OCR error (C:\Conversion\JobRoot\00000CDU\tiff\20013KK3.tif): Unspecified error
-------
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
C«K«.««. 00 4nnt OFFICE OF THE ADMINISTRATOR
February 22, 1993 SCIENCE ADVISORY BOARD
EPA-SAB-EHC-93-007
Honorable Carol M. Browner
Administrator
U.S. Environmental Protection Agency
401 M Street, S.W
Washington, DC 20460
Subject: Science Advisory Board Review of the Office of Solid Waste and
Emergency Response draft Risk Assessment Guidance for Superfund (RAGS),
Human Health Evaluation Manual (HHEM),
Dear Ms. Browner:
Early in the implementation of the Comprehensive Emergency Response
Compensation and Liability Act, the EPA's Office of Solid Waste and Emergency
Response (OSWER) decided to rely heavily on site-specific assessments of human and
environmental risk to determine the need for remedial action, and to set protective
cleanup levels. This approach was utilized because of the substantial differences in
land use activities, terrain, hydrogeology, and nature and extent of contamination from
site to site.
OSWER developed risk assessment guidance for Superfund to increase
consistency in the way risk assessments are conducted within and across EPA
Regional offices. The health risk assessment guidance for the Superfund program is
codified in several documents, hereafter referred to collectively as the Risk Assessment
Guidance for Superfund, Human Health Evaluation Manual (RAGS/HHEM). To ensure
that the final RAGS/HHEM document reflects state-of-the-art technical guidance, and to
comply with the recommendations of the Superfund 30-Day Study Task Force that
OSWER should seek internal and external review of the Superfund risk assessment
guidance, OSWER officials requested that the Science Advisory Board review selected
issues addressed by the RAGS/HHEM document. Consequently, the Science Advisory
-------
Board's Environmental Health Committee (EHC) met on April 7-8, 1992 in Bethescz
Maryland to review four broad issues relating to Superfund human health risk asses -
ment:
*v
a) Defining and calculating the Reasonable Maximum Exposure (RME) -- a
conservative exposure case (i.e., well above the average) that is still
within the range of possible exposures.
b) Assessing and dealing with exposures to multiple chemicals -- Using the
Hazard index (HI)/Hazard Quotient (HQ) to assess risk.
c) Reference doses in goal-setting - use of chronic/sub-chronic RfDs and
specific populations to set risk-driven remediation goals.
d) Short-term toxicity values - use of appropriate defaults for characterizing
less-than-lifetime exposure to toxicants.
The Committee found OSWER's attempts to improve the consistency of its risk
assessment methodology praiseworthy. The OSWER has (understandably, given the
range and complexity of the issues addresed) not succeeded fully in meeting their
goals, but has made a good start. The Committee's findings note where a redirection m
approach is called for, and provide advice where possible.
The Committee is of the opinion that there are some serious conceptual and
practical problems with the proposal to calculate an RME based :n an upper confi-
dence limit (UCL) on the average concentration at a site. Given the proposed methods
for computing the UCL, its underlying statistical assumptions, and problems in dealing
with the spatial distribution of contamination v/a-a-v/s the relative frequency with which
people are likely to visit various parts of the site, the resulting UCL may have little, if
any, relation to actual concentrations to which people may be exposed at a site. The
Committee recommends that the EPA move to a distributional approach for calculating
the RME, i.e., develop distributions for each of the terms or variables needed to
calculate individual exposures and their distributions. These distributions determine a
subsequent distribution for exposure, which can be calculated using Monte Carlo
methods. A particular percentile of this exposure distribution, such as the 90th percen-
tile, could then be used as the definition of the RME. To implement such an approacn.
EF \ should develop default distributions for exposure parameters unlikely to vary
-------
significantly from site to site. In situations where site-specific conditions may be ur
however, the collection of site-specific data is encouraged. Until this approach can *
put into use, a modification of the current approach may have to be used. The
Committee agrees with OSWER that, as long as some type of mean concentration is u.
be employed to estimate human exposure, an arithmetic mean is more appropriate than
a geometric mean.
. The Committee is concerned about the approach of using Reference Dose (RfD)-
derived Hazard Quotients/Hazard Indices as a basis for adding "risks" from exposure to
complex mixtures; it is not truly risks which are being added when the proposed ap-
proach is used. Quantitative applications using dose-response data (not the "point"
data represented by Lowest Observed Adverse Effects Level/No Observed Adverse
Effects Level (LOAEUNOAEL)-derived RfDs) would be preferable, as would the use of
alternatives to the current default approaches that assume risk additivity. The use of
the HI itself can be misleading, and it should be used as a "fallback," with full recogni-
tion of its flaws, only when more refined lexicological data are not available.
The HHEM recommends using the RfD as the toxicity criterion for each of the
other effects believed to be caused by a given agent in a chemical mixture. This
proposal does not deal with effect interactions, nor with the fact that many RfDs are
based on non-specific endpoints which can stem from many different causes. It is not
nearly as seriously flawed as the alternatives presently in use, however, and cjespite its
flaws is an improvement.
Three approaches for using RfDs to develop risk-based remediation goals for
contaminated soil, involving differing exposure scenarios and target populations, were
presented. The most supportable of these uses a 30-year time-weighted average with
a chronic RfD; differences between the three approaches are not dramatic however,
and OSWER should study all three approaches to verify its ultimate choice (or range of
choices).
The Committee sees no particular problems in the existing approach for dealing
with short-term toxicity estimates. OSWER should take cognizance of the EPA-spon-
sored work at the National Academy of Sciences on Community Emergency Exposure
Levels, and of the work on Emergency Response Planning Guidelines by the American
Conference of Governmental Industrial Hygienists.
-------
We appreciate having been given the opportunity to address these issues, and
look forward to receiving your response to our comments.
»
Sincerely,
Dr.^Raymond Loehr; uhair
Science Advisory Board
Dr. Arthur C. Upton, Chair
Environmental Health Committee
-------
ABSTRACT
The Office of Solid Waste and Emergency Response (OSWER) developed the
Risk Assessment Guidance for Superfund (RAGS), Human Health Evaluation Manual
(HHEM), Part A-Baseline Risk Assessment (December 1989), supplemented in
March 1991 with Standard Default Exposure Factors guidance, and Part B-Develop-
ment of Risk-based Preliminary Remediation Goals, and Part C-Risk Evaluation of
Remedial Alternatives in December 1991 (asjnterim documents) to guide Agency staff
performing site-specific assessments of human and environmental risk to determine
the need for remedial action. At the request of OSWER, the Science Advisory Board's
Environmental Health Committee (EHC) met on April 7-8, 1992 to review four broad
issue areas relating to Superfund human health risk assessment: a) Defining and
calculating the Reasonable Maximum Exposure (RME); b) Assessing and dealing with
exposures to multiple chemicals and using the Hazard index (HI)/Hazard Quotient
(HQ) to assess risk; c) reference doses in remediation goal-setting; and d) Use of
appropriate defaults for characterizing less-than-lifetime exposure to toxicants. The
Committee found OSWER's attempts to improve the consistency of its risk assess-
ment methodology to be praiseworthy and a good start, but noted areas where a
revised approach is recommended.
The Committee is of the opinion that there are some serious conceptual and
practical problems with the proposal to calculate an RME based on an upper confi-
dence limit (UCL) on the average concentration at a site. The EHC recommends that
the EPA move to a distributional approach to calculating the RME, i.e., developing
distributions for each of the terms or variables needed to calculate individual expo-
sures and their distributions. Given the difficulty in interpreting the RME as presently
calculated, the Committee recommends that some type of 'most reasonable' estimate
of exposure also be calculated and made available to risk managers along with the
RME. The Committee agrees with OSWER that, as long as some type of mean
concentration is to be employed to estimate human exposure, an arithmetic mean is
more appropriate than a geometric mean.
The Committee is concerned about the approach of using RfD-derived Hazard
Quotients/Hazard Indices as a basis for adding "risks" fr-om exposure to complex
mixtures. Quantitative applications using dose-response data (not the "point" data
represented by LOAEL/NOAEL-derived RfDs) would be preferable, as would the use
of alternatives to the current default approaches that assume risk additivity. The use
of the HI itself can be misleading, and it should be used as a "fallback," with full
recognition of its possible inapplicability, only when more refined lexicological data are
not available. Interpretation of an HI greater than 1 can vary depending on several
lexicological factors. Although it is likely that risk increases as the HI exceeds 1, we
can not state (without a more complete understanding of interaction mechanisms) how
rapidly this increase occurs, nor can we rely on Hl-based comparisons of risks when
the His are greater than 1. -
-------
Three approaches for using RfDs to develop risk-based remediation goals for
contaminated soil were presented. The most supportable of these uses a 30-year
time-weighted average with a chronic RfD; differences between the three approaches
are not dramatic however, and OSWER should study all three approaches to verify its
ultimate choice (or range of choices).
The Committee sees no particular problems in the existing approach for dealing
with short-term toxicity estimates. OSWER should take cognizance of the EPA-spon-
sored work at the National Academy of Science on Community Emergency Exposure
Levels, and of the work on Emergency Response Planning Guidelines by the Ameri-
can Conference of Governmental Industrial Hygienists.
KEYWORDS: complex mixtures; exposure; Hazard Index (HI); Hazard Quotient (HQ);
reasonable maximum exposure; risk assessment; Reference Dose (RfD); Super-fund;
site assessment
-------
NOTICE
This report has been written as a part of the activities of the Science Advisory
Board, a public advisory group providing extramural scientific information and advice
to the Administrator and other officials of the Environmental Protection Agency. The
Board is structured to provide balanced, expert assessment of scientific matters
related to problems facing the Agency. This report has not been reviewed for
approval by the Agency and, hence, the contents of this report do not necessarily
represent the views and policies of the Environmental Protection Agency, nor of other
agencies in the Executive Branch of the Federal government, nor does mention of
trade names or commercial products constitute a recommendation for use.
HI
-------
ENVIRONMENTAL PROTECTION AGENCY
SCIENCE ADVISORY BOARD
ENVIRONMENTAL HEALTH COMMITTEE
''April 7-8, 1992
CHAIRMAN
Dr. Arthur C. Upton, New York University Institute of Environmental Medicine, New
York City, NY
MEMBERS & CONSULTANTS
Dr. William Bunn, Mobil Oil, Princeton, NJ
x
Dr. Kenny Crump, Clement International Corporation, Ruston, LA
Dr. Paul Deisler, Retired, Shell Oil, Houston, TX
Dr. Adam Finkel, Resources for The Future, Washington, DC
Dr. David Gaylor, National Center for Toxicological Research, Jefferson, AR
Dr. Rolf Hartung, University of Michigan, Ann Arbor, Ml
/
Dr. Marshall Johnson, Jefferson Medical College, Philadephia, PA
Dr. Nancy K. Kim, New York Department of Health, Albany, NY
Dr. Rogene Henderson, Lovelace Inhalation Toxicology Research Institute, Albuquer-
que, NM
Dr. Richard Monson, Harvard University School of Public Health, Boston, MA
Dr. Martha J. Radike, University of Cincinnati Medical Center, Cincinnati, OH
Dr. Bernard Weiss, University of Rochester School of Medicine, Rochester, NY
Dr. Ronald Wyzga, Electric Power Research Institute, Palo Alto, CA
DESIGNATED FEDERAL OFFICIAL
Mr. Samuel Rondberg, Environmental Health Committee, Science Advisory Board.
U.S. Environmental Protection Agency
STAFF SECRETARY
Ms. Mary L. Winston, Science Advisory Board, U.S. Environmental Protection Agency
iv
-------
ASSISTANT STAFF DIRECTOR
Mr. Robert Flaak, Science Advisory Board, U.S. Environmental Protection Agency
STAFF DIRECTOR
Dr. Donald G. Barnes, Science Advisory Board, U.S. Environmental Protection
Agency
-------
TABLE OF CONTENTS
1. EXECUTIVE SUMMARY 1
2. INTRODUCTION > 5
2.1 Background 5
2.2. Charge To The Committee 6
3. SPECIFIC FINDINGS 11
3.1 Reasonable Maximum Exposure-Issue One 11
3.1.1 Arithmetic Mean Concentration 11
3.1.2 Use of the 95% Confidence Limit/Estimating Exposure-
Issues Two and Five 14
3.1.3 Characterizing Contaminant Concentrations-Issue Three ... 18
3.1.4 Determining Exposure Factors--! ssue Four 18
3.2 Chemical Mixtures 19
3.2.1 Additivity of Risk-Issue Six 19
3.2.2 Hazard Indexes and Thresholds of Concern-Issue Seven ... 20
3.2.3 Interpretation of the Hazard Index-Issue Eight 22
3.2.4 The RfD As a Criterion of Toxicity-lssue Nine 23
3.3 RfDs in Goal Setting 28*
3.3.1 Exposure Scenarios-Issue Ten 28
' 3.4 Short-Term Toxicity Values 29
3.4.1 Interim Estimates of Toxicity-lssue Eleven 29
3.4.2 Short-Term Air Action Levels-Issue Twelve 30
4. CONCLUSIONS .. . . . 32
5. REFERENCES R-1
APPENDIX A - Characteristics of the Hazard Index A-1
APPENDIX B - GLOSSARY .' A-4
VI
-------
1. EXECUTIVE SUMMARY
Early in the implementation of the Comprehensive Emergency Response
Compensation and Liability Act (CERCLA), the EPA's Office of Solid Waste and
Emergency Response (OSWER) decided to rely heavily on site-specific assessments
of human and environmental risk to determine the need for remedial action. OSWER
developed the Risk Assessment Guidance for Superfund (RAGS), Human Health
Evaluation Manual (HHEM), Part A-Baseline Risk Assessment (December 1989),
supplemented in March 1991 with Standard Default Exposure Factors guidance, and
Part B-Development of Risk-based Preliminary Remediation Goals, and Part C--Risk
Evaluation of Remedial Alternatives in December 1991 (as interim documents).
At the request of OSWER program officials, the Science Advisory Board's
Environmental Health Committee met on April 7-8, 1992 to review four broad issue
areas relating to Superfund human health risk assessment:
a. Defining and calculating the Reasonable Maximum Exposure (RME) --
The goal of the RME is to estimate a conservative exposure case (i.e.,^^.
well above the average case) that is still within the range of possible
exposures.
b. Assessing and dealing with exposures to multiple chemicals - Using the
Hazard Index (HI)/Hazard Quotient (HQ) to assess risk.
c. Reference doses in goal-setting - use of chronic/sub-chronic Reference
Doses (RfDs) and specific populations to set risk-driven remediation
goals.
d. Short-term toxicity values - use of appropriate defaults for characterizing
less-than-lifetime exposure to toxicants.
The Committee found OSWER's attempts to improve the consistency of its risk
assessment methodology praiseworthy. The OSWER has (understandably, given the
range and complexity of the issues addressed) not succeeded fully in meeting their
goals, but has made a good start. The Committee's major findings follow, specifically
noting where a redirection in approach is called for, and providing advice where
possible.
-------
The Committee is of the opinion that there are some serious conceptual and
practical problems with the proposal to calculate an RME based on an upper confi-
dence limit (UCL) on the average concentration at a site. The current approach
assumes that the samples taken are representative of those areas where exposures
are most likely to occur. The RME'ls also a function of the number of samples
available; a larger number of samples will result in a smaller RME even if the samples
are not representative of exposure opportunities. This approach does not deal with
"hot spots" at a site which could cause visitors/residents to be exposed to levels
significantly higher than the UCL. Given the proposed methods for computing the
UCL, its underlying statistical assumptions, and problems in dealing with the spatial
distribution of contamination via-a-vis the relative frequency with which people are
likely to visit various parts of the site, the resulting UCL may have little, if any, relation
to actual concentrations to which people may be exposed at a site. The Committee
recommends that the EPA move to a distributional approach to calculating the RME,
i.e., developing distributions for each of the terms or variables needed to calculate
individual exposures and their distributions. These distributions determine a distribu-
tion for exposure, which can be calculated using Monte Carlo methods. A particular
percentile of this exposure distribution, such as the 90th percentile, could be used av**
the definition of the RME. Kriging and triangulation are two statistical methods for
quantifying spatial distributions of contaminant concentrations which could be used as
part of this approach to address the issue of hot spots.
To implement a distributional approach, EPA should develop default distribu-
tions for exposure parameters unlikely to vary significantly from site to site. The
collection of site-specific data is encouraged in instances where site-specific conditions
may be unique, however. Until this approach can be put into use, a modification of
the current approach may have to be used. Given the difficulty in interpreting the
RME as presently calculated, the Committee recommends that some type of 'most
reasonable1 estimate of exposure also be calculated and made available to risk
managers along with the RME. The Committee agrees with OSWER that, as long as
some type of mean concentration is to be employed to estimate human exposure, an
arithmetic mean is more appropriate than a geometric mean.
The Committee is concerned about the approach of using RfD-derived Hazard
Quotients/Hazard Indices as a basis for adding "risks" from exposure to complex
mixtures; it is not truly risks which are being added when the proposed approach is
used. Quantitative applications using dose-response data (not the "point" data
-------
represented by LOAEL/NOAEL-derived RfDs) would be preferable, as would the use
of alternatives to the current default approaches that assume risk additivity.
The use of the HI itself can be misleading, and it should be used as a
"fallback," with full recognition of its possible inapplicability, only when more refined
toxicological data are not available. The condition "HI = 1" defines a "threshold of
concern" that is not shared by any other value for HI, and for which, under specified
conditions, the uncertainty in the HI approach is no greater than that of the component
RfDs. The HI approach is invalid, however, if the chemicals in the mixture cannot be
fully characterized by a combination of dilution-type interactions and independent
mechanisms of action. Interpretation of an HI greater than 1 can vary depending on
several toxicological factors. Although it is likely that risk increases as the HI exceeds
1, we can not state (without a more complete understanding of interaction mecha-
nisms) how rapidly this increase occurs, nor can we rely on Hl-based comparisons-of
risks when the His are greater than 1.
The HHEM recommends using the RfD as the toxicity criterion for each of the
other effects believed to be caused by a given agent in a chemical mixture. This ,~^
proposal does not deal with effect interactions, nor with the fact that many RfDs are
based on non-specific endpoints which can stem from many different causes. It is not
nearly as seriously flawed as the alternatives presently in use, however, and despite
its flaws is an improvement.
Three approaches for using RfDs to develop risk-based remediation goals for
contaminated soil, involving differing exposure scenarios and target populations were
presented. The most supportable of these uses a 30-year time-weighted average with
a chronic RfD; differences between the three approaches are not dramatic however,
and OSWER should study all three approaches to verify its ultimate choice (or range
of choices).
The Committee sees no particular problems in the existing approach for dealing
with short-term toxicity estimates. OSWER should take cognizance of the EPA-spon-
sored work at the National Academy of Science on Community Emergency Exposure
Levels, and of the work on Emergency Response Planning Guidelines by the Ameri-
can Conference of Governmental Industrial Hygienists. The method proposed by
Region 6 for setting short-term air action levels calls for the possible use of OSHA
standards, such as Permissible Exposure Limits (PELs) and Short-Term Exposure
-------
Levels (STELs). The data used to derive any OSHA standard so used should be
examined to see if the same data can be used to derive an appropriate RfD. The use
of EPA derivation methods would help promote consistency across various hazardous
substances; hence the use of EPA^ methods should be encouraged. Where this is not
possible or practical, the use of health data on which OSHA standards have been
based could be considered, taking into account differences between the worker
population and the general population.
-------
2. INTRODUCTION
2.1 Background
Risk assessment is an essential component of the Superfund site remediation
process. Early in the implementation of the Comprehensive Emergency Response
Compensation and Liability Act (CERCLA), the EPA's Office of Solid Waste and
Emergency Response (OSWER) decided to rely heavily on site-specific assessments
of human and environmental risk to determine the need for remedial action, to identify
contaminants of concern and critical exposure pathways, and to determine protective
cleanup levels. This approach was utilized because of the substantial differences in
land use activities, terrain, hydrogeology, and nature and extent of contamination from
site to site. OSWER believed that decisions regarding the protectiveness of contami-
nant concentrations in the environment were best made on the basis of specific site
circumstances.
OSWER developed risk assessment guidance for Superfund to reflect extensive
experience obtained from conducting health and environmental risk assessments at**""
Superfund sites, utilizing existing Agency risk assessment methods and data bases.
The guidance is designed to increase consistency in the way risk assessments are
conducted within and across EPA Regional offices. The health risk assessment
guidance for the Superfund program is presented in several documents. Risk
Assessment Guidance for Superfund (RAGS), Human Health Evaluation Manual
(HHEM), Part A--Baseline Risk Assessment, was published as interim final in Decem-
ber 1989. This Part was supplemented in March 1991 with Standard Default Expo-
sure Factors guidance (OSWER Directive 9285.6-03). Two additional Parts of the
Human Health Evaluation Manual strengthen the relationship among human health
evaluation, cleanup goals, and remedy selection. Part B-Development of Risk-based
Preliminary Remediation Goals, and Part C--Risk Evaluation of Remedial Alternatives
were published as interim documents in December 1991. In the near future, all three
Parts of HHEM will be integrated into a single final HHEM document that will incorpo-
rate new policy information and other technical guidance that has been issued by the
Agency since 1989, as well as comments received from field users.
Unlike risk assessments conducted by other EPA Program Offices that often
focus on single contaminants or single exposure pathways, Superfund assessments
must address multiple contaminants and multiple pathways for each site on the
National Priorities List (currently more than 1200 sites). The Superfund program deals
-------
with diverse problems including waste from spills, illegal dumping, landfills, surface
impoundments, drum/container storage, as well as other contaminant sources. The
Superfund program typically conducts risk assessments at well over 100 sites per
year, in all regions of the country, gnder varied environmental conditions, and for
multiple land uses. There is strong public and political pressure on this program to
address these sites as quickly as possible. Efforts to characterize baseline health
risks at a site typically take two to three months once data are in hand, but are
sometimes delayed because of the need to collect better sampling data, or negotia-
tions with "potentially responsible parties" over land use, exposure assumptions, and
chemical toxicity.
To address the needs of Superfund risk assessors, Superfund's risk assess-
ment guidance must be flexible enough to encompass the wide variety of conditions
present at sites in all Regions yet specific enough to assure a reasonable degree of
consistency. Assessors must balance the pressures to gather additional data (usually
resource intensive) with the need to move quickly toward site cleanup. Risk asses-
sors and site managers are often required to make real-time cleanup decisions at
Superfund sites with imperfect data. """
' In the interest of continuous improvement, to ensure that the final RAGs-HHEM
document reflects state-of-the-art technical guidance, and to comply with the recom-
mendations of the Superfund 30-Day Study Task Force that OSWER should seek
internal and external review of the Superfund risk assessment guidance, program offi-
cials requested that the Science Advisory Board review selected issues addressed by
the Risk Assessment Guidance documents noted above.
2.2. Charge To The Committee
Four broad issue areas relating to Superfund human health risk assessments
were identified for review; within each issue, specific questions were posed to the
Committee.
a) Reasonable Maximum Exposure: Superfund's approach for calculating
the Reasonable Maximum Exposure (RME) is presented in HHEM Part A
(Chapters 6 and 8) and in its supplement "Standard Default Exposure
Factors" (OSWER Directive 9285.6-03). The goal of the RME is to
6
-------
estimate a conservative exposure case (i.e., well above the average
case) that is still within the range of possible exposures.
1) The HHEM Part A identifies the arithmetic mean concentration,
within an appropriate averaging zone, as the measure of concen-
tration that generally represents the integrated long-term exposure
that will be received by an individual within that zone. Is this
approach reasonable?
2) Part A also indicates that given the typical distribution and spa-
tial/temporal variability of contaminant concentration and the often
limited number of samples collected at Superfund sites, it is
important that the health risk assessments explicitly address
uncertainty in the mean concentration. Do 95% confidence limits
on the mean concentration provide an appropriate tool for ad-
dressing uncertainty in concentration?
3) Superfund is currently investigating alternate approaches for
characterizing contaminant concentrations, such as kriging and
triangulation methods. Are these alternative approaches for esti-
mating average concentrations (or others) appropriate to consider
for future guidance?
4) Superfund guidance indicates that valid site-specific information on
exposure factors - particularly human behavior patterns -- be
used in exposure assessments. In the absence of site-specific
survey data, or in cases where the assessment must determine
projected changes in land use, guidance has relied on survey data
for other populations, commonly at the national level. Is this
approach reasonable?
5) The Agency's new "Guidelines for Exposure Assessment" (Section
5.3.5.1) discuss three approaches towards making estimates of
exposure for highly exposed individuals. First, upper percentiles
of population exposure may be directly determined from surveys
where direct- measurement data on exposure were obtained.
Second, mathematical simulation techniques can be used to
-------
combine adequate distributional data of individual exposure factors
to estimate the overall distribution of exposure. Third, an estimate
may be constructed using a combination of upper-bound and mid-
range values for exposure factors. In this case, the few most
variable factors would generally be set to upper-range values and
the remaining factors set to mid-range values. Experience indi-
cates that under appropriate circumstances each of the appro-
aches can provide useful information on Superfund site risk. How-
ever, since data on exposed populations are limited, HHEM Part A
provides additional details for the third tier situation. Is the Super-
fund approach of combining high-end and mid-range values pre-
sented in HHEM Part A (Chapter 6, Sections 6.1.2 and 6.4) con-
sistent with the third approach for estimating high-end exposures
mentioned above and described in the "Guidelines for Exposure
Assessment" (Section 5.3.5.1)?
b) Exposures to Multiple Chemicals: The current approach to assessmenj^of
hazard or risk of concurrent exposure to multiple chemicals (HHEM Part
A, Chapter 8), is based on the assumption of dose additivity, as recom-
mended in EPA'S "Guidelines for Health Risk Assessment of Chemical
Mixtures." For carcinogens, simple additivity of risk is used. For non-
carcinogens, a Hazard Index (HI) approach has been developed, in
applying the HI approach, the potential for non-carcinogenic effects is
evaluated by comparing estimated exposure to a reference dose (RfD).
The resulting ratio for each chemical is called a hazard quotient, if the
sum of these hazard quotients (called a hazard index or HI) for several
chemicals with the same toxic endpoint exceeds unity, there is a concern
for potential adverse effects.
6) Given what is currently understood about the potential interactive
effects of chemicals, is it appropriate to add risk estimates for
multiple contaminant exposures (i.e., calculation of Hazard indices
for chemicals with similar toxic endpoints, and simple additivity of
cancer risks)?
8
-------
7) Given the uncertainties in determining His, is it appropriate
to use HI values greater than a specific and constant num-
ber (i.e., "1") as a threshold of concern?
8) What does an HI greater than 1 represent when the hazard quo-
tients used to calculate the HI are individually less than 1 (i.e.,
when the estimated "dose" from an individual chemical does not
exceed the RfD)?
9) Existing guidance specifies that chemicals should be grouped
according to their major toxic effects, including those seen at
doses or exposures higher than those associated with the critical
effect (upon which the RfD was based). As a conservative and
simplifying step, the guidance recommends that the RfD be used
as the toxicity criterion for each of the other effects believed to be
caused by that chemical. Is this a reasonable approach? The
issue paper provided to the Committee (Derivation of Effect-Spa^,
cific RfDs and Their Use in Risk Assessment for Chemical Mixt
ures) presents an alternative approach. Is this approach (or are
others) reasonable to consider for future guidance?
c) Reference Doses in Goal-setting: During the development of the HHEM
Part B, two approaches were considered for using RfDs in setting risk-
based remediation goals in soil: 1) comparison of a 6-year, childhood
exposure to contaminants in soil with a sub-chronic RfD; and, 2) compar-
ison of a 30-year, time-weighted average exposure to contaminants in
soil (including exposures to both children and adults) with a chronic RfD
At the time, the second approach was chosen, as it provided a more
conservative cleanup goal. Since that time, a third approach has been
proposed: comparison of a 6-year, childhood exposure with a chronic
RfD.
10) Given EPA's definition of an RfD [i.e., "an estimate (with uncer-
tainty spanning perhaps an order of magnitude) of a daily expo-
sure to the human population (including sensitive subgroups) that
is likely to be without an appreciable risk of deleterious effects
during a lifetime"], what is the panel's opinion regarding the
-------
appropriateness of the second approach? Under what conditions should
other approaches be considered?
d) Short-term Toxicitv VaJues: Current guidance in HHEM Part C (Appendix
C, p. 49) refers risk assessors to the Superfund Technical Support
Center at ECAO-Cincinnati to obtain any toxicity criteria needed for risk
assessment of less-than-lifetime exposures. Since data on assessing
short-term exposure risks are extremely limited and no Agency-wide
guidance is available, ECAO often must derive interim toxicity criteria
based on the methods outlined in the Interim Methods for Development
of Inhalation Reference Concentrations (p. 4-37).
11) In the absence of chemical- and duration-specific data, is this
method reasonable? Are alternate procedures available? -
There is currently no national guidance for setting short term air action
levels that would guide activity (and emergency shutdown) during remefc
dial action at Superfund sites. EPA Regions are currently using various
approaches for deriving such levels. For example, Region 6 has issued
a policy statement that discusses the derivation of air action levels from
EPA's chronic health risk values; from OSHA Permissible Exposure
Limits, Short Term Exposure Levels, or Ceiling Values; or from RfD/RfC
values.
12) Is the Region 6 approach an appropriate method for deriving
short-term air action levels? Is there a more appropriate method
for developing short-term air action levels that could be used to
trigger shutdown of cleanup operations and/or the evacuation of
the general population near a Superfund site should unanticipated
releases occur (e..g., toxicity values relevant to 15-minute human
exposures)?
In addition to the specific questions noted above, the OSWER requested that
this review take into consideration the context within which these issues will be dealt
with operationally, i.e., the pressures and real-time decisions that must be made at
Superfund sites.
10
-------
3. SPECIFIC FINDINGS
3.1 Reasonable Maximum Exposure-Issue One
3.1.1 Arithmetic Mean Concentration
It is difficult for the Committee to evaluate whether OSWER's use of the
arithmetic mean concentration is appropriate, because OSWER has neither adequate-
ly explained the rationale for moving from an upper-bound estimate of concentration to
a mean estimate (and the ramifications thereof), nor sufficiently documented over what
measurements the mean is to be estimated. The following briefly discusses each of
these two issues in turn.
a) The "appropriate" mean
OSWER is attempting to account for several fundamentally different
kinds of uncertainty or error with one rather vague policy. The combination of
"reasonable high-end" values for those parameters that vary across individual?*'
(e.g., breathing rate, ingestion rates, body weight, contact rate, exposure
frequency, exposure duration), with a mean concentration within an "appropriate
averaging zone" co-mingles different kinds of variability (inter-individual versus
spatial/temporal) with various sources of uncertainty into a single measure that
is supposed to be "conservative" but not wildly so. This is a worthwhile end.
but the means to that end are arbitrary and unverified, and this measure has no
consistent interpretation. First, OSWER needs to show (preferably by Monte
Carlo simulations using both actual and stylized data sets) that the short-hand
combination of "high-end" default values and mean concentration does in fact
yield a reasonably conservative output1. More importantly, however, OSWER
must demonstrate that the current spatial average (which needs to be defined)
is in fact the long-term average for the "average individual" (or whatever
individual it intends to model).2
'Such • demonstration should at l«a»t attampt to model the likely correlation* among the behavioral variable* (both positive
and negative) and the possible correlation* between the behavioral and concentration inputs.
1 The burden of proof or of validation « definitely on EPA in thi* instance. Intuition tell* us that the current spatial average
itself is likely to change over time; to this, EPA add* the additional assumption that spatial average* and inter-individual lifetime
average* are related
-*
11 '
-------
Clearly, part of this demonstration will come about from a better definition
of "appropriate averaging zone." In addition, OSWER must also investigate
whether even an "appropriate" spatial average yields a representative long-term
concentration estimate for the relevant person or group. In other words, in
addition to needing to specify over what area a person or group's "random
walk" is occurring, OSWER needs to justify that the "walk" is in fact random.
This might be more or less true if the issue was sporadic contact with an
industrial site, but it may be inappropriate to average over a broad "averaging
zone" in the case of widespread contamination in a residential area. Here the
only "random walk" would be over small portions of each homeowner's proper-
ty, not over an entire neighborhood.
Consequently, we do not believe th-at the proposed methods are consis-
tent with the Guidelines for Exposure Assessment, particularly the proposed
approach for estimating the appropriate concentration term in the RME. Using
the mean, even allowing for its upper 95% confidence interval, does not
estimate the reasonable maximum concentration to which an individual is
exposed, but merely gives a confidence limit on the "average" site-wide concwrr*
tration over the time period during which measurements were taken-jf one
assumes that the sites sampled were representative of the site of concern. If
the sites sampled are not random, or are not part of a systematic design to
characterize the site as a whole, we do not know how to interpret either the
resulting average concentration or its upper 95% confidence interval, and we
have no idea how it relates to the appropriate RME concentration.
Me other technical question also needs to be addressed. It relates to
the I ations of Haber's Rule3, which forms the conceptual basis for time-
weighted averaging. An implication of Haber's Rule is that the total amount of
a dose governs the effect, and that the time-course of administration is irrele-
vant. Haber's Rule tends to work over relatively narrow ranges of exposures
and duration, but, as the ranges increase, it begins to fail more and more.
Because it does not take into account pharmacokinetics and biotransformations,
its applications are limited, and the Agency should consider these limitations
carefully whenever it proposes the use of time-weighted averages in its risk
assessment procedures.
1 Habam Rule: W» K x C x T, wher« W« Wlrfcung (a con»tant effect); K* a proportionality constant; C» concentration or
dote/unX time; and T« duration of «xpo«ur«.
12
-------
To summarize, OSWER may have modeled the randomness of exposure
to hazardous waste sites incorrectly, and could probably do a much better job
by at least explaining whose behavior and exposures are being modeled, and
why. The above issues notwithstanding, if EPA decides to use some type of
mean concentration to estimate human exposure, the Committee agrees with
OSWER that an arithmetic mean is more appropriate than other types of
means, such as a geometric mean. The reasons for this are two-fold. First,
an arithmetic mean may be more likely to correlate with health effects than a
geometric mean. For example, if a person is exposed to 100 ppm of a chemi-
cal for 12 hours of the day and to 0.001 ppm for the other 12 hours per day,
the arithmetic mean exposure is 50 ppm and the geometric mean exposure is
0.316 ppm. The arithmetic mean clearly reflects higher exposure levels that
most likely have a greater influence on any health response. The second
reason is that the geometric mean of a group of samples is highly sensitive to
values assigned to analytical results below the detectable level of the assay.
Theoretically, if a single value is truly zero, then the geometric mean is likewise
zero, regardless of the remaining values.
«»
b) Implications of using the mean estimate
Under certain conditions, involving both statistical factors and science-
policy judgments, the mean is the preferred estimator of an uncertain and/or
variable quantity. Within the context of hazardous waste-site risk assessment,
the mean would be appropriate if: (1) the judgment were made that population
risk, rather than individual risk, were the relevant metric with which to measure
hazard; or (2) the assessor had reason to believe that the (spatial and tempo-
ral) variability of the concentration term were sufficiently small so that upper-
bound individual risks would not deviate significantly from the mean. In addi-
tion, the assessor would have to believe (3) that the uncertainty of the concen-
tration term (due to measurement or modeling errors) was sufficiently well
understood so that the mean was neither an overly conservative estimate of the
representative concentration (i.e., not unduly affected by a few outliers) nor
biased low due to small sample sizes, incorrect model form, etc. OSWER has
attempted to address this third condition by using the upper confidence limit
(UCL) of the mean, but the Committee does not regard this as an adequate
solution (see discussion following concerning the use of a designated percentiic
on the distribution in place of a confidence limit of the mean).
13
-------
In addition, OSWER needs to communicate better to senior managers
and the public that adopting any type of mean concentration represents a
fundamental shift in the Superfund program, making it more like the pesticides
program, for example, and less like the air toxics program (which considers the
risk to the "maximally exposed individual" to be of equal or greater import than
the population average risk or the total number of "lives saved"). If OSWER
believes that by "appropriate" definition of the "averaging zone" (see below), it
is circumscribing the risk analysis to address a relatively "tight" concentration
distribution that accounts for "reasonable high-end exposures," it needs to make
that assumption explicit and provide some empirical support for it. Otherwise, it
should acknowledge that the use of the mean exposure implicitly embodies the
controversial, value judgment that the distribution of individual risks is irrelevant
for cleanup decisions or comparative risk analysis.
3.1.2 Use of the 95% Confidence Limit/Estimating Exposure-Issues Two and
Five
The HHEM addresses the estimation of the Reasonable Maximum Exposure,^.
(RME) as an alternative to use of the Maximally Exposed Individual (MEI). We
applaud the consideration of this concept and believe it could be of considerable value
in evaluating health risks at Superfund sites.
The proposed estimate of the RME in the HHEM is a product of several terms,
including the upper 95% confidence interval on the arithmetic mean of the contaminant
concentration at the site of concern. The mean is averaged over temporal and spatial
scales and can be applied to concentrations in all media. Question two of the Charge
asks if this 95% upper confidence limit is appropriate; question five, which is closely
related, asks about the consistency of this estimate, and the approaches to combining
high- and mid-range exposure values. Both issues are discussed below.
a) Use of the UCL on the mean concentration
If the sampling design allows an estimate of the average concentration,
we do not believe that the upper 95% confidence interval of this average is
appropriate, unless there is evidence that individual exposures are equally likely
across all parts of the site over time periods similar to that over which the
samples were obtained. Another problem with the current approach is that the
14
-------
number of samples taken will have undue influence on the upper confidence
limit even if a large number of samples will not lead to good estimates of actual
exposure. For example, consider a site where contaminant concentrations are
highly non-uniform: there may be a hotspot where concentrations are high, and
the remainder of the site is not highly contaminated. Assume that the hotspot
is highly attractive to visitors (e.g., a waterhole or pond). If there are many
samples taken across other areas of the site, as well as the hotspot (n is large)
then the mean is not a good estimate of exposure, and the upper confidence
limit on the mean is a poor estimate of the RME despite the fact that there are
many sample observations. Estimation of the RME cannot ignore the distribu-
tion of contaminants at the site and the distribution of individual behaviors which
lead to exposure.
The preceding discussion implies that the estimate of an overall mean
concentration at the site, or an upper confidence limit thereon as proposed in
the RAGs draft document, can be inadequate for calculation of an RME.
Rather, the spatial distribution of the concentration over the site must be
considered along with a distribution reflecting the relative frequency with whiciw-
people are likely to visit different parts of the site. For this reason, any summa-
ry measure of concentration (such as the average proposed in the RAGs
document) that does not take into account the spatial distribution of the underly-
ing samples is likely to be inadequate. Therefore, the Committee believes that
the Agency should give strong consideration to incorporating methods, such as
kriging or triangulation (procedures that are discussed in more detail in Section
3.1.3) that take into account the spatial distribution of contamination to charac-
terize exposure.
A related problem with calculating an upper confidence limit for a mean,
and one that can have very severe practical consequences, is the fact that the
statistical assumptions required to calculate an UCL for a mean exposure are
typically not met by sampling plans at a Superfund site. First, the statistical
procedures generally require that sample locations be selected randomly.
However, in typical cases, sampling is done in several stages and is generally
non-random. An initial screening may be performed to identify areas of particu-
lar concern, and follow-up sampling may be performed to characterize more
completely the extent of contamination. Random sampling may not be used (or
even be appropriate) at any stage of sampling.
15 *
-------
Second, in order to calculate a statistical confidence limit for the mean, a
particular distributional form must be assumed for the data (e.g., normal, log-
normal, or Weibull). There is a priori no reason that sampling data should have
a particular distributional form. Different distributional assumptions can some-
times produce quite disparate results.
In some cases, routine application of a standard procedure for calculat-
ing an UCL on the mean concentration can produce an anomalous result in
which the UCL is literally thousands of times larger than the maximum observed
value (MOV). In fact, such extreme behavior can even occur when obtaining a
point estimate of the mean. When this anomaly occurs, it is likely to be due to
the fact than one or more of the underlying statistical assumptions are violated,
in which case the relationship between the. computed UCL and the actual
concentrations at the site are likely to be purely coincidental:
The RAGs manual suggests that the upper limit be replaced by the MOV
in such cases. However, this default could result in a non-conservative esti-
mate of the average concentration. Whenever the estimated mean, or UCL £rT
the mean, exceeds the MOV, EPA should have concern as to whether or not
the concentrations at the site have been adequately characterized.
A related problem that may be critical when calculating the mean
concentration, or a UCL for the mean, is the statistical treatment of "non-
detects" (samples in which no contamination is detected). These cases are
often treated in an ad hoc fashion by assigning either the smallest concentra-
tion that could have been detected (the "detection limit") or some fixed fraction
of the detection limit to these samples. However, the value assumed for
non-detects can sometimes have an enormous effect upon the UCL for the
mean concentration. However, likelihood methods are available for such data
that do not require assigning a particular value to non-detects (see, for exam-
ple, Crump, 1978). The Committee recommends that such methods be consid-
ered by EPA when calculating UCLs for mean concentrations from data con-
taining non-detects.
Considering the estimation issue de novo by media of potential expo-
sure, the proposed methodology is probably most relevant for soil contamina-
tion and dermal exposure. The calculation of the RME must consider the likely
16
-------
exposure patterns of individuals visiting a site, and all parts of the site may not
be equally likely of receiving a visit. An average concentration weighted by the
likelihood of that sampling site being visited is a much better estimate of the
concentration to be used for calculating the RME. This can be improved even
further by looking at a distribution of average concentrations to which visitors at
a site are exposed and taking a given percentile, such as the 90th percentile, of
that distribution. We would advocate such an approach in place of the ap-
proach proposed in the HHEM. In addition, we believe that the alternative
approach proposed above is more consistent with the exposure assessment
guidelines, and is in the spirit of the Exposure Factors Guidelines.
b) Approaches to combining various exposure inputs
EPA is treating different factors in the exposure calculation in different
manners (upper 95th confidence limit for one factor, best estimates for others,
and some not clearly defined type of conservative limit for others), so that the
result is very difficult to interpret. The Committee recommends that EPA move
towards a full distributional approach in which distributions are developed for*—"
each of the terms in the exposure equation and a Monte Carlo analysis be
applied to obtain the resulting distribution for exposure (and thus any desired
percentile of this distribution, such as the 90th percentile). In this manner, EPA
can consciously choose the desired degree of "conservatism." An EPA-spon-
sored effort in which such distributions are developed and applied to d few sites
would illustrate the methods, expose the strengths and weaknesses b'f the
methods in more detail, and provide guidance on the appropriate data needs to
facilitate the calculation of the RME. To facilitate this, EPA should develop
default distributions for terms that are not likely to vary greatly from site to site
(amount of water consumed, body weight, etc.). It has been argued that the
data are not always available to derive the estimates of the type we suggest.
We believe that the costs of generating these data are not very great, and the
burden of providing these data could be placed upon the Potentially Respon-
sible Parties (PRPs). It would generally be in their interest to facilitate the
estimation of the RME in place of the more conservative MEI.
For other media of exposure (air and water), the proposed methods need
to be adapted because the RMEs may be associated less with exposures on
17
-------
site than with exposures in areas adjacent to the site. The approach proposed
in the HHEM ignores this possibility.
Given the difficulties in interpreting any single number as a measure of
exposure, and in particular the RME as presently calculated, the Committee
believes that some type of 'best' or 'most reasonable' estimate of exposure
would provide useful additional information to a risk manager, and should be
presented along with the "conservative" estimate we advocate above, if a
Monte Carlo analysis is available, such an estimate could be obtained from the
median of the uncertainty distribution. If a Monte Carlo approach has not been
implemented, a 'best' or 'most reasonable' estimate could be developed by
assigning 'best' or 'reasonable' values to each component of the estimating
expression.
3.1.3 Characterizing Contaminant Concentrations-Issue Three
It is worth investigating alternate approaches for characterizing underground
concentrations for possible use in future guidance. Of the two mentioned in the "***
HHEM, kriging and triangulation, the former has been in use for many years, first in
the extractive industries to map ore or coal deposits, for example, and more recently
in characterizing underground contamination. The calculational method is well
developed and the interpretation of the results is well understood. Typically, kriging
yields not only concentration contours but also standard error contours, which facilitate
developing sampling designs from a few initial samples, if these are available, as well
as the determination of where additional samples may be needed to increase the
confidence of definition of "hot spots" or other features of underground concentrations
of contaminants. The Committee recommends that this technique be included m
future guidance as a useful, already developed tool. Trianqulation is a newer tech-
nique: different in its approach to data analysis from kriginq. it nonetheless produces
similar results and is worth further examination as a possible tool. Case studies.
which illustrate the differences in results and their implications, should be developed
before final judgements are made
3.1.4 Determining Exposure Factors-Issue Four
Although a well done survey is the best way to characterize behavior patterns
of individuals working, living or otherwise present in the vicinity of Superfund sites >t
18
-------
may not be necessary, in many cases, to carry one out. National survey information,
suitably altered for known gross differences between the specific site and the nation
as a whole -- such as behavioral differences related to climate -- will often give useful
guidance. Other local data may be readily obtained on population distribution,
economic level, types of industrial or other work activities in the area, geographic
details, possible future developments, and so forth from such sources as local
chambers of commerce, planning commissions, departments of motor vehicles, police
departments, sheriffs offices or other local bodies, without the need for a physical visit
to the site area; such information would provide further basis for the modification of
nationally observed trends. Such information would also be helpful in deciding if a
local survey was needed or not. Some regional and national survey data are under
development. These data could be used to derive default exposure values in the
absence of site-specific data, although the collection of site-specific data is encour-
aged in instances where site-specific conditions are unique and there is reasonable
possibility that use of regional or national data could lead to large errors.
3.2 Chemical Mixtures
3.2.1 AdditMty of Risk-Issue Six
An initial issue to address in considering additivity of risk is the accuracy and
precision of the data on which the risk estimates are based. If the risk estimates are
based on outdated, imprecise, or inaccurate studies, then their reliability will be
constrained accordingly. Rigorous analysis of the existing scientific database requires
not only diligence and time, but also adequate resources. The ultimate commitment
for a Superfund site warrants a system that reflects state-of-the-art science and risk
estimation.
The EPA Hazard Index (HI) depends upon Reference Doses (RfD) which, in
turn, depend upon effect levels, such as the No Observed Adverse Effects Level
(NOAEL), divided by an uncertainty factor. The ambiguities inherent in such a proc-
ess had earlier led the Environmental Health Committee to urge greater reliance on
the total dose-response function (when available) to calculate values such as Bench-
mark Doses (EPA, 1990; Crump, 1984) or effect level specifications that incorporate
dose-response information (Barnes and Dourson, 1988).
19
-------
A Hazard Index is often presumed to be a measure of the toxic potency of a
mixture. But the RfD is only an indirect index of potency. It is only indirect because
potency should be defined by the form of the dose-response (effect) relationship, not
simply by a single, often equivocaj, point on the dose-response function.
1 -,
Given the interpretive flaws of the RfD, it is more appropriate to apply the
strategy of the Benchmark Dose to mixtures: If a data set can support the derivation
of a NOAEL, it should also provide enough information to calculate an ED10 or ED01.
(Comments by Dr. Gaylor discussed at previous Committee meetings indicate that, in
fact, NOAELs approximate the lower confidence limits of the EDM)
Of course, exposure to any of the components of most environmental mixtures
rarely approach even an ED01, and the shape of the low-dose portion of the dose-res-
ponse function is typically unknown. Moreover, the Hazard Quotient is not, as posed
in Question 6, a risk estimate, and it is not risks that are added.
By utilizing dose-response information, more quantitative approaches would be^
possible. Such approaches might also help address the question of how to apply
effect-specific RfDs. If separate dose-response functions can be fitted to individual
toxic criteria, it should be possible to combine the available data in some form of
meta-analysis. Another potential virtue of using dose-response information for
individual effects is the possibility of examining potency ratios between effects at
different positions of the dose-response function. Such an examination might offer
clues about where to search for interactions. The Committee encourages EPA to
utilize alternatives to the default approaches that involve additivity of risks or doses in
specific cases whenever there is a reasonable scientific basis for so doing..
3.2.2 Hazard Indexes and Thresholds of Concern-Issue Seven
The Hazard Index approach to evaluating the non-cancer hazard of chemical
mixtures is crude and affords far from universal protection for all mixtures. The
Committee recommends, therefore, that if toxicological data on a particular mixture are
adequate to derive an RfD for that mixture, the RfD should be used in place of the Hi
approach. Similarly, if a situation were to arise in which the scientific understanding of
the interactions of chemicals comprising a mixture were sufficient to predict, theoreti-
cally, an RfD for a chemical based on its potential for interaction, the application of
that RfD would be encouraged by the Committee.
20
-------
However, the Committee also recognizes that neither of these cases is apt to
occur frequently in practice. Therefore, some less precise approach, such as the use
of the HI, seems unavoidable. The HI approach is not necessarily arbitrary, and in
fact provides a very rational answer for an important class of interactions. Some of
the useful features of the HI approach may be summed up in the three following
characteristics. The validity of these characteristics is demonstrated in Appendix 1.
a) Characteristic 1: If all the chemicals in a mixture act lexicologically as if they
are dilutions of a single chemical (This type of interaction will be referred to as
"dilution-type" interaction), then the criterion HI = 1 should afford a level of
protection that is intermediate between the levels of protection that are experi-
enced as a result of exposure (limited to the levels associated with their
respective RfDs) to the individual chemicals in the mixture. Consequently, if
each RfD affords adequate protection for exposure to that individual chemical
— as is intended in setting the RfD — then the condition HI = 1 should likewise
afford adequate protection for exposure to the mixture.
b) Characteristic 2: If 1) each chemical in a mixture has an effective threshold
when administered in isolation and 2) the RfD for each chemical in the mixture
is below the threshold for that chemical and 3) the interactions among chemi-
cals in the mixture involve a combination of independent mechanisms of action
and dilution-type interactions (e.g., no synergistic interactions), then the thresh-
old for the mixture should not be exceeded as long as HI < 1. [See the Appen-
dix for a rigorous definition of this class of chemicals.]
c) Characteristic 3: Even if all of the other conditions on the mixture in Charac-
teristic 2 hold, if the hazard index of the mixture is greater than one (HI > 1),
the threshold of the mixture may be exceeded.
Stated more broadly, Characteristic 2 implies that whenever the interactions in
a mixture involve some combination of independent action and dilution-type interac-
tions, then the HI approach will always afford at least as much protection as the least
protective individual RfD. Characteristic 3 implies that under these types of interac-
tions, the condition HI = 1 defines the least conservative approach that is still guaran-
teed to be protective, and that can be obtained without more detailed mechanistic
information. (A less conservative approach [i.e., one consistent with HI > 1] can in
some cases afford less protection than that afforded by any of the individual RfDs )
21
-------
An example of additional mechanistic information would be knowledge that all
of the chemicals in the mixture operate through independent mechanisms. If that
were known to be the case, then simply requiring the doses of the component
chemicals in the mixture to be below their respective RfDs would be less conservative
than the HI approach and still provide the same protection as that afforded by the
individual RfDs. However, if we cannot rule out the possibility that some of the
chemicals may operate by other mechanisms of joint action, then we cannot be sure
that this approach affords adequate protection.
These considerations demonstrate that the value "1" (as in HI = 1) has a
rational and meaningful basis for defining a "threshold of concern" for the HI that is not
shared by any other number. Under the stated conditions, the uncertainty in the HI
approach is, in a sense, no greater than that of the component RfDs. Given this, the
Committee does not see any value, generally speaking, to use numbers other than "1"
in defining a "threshold of concern." However, if there are interactions of the chemi-
cals in the mixture which cannot be fully characterized by a combination of dilution-
type interaction and independent mechanisms of action, then the entire HI approach _.
may be inappropriate. Moreover, the use of the number "1" in defining a threshold 6T
concern for a mixture, as derived in the Appendix, does not take into account the
possibility that the joint severity to a particular individual subjected to a number of
unrelated adverse effects, whether they exhibit dilution type interactions or not, may
be very great, and the safety of the mixture may therefore be questionable.
It should also be noted that none of these properties assume a linear dose
response. The hazard index approach is not predicated upon a linear dose response
3.2.3 Interpretation of the Hazard Index-Issue Eight
When the hazard index (HI) is greater than unity, the toxicologic connotations
and interpretations may be different depending on several factors:
a) If two or more agents involved share the same mechanism of toxioty
their doses could well be additive, possibly resulting in more than the
sum of the additive risks (see discussion in the Appendix).
22
-------
b) If they instead act upon two or more sites along a mechanism sequence
they could well be supra-additive, as are Malathion/Parathion (Murphy,
1969).
c) If they each act by different toxicologic mechanisms, additivity of risks for
a common endpoint is not necessarily to be expected; for different
endpoints, the potential hazard could be overestimated.
It should be noted that the occurrence of agents acting in the same target
organ does not imply that they share the equivalent mechanistic properties. For in-
stance, there are many different types and mechanisms of hepatotoxicity. One should
never consider that because two agents cause reproductive (e.g., testicular) or
developmental toxicity (e.g., cleft palate), they operate by the same or even similar
mechanisms. On the other hand, diverse types of developmental (for instance) toxicity
can result from the same apparent mechanism, e.g., adenosine tri-phosphate deficien-
cy and mitotic arrest, or interference with directional cell migration.
In the absence of experimental data sets, it is not yet possible to contemplate"""
whether the slope of increasing risk would be steep or shallow. With adequate
experimental data sets, one could still have different levels of concern, depending on
both the slopes of the dose-response curves for mixture components as well as the
types of resulting effects. The most prudent mode would be to consider that as the HI
exceeds unity, the potential for risk increases; without a more complete understanding
of interaction mechanisms, however, we can not state how rapidly this increase
occurs. Also, we cannot rely on comparisons of risk using His for His greater than
unity.
3.2.4 The RfD As a Criterion of Toxicrty-lssue Nine
The OSWER proposal suggests that EPA modify its current practice of calculat-
ing the Hazard Index for mixtures. At present, chemicals in a mixture are assessed
for joint action by computing the ratio of exposure to RfD, then adding these quotients
to obtain the Hazard Index. Because the RfDs are based on doses such as NOAEls
that are derived from the critical effect in an assay (that is, the effect showing the
greatest sensitivity to exposure), the resulting Hazard Index may encompass a
spectrum of toxic endpoints and risk levels. Such a melding of disparate endpomts
may overestimate the magnitude of risk. The proposed modification would caicuia'e
23
-------
effect-specific RfDs instead, and is stated to be more consistent with the assumption
of dose-additivity that guides the Hazard Index calculation.
The argument presented in tfie issue paper has merit, but offers practical
difficulties. One reason is that some RfDs are based on non-specific endpoints such
as weight loss, which can arise from many causes. Another is the problem of effect
interactions; for example, kidney damage may lead to peripheral neuropathy, and liver
dysfunction may prolong the CMS action of a solvent.
More descriptive, flexible indices of joint action might prove useful. Assume, for
example, at least 100 identifiable contaminants at a site. If each is present in
groundwater, say at an average of 1% of its RfD, does it make sense to add all the
Hazard Quotients to derive a single sum expressed as a Hazard Index of 1.0? Would
it be more reasonable to attempt to derive some estimate of the degree to which these
substances act jointly, or embrace common modes of toxicity?
As an example of such an approach, assume a mixture of five chemicals, each
present at 0.2 of the individual RfDs. If the biological effects of all five were totally **~~
independent, that is, affecting different organs, systems, or receptors, then the total
effect of the mixture would still be 0.2 times the RfD. If the biological effects of all five
overlapped completely, and acted in the same way on the same system, then the total
effect of the mixture would be the sum of the individual RfDs, or 1.0.
Another approach would be to conceive of the overlap estimates as correla-
tions. Under this assumption, the matrix might then be subjected to Principal Compo-
nents analysis. That is, the cell entries would represent shared variance.
Assume, however, that the mutual overlap = 0.01. That is, any one of the
chemicals enhances the effect of one of the others by 1-.0%. Then, the combined
effects of chemicals 1 and 2 = 0.2 + 2 (.01) = 0.22, because chemical 1 enhances
chemical 2 by 1.0% and the reverse.
Filling in each of the interaction cells in the 5 x 5 matrix (10 cells) yields 02 x
10 = 0.2. Therefore, the total proportion of biologically effective RfD is 0.4 rather than
1.0, which would be the current default Hazard Index.
24
-------
Perhaps such a value could be expressed by a matrix (100 x 100 in this exam-
ple) in which each cell contains an estimate of overlap or joint action, defined else-
where as "commonality" (Weiss, 1986). We would expect commonality coefficients
within chemical classes, for example, to exceed commonality estimates between
classes. Although these estimates would be expert judgments, they need not be any
more vague than the EPA approaches to probabalistic risk assessment or any less
cogent than Office of Toxics Substances' reviews of chemical structures. Such a
matrix might substitute for, or be an accessory to, effect-specific RfDs.
A first step in this ambitious project would be to group chemicals for specific
effects into groups that are likely to operate through some combined (i.e., non-
independent) action. The next step would be to assign some type of interaction term
to all pairs of chemicals in the same group as suggested above. This step is likely to
be more difficult as well as controversial. Until these interaction terms are developed
and adopted, EPA could make an incremental improvement just by taking advantage
of the presumption that chemicals ir> different groups operate by independent mecha-
nisms of action, but otherwise retaining other facets of the current approach. The
result would be an approach that is similar to that proposed by EPA, except that •**"
chemicals which cause the same endpoint would not necessarily be assumed to
operate by a common mechanism of action.
Finally, the proposed modifications remain captives of NOAELs and their
relatives. Contrary to what is stated in the issue paper, effect levels are not'thresh-
olds as is noted repeatedly in EPA documents. Examine the case study of chloroform
-- the LOAEL for liver damage is 12.9 mg/kg daily, and the RfD is calculated on (he
basis of an uncertainty factor of 1000. The NOAEL for kidney effects (based on the
same dog study) is also 12.9 mg/kg, but in accordance with EPA practice, it is divided
by an uncertainty factor of 100. The comparisons imply the assignment of a seventy
index that probably is unjustified on the basis of these kinds of data; moreover, the
basis for such an assignment is not adequately explained. The absence of dose-
response factors is a further defect.
Despite these problems with the proposed approach, it is not nearly as serious-
ly flawed as the two alternatives presently in use. The "critical" effect approach
ignores the fact that a chemical can cause an effect even when that effect is not the
critical effect and therefore is not health protective. On the other hand, the "Super-
fund" approach assigns the RfD of the critical effect to all effects of a chemical, wh en
25
-------
is clearly overly conservative. The proposed approach, which involves calculating an
RfD for each organ site for each chemical and then combining over the chemicals for
each organ site, and then combining over organ sites is, despite its flaws, an im-
provement over either of these two alternatives.
RfDs for individual agents have clear value in risk management in setting "safe"
limits of exposure for such agents (that is: very low, essentially zero -- or perhaps
even zero -- risk limits) and HQs for individual agents clearly measure how far from or
close to the RfDs the actual exposures are for those agents. The HQs give a
measure of risks at the actual exposures relative to those if the exposures were equal
to the RfDs -- at least for HQs above 1.0 and possibly below, depending on the
existence or nonexistence of thresholds and where the RfDs are located with respect
to the thresholds. Ideally, in assessing a mixture of agents using the RfD concept, the
RfD of the mixture, as determined from exposures using the mixture itself, is what is
needed to assess the safety or lack of safety (the lack or presence of risk) of actual
exposures to the mixture. If such a "mixture RfD" were available, then the HQ for the
mixture would indicate risk relative to the RfD, an HQ of 1.0 being, again, the dividing
line between "safe" and "unsafe". Experimentally determined RfDs for specific **-~
mixtures are generally not available, and it is not generally practical during a particular
assessment to obtain them. It has been suggested that His greater than 1.0 should
suggest particular concern, an HI for a mixture, In this sense, being taken as a kind of
surrogate for the actual HQ of the mixture.
Considering two cases helps answer the questions raised here: (a) In deriving
RfDs from experimental data the original idea was that, for non-carcinogenic adverse
effects, thresholds exist and that the RfDs represent safe doses located somewhere
below the threshold doses for each agent -- how far below not being known; and (b) it
may be, instead,that there is residual risk below the RfD because there is no threshold
or, alternatively, that there is a threshold but the RfD is set, unintentionally, some-
where above it.
In case (a), it is easy to show, with a few numerical examples, that a higher Hi
does not necessarily imply a higher risk. Considering (b), the actual risk -- the proba-
bility of a particular adverse effect -- corresponding to the RfD is not known and can '
differ significantly from one RfD to another.
26
-------
Nothing in the various procedures for determining RfDs is aimed at ensuring
that RfDs are determined at either known levels, or a fixed level, of risk. Comparisons
of HQs for different agents is therefore not meaningful from a risk standpoint since the
ratios depend on the RfDs determined at varying but unknown levels of risk and reflect
this variation. An HQ of 3.0 for one agent, for example, might not indicate a higher
risk than for another agent having an HQ of 2.0 if the RfD for the first agent was, by
chance, determined at a sufficiently lower value of risk than was the second (depend-
ing, also, on the shapes of the dose response curves at low doses). Summing the
HQs to produce an HI inherently assumes that the HQs are comparable. Since it
cannot be generally known that there is comparability of HQs and since comparability
does not therefore exist in general, the significance of the sums is, in turn, in doubt. It
cannot be known whether a given mixture having a larger HI than another mixture
represents a greater or lesser risk compared to that mixture. If it were possible to
derive RfDs at some standardized risk level, it might be possible to develop a basis for
comparing HQs and for forming meaningful His. In the case that the dose responses
of individual mixture components at low doses are linear and extend to zero, for
example, HQs are just the ratios of the probabilities of effect at the actual exposures
to those at the RfDs; if the RfDs were all determined at a standardized probability
level then the His would be proportional to the sum of the probabilities of effect at the
actual exposures and under these circumstances the His would offer measures of
relative risk. No accounting for synergism, antagonism or of joint severity is included
in this approach (nor are they in the HI as now defined), and the implicit assumption of
dilution-type interaction may well be invalid for many interactions.
From the foregoing discussion, we can state that His do not have general
meaning with regard to relative risk, and their undiscriminating use can lead to giving
a degree of unwarranted comfort to the unwary. A high HI may or may not mean a
high relative risk and a low one must not be taken to mean that the relative risk is
necessarily low. The contributions of individual agents need to be considered
individually; in addition, the combined effects of agents present need also to be
considered, taking into account at least the modes of action, when known, of the
organs or systems affected, and the possible joint severity of the effects -- all this
aside from any synergism or antagonisms that may be present. Even if synergism
and antagonisms are not present, the total severity of being affected independently by
more than one adverse effect (or by effects caused by more than one agent) can be
much greater than the severities'of the individual effects might lead one to believe;
and the risk (defined as including both severity and probability) will therefore be mucn
27
-------
greater, too, than might be anticipated by considering the array of individual HQs,
alone.
Mixtures containing one or more agents, the HQs of which are greater than 1 0
may pose a risk on the basis of the RfD concept; those not containing such agents
cannot be considered risk-free, however, and must be subjected to further examination
-- as must the former, since reducing exposures to reduce the HQs greater than 1.0 to
values less than 1.0 will not necessarily bring the total risk to a low enough level.
Under conditions described in full in the Appendix and elsewhere in the body of
this report, a Hazard Index of 1 for a mixture can have special meanings in relation to
the RfDs of the individual chemicals, the criterion HI = 1 affording a level of protection
that is at least as protective as the jst protective RfD for single chemical in the
mixture. However, this may not be tne case if the HI exce?. .s 1.0 for a mixture.
Moreover, if some of the chemicals interact synergistically, then the condition HI = 1.0
may not afford adequate protection. On the other hand, if the individual chemicals
have independent mechanisms of action, then the criterion HI = 1.0 may be overly
protective. Whenever there is no, or very limited, information on the types of inter- ^_
actions that exist among the chemicals within a specific mixture, the Committee
suggests that the HI, including the criterion of HI = 1, be considered (along with
whatever other information may be available) for possible guidance unless information
becomes available which confirms the validity of HI = 1, or unless information on
interactions becomes available which permits the development of an appropriate value
of HI or another criterion.
The Committee realizes that most Superfund decisions must be made on the
basis of incomplete (and usually highly imperfect) information. This situation is, of
course, the root cause of the estimation problem with which we must cope.
3.3 RfDs in Goal Setting
3.3.1 Exposure Scenarios-Issue Ten
OSWER initially considered two approaches for using RfOs in setting risk-based
remediation goals in soil: 1) comparison of a 6-year, childhood exposure to contami-
nants in soil with a sub-chronic RfD; and, 2) comparison of a 30-year, time-weighted
average exposure to contaminants in soil (including exposures to both children and
28
-------
adults) with a chronic RfD. Now, a third approach has been proposed: comparison of
a 6-year, childhood exposure with a chronic RfD.
The second approach proposed by the OSWER probably is the more reason-
able. That is, to compare a 30-year time-weighted average (TWA) exposure with a
chronic RfD. It is likely to be adequately conservative. Comparison of a six-year old's
exposure with a chronic RfD (a third approach) may be overly conservative. It also
assumes the six-year old is the more vulnerable. The second approach accounts for
variable susceptibility with age in a more conservative manner than does method
three. Actually, all three methods could be considered, and the one giving the most
conservative (in the absence of specific information) or the most reasonable estimate
(in the presence of such information) used. It would be helpful to see a group of
diverse examples for all three approaches. Perhaps the most relevant exposure -
scenario could then be selected (i.e., childhood vs. lifetime). Clearly, the model
selected must be one that accommodates the most intense future land utilization, e g.,
housing, lest repeat remediation become necessary.
3.4 Short-Term Toxicity Values
3.4.1 Interim Estimates of Toxicity-lssue Eleven
The methods outlined in the Interim Methods for Development of Inhalation
Reference Concentrations are a reasonable approach to determining short-term
toxicity values. Basically, the approach is to find human toxicity data (if possible) or
animal data of the appropriate duration (or as close as possible to the appropriate
duration) that indicate a NOAEL The NOAEL is then used to set the RfC, based on
the NOAEL (adjusted for duration) divided by uncertainty or modifying factors. Care
must be taken in adjusting the NOAEL for the duration of exposure for compounds
that cause acute effects based mainly on concentration and not duration of exposure
As noted in section 3.3.1, concerning time-weighted averaging, caution must be used
when the relevant extrapolation ranges are fairly broad.
A major concern in the calculation'of short-term toxicity values is to choose
appropriate uncertainty and modifying factors so as not to exaggerate the potential
toxicity associated with a site or a specific chemical and yet protect the public health
It will also be important in assessing risk from short-term exposures that the exposure
29
-------
time and the averaging time used are consistent with the shortest period of time that
could produce an effect (section 6.4.2 of RAGS-A).
There are two additional squrces of information on short-term toxicity values.
One is the Emergency Response Planning Guidelines (ERPG) of the American
Conference of Governmental Industrial Hygienists. The second is the Community
Emergency Exposure Levels (CEELs) being developed by a National Academy of
Sciences Committee sponsored by the EPA. The approaches of both of these
committees should be considered in developing short-term toxicity values. Although
some of the short-term exposures of interest to the EPA at Superfund sites are of a
longer duration than one to eight hours, the concept of providing some measure of the
dose-response characteristics of a chemical is an excellent one. For persons in risk
management, it is important to know if the level of a chemical that causes mild
irritation is two times or one thousand times lower than the level that is life threaten-
ing. The ERPG and CEEL methods use multiple guidance levels (ERPG-1, ERPG-2,
ERPG-3 or CEEL-1, CEEL-2, CEEL-3) to provide some dose-response information for
the chemical of concern. The lower guidance level, such as CEEL-1 or ERPG-1 is the
level below which the chemical is unlikely to cause mild effects such as discomfort o7*~
irritation. The second level is the level below which the chemical is unlikely to cause
toxic effects leading to disability that could interfere with taking protective actions The
third level is the level that is life threatening. These short-term toxicity guidance levels
are set for specific times of exposure (one or eight hours).
3.4.2 Short-Term Air Action Levels-Issue Twelve
The method proposed by Region 6 suggests the possible use of OSHA
standards, such as Permissible Exposure Limits (PELs) and Short-Term Exposure
Levels (STELs), to derive short-term action levels when alternative EPA RfD data are
not available. Clearly the absence of an RfD does not mean that there should be no
short-term action level. Ideally the data used to derive the OSHA standard should be
examined to see if the same data can be used to derive an appropriate RfD. The use
of EPA derivation methods would help promote consistency across various hazardous
substances; hence the use of EPA methods should be encouraged. Where this is not
possible or practical, the use of health data on which OSHA standards have been
based could be considered, taking into account differences between the worker
population and the general population. In its consideration of adapting EEGLs to the
general population, the National Research Council (Criteria and Methods for Precanng
30
-------
Emergency Exposure Guidance Level (EEGL), etc., 1986) applied a safety factor of
two for sensitive subgroups of the general public and a safety factor of ten for
newborn infants. The rationale used to derive these safety factors should be reviewed
to see if it is appropriate for PELs and STELs as well.
31 '
-------
4. CONCLUSIONS
The OSWER's attempts to codify, and so increase the consistency, of its site-
specific risk assessment methodology are praiseworthy. The need to deal with wide
ranges of contaminants, often in complex mixtures with exposure through multiple
media, poses daunting problems for risk assessment. Although the OSWER has
(understandably) not succeeded fully in meeting the goals they set for themselves in
producing the HHEM document, they have made a good start. The following com-
ments summarize the Committee's major findings, specifically noting where we believe
a redirection in approach is called for, and providing advice where possible.
The Committee is of the opinion that there are some serious difficulties, both
conceptual and practical, with the approach recommended in the RAGs document for
calculating an RME based on a UCL on the average concentration at a site.
First, a UCL for the mean concentration does not lead logically to a "reasonable
maximal exposure." The current approach assumes that the samples taken are
representative of those areas where exposures are most likely to occur. The RME is
also a function of the number of samples available; a larger number of samples will
result in a smaller RME even if the samples are not representative of exposure
opportunities. For example, if a site is well-characterized, so that the UCL is very
close to the true mean, but a "hot spot" is very attractive to visitors, then a significant
fraction of visitors to the site (perhaps the majority) could be exposed to levels
significantly higher than the UCL.
Second, to calculate the RME, the UCL on concentration is combired with 90th
percentile values for some variables and 50th percentile values for other -ariables -
an ad hoc fashion, making the resulting RME very difficult to interpret.
Third, the UCL on the mean concentration does not take into account the
spatial distribution of contamination via-a-vis the relative frequency with which people
are likely to visit various parts of the site.
Fourth, the calculation of a UCL requires statistical assumptions that are
generally not met by sampling plans at a Superfund site. As a consequence, statisti-
cal procedures can sometimes produce a UCL that has little, if any, relation to actual
concentrations at a site.
32
-------
Because of these difficulties, the Committee recommends that the EPA move
towards a 'full distributional approach" to calculating the RME. In such an approach, a
distribution is developed for each of the input terms in the exposure equation. These
distributions determine a distribution for exposure, which can be calculated using
Monte Carlo methods. A particular percentile of this exposure distribution, such as the
90th percentile, could be used as the definition of the RME.
If people are more likely to visit certain areas of the site than others, then the
spatial distribution of contaminant concentrations about the site needs to be consid-
ered when quantifying human exposure. Kriging and triangulation are two statistical
methods for quantifying concentration that take into account the spatial distribution of
samples. These approaches should be considered for adoption by EPA.
As part of the effort to implement a distributional approach for quantifying
exposure, EPA should develop default distributions for exposure parameters that are
unlikely to vary significantly from site to site. However, the collection of site-specific
data is encouraged in instances where site-specific conditions are unique and there is
a reasonable possibility that use of default distributions developed from regional or
national data could lead to gross errors.
The Committee recognizes that some time will be required to implement a
distributional approach to quantifying exposure. In the meantime., some version of the
current approach may have to be used. The Committee has two recommendations
regarding the application of the current approach during this interim period. First, as
long as some type of mean concentration is to be employed to estimate human
exposure, the Committee agrees with OSWER that an arithmetic mean is more
appropriate than a geometric mean. Second, given the difficulty in interpreting (he
RME as presently calculated, the Committee recommends that some type of 'most
reasonable' estimate of exposure also be calculated and made available to risk
managers along with the RME.
The issue of risk additivity from exposure to complex mixtures remains a
difficult question. The Committee is concerned about the approach of using RfD-
derived Hazard Quotients/Hazard Indices as a basis for adding "risks." In our opinion
the Hazard Quotient is not a risk estimate, and it is not truly risks which are being
added when the proposed approach is used. We would much prefer to see quantita-
tive applications using dose-response data (not the "point" data represented by
33
-------
LOAEL/NOAEL-derived RfDs) to drive risk estimates. Further, we suggest that the
Agency develop and use alternatives to the current default approaches that assume
risk additivity whenever there is a reasonable scientific basis for so doing.
The use of the HI itself can be misleading, and it should used as a "fallback,"
with full recognition of its possible inapplicability, only when more refined lexicological
data are not available. As noted in section 3,.2.2 of the report (and demonstrated in
the Appendix), the condition "HI = 1" defines a "threshold of concern" that is not
shared by any other value for HI, and for which, under specified conditions, the uncer-
tainty in the HI approach is no greater than that of the component RfDs. The HI
approach is invalid, however, if the chemicals in the mixture cannot be fully character-
ized by a combination of dilution-type interactions and independent mechanisms of
action. Another instance in which the HI = 1 approach may not adequately address
the safety of a mixture occurs when there is a high joint severity of total effect in
individuals subjected to a number of unrelated adverse effects (regardless of the
mechanisms of action).
Interpretation of an HI greater than 1 can vary depending on several toxicologT""
cal factors: the existence of common mechanisms of toxicity (in which case doses
could be additive; cases in which the agents act upon two or more sites along a
mechanism sequence (resulting in supra-additivity); and instances in which the
substances act by different toxicologic mechanisms (in which case, potential hazard
could be overestimated. Although it is likely that risk increases as the HI exceeds 1,
we can not state without a more complete understanding of interaction mechanisms)
how rapidly this increase occurs. For this same reason we cannot rely on Hl-based
comparisons of risks for His greater than 1.
The proposed guidance recommends modifying current policy and using the
RfD as the toxicity criterion for each of the other effects believed to be caused by a
given agent in a chemical mixture. This proposal is not without merit, since the
current approach melds many disparate endpoints and may overestimate risk. The
proposal suffers however, from problems of its own-it does not deal with effect
interactions, nor with the fact that many RfDs are based on non-specific endpoints
which can stem from many different causes. The proposed approach, despite its
limitations, is an improvement over the alternatives presently in use.
34
-------
OSWER discusses three approaches for using RfDs to develop risk-based
remediation goals for contaminated soil, involving differing exposure scenarios and
target populations: 1) comparison of a 6-year, childhood exposure to contaminants in
soil with a sub-chronic RfD; 2) comparison of a 30-year, time-weighted average
exposure to contaminants in soil (including exposures to both children and adults) with
a chronic RfD; and 3)comparison of a 6-year, childhood exposure with a chronic RfD
The most reasonable and supportable approach appears to using a 30-year
time-weighted average with a chronic RfD, but differences between the three propos-
als are not dramatic. OSWER should study all three approaches, applied to a diverse
set of examples in order to verify its ultimate choice (or range of choices).
The Committee sees no particular problems in the existing approach for dealing
with short-term toxicity estimates. As in all cases when using time-weighted averag-
ing, care must be taken when the extrapolation ranges are broad. The Committee
suggests that OSWER take cognizance of the EPA-sponsored work at the National
Academy of Science on Community Emergency Exposure Levels, and of the work on
Emergency Response Planning Guidelines by the American Conference of Govern-
mental Industrial Hygienists.
The method proposed by Region 6 for setting short-term air action levels calls
for the possible use of OSHA standards, such as Permissible Exposure Limits (PELs)
and Short-Term Exposure Levels (STELs). The data used to derive any OShfA stan-
dard so used should be examined to see if the same data can be used to derive an
appropriate RfD. The use of EPA derivation methods would help promote consistency
across various hazardous substances; hence the use of EPA methods should be
encouraged. Where this is not possible or practical, the use of health data on which
OSHA standards have been based could be considered, taking into account differen-
ces between the worker population and the general population.
35
-------
6. REFERENCES
Barr . D.G., and M. Dourson. 1986. Reference Dose (RfD): Description and use in
wealth risk assessments. Regulatory Toxicol. Pharamacol. (8): 471-486.
Crump, K. S. 1978. Estimation of Mean Pesticide Concentrations When Observations
are Detected Below the Concentration Limit. Prepared for the Food and Drug
Administration.
Crump, K.S. 1984. A new methods for determining allowable daily intakes. Fund. Appl.
Toxicol. (4):854-871.
EPA (Environmental Health Committee, U.S. EPA Science Advisory Board). 1990. Use
of Uncertainty and Modifying Factors in Establishing Reference Dose Levels.
EPA-SAB-EHC-90-005.
Murphy, S.D. 1969. Mechanisms of pesticides interaction in vertebrates. Residue
Reviews. (25):201-221.
National Research Council. 1986. Criteria and Methods for Preparing Emergency •&=-
Exposure Guidance Levels (EEGL).
Weiss, B. 1986. In: Annau, Z. (ed.) Neurobehaviorial Toxicology. Baltimore: Johns
Hopkins University Press, pp. 1-20.
R-1
-------
APPENDIX A - CHARACTERISTICS OF THE HAZARD INDEX
If n chemicals in a mixture operate by "dilution-type interaction" (i.e., have the
same toxic properties as if they were all dilutions of the same chemical), then the
probability of an adverse effect from simultaneous exposure to doses E, En of the
n respective chemicals can be expressed in terms of a single dose response, P(d),
), (1)
where T,, .... TB are toxicity equivalent factors (TEFs) for the respective chemicals.
TEFs are used by EPA to assess risks of mixtures such as those comprised of dioxin,
furans and some PCBs that are thought to have a common mechanism of action.
Note that this formulation makes no assumptions regarding the shape of the dose
response, P(d). E.g., it can be linear, nonlinear or even include a threshold. The
shapes of the dose-response curves of individual components, relative to each other,
must be such as to result in the "dilution type interaction."
It is clear from this formulation that the RfD for chemical i is equally protective
as the RfD for chemical 1 if •*
(2)
where R, is the RfD of chemical i (that is, that each chemical, at its RfD, elicits the
same response as others do at their RfDs).
The validity of Characteristic 1 is now demonstrated. Suppose that a mixture
has HI = 1 and that all RfDs from each of the n chemicals are equally protective. That
means that the component doses E1 satisfy
HI - i + ... * a - i (3)
If we solve for R, in expression (1) for i > 1 and substitute this for R, into expression
(2), we get
HI » —i + -i-1 + ... + -2-2 » i
Rt R1T1 R^
or, equivalently,
A-1 '
-------
(5)
However, this expression says that exposure to the mixture provides exactly the
same risk as exposure to the RfD of chemical 1 alone, which is by assumption the
same as the risk from exposure to the RfD of any other of the n chemicals.
Next consider the case in which all RfDs of the n chemicals are not equally
protective (i.e., the product of the RfD and the corresponding TEF are not the same
for all chemicals). Suppose, without loss of generality, that R,T, < R,T,, or, equiva-
lently, R, > R^/T,, for i=1,...,n. Replacing R,, i>1, in equation (2)"by the right side of
this inequality yields
2. 1
(6)
or, equivalently,
. . . ,RaTa)
(7*—
Similarly,
,RaTa
(8)
These two inequalities indicate that exposure to the mixture with an HI of 1
poses a risk that is intermediate between that posed by the most protective and the
least protective of the RfDs of the individual chemicals. However, if the RfD for each
individual chemical is sufficiently protective, then the protection afforded by the
condition HI = 1 must be sufficiently protective as well (because it is more protective
than the least protective RfD).
Next we demonstrate the validity of Characteristic 2. The assumption here is
that the chemicals in the mixture can be divided into subgroups, such that chemicals
within each subgroup have dilution-type interactions and chemicals in different
subgroups have independent mechanisms of action. Independence of action implies'
that thresholds for a given chemical are unaffected by exposures from other chemicals
having independent mechanisms of action. I.e., if a given dose of a particular
chemical (or combined dose of chemicals having dilution-type interaction) is below the
threshold for an effect (i.e., does not increase the likelihood of an effect) when given m
A-2
-------
isolation, then that same dose will have no effect when given in combination with other
chemicals having independent mechanisms of action.
Now suppose a mixture has HI = 1 and is composed of doses E,, E2 En of
its component chemicals. This means that the component doses satisfy expression
(2). Without loss of generality, suppose that E,, E2,...,Eml where m < n, are the
component doses from one of the subgroups of chemicals that have dilution-type
interactions. Since HI = 1 for the complete mixture, it follows that, for this subgroup,
we have
which, reasoning exactly as before, implies that
E2T2
iMAX(R1T1/ .
This, in turn, implies that exposure to doses E, ..... Em in this subgroup is no
less protective (i.e., is no more likely to cause an adverse effect) than exposure to the
least protective RfD in this subgroup. But if all of the RfDs are below the respective
thresholds, then exposure to this sub-mixture will likewise not increase the likelihood
of an adverse effect, at least when exposure to the sub-mixture is in the absence of
other exposures. However, since we are assuming the other chemicals have mecha-
nism of action that are independent of those in this subgroup, exposure to this sub-
mixture will have no effect even in the presence of the remaining chemicals in the
mixture. This demonstrates the validity of Characteristic 2.
To demonstrate the validity of Characteristic 3, we simply note that if a mixture
is comprised of chemicals with dilution-type interactions and if the RfD for each
chemical happens to coincide with this threshold, then, by the same reasoning that led
to expression (4), we have
+ . . . ET>*T (ID
which- implies that the threshold for the mixture has been exceeded.
A-3
-------
APPENDIX B - GLOSSARY
CEEL Community Emergency Exposure Levels
CERCLA Comprehensive Environmental Response, Compensation, and Liability
Act
CMS Central Nervous System
ECAO Environmental Criteria and Assessment Office
ED01 Effective Dose (level at which 1% of test animal subjects are affected)
ED10 Effective Dose (level at which 10% of test animal subjects are affected)
EEGL Emergency Exposure Guidance Level
EPA Environmental Protection Agency
ERPG Emergency Response Planning Guidelines
HHEM Human Health Evaluation Manual
HI Hazard Index
HQ Hazard Quotient
LOAEL Lowest Observed Adverse Effects Level
MEI Maximally Exposed Individual
MOV Maximum Observed Value
NOAEL No Observed Adverse Effects Level
OSHA Occupational Safety and Health Administration v
OSWER Office of Solid Waste and Emergency Response
PEL Permissable Exposure Limit
PRP Potentially Responsible Parties
RAGS Risk Assessment Guidelines for Superfund
RfC Reference Concentration
RfD Reference Dose
RME Reasonable Maximum Exposure
STEL Short-Term Exposure Levels
TEF Toxicity Equivalent Factor
TWA Time-weighted Average
UCL Upper Confidence Limit
B-1
-------
Distribution List
Administrator
Deputy Administrator
Assistant Administrators
Deputy Assistant Administrator for Research and Development
Deputy Assistant Administrator for Water
EPA Regional Administrators
EPA Laboratory Directors
EPA Headquarters Library
EPA Regional Libraries
EPA Laboratory Libraries LIBRARY
RSKERL
ECW" AGENCY
------- |