioxins
and serum. EPA agrees that the analysis done on UMDES data support this finding. However,
the issues about the relative contribution of diet and the design issues noted above for soils
would also apply to dusts. Also, it is noted that two key documents, the 2006 report (UM, 2006)
and the 2009 journal article (Garabrant, 2009b), have slightly different results. The 2006 report
suggested that a relationship might exist between PCB 1 18 levels in serum and household dust,
whereas Garabrant et al. (2009b) suggest that the relationship is with PCB 126.
A variety of findings addressed serum/fish relationships including: where the fish came from
(store bought or from the impacted water bodies), what kind of fish, whether the individuals
recreationally fished, and so on. The general comment that there is a relationship between
consumption of fish and serum dioxin levels is generally supported by the several findings that
were outlined in the primary publication detailing the statistical correlation analysis of the data
set (Garabrant et al, 2009b). While the correlation was noted, the statistical relationship was
weak. There appeared to be some slight differences and inconsistencies between the 2006 report
(UM, 2006) and the 2009 journal article (Garabrant et al., 2009b).
Certain recreational activities were found to be associated with dioxin serum levels. This
was supported by a positive association between recreational activities and dioxin body burden
that appeared in the 2006 report (UM, 2006). However, like the conclusion noted above on
consuming fish from the impacted water bodies, this conclusion was not supported in the
Garabrant et al., 2009.
Page 47

-------
SECTION 8: REFERENCES
Anderson, H.A.; Falk, C.; Hanrahan, L.; Olson, J.; Burse, V.W.; Needham, L.; Paschal, D.;
Patterson, D.; Hill, R.H.; Boddy, J.; Budd, M.; Burkett, M..; Fiore, B.; Humphrey, H.E.B.;
Johnson, R.; Kanarek, M..; Lee, G.; Monaghan, S.; Reed, D.; Shelley, T.; Sonzogni, W.; Steele,
G.; Wright, D. (1998) Profiles of Great Lakes critical pollutants: A sentinel analysis of human
blood and urine. Environmental Health Perspectives. 106(5):279-289.
Ann Arbor Technical Services, Inc (ATS). (2009). Final Geomorph site characterization report.
Tittabawassee River and floodplain soils. Midland, MI. Volume I Site Characterization Report.
Chen, Q; Garabrant, D; Hedgeman, E; Gillespie, B; Hong, B; Knutson, K; Lepkowski, J;
Franzblau, A; Jolliet, O. Chang CW, Garabrant D, Gillespie B, Hong B, Chen Q, Franzblau A,
Hedgeman E, Knutson K, Lee SY, Adriaens P, Demond A, Towey T, Trinh H, Sima C and
Lepkowski J. (2007). Association of dioxin concentrations with serum dioxin concentrations in
Midland, MI, USA. Poster presentation at Dioxin 2007. Tokyo, Japan.
Chen Q, David H. Garabrant, Elizabeth Hedgeman, Brenda W. Gillespie, Biling Hong, Kristine
Knutson, James Lepkowski, Alfred Franzblau, and Olivier Jolliet. (2008). Serum Dioxin, Furan,
and PCB Concentrations and Half-Life Study Among the U.S. General Population.
Epidemiology. 16(5), S233-234, November Supplement.
Cummings, P. and Rivara, F.P., Reporting Statistical Information in Medical Journal Articles
Arch Pediatr AdolescMed. 2003;157:321-324.
Demond A, Adriaens P, Towey T, Chang S, Hong B, Chen Q, Chang C, Franzblau A, Garabrant
DH, Gillespie BW, Hedgeman E, Knutson K, Lee S, Lepkowski J, Olson K, Ward B, Zwica L,
Luksemburg W, Maier M. (2008). Statistical comparison of residential soil concentrations of
PCDDs, PCDFs and PCBs from two communities in Michigan. Environ. Sci. Technol.
42(15):5441-5448.
EPA, 2003. Dioxin Reassessment National Academy of Sciences (NAS) Review Draft 2003.
U.S. Environmental Protection Agency, Washington, D.C., EPA/600/P-00/001Cb.
EPA, Environmental Protection Agency (EPA). (2007). Pilot survey of levels of PCDDs,
PCDFs, PCBs and mercury in rural soils of the U.S. EPA/600/R-05/048F.
Franzblau A; Hedgeman E; Chen Q; Lee, S; SY, Adriaens P; Demond A; Garabrant, D; DH,.
Gillespie, B; BW, Hong B; Jolliet O; Lepkowski J; Luksemburg W; Maier, M; Wenger, Y.
(2008). Case Report: Human Exposure to Dioxins from Clay. Environmental Health
Perspectives. 116(2):238-42.
Franzblau, A; Zwica, L; Knutson, K; Chen, O; Lee, S; Hong, B; Adriaems, P; Demond, A;
Garabrant, D; Gillespie, B; Lepkowski, J; Luksemburg, W; Maier, M; Towey, T. (2009a). An
Page 48

-------
investigation of homes with high concentrations of PCDDs, PCDFs, and/or dioxin-Like PCBs in
house dust. Journal of Occupational and Environmental Hygiene, 6 (3): 188 - 199.
Franzblau, A; Hedgeman, E; Knutson, K; Towey, T; Chen, Q; Hong, B; Adriaens, P; Demond,
A; Garabrant, D H; Gillespie, B W; Lepkowski, J. (2009b). Abstract # 1211. The University of
Michigan Dioxin Exposure Study: A Follow-Up Investigation of Cases with High Serum
Concentrations of 2,3,4,7,8-PENTACDF. Epidemiology 19(6Supplement):S236.
Franzblau, A; Demond, A; Towey, T; Adriaens, P; Chang, S; Luksemburg, W; Maier, M;
Garabrant, D; Gillespie, B; Lepkowski, J; Chang, C; Chen, Q; Hong, B. (2009c). Residences
with anomalous soil concentrations of dioxin-like compounds in two communities in Michigan,
USA: A case study. Chemosphere. 74(3), 395-403.
Garabrant DH, Franzblau A, Gillepsie B, Lin X, Lepowski J, Adriaens, P and Demond A. (2005)
The University of Michigan dioxin exposure study: Study protocol.
Garabrant, D; DH, Biling Hong, B; Chen, Q; Chang, C; Jiang, X; Franzblau, A; Lepkowski,
J.; Adriaens, P.; Demond, A.; Hedgeman, E.; Knutson, K.; Towey, T.; Gillespie, B. . (2008).
Factors that Predict Serum PCB, PCDD, and PCDF Concentrations in Michigan, USA
Epidemiology. 16(5), S265, November Supplement.
Garabrant DH, Franzblau A, Lepkowski J, Gillespie BW, Adriaens P, Demond A, Ward B,
LaDronka K, Hedgeman E, Knutson K, Zwica L, Olson K, Towey T, Chen Q, Hong B. (2009a).
The University of Michigan Dioxin Exposure Study: Methods for an Environmental Exposure
Study of Poly chlorinated Dioxins, Furans and Biphenyls. Environ Health Perspect 117:803-810.
Garabrant DH, Franzblau A, Lepkowski J, Gillespie BW, Adriaens P, Demond, Hedgeman E,
Knutson K, Zwica L, Olson K, Towey T, Chen Q, Hong B, Chang C, Lee S, Ward B, LaDronka
K, Luksemburg W, Maier M. (2009b). The University of Michigan Dioxin Exposure Study:
Predictors of Human Serum Dioxin Concentrations in Midland and Saginaw, Michigan. Environ
Health Perspect: 117:818-824.
Garabrant D, B Hong, O Jolliet, QChen, X Jiang, A Franzblau, J Lepkowski, P Adriaens, A
Demond, E Hedgeman, K Knutson, T Towey, B Gillespie. (2009). Public Health Impact of
Dioxin Exposure Pathways in the UMDES, Based on Linear Regression Models. Presentation
at Dioxin 2009, Beijing, China.
Garabrant DH. (2009c). UMDES response to comments by John Kern, PhD.
http://www.sph.umich.edu/dioxin/datarequests.html):
Hedgeman, E; Chen, Q.; Hong, B.; Chang, C.; Olson, K.; LaDronka, K.; Ward, B.; Adriaens,
P.; Demond, A.; Gillespi, B.; Lepkowski, J.; Franzblau, A.; Garabrant, D.. . (2009). The
University of Michigan Dioxin Exposure Study: Population Survey Results and Serum
Concentrations for Polychlorinated Dioxins, Furans, and Biphenyls. Environ Health Perspect
117: 811-817.
Page 49

-------
Hedgeman E , B Hong, Q Chen, K Knutson, SY Lee, K Olson, B Lohr-Ward, K Ladronka, M
Maier, W Luksemburg, J Lepkowski, B Gillespie, A Franzblau, D Garabrant. (2007). Change
in background serum levels with the new 2005 TEFs. Poster presentation at Dioxin 2007.
Tokyo, Japan.
Hedgeman E.; Chen, Q.; Hong, B.; Knutson, K.; Lee, S.; Olson, K.; Lohr-Ward, B.;
LaDronka, K.; Lepkowski, J.; Gillespie, B.; Franzblau, A.; Garabrant, D. . (2008). Current
Estimates of Population Serum PCDD, PCDF, and Dioxin-like PCB Levels in the United States.
Epidemiology. 16(5), S235, November Supplement.
Hong B, D Garabrant, X Jiang, Q Chen, A Franzblau, B Gillespie, J Lepkowski, P Adriaens, A
Demond. (2009). Factors That Predict Serum Concentration of 2,3,7,8-TCDD in People from
Michigan, USA. Presentation at Dioxin 2009, Beijing, China.
Jolliet O.; Wenger, Y.; Milbrath, M.; Garabrant, D.; Jiang, X.; Gillespie, B. . (2008).
Pharmacokinetic Modeling to Support the Statistical Analysis of Blood Dioxin Concentration.
Epidemiology. 16(5), S372, November Supplement.
Knutson K, Zwica L, Lee S, Hong B, Chen Q, Towey T, Gillepsie B, Demond A, Adriaens P,
Franzblau A and Garabrant D (2007). Linear regression modeling to predict household dust TEQ
and TCDD concentration. Poster presentation at meeting of the Society for Risk Analysis:
December 2007, San Antonio, TX.
Knutson K.; Hong, B.; Chen, Q.; Chang, C.; Hedgeman, E.; Towey, T.; Jolliet, O.; Gillespie,
B.; Franzblau, A.; Lepkowski, J.; Adriaens, P.; Demond, A.; Garabrant, D. . (2008). The
Relationship Between Blood Serum Dioxin Levels and Breast Feeding. Epidemiology. 16(5),
SI79, November Supplement.
LaKind, J.; Berlin, C.; Naiman, D. . (2001). Infant Exposure to Chemicals in Breast Milk in the
United States: What We Need to Learn From a Breast Milk Monitoring Program Env. Health
Persp 109: 75-88.
Lorber M. (2002). A pharmacokinetic model for estimating exposure of Americans to dioxin-
like compounds in the past, present, and future. Science of the Total Environment 288:81-95.
Michigan Department of Community Health (MDCH). (2007). Fish Consumption Survey of
People Fishing and Harvesting Fish from the Saginaw Bay Watershed. Michigan Department of
Community Health, Lansing, MI. June 14, 2007.
Milbrath, M.; MO, Yvan Wenger, Y.; Chang, C.; Emond, C.; Garabrant, D.; Gillespie, B,
Jolliet, O. 2008. Apparent Half-Lives of Dioxins, Furans, and PCBs as a function of Age, Body
Fat, Smoking Status, and Breastfeeding. (2009). Environmental Health Perspectives. 117(3):
417-425.
Page 50

-------
University of Michigan (UM). (2006). Measuring people's exposure to dioxin contamination
along the Tittabawassee River and surrounding areas: Findings from the University of Michigan
dioxin exposure study.
Cummings, P. and Rivara, F.P., Reporting Statistical Information in Medical Journal Articles
Arch Pediatr AdolescMed. 2003;157:321-324.
Page 51

-------
APPENDIX A.
EVALUATION OF THE IMPACT OF SOIL CONCENTRATIONS
ON BODY BURDENS OF INDIVIDUALS
Introduction and Approach:
The Dioxin Reassessment (US EPA, 2003) included pharmacokinetic (PK) modeling which
predicted the lipid concentration of dioxin-like compounds during an individual's life, starting
from the impacts of breast-feeding through adulthood and old age. This simple PK model was
retrieved and used to model the body burdens of individuals who lived in two settings which
differed only in the soil concentration to which the individuals were exposed. Specifically, the
individuals that are modeled here live in: a fully background setting with regard to lifetime
intakes including intakes via ingestion of soil at background levels, and a setting where intakes
are the same as the first scenario except the soil-related intakes (ingestion plus dermal contact) is
based on living on a soil that has a toxic equivalent (TEQ) concentration of 223 pg/g (ppt).
The concentration of 223 ppt TEQ was selected since this is the level that was
characterized as the 95% percentile flood plain soils (Demond et al, 2008). This compares to the
11.6 ppt TEQ (17 D/Fs, 13 PCBs) assumed for background exposures in the Reassessment (US
EPA, 2003). At this soil level, background soil impacts explained about 1% of total exposures
for the various age ranges of individuals, with the exception of breast-feeding, which dominated
exposures when it was modeled. It was above 1% for the children 1-5 whose soil ingestion rate
was 100 mg/day, but below 1% for older age ranges, where ingestion rates were 50 mg/day. For
example, the Reassessment estimated a total exposure of 50 pg TEQ/day for children ages 1-5.
Of that total, soil ingestion (100 mg/day) plus soil dermal contact (2.2 mg/day) explained 1.2 pg
TEQ/day (102.2 mg/day * 11.6 pg TEQ/g * 0.001 g/mg = 1.19 pg TEQ/day), which was over 2%
of the total. If instead the child was living on a property whose soil level was 223 ppt TEQ, then
his/her total exposure between ages 1-5 would be:
50 pg/day - 1.2 pg/day + (102.2 mg/day * 223 pg/g * 0.001 g/mg) = 71.5 pg/day
In this exercise, we will model the body burdens of a 20 year-old, a 40 year-old, and a 60
year-old whose blood was sampled in 2007. Therefore, the 20 year-old was born in 1988, the 40
year-old was born in 1968, and the 60 year-old was born in 1948. We will model the impacts of
being raised in two settings: a background setting with all exposures similar to that of the
Reassessment, and another background setting with all exposures the same except the soil-
related exposures which are based on a concentration of 223 ppt TEQ. This results in a total of 6
scenarios:
20 year-old, 11.6 ppt TEQ soils
20 year-old, 223 ppt TEQ soils
40 year-old, 11.6 ppt TEQ soils
Page 52

-------
40 year-old, 223 ppt TEQ soils
60 year-old, 11.6 ppt TEQ soils
60 year-old, 223 ppt TEQ soils
The outcome of interest in these scenarios is simply the predicted body burdens of total dioxins
in these individuals as though they were sampled in 2007.
In order to do this exercise, and also to be clear on the steps necessary to model the doses
to which these individuals are exposed to, the construct of the exercise will be described in the
following series of steps:
1) Step 1: Develop a "multiplier "for converting current doses to past doses
This exercise must consider the differences in environmental levels of dioxins in past
years; specifically, higher exposures occurred in the past because of higher environmental levels.
This was studied in Lorber (2002), who modeled the exposure of Americans to dioxins
throughout the 20th into the 21st century using a simple pharmacokinetic model. He found that
individuals in America who lived in the second half of the 20th century had higher body burdens
than currently because of much higher exposures in the past. Table 1 is reproduced from Lorber
(2002) and it shows the results from various blood monitoring studies of Americans in
background settings throughout the 20th century. The earliest studies measuring the 17 dioxin
and furan congeners (this table did not include dioxin-like PCBs) in background settings was a
control group for a systematic study of Vietnam Veterans in sampling which occurred in several
years in the 1970s. As seen in Table 1, background body burdens in the 1970s was consistently
higher than 50 ppt TEQ, while in later years, it was under 30 ppt TEQ. The latest NHANES data
from the early 2000s shows body burdens mostly near and under 20 ppt TEQ, except for the
oldest in the population. In any case, Lorber (2002) studied temporal trends in exposure using
the simple pharmacokinetic model used in this exercise. He calibrated a dose which was able to
reproduce body burdens that were seen in Table 1. Figure 1 is from Lorber (2002) and it shows
this calibrated background dose of dioxins and furans that was derived for adults for exposures
which occurred in the past. While this calibration curve may suggest a degree of certainty in the
knowledge of past exposure, Lorber (2002) describes the uncertainties associated with its
generation, including (for example) the use of the simple pharmacokinetic model, the modeling
of TEQ as though it were a single compound, the lack of data for historical serum and exposure
media data to back up this historical intake, and so on. In any case, use of this intake did result
in a good fit between model predictions and past body burdens compared against those that are
shown in Table 1. As seen in Figure 1, the dose in the past was modeled to peak at about year
1964 and to decline steadily from that point until 1980. Then, the decline was more gentle until
year 2000 and then flattened out past that. Specifically and for current purposes, the CDD/F
dose in the year 1948 was 4.1 pg TEQ/kg-day, it peaked in 1964 at 6.5 pg TEQ/day, in 1968 the
dose was 5.5 pg TEQ/kg-day, in 1980 the dose was 1.0 pg TEQ/kg, and in 2000, the dose
flattened at 0.65 pg TEQ/kg-day. This 0.65 pg TEQ/kg-day was, in fact, derived in Lorber
(2002) to be the same as current CDD/F background dose of 43 pg CDD/F TEQ/day that was
developed as part of the Dioxin Reassessment effort. In the Reassessment, the current
background dose of PCBs was derived as 23 pg PCB TEQ/day, resulting in a total current dose
Page 53

-------
of 66 pg CDD/F/PCB TEQ/day. The dose of PCBs turned out to be about 33% of total adult
dose. If we assume the ratio between dioxin/furan and dioxin-like PCBs to be constant over
time, then the total doses (CDD, CDF and PCB) over time for an adult can be derived as the
CDD/F dose times 1.5. The results of this multiplication are: 6.15 pg TEQ/kg-day in 1948, 9.75
pg TEQ/kg-day in 1964, 8.25 pg TEQ/kg-day in 1968, 1.5 pg TEQ/kg-day in 1980, and 1.0 pg
TEQ/kg-day in 2000. These "total doses" can now be used as "multipliers" that correlate current
dose levels to past levels. For example, the adult dose in 1980 is calculated as the current adult
total dose, 66 pg TEQ/day, times the 1980 multiplier, 1.5, and in 1968, the total adult dose is
calculated as 66 pg TEQ/day * 8.25 = 545 pg TEQ/day. Needless to say, the doses from 2000 on
are the same as derived in the Reassessment, since the multiplier is 1.0.
2) Step 2: Use the "multiplier " with current background exposures to model past exposures
This step will now map out the strategy to model past background exposures, not yet
considering being raised on elevated soil levels. First, the Dioxin Reassessment provided these
background D/F/P exposures, on a mass/day basis (not on a body weight basis) as a function of
age:
If we neglect breast-feeding, we can assume that the dose from year 0-1 equals the dose for the
1-5 year-old, even though the dose from 1-5 has a soil ingestion component. This is obviously a
simplification for this exercise. However, the Reassessment showed with modeling that while
breast-feeding greatly impacted the body burden of an infant and child, the body burdens of a
breast-fed versus a formula-fed individual were very similar by the time the individuals reached
20 years old or thereabouts. If an individual was born in 1968, their initial exposure is calculated
as 50 pg TEQ/day * 8.25 = 412.5 pg TEQ/day. This calculation employs the unitless multiplier
of 8.25 derived above. When this individual turns 6 years old, their dose is recalculated as 54 pg
TEQ/day times a new multiplier, extrapolated downward from 8.25 to be about 6.0. Similarly
when the individual turns 12 in 1980, his/her dose is now calculated as 61 pg TEQ/day times a
multiplier of 1.5. We can continue this calculation for all ages up until 2007 when the individual
turns 40. Doing these calculations entailed a linear decline in the multiplier, as well as linear
changes in dose as a function of age. This procedure was repeated for an individual born in 1948
and 1988.
3) Step 3: Incorporate elevated soil levels into these calculations
Step 2 above described the derivation of daily doses for an individual born in 1948 (the 60 year-
old), 1968 (the 40 year-old), and 1988 (the 20 year-old), being exposed to background levels of
dioxins, furans, and PCBs. Reconstructing the past exposures was accomplished by using the
temporal dose regime constructed by Lorber (2002) in combination with the Dioxin
Reassessment's evaluation of current dose for different age ranges. This procedure was laid out
in steps 1 and 2 above. The next step will be to amend the three dose scenarios developed (i.e.,
Age
1-5
6-11
12-19
>19
50 pg TEQ/day
54 pg TEQ/day
61 pg TEQ/day
66 pg TEQ/day
Page 54

-------
for the 20, 40, and 60 year-old) to consider the impact of being raised on soils typical of the 95%
soil concentration of floodplain soils: 223 ppt TEQ. For the background scenarios, exposures
via soil ingestion plus dermal contact was 102.2 mg soil/day contact for children ages 1-5, and
about 51 mg soil/day thereafter. In the Introduction above, it was shown how the daily exposure
for a child ages 1-5 rose from 50 pg TEQ/day to 72 pg TEQ/day considering being raised on soil
at concentrations of 223 ppt TEQ. If we now change the exposures for the other age ranges, we
arrive at:
If we repeat the analysis as outlined in steps 1 and 2 above, we arrive at exposures for the three
age ranges over time.
It is important to note that this three-step procedure results in past exposures that were
uniformly higher in all pathways as compared to today's exposure. Specifically, the
"multiplier " gets applied to the intakes displayed above. The proportion of soil intakes to total
intakes remains the same throughout time. What this means, however, is that this simulation
assumes higher soil concentrations in the past (assuming that the soil exposure factors of soil
ingestion and soil dermal contact remain the same over time). With a multiplier of 8, for
example, it is assumed that back in the 1960s, soil that is today at 223 ppt TEQ was more like
2000 ppt TEQ then. At the very least, this is conservative, but it is also plausible. Dioxins do
degrade in surface soils with a half-life of roughly 25 years (Paustenbach et al. 1992), so with
minimal inputs over the past 25 years, a concentration of223 ppt TEQ today might have been
450 ppt TEQ 25 years ago. It is plausible as well that emissions into the air as well as into the
river by the Dow facility were much higher in the 1960s as compared to today, leading to higher
depositions to soil from the atmosphere and higher sediment concentrations, and hence higher
surface soil concentrations in affected areas. In any case, the discussion of all results on
impacts to body burdens of living on soils at 223 ppt TEQ is really a discussion on living on soils
that are at that level today but were higher in the past. If the soil in fact had been constant at
223 ppt TEQ, than the predicted impacts to body burdens increments would be lower than
presented in this Appendix.
Table 2 shows the final results of this procedure to develop daily doses for the six
scenarios. The table includes the daily doses in 5-year increments for each of the 6 scenarios.
Step 4: Run simulations incorporating all PK assumptions as previously developed
The Dioxin Reassessment provides details on the simple 1st order, single-compartment
pharmacokinetic (PK) model used to simulate the accumulations of dioxins throughout a lifetime
of exposure, including breast-feeding. The simplified modeling approach treats "TEQ" as
though it were a single compound with a half-life (and elimination rate) that varies over time as a
function of the person's age. Also varying over a person's age are body weight and body lipid
Age
11.6 ppt TEQ
223 ppt TEQ
1-5
6-11
12-19
>19
50 pg TEQ/day
54 pg TEQ/day
61 pg TEQ/day
66 pg TEQ/day
72 pg TEQ/day
65 pg TEQ/day
74 pg TEQ/day
78 pg TEQ/day
Page 55

-------
reservoir. Table 3 shows how these key parameters vary over the course of a lifetime. Another
key parameter for PK modeling is the absorption fraction. Intakes are multiplied by this factor to
consider what portion of the intake dose gets absorbed into the body. Simplifications were
made in this exercise to consider absorption. First, here is how it was done in the Reassessment.
Food and inhalation ingestion amounts were multiplied by an absorption fraction of 0.8 before
doing PK modeling. Soil ingestion exposures were multiplied by 0.5. The dermal contact factor
of 2.2 mg/day described above considers several parameters including surface area of skin which
contacts dirt, contact events per day, adherence of soil to skin, and most importantly, it also
considers absorption at 3%. In other words, the value of 2.2 mg/day already considers
absorption and an additional factor is not needed. For this exercise, the total doses derived above
were all simply multiplied by 0.8 before doing PK modeling. This results in a slight
overprediction, as compared to the Reassessment approach, because soil ingestion exposures
were not reduced by 50%, but only by 20%. As in other aspects of this simulation, values
between the dates shown in Tables 2 and 3 are interpolated in order to arrive at (somewhat)
smooth simulations. The model was incorporated into a spreadsheet using a 1-month time-step,
with interpolations as necessary.
Results:
The results from the simulation are shown in Figures 2-4 and Table 4. Figures 2-4 show
the body burdens from birth to 20 (Figure 2), 40 (Figure 3), and 60 (Figure 4), given that these
individuals reached these ages in 2007. The ages at that last point are reproduced in Table 4,
along with comparisons to NHANES data and to data from the Jackson/Calhoun cohort of the
UMDES. The NHANES data are from 2001-2002 (Scott et al, 2008) and also from 2003-2004
(Lakind et al, 2008). All NHANES data are derived using the 2005 TEF scheme and include the
17 CDD/Fs and 9 coplanar PCBs that have been measured in NHANES. A quick evaluation of
the NHANES data show that about Vi of total TEQ is comprised of the 9 dioxin-like PCBs. This
is not an exact match with the assumption that 1/3 of dose is derived from these PCBs, but given
the simplicity and generality of the overall approach (modeling TEQs as one compound,
assuming the 1/3 contribution by PCBs to total dose is consistent over time, etc), the important
aspect is that past exposures to CDD/F/PCB TEQ is meaningfully higher than just CDD/F TEQ.
In any case, some immediate observations from the information presented in Table 4 include:
1.	Although the age breakdown of NHANES 2001-2002 is not the same as NHANES 2003-
2004, and the available metric for NHANES 2001-2002 - geometric mean, is not the same as the
metric for NHANES 2003-2004 - median, it does appear that body burdens for the later survey
date are meaningfully lower. For example, the geometric mean for 20-29 year olds in NHANES
2001-2002 is 12.5 pg/g TEQ while for medians 12-19 and 20-39 ages are both under 10 pg/g
TEQ for NHANES 2003-2004. Similar trends are noted for the other two age categories shown
in Table 4.
2.	Although again not a one-to-one correspondence, the medians for the age ranges in
Jackson/Calhoun appear quite comparable to the similar age ranges from the NHANES survey.
For example, the median concentrations in the 45-59 and >60 age ranges in Jackson/Calhoun of
20.8 and 31.3 pg/g TEQ, respectively, compare well to geometric means of 20.7 and 33.7 pg/g
Page 56

-------
TEQ from NHANES 2001-2002. Interestingly, the lower median of 7.8 pg/g TEQ for the 18-29
age range compares better to the NHANES 2003-2004 results (7.1 and 8.9 pg/g TEQ medians for
12-19 and 20-39 ages, respectively) than the NHANES 2001-2002 results (12.5 geometric means
for 20-29 year olds).
3. The modeled concentrations for the 20, 40, and 60 year-old are certainly within the range of
the Jackson/Calhoun and NHANES data. Further, the span of concentrations seem to be
comparable to both sets of NHANES data. Specifically, the three modeled body burdens are
13.6 (for the 20 year-old), 16.7 (40 year-old), and 19.4 pg /g TEQ (60 year-old), which is about a
6 pg/g TEQ spread. This is similar to the range implied by the 7.1-8.9 pg/g TEQ medians for the
lower ages in NHANES 2003-2004 to the 15.0 pg/g TEQ as the median for the 40-59 age range.
For NHANES 2001-2002, the range appears to be from about 12.5 to 20.7 pg/g TEQ, also
similar to the modeled range. However, for both NHANES data sets, the concentration for ages
>60 then jump somewhat, to 33.7 pg/g TEQ for NHANES 2001-2002 and 26.9 for NHANES
2003-2004. It is unclear whether the model would simulate this high a body burden for older
individuals, but that remains untested. I suspect it would not.
This brief analysis focused only on the model predictions, the Jackson/Calhoun data, and
the NHANES data. What is missing is the listing of the appropriate data from the
Midland/Saginaw cohorts. Strictly speaking, what is required are data for individuals in these
locations who live on soils that are elevated in concentration, say greater than 200 pg/g TEQ as
an example. What is missing even still are the overall population summaries broken out by age
and TEQ for the 4 Midland/Saginaw cohorts. There does appear to be a difference between the
Jackson/Calhoun cohort and the Midland/Saginaw cohorts. The overall TEQ of these two
populations are broken out by overall quartile (not age) in Hedgeman et al (2009). From their
summary, it does appear that the upper percentiles in the Midland/Saginaw population diverge
from both NHANES and Jackson Calhoun. Hedgeman et al (2009) make this observation:
"The population serum data from the present study bracket the NHANES 2001-2002 serum data
published by Ferriby et al. (2007). With the exception of the maximum TEQ(DFP- 1998)
observation, serum percentiles from the Jackson/Calhoun population vary from the U.S. adult
population by < 5 ppt. Within the Midland and Saginaw county populations, serum
concentrations are similar to the NHANES 2001-2002 data up to the 50th percentile, but then
diverge from the U.S. population data with an increase of 10-30 ppt."
It is noted that the overall summary report from 2006 does contain Figure 2 which shows
the age breakdown comparison between Jackson/Calhoun and the Floodplain cohort. This figure
is in TEQ-1998, which makes a difference: they have shown in separate publications that it can
mean several ppt TEQ depending on what system is used. Since both are in the same TEF, both
populations would be adjusted similarly when going to the newer TEF, so it would be expected
that the general trends would not change. In any case, this figure is curious in that it suggests
little difference in the compared cohorts, not like the 10-30 ppt TEQ noted above. This could be
due to the fact that the statement above covers all 4 cohorts in the Midland/Saginaw region while
Figure 2 only looks at the Floodplain cohort. Certainly, there is no 10-30 ppt TEQ spread in the
data shown in Figure 2. It is a curious result that merits further investigation - why would the
Floodplain as a cohort be similar to Jackson/Calhoun while the other 3 cohorts seem to
significantly drive up the concentrations by the overall Midland/Saginaw 4-cohort group?
Page 57

-------
So besides getting clarity on the entire Midland/Saginaw age range breakdown, as a
whole group and then by each of the four subgroups perhaps, an important next step in this
evaluation is to retrieve the data in detail so the blood levels of individuals in homes with high
soil concentrations can be separated from the rest of the population. Initial evaluations of this
type are provided in Appendix B of this report, where individuals living on properties with a soil
concentration greater than 1000 ppt TEQ were retrieved and studied. But it would appear that a
majority of individuals from the Floodplain did not live on elevated soils. As noted in
Hedgeman et al (2009), the median soil level in the floodplain is 11 ppt TEQ (CDD/F/PCB).
This is the same as the national background soil level used in the Reassessment and in these
simulations to characterize background exposures. Even though the median soil level in the
floodplain is the same concentration, it cannot be surmised that the median blood level
corresponds to the median soil level. To complete the evaluation, we need to retrieve the
detailed data and then we need to compare, by age range ideally, the blood concentrations of
individuals living in homes with elevated soils, perhaps 200 ppt and greater, with the model
predictions for individuals exposed to 223 ppt TEQ soils.
It is also noted that the median soil levels in the Jackson/Calhoun cohort is 3.6 ppt TEQ
(Hegeman et al, 2009). The difference in modeling impacts from living on 11.6 ppt TEQ versus
3.6 ppt TEQ is miniscule - there would be 0.1 ppt TEQ lipid difference in predicted lipid levels
(I did a quick test to confirm this). Again, the most valid comparison of modeling and measured
data can come once we retrieve the detailed data and look at blood levels only for individuals
living on soils with elevated dioxin levels.
Notwithstanding the absence of this more valid comparison, it can be stated that the
approach appears reasonable in its ability to capture the variation in a central tendency measure
of dioxin body burdens by age for a cross section of adult ages for a survey that might occur in
the 2000s. This statement is mainly supported by the comparison of model results with
NHANES as well as the Jackson/Calhoun referent population. This does not necessarily mean,
however, that the procedure to characterize exposures to dioxins in soil is accurate. Rather, it is
a statement that the general procedure to derive a history of total dose (including mainly food
intakes, but also soil intakes, inhalation, etc), and the impact of that dose on body burden over a
lifetime, might be reasonable.
Even though we do not yet have the more detailed and appropriate data to compare the
model results with those from the UMDES, we can still make some observations. It can be
stated, for example, that the modeled difference between body burdens of individuals exposed to
background only compared to background except with exposures to elevated soils (equal to the
95% soil found in the floodplains, at 223 ppt) appears to be only about 3 ppt TEQ body burden,
on a lipid basis. As was noted earlier, this finding may be conservative because it assumed that
all exposures in the past were proportionally higher, including soil exposures. A simple exercise
can be done to show that constant exposure to soils at 223 ppt TEQ would result in an increment
even less than 3 ppt TEQ. First, the aannual dioxin uptake from ingested soil is calculated as 223
pg/g * 0.05 g/d * 365d/yr * .8 absorption = 3256 pg/yr. The total amount of dioxin that would be
deposited into the body over 30 years is given as:
Page 58

-------
TD = (ANN/k) (1 - e"kt)
Where TD
total amount of dioxin remaining in the body over a period of exposure, pg
the annual intake at 3256 pg/yr
first order elimination rate, 0.0815/yr (given a half-life of 8.5 yr, Table 2)
years of exposure, yr
ANN
k
t
If we assume 30 years of exposure, the TD is solved above as 36486 pg. If this gets deposited
into an adult at 70 kg at about 0.25 body fat (see Table 2), than the resulting incremental body
lipid concentration would be 2.1 pg/g.
Questions remain that could be answered with PK modeling, with further evaluation of
data from the UMDES study area, or maybe never answered because specifically-desired data is
not available. For example, monitoring did not occur for young children. Figure 2 showing
differences for the 20 year-old may be useful to hypothesize about possible impacts to children.
It is seen that the peak body burden occurs for about a 5 year-old. At that age, the difference in
body burdens is about 6 ppt TEQ, at 23.4 ppt TEQ for the child living on elevated soil and 17.4
for the child living on background soils. Although the simulation was set up for the 5 year-old in
this figure to have turned 5 years old in 1993, in fact this is about what the model would predict
for an individual who was born in 2002 and turned 5 in 2007. This is because, as seen in Figure
1, background doses to dioxins were surmised to have largely diminished to near current levels at
about 1980. Therefore, the body burdens modeled in Figure 2 are nearly what would be
predicted to occur from birth to age 20 given today's background dose levels.
Another key question is, what might the impact be if living at much higher soil levels, say
at 1000 ppt TEQ? This can be initially answered with the PK modeling framework developed
here. Results for a 20 year-old using this framework, making all relevant changes to dose, are
shown in Figure 5. Here the peak concentration for the 5 year-old is 36.1 ppt TEQ, now more
than double that of a child of that age living in a background setting. Even at age 20, the
individual living on 1000 ppt soil has a body burden of 24.5 ppt, compared to 13.6 and 16.5 ppt
TEQ for the other two scenarios. This is yet another question that might be asked of the
UMDES data - we can look at blood levels of individuals who not only lived in soils greater than
223 ppt TEQ, but of only those who lived on soils greater than 1000 ppt TEQ.
This exercise was a start at using PK models to "ground truth" the body burden results
that have come out of the UMDES. If an acceptable measure of validity can be obtained with
use of the PK model, in addition to or other than what this exercise has already shown, than the
"bottom line" finding of the UMDES that living in soils elevated in dioxins would not greatly
impact body burdens, can be supported. Also, the model can be used to do further "what if'
scenarios and answer questions about exposure not necessarily captured with the monitoring
data. Two questions that this exercise has already looked at are the impacts to the 5 year-old and
impacts to having lived on soils as high as 1000 ppt TEQ.
Page 59

-------
REFERENCES
Demond, et al. 2008. Statistical Comparison of Residential Soil Concentrations of PCDDs,
PCDFs, and PCBs from Two Communities in Michigan. Environ. Sci. Technol., 2008, 42 (15),
5441-5448.
Hedgeman, et al. 2009. The University of Michigan Dioxin Exposure Study: Population
Survey Results and Serum Concentrations for Polychlorinated Dioxins, Furans, and Biphenyls.
EHP 117: 811-817.
Lakind, JS, SM Hays, LL Aylward, DQ Naiman. 2008. Perspective on serum dioxin levels in
the United States, an evaluation of the NHANES data. J. Exp. Sci and Env. Epi. 2008: 1-7.
Lorber M. 2002. A pharmacokinetic model for estimating exposure of Americans to dioxin-like
compounds in the past, present, and future. Science of the Total Environment 288:81-95.
Paustenbach, D.J.; Wenning, R.J.; Lau, V.; Harrington, N.W.; Rennix, D.K.; Parsons, A.H.
(1992) Recent developments on the hazards posed by 2,3,7,8-tetrachlorodibenzo-p-dioxin in
soil: implications for setting risk-based cleanup levels at residential and industrial sites. Journal
Toxicology and Environmental Health 36:103-149.
Scott LLF, KM Unice, P Scott, LM Nguyen, LC Haws, M Harris, D Paustenbach. 2008.
Addendum to: Evaluation of PCDD/F and dioxin-like PCB serum concentration from the 2001-
2002 National Health and Nutrition Examination Survey of the United States population. J. Exp.
Sci and Env Epi. 18: 524-532.
US EPA. 2003. Exposure and Human Health Reassessment of 2,3,7,8-Tetrachlorodibenzo-p-
Dioxin (TCDD) and Related Compounds. United States Environmental Protection Agency,
Office of Research and Development, National Center for Environmental Assessment. NAS
Review Draft. December, 2003. EPA/600/P-00/001C(a-f). Available at,
http://www.epa.gov/ncea/dioxin.htm.
University of Michigan (UM) 2006 . Measuring People Exposure to Dioxin Contamination
Along the Tittabawassee River and Surrounding Area's. Findings from the University of
Michigan Dioxin Study. August 2006 .
Page 60

-------
Table 1 from Lorber (2002)
Table 1
Summary of studies with body burden data of dioxiirj- ami furans
Y.ai
Mm TEQ,
p?/z lipid
N
Ages
Reference; Location.
19~2
:r-
19'5
iv'6
19"'"'
1?45 4
666
XA
Orban m al, 1594; EPA, 1991
19S8
50
26
XA
Selector et al, 1989; Massachusetts
1988
37
57
.7-50
i12-88)
Stanley et al., 1989; San Francises md Lei
Angeles. CA
1989
50
100
XA
Scliecter, I9T-1: Syistir.e, NT
1991
29
44
a ~ 48
Ml-661
Schecter «t . 199J: M:cl'.:g5.a Vietnam ntaa
1996
32
100
XA
Scliecter et u... iS97; Bmghamtccp, NY
1996
20
316
j = 45
123-70)
Personal communication bat D. Patterson,
Center for Disease Control, Atlanta, GA, to M.
Laarbei'. US EPA, Washington, DC. April. 2000;
background populations inn site-specific
stadias in MO, OR, WS» AK and NC.
1998
25
45
.7=45
(28-67)
Fetreas et al. 2600; San Francisco, CA
Page 61

-------
Table 2. Daily background dose, described in 5-year increments, for each of the individuals in
the 6 scenarios; dose in pg TEQ/day.

Scenario




4') \ ear-
2<> \ ear-
2<) \ ear-




old.
old.
old.

Mi \ ear-old.
(id \ car-old.
4<> \ ear-old.
ele\ a led
hkmnd
ele\ aled
Year
hkurnd soil
ele\aled soil
bkurnd soil
soil
soil
soil
1948
200
288




1953
282
376




1958
405
524




1963
544
685




1968
572
643
413
594


1973
396
424
290
360


1978
210
248
154
186


1983
96
113
90
108


1988
88
103
86
101
65
94
1993
79
94
78
92
63
78
1998
71
84
69
82
62
75
2003
66
78
66
78
63
76
2007
66
78
66
78
65
78
Table 3. Modeling parameters for pharmacokinetic modeling.
Time
IJodv
IJody
Ti:n Mall-life
After
\Yciulil
Lipid
(yrs)
1 ii nh
(ku)
Traction

At birth
3.3
0.14
0.40
1 year
11.3
0.23
1.06
2 years
13.3
0.20
1.72
5 years
19.7
0.15
3.70
12 years
41.1
0.15
7.12
19 years
65.1
0.13
7.54
35 years
71.5
0.21
8.50
60 years
73.8
0.27
10.00
Page 62

-------
Table 4. Final results of PK modeling exercise including lipid concentrations of 6 modeled individuals:
20, 40, and 60 year-old individuals living on background and elevated soil concentrations, compared
against the UMDES results (from Hedgeman et al, 2009), NHANES 2001-2002 (from Scott et al, 2008;
17 CDD/Fs, 9 PCBs), and NHANES 2003-2004 (Lakind et al, 2008).

Scrum Lipid Levels (ppt TKQ)
Description
Modeled
i Mi>i:s
Ml AMIS
20 yr old exposed to
background including
soils at 11.6 ppt TEQ
13.6
7.8
(median for 18-29 yr
olds in
Jackson/Calhoun)
13.5
(geometric mean for 20 -29
yr olds)
7.1 and 8.9 (median for 12-
19 and 20-39 in NHANES
2003-2004)
20 yr old exposed to
background but soils at
223 ppt
16.5
Comparable data not
yet available
No comparable data
40 yr old exposed to
background including
soils at 11.6 ppt TEQ
16.7
14.0
(median for 30-44 yr
olds in
Jackson/Calhoun)
17.1 (geometric mean for
30-44 yr olds in NHANES
2001-2002)
8.9 and 15.0 (median for
20-39 and 40-59 in
NHANES 2003-2004)
40 yr old exposed to
background but soil at 223
ppt.
20.0
Comparable data not
yet available
No comparable data
60 yr old exposed to
background including
soils at 11.6 ppt TEQ
19.4
20.8; 31.3
(median for 45-59 &
>60 yrs in
Jackson/Calhoun)
20,7, 33.7
(geometric mean for 45-59
and >60 yrs; NHANES
2001-2002)
15.0 and 26.9 (median for
40-59 and >60 in
NHANES 2003-2004 )
60 yr old exposed to
background but soils at
223 ppt
22.7
Comparable data not
yet available
No comparable data
Page 63

-------
Figure 1. The variation in adult exposures exposure levels as a function of time (from Lorber,
2002).
>	7
CO
-o	6
i	5
^	5
3	4
LU
° 1
 1
o _
1900 1920 1940 1960 1980 2000 2020
EPA (2000):
0.65 pg TEQ/kg-d
Page 64

-------
Figure 2. Difference in the body burden of a 20 year-old sampled in 2007 as a result of being
exposed to soil at two concentration levels, along with all other background exposures.
223 ppt TEQ
12 ppt TEQ
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Age, years
Page 65

-------
Figure 3. Difference in the body burden of a 40 year-old sampled in 2007 as a result of being
exposed to two different soil levels, along with all other background exposures.
140
O
W 120
CL
CL
100
c
o
CO
!	
-t—'
c
0
o
c
o
o
¦g
Q.
223 ppt TEQ
_i
12 ppt TEQ
5
10
15
20
25
30
35
40
Age, years
Page 66

-------
Figure 4. Difference in the body burden of a 40 year-old sampled in 2007 as a result of being
exposed to two different soil levels, along with all other background exposures.
223 ppt TEQ
10
20
30
40
50
60
Age, years
Figure 4. Difference in the body burden of a 40 year-old sampled in 2007 as a result of being
exposed to two different soil levels, along with all other background exposures.
Page 67

-------
Figure 5. Impact of being exposed to 1000 ppt TEQ rather than 223 or 12 ppt TEQ.
1000 ppt TEQ
223 ppt TEQ
12 ppt TEQ
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Age, years
Page 68

-------
APPENDIX B
EVALUATION OF THE BLOOD SERUM DATA FROM SUBJECTS LIVING ON
PROPERTIES WITH MAXIMUM SOIL TEQ VALUES GREATER THAN 1000 PPT .
In response to an EPA request, Dr. Garabrant provided (September 11, 2009 email from
Garabrant to Dr. Frithsen) a graph of the data presented in Figure 1 from Garabrant et al., 2009b
which had been modified to identify the data points associated with subjects living on properties
with a maximum soil TEQ concentration greater than 1000 ppt. The graph provided, labeled as
'Plot 1' is copied below. Data for subjects living on properties with maximum soil TEQ
concentration greater than 1000 ppt are identified in the plot by the large blue and red dots.
There are 23 of these in the data set.
Plot 1: Serum TEQDFP29-2005 by age, identifying subjects whose maximum soil TEQ DFP29-
2005 is > 1000 ppt. (without Clay Lady)
Q.
Q.
UO
O
o
CM
I
CD
CM
0-
O
LU
T3
o
o
CO
110
105-
100
95
90-
85:
SO
75
70
65
60
55
50
45
40
35]
30
25
20
15
10
5
01
20
•	• Female, Soil Max>=1000 (ppt)
•	" Male, Soil Max>=1000 (ppt)
*	* Female, Soil Max<1000 (ppt)
*	* Male, Soil Max<1000 (ppt)
	Female, BMI_center = 20
	Male, BMI_center = 20
	Female, BMI_centei = 40
	Male, BMI center = 40
* * * %*+ *	•* *E-
..•«»««	,* i ¦"***> ** * *
*****	*	*	®++t	*	*	+ **
h* *	**
* ** ** * *
30
40
50
Age
60
70
80
Notes:
Page 69

-------
1.	The maximum soil is the soil sample that has the highest TEQ of all soil
samples taken on the participant's property.
2.	This plot shows all data points except for the "Clay Lady" whose serum TEQ
was 210.7 ppt. She did not have a soil value > 1000 ppt. She was omitted from
the graph so that the Y axis was not compressed by her high value.
Visual examination of this plot indicates a pattern in the greater than 1000 ppt subset that
is distinct from the overall data set. That is, the relationship between serum levels and age in the
subset is visually different from the overall data in that the majority of the points lie above the
averages represented by the lines in the plot and appear to increase with age at a greater rate. In
stable populations, such as the communities in the UMDES, time in a residence would be
generally expected to increase with age which would suggest that age is likely to be a surrogate
measure for time lived on the property. This, in turn, suggests that these data are an indication of
an effect on serum levels due to extended exposure to high soil levels.
The specific data for the 23 subjects were not provided but it was possible to read
approximate values for these data from the plot. The blood serum and age values read from the
plot for the 23 subjects living on soils with greater than 1000 ppt TEQ concentrations are shown
below in the graph titled "Figure: Serum TEQ DFP29-2009 vs. Age for subjects whose
maximum soil TEQDFP29-2009 is > 1000 ppt". The line overlaying the plot is the least squares
fit to the data of the linear model
Y= Bo + BiX,
where Y= blood serum TEQ, X = Age, B0 = intercept and Bi = coefficient for Age.
Page 70

-------
Figure: Serum TEQ DFP29-2009 vs Age for subjects whose maximum soil
TEQDFP29-2009 is > 1000 ppt
100
•
Female
•
Male

Linear LS Fit

80
60
40
20
20
~T
40
Age
60
~T
80
The results are approximate because the data were read from the graphic provided by Dr.
Garabrant. Never the less, the estimated regression relationship between blood serum level and
age is strong. The regression is highly significant, and R-squared is 0.44. The statistics for this
regression are shown in Table 1, below. Garabrant et al. (2009b) reported an R2 of 0.396 for
serum levels versus a set of nine demographic factors of which age was only one, for the overall
data set. More importantly, the graph in the Figure shows a reasonable regression relationship in
the subset data, i.e., a relatively narrow pattern of data points clustered around an increasing
regression line. Regardless of the approximate nature of the data used in this analysis, there is an
apparent relationship in the subset that should be investigated further.
Page 71

-------
Table 1. Regression Statistics: Blood Serum TEQ versus Age for Adults Living on Properties
with Maximum Soil Concentrations Greater than 1000 TEQ


Regression Coefficients:
Estimate
Std. Error
t value
Pr(>|t|)
(Intercept)
-27.2347
14.5782
-1.868
0.075754
Age
1.0762
0.2505
4.296
0.000320 ***
Residual standard error: 12.43 on 21 degrees o!
¦ freedom
Multiple R-squared: 0.4677,
Adjusted R-squared: 0.4424


F-statistic: 18.45 on 1 and 21 DF, p-value: 0.0003203
*** Significant at less than 0.001 level
In the September 11, 2009 email, Dr. Garabrant also provided the graph shown below, labeled as
Plot 2 which shows residuals from the base regression model of serum TEQdfp29-2oos
versus maximum soil TEQdfp29-2oos. The residuals are the differences between the actual
measured blood values and the values predicted by the model. Notes for this plot included the
following statement: "This plot shows that after accounting for the factors in the base model,
there is no relationship between maximum soil TEQ and serum TEQ". The basis for this
statement is not clear from the information provided in the email although it would likely be
difficult to discern an effect in a subset of 23 subjects in an overall analysis of the data. It is
clear that the base model referred to in the email is somewhat different from that in Garabrant et
al. (2009b). It is also clear from the pattern of the residuals that the model tends to under predict
blood levels for the high soil concentration subset. Specifically, residuals for 15 of the 23 serum
levels predicted for subjects living on soils with greater than 1000 ppt concentration are greater
than zero. If the fitted model was correct, the residuals should be equally distributed about zero.
Furthermore, Plot 2 shows an overall pattern of increasing residuals as soil concentration
increases. This is an indication of a fundamental error in the regression model [e.g., Draper and
Smith, Applied Regression Analysis (1996)].
Page 72

-------
Plot 2: Bubble plot of studentized residuals from base regression model of serum TEODFP29-
2005 versus maximum soil TEODFP29-2005.
For serum TEQ2006 base model (R2= 65.5 %)
Ref Fu I I Model R 2 = 7 D.3 *

"O
O
E
a>
CO
CO
-Q
•—
o
M—
~aj
3
~0
V)
a>
on
Tj
a>
N
a>
Tj
CO
°0 ¦


WL ° o
¦L o
m /

o
H
H a
o

i)


0


2000 4000 6000 8000
Soil Max for TEQ2006
10000
12000
Notes:
1.	The maximum soil is the soil sample that has the highest TEQ of all soil samples
taken on the participant's property.
2.	The base model is the linear regression model in which logio(serum TEQ) is the
outcome variable, and age, age2, sex, BMi, pack-years of smoking, lifetime breast
feeding, sex by age interaction, and sex by BMI interaction are independent variables.
3.	The size of the bubble is proportional to the survey sample weight of the observation.
4.	This plot shows that after accounting for the factors in the base model, there is no
relationship between maximum soil TEQ and serum TEQ.
5.	This plot shows all data points including the "Clay Lady".
In his email of September 11, 2009, Dr. Garabrant also provided the graph shown below
identified as Plot 4 which shows the data from Figure 1 from Garabrant et al., 2009b modified to
Page 73

-------
highlight the subjects in the overall data living on properties with homogenized perimeter soil
samples that were measured at greater than 1000 ppt TEQ. The blood serum values for these six
subjects appear to be somewhat higher than the average but no analysis was attempted with these
data.
Plot 4: Serum TEQDFP29-2005 by age, identifying subjects whose house perimeter top 1
inch soil TEQDFP29-2005 is > 1000 ppt. (without Clay Lady)
•	• Female, Soil HP1>=1000 (ppt)
•	• Male, Soil HP1>=1000 (ppt)
*	* Female, Soil HP1<1000 (ppt)
*	* Male, Soil HP1<1000 (ppt)
—	Female, BMI_centei = 20
—	Male, BMI_center = 20
—	Female, BMI_center = 40
—	Male, BMI center = 40
* * =»*¦¦¦'*.		
# & f * * *
'V- - -w
***? ** %*
gp ** ** t ; % **
**
20
30
40
50
Age
_1—I—r
60
70
80
Notes:
1.	The house perimeter top 1 inch soil is the homogenized sample of the top 1
inch soil from all soil cores taken on four sides around the house on the
participant's property.
2.	This plot shows all data points except for the "Clay Lady" whose serum TEQ
was 210.7 ppt. She did not have a soil value > 1000 ppt. She was omitted from
the graph so that the Y axis was not compressed by her high value.
Page 74

-------