! United States
Environmental Protection
I Agency
Office of Research and
Development;
Washington DC 20460
EPA/6307R-00/002
August 2000
Supplementary
Guidance for Conducting
Health Risk Assessment of
Chemical Mixtures
RISK ASSESSMENT FORUM
-------
EPA/630/R-00/002
August 2000
Supplementary Guidance for Conducting
Health Risk Assessment of Chemical
Mixtures
Risk Assessment Forum Technical Panel
Authors:
Office of Research and Development - NCEA
Harlal Choudhury Jim Cogliano
Richard Hertzberg (Chair) Debdas Mukerjee
Glenn Rice Linda Teuschler
Office of Pesticide Programs
Elizabeth Doyle (OPP)
Office of Pollution Prevention and Toxics
Yintak Woo (OPPT)
Office of Water
Rita Schoeny (OST)
Contributors:
Office of Pollution Prevention and Toxics
Elizabeth Margosches (OPPT)
Office of Research and Development
Jane Ellen Simmons (NHEERL)
Office of Water
Charles Abernathy (OST)
Regional Offices
Debra Forman (Region III) Mark Maddaloni (Region II)
Risk Assessment Forum Staff
William P. Wood
Risk Assessment Forum
U.S. Environmental Protection Agency
Washington, DC 20460
-------
DISCLAIMER
This document has been reviewed in accordance with the U.S. Environmental Protection
Agency's peer and administrative review policies and approved for publication. Mention of trade
names or commercial products does not constitute endorsement or recommendation for use.
-11-
-------
CONTENTS
List of Abbreviations vii
Preface ix
Peer Reviewers xii
Executive Summary xiv
1. INTRODUCTION 1
1.1. BACKGROUND 1
1.2. OVERVIEW 2
2. APPROACH TO RISK ASSESSMENT OF CHEMICAL MIXTURES 4
2.1. THE RISK ASSESSMENT PARADIGM FOR MIXTURES 4
2.1.1. Problem Formulation 4
2.1.2. Hazard Identification and Dose-Response Assessment 4
2.1.3. Exposure 5
2.1.4. Risk Characterization and Uncertainty 5
2.1.5. Incorporating the Paradigm Into Mixtures Guidance 5
2.2. PROCEDURE FOR SELECTING A RISK ASSESSMENT METHOD 6
2.2.1. Introduction 6
2.2.2. Steps for Selection 7
2.2.3. Key Concepts 10
2.2.4. Qualitative Assessments 11
2.2.5. Defaults 11
2.3. DATA QUALITY ASSESSMENT 13
2.3.1. Quality of Exposure Information 13
2.3.2. Quality of Health Effects Information 15
2.3.3. Quality of Interactions Information 16
2.4. CHEMICAL MIXTURE EXPOSURE ASSESSMENT ISSUES 16
2.4.1. Environmental Fate and Transport 17
2.4.2. Importance of the Exposure Sequence for Multiple Chemicals 21
2.4.3. Routes of Exposure 22
2.4.4. Exposure Assessment Summary 23
2.5. DATA AVAILABLE ON WHOLE MIXTURES 23
-in-
-------
CONTENTS (continued)
2.5.1. Data Available on the Mixture of Concern 23
2.5.2. Data Available on a Sufficiently Similar Mixture 24
2.5.3. Data Available on a Group of Similar Mixtures 25
2.6. DATA AVAILABLE ON MIXTURE COMPONENTS 27
2.6.1. Toxicologic Similarity and Dose Addition 28
2.6.2. Independence and Response Addition 29
2.6.3. Interactions Data 30
2.7. FUTURE DIRECTIONS 31
2.7.1. Overview 31
2.7.2. Research Suggestions for Improving Mixture Risk Assessment 33
3. METHODS FOR WHOLE-MIXTURES DATA 37
3.1. INTRODUCTION 37
3.1.1. Data Available on the Mixture of Concern 37
3.1.2. Data Available on a Sufficiently Similar Mixture 37
3.1.3. Data Available on a Group of Similar Mixtures 38
3.1.4. Environmental Transformations for Whole Mixtures 39
3.1.5. Uncertainties With Whole-Mixture Studies 39
3.2. WHOLE-MIXTURE RFD/C AND SLOPE FACTORS 40
3.2.1. Introduction 40
3.2.2. Examples of RfD Development for a Whole Mixture 41
3.2.3. Example of Cancer Assessment for a Whole Mixture 42
3.2.4. Procedure for a Whole Mixtures Dose-Response Assessment 43
3.3. COMPARATIVE POTENCY 45
3.3.1. The Comparative Potency Method 45
3.3.2. Theoretical Development 46
3.3.3. Procedures for Applying the Comparative Potency Approach 53
3.4. ENVIRONMENTAL TRANSFORMATIONS 58
3.4.1. Using Environmental Process Information to Determine Mixture
Similarity 58
3.4.2. Procedures for Incorporating Environmental Process Information .... 59
-iv-
-------
CONTENTS (continued)
3.4.3. Geographic Site-Specific Modifications: An Example Using PCB
Mixtures 60
4. METHODS FOR COMPONENT DATA 66
4.1. INTRODUCTION 66
4.1.1. Criteria for Dose Addition vs. Response Addition 66
4.1.2. Toxicologic Interactions 74
4.1.3. Risk Assessment Strategy 76
4.1.4. Cautions and Uncertainties With Component-Based Assessments ... 76
4.2. HAZARD INDEX 79
4.2.1. Definition 79
4.2.2. Information Requirements 80
4.2.3. Alternative Formulas 80
4.2.4. Comparison of the Hazard Index Formulas 85
4.2.5. Interpretation 87
4.2.6. Reference Value for a Mixture 88
4.3. INTERACTION-BASED HI 90
4.3.1. HI Definition 90
4.3.2. Information Requirements 102
4.3.3. Interpretation 103
4.4. RELATIVE POTENCY FACTORS 103
4.4.1. Introduction 103
4.4.2. Procedures for Developing a Relative Potency Factor (RPF)
Approach 109
4.4.3. Risk Characterization Using RPFs 115
4.4.4. Hypothetical Example of RPF Approach 116
4.5. RESPONSE ADDITION 119
4.5.1. Background 119
4.5.2. Individual Toxicity 121
4.5.3. Population Toxicity 122
4.5.4. Application 124
4.5.5. Use of Upper Bound Response Estimates 125
4.5.6. Qualitative Judgments of Interaction Potential 126
-v-
-------
CONTENTS (continued)
REFERENCES 131
APPENDIX A: GUIDELINES FOR THE HEALTH RISK ASSESSMENT OF CHEMICAL
MIXTURES A-l
APPENDIX B: DEFINITIONS B-l
APPENDIX C: PHARMACOKINETICS C-l
C.I. PHARMACOKINETIC/PHARMACODYNAMIC MODELING ... C-l
C.2. PHARMACOKINETIC PRINCIPLES:
CHEMICAL MIXTURES C-5
C.2.1. Absorption C-5
C.2.1.1. Gastrointestinal C-5
C.2.1.2. Pulmonary C-7
C.2.1.3. Dermal C-7
C.2.1.4. Elimination C-8
-VI-
-------
LIST OF ABBREVIATIONS
ACGIH American Conference of Government Industrial Hygienists
AHH Aryl Hydrocarbon Hydroxylase
ATSDR Agency for Toxic Substances and Disease Registry
B[a]P Benzo(a)pyrene
BINWOE Binary Weight-of-Evidence
BMD Benchmark Dose
CRAVE Carcinogen Risk Assessment Verification Endeavor
EDX Effective Dose in x Percent of Test Animals
GSH Glutathione
HI Hazard Index
HQ Hazard Quotient
IRIS Integrated Risk Information System
LDX Lethal Dose in x Percent of Test Animals
LOAEL Lowest-Observed-Adverse-Effect Level
MFO Mixed Function Oxidase
MOAEL Minimum-Observed-Adverse-Effect Level
MOE Margins of Exposure
MT Metallothionein
NAS National Academy of Sciences
NOAEL No-Observed-Adverse-Effect Level
NRC National Research Council
-Vll-
-------
LIST OF ABBREVIATIONS (continued)
OSHA Occupational Safety and Health Administration
PAH Polycyclic Aromatic Hydrocarbon
PBPK Physiologically Based Pharmacokinetics
PBPK/PD Physiologically Based Pharmacokinetics and Pharmacodynamics
PCB Polychlorinated Biphenyl
POM Polycyclic Organic Material
RfC Reference Concentration
RfD Reference Dose
RPF Relative Potency Factor
TEF Toxicity Equivalence Factor
TEQ 2,3,7,8-TCDD Toxicity Equivalents
TOC Total Organic Carbon
TTC Toxicity-Specific Concentration
TTD Target Organ Toxicity Dose
UF Uncertainty Factor
WHO World Health Organization
WOE Weight of Evidence
-Vlll-
-------
PREFACE
The U.S. EPA's Risk Assessment Forum (Forum) is publishing the Supplemental
Guidance for Conducting Health Risk Assessment of Chemical Mixtures as a supplement to the
EPA's Guidelines for the Health Risk Assessment of Chemical Mixtures (Guidelines)(\J.S. EPA,
1986) (Appendix A). The 1986 Guidelines represent the Agency's science policy and are a
procedural guide for evaluating data on the health effects from exposures to chemical mixtures.
The principles and concepts put forth in the Guidelines remain in effect. However, where the
Guidelines describe broad principles and include few specific procedures, the present guidance is
a supplement that is intended to provide more detail on these principles and procedures.
To address concerns over health risks from multichemical exposures, the U.S.
Environmental Protection Agency published the Guidelines for the Health Risk Assessment of
Chemical Mixtures in 1986 (U.S. EPA, 1986) (Appendix A). The Guidelines describe broad
concepts related to mixture exposure and toxicity and include few specific procedures. In 1989
EPA published guidance for the Superfund program on hazardous waste that gave practical steps
for conducting a mixtures risk assessment (U.S. EPA, 1989a). Also in 1989, EPA published the
revised document on the use of Toxicity Equivalence Factors for characterizing health risks of
the class of chemicals including the dibenzo-dioxins and dibenzofurans (U.S. EPA, 1989b). In
1990, EPA published a Technical Support Document to provide more detailed information on
toxicity of whole mixtures and on toxicologic interactions (e.g., synergism) between chemicals in
a binary (two-chemical) mixture (U.S. EPA, 1990). The concept of toxicologic similarity was
also discussed. The Environmental Criteria and Assessment Office (now the National Center for
Environmental Assessment) followed this with the production of a Technical Support Document
on Health Risk Assessment of Chemical Mixtures (U.S. EPA, 1990b).
This supplementary guidance document is a result of several influences. Because the
science of environmental risk assessment has continued to evolve and EPA has learned from an
array of experiences, the Agency charged the Risk Assessment Forum with developing guidance
on challenging issues such as cumulative risk assessment. Part of the Forum's response to this
charge was to establish a Technical Panel to ensure that the advances in the area of chemical
mixtures health risk assessment are reflected in Agency-wide guidance materials. Through the
evaluation of waste sites for mixtures risks it has become apparent that the exposure scenarios for
these sites are extremely diverse. Moreover, the quality and quantity of pertinent information
available for risk assessment has varied considerably for different mixtures. Other Agency and
external initiatives have influenced the development of the chemical mixtures supplementary
guidance:
-IX-
-------
• The National Academy of Sciences has issued a recommendation to move away
from single-chemical assessments. (NRC, 1994)
• In 1997, EPA's Science Policy Council issued a policy statement on cumulative risk
assessment. This policy addressed the first step in the overall assessment process
(i.e., problem formulation) (U.S. EPA, 1997a).
• Siting activities have raised the issue of multiple chemical exposures. Parties are
concerned not only about what risks are associated with releases from a particular
facility, but also the potential combined effects of exposures from other sources in
the area.
• EPA's research strategy for 2000 and beyond emphasizes research on chemical
mixtures.
When the 1986 Guidelines were published, the Agency recognized that the Guidelines
would need to be updated as the science of chemical mixture assessment evolved. Research
efforts were undertaken immediately and by 1988 Agency offices were discussing revision
topics. By 1989, under the auspices of the Risk Assessment Forum, efforts were underway to
revise the Guidelines. Updates to the Guidelines were reviewed in a June 1997 Internal Risk
Assessment Forum Review Draft of the Guidance on Health Risk Assessment of Chemical
Mixtures. The Technical Panel revised the document in accordance with comments received
during the July 1997 review. In June 1998 the Forum sponsored an Agency review and
colloquium. Over the next months the Technical Panel worked with commenters to address
issues raised during the 1998 colloquium to prepare the document for external peer review. It
was determined at this time that the broad principles and concepts put forth in the 1986
Guidelines remained applicable, but needed more detail. As a result it was determined that the
document would supplement, and not replace the 1986 Guidelines. An external peer review was
convened in May 1999. Twelve independent experts representing consulting, academia, industry,
the U.S. Department of Health Agency for Toxic Substances and Disease Registry, and the TNO
Nutritional and Food Research Institute of the Netherlands, reviewed the revised supplementary
document dated April 1999. The experts provide comments that reflected their experience and
expertise in toxicology, mechanistic and pharmacokinetic modeling, statistics, and risk
assessment (risk assessment of chemical classes, of complex and unidentifiable mixtures, and of
multi-chemical exposures at Superfund sites). Their comments are documented in the report
entitled, Report of the Peer Review Workshop on the Guidance for Conducting Health Risk
-x-
-------
Assessments of Chemical Mixtures (Eastern Research Group Inc., 1999). During the summer of
1999 the Technical Panel considered comments from the external experts and from the Forum in
revising and reorganizing the supplementary document. This series of internal and external
reviews has ensured that the supplementary guidance is consistent with related science and
Agency guidance developments.
After an abbreviated overview of the background and scope, the Supplementary Guidance
for Conducting Health Risk Assessment of Chemical Mixtures document puts forth the risk
assessment paradigm for mixtures. This paradigm begins with problem formulation, then briefly
discusses hazard identification, dose-response assessment, exposure, and risk characterization.
The document is organized according to the type of data available to the risk assessor, ranging
from data-rich to data-poor situations. (See Figure 2-1). Procedures are described for assessment
using data on the mixture of concern, data on a lexicologically similar mixture, and data on the
mixture component chemicals. The state of the science varies dramatically for these three
approaches. The whole-mixture procedures are most advanced for assessing carcinogenic risk,
mainly because of the long use of in vitro mutagenicity tests to indicate carcinogenic potency. In
vitro test procedures for noncancer endpoints are still in the pioneering stage. In contrast, the
component-based procedures, particularly those that incorporate information on toxicologic
interactions, are most advanced for noncarcinogenic toxicity. No single approach is
recommended in this supplementary guidance. Instead, guidance is given for the use of several
approaches depending on the nature and quality of the data. The appendices contain definitions,
a discussion on toxicologic interactions and pharmacokinetic models, and a reprint of the 1986
Guidelines.
-XI-
-------
PEER REVIEWERS
The following individuals reviewed the April 1999 draft, Guidance for Conducting
Health Risk Assessment of Chemical Mixtures.
Kenneth Brown
Principal
Kenneth G. Brown, Inc.
4917ErwinRoad
Durham, NC 27707
Gail Charnley
HealthRisk Strategies
826 A Street, SE
Washington, DC 20003
K.C. Donnelly
Associate Professor
Texas A&M University
Department of Veterinary Anatomy
and Public Health
College Station, TX 77845-4458
Michael Dourson
Director
Toxicology Excellence for Risk
Assessment
4303 Hamilton Avenue
Cincinnati, OH 45223
Hi sham El-Masri
Environmental Health Scientist
Division of Toxicology
ATSDR
1600 Clifton Road, NE (E-29)
Atlanta, GA 30333
John Groten
Head, Dept. of Exploratory Toxicology
TNO Nutrition and Food Research Inst.
P.O. Box 48
Utrechtseweg 48
Zeist, The Netherlands 3700 AJ
Kannan Krishnan
Faculty of Medicine, Department of
Occupational and Environmental Health
University of Montreal
2375 chemin de la Cote Ste-Catherine
Room 4105
Montreal, PQ Canada H3T 1A8
Moiz Mumtaz
Science Advisor, Division of Toxicology
ATSDR
1600 Clifton Road, NE (E-29)
Atlanta, GA 30333
Michael Pereira
Professor, Director, Center for
Environmental Medicine
Medical College of Ohio
Department of Pathology
3055 Arlington, HEB - Room 200F
Toledo, OH 43614-5806
Resha Putzrath
Principal
Georgetown Risk Group
3223 N Street, NW
Washington, DC 20007
Paul Feder
Research Leader, Statistics and
Data Analysis Systems
Battelle
505 King Avenue - Room 11-7060
Columbus, OH 43201-2693
Toxicology Division
-Xll-
-------
Jay Silkworth
Research Scientist/Toxicology
General Electric Company
Corporate Research and Development
1 Research Circle, Building K-l, 3C13
Niskayuna, NY 12309
-Xlll-
-------
EXECUTIVE SUMMARY
This supplementary guidance document is organized according to the type of data
available to the risk assessor, ranging from data rich to data poor situations. This organization
reflects the approaches to chemical mixture risk assessment recommended in the 1986
Guidelines for the Health Risk Assessment of Chemical Mixtures (Appendix A). This document
describes more detailed procedures for chemical mixture assessment using data on the mixture of
concern, data on a lexicologically similar mixture, and data on the mixture component chemicals.
The state-of-the-science varies dramatically for these three approaches. It is recommended that
the risk assessor implement several of the approaches that are practical to apply and evaluate the
range of health risk estimates that are produced.
This document suggests that the selection of a chemical mixture risk assessment method
follows the outline in the flow chart shown in Figure 2-1, which begins with an assessment of
data quality and then leads the risk assessor to selection of a method through evaluation of the
available data. The major concerns for the user are whether the available data are on components
or whole mixtures, whether the data are composed of either similar components or similar
mixtures that can be thought of as acting by similar toxicologic processes, and whether the data
may be grouped by emissions source, chemical structure, or biologic activity. Method-specific
user fact sheets for quantitative risk assessment can be found in Sections 2.5 and 2.6 and are
intended to provide a concise overview of each currently available method. These fact sheets
provide the following information relative to the risk assessment approach:
• Type of Assessment
• Data Requirements
• Section(s)
• References
• Strategy of Method
• Ease of Use
• Assumptions
• Limitations
• Uncertainties
In Figure 2-1, an evaluation of the data may lead the user to decide that only a qualitative
analysis should be performed. This generally occurs in cases where data quality is poor,
inadequate quantitative data are available, data on a similar mixture cannot be classified as
-xiv-
-------
"sufficiently similar" to the mixture of concern, exposures cannot be characterized with
confidence, or method-specific assumptions about the toxicologic action of the mixture or of its
components cannot be met. When this occurs, the risk assessor can still perform a qualitative
assessment that characterizes the potential human health impacts from exposure to that mixture.
Such a risk characterization should discuss each element of the risk assessment paradigm,
including available information on the mixture itself, on its components, and on potential
interactions among the components. Any information on fate and transport of the mixture that
would affect its final composition at the time of exposure should be noted.
The assessment of chemical mixtures is an area of active scientific investigation. As new
information relevant to health risk from exposure to chemical mixtures becomes available,
additional guidance documents will be published.
-xv-
-------
1. INTRODUCTION
1.1. BACKGROUND
Although some potential environmental hazards involve significant exposure to only a
single compound, most instances of environmental contamination involve concurrent or
sequential exposures to a mixture of compounds that may induce similar or dissimilar effects
over exposure periods ranging from short-term to lifetime. For the purposes of this guidance
document, a mixture will be defined as any combination of two or more chemical substances,
regardless of source or of spatial or temporal proximity, that can influence the risk of chemical
toxicity in the target population (U.S. EPA, 1986). In some instances, the mixtures are highly
complex, consisting of scores of compounds that are generated simultaneously as by-products
from a single source or process (e.g., coke oven emissions and diesel exhaust). In other cases,
complex mixtures of related compounds are produced as commercial products (e.g., PCBs,
gasoline and pesticide formulations) and eventually released into the environment. Another
category of mixtures consists of compounds, often unrelated chemically or commercially, that are
placed in the same area for disposal or storage, and have the potential for combined exposure to
humans. Multichemical exposures are ubiquitous, including air and soil pollution from
municipal incinerators, leakage from hazardous waste facilities and uncontrolled waste sites, and
drinking water containing chemical substances formed during disinfection.
To address concerns over health risks from multichemical exposures, the U.S.
Environmental Protection Agency, hereafter referred to as EPA, issued Guidelines for the Health
Risk Assessment of Chemical Mixtures in 1986 (U.S. EPA, 1986) (Appendix A). Those
Guidelines described broad concepts related to mixture exposure and toxicity and included few
specific procedures. In 1989, EPA published guidance for the Superfund program on hazardous
waste that gave practical steps for conducting a mixtures risk assessment (U.S. EPA, 1989a).
Also in 1989, EPA published the revised document on the use of Toxicity Equivalence Factors
for characterizing health risks of the class of chemicals including the dibenzo-dioxins and
dibenzofurans (U.S. EPA, 1989b). In 1990, EPA published a Technical Support Document to
provide more detailed information on toxicity of whole mixtures and on toxicologic interactions
(e.g., synergism) between chemicals in a binary (two-chemical) mixture (U.S. EPA, 1990). The
concept of toxicologic similarity was also discussed.
As more waste sites were evaluated for mixtures risks, it became apparent that the
exposure scenarios for these sites were extremely diverse. Moreover, the quality and quantity of
pertinent information available for risk assessment varied considerably for different mixtures.
Such difficulties continue. Occasionally, the chemical composition of a mixture is well
characterized, levels of exposure to the population are known, and detailed toxicologic data on
-1-
-------
the mixture are available. Most frequently, some components of the mixture are unknown,
exposure data are uncertain or vary over time, and toxicologic data on the known components of
the mixture are limited. Consequently, this document has been developed to supplement the
earlier guidance documents and is organized according to the type of data available to the risk
assessor, ranging from data-rich to data-poor situations. Procedures are described for assessment
using data on the mixture of concern, data on a lexicologically similar mixture, and data on the
mixture component chemicals. The state of science varies dramatically for these three
approaches. The whole-mixture procedures are most advanced for assessing carcinogenic risk,
mainly because of the long use of in vitro mutagenicity tests to indicate carcinogenic potency. In
vitro test procedures for noncancer endpoints are still in the pioneering stage. In contrast, the
component-based procedures, particularly those that incorporate information on toxicologic
interactions, are most advanced for noncarcinogenic toxicity.
Mixture risk assessments usually involve substantial uncertainties. If the mixture is
treated as a single complex substance, these uncertainties range from inexact descriptions of
exposure to inadequate toxicity information. When viewed as a simple collection of a few
component chemicals, the uncertainties include the generally poor understanding of the
magnitude and nature of toxicologic interactions, especially those interactions involving three or
more chemicals. Because of these uncertainties, the assessment of health risk from chemical
mixtures should include a thorough discussion of all assumptions and the identification, when
possible, of the major sources of uncertainty. No single approach is recommended in this
supplementary guidance. Instead, guidance is given for the use of several approaches depending
on the nature and quality of the data.
1.2. OVERVIEW
The primary purpose of this document is to generate a consistent Agency approach for
assessing health risks from exposures to multiple chemicals, denoted in this guidance by the
general term "mixtures." The resulting mixtures risk assessments are intended to assist decision
makers by characterizing health risks for the particular exposure conditions of interest. Because
exposure scenarios and the available supporting data are highly diverse, this document has been
developed as a procedural guide that emphasizes broad underlying principles of the various
science disciplines (environmental chemistry, toxicology, pharmacology, statistics) necessary for
providing information on the relationship between multichemical exposure and potential health
effects. Specific approaches to be used for the evaluation of the various kinds of mixture data are
also discussed.
This document addresses only risks to human health from multichemical exposures.
Ecological effects are beyond its scope, even though many of the procedures might be adaptable
-2-
-------
to ecological risk assessment from multiple stressors. Because other Agency guidelines exist that
address exposure assessment and specific toxic endpoint evaluations, this guidance focuses on
procedures for dose-response assessment and risk characterization.
It is not the intent of this guidance document to regulate any social or economic aspects
concerning risk of injury to human health or the environment caused by exposure to a chemical
agent(s). All such action is addressed in specific statutes and federal legislation and is
independent of this guidance.
This guidance document represents a supplement to the original Guidelines of 1986 and
is intended to reflect the evolutionary scientific development in the area of chemical mixtures
risk assessment. New guidance has been provided that gives more specific details on the nature
of the desired information and the procedures to use in analyzing the data. Among these are
methods for using whole-mixture data on a lexicologically similar mixture, methods for
incorporating information on toxicologic interactions to modify a Hazard Index (HI), and
generalized procedures for mixtures involving classes of similar chemicals. There are also
expanded discussions of the concerns when using only whole-mixture data as well as when using
only data on the individual chemical components.
The assessment of chemical mixtures is an area of active scientific investigation. Some
of the procedures herein for chemical mixtures have had little or no application to date in actual
health risk assessments. Their use is encouraged, along with research on new procedures to
improve or replace those discussed here. As new information relevant to health risk from
exposure to chemical mixtures becomes available, additional guidance documents will be
published.
-3-
-------
2. APPROACH TO RISK ASSESSMENT OF CHEMICAL MIXTURES
2.1. THE RISK ASSESSMENT PARADIGM FOR MIXTURES
Human health risk assessments done by EPA generally follow the paradigm established
by the National Academy of Sciences (NRC, 1983). This paradigm describes a group of
interconnected processes for performing a risk assessment that include hazard identification,
dose-response assessment, exposure assessment, and risk characterization. These four parts of
the paradigm are used as the foundation for the procedures presented in this guidance. Preamble
to all is problem formulation, which is defined in EPA's (1998a) Ecological Risk Assessment
Guidelines as "a process for generating and evaluating preliminary hypotheses about
why...effects have occurred or may occur." This EPA guidance for assessing risks from
exposures to chemical mixtures begins with problem formulation as the initial step; much of the
information about this key step has been adapted from the Ecological Risk Assessment
Guidelines, and the reader is referred to Chapter 3 of that document for a more comprehensive
discussion (U.S. EPA, 1998a).
2.1.1. Problem Formulation
Problem formulation, which provides the foundation for the entire risk assessment,
consists of three initial steps: (1) evaluate the nature of the problem, (2) define the objectives of
the risk assessment, and (3) develop a data analysis and risk characterization plan. The quality,
quantity, and pertinence of information will determine the course of problem formulation. It
concludes with three products: (1) selection of assessment endpoints, (2) review of the
conceptual models that describe the relationship between exposure to a mixture of chemicals and
risk, and (3) adjusting the analytic plan. (The pertinence of the information that is available at
the outset of the assessment, in combination with the assessment objectives, will identify the
types of information that should be collected through the analytic plan.) Ideally, the problem is
formulated jointly by risk analysts and risk managers. While the steps and outcomes associated
with problem formulation are presented separately, experiences from ecological applications and
Superfund site assessments show the process to be frequently interactive and iterative rather than
linear.
2.1.2. Hazard Identification and Dose-Response Assessment
In hazard identification., available data on biological endpoints are used to determine if a
material is likely to pose a hazard to human health. These data are also used to define the type of
potential hazard (e.g., does the material induce tumor formation or act as a kidney toxicant). In
the dose-response assessment, data (most often from animal studies and occasionally from
-4-
-------
human studies) are used to estimate the amount of material that may produce a given effect in
humans. The risk assessor may calculate a quantitative dose-response relationship usable for
low-dose exposure, often by applying mathematical models to the data.
2.1.3. Exposure
The exposure assessment seeks to determine the extent to which a population is exposed to
the material. Exposure assessment uses available data relevant to population exposure, such as
emissions data, measurement of the material in environmental media, and biomarker information.
Fate and transport of the material in the environment, as well as media, pathways, and routes of
exposure, may all be considered in the exposure assessment. Data limitations on the
environmental concentrations of interest often necessitate the use of modeling to provide relevant
estimates of exposure.
2.1.4. Risk Characterization and Uncertainty
Risk characterization is the integrating step in the risk assessment process that
summarizes assessments of effects on human health and ecosystems and assessments of exposure
from multiple environmental media, identifies human subpopulations or ecological species at
elevated risk, combines these assessments into characterizations of human and ecological risk,
and describes the uncertainty and variability in these characterizations. In March 1995, the
Administrator of EPA issued the Policy for Risk Characterization at the U.S. Environmental
Protection Agency (U.S. EPA, 1995). The purpose of this policy statement was to ensure that
critical information from each stage of a risk assessment be presented in a manner that provides
for greater clarity, transparency, reasonableness, and consistency in risk assessments. Most of
the 1995 Policy for Risk Characterization at the U.S. EPA was directed toward assessment of
human health consequences of exposures to an agent. Key aspects of risk characterization
identified in the 1995 Policy for Risk Characterization at the U.S. EPA include these: bridging
risk assessment and risk management, discussing confidence and uncertainties, and presenting
several types of risk information. Another publication, Science and Judgment in Risk Assessment
(NRC, 1994), produced primarily for implementation of the 1990 Amendment to the Clean Air
Act but applicable more generally, emphasized that the goal of risk characterization is to provide
understanding of the type and magnitude of potential adverse effects of an agent under the
particular circumstances of its release.
2.1.5. Incorporating the Paradigm Into Mixtures Guidance
EPA regularly publishes guidelines to provide for consistency of application and
communication of risk assessment. Guidelines were published in 1986 on assessment of the
-5-
-------
following areas: exposure, developmental effects, germ cell mutagenicity, carcinogenic effects,
and chemical mixtures (U.S. EPA, 1986, 1987). Since that time, the Agency has revised some of
these Guidelines and also published new Guidelines. These include Guidelines on
developmental toxicity (U.S. EPA, 1991a), exposure assessment (U.S. EPA, 1992), cancer
(proposed revisions) (U.S. EPA, 1996a), reproductive toxicity (U.S. EPA, 1996c), and
neurotoxicity (U.S. EPA, 1998b). All of the EPA guidelines for human health risk assessment
incorporate the four parts of the NAS paradigm.
For this supplemental guidance on the risk assessment of chemical mixtures, the four
parts of the paradigm are interrelated and will be found within the assessment techniques that are
presented. For some methods described herein, assessment of dose-response relies both on
decisions in the area of hazard identification and on assessment of potential human exposures.
For mixtures, the use of pharmacokinetics data and models in particular differs from single-
chemical assessment, where they are often part of the exposure assessment. For mixtures, the
dominant mode of toxicologic interaction is the alteration of pharmacokinetic processes, which
strongly depends on the exposure levels of the mixture chemicals. In this guidance, there has
been no effort to categorize methods strictly or arbitrarily into one part of the paradigm. The
methods are organized instead according to the type of available data. In general, risk
characterization takes into account both human health and ecological effects, and also assesses
multiroute exposures from multiple environmental media. This guidance focuses only on the
human health risk assessment for chemical mixtures and only discusses multiroute exposures in
terms of conversions from dermal to oral.
2.2. PROCEDURE FOR SELECTING A RISK ASSESSMENT METHOD
2.2.1. Introduction
The 1986 Guidelines for the Health Risk Assessment of Chemical Mixtures (U.S. EPA,
1986) (Appendix A) recommend three approaches to quantitative health risk assessment of a
chemical mixture, depending upon the type of available data. In the first approach, toxicity data
on the mixture of concern are available; the quantitative risk assessment is done directly from
these preferred data. In the second approach, when toxicity data are not available for the mixture
of concern, the Guidelines recommend using toxicity data on a "sufficiently similar" mixture. If
the mixture of concern and the proposed surrogate mixture are judged to be similar, then the
quantitative risk assessment for the mixture of concern may be derived from health effects data
on the similar mixture. Finally, the third approach is to evaluate the mixture through an analysis
of its components, e.g., using dose addition for similarly acting chemicals and response addition
for independently acting chemicals. These procedures include a general assumption that
interaction effects at low dose levels either do not occur at all or are small enough to be
-------
insignificant to the risk estimate. The Guidelines recommend the incorporation of interactions
data when available, if not as part of the quantitative process, then as a qualitative evaluation of
the risk.
No single approach is recommended in this guidance document. Instead, guidance is
given for the use of several approaches depending on the nature and quality of the available data,
the type of mixture, the type of assessment being made, the known toxic effects of the mixture or
of its components, the toxicologic or structural similarity of mixtures or of mixture components,
and the nature of the environmental exposure. The approaches presented herein represent a mix
of well-known, routine methods with several newer, less well-established techniques. As a
collection, they provide the risk assessor with a number of reasonable options for evaluating risk
for chemical mixtures.
2.2.2. Steps for Selection
This guidance suggests that the selection of a chemical mixture risk assessment method
follow the outline in the flow chart shown in Figure 2-1, which begins with an assessment of data
quality and then leads the risk assessor to selection of a method through evaluation of the
available data. The major concerns for the user are whether the available data are on components
or whole mixtures, whether the data are composed of either similar components or similar
mixtures that can be thought of as acting by similar toxicologic processes, whether the mixture
components act by the same mode of action or are functionally independent, or whether the data
may be grouped by emissions source, chemical structure, or biologic activity.
This document is organized around the decision points in Figure 2-1, so that the user can
refer to specific sections and find guidance on the issues to consider when working through the
flow chart. Appendix B also offers the user a number of definitions to help clarify the
terminology that is unique to chemical mixtures risk assessment. Table B-l presents chemical
mixture definitions in terms of specific criteria including the complexity of the mixture,
similarity of biologic activity, similarity of chemical structure or mixture composition, the
environmental source of the mixture, toxic endpoint, etc. Table B-2 provides definitions for
terms that are used to describe various types of toxicologic interactions including forms of
additivity, antagonism, synergism, and other toxicologic phenomena.
Method-specific user fact-sheets in Sections 2.5 and 2.6 are intended to provide a concise
overview of each currently available method. These fact-sheets provide the following
information relative to the risk assessment approach:
• Type of Assessment: distinguishes whether the approach is a dose-response
assessment or whether it combines dose response and exposure information to
perform a risk characterization.
-7-
-------
Assess Data Quality
adequate
Only Qualitative Assessment
_
CT Components "^3
\
Mixture
of Concern
SufficientlyX
Similar I
Mixture /
/ Group of
Similar
\ Mixtures
f Toxicologically] f Toxicologically] , ,nteractjons i
I Similar J I Independent J I [nteract!OnsJ
oo
Mixture
RfD/C;
Slope
Factor
Comparative
Potency
Environmental
Transformation
Hazard
Index
Relative
Potency
Factors
Response
Addition
Interactions
Hazard
Index
Compare and Identify Preferred Risk Assessment,
Integrate Summary with Uncertainty Discussion
Figure 2-1. TSie different types of mixtures assessments based on the availability and quality of the data.
Ail possible assessment paths should be performed.
-------
• Data Requirements: details the types and amount of data that are needed to carry
out the procedure.
• Section(s): refers the user to sections of this document that provide greater detail
on the approach.
• References: cites reports or publications in which the approach has been applied
in practice or indicates that this is a new procedure.
• Strategy of Method: provides concise directions on how the calculations are
performed.
• Ease of Use: gives a sense of how much effort, expertise, and data are required in
order to apply the approach.
• Assumptions: lists the toxicologic or statistical assumptions that are inherently
made when the data are treated by applying the approach; the user can then decide
if the approach is appropriate for the available data.
• Limitations: suggests problems the user may encounter relative to data gaps or
quality deficiencies, and statistical modeling requirements or goodness-of-fit
issues.
• Uncertainties: indicates unknown elements of the analysis that should be
considered and characterized in the presentation of the risk assessment (e.g., data
are not available, mode of action is unknown, scientific judgments are made,
exposures are not well characterized, extrapolations are made, etc.).
Following an assessment of data quality, the first major distinction addressed in Figure
2-1 is whether the type of available data is whole mixture data or mixture component
information. This distinction points the risk assessor toward methods that are available for these
specific types of data. Methods available for whole mixtures then depend on whether there is
information directly available on the mixture of concern or only on sufficiently similar mixtures
or groups of similar mixtures. Methods available for component data then depend on whether
there are interactions data available or whether the components act with a similar mode of action
or are lexicologically independent. In these cases, the outcome is a quantitative assessment with
a complete risk characterization and uncertainty discussion presented.
Figure 2-1 is deceptively simple, however, as many of the issues that are represented in
the diagram require the use of scientific judgment or data that may not be readily available. In
addition, there will often be mixtures for which there exist both whole-mixture and component
data, so that the choice of method will not be clear (for example, both epidemiologic data and
component toxicity data exist for evaluation of health effects from exposure to chlorinated
drinking water). Furthermore, the true toxicologic mechanism of action (see Section 2.2.3) is
rarely known for a given mixture or even for most of its components; thus the judgments that are
made of toxicologic similar action or independence of action, for example, will be uncertain. It
is recommended, therefore, that the risk assessor implement several of the approaches that are
practical and evaluate the range of health risk estimates that are produced.
-9-
-------
2.2.3. Key Concepts
There are several concepts that must be understood in order to evaluate a chemical
mixture (see Appendix B). The first is the role of toxicologic similarity which, in this document,
is considered along a continuum of information. The term mode of action is defined as a series
of key events and processes starting with interaction of an agent with a cell, and proceeding
through operational and anatomical changes causing disease formation. "Mode" of action is
contrasted with "mechanism" of action, which implies a more detailed understanding and
description of events, often at the molecular level, than is meant by mode of action. The specific
term toxicologic similarity represents a general knowledge about the action of a chemical or a
mixture and can be expressed in broad terms such as at the target organ level in the body (e.g.,
enzyme changes in the liver). In this document, assumptions about toxicologic similarity are
made in order to choose among risk assessment methods. In general, we assume a similar mode
of action across mixtures or mixture components and, in some cases, this requirement may be
relaxed to require that these chemicals act only on the same target organ.
The second key concept in understanding mixtures risk assessment is the assumption of
similarity or, in contrast, independence of action. The term sufficiently similar mixture refers to a
mixture that is very close in composition to the mixture of concern, such that differences in their
components and their proportions are small; the risk assessor can then use the data from the
sufficiently similar mixture to make a risk estimate about the mixture of concern. The term
similar components refers to the single chemicals within a mixture that act by the same mode of
action and may have comparable dose-response curves; the risk assessor can then apply a
component-based approach that uses these characteristics to form the basis of the risk
assessment. The term group of similar mixtures refers to chemically related classes of mixtures
that act by a similar mode of action, have closely related chemical structures, and occur together
routinely in environmental samples, usually because they are generated by the same commercial
process; the risk assessor can use what is known about the shifts in chemical structure and
relative potency of the components to perform a risk assessment. Finally, the term independence
of action is defined as mixture components that cause different kinds of toxicity, or effects in
different target organs; the risk assessor may then combine the probabilities of toxic effects for
the individual components.
Another key concept for this document is the understanding of language referring to
toxicologic interactions, which is defined here as any toxic responses that are greater than or less
than what is observed under an assumption ofadditivity. The term additivity is used when the
effect of the combination of chemicals can be estimated directly from the sum of the scaled
exposure levels (dose addition) or of the responses (response addition) of the individual
components. There are a myriad of terms (see Appendix B, Table B-2) that represent various
-10-
-------
kinds of interaction effects (e.g., inhibition, antagonism, masking). The most common and
general of these refer to effects that are greater than additive (i.e., synergistic) or less than
additive (i.e., antagonistic).
2.2.4. Qualitative Assessments
In Figure 2-1, an evaluation of the data may lead the user to decide that only a qualitative
analysis should be performed. This generally occurs in cases where data quality is poor, there are
inadequate quantitative data available, data on a similar mixture cannot be classified as
"sufficiently similar" to the mixture of concern, exposures cannot be characterized with
confidence, or method-specific assumptions about the toxicologic action of the mixture or of its
components cannot be met. When this occurs, the risk assessor can still do a qualitative
assessment that characterizes the potential human health impacts from exposure to that mixture.
Such a risk characterization should discuss each element of the risk assessment paradigm,
including available information on the mixture itself, on its components, and on potential
interactions among the components. Any information on fate and transport of the mixture that
would affect its final composition at the time of exposure should be noted.
2.2.5. Defaults
The development of a risk assessment for a chemical mixture will generally involve the
examination of complex exposures and toxicities and the application of specific methods as well
as scientific judgment. This process necessarily involves a thorough examination and discussion
of the uncertainties, limitations, and assumptions inherent in exposure assessment, fate and
transport, uptake and pharmacokinetics, and the magnitude and nature of toxicity and toxicant
interactions. Because of the complexity of considerations that must be undertaken to develop a
chemical mixtures health risk assessment, it is not practical to recommend a clear listing of
default procedures that covers all cases. In many cases, information gaps will be too substantial
to allow use of defaults, so that only a qualitative risk assessment can be performed.
Nonetheless, for some restricted situations, default values and methods can be recommended.
This section outlines the philosophy underlying their choice.
For low exposure levels when no interactions information is available, default methods
using an additivity assumption are given. For the component chemicals in a mixture that show
dissimilar toxicity, response addition (Sections 2.6.2, 4.1, and 4.5) is recommended. For the
component chemicals that show similar toxicity, dose addition (Sections 2.6.1, 4.1, 4.2, and 4.4)
is recommended. Under dose addition, the general procedure is to scale the doses of the
components by their relative potency and add the scaled doses together; the mixture response is
then estimated for the combined dose. Under response addition, the general procedure is to first
-11-
-------
determine the risks per the exposure for the individual components; the mixture risk is then
estimated by adding the individual risks together. These processes are fundamentally different
and require different assumptions of the data in order for them to be used appropriately. Finally,
if interactions data are available, the default recommendation is that they be incorporated into the
risk assessment by using the interaction-based Hazard Index (HI) (Sections 2.6.3, 4.1, and 4.3).
Dose addition is the default approach in situations where the dose for each individual
component is at a level at which effects are not expected to occur, be observable, or be of
concern; however, when the doses are combined, effects of concern may be expected or observed
in response to the higher dose level of the mixture. A method based on dose addition that has
been used most often by EPA is the HI, where HI < 1 indicates a mixture exposure of no
significant concern (U.S. EPA, 1989a). True dose addition is applied by scaling the potencies of
all the components in the mixture with the same mechanism of action to an index chemical,
adding the scaled doses together to give the equivalent dose in terms of the index chemical, and
using the index chemical's dose-response curve to estimate the response for the equivalent total
mixture dose. Dose addition is different from response addition because two assumptions are
made: that all of the components have similar uptake, pharmacokinetics, and toxicologic
processes; and that the dose-response curves of the components have congruent or similar shape
(Teuschler and Hertzberg, 1995). This means that, for equal effects, the dose of one component
is a constant multiple of the dose of a second component.
The interaction-based HI is the default approach for using interactions data to modify
simple dose addition. This approach uses binary interactions data for the components of the
mixture to modify the HI. The factors that are used include the interaction magnitude at low
doses, the toxicity of each component relative to each other component, the weight of evidence
of the interactions data, and the relative proportions of the components in the mixture.
Response addition is the default approach when the component chemicals are functionally
independent. It is most often applied when an effect that is of concern is expected to be present
at low dose levels for each of the component chemicals, even though it is highly unlikely to be
observable at these low levels in either epidemiologic or toxicologic studies; the mixture risk is
then usually approximated by the sum of the individually low risks of the independently acting
component chemicals. For example, response addition has often been used for the risk
assessment of mixtures of carcinogens (Gaylor et al., 1997; U.S. EPA, 1989a). Response
addition is different from dose addition in that it does not assume similar kinetics or a similar
mode of action and does not assume that the dose-response curves have similar shape. It
assumes that the components of the mixture are functionally independent of one another at low
exposure levels (Mumtaz and Hertzberg, 1993), so that the risks may be added together (see
Section 4.5 for details on interpretation and calculation). Because response addition does not
-12-
-------
require a similar mode of action across the chemicals in the mixture, it allows for combining
risks across chemicals even if they have different types of endpoints. An example is the
combined risk of any kind of reproductive toxicity for a set of chemicals with different modes of
action.
2.3. DATA QUALITY ASSESSMENT
The first consideration in Figure 2-1 is the assessment of data quality relative to its
relevancy, completeness, quantitative nature, and certainty in three areas: exposure information,
health effects information, and information on interactions. Table 2-1 presents a classification
scheme for assessing the quality and nature of the available mixtures data. Consideration of the
factors presented in Table 2-1 can be used to guide the risk assessor through Figure 2-1. This
evaluation can assist the decision of whether to quantify the risk (the first step in Figure 2-1), and
can be included in a discussion of overall quality of the risk assessment. Usually a classification
of "FAIR" or better is required for quantitative risk assessment. For example, a "GOOD"
classification for each of exposure information, health effects information and information on
interactions would lead the risk assessor to consider the data quality to be adequate for
quantification, with good data available for both the exposure and toxicity aspects of the mixture
of concern. Figure 2-1 would then guide the risk assessor to perform a risk assessment directly
on the mixture of concern by calculating, for example, a toxicity value for the mixture, such as a
Reference Dose (RfD) or slope factor. A "POOR" classification for one or more of these
categories would likely lead the risk assessor to decide that data quality was inadequate; in this
case, Figure 2-1 directs the risk assessor to perform only a qualitative risk assessment. With
"FAIR" information on each of exposure, health effects, and interactions, the risk assessor would
conclude that data quality was adequate to estimate both the exposure and toxicity of the
components of the mixture, and furthermore to use the available interactions data on the
components in the assessment. Under these conditions, Figure 2-1 indicates that an interaction-
based HI approach would be appropriate. It is the purview of the risk assessor to decide at what
point the validity of the risk assessment is compromised by the data quality to such a degree that
only a qualitative assessment should be performed.
2.3.1. Quality of Exposure Information
Exposure information ideally includes all data needed to characterize the human exposure
to the mixture of concern from the point of environmental release to the point of human intake.
There are several details needed to quantify exposure to chemical mixtures; these include:
-13-
-------
Table 2-1. Classification scheme for the quality of available mixtures data3
Exposure Information6
GOOD - Monitoring information either alone or in combination with modeling information is sufficient
to accurately characterize human exposure to the mixture or its components.
- Modeling information is sufficient to reasonably characterize human exposure to the mixture
or its components.
FAIR - Exposure estimates for some components are lacking, uncertain, or variable. Information on
health effects or environmental chemistry suggests that this limitation is not likely to
substantially affect the risk assessment.
- Not all components in the mixture have been identified, or levels of exposure are highly
uncertain or variable. Information on health effects or environmental chemistry is not
sufficient to assess the effect of this limitation on the risk assessment.
POOR - The available exposure information is insufficient for conducting a risk assessment.
Health Effects Information
GOOD - Full health effects data are available and relatively minor extrapolation is required.
- Full health effects data are available but extensive extrapolation is required for route or
duration of exposure or for species differences. These extrapolations are supported by
pharmacokinetic considerations, empirical observations, or other relevant information.
FAIR - Full health effects data are available, but extensive extrapolation is required for route or
duration of exposure or for species differences. These extrapolations are not directly
supported by the information available.
- Certain important health effects data are lacking and extensive extrapolations are required for
route or duration of exposure or for species differences.
POOR - A lack of health effects information on the mixture and its components in the mixture
precludes a quantitative risk assessment.
Information on Interactions
GOOD - Assessment is based on toxicologic data on the mixture of concern.
- Assessment is based on data on a sufficiently similar mixture.
FAIR - Quantitative interactions of all components are well characterized.
- The assumption of additivity is justified based on the nature of the health effects and on the
number of component compounds.
POOR - Interactions information is inadequate, an assumption of additivity cannot be justified, and no
quantitative risk assessment can be conducted.
aSee text for discussion of sufficient similarity, adequacy of data, and justification for additivity
assumptions.
bSee the Agency's guidelines for exposure assessment (U.S. EPA, 1992) for more complete
information on performing exposure assessments and evaluating the quality of exposure data.
Concentration of the chemical mixture in the medium/media of concern at the
point(s) of human contact
The duration and frequency of exposure should be developed from repeated
measurements or validated models of environmental fate in media to which
individuals are exposed, as well as human activity pattern data. The media
concentrations should be determined at the points of human exposure. If the
exposure data are limited, the analyst should address the degree to which the data
-14-
-------
represent the environmental chemical mixture over space and time.
Environmental transformation of the mixture over time is a key concern.
Analytic chemistry
The analyst should consider both the accuracy and reliability of the measurement
techniques and determine if all of the components have been identified (i.e., are
there unidentified components of the mixture?). The analyst should also
determine if the key environmental reactions have been identified and reaction
rates measured (e.g., environmental half-life) that govern the fate of the mixture.
If components of the environmental mixture have not been detected analytically,
the analyst should describe if and how they were included in the assessment (e.g.,
the compounds were assumed to be present at one-half the detection limit).
Uptake from the environment
The analyst should examine the bioavailability of the mixture for the medium and
route of concern. The ideal data set would be derived from well-conducted
studies that measure either the entire mixture or all the components in the
pertinent exposure media and over the timeframe of concern. (The ideal data may
be derived from accurate analytic measurements at points of human contact or
from validated environmental fate models.) The magnitude of the human exposure
would be measured or modeled on the basis of human activity patterns. Finally,
the bioavailability of the mixture or the components would be known.
Unfortunately, a complete data set is rarely available. The analyst should identify
(and perhaps quantify) uncertainty based on imperfect analytic methods (e.g.,
some constituents may not be characterized by the analytic technique that
represents the current state of the science), extrapolations between concentrations
at measurement points and points of human exposure, incompletely understood
transformation reactions to the mixture in the environment, and bioavailability.
Each of these uncertainties in the risk assessment should be discussed and
accounted for in the final risk characterization.
2.3.2. Quality of Health Effects Information
Health effects information includes both hazard identification and dose-response data on
the complex mixture, a similar mixture, or the components of the mixture. The best data would
be human epidemiologic or human clinical data directly on the complex mixture for which the
health effects of concern are causally linked to the mixture exposure and a dose-response
relationship can be established for the exposure route of interest. Unfortunately, such high-
quality direct information is rarely available, so the risk assessor usually performs one or more
extrapolations. Examples of such extrapolations include using animal data to project potential
human health effects, using inhalation data to predict risks from oral exposure, using component
data to estimate risks for the complex mixture, and using data from short-term human clinical
-15-
-------
studies or subchronic animal bioassays to project human health risks from chronic exposure.
Each of these extrapolations introduces uncertainty into the risk assessment that should be
discussed and accounted for in the final risk characterization.
2.3.3. Quality of Interactions Information
Interactions information includes any data indicating that the toxicologic action of the
complex mixture is greater than or less than what might be expected from exposure to a
colleciton of individual components of the mixture. Thus, human or animal data directly on the
whole mixture implicitly provides interactions information for use in risk assessment. However,
since such data are rarely available, the risk assessor must often rely on component information,
the vast majority of which is laboratory toxicity data on binary combinations of chemicals
(Teuschler and Hertzberg, 1995). The quality of interactions data, whether it be data on the
complex mixture, a sufficiently similar mixture, or simple combinations of the components, can
be judged according to the strength of evidence for three criteria. First, there should be adequate
toxicity data that not only provide information on dose response, but also on the mechanism of
action for the mixture. Second, interactions data should be for the same route of exposure as the
mixture of concern. Furthermore, when data on several different component mixtures are
evaluated, these data should be from comparable studies, such as the same species, same
endpoint of concern, similar laboratory conditions, or comparable study duration. Finally,
observed interactions data that are usable for risk assessment purposes should be lexicologically
significant (i.e., show definite adverse effects). The strength of the evidence for each of these
criteria should be discussed and accounted for in the final risk characterization.
2.4. CHEMICAL MIXTURE EXPOSURE ASSESSMENT ISSUES
While this guidance document is intended to serve risk assessors primarily by informing
them of dose-response and risk characterization methods associated with exposures to chemical
mixtures, the purpose of this section is to highlight additional exposure issues of a general nature
that should be considered when developing a risk assessment for chemical mixtures. The issues
presented in this section should be considered in addition to those normally followed in an
exposure assessment. The Agency's primary guidance in this area is the Exposure Assessment
Guidelines (U.S. EPA, 1992); however, that document primarily focuses on issues pertaining to
single-chemical exposures. Other, more specific exposure assessment issues involving multiple
chemicals will be discussed by the Agency more comprehensively in separate future efforts (e.g.,
the EPA's Risk Assessment Forum is developing a cumulative risk assessment framework as this
guidance goes to press). While there are other important issues related to exposures to chemical
-16-
-------
mixtures, three critical areas will be discussed briefly here: environmental fate, temporal patterns
of exposure, and routes of exposure.
The wide diversity in mixture compositions and site characteristics precludes any
recommendation for a single approach for site-specific modification of the mixture assessment.
Through examples, some steps that should be considered can be articulated. The example in
Section 3.4 demonstrates some of the considerations that should be part of such a modification.
Other modifications based on the exposure and mixture characteristics are encouraged, as long as
they are clearly described and supported with plausible concepts and empirical measurements.
Clearly, the analyst should report the significance of any assumptions utilized as well as the
potential uncertainty and variability associated with the exposure modifications developed for the
risk assessment.
2.4.1. Environmental Fate and Transport
The composition and quantity of a mixture of chemicals may change after release into the
environment. The environmental fate of chemical mixtures released into the environment can be
conceptualized as being composed of three interrelated components: (1) transport through an
individual compartment (e.g., atmospheric dispersion); (2) transfer between environmental
compartments (i.e., partitioning); and (3) transformation mediated by biological, chemical, or
physical processes (e.g., weathering) (Crawford-Brown, 1997, Chapter 2). Even though the
environmental processes that occur within these three components of environmental fate are not
unique to chemical mixtures, the analyst should assess compositional and quantitative changes
that may occur to the chemical mixture of interest in the environment (particularly with respect to
the time from release to exposure), and the impact these will have on exposure and toxicity.
This is particularly important when considering the appropriateness or relevance of an
analytic measurement of quantity or composition of a chemical mixture; the analyst needs to
consider the possible changes to the mixture between the time the measurement was conducted
and the time over which exposures are expected to occur. These environmentally mediated
changes are also important when comparison is made in the assessment to the dose response
exhibited by either a sufficiently similar whole mixture (e.g., comparison of the dose response of
the commercial mixture that has been lexicologically tested to that of the environmental mixture)
or mixture components. The concept of sufficient similarity is not discussed in the 1986
mixtures guidelines (U.S. EPA, 1986, 1987) (Appendix A). Common sense dictates that
sufficient similarity entails the assumption that the toxicologic consequences of exposure to the
two mixtures (i.e., the mixture of concern and the mixture on which data are available) will be
identical or at least indistinguishable from one another. In practice, some degree of chemical
similarity or at least an understanding of how chemical differences between the mixtures affect
-17-
-------
toxicological activity is required. The acceptability of a surrogate, given the degree of accuracy
desired in the risk assessment, should be identified in the analysis.
When the effects of such environmental processes cannot be directly measured or
modeled on the mixture of interest, there is potential for substantial error in the risk assessment.
The risk assessment can sometimes be modified by knowledge of the process that is generating
the mixture exposure, or by information on the original mixture chemicals along with the
geochemical and biochemical processes operating during their transport and over time. The
degree to which environmental fate alters the exposure or the dose response changes a basic
assumption of risk assessment of chemical mixtures, that of sufficient similarity. Under some
circumstances, sufficiency of similarity may be gauged by the gradient of costs (monetary or
environmental) of misjudging similarity, although such analyses will not be discussed here.
Whenever the mixture risk assessment is based on chemical component information and
the mixture composition cannot be fully identified, the uncertainty and possible bias in the
resulting risk assessment should be clearly described. Attention should also be given to the
persistence of the mixture in the environment as well as to the variability of the mixture
composition over time or from different sources of emissions. The assessment should also
discuss methods for improving the assessment, including gathering of more data as well as
employing other measurement or extrapolation techniques.
2.4.1.1. Transport Through an Environmental Compartment
Transport of a chemical mixture through the environmental compartments of air, soil, and
water will depend upon the physical and chemical properties of the individual components or the
unique properties of the chemical mixture (e.g., nonaqueous-phase liquids [NAPLs]) and the
environmental medium. There are a number of examples of changes in composition or quantity
of a chemical mixture as a result of environmental fate. The changes in the quantities and
concentrations of chemical disinfectant by-products (occuring in chemically disinfected drinking
water over time) during transport through the drinking water distribution system provide an
example of the changes that can occur to a mixture during transport through an environmental
compartment.
2.4.1.2. Intercompartmental Transfer Between Environmental Compartments
All components of a chemical mixture may not be transferred between environmental
compartments at the same rate. Once released to the environment, a mixture of chemicals may
be partitioned on the basis of the physical/chemical properties of each component of the mixture
and the condition of the microenvironment into which the components are partitioned.
-18-
-------
Selective movement of components can occur primarily during transport between soil,
air, or water environments. For example, volatilization from the soil surface compartment to the
atmospheric compartment could be important initially for the more volatile compounds in the
mixture. Volatilization from dry soil surfaces is dependent on both the vapor pressure (more
volatile compounds will volatilize more readily) and the ability of a compound to adsorb to soil.
Volatilization from moist soil surfaces is driven by the Henry's Law constant at steady state
(volatilization increases with a larger Henry's Law constant) and, as with dry soil surfaces, the
ability of a compound to adsorb to the soil. Because the Henry's Law constant is defined as the
ratio of a compound in air to that in water, compounds with either a high vapor pressure or
compounds that have a low vapor pressure together with a low water solubility may volatilize
from both moist soil and water surfaces. The rate at which a compound can volatilize from the
soil surface may be attenuated if that compound is also able to adsorb strongly to soil particles.
Compounds that adsorb strongly to the soil may also be physically entrained in the air as dust or
moved to aquatic environments via sediment runoff. Compounds that do not adsorb strongly to
the soil may leach readily through the soil column to groundwater systems if processes such as
volatilization and biodegradation do not occur rapidly enough. (There are exceptions, such as
where some vapor-phase pollutants in stack emissions adsorb to particulates.) The extent of soil
adsorption is generally based on the organic content of the soil, although some compounds (those
with a positive charge) can also adsorb to clays. A soil adsorption coefficient is defined in terms
of the soil organic carbon and can be used to estimate the ability of a particular compound to
leach into the soil column. The more volatile components of a chemical mixture in soil may
volatilize over a several-year period and no longer be present. A risk assessment based only on
the original mixture composition could then overestimate the long-term risk if the volatile
chemicals were the primary toxicants. Adjustments based on other factors such as exponential
decay models calibrated for the soil composition being assessed might improve the risk estimate.
The analyst should also consider differential transfer of chemicals comprising a mixture
between abiotic and biotic compartments and between two different biotic compartments. For
example, certain dioxin congeners released from the stacks of combustion sources appear to be
selectively taken up and retained in plant tissues (Lorberetal., 1996; 1998). The relative
proportions of dioxin congeners in the mixture to which humans and grazing animals are
exposed through the consumption of these contaminated plants vary considerably from the
original congener mixture released to the environment. The proportions of dioxin congeners in
human exposures that result from consumption of the tissues of the grazing animals (e.g., beef
cattle) will differ from the proportions released from the stack as well as those in the
contaminated plants.
-19-
-------
2.4.1.3. Transformation of a Chemical Mixture or Individual Compound Into Degradation
Products
In the environment, chemical mixtures may arise or change as a result of transformation.
If the various compound/s are susceptible to degradation via photolysis, hydrolysis, or
biodegradation (both aerobic and anaerobic), both alteration of the profile of the original
compounds in the mixture and changes in the quantity of the mixture present are possible. The
processes acting to change the profile of a mixture may be affected by the point of release of the
mixture (i.e., the profile from a mixture directly released to a lake may be different from that
from the same mixture following long-range atmospheric transport). Transformation reactions
that may differentially affect mixtures components in air, soil, and water are presented below,
followed by an example using the transformation of toxaphene.
• Atmosphere: Compounds can be transformed by direct photolysis, if the
compound is able to absorb light in the visible region of the spectrum, and/or by
reaction with reactive photochemically generated hydroxyl radicals, nitrate
radicals, and ozone (Atkinson, 1994). Reaction with hydroxyl radicals is expected
to be the major degradation process in the troposphere for most molecules, and the
rate of this reaction depends primarily on the chemical structure (Atkinson, 1994).
Unsaturated compounds also are expected to react quickly with nitrate radicals
and ozone.
Soil: Compounds can be transformed through aerobic and anaerobic
biodegradation at the soil surface. Aerobic biodegradation is controlled by
concentrations of oxygen and nutrients; compounds susceptible to anaerobic
biodegradation may be transformed in anaerobic microsites, which may be found
within the soil column and when the soil is flooded.
Water: Susceptible compounds may be transformed through hydrolysis (e.g.,
structures such as amides, alkyl halides, carbamates, and phosphoric acid esters
[Lyman et al., 1990] are particularly vulnerable), direct photolysis at the water
surface, and aerobic biodegradation.
The assessment of environmentally degraded or "weathered" toxaphene, previously the
most heavily used pesticide in the United States, exemplifies the concerns of transformation as
well as other environmental fate processes when developing a chemical mixtures risk
assessment. Toxaphene is a formulation of multiple ingredients. The relative amounts of these
components and their character change after toxaphene is released to the environment and the
original components of the mixture are exposed to differential partitioning and transformation
processes in air, water, and soil environments (U.S. EPA, 1997b).
-20-
-------
Toxaphene congeners are generally biologically degraded under anaerobic
conditions through reductive dechlorination. Anaerobic degradation rates in soils
and sediments are expected to be determined largely by qualities of the original
component molecules and the environment's potential to interact and change the
molecules' structure (Fingerling et al., 1996; Smith and Willis, 1978). The
stability of reaction products, whether in soil or sediment, seems to depend on the
position of the various chlorine atoms.
• Under aerobic conditions toxaphene degrades slowly, if at all (Parr and Smith,
1976; Bidleman et al., 1981; Mirsatari et al., 1987; Nash and Woolson, 1967).
• In general, the lower chlorinated toxaphene congeners are more easily vaporized
than are the higher chlorinated congeners (Seiber et al. [1979] showed soil
surface enrichment of the less volatile, more chlorinated compounds through GC
analysis); however, both can be atmospherically transported.
• Toxaphene, particularity the more volatile components, may be transported far
from the initial source by long-range atmospheric transport processes.
• Once deposited in water, the higher chlorinated congeners can bioaccumulate in
the food chain because of their lipophilicity.
The composition of "weathered" toxaphene samples may be different, depending on the
environmental processes to which the original agent was exposed. For example, toxaphene
extracted from an anaerobic soil does not resemble that from an aerobic soil, and toxaphene
detected in an air sample from the Arctic does not resemble the toxaphene residue obtained from
the blubber of an Arctic seal. Site-specific consideration of the partitioning and transformation
processes is needed for different environments. The resulting changes in chemical composition
of the original mixture over time will affect the toxicity of the mixture.
For another example, when the primary change to a mixture is believed to be the degree
of halogenation or other substitution, some adjustment of the estimated exposure or toxic
potency may be possible. One example (discussed in Section 3.4) concerns combinations of
PCBs, for which EPA has developed specific methodology to alter the toxic potency on the basis
of site-specific environmental factors.
2.4.2. Importance of the Exposure Sequence for Multiple Chemicals
The order in which chemical exposures occur and the time between exposures to different
chemical agents may affect the nature of the response to the chemical insult. For example, the
sequence or pattern of exposure is important for compounds that have been described as initiators
and those described as promoters of carcinogenicity. There is evidence to suggest that exposure
to certain compounds results in an irreversible change in the affected cells and progeny (the cell
is said to be initiated). When the initial exposure is followed by repeated doses of a second
chemical agent (i.e., the promoter), tumors occur. In the absence of either the initiator or the
-21-
-------
promoter, or if the order is reversed, tumors do not occur. An example of an initiator-promoter
sequence is the application of a PAH (initiator) (e.g., benzo[b]fluoranthene) followed by
repeated applications of 12-o-tetradecanoyl phorbol-13-acetate (TPA) to the skin of shaved mice
(Aminetal., 1985).
2.4.3. Routes of Exposure
In environmental health risk assessments, analysts typically consider three routes of
human exposure: oral, dermal, and inhalation. Differences in the properties of the cells that line
the surfaces of the gastrointestinal tract, the skin, and the air pathways and lungs may result in
different intake patterns of chemical mixture components depending on the route of exposure.
Additionally, chemicals in a mixture may partition to contact media differently, resulting in
different potential routes of exposure (see Section 2.4.1). In chemical mixtures risk assessment,
the issue becomes how and when to combine routes. EPA is still developing approaches for this.
EPA (1998c) recommends that route-to-route conversion should be attempted only for dermal
exposures at this time. Adequate inhalation-to-oral conversion methods for steady-state
conditions have not yet been developed. A general outline of the oral-to-inhalation extrapolation
process and a discussion of route-to-route extrapolation issues can be found in Gerrity and Henry
(1990) and in EPA's Reference Concentration methodology document (U.S. EPA, 1994a). Until
such methodology is developed, inhalation and oral risk characterization should be carried out
separately. The assessor should note, however, that total risk from the mixture could be
underestimated by not combining all routes of exposure, because the total exposure is not
characterized and the chemical interactions may not be considered.
Multiple-route exposures can be combined in two different ways: summing the absorbed
daily doses or summing the (external) oral equivalent daily doses. Both approaches require an
estimate for the oral absorption fraction, but the latter is adopted here as it is simpler for
consideration with standard toxicity comparison values based on ingestion (e.g., RfD).
A number of factors might contribute to differences in toxicologic effectiveness between
oral and dermal exposures at equal dosages. The most obvious relates to differences in
absorption rates between the two routes. Other potential contributing factors include differing
sensitivity of absorption sites to damage and differences in toxicokinetics (i.e., distribution,
metabolism, elimination) between exposure routes. Ideally, the conversion from dermal to
equivalent oral dose would be based on experimentally derived values that characterize the
relationship between the doses that produce a particular toxicity by each of the different routes.
In practice, however, the conversion usually will be based on absorption factors because of a
general absence of appropriate data.
-22-
-------
2.4.4. Exposure Assessment Summary
This section summarizes a few important concepts related to chemical mixtures exposure
assessment. Once a chemical mixture is released to the environment, its concentration and
composition may change through partitioning into abiotic and biotic compartments and through
transformation mediated by the environment and biota. The physical/chemical properties of each
component of the mixture (or the properties of the mixture as a whole) and the condition of the
microenvironment into which the components are partitioned may change the magnitude and the
routes of human exposure. Partitioning and transformation of the mixture components will affect
the routes of exposure. Ideally, chemical mixture exposures through different routes can be
integrated through measurement data or a validated physiologically based pharmacokinetic
(PBPK) model; at this time, approaches are still evolving, particularly for combining inhalation
and oral exposures. The sequence of exposures to different chemical agents is clearly important
for some responses. A number of other issues will be deferred for later discussion by the
Agency; these include chemical mixtures with intrinsically unique properties (e.g., NAPLs), mass
balance within chemical mixtures assessments, assessing risk of unidentified components of
chemical mixtures, measurement issues, and component bioavailability.
2.5. DATA AVAILABLE ON WHOLE MIXTURES
Whenever possible, the preferred approach to the health risk evaluation of chemical
mixtures is to perform the assessment using health effects and exposure data on the whole
mixture. Such data include human epidemiologic, clinical, or occupational studies; animal
studies on the complex mixture; or in vitro data on the complex mixture. Figure 2-1 shows that
the whole-mixtures data can then be divided into subsets of data directly on the mixture of
concern, data on a sufficiently similar mixture, or data on a group of similar mixtures. This
guidance document discusses these situations and offers some examples of how to approach a
whole-mixture health risk assessment.
2.5.1. Data Available on the Mixture of Concern
Exposure and toxicity data directly on the mixture of concern are most likely to be
available for highly complex mixtures, such as coke oven emissions, which are generated in large
quantities and associated with or suspected of causing adverse health effects. The evaluation of
such a mixture requires scientific judgment regarding the stability of the mixture in the
environment and the linkage of the observed human health effect to the mixture exposure.
Toxicity data obtained from concentrates or extracts of the original mixture of concern may not
be predictive of human toxicity to the original mixture. Such data are more properly handled
using procedures developed for lexicologically similar mixtures (Sections 2.5.3, 3.3).
-23-
-------
2.5.1.1. User Fact Sheet: Mixture of Concern RfD/C or
Slope Factor
The user of this guidance document can use
Figure 2-1 to determine if data are available directly on the
mixture of concern. Then a procedure is suggested for
estimating either a cancer slope factor or a reference
dose/concentration (RfD/C), as encapsulated in the
following user-information fact sheet.
Approach:
Type of Assessment
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Mixture of Concern RfD/C or Slope
Factor
Dose-Response Toxicity Value
3.1,3.2
Examples can be found on IRIS
(U.S. EPA, 2000a).
Toxicity data are available on the
mixture of concern. Examples of
such data are human
epidemiologic data from an
occupational setting, human data
from a clinical study, or animal
toxicology data on the complex
mixture.
Estimate dose-response toxicity
value directly from data on
complex mixture of concern, using
the same procedures as those
used for single chemicals.
Calculations are simple.
Composition of the test mixture is
functionally the same as what is
found in the environment. Test
data are adequate to account for
all sensitive endpoints.
Data are rarely available.
Scientific judgments of the
chemical composition of the
mixture; toxicologic relevance of
the laboratory data to the
environmental mixture.
2.5.2. Data Available on a
Sufficiently Similar Mixture
If data are not available on
the mixture of concern, the risk
assessment may be based on data on
a sufficiently similar mixture. A
mixture is sufficiently similar to the
mixture of concern when its
components are not very different
and are contained in about the same
proportions as the mixture of
concern. In addition, if information
exists on differences in
environmental fate, uptake and
pharmacokinetics, bioavailability, or
toxicologic effects for either of these
mixtures or their components, it
should be considered in the
determination of sufficient similarity.
If such data are available, an attempt
should be made to determine if
significant and systematic differences
exist between the chemical mixtures.
If no significant differences are
noted, then a risk assessment may be
performed using data on the
sufficiently similar mixture as a
surrogate for the mixture of concern.
-24-
-------
2.5.2.1. User Fact Sheet: Sufficiently Similar Mixture
RfD/C or Slope Factor
The user of this guidance document can use Figure
2-1 to determine that the data available are on a mixture that
is sufficiently similar to the mixture of concern. Then a
procedure is suggested for estimating either a cancer slope
factor or a reference dose/concentration (RfD/C), as
encapsulated in the following user-information fact sheet.
Approach:
Type of Assessment:
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Sufficiently Similar Mixture RfD/C or
Slope Factor
Dose-Response Toxicity Value
3.1,3.2
New procedure.
Toxicity data are available on a
mixture that is judged as sufficiently
similar to the mixture of concern in
the environment. No data are
available on the mixture of concern.
Examples of such data are human
epidemiologic data from an
occupational setting, human data
from a clinical study, or animal
toxicology data on the complex
mixture.
Estimate dose-response toxicity
value using data on the sufficiently
similar mixture as a surrogate for
data on the mixture of concern,
using the same procedures as those
used for single chemicals.
Calculations are simple.
Composition of the sufficiently
similar mixture is functionally the
same as what is found in the
environment. Test data are
adequate to account for all sensitive
endpoints. Similarity judgment
across the mixtures must be made
and supported.
Availability of data is limited.
Scientific judgments of sufficient
similarity, chemical composition and
stability of the two mixtures;
toxicologic relevance of the
laboratory data to the environmental
mixture.
2.5.3. Data Available on a
Group of Similar Mixtures
In some cases, data are
available on a group of similar
mixtures that are known to be
generated by the same commercial
process or emissions source but
that vary slightly in composition
depending on factors such as time
since emission, environmental
transformation, or geographic
location of emission sources. Data
are then available on several
mixtures with approximately the
same components but with slightly
different component exposure
levels, so that the likely range of
compositional variation is
covered. Thus, risk assessors can
use toxicity and exposure data that
exist on the group of similar
mixtures and extrapolate in order
to perform a risk assessment on
the less well-studied or
environmentally transformed
mixtures that belong to that same
group.
-25-
-------
2.5.3.1. User Fact Sheet: Comparative Potency
The user of this guidance document can use
Figure 2-1 to determine that the data available are on a
group of similar mixtures. Then a procedure is suggested
for using a comparative potency approach to estimating a
cancer slope factor, as encapsulated in the following user-
information fact sheet.
Approach:
Type of Assessment:
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Comparative Potency
Dose-Response Toxicity Values
for Cancer, Genetic Toxicity
3.1,3.3
Used for combustion mixtures
(Lewtas, 1985, 1988; Nesnow,
1990).
Method requires short-term data
on several similar mixtures
including the mixture of concern,
and at least one data point from a
chronic in vivo study on one of
these mixtures. Examples of such
data are in vitro mutagenicity
assays and chronic rodent
bioassays.
Estimate dose-response value
using relationships across similar
mixtures and similar assays to
extrapolate to a value for the
mixture of concern.
Calculations involve some
statistical modeling and toxicologic
judgment. Method is data
intensive with short-term assay
data required.
Assumes the potency change for
similar mixtures across assays is
the same for all similar mixtures.
Test data are adequate to account
for all sensitive endpoints.
Similarity judgment across the
mixtures must be made and
supported.
Availability of data is limited.
Scientific judgments of sufficient
similarity relative to chemical
composition and toxicologic
activity of the mixtures.
-26-
-------
2.5.3.2. User Fact Sheet: Geographic Site-
Specific Assessments
The user of this guidance document can
follow Figure 2-1 to determine that the data available
are on a group of similar mixtures. Then a procedure
is suggested for estimating risk from exposure to the
mixture by using a geographic site-specific
assessment, as detailed in the following user-
information fact sheet.
Approach:
Type of Assessment:
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Geographic Site-Specific
Assessment
Risk Characterization for Any
Toxic Endpoint
3.1,3.4
Used for cancer assessment
ofPCBs(U.S. EPA, 1996c)
Method requires both toxicity
and exposure data on the
mixture's components.
Toxicity data on the
commercial mixture are used
to estimate a range of toxicity
values that are then adjusted
for alterations in the
mixture's composition
because of environmental
factors to produce a risk
estimate for the total mixture.
Complicated to use. Data
intensive.
Requires the user to make
assumptions about the fate
and transport of groups of
chemicals.
Some data restricted by
similarity. Restricted to
specific conditions. Limited
by data quality.
Scientific judgment of fate
and transport. Accuracy of
exposure data.
2.6. DATA AVAILABLE ON
MIXTURE COMPONENTS
If data are not available on an
identical or reasonably similar mixture,
the risk assessment may be based on the
toxic or carcinogenic properties of the
components in the mixture. When
quantitative information on toxicologic
interaction exists, even if only on
chemical pairs, it should be incorporated
into the component-based approach.
When there is no adequate interactions
information, dose- or risk-additive
models are recommended. The primary
criterion for choosing between dose
addition and response addition is the
toxicologic similarity among the
chemicals in the mixture. This decision
should be based on information about the
toxicologic and physiologic processes
involved, the single-chemical dose-
response relationships, and the type of
response data available. The risk
assessment using component data should
then begin with selection of the most
appropriate concept for the chemicals in
the mixture.
-27-
-------
2.6.1. Toxicologic Similarity and
Dose Addition
In the simplest terms,
chemicals can be considered as dose
additive if each chemical can be
thought of as a concentration or
dilution of every other chemical in
the mixture. The chemicals are
assumed to behave similarly in terms
of the primary physiologic processes
(uptake, metabolism, distribution,
elimination) as well as the
toxicologic processes. The
mathematical description of dose
addition requires a constant
proportionality between the
effectiveness of the two chemicals.
Three component methods that are
based on dose addition are discussed
in this document: the HI, the
Relative Potency Factor (RPF)
method, and the Toxicity
Equivalence Factor method, which is
a special case of the RPF method.
They differ in the required
knowledge about toxic mechanism
and in the extent over which
toxicologic similarity is assumed. In
each method, the exposure levels are
added after being multiplied by a
scaling factor that accounts for
differences in toxicologic potency.
2.6.1.1. User Fact Sheet: Hazard Index
The user of this guidance document can follow
Figure 2-1 to determine that the data available are on the
components of the mixture of concern and that there is
evidence of toxicologic similarity of the components. Then
a procedure is suggested for estimating a Hazard Index,
an indication of risk from exposure to the mixture, as
encapsulated in the following user-information fact sheet.
Approach:
Type of Assessment:
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Hazard Index
Risk Characterization for Any
Toxic Endpoint
4.1,4.2
Used in Superfund site
assessments (U.S. EPA, 1989a).
Method requires both toxicity and
exposure data on the mixture's
components. Good dose-
response data are needed, such
as what is available on IRIS (U.S.
EPA, 2000a).
Scale individual component
exposure concentrations by a
measure of relative potency
(typically, divide by a Reference
Dose/Concentration [RfD/C]) for
components with a similar
mechanism-of-action. Add scaled
concentrations to get an indicator
of risk from exposure to the
mixture of concern.
Easy to calculate.
Applies dose addition, which
carries with it assumptions of
same mode of action and similarly
shaped dose-response curves
across the components. The
"common mode-of-action"
assumption can be met by using a
surrogate of same target organ.
Exposure data should be at
relatively low levels (near no-
adverse-effect levels) at which
interaction effects are not
expected. RfD/C values across
components vary in their
uncertainty, so other measures of
potency may be more
appropriate.
Similarity of mechanism-of-action.
Accuracy of exposure data.
-28-
-------
2.6.1.2. User Fact Sheet: Relative Potency Factors
The user of this guidance document can follow
Figure 2-1 to determine that the data available are on the
components of the mixture of concern and that there is
evidence of toxicologic similarity of the components. Then
a procedure is suggested for estimating risk from exposure
to the mixture by using Relative Potency Factors, as
encapsulated in the following user-information fact sheet.
Approach:
Type of Assessment:
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Relative Potency Factors
Dose-Response Assessment for
Any Toxic Endpoint
4.1,4.4
New Procedure
Method requires both toxicity and
exposure data on the mixture's
components. Toxicity data are
missing for some components.
Scale component exposure
concentrations relative to potency
of an index chemical (typically the
best-studied component) following
expert committee consensus.
Add scaled concentrations. Use
dose-response curve of index
chemical to generate response
estimate for sum of scaled
concentrations.
Complicated to use. Requires
some statistical modeling and
judgment of relative potency
factors.
Based on dose addition which
carries with it assumptions of
same mode of action and similarly
shaped dose-response curves
across the components. The
"common mode-of-action"
assumption can be met using a
surrogate of toxicologic similarity,
but for specific conditions
(endpoint, route, duration).
Limited by data quality and
similarity. May not have data
from all routes of exposure of
interest. Same mode-of-action
across components may not be
known.
Judgment of relative potency
factors. Similarity of toxicologic
action. Missing data on some
components.
-29-
2.6.2. Independence and
Response Addition
Response addition may apply
when components act on different
systems or produce effects that do
not influence each other. Under
response addition, the chemicals in
the mixture are assumed to behave
independently of one another, so that
the body's response to the first
chemical is the same whether or not
the second chemical is present.
Mathematically, response addition
can be described by the statistical
law of independent events, with
"response" measured by the
percentage of exposed animals that
show toxicity or the proportion of
the population responding.
Response addition is particularly
useful when the effects of concern
are thought to be present at low dose
levels for each of the component
chemicals, even though it is highly
unlikely the effects are capable of
being observed at these low levels in
the environment. When interaction
data are available on any of the
components in the mixture, the risk
assessor may provide a qualitative
discussion of the likely effect of
these data on the outcome of the
mixture risk assessment under
response addition (see Sections
2.2.4,4.5.4).
-------
2.6.2.1. User Fact Sheet: Response Addition
The user of this guidance document can follow
Figure 2-1 to determine that the data available are on the
components of the mixture of concern and that there is
evidence of toxicologic independence of action. Then a
procedure is suggested for estimating risk from exposure
to the mixture by using Response Addition, as
encapsulated in the following user information fact sheet.
Approach:
Type of Assessment:
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Response Addition
Risk Characterization for Any
Toxic Endpoint
4.1,4.5
Used extensively for cancer.
Used in Superfund site
assessments (U.S. EPA, 1989a).
Method requires both toxicity data
(measured in percent responding)
and exposure data on the
mixture's components. Good
dose-response data are needed,
such as what is available on IRIS
(U.S. EPA, 2000a).
Risk of an effect is estimated for
each component using its dose-
response curve at the
component's exposure
concentration. Component risks
are added, using the
independence formula, to yield a
risk estimate for the total mixture
for the specific exposure.
Easy to calculate.
Assumes toxicologic
independence of action.
Assumes interactions are not
significant at low exposures.
Limited to low exposure
concentrations. Slight
overestimate of mixture's upper
bound on risk when adding
individual component upper
bound estimates. Restricted to
independence of action.
Independence of action.
Accuracy of exposure data.
Individual risk estimates may vary
in quality.
2.6.3. Interactions Data
Toxicologic interactions are
operationally defined by the
existence of data showing significant
deviations from a "no interaction"
prediction; that is, the response is
different from what would be
expected under an assumption of
additivity (e.g., dose-additive,
response-additive). Types of
interactions among mixture
components that can affect
toxicologic response to the whole
mixture include chemical-to-
chemical, toxicokinetic, and
toxicodynamic interactions (see
Table B-2 and Appendix C). The
impact of these constituent
interactions on toxicologic response
can be less than additive (e.g.,
antagonistic) or greater than additive
(e.g., synergistic). The component-
based method discussed in this
document that incorporates
interactions information is the
interaction-based HI.
-30-
-------
2.6.3.1. User Fact Sheet: Interaction-Based Hazard
Index
The user of this guidance document can follow
Figure 2-1 to determine that the data available are on the
components of the mixture of concern and that interactions
data are available. Then a procedure is suggested for
estimating risk from exposure to the mixture by
incorporating information on binary combinations of the
components using an interaction-based hazard index, as
encapsulated in the following user information fact sheet.
Approach:
Type of Assessment:
Section(s):
References:
Data Requirements:
Strategy of Method:
Ease of Use:
Assumptions:
Limitations:
Uncertainties:
Interaction-Based Hazard Index
Risk Characterization for Any
Toxic Endpoint
4.1,4.3
New procedure (Hertzberg et al.,
1999).
Method requires both toxicity and
exposure data on the mixture's
components, and interactions
data on at least one pair of
components.
Scale component exposure
concentrations by a measure of
relative potency (typically, divide
by a reference dose/concentration
[RfD/C]) for components with a
similar mechanism-of-action.
Modify this term with data on
binary interactions. Add
scaled/modified concentrations to
provide an indicator of risk from
exposure to the mixture of
concern.
Complicated to use.
Assumes binary interactions are
the most important. Assumes
interaction magnitude is not dose
dependent, but depends on
component proportions.
Limited interactions data are
available. Model with relative
proportions is untested.
Interaction magnitude is often a
default because of lack of
measurement data.
Binary interactions used to
represent the interactions for the
whole mixture. Accuracy of
exposure data. Accuracy of
default for interaction magnitude.
2.7. FUTURE DIRECTIONS
2.7.1. Overview
Risk assessment methods for
chemical mixtures are progressing
along paths similar to risk
assessment for single chemicals, by
incorporating more knowledge of
specific modes of toxicologic action
of the chemicals and by greater use
of statistical methods and
mathematical models. Where the
field differs, however, is in the more
extensive use of quantitative
inference from tested chemicals to
untested chemicals. Mixture
exposures can be extremely varied,
with differences in total dose,
composition, and relative
proportions. Consequently, only a
small fraction of environmental
mixtures can actually be tested for
dose-response characteristics. Two
options then seem feasible: directly
investigating a few high-priority
mixtures, and, for the remainder,
developing extrapolation methods
for using available data on the
mixture components or on similar
mixtures.
The first option requires
priority setting, which for mixtures
is its own research area (Cassee et
al., 1998). The characteristics to
include in a mixture prioritization
scheme should parallel those often
cited for single chemicals: target
-31-
-------
those mixtures posing the highest public health risk. The supporting data could include annual
emissions of mixtures, frequency of occurrence of mixtures in the environment, identity of
mixtures containing highly toxic chemicals, or documented health problems in populations
exposed to identified mixtures. Because most interaction data are on chemical pairs, one
approach would include the frequency of occurrence of chemical pairs in the media associated
with the exposure scenario to be regulated. The prioritization should also consider the
availability of interaction data. For high-priority mixtures lacking such data, other assessment
methods may be needed. The various regulatory program areas, such as Superfund waste sites,
ambient air, and drinking water, pose substantially different kinds of mixtures and exposure
conditions, so that a priority list for one program may not be appropriate for a different
regulatory program.
Once a few mixtures posing the highest concern have been identified, researchers should
seek to evaluate their exposure, toxicity, and risk characteristics. Because even the highest
priority mixtures are likely to pose complex and varied exposure possibilities, much of the
research effort should involve developing highly efficient experimental designs, short-term
toxicity assays, and uncertainty methods so that several scenarios can be characterized for each
mixture.
The second option, for addressing all the remaining mixtures, is to develop methods that
can extrapolate exposure and toxicity estimates from available data to the scenario of concern. In
addition to the issues being addressed by extrapolation methods for single chemicals (e.g., cross-
species, cross-route), mixtures issues also include interactions and changes in composition.
Interactions issues include the commonly observed toxicologic interactions that influence
pharmacokinetics, as well as the less-studied areas of physiological interactions between affected
tissues or organs, and the biochemical and physical interactions affecting degradation and
transport of mixtures in environmental media. Because of the wide variety of mixture exposures,
all relevant information should be tapped to improve the understanding of the basic biological
and chemical processes. For example, to improve dose-response extrapolation, toxicology
experiments, epidemiology and occupational studies, and mathematical model development
should be pursued simultaneously.
Mixtures research should be efficient. The complexity of the issues is beyond the reach
of any single agency. Sharing of resources and information within different sectors of EPA as
well as with other agencies is essential. Several such efforts are underway. The Integral Search
System (Arcos et al., 1988) and the Mixtox database (Marnicio et al., 1991) are two EPA
collections of bibliographic summaries of interaction studies that are available to the public.
Additional databases should be developed, perhaps jointly with the public, on mechanisms and
modes of toxicologic interaction and on mathematical models of biological processes influencing
-32-
-------
the interactions. The National Institute for Occupational Safety and Health (NIOSH) has a
Mixed Exposures research program whose advisory committee includes representatives from
EPA, other federal agencies, and research institutions. EPA, NIOSH, and the Agency for Toxic
Substances Disease Registry (ATSDR) have organized the Mixed Exposures Research Group
(MERG), composed of almost 20 federal and state agencies, to share regulatory approaches.
MERG seeks to facilitate interagency communication and jointly sponsored research projects on
mixtures. Additional cooperative efforts should be pursued with the public and foreign agencies.
Mixture risk assessment methods should ideally be developed in conjunction with those
laboratory and field studies that are needed for implementation as well as validation. Otherwise,
the methods become conceptual models that cannot feasibly be applied, or decision tools whose
accuracy cannot be tested. One example concerns interaction studies, such as those detailed in
the EPA's Mixtox database (Marnicio et al., 1991; U.S. EPA, 1990) of in vivo toxicologic
interaction studies. In the Mixtox database, 95% of the studies involve only pairs of chemicals
(Teuschler and Hertzberg, 1995). Consequently, the interaction-based Hazard Index (Section
4.3) was developed for pairwise interactions to allow use of available data. Interaction studies
are in progress by research groups in EPA's National Center for Environmental Assessment
(NCEA) and National Health and Environmental Effects Research Laboratory (NHEERL) to
provide the toxicity data and data analysis methods for validation of the index.
The information required for evaluation of the extrapolation methods in this document is
generally not yet available. The number of pairs studied for interactions is a small fraction of the
number of possible chemical combinations, and the number of whole mixtures studied is far
smaller yet. For example, with a simple mixture of only 20 chemicals, there are 190 pairs, but
over a million possible combinations (pairs, triples, etc.). Because of this sparseness of existing
data, both on whole mixtures and on interactions, the accuracy of these extrapolation methods
will be difficult to judge. The inferential procedures for mixture risk discussed in this document
are then likely to be adopted based on scientific plausibility and on relatively few validation
studies. The validation process is valuable, even when incomplete. As was found with the
analysis of the consistency of pairwise interactions (Durkin et al., 1995), the evaluation of the
mixture risk tools will likely spawn research questions that lead to new statistical, exposure, and
toxicologic studies, and subsequently to better risk tools.
2.7.2. Research Suggestions for Improving Mixture Risk Assessment
Several research directions have been suggested during the development of this guidance
document. Although specific projects have been identified related to dose-response assessment,
the highest priority was the preparation of guidance on exposure assessment of mixtures. Some
of the key concerns with exposure assessment are discussed in this document (Section 2.4). The
-33-
-------
need is for specific procedures for measurement and modeling of exposures for various
scenarios, along with the corresponding methods for characterizing the uncertainties. The Risk
Assessment Forum created an advisory panel in 1999 to decide the scope and project
requirements for a framework for cumulative risk assessment. A major component of that
framework is the exposure assessment of mixtures. Some specific areas for exposure assessment
that have been suggested during review of this guidance are given in the list below.
Among the next highest priorities was research aimed at the evaluation and improvement
of the dose-response methods in this guidance document. In particular, the comparative potency
method for whole mixtures and the interaction-based Hazard Index need to be demonstrated with
different kinds of mixtures. Methods for validation of these two methods also need to be
developed, followed by the validation exercise itself for several different mixtures.
The most often mentioned research area was uncertainty analysis. Each of the methods in
this guidance document produces a single risk estimate. An initial goal is to present that risk
estimate as a plausible range in addition to the single recommended value. A related goal is to
present a range of risk estimates that reflects all the risk methods applied to the mixture of
concern, i.e., the uncertainty in model selection. Data uncertainties should also be addressed, at
least by sensitivity analysis. Subsequent efforts should pursue more complete uncertainty
characterization, including methods for choosing the default distributions for the parameters and
variables in each method. Uncertainty characterization is also one of the components of the
Forum's cumulative risk framework project, so further work will commence in this area over the
next few years.
The other main research needs raised during the authoring and review of this guidance
document covered a wide range of scientific areas. The most commonly discussed topics are in
the following list. The research areas are roughly grouped by scientific discipline or application.
Exposure assessment
• data and models for degradation over several years (e.g., pathogens in
groundwater, pesticide mixtures in soil).
• models/data for chemical and biological interactions influencing mixture
transport.
• mixture changes (chemical composition, relative proportions) from facility
failures (e.g., drinking water, municipal combustors).
• procedures for artificial degradation or weathering of complex mixtures.
procedures for monitoring mixtures when there are hot spots with each spot
having a different driver chemical.
biomarkers of exposure that are specific to single chemicals or chemical classes
and mathematical models that relate the biomarker to existing or prior external
exposure levels, and to tissue levels and/or tissue-specific toxic effects.
-34-
-------
Statistical/mathematical methods
formulas for incorporating independence when adding upper-bound risks (n > 3).
concepts and methods for tolerance distributions for n > 2 chemicals.
uncertainty analysis, i.e., Bayesian, Monte Carlo simulation for each of the
mixture risk assessment procedures.
efficient and stable numerical methods for modeling highly complex interacting
systems (hundreds of chemicals, multiple tissues, time-variable exposures).
statistical graphics methods for demonstrating and displaying interactions in
multichemical mixtures (n > 5).
Biomathematical models
models for describing the dependence of interaction magnitude on total dose and
on component fractions.
biologically based models that separate out the relative differences of chemicals in
terms of pharmacokinetics and pharmacodynamics.
models that incorporate aging and growth, and more physiological processes and
factors than just flows to major organs and tissues.
models for initiation-promotion interactions that include background exposures to
initiators or promoters.
Human studies
Toxicology
database of epidemiology studies with exposure-response information on
mixtures.
database of occupational health studies with exposure-response information on
mixtures.
methods for estimating interaction magnitudes in epidemiology studies that relate
to (are consistent with) physiologic measures of interaction magnitude.
information on background exposure levels, background prevalence of health
conditions, and those population characteristics that indicate increased
susceptibility to toxic chemicals, including models that quantify the influence of
population characteristics on toxicology.
modes and mechanisms of interaction for carcinogens.
data describing the dependence of interaction magnitude on total mixture dose and
on component fractions.
concordance across animal species of specific toxic effects, modes of action, and
modes of interaction.
data and modes of interaction for inhibition (one chemical is nontoxic).
data and concepts for particulate interactions with other airborne chemicals.
-35-
-------
more examples and methods for short-term whole-mixture toxicity testing,
particularly data showing the representativeness of in vitro studies to represent in
vivo toxicity.
relationships between mode of toxic action and mode of interaction.
concepts, mechanisms or modes of action, or toxicity data to explain the
mathematical interaction models of proportional response addition and straight-
line isoboles that are not parallel.
interaction studies on major chemical classes to establish empirical interaction
classes based on interaction patterns.
test procedures that mimic real-world exposures (e.g., species-adjusted
intermittent exposures to correspond to occupational exposure patterns)
biomarkers of toxicity that are specific to single (or related) toxic effects and
mathematical models that relate the biomarker to actual measurable toxic
endpoints.
Risk methods
development of screening assays for mixtures to identify combinations of
chemicals that are most toxic or that potentially interact.
risk estimation for a mixture of mixed types, including similar, independent, and
interacting chemicals with same target organ, e.g., for classes with similar (RPF)
chemicals and other chemicals.
risk estimates or qualitative risk indicators for unidentified chemicals in a mixture
(see U.S. EPA, 1998d. Comparative risk framework methodology and case study.
SAB external review draft. NCEA-C-0135).
MOE methods for carcinogens using response addition.
RPFs from dose-response data on all chemicals, as improvement over HI because
it allows actual estimate of toxicity from the index chemical's dose-response
curve.
use of interaction patterns for estimating interaction direction in a chemical class.
methods for prioritizing chemical pairs (air, drinking water) for further study on
the basis of health risk.
methods for prioritizing complex mixtures for further study on the basis of health
risk.
methods for prioritizing complex mixtures for further study on the basis of
degradation potential.
-36-
-------
3. METHODS FOR WHOLE-MIXTURES DATA
3.1. INTRODUCTION
If whole-mixture data are available, then one approach to the health risk evaluation of a
chemical mixture is to perform a risk assessment using health effect, dose response, and exposure
data on the complex mixture. Health effect and dose-response data include human
epidemiologic, clinical, or occupational studies; animal studies on the complex mixture; or in
vitro data on the complex mixture. Exposure data include both environmental measurements and
human activity patterns that take into account environmental fate, temporal patterns of exposure,
and routes of exposure. The evaluation of whole mixtures in this document is subdivided into
categories depending on data availability: data directly on the mixture of concern, data on a
sufficiently similar mixture, or data on a group of similar mixtures.
3.1.1. Data Available on the Mixture of Concern
For predicting the effects of subchronic or chronic exposure to mixtures, the preferred
approach is to use subchronic or chronic health effect, dose-response, or exposure data on the
mixture of concern and adopt procedures similar to those used for single compounds, either
systemic toxicants or carcinogens (see U.S. EPA, 1987, 1989a, 1996a,d). Exposure and toxicity
data on the mixture of concern are most likely to be available on highly complex mixtures such
as coke oven emissions, which are generated in large quantities and associated with or suspected
of causing adverse health effects. Issues that need to be considered in order to justify performing
a risk assessment directly on the mixture of concern include bioavailability to humans of the
mixture in the environment, stability or variability of the mixture composition over time,
consistency of the mixture composition relative to its source, and potential differences between
the mixture tested in the laboratory and the mixture found in the environment. These factors
should be taken into account or the confidence in and applicability of the risk assessment is
diminished.
3.1.2. Data Available on a Sufficiently Similar Mixture
If adequate data are not available on the mixture of concern, but health effects data are
available on a similar mixture, a decision should be made whether the mixture on which health
effects data are available is or is not "sufficiently similar" to the mixture of concern to permit a
risk assessment. The determination of "sufficient similarity" should be made on a case-by-case
basis, considering not only the uncertainties associated with using data on a surrogate mixture,
but also contrasting the inherent uncertainties if one were to use other approaches, such as
component-based methods.
-37-
-------
In determining whether a mixture is sufficiently similar, consideration should be given to
any available information on the components that differ or are contained in markedly different
proportions from the mixture of concern. In addition, if information exists on differences in
environmental fate, uptake and pharmacokinetics, bioavailability, or toxicologic effects for either
of these mixtures or their components, it should be considered in deciding on a risk assessment
approach. If such information is not available, it should be identified as a source of uncertainty.
If toxicity data for the sufficiently similar mixture are only available for a different exposure
route than the environmental route being addressed, extreme care should be used to ensure that
the results are applicable, and that any effects restricted to the portal of entry to the body are
appropriately discounted.
3.1.3. Data Available on a Group of Similar Mixtures
In some cases, data are available on a group of similar mixtures that are known to be
generated by the same commercial process or emissions source, but that vary slightly in
composition, depending on factors such as time since emission, environmental transformation, or
geographic location of emission sources. Data are then available on several mixtures with the
same components but with different component exposure levels, so that the likely range of
compositional variation is covered. If such data are available, an attempt should be made to
determine if significant and systematic differences exist among the chemical mixtures. If
significant differences are noted, ranges of risk can be estimated based on the environmental fate
data, chemical structures, and toxicologic data of the various mixtures (Section 3.4). If no
significant differences are noted, then a risk estimate can be made by extrapolating across these
similar mixtures by comparing toxicity across various assays (Section 3.3).
A group of mixtures may be considered similar if they have the same components but in
slightly different ratios or have several common components but a little fewer or additional
components. This judgment can be based on empirical measurements or on indirect evidence.
The risk assessor should be able to support the assumption of toxicologic similarity and can do so
by using any of a number of approaches: (1) establishing that a common mode of action exists
across the mixtures or their components; (2) showing consistency in results of short-term
screening assays; (3) distinguishing chemical class or chemical structure similarity; (4)
identifying common components across the mixtures in similar proportions; and (5) establishing
a common source of formation or emission for the group of mixtures.
-38-
-------
3.1.4. Environmental Transformations for Whole Mixtures
A mixture's composition can change over time in the environment and thus become an
issue for the development of a whole-mixture risk assessment. The impact of this phenomenon
is that the exposure assessment will not fully characterize the mixture in terms of its chemical
components, often because of suspected changes over time in the mixture composition or
because of incomplete identification of the individual chemical components (see Section 2.4 on
exposure issues).
Whenever the mixture composition is affected by environmental factors, the uncertainty
and possible bias in the resulting risk assessment should be clearly described. Attention should
also be given to the persistence of the mixture in the environment as well as to the variability of
the mixture composition over time or from different sources of emissions. The assessment
should also discuss methods for improving the assessment, including gathering of more data as
well as employing other measurement or extrapolation techniques.
3.1.5. Uncertainties With Whole-Mixture Studies
Even if a risk assessment can be made using whole-mixture data, it may be desirable to
also conduct a risk assessment based on toxicity data on the components in the mixture using
procedures outlined in Chapter 4. When a mixture contains component chemicals whose critical
effects are of major concern, e.g., cancer or developmental toxicity, an approach based on the
mixture data alone may not be sufficiently protective in all cases. For example, the whole-
mixture approach for a two-chemical mixture of one carcinogen and one toxicant would use
toxicity data on the mixture of the two compounds. However, in a chronic study of such a
mixture, the presence of the toxicant could mask the activity of the carcinogen. That is to say, at
doses of the mixture sufficient to induce a carcinogenic effect, the toxicant could induce
mortality so that at the maximum tolerated dose of the mixture, no carcinogenic effect could be
observed. Since carcinogenicity is generally considered by the Agency to be an effect of concern
even at extremely low doses, it may not be prudent to conclude that the lack of a carcinogenic
effect from such a bioassay indicates the absence of cancer risk at lower doses. (The type of
carcinogenic effect is also a factor here; for example, low doses of a promoter are generally less
of a concern than of a genotoxic carcinogen.) Consequently, the mixture approach should be
modified to allow the risk assessor to evaluate the potential for masking, of one effect by another,
on a case-by-case basis.
For most noncarcinogenic effects, reduced exposure levels lead to reduced severity of the
effects. Carcinogenic effects have traditionally been assumed by EPA to be potentially fatal, so
that reducing the exposure only lowers the expected response rate; the effect severity remains
high. Environmental exposures, even at lower levels than those in the study, to a mixture with a
-39-
-------
known carcinogenic component then may pose a cancer risk in spite of negative results from a
whole-mixture study. Another example is a whole-mixture assay that did not show
developmental effects. Any developmental toxicity is considered an effect of major concern. If a
component chemical is a known developmental toxicant, then the whole-mixture data should be
carefully reviewed for a possible lack of statistical power or toxicologic sensitivity.
Environmental exposures to such a mixture may then pose a risk of developmental toxicity in
spite of the lack of developmental effects in the whole-mixture study. In such cases, the
uncertainty caused by the known effects of the component chemicals should be discussed.
Additional evaluation may be warranted before developing the risk characterization.
3.2. WHOLE-MIXTURE RFD/C AND SLOPE FACTORS
3.2.1. Introduction
A dose-response assessment has been done by the Agency for several whole mixtures (see
Sections 3.4.2 and 3.4.3 below). Under certain conditions, a dose-response assessment can be
determined for the mixture itself; a major requirement is that the mixture composition be stable.
This implies that for the exposure duration addressed by the risk assessment, the relative
proportions of the mixture component chemicals are roughly constant so that the mixture can be
treated as though it were a single chemical.
The use of such a dose-response estimate depends on whether the environmental mixture
of concern and the mixture whose data are used to derive the dose-response assessment can be
considered either exactly the same or sufficiently similar. This concept of "sufficient similarity"
can be viewed along a continuum beginning with exposure and dose-response data directly on the
environmental mixture of concern (e.g., human data from an occupational study) to comparing a
mixture for which laboratory dose-response data are available to an environmental mixture (e.g.,
animal toxicity data on a commercial mixture as compared with the same product that has
chemically degraded to some degree in the environment). If the mixtures are highly similar, the
dose-response assessment can be applied with high confidence. As the mixtures being compared
become less similar, there would be less confidence in applying a dose-response assessment
because the mixtures would have different components, or different concentrations of the same
components, so that there would be a greater potential for different toxic effects to occur that
would mask the toxic effect from exposure to the mixture of concern. Thus, the risk assessor
should be able to apply dose-response assessments with confidence from highly similar mixtures,
know the problems of applying them for less similar mixtures, and make some judgment about
where on this continuum each case lies.
-40-
-------
A dose-response assessment for a single chemical by an oral route of exposure may result
in the calculation of a reference dose (RfD), defined on the Agency's Integrated Risk Information
System (IRIS) as follows (U.S. EPA, 2000a):
The RfD is an estimate (with uncertainty spanning perhaps an order of magnitude)
of a daily exposure to the human population (including sensitive subgroups) that
is likely to be without an appreciable risk of deleterious effects during a lifetime.
The RfD is used for oral exposures. For inhalation exposures, the analogous value is the
reference concentration (RfC) (U.S. EPA, 1994a). The RfD is based on the assumption that for a
critical effect, such as cellular necrosis, there exists a dose level at which the effect is not
observed, not expected to occur, or is at a level of severity that is not of concern (e.g., the effect
is reversible or is a mild precursor effect). The mixture RfD is then given as a daily dose (e.g.,
mg/kg/day), where the mg exposure is for the mixture as a whole. The mixture RfD can be
interpreted as an RfD for a single chemical, and its use in a risk characterization, e.g., a Hazard
Index calculation (see Section 4.2), judged similarly. An analogous approach can be taken to
calculate an RfC or a slope factor (U.S. EPA, 1987, 1996a). Data either on the mixture of
concern or on a sufficiently similar mixture can be considered for developing these dose-
response assessments with accompanying discussions of similarity judgment and uncertainty.
3.2.2. Examples of RfD Development for a Whole Mixture
Among the first mixture RfDs were those developed by the Agency's Reference
Dose/Reference Concentration Work Group (RfD/C WG) for the commercial PCB mixtures
Aroclor 1016, Aroclor 1248, and Aroclor 1254 in the early 1990s, with the resulting information
made available on IRIS (U.S. EPA, 2000a). RfDs were derived for Aroclor 1016 and Aroclor
1254, but Aroclor 1248 was deemed "not verifiable." Some details on Aroclor 1016 are
provided below to illustrate this procedure for a whole mixture. For additional information, see
the IRIS database.
3.2.2.1. Aroclor 1016
After a review of the spectrum of effects found in available studies on Aroclor 1016, the
RfD/C WG selected a critical effect of reduced birth weights in a monkey reproductive bioassay
(Barsotti and van Miller, 1984) to establish an RfD of 7E-5 mg/kg/day. This assessment was
supported by a series of reports that evaluated perinatal toxicity and long-term neurobehavioral
effects of Aroclor 1016 in the same groups of infant monkeys (Levin et al., 1988; Schantz et al.,
1989, 1991). An uncertainty factor (UF) of 100 was used: a 3-fold factor is applied to account
-41-
-------
for sensitive individuals; a 3-fold factor for extrapolation from rhesus monkeys to humans; a
threefold factor for limitations in the database, particularly relative to the issue of male
reproductive effects; and a threefold factor for extrapolation from a subchronic exposure to a
chronic RfD.
The NOAEL was selected and UFs applied as if Aroclor 1016 were a single chemical.
The RfD/C WG did, however, provide statements concerning the uncertainty in this assessment,
its applicability to humans, and its use by risk assessors given that the substance is a mixture.
The guidance that was provided on IRIS includes:
"Confidence in the critical studies is rated medium since essentially only one group of
monkeys has been examined. The initial study was well conducted in a sensitive animal
species (rhesus monkeys) that closely resembles humans for many biological functions.
These studies evaluated many sensitive endpoints of PCB toxicity and the effects
observed have also been documented for human exposure.
"The database for PCBs in general is extensive. Studies examining Aroclor 1016 have
been performed in rhesus monkeys, mice, rats, and mink. However, despite the extensive
amount of data available, only medium confidence can be placed in the database at this
time. It is acknowledged that mixtures of PCBs found in the environment do not match
the pattern of congeners found in Aroclor 1016, therefore the RfD is only given medium
confidence. For those particular environmental applications where it is known that
Aroclor 1016 is the only form of PCB contamination, use of this RfD may rate high
confidence. For all other applications only medium confidence can be given."
3.2.3. Example of Cancer Assessment for a Whole Mixture
A dose-response assessment was performed for coke oven emissions, with the results
loaded onto IRIS in 1989 (U.S. EPA, 2000a). Coke oven emissions were determined to be a
human carcinogen, causing increased risk of mortality from cancer of the lung, trachea, and
bronchus; cancer of the kidney; cancer of the prostate; and cancer at all sites combined in coke
oven workers. The inhalation unit risk, defined as the quantitative estimate in terms of
incremental or excess risk per |j.g/m3 air breathed, of 6.2E-4 per |J.g/m3) was based on respiratory
cancer in males exposed in an occupational setting to coke oven emissions. This assessment is
different from most cancer quantitative assessments found on IRIS because it is based on
epidemiologic data on the exposure of concern and because the coke oven emissions mixture is
evaluated as if it were a single chemical. The IRIS description of the quantitative assessment of
the Lloyd-Redmond cohort data (Lloyd et al., 1970; Lloyd, 1971) is as follows:
"Respiratory cancer was considered the most appropriate basis for quantitation, as it was
the common finding among epidemiologic studies. U.S. EPA (1984) calculated an
inhalation unit risk estimate based on the Lloyd-Redmond cohort data assembled by
-42-
-------
Mazumdar et al. (1975) and sorted by Land (1976). The total background U.S. death rate
was used as a basis of comparison rather than the death rate for nonwhite males. A
composite unit risk estimate of 6.2E-4 per |J,g/cu.m was obtained by calculating the
geometric mean of the 95% upper bound estimates obtained for four latency periods (0, 5,
10, and 15 years). This value estimates the human lifetime respiratory cancer death rate
due to continuous exposure to 1 |_ig/cu.m of the benzene-soluble organics extracted from
the paniculate phase of coal tar pitch volatiles from coke oven emissions."
Although coke oven emissions are known to be a complex mixture, differences in
components for the various mixtures exposures were not a part of this assessment. As indicated
in IRIS, the exposures consist of direct exposure to either coke oven emissions by workers or to
the emissions' extracts and condensates in animal inhalation studies and skin-painting bioassays.
The general composition of these emissions is assumed to be stable. The only mention of
components is made in reference to mutagenicity studies of whole extracts and condensates,
where these studies were also done on individual components. These studies provided
supportive evidence for carcinogen!city.
3.2.4. Procedure for a Whole-Mixture Dose-Response Assessment
If a risk assessor wants to calculate an RfD, RfC, slope factor, or other dose-response
estimate for a whole mixture, the general process is to assume the mixture can be treated the
same as a single chemical and proceed with the established methodology for generating that
estimate. This procedure is essentially the same whether the available data are directly on the
mixture of concern or on a sufficiently similar mixture. In the latter case, the risk assessor must
support the similarity assumption in addition to following the single-chemical procedure. The
difference for the mixture assessment lies in several areas: data requirements, the establishment
of the stability of the mixture, cautions relative to dose-response models for mixtures data,
discussions of the uncertainty relative to the mixture assessment, and the need for guidance on
the use of the estimate given that it is based on mixtures data. The following procedural
requirements must be considered:
(1) Data collection and requirements: Human data are preferred for the assessment
from either epidemiologic studies on the exposure of concern or from human
clinical studies directly on the mixture of concern (e.g., clinical studies on
pesticide mixtures). In their absence, a strong animal database, such as the
primate data that were used for the Aroclors, is needed. These data should be
supported by either animal toxicity data on the commercial mixtures or on extracts
from the environmental/occupational exposure, or by human or animal toxicity
data on the major components of the mixture that are deemed to be responsible for
the majority of its toxic effects. Assays that describe the mode of action for the
-43-
-------
mixture are also desirable. In addition, there may be other data requirements for
the methodology of the toxicity value that is being estimated, and these should be
met.
(2) Stability of the mixture: The risk assessor must ascertain that the mixture in
question is relatively stable. Some of the issues that need to be considered include
stability of the mixture in the environment, variability of the mixture composition
over time, sources of the mixture, and potential differences between mixtures
tested in the laboratory and those in the environment (e.g., bioavailability and
route of exposure). In determining stability, consideration should be given to any
information on the environmental exposure that may cause the components to
occur in markedly different concentrations or proportions; if this is the case,
information should be gathered to examine any differences in environmental fate,
in uptake and pharmacokinetics, or in toxicologic effects.
(3) Sufficient similarity (when available data are on a similar mixture): A decision
must be made whether the mixture on which health effects data are available is or
is not "sufficiently similar" to the mixture of concern, using the criteria discussed
in Section 3.1.2. The risk assessor must consider the number of components that
are the same across the mixtures, the differences in their proportions, common
modes of action across the mixtures or their components, and common sources of
formation or emission for the group of mixtures. Whatever judgment is made
must be supported by the risk assessor.
(4) Dose-response assessment: The same procedures may be used as is common for
single-chemical dose-response assessments. The NOAEL RfD/C approach or
benchmark dose methodology, with the application of appropriate uncertainty
factors, can be used for development of one of these values (U.S. EPA, 1996d,
1999). The approaches recommended in the Proposed 1996 Cancer Guidelines
(U.S. EPA, 1996a) may be used to develop estimates of cancer dose response.
There should be some caution, however, in applying dose-response models to
whole-mixture data (e.g., applying a Weibull model to generate a benchmark dose
or using the linearized multistage procedure). Dose-response models that are
empirical and are based on toxicity data similar to the environmental exposure of
interest are more reliable than those requiring substantial extrapolation, either to a
different exposure route or to a much lower dose (concentration) than was used in
the original toxicity studies. The risk assessor must recognize that dose-response
models used for single compounds are often based on biological modes of action
of the toxicity of single compounds, and may not be as well justified when applied
to the mixture as a whole.
(5) Guidance on the uncertainties and usefulness of the assessment: The risk assessor
must fully characterize the nature of the data upon which the estimate has been
made, noting the relevance of the animal, epidemiologic, or clinical data to
environmental exposures. Investigations that were made into establishing the
stability of the mixture should be disclosed, with uncertainties discussed. The risk
-44-
-------
assessor must also be aware of environmental fate issues that may make the
mixture too unstable to be characterized by laboratory toxicity or epidemiologic
data (e.g., the mixture may exist only up to a certain distance from the emissions
source). Attention should be given to the persistence of the mixture in the
environment as well as to the variability of the mixture composition over time or
from different sources. If the components of the mixture are known to partition
into different environmental compartments or to degrade or transform at different
rates in the environment, then those factors must also be taken into account, or the
confidence in and applicability of the risk assessment are diminished. The
confidence in the assessment must be discussed, along with any cautions relative
to its use in risk characterizations (see example in Section 3.2.2 for Aroclor 1016).
3.3. COMPARATIVE POTENCY
3.3.1. The Comparative Potency Method
One of the few procedures for similar mixtures that has been developed and applied to
data on environmental mixtures is the comparative potency method. In this procedure, a set of
mixtures of highly similar composition is used to estimate a scaling factor that relates toxic
potency between two different assays of the same toxic endpoint. The mixture of concern can
then be tested in one of the assays (perhaps a simple assay, e.g., in vitro mutagenicity), and the
resulting potency is then adjusted by the scaling factor to estimate the human cancer potency.
Comparative potency approaches were developed as a means of estimating the toxicity of
a complex mixture in its entirety. Thus far, this method has been applied to data from the testing
of mixtures of emissions released upon the combustion of organics (Albert et al., 1983; Lewtas,
1985, 1988). In addition, the comparative potency procedure has only been applied to estimation
of long-term cancer unit risks, using surrogate test information from short-term cancer bioassays
and in vitro mutagenicity assays. Comparable efforts for noncancer effects are just beginning to
be developed (Gandolfi et al., 1995).
The comparative potency method involves extrapolation across mixtures and across
assays. It is restricted to a set of different assays that monitor the same, single type of health
effect, and to different mixtures that are considered lexicologically similar. The basic
assumption is that the curves of dose response for the assays are the same shape and that the
relationship between any two mixtures will be the same, whichever assay is used. That means, if
you stretch the curve of assay 1 to get the curve of assay 2 for mixture X, then you will stretch it
by the same amount for mixture Y. You also assume the curve of assay 1 for mixture Y is the
same shape as for mixture X. Similarly, if you move the curve for X by a certain amount to
obtain the curve of assay 2 from assay 1's curve, you would do the same for mixture Y. A toxic
potency is one common single-numeric summary of the dose-response curve. Using a numeric
summary allows multiplication and division to move from one assay or mixture to another.
-45-
-------
Thus, if mixture X is twice as potent as mixture Y in assay 1, then X is twice as potent as Y in
assay 2. This constancy of potency ratios can then be used to estimate potency for one mixture in
one assay by using data from other assays and on other similar mixtures.
The comparative potency approach is an example of a similar-mixtures approach to risk
assessment. It is assumed that the mixture of concern can be considered a member of a class of
similar mixtures based on similarity of biologic activity, or reasonable expectation of a type of
biologic activity based on chemical composition. In order to use a comparative potency method,
the risk assessor must test the consistency of dose response for the class of mixtures in question
and test the assumption of a uniform proportionality constant between assays for all mixtures in
the similarity class and for the series of bioassays under consideration.
3.3.2. Theoretical Development
The major assumption in the comparative potency method is that there exists a simple
linear relationship between the mixtures' potencies from each assay for all members of the group
of similar mixtures. The assays themselves, however, need not provide linear dose-response
relationships. Consider an application to cancer unit risk estimation. A mixture with zero
potency (i.e., not carcinogenic) must have zero potency in each bioassay for carcinogenicity, so
the linear relationship across assays must pass through the origin (0,0) of the assay l-assay2 axes
and is then a simple proportionality constant. This relationship is not chosen because it is
simple, but is used because the mixtures are deemed lexicologically similar, and thus can serve
as surrogates for one another. These mixtures must then change in potency from one assay to
another in the same fashion.
In general, this assumption can be expressed as follows. Define:
{ Xi } = group of m similar mixtures, where i = l,...,m (3-1)
{ Aj } = the group of n bioassays, where j = l,...,n (3-2)
Let P represent the toxic potency. Then the above proportionality assumption can be written as:
= k* PAI(XI), for any Xi in the similarity group (3-3)
where k is the proportionality constant that relates the potencies across the two assays. When
there are only two assays and two mixtures, this can be illustrated as in Figure 3-1, where k12
represents the constant proportionality between assays Al and A2, and c12 represents the constant
difference in potency between mixtures Xx and X2.
-46-
-------
PA1(X2) = c,2PA1(X1)
Relative Toxlcltles
Mixture 1 (X1)
\
\
Assay 1 (A1) v
'12
Mixture 2 (X2)
Assay 1 (A1)
ic^ «?
i
Assay 2 (A2)
Assay 2 (A2)
c12 \
Mixture 1(X1) > Mixture 2 (X2)
PA2(X2) =
Figure 3-1. Proportionality assumption for two assays and two mixtures.
-------
When three or more assays are used to establish the necessary relationships, there will be
several such proportionality constants. In general, for assays Ar and As (where r and s are
different and each in the range l,...,n), the constant is ksr:
PAr(Xi) = kSr*PAS(Xi) (3_4)
3.3.2.1. Example With Two Assays
Suppose that we wish to estimate the human cancer potency for mixture X2; thus X2 is the
mixture of concern. Although direct estimation of human cancer potency usually comes from
epidemiological or occupational studies, not actual bioassays on humans, we will stay with that
nomenclature for consistency with the preceding discussion. Suppose that the available
information is the following:
• the group of similar mixtures contains four mixtures Xx through X4.
mixture Xx is twice as potent for human cancer (assay A2) as it is for tumors from
mouse skin painting (assay Al), and the cross-assay potency ratios for mixtures
X3 and X4 are also roughly 2.
• the only potency estimate for X2 is from mouse skin painting studies.
The human cancer potency for X2 is then estimated as follows. First, k in Equation 3-3 (or k12 in
Figure 3-1) can be estimated to be 2. Because X2 is a member of the similarity class that includes
mixtures Xl5 X3, and X4, the same cross-assay ratio holds for X2 as for all the other similar
mixtures. From Equation 3-3 and the estimate of k=2, we then have the human potency estimate
for X2 as:
PA2(X2) = 2 *PA,(X2) (3.5)
Note that if a graph were created plotting the data for these mixtures as points with the potency
for A2 on the y-axis and the potency for Al on the x-axis, then the slope would be roughly 2.
The decision to use this risk (potency) estimate from Equation 3-5 is better substantiated as the
graph becomes more linear.
3.3.2.2. Example With Three Assays (see Figure 3-2)
A slightly more complicated situation involves three assays, with incomplete data for
each one. Suppose again that we wish to estimate the human cancer potency for mixture H, and
that the available data are as follows:
-48-
-------
slope=k21
CD
-t—>
O
Q.
03
O
co
E
Q.
6 "^"2
VO
O
In vivo animal cancer potency
o
c
CD
CD
O
c
CD
O
"CD
E
'c
CD
slope=k
32
In vitro mutagenic potency
Hypothetical comparative potency example using proportionality constants with two assays.
(Left) Human potency estimated from animal data for four mixtures.
(Right) Animal potency estimated from in vitro data for four mixtures.
pH = human potency for mixture H estimated not from the animal data
but from the estimated animal potency for H, aH, which is estimated from the in vitro
potency, mH, so that
V32
m,
Figure 3-2. Comparative potency method - three assays.
-------
• a potency estimate for mixture H has only been measured with the in vitro study
(assay A3).
• three or more mixtures (A, B, C, G in Figure 3-2 right) have been studied with
both assays A3 and A2 (short term in vivo rodent study), and three or more
mixtures (not the same group; A, B, C, D in Figure 3-2 left) have been studied
with both assays A2 and Al (human cancer study).
the two "cross-assay" constants k32 and k21 have been estimated separately using
these two subsets of the class of similar mixtures.
The estimate of human potency (assay Al), using the notation in Equation 3-4, is then calculated
by extrapolating from assay A3 to A2 and then from assay A2 to Al . The calculation is just the
potency of H from assay A3 multiplied by the product of the two cross-assay constants:
PAi(H) = k32 * k2i * PAs(H) (3_6)
As shown in Figure 3-2, the two graphs can be used together as a nomogram where the potency
of H on Al is plotted from its potency on A3 (see dashed lines in the figure). Note that because
data for H exist only with assay A3, the constants k32 and k21 are based only on data for the other
mixtures (A, B, C, D, G) and do not use data on mixture H at all.
3.3.2.3. Example With Combustion Emissions
In this section, this methodology is applied to the estimation of human cancer unit risk
from exposure to polycyclic organic matter (POM) from such mixtures as cigarette smoke, coke
oven emissions, internal combustion engine emissions, and coal burned for heat and cooking
(Nesnow, 1990). This example is only presented to illustrate the application of the comparative
potency method. The unit risk estimates presented here are those published and do not
necessarily represent the current EPA risk estimates for the chemicals involved.
The data for this example are given in Table 3-1 and plotted in Figure 3-3. The diesel
estimate for human cancer unit risk in Table 3-1 was derived based on a rat inhalation study,
from a different species than the other mixtures' values. The human potency estimates for the
other three mixtures are based on epidemiologic data, which allows us to gauge how this potency
prediction compares to the standard species-to-species extrapolation. The regression line in
Figure 3-3 is based on the data without diesel, and its slope represents the cross-assay
proportionality constant, or the way to scale from the mouse skin potency (A2) for diesel via the
remaining mixtures to the human unit risk (Al) from diesel. This particular proportionality
constant (k = 4 x 10"4) is not significantly different from zero at one typical level of 0.05 (p =
0.14), though the adjusted model r-square is 0.91, which suggests the model explains a lot of the
-50-
-------
Table 3-1. Comparative potency method for emission extracts'1
Combustion product
Coke oven emissions
Roofing tar
CSC
Diesel
Mouse skin tumor initiation1'
2.1
0.40
0.0024
0.31
Human lung cancer unit risk0
(Hg/m3)-1
9.3 x 1Q-4
3.6 x 1Q-4
2.2 x 1Q-6
(0.7 x io-y
a From Ne snow, 1990.
b Expressed as number of papillomas/mouse at 1 mg organics.
0 Direct estimates from human data.
d The diesel value was based on rat inhalation data (Albert and Chen, 1986) and was adjusted
for the percentage of organics on the particulates.
variability. For our purposes, however, with only three points, a more relaxed significance level
(type I error rate) (e.g., a = 0.20) may well be good enough. So we could substitute this value of
k in Equation 3-3 to get:
PAi(diesel) = (4 x 10'4 )* PA2(diesel).
(3-7)
This estimate using comparative potency compares reasonably well with an estimate of 0.7 x
10"4 derived by traditional single-sub stance methods from rodent data (Table 3-1).
3.3.2.4. Use of Relative Potencies
Previous publications on comparative potency (Lewtas, 1985; Schoeny and Margosches,
1989) have performed the calculations using the "relative potency" (i.e., the ratio of the potency
of the mixture of concern to that of a "reference mixture") in the same assay, instead of using the
actual mixture potencies. Such scaling of the actual potencies does not add any information, nor
does it increase the flexibility of the approach. Consider a graph of PA2 versus PA1 (i.e., the
mixture potencies for assay A2 plotted against the mixture potencies for assay Al; two such
graphs are shown in Figure 3-2). Scaling a quantity by a constant (e.g., the reference mixture)
only changes the numbers on the axes of the graph, but the shape of the curve through the data
points remains unchanged. Thus, regardless of the reference mixture used for scaling the
potencies, even if different in each assay, the only relationship required is that the same
proportionality constant across assays holds for all the similar mixtures.
-51-
-------
~5b
10.0
8.0
S 6.0
S3 4.0
2.0
03
O
3
H-l
0.0
-2.0
Coke Oven
footing Tat-
Ciga/ette Smoke
-1.0 .0 1.0 2.0 3.0
Mouse Skin Tumors (Papillomas/mouse)
4.0
Linear regression shown with 95% confidence bands based on nondiesel data.
Figure 3-3. Combustion mixtures (PAH).
Data from Nesnow (1990).
The use of a scaled potency for comparing assays has some advantages, however, because
all potencies are then "standardized" to be numbers near one (1.0), and the differences are more
easily visualized. The problem occurs when tables of these standardized values are
used for calculations instead of for carrying out such statistical methods as a regression. The
weakness with using relative potencies is that the relative potency for the reference mixture
(relative to itself) is always viewed as exactly 1.0; it is no longer perceived as a measured random
variable but is presumed to be exact, and the variation is all assumed to lie with the other
mixtures' potencies. This is clearly wrong. Consequently, regression across all mixtures should
be used instead. But even when regression is used, and the index mixture value is displayed with
a confidence interval (e.g., 1.0 [0.5-2.8] ), the visual comparison will still tend to focus on other
values in comparison to 1.0. To avoid misinterpretation, it is better to give an analysis of the
-52-
-------
"constant ratio" assumption (i.e., the assumption of Equation 3-3) separately from the table of
potency data.
3.3.3. Procedure for Applying the Comparative Potency Approach
Using the comparative potency method requires gathering and analyzing data on several
mixtures along with considerable judgment of toxicologic similarity. The approach should be
limited to the assessment of a mixture for which whole-mixture in vivo toxicity studies have not
been done, and where the composition of the mixture is deemed too complex for the application
of component-based assessment methods. Because this is a methodology based on the
comparison of different mixtures and different types of data, and not on an extrapolation from
directly related human health data, it is expected that these estimates will be accurate only within
an order of magnitude. The following main steps have been identified:
• Similarity of Mixtures: Develop the class characteristics or other similarity
criteria for the group of mixtures, including the mixture of concern, in order to
support the assumption that the group of mixtures can be judged as
"lexicologically similar."
• Data Collection: Compile the available toxicity data on the mixtures in the
similarity class and evaluate them for general quality and applicability to the toxic
endpoints of interest for the mixture of concern.
• Potency Relationship: Describe the degree of consistency within the mixture
group of the cross-assay potency ratios, and estimate values to support the
constant potency ratio relationship.
• Dose-Response Characterization: Describe the best estimates of the cross-assay
ratios along with all uncertainties in their application to human risk assessment for
the mixture of concern.
3.3.3.1. Similarity of Mixtures
The comparative potency approach is built on the assumption that the mixtures under
consideration, including the mixture of concern, act in a similar manner lexicologically. A
determination can be made that a group of mixtures is lexicologically similar by establishing
criteria that any given mixture must satisfy in order to be designated as a member of that group.
The risk assessor must be able to support the assumption that the mixtures are similar, and can do
so by using any of a number of approaches that define chemical structure or biologic criteria: (1)
establishing that a common mode of action exists across the mixtures; (2) showing consistency in
results of short-term screening assays; (3) distinguishing chemical class or chemical structure
similarity; (4) identifying common components across the mixtures in similar proportions; and
(5) establishing a common source of formation or emission for the group of mixtures. Although
there are references to the use of comparative potency for endpoints other than cancer (Albert,
-53-
-------
1985), the methodology has been used by EPA only for cancer potency prediction. Use of
comparative potency for noncancer endpoints depends on the availability of accepted short-term
tests relevant to those endpoints.
The mixture class characteristics that are thought most useful for prediction are those
determined from data on biologic activity of the mixtures, specifically including whether the
mixtures cause an effect by the same mode of action. It should be emphasized that, in estimating
human potency by extrapolating from in vivo or in vitro test data, expert judgment will be needed
to verify that a common mode of action may be expected to operate for the mixtures of interest
across the test systems. For example, the mouse skin tumor bioassay has been shown to be an
appropriate system for estimating human lung tumor potency for PAH mixtures and alkylating
agents, but not for metal carcinogens (Nesnow and Lewtas, 1991); the conclusion is that different
modi operandi obtain for metals in humans than are seen in mouse lung.
Short-term screening tests can be used to determine similarity, including in vitro and in
vivo models. Short-term testing to evaluate genetic toxicity (e.g., tests for DNA damage, gene
mutation, cell transformation) have been suggested to characterize similar mixtures (Nesnow,
1990). Other test systems for carcinogenicity screening, such as the Syrian Hamster Embryo
(SHE) Cell Transformation Assay or the Japanese Medaka (Oryzias latipes\ would also be
candidates for short-term screening of similarity.
The identification of the major components in common for the group of mixtures can be a
useful way to screen for similarity. For example, a simple chemical fractionation that indicates
substantial amounts of polycyclic aromatic hydrocarbons (PAH) or aromatic amines are present
may be the basis for a preliminary grouping of similar mixtures. Nesnow (1990) suggests that
common indicator constituents may be used to predict similar effects across mixtures when it can
be assumed that the indicator constituents are responsible for a significant amount of the adverse
effect. As the number of major components within the group of mixtures increases and the
mixture becomes more complex, these methods are less reliable. EPA researchers have
evaluated mixtures of up to 25 chemicals (Simmons et al., 1994) and describe difficulties in
toxicologic evaluation of complex mixtures (Simmons et al., 1995). When this type of
component identification is performed, care must be given to the relative proportions of the
components within each of the mixtures to determine if differences in proportions are significant
enough to change the type or magnitude of the effects.
Another potential screening method for similarity of mixtures is to examine the
similarities of individual major chemical components by activity profile and/or structure-activity
relationships (SAR) analysis. Nesnow (1990) suggests that EPA's genetic activity profile (GAP)
software can be used to identify structurally and or biologically similar chemicals (Waters et al.,
1988a,b). The OncoLogic Cancer Expert System developed for EPA (Woo et al., 1995a) can be
-54-
-------
used to screen for structurally and/or functionally similar chemicals with respect to
carcinogenicity as the toxicity endpoint. Other SAR models can also be applied that will give
indications of expected toxicity. For example, one module of the TOPKAT® structure-activity
relationship software that was developed for the EPA predicts the chronic rat LOAEL for
chemicals by using a linear regression of the LOAEL on chemical structure descriptors (Mumtaz
et al., 1995). Other endpoints, such as the probability of carcinogenesis, can also be predicted
using the TOPKAT® model (Enslein et al., 1990). Note that these SAR models are limited in that
they only generate predictions for single chemicals, which must be extrapolated to infer
similarity among a group of mixtures.
Consideration of the origin of the mixture provides another means for grouping; for
example, mixtures resulting from incomplete combustion of organics are expected to show some
degree of similarity. The degree of similarity can be pursued by combining information from the
origin of mixture and chemical composition of archetypal mixtures. Thus, the risk assessor could
expect mixtures of POM from various types of diesel engines to constitute a similarity class; one
could expect more common characteristics within this similarity subclass than across the whole
universe of combustion mixtures or with another combustion subclass (e.g., tobacco smoke
condensates).
3.3.3.2. Data Collection
The act of collecting data for use in the comparative potency approach involves
compiling the available toxicity data on the mixtures in the similarity class and evaluating them
for general quality and applicability to the toxic endpoints of interest for the mixtures of concern.
The data must be evaluated for relevance in two areas: (1) to the toxic endpoint being assessed;
and (2) for the mixture class. Assays most useful are those that can be shown to provide
measures of toxicologic changes generally accepted as relevant to the mode of action. For
carcinogenicity there are many short-term or limited-scale assays generally considered to be
relevant to processes in humans: skin-painting in rodents, in vitro cell transformation, and
development of preneoplastic liver cell foci, to name a few. For certain carcinogens that act by
altering genetic material, it is generally accepted that mutagenicity tests in vitro can provide
relevant data. For noncancer endpoints there are fewer well-established short-term tests, but
changes in appropriate cellular receptor binding or enzyme levels are among those that could be
used.
A consideration for the suitability of assay systems is similarity of pharmacokinetics
among the systems and to the human situation. For most assurance of similarity, the metabolites
produced and/or absorption characteristics for the chemicals/mixtures of interest should be
identical (or at least comparable) across the test systems.
-55-
-------
The data must also be evaluated in terms of providing information relevant to the human
health risk assessment of the particular mixture. For example, Salmonella typhimurium strains
widely used for in vitro mutation tests have an endogenous nitroreductase enzyme system not
found in human cells. One would need to consider relevance of data from Salmonella tests when
evaluating mixtures high in nitropyrenes that are easily activated by the bacteria, but may not be
metabolized to carcinogens by humans.
There are numerous points in deciding whether or how to apply comparative potency.
Some of these are described in Schoeny and Margosches (1989). The NRC (1988) publication
Complex Mixtures—Methods for In Vivo Toxicity Testing provides guidance not only for testing
but for sampling and interpretation of data. Some decision issues are considered below.
1. Use of extrapolation procedures. Extrapolations that are used for the comparative
potency approach should be carefully applied and justified. For example, these may include
using animal data to estimate human risk, using subchronic data to estimate risk from chronic
exposures, using oral or dermal data to estimate inhalation risks, or using high-dose exposures
from long-term or short-term in vitro or in vivo tests to estimate risks from low exposures that
humans would typically encounter in environmental media. Processes and considerations for
some such extrapolations may be found in the original U.S. EPA Risk Assessment Guidelines
(U.S. EPA, 1986, 1987) (Appendix A) and subsequent guidelines for carcinogenicity,
developmental toxicity, reproductive toxicity, and neurotoxicity (U.S. EPA, 1996a, 199la,
1996b, and 1998b, respectively).
2. Availability of human data suitable for a quantitative assessment. The original
demonstration of the comparative potency method used three combustion-related mixtures for
which there were human data sufficient for derivation of a human cancer unit risk estimate (as
shown in Section 3.3.2.3). Human cancer unit risk estimates for diesel emissions from specific
engine types were then derived from a central tendency estimate of the three existing human
cancer unit risks on the similar combustion mixtures (Schoeny and Margosches, 1989). Greater
confidence can be attached to a comparative potency approach that relies at some point on at
least one human cancer unit risk estimate based on human data.
Compounds for which there are no quantitative human data could be used in the process
if they are known to have a well-characterized response in an animal model that is a known
reflection of human toxicity. Cancer response data from animal testing of the mixture should be
evaluated following the Agency's Guidelines for Cancer Risk Assessment (U.S. EPA, 1986) and
supplemented by the revised Proposed Guidelines for Cancer Risk Assessment (U.S. EPA,
1996a). In using data from animals for comparative potency, care must be taken to utilize
reasonable, scientifically based dose extrapolation processes. In particular, uncertainties
-56-
-------
introduced when extrapolating across exposure routes can be excessive and hence must be
articulated and quantified when possible.
3. Form, source, and preparation of the environmental mixture sample. Ideally the risk
assessor would use data on the form of the mixture and mode of exposure most like those
encountered by humans. For combustion-related mixtures, for example, the risk assessor would
prefer data from inhalation assays of vapor phase plus particulate. This type of assay is least
likely to be encountered in the literature, as its development is most resource intensive. The use
of data from testing of the mixture in a form not presented to humans is also a source of
uncertainty. For example, in the original demonstration of the comparative potency method,
POMs, organic extracts of combustion particulate, were tested in mouse skin initiation/promotion
studies and in vitro. By contrast, humans would be most often exposed (at least through
inhalation) to a combustion mixture consisting of volatile materials and mixed sizes of particles
associated with organic and inorganic compounds. The NRC (1988) gives useful guidance on
collecting representative samples and their preparation for bioassay. In choosing to use data from
fractions (such as organic extractables from particulate matter) or more feasible modes of
administration (such as skin painting), the risk assessor introduces further areas of uncertainty
into the estimate of risk. It is necessary to describe these uncertainties, limit and quantify them to
the extent possible, and provide justification for decisions made in data or assay choice. Point of
sampling and preparation of sample must also be considered and the decisions explained. An
example of a decision-making process and justification for decisions is found in Albert et al.
(1983). Some considerations for data collection specific to short-term tests are found in Schoeny
and Margosches (1989) and Nesnow (1991).
3.3.3.3. Potency Relationships
The next step is to estimate the degree of consistency in the assay ratios across the similar
mixtures and estimate values to support the constant relative potency relationship. Having
selected appropriate data types, the risk assessor then evaluates the hypothesis of consistent
relative potency. If relative potency ratios are consistent across similar mixtures for one type of
assay but not others, it indicates the limitations of application of comparative potency. In other
words, if only assays relating to cancer as an endpoint are consistent, the comparative potency
estimation should be limited to cancer; if only receptor binding is consistent, the application
should be limited to health endpoints associated with receptor binding. If there are data
applicable to only one health endpoint, the methodology should not be extended to other health
endpoints. In order to estimate a constant for the relative potency assay ratios for the similar
mixtures, it is recommended that a linear regression model without an intercept parameter be
used, as illustrated in Section 3.3.2.3.
-57-
-------
3.3.3.4. Dose-Response Characterization
This final stage of the comparative potency approach is the most important for
communication and risk management decisions. Where environmental issues are significant, the
risk assessment is incomplete without a characterization of the process used to determine the
dose-response value. This stage includes the calculating of human potency estimates, with a full
description of the uncertainty and variability of the application. The dose-response
characterization should include such information as the following:
• data quality and availability,
criteria used to determine consistency of relative potency ratios and the parallel
relationship between types of assays,
basis for the determination that the class of mixtures qualified as sufficiently
similar,
description of any extrapolations that were made, such as route-to-route or animal
to human,
full disclosure of statistical procedures that were used, any assumptions made, and
significance levels used for any hypothesis testing (e.g., significant slope
parameter for the linear regression), and
• explanation of the level of confidence in the final human potency estimates and an
estimate of the variability inherent in these numbers.
3.4. ENVIRONMENTAL TRANSFORMATIONS
3.4.1. Using Environmental Process Information to Determine Mixture Similarity
Environmental processes can affect the exposure, and thus the toxicity, of a mixture in the
environment, so one approach to a whole-mixture assessment is to adjust the risk assessment
based on what is known about the mixture because of environmental transformations. When a
mixture is altered in the environment, it is not practical to expect toxicity information to be
available for each specific environmental mixture to which humans are exposed. It is more likely
that there will be toxicity information for only a few standard mixtures or mixture components.
If information is available on some similar standard mixtures, then a feasible approach would be
to determine which standard mixtures best resemble the environmental mixture and use the
toxicity information from those standard mixtures as a surrogate for the environmental mixture's
toxicity. In the case of information available on mixture components, then a component-based
approach may be feasible.
In either case, it is important to discuss how the mixture is altered in the environment,
and which source of toxicity information provides the best surrogate. It is also important to
discuss what uncertainties remain even after the best surrogate information is used to estimate
risks from the environmental mixture, as mixtures encountered in the environment can be
markedly different from the mixtures originally released into the environment or the mixtures
-58-
-------
subjected to toxicity testing. Partitioning and bioaccumulation, for example, can cause
substantial changes in an environmental mixture. When partitioning is involved, different
exposure pathways can involve exposure to different mixture fractions; for example, the mixture
fraction adsorbed to soil can be different than the mixture fraction dissolved in drinking water.
When bioaccumulation is involved, the mixture fraction to which humans are exposed can be
more persistent than the original mixture, as the bioaccumulated mixture can contain a higher
proportion of the mixture components that resist metabolism and elimination. Note that this
approach makes a link between dose-response assessment and exposure assessment, as the
circumstances of exposure can alter the potency of a mixture in the environment.
3.4.2. Procedures for Incorporating Environmental Process Information
Different procedures should be followed depending on the degree to which most of the
components in the mixture have toxicity data available for evaluation. Guidance on approaches
for using environmental process information to determine mixture similarity, given certain data
scenarios, are given below:
Data scenario/approach: Toxicity information is available on most mixture component
chemicals/use component-based approaches.
If all relevant component chemicals have toxicity information and have been measured at
the time and location where population exposure is expected, then estimate the mixture
toxicity by combining the component chemical toxicities. One way is to develop a
Hazard Index for each toxic endpoint of interest (Section 4.2). If the chemicals are
sufficiently similar to form a toxicologic class, then relative potency factors can be
estimated (Section 4.4).
Data scenario/approach: Toxicity information is available on only a few mixture
components/use bounding estimates and similar mixture data.
(a) If too many chemicals lack specific exposure or toxicity information but some
sense of total exposure can be obtained, then a bounding approach can be used.
The mixture toxicity is estimated then as a range, from the worst case (assume all
components are as toxic as the most toxic component) to the least case (assume all
components are as toxic as the weakest component). Consider the environmental
influences to determine how the components and mixture composition will
change over time and during transport to the receptor population. Determine
which chemical components will be dominant in the population exposure, and
reflect that determination by a recommendation of how close to each extreme the
mixture toxicity is likely to be.
-59-
-------
(b) If the mixture can be characterized by its source, for example as a specific
commercial mixture, then the mixture exposure and toxicity might be estimated
by using data on an environmentally transformed similar mixture. The use of
toxicity data on transformed whole mixtures is encouraged because it obviates the
need for full identification and measurement of the mixture components. The
decision regarding similarity must consider information and uncertainties on
differences in total exposure level, in relative proportions of components, in
exposure levels of key components (high toxicity and/or exposure level), and in
the proportion of unknown chemical components. These differences should be
judged for the transformed mixture to which the population is exposed, not for the
original mixture.
(c) If a high fraction (e.g., >30%) of chemicals in the environmental exposure cannot
be identified, the assessor must judge whether the source mixture could have been
altered by some components being transformed into chemicals not in the source
mixture. In that case, the unidentified chemicals should be investigated further,
using test methods that artificially degrade the mixture or using extrapolation
methods such as QSAR on the source mixture components. If such an
investigation is not feasible, then the unknown chemicals constitute a major
uncertainty in the mixture assessment, which must be clearly stated.
In addition to the uncertainties described in the procedural sections for the Hazard Index
(Section 4.2), relative potency factors (Section 4.4), and whole-mixture testing (Section 3.1.5),
the risk characterization must also discuss the extent of understanding of the transport and
transformation of the component chemicals from the source to the exposed population. In
particular, the characterization must include the identification of the chemical components and
the assumptions and errors in determining concentrations at the point of population exposure.
3.4.3. Geographic Site-Specific Modifications: An Example Using PCB Mixtures
EPA's approach to assessing the cancer risk from environmental PCBs (U.S. EPA,
1996c; Cogliano, 1998) illustrates both the similar-standard-mixture approach and the relative
potency approach described above. There have been no cancer bioassays for PCB mixtures as
encountered in the environment, but these environmental mixtures are being assessed using both
approaches. The similar-standard-mixture approach relies on cancer bioassays for a few standard
PCB mixtures formerly used in commerce, whereas the relative potency approach is based on a
large body of experimental information that elucidates modes of action or mechanisms of toxicity
and quantifies their potency for a small number of PCB congeners that act like dioxin.
-60-
-------
3.4.3.1. Composition of PCB Mixtures
PCBs are chemical mixtures of variable composition. Mixture components are called
"congeners," with 209 different congeners possible. Although their chemical properties vary
widely, different mixtures can have many common components. Table 3-2 shows the
overlapping composition of some commercial mixtures in terms of congeners with 1 to 10
chlorines. PCB mixtures manufactured in the United States carried the trademark "Aroclor"
followed by a four-digit number; the first two digits were "12," and the last two digits indicated
the percent chlorine by weight. Aroclor 1016, with approximately 41% chlorine, is an exception
to this scheme.
3.4.3.2. Hazard Assessment and Dose-Response Assessment for PCBs
Toxicity information is available for several Aroclors. Among the many studies that
implicate PCBs as likely to cause cancer in humans, a recent study comparing four Aroclors
(Brunner et al., 1996; Mayes et al., 1998) provides the best information for distinguishing the
cancer potential of different mixtures. Groups of 50 male or female Sprague-Dawley rats were
fed diets with different concentrations of Aroclor 1016, 1242, 1254, or 1260; there were 100
controls of each sex. Exposure began when the rats were 6 to 9 weeks old, and the animals were
killed 104 weeks later. Statistically significantly increased incidences of liver tumors were found
in female rats for all Aroclors and in male rats for Aroclor 1260 (Table 3-3). In female rats,
Aroclor 1254 appeared most potent, followed by Aroclors 1260 and 1242, with Aroclor 1016
markedly less potent. In male rats, only Aroclor 1260 caused liver tumors.
Because these Aroclors contain overlapping groups of congeners that together span the
range of congeners most often found in environmental mixtures, EPA concluded that all
environmental PCB mixtures pose a risk of cancer. The dose-response assessment, however, was
able to make distinctions in the potencies of these mixtures. Using the increased incidences of
liver tumors in female Sprague-Dawley rats, central-estimate and upper-bound slope factors were
calculated for each of the four tested Aroclors (Table 3-4).
3.4.3.3. Exposure Assessment and Risk Characterization for PCBs
In the environment, PCBs occur as mixtures whose compositions differ from the
Aroclors. This is because after release into the environment, mixture composition changes over
time, through partitioning, chemical transformation, and preferential bioaccumulation.
Partitioning refers to processes by which different fractions of a mixture separate into air, water,
sediment, and soil. Chemical transformation can occur through biodegradation of PCB mixtures
in the environment. Preferential bioaccumulation occurs in living organisms, which tend to
-61-
-------
Table 3-2. Typical composition of some commercial PCB mixtures
Aroclor
Mono-CBs (% wt)
Di-CBs
Tri-CBs
Tetra-CBs
Penta-CBs
Hexa-CBs
Hepta-CBs
Octa-CBs
Nona-CBs
Deca-CBs
PCDFs (ppm)
Chlorine content (%)
Production, 1957-1977 (%)
1016
2
19
57
22
-
-
-
-
-
-
ND
41
13
1242
1
13
45
31
10
-
-
-
-
-
0.15-4.5
42
52
1248
-
1
21
49
27
2
-
-
-
-
NR
48
7
1254
-
-
1
15
53
26
4
-
-
-
0.8-5.6
54
16
1260
-
-
-
-
12
42
38
7
1
-
0.8-5.6
60
11
- = less than 1%.
ND = not detected.
NR = not reported.
Sources: Compiled by U.S. EPA (1996c) from other sources.
concentrate congeners of higher chlorine content, producing residues that are considerably
different from the original Aroclors. Thus, an Aroclor tested in the laboratory is not necessarily
the best surrogate for assessing that Aroclor as altered in the environment.
EPA encourages risk assessors to consider how environmental processes alter PCB
mixture composition and toxicity. Through partitioning, different portions of a PCB mixture are
encountered through each exposure pathway. The mixture fraction that adsorbs to sediment or
soil tends to be higher in chlorine content and persistence than the original mixture; it tends also
to be less inclined to metabolism and elimination, and thus higher in persistence and toxicity.
Consequently, ingesting contaminated sediment or soil or inhaling contaminated dust can pose
relatively high risks. On the other hand, the mixture fraction that dissolves in water or
evaporates into air tends to be lower in chlorine content and persistence, so risks from ingesting
water-soluble congeners or inhaling evaporated congeners would tend to be lower, in the absence
-62-
-------
Table 3-3. Liver tumora incidences for Aroclor mixtures
Mixture
Aroclor 1260
Aroclor 1254
Aroclor 1242
Aroclor 1016
Dose
Control13
25 ppm
50 ppm
100 ppm
Control13
25 ppm
50 ppm
100 ppm
Control13
50 ppm
100 ppm
Control13
50 ppm
100 ppm
200 ppm
Females
1/85 (1%)C
10/49 (20%)
1 1/45 (24%)
24/50 (48%)
1/85 (1%)C
19/45 (42%)
28/49 (57%)
28/49 (57%)
1/85 (1%)C
1 1/49 (24%)
15/45 (33%)
1/85 (1%)C
1/48 (2%)
6/45 (13%)
5/50 (10%)
Males
7/98 (7%)c
3/50 (6%)
6/49 (12%)
10/49 (20%)
7/98 (7%)
4/48 (8%)
4/49 (8%)
6/47 (13%)
7/98 (7%)
1/50 (2%)
4/46 (9%)
7/98 (7%)
2/48 (4%)
2/50 (4%)
4/49 (8%)
a Hepatocellular adenomas, carcinomas, cholangiomas, or cholangiocarcinomas in rats alive
when the first tumor was observed.
b One control group supported all experiments.
0 Statistically significant (p<0.05) by Cochran-Armitage trend test.
Source: Brunner et al., 1996, reported by U.S. EPA, 1996c.
Table 3-4. Human slope estimates (per mg/kg-day) for Aroclor mixtures
Mixture study
1016, Brunner etal., 1996
1242, Brunner et al., 1996
1254, Brunner et al., 1996
1260, Brunner et al., 1996
1260, Norback, 1985
Central slope
0.04
0.3
1.2
0.4
1.6
Upper-bound slope
0.07
0.4
1.5
0.5
2.2
Source: U.S. EPA, 1996c.
-63-
-------
of contaminated sediment or dust. Preferential bioaccumulation can have even more pronounced
effects, as each species in the food chain retains persistent congeners that prove resistant to
metabolism and elimination. Bioaccumulated PCBs appear to be more toxic than Aroclors and
more persistent in the body. The Aroclors tested in laboratory animals were not subject to prior
selective retention of persistent congeners through the food chain. For exposure through the food
chain, therefore, risks can be higher than those estimated in an assessment. (This last statement
is an example of characterizing uncertainties that remain even after the best surrogate
information is used to estimate risks from an environmental mixture.)
To reflect these environmental processes, EPA developed a tiered approach that considers
how partitioning and bioaccumulation affect each exposure pathway or situation. Three tiers are
provided:
High risk and persistence (upper-bound slope, 2 per mg/kg-d; central-estimate slope, 1
per mg/kg-d). The highest slope from Table 3-4 is used for pathways where
environmental processes tend to increase risk: food chain exposure, sediment or soil
ingestion, dust or aerosol inhalation, exposure to dioxin-like, tumor-promoting, or
persistent congeners, and early-life exposure (all pathways and mixtures).
Low risk and persistence (upper-bound slope, 0.4 per mg/kg-d; central-estimate slope,
0.3 per mg/kg-d). A lower slope is appropriate for pathways where environmental
processes tend to decrease risk: ingestion of water-soluble congeners and inhalation of
evaporated congeners. Dermal exposure is also included, because PCBs are incompletely
absorbed through the skin; however, if an internal dose has been calculated by applying
an absorption factor to reduce the external dose, then the highest slope would be used
with the internal dose estimate.
Lowest risk and persistence (upper-bound slope, 0.07 per mg/kg-d; central-estimate
slope, 0.04 per mg/kg-d). The lowest slope from Table 3-4 is used when congener or
homologue analyses verify that congeners with more than four chlorines comprise less
than one-half percent of total PCBs. Such a mixture composition is used to established
sufficient similarity to the tested mixture Aroclor 1016.
3.4.3.4. Relative Potency Approach for PCBs
The World Health Organization has developed toxic equivalency factors for 13 dioxin-
like PCB congeners. When dioxin-like congener concentrations are reported for an
environmental sample, the mixture-based approach can be supplemented by an analysis of the
dioxin toxic equivalents contributed by the dioxin-like PCB congeners. Such an analysis is
particularly important when environmental processes have increased the concentrations of
dioxin-like congeners as a fraction of the total PCB mixture.
-64-
-------
Because PCBs can cause cancer through both dioxin-like and non-dioxin-like modes of
action, it is important to consider the contribution from both dioxin-like and non-dioxin-like
modes of action to the total risk. Risks for the dioxin-like and non-dioxin-like portions of the
mixture are calculated separately. For the dioxin-like portion, a relative potency approach is
used. The dose of each dioxin-like congener is multiplied by its toxic equivalency factor, then
these products are summed to obtain the total dioxin toxic equivalents present in the PCB
mixture. This, in turn, is multiplied by the dioxin slope factor to estimate the risk from dioxin-
like modes of action. For the non-dioxin-like portion, a similar-standard-mixture approach is
used. The total dose of PCBs, less the dose comprising the 13 dioxin-like congeners already
considered, is multiplied by the appropriate PCB slope factor as determined in the previous
section. U.S. EPA (1996c) provides a detailed example of these calculations.
3.4.3.5. On Estimating a Mixture's Persistence
The persistence of PCB mixtures is sometimes characterized by a measure of half-life.
EPA's assessment cautions that ascribing a half-life to a mixture is problematic if the half-lives of
its components differ widely. More specifically, half-life estimates for a mixture will
underestimate its long-term persistence. To illustrate, consider a mixture of two components in
equal parts: one component has a half-life of 1 year; the other, 100 years. If the mixture
concentration is sampled after 10 years, the half-life of the total mixture will appear to be
approximately 10 years: virtually all the first component will be gone, and virtually none of the
second, so about half the original mixture will remain. This half-life, however, overestimates the
slow rate of decrease in the more persistent mixture fraction that remains.
-65-
-------
4. METHODS FOR COMPONENT DATA
4.1. INTRODUCTION
If data are not available on an identical or reasonably similar mixture, the risk assessment
may be based on the toxic or carcinogenic properties of the components in the mixture. When
quantitative information on toxicologic interaction exists, even if only on chemical pairs, it
should be incorporated into the component-based approach. When there is no adequate
interactions information, dose- or response-additive models are recommended. Several studies
have demonstrated that dose (or concentration) addition often predicts reasonably well the
toxicities of mixtures composed of a substantial variety of both similar and dissimilar compounds
(Pozzani et al., 1959; Smyth et al., 1969, 1970; Murphy, 1980; Ikeda, 1988; Feron et al., 1995),
although exceptions have been noted. For example, Feron et al. (1995) discuss studies where
even at the same target organ (the nose), differences in mode of action led to other than dose-
additive response. Dose-additive models may be an adequate default procedure for chemicals
affecting the same target organ, but may not be the most biologically plausible approach if the
compounds do not have the same mode of toxicologic action. Consequently, depending on the
nature of the risk assessment and the available information on modes of action and patterns of
joint action, the most reasonable dose-response model should be used.
The mixtures methods in this chapter rely heavily on existing EPA risk assessment
information on single chemical toxicity, such as that in the EPA IRIS files. Levels of exposure
for the mixture component chemicals are assumed to be estimates obtained following the
appropriate Agency exposure assessment guidance (e.g., U.S. EPA, 1992). The procedures and
terminology associated with dose response and risk characterization for single chemicals, such as
the RfD, RfC, and cancer potency values, have the same interpretation in the mixture procedures
in this chapter. The following descriptions of component-based mixture methods include
references, but assume the reader is familiar with these single-chemical risk assessment concepts
and practices.
4.1.1. Criteria for Dose Addition vs. Response Addition
Toxicologic interactions are defined in this guidance document (Appendix B) to facilitate
the selection and application of specific risk assessment methods. When adequate evidence for
toxicologic interactions is not available, the most appropriate no-interaction approach (dose
addition or response addition, as detailed below) will be employed. Toxicologic "interactions"
are then operationally defined by mixture data showing statistically or lexicologically significant
deviations from the "no-interaction" prediction for the mixture.
-66-
-------
Several differing definitions of "no interaction" are discussed in the scientific literature.
Plaa and Venzina (1990) provide a nice historical overview of the differences in definitions, and
Kodell and Pounds (1991) discuss some of the implications of these differences. Muska and
Weber (1977) introduced the terms "concentration addition" and "response addition." Their
definitions are based on ideas related to general toxicologic modes of action; i.e., concentration
addition (also termed dose addition) applies when the components act on similar biological
systems and elicit a common response, whereas response addition applies when components act
on different systems or produce effects that do not influence each other.
In this guidance, "no interaction" is defined using the two common concepts of Muska
and Weber (1977): dose addition and response addition. These definitions have been selected
because the underlying concepts are straightforward and in common use, and because hypothesis
tests exist to determine whether data are consistent with each of these concepts (see Gennings,
1995; Gennings and Carter, 1995). These definitions do not indicate specific toxicologic modes
of action, although they should be consistent with the major examples and concepts of
toxicologic interaction. Dose addition and response addition then represent default approaches
for lexicologically similar and lexicologically independent chemicals, respectively. The risk
assessment using component data should then begin by selecting the most appropriate concept
for the chemicals in the mixture. There will be many cases where the information does not
support either dose or response addition. In those cases, the mixture should be further
investigated, and consideration should be given to using methods that incorporate combinations
of dose and response addition as well as toxicologic interactions. Information on interactions can
be included as modifications of the "no-interaction" approach that was selected (see Sections 4.3
and 4.5.4).
The primary criterion for choosing from dose or response addition as the no-interaction
approach is the similarity or independence among the chemicals in the mixture. This judgmental
decision, detailed further in Sections 4.1.1.1 and 4.1.1.2, should be based on information about
the toxicologic and physiological processes involved, the single-chemical dose-response
relationships, and the type of response data available. If tissue levels can be estimated, then the
judgment of similarity or independence can focus on the toxicologic mode of action. If external
exposure levels are used instead of tissue doses, then the judgment of toxicologic similarity or
independence must consider all the processes from contact with the environmental media to the
toxicity itself (i.e., uptake, metabolism, distribution, elimination, and toxicologic mode of
action). To facilitate understanding, the discussions that follow will initially consider only two-
chemical mixtures. For additional explanation of these concepts, see Svendsgaard and Hertzberg
(1994).
-67-
-------
4.1.1.1. Dose Addition
In the simplest terms, two chemicals are dose additive if chemical B is functionally a
clone of chemical A. In this ideal case, the chemicals are assumed to behave similarly in terms
of the primary physiologic processes (uptake, metabolism, distribution, elimination) as well as
the toxicologic processes. The mathematical characterization of dose addition requires a
constant proportionality between the effective doses of the two chemicals. This means that, for
equal effects, the dose of chemical B is a constant multiple of the dose of chemical A. The dose-
response functions are then congruent in shape. Let t be the proportionality constant that denotes
the relative effectiveness of chemical B to chemical A as estimated by the ratio of their iso-
effective doses, e.g., their ED10s. Let px and p2 be response measures and f(d) and g(d) be the
dose-response functions for chemicals A and B, respectively. Then for doses dx and d2 of
chemicals A and B, respectively, we have:
p> =
P2 =g(d2) = f(t*d2) -
(4-2)
The last equation (4-2) illustrates dose addition by converting dose d2 into an equivalent dose of
chemical A and then using the dose-response function f of chemical A to predict the response.
For a mixture of the two chemicals, the mixture response p,^ is then given in terms of the
equivalent dose and dose-response function for chemical A:
PMIX = f(di + t*di) (4.3)
Among the many ways to decide dose-addition, the isobole is one of the more common
graphical methods (see Figure 4-1). The isobole for a two-chemical mixture is the graph of the
various combinations of doses (dl3 d2) at which a fixed response is observed (Gessner, 1995). In
other words, the x-coordinate is the dose of chemical A and the y-coordinate is the dose of
chemical B such that the joint exposure (dl3 d2) produces the fixed response. This means that for
all points plotted on the isobole, the same response occurs. For example, in Figure 4-1, the
straight-line isobole represents the mixture doses in mg/kg that elicit a 10% response in the test
animals. If a point, say (2000,50), is on the isobole, then the dose combination of 2000 mg/kg of
chemical A and 50 mg/kg of chemical B will yield a 10% response in the test animals. Note that
this decision tool can be applied to any fixed response measure, whether percent responding in a
-68-
-------
250
200
Dose 2
100
i i r
500 1,000 1,500 2,000
Dose 1
INTERACTION
— — - Antagonism
~— Dose addition
Synergism
Isobole for 10% response:
(Dose2)/(200) + (Dose 1)7(2,700) = 1
Equivalent linear formula:
Dose2 = 200 - Dosel*0.074
2,500 3,000
Figure 4-1. Isoboles for 10% response level of combination doses (dl, d2) of two
chemicals snowing the possible types of interaction.
group, deficit of functionality, severity of a lesion, or any measure of toxicity that is constant
along the isobole.
When the set of equal-response points is a straight line, the two chemicals are said to be
dose-additive. Although in Figure 4-1 the other two isoboles show clear curvature, in many plots
the nonlinearity is less obvious. Statistical methods exist that help in deciding whether the points
indicate a departure from dose additivity (Gennings, 1995), and their use is strongly
recommended. Note that in the simple "clone" definition of dose addition, all isoboles for
different response rates will be parallel. Other more general definitions of dose addition have
also been proposed (Svendsgaard and Greco, 1995), including where the lines for different
response rates are linear, but not parallel (Svendsgaard and Hertzberg, 1994). When reviewing
the literature for evidence supporting dose addition, the assessor should ensure that the
definitions and assumptions are consistent with those used in this document. Foremost is that the
isoboles should be linear. Second, unless the isoboles for a wide range of response levels are all
parallel, the reported dose combinations used in generating the isobole should be comparable to
the environmental doses being assessed. If the published isoboles only reflect doses associated
with unacceptable toxicity (e.g., LD10s) or exposure levels much higher than the environmental
-69-
-------
levels of concern, then justification must be given for extrapolating the dose-addition property to
the lower environmental levels.
Recent work has demonstrated the issues that must be considered when assuming dose
addition (Feron et al., 1995). Feron and colleagues tested various simple mixtures (n=4 or 9
components) at levels near the no-observed-adverse-effect levels (NOAELs). Studies in their
laboratory on mixtures of chemicals with different target organs, or same target organ but
different toxicity modes of action, showed interactions when chemicals were at their minimum-
observed-adverse-effect levels (MOAELs), and no effects when component chemicals were at
1/10 or 1/3 their respective NOAELs. Mixtures of chemicals with the same target organ (kidney)
and similar toxic modes of action showed consistency with dose addition when each chemical
was at or slightly below its NOAEL. Similarity of toxic modes of action is then stronger support
for dose addition than is similarity of target organs. When exposures are near the NOAELs of
the components, target organ similarity seems to be sufficient justification for dose addition.
Three component methods are discussed in this document that are based on dose addition:
the RPF method, the TEF method, which is a special case of the RPF method, and the HI
method. They differ in the required knowledge about toxicologic processes and in the extent
over which toxicologic similarity is assumed. In each method, the exposure levels are added
after being multiplied by a scaling factor that accounts for differences in toxicologic potency
(also called toxic strength or activity).
The RPF method uses empirically derived scaling factors that are based on toxicity
studies of the effect and exposure conditions of interest in the assessment. When extensive
mechanistic information shows that all the toxic effects of concern share a common mode of
action, then one scaling factor is derived for each chemical that represents all toxic effects and all
exposure conditions. This special case is the TEF method, where actual toxicologic equivalence
between the component chemicals is assumed once the scaling factor is applied. When data are
conflicting or missing, or indicate that different modes of action may apply to different effects or
exposure conditions, separate factors may be derived for each effect or exposure condition,
which are distinguished from the special TEFs by being called RPFs. In the general RPF and
specific TEF methods, the scaling factor represents the toxicity relative to the toxicity of one of
the chemical components, called the index chemical, which is usually the best-studied chemical.
The mixture exposure, given by the sum of the scaled exposure levels, is then the equivalent
exposure in terms of the index chemical. This equivalent exposure is the exposure level of the
index chemical that elicits the same response as the mixture exposure. The risk assessment then
evaluates the equivalent index chemical exposure on that chemical's dose-response curve in
order to predict the mixture response.
-70-
-------
The Hazard Index method has weaker assumptions and data requirements, is more
generally applicable, and has more uncertainty in the resulting assessment. Instead of requiring
knowledge of similar mode of action, the Hazard Index method requires only similarity in target
organ. As with the general RPF method, a separate Hazard Index is determined for each target
organ of concern. Instead of converting the component exposure levels into an equivalent index
chemical exposure, the scaling factors are standardized so that the resulting sum is
dimensionless, and the Hazard Index is interpreted by whether or not it is greater than 1. The
scaling factors for the Hazard Index are based only on each component's toxicity, preferably
related to the target organ being assessed so that the interpretation of the Hazard Index value can
be tied to the target organ risk. For example, if the ED10 for liver effects is used (so that 1/ED10 is
used as the toxicity scaling factor), then when HI=1, the mixture is at its ED10 for liver toxicity.
Similarly, if some estimate of a practical threshold exists for each component, then HI=1
indicates that the mixture is at its practical threshold. The scaling factors for the Hazard Index
method should then be defined so that the resulting interpretation of HI=1 allows a clear risk
assessment interpretation for the mixture. In previous EPA applications of the Hazard Index
method, the Hazard Index has served only as a decision index, where HI>1 leads to more
investigation or to remedial action. If enough information becomes available on the components
to assume a similar toxic mode of action, then RPFs could be developed instead.
4.1.1.2. Response Addition
Under response addition, the chemicals are assumed to behave independently of one
another, so that the body's response to the first chemical is the same whether or not the second
chemical is present. In simplest terms, classical response addition is described by the statistical
law of independent events, with "response" measured by the percentage of exposed animals that
show toxicity. Using the same notation defined above for Equations 4-1 through 4-3, the
statistical law of independence is, for two chemicals:
PMIX = 1 - (1 - pi)*(l - pi) (4.4)
In terms of mixture response, this equation says that the response to either chemical A or B is 1
minus the probability of not responding to either chemical. Expanding the right-hand-side, one
obtains:
PMIX = pi + pi - pi* pi \ ~ )
which, for small single-chemical responses and only two chemicals in the mixture, is well
approximated by the simple summation:
-71-
-------
PMIX = pi + p2 (4-6)
The generalization of Equation 4-4 to an arbitrary number (n) of chemicals is:
PMIX = 1 - (1 - pi)*(l - p2)*. . . *(1 - p.)
(4-7)
Unless the number of mixture components is small and the individual risks are very small,
Equation 4-7 should be used for the response addition mixture estimate.
Response addition has also been reported where "response" is a measured effect (Ikeda,
1988), but no publications have been located that explain this approach in any detail. The
component effects are numerically added to give an estimated measured effect for the mixture.
For example, if 20 mg/kg of chemical A causes a 5% increase in liver weight and 30 mg/kg of
chemical B causes a 3% increase, then the prediction for a mixture of 20 mg/kg of A and
30 mg/kg of B is a liver weight increase of 8%. The simple summation implies that each
component effect is small so that the effects caused by different components are not influenced
by each other. Because this "effect addition" is not well characterized or investigated, this
approach is not recommended for general use at this time. Any risk assessment based on effect
addition should be restricted to the specific effects and dose ranges given in the supporting
studies.
Several variations of response addition have been developed (see U.S. EPA, 1986,
Appendix B). Some of these variations require additional information and assumptions. When
reviewing the literature for evidence supporting response addition, the assessor should ensure
that the definitions and assumptions are the same as those used in this document, or at least that
the interpretations are consistent with the procedures in this guidance document.
4.1.1.3. Low-Dose and Low-Response Risk Assessments
One of the important differences between risk assessment for individual chemicals vs. a
mixture assessment occurs when exposure levels are below the risk criteria values for the
individual components of the mixture. The individual chemical assessments, performed
separately, would conclude that none of the chemicals poses a significant risk. If the mixture
contains several lexicologically similar chemicals with no evidence of interaction, then dose
addition would be applied and the higher combined mixture dose could lead to an assessment of
significant risk of toxic effects.
If the mixture contains only lexicologically dissimilar chemicals, then response addition
would usually be applied because of the assumption of independent action. For example,
consider the case where decreasing the exposure reduces the probability of an effect, but not its
-72-
-------
severity (as EPA traditionally assumed for carcinogens). Simultaneous exposure to several of
these chemicals could then accumulate many small risks and be unacceptable in combination
even though the individual risks were acceptably small.
In contrast, consider the case where decreasing exposure results in a decrease in toxic
severity so that there is a practical threshold below which the effects are considered nonadverse.
If these chemicals are lexicologically independent and at individual exposure levels below their
respective practical thresholds, then an assessment of simultaneous exposure to several of these
chemicals may conclude there is no significant risk. This conclusion is plausible not only
because of the very low percent response for each chemical, but also because the intensity of the
effect decreases with dose.
In some cases, the sensitivity or resolution of the toxicity test may be worse as exposure
level decreases. In such cases, the exposure level labeled as an apparent toxicity threshold may
only reflect the reduced ability to discern that dose-related toxicity has occurred. Any risk
assessment based on evaluations near these practical thresholds should reflect the uncertainty
caused by the reduced sensitivity or resolution of the underlying toxicity tests. When
quantitative corrections are not possible, the risk characterization must include these study
weaknesses in the discussion of uncertainties.
4.1.1.4. Evidence for Dose or Response Additivity
Several studies have been published that suggest that dose or response additivity
adequately characterizes mixture risk. The large variety of possible mixtures, however, precludes
any strong characterizations of the accuracy of additivity methods. Some sense of the opinion of
toxicologists, however, can be gained from some key publications, in which dose or response
addition is recommended as a plausible default procedure. Ikeda (1998) surveyed the literature
and found few cases, by his judgment, that showed "clear-cut cases of potentiation" and he
concluded (p. 418): "Thus, the most practical approach in evaluating the combined effect of
chemicals seems to be the assumption of additive effects." He also noted that assuming
additivity of effects for chemicals with dissimilar modes of action is more protective than
independence. Furthermore, except for their initial overview, Plaa and Venzina (1990) focus on
concentration (i.e., dose) addition. The NAS book (NRC, 1988, p. 100) on complex mixtures is
less precise. NAS notes that "no-interaction" in its Chapter 1 is dose addition, while in its
discussion of ordinary linear statistical models, no-interaction refers to response addition. The
original U.S. EPA guidelines for mixture risk assessment (U.S. EPA, 1986) (Appendix A)
recommend default no-interaction approaches of dose addition for nongenotoxic toxicants acting
by similar modes of action or affecting common organs, and response addition for carcinogenic
risk.
-73-
-------
Reviews of toxicologic interaction studies do not often evaluate additivity, or are not able
to develop general conclusions. In too many cases, a study was not designed properly for
detecting departures from additivity. For example, in a review of statistical methods in 462
interaction studies (U.S. EPA, 1990), roughly one-third of the reported results indicated no
interaction or some kind of additivity, but nearly half of the studies used no statistical analysis or
did not report what procedures were used in determining statistical significance. As a result, it is
presently difficult to guess how common some kind of additivity might be for pairwise
interactions.
The decision to use dose addition and response addition as default "no-interaction"
definitions is primarily based on scientific plausibility when their assumptions are met (i.e., toxic
similarity for dose addition, independence for response addition). In addition, these default
approaches have clarity, simplicity, and ease of implementation. The evidence for either dose
addition or response addition as a good approximation for a mixture risk assessment is not
strong, and clearly is not comprehensive in representing the varying types of chemicals
considered in environmental risk assessment. Whenever evidence exists that clearly disagrees
with both dose and response addition, then alternative approaches should be considered, such as
those presented later that incorporate data on pairwise interactions.
4.1.2. Toxicologic Interactions
Regulatory decisions usually involve the assessment of chemical mixtures, though often
on a chemical-by-chemical basis. Typical exposures, in contrast, are composed of a combination
of biological, chemical, and physical agents that may influence each other's adverse effects.
Several quantitative descriptions of interaction have been proposed during the past 50 years.
Plaa and Venzina (1990) provide a historical overview of the differences in definitions, and
Kodell and Pounds (1991) discuss some of the implications of these differences. One of the
earliest quantitative characterizations of interactions was by Bliss (1939): similar joint action,
independent joint action, and synergistic or antagonistic joint action. Plaa and Venzina (1990)
propose the terms additive (sum of individual effects, an admittedly vague definition), infra-
additive, and supra-additive as having the advantage of not requiring consideration of
mechanisms. Table B-2 (Appendix B) recommends a set of definitions for use in chemical
mixture risk assessment. It clarifies the terminology related to additivity and interaction effects
for both cancer and noncancer endpoints.
Types of interactions among mixture components that can affect toxicologic response to
the whole mixture include chemical-to-chemical, toxicokinetic, and toxicodynamic interactions
(see Appendix C). The impact of the joint exposure on toxicologic response can be additive
(e.g., dose-additive, where chemicals act as dilutions of each other and cause toxicity by the same
-74-
-------
mode of action), less-than-additive (e.g., dietary zinc that inhibits cadmium toxicity through
toxicokinetic interactions that reduce the amount of dietary cadmium absorbed), or greater-than-
additive (e.g., enhanced carcinogenicity for asbestos and tobacco smoke). It must be emphasized
that antagonism is not the same as inhibition. Antagonism only implies a lesser joint response
than predicted from dose addition. Presence of antagonism does not justify lowering of risk
estimates of an affected chemical, say by increasing its Reference Dose. An antagonistic
chemical is also toxic. In contrast, the inhibitor chemical is not toxic by itself, but does reduce
the toxicity of the second chemical. Only for inhibition could risk levels for the second chemical
be adjusted because of reduced toxicity. Additional information and examples of data on
interactions can be found in Appendix C.
Interaction effects may result from events taking place at many possible loci in the body,
including the site of toxic action or during the processes of absorption, tissue distribution,
metabolism, excretion, or repair. Any or all of these can vary with route of administration, age,
sex, health, nutritional status, etc. With the almost infinitely large number of chemical mixtures
in the environment, systematic studies relevant to the toxicology of these chemical mixtures
using conventional methodologies and approaches are impossible; the development of predictive
and alternative toxicology methods is imperative. An evolving approach is the utilization of
physiologically based pharmacokinetic/pharmacodynamic (PBPK/PD) modeling, coupled with
model-oriented toxicology experiments (Tardif et al., 1997). Tissue dosimetry at the PK and PD
levels is achievable with simple and complex, but chemically defined mixtures. Further
discussions pertinent to the available PBPK/PD modeling and the metabolic processes have been
presented in Appendix C.
Evidence of toxicologic interaction should be reflected in the mixture risk assessment
(U.S. EPA, 1986). Previous risk assessments of multichemical exposures by EPA have
considered the information on interactions only in a qualitative sense. For example, a Superfund
site may receive more scrutiny or its remediation may proceed faster if there were several
indications of potential synergism among the detected chemicals. The cleanup goals and the
estimated risk, however, would not change. Consequently, most mixture risk assessments do not
include interactions information. No standard methods are yet in place in regulatory agencies to
incorporate interactions and no biologically motivated mathematical models have been
developed that could serve as a default method. The method described in this chapter is new. Its
use is encouraged so that EPA can gain experience regarding the difficulties and advantages of
an interaction-based approach and then identify ways to improve the approach.
In developing an interaction-based risk assessment method, the following constraints
were established:
-75-
-------
• the method should use readily available data, or at least information that can be
feasibly obtained.
the method should include several steps, each of which could be modified or
replaced when more data or biological models became available.
• the method should be plausible, either supported by some empirical cases or
supported by consensus among practicing mixtures toxicologists and risk
assessors.
4.1.3. Risk Assessment Strategy
Approaches based on the mixture's chemical components are recommended for relatively
simple, identified mixtures with approximately a dozen or fewer chemical constituents. For
exposures at low doses with low component risks, the likelihood of significant interaction is
usually considered to be low. Interaction arguments based on saturation of metabolic pathways
or competition for cellular sites usually imply an increasing interaction effect with dose, so that
the importance at low doses is probably small. The default component procedure at low
exposure levels is then to assume response addition when the component toxicological processes
are assumed to act independently, and dose (or concentration) addition when the component
toxicological processes are similar. For dose (concentration) addition, a specific Hazard Index
procedure is recommended. For higher exposure levels, or when adequate data on interactions
suggest other than dose or response additivity at low doses, such information must be
incorporated into the assessment. Specific procedures are recommended for interactions based
on the available data (Sections 4.4 and 4.5).
4.1.4. Cautions and Uncertainties With Component-Based Assessments
The component-based procedures discussed earlier for dose-response assessment and risk
characterization are intended only for simple mixtures of a dozen or so chemicals. The
uncertainties and biases for even a small number of chemical components can be substantial.
Component-based methods are particularly susceptible to misinterpretation because the listing of
chemical components in a mixture is often misconstrued as implying a detailed understanding of
the mixture toxicity and, by inference, the estimated mixture risk. The risk characterization must
include a discussion of what is known as well as what is missing or poorly understood in order to
convey a clear sense of quality and confidence in the risk assessment.
4.1.4.1. Exposure Uncertainties
The general uncertainties in estimating mixture exposure are addressed in the Agency's
guidelines related to exposure assessment (U.S. EPA, 1992). The risk assessor should discuss
these exposure uncertainties in terms of the strength of the evidence used to quantify the
-76-
-------
exposure. When appropriate, the assessor should also compare monitoring and modeling data
and discuss any inconsistencies as a source of uncertainty. For mixtures, these uncertainties may
be increased as the number of compounds of concern increases.
If levels of exposure to certain compounds known to be in the mixture are not available,
but information on health effects and environmental persistence and transport suggests that these
compounds are not likely to be significant in affecting the toxicity of the mixture, then a risk
assessment can be conducted based on the remaining compounds in the mixture, with appropriate
caveats. If such an argument cannot be supported, no final risk assessment can be performed
with high confidence until adequate monitoring data are available. As an interim procedure, a
risk assessment may be conducted for those components in the mixture for which adequate
exposure and health effects data are available. If the interim risk assessment does not suggest a
hazard, there is still concern about the risk from such a mixture because not all components in the
mixture have been considered.
In perhaps a worst-case scenario, information may be lacking not only on health effects
and levels of exposure, but also on the identity of some components of the mixture. Analogous
to the procedure described in the previous paragraph, an interim risk assessment can be
conducted on those components of the mixture for which adequate health effects and exposure
information are available. If the risk is considered unacceptable, a conservative approach is to
present the quantitative estimates of risk, along with appropriate qualifications regarding the
incompleteness of the data. If no hazard is indicated by this partial assessment, those partial
results should be conveyed to the risk manager, but the risk assessment should not be quantified
until better health effects and monitoring data are available to adequately characterize the
mixture exposure and potential hazards.
4.1.4.2. Dose-Response Uncertainties
For many simple mixtures for which a component-based approach might be applied,
studies on interactions, even pairwise interactions, will be missing. Use of a dose- or response-
additive model is easily implemented, but justification for such approaches is largely based on
conceptual arguments, not empirical studies. In the review cited previously on available
interaction studies (U.S. EPA, 1990), statistical tests were used to decide the presence of
toxicologic interaction, but dose-response models for interactions were extremely rare. For
example, of the 462 studies reviewed, only four gave a prediction under no interaction (using
response addition as the default). As indicated previously, recent studies by Feron et al. (1995)
show that there are exceptions to most rules regarding interactions, even the common assumption
that additivity is acceptable if chemicals target the same organ. Recent studies on dose additivity
have focused on very simple mixtures of chemically and metabolically similar chemicals
-77-
-------
(Gennings et al., 1997; Simmons et al., 1995). Improvements in experimental design and
statistical hypothesis testing for dose additivity, along with better understanding of the chemical
characteristics that accompany observed dose additivity, should lead to improved predictive
ability and justification for dose addition as a default approach.
Conclusions regarding toxicologic interaction are also only weakly supported by
empirical studies. Based on a review of EPA's Mixtox database (U.S. EPA, 1990), reflecting
437 articles on interactions between pairs of environmental chemicals, many studies failed to
identify what the "no-interaction" hypothesis was, so that any conclusions regarding nonadditive
interaction were difficult to interpret. Other studies identified the no-interaction hypothesis, but
employed incorrect experimental designs, so that the conclusions were questionable. Perhaps the
most substantial weakness in the understanding of toxicologic interactions is the lack of studies,
models, and concepts for interactions involving more than two chemicals. The key assumption
in both of the interaction methods described in Section 4.3 (Mumtaz and Durkin, 1992;
Hertzberg, 1996) is that, at least for low doses, the resulting influence of all toxicologic
interactions in a mixture is well approximated by the pairwise interactions. No studies have been
located to date that investigate that assumption, although two studies are in progress at EPA and
ATSDR.
Toxicologic understanding of interaction is also limited. Although interaction modes of
action are commonly assumed to involve either pharmacokinetics and metabolism or toxicologic
receptors, nearly all studies on mechanisms and modes of interaction focus on pharmacokinetics
(El-Masri et al., 1995). Current pharmacokinetic models for interactions usually address two- or
three-chemical mixtures. Clearly, more research on complex interactions is necessary to improve
risk assessment interactions information.
4.1.4.3. Presenting Component-Based Risk Characterization
The consequence of this early stage of mixture risk research is that the risk assessor must
use considerable judgment along with plausible approaches. The results, however, must be
presented transparently. Although the procedures described in this chapter are developed from
available concepts and data on simple mixtures, all component-based quantitative mixture risk
assessments should be limited to one significant digit for the risk value, unless substantial
justification is given for higher precision.
Mixtures composed of chemicals with RfDs or RfCs must be assessed and presented
carefully. A common interpretation is that mixtures with few components, each less than its RfD
or RfC, pose no significant risk. As discussed above, for lexicologically similar chemicals this
conclusion can be in error because the joint exposures contribute to the same potential toxicity
and effectively represent a cumulative dose; thus a dose-additive assessment should be
-78-
-------
performed. For a mixture of a few dissimilar chemicals, where an assessment is based on
response addition, the mixture risk would likely be judged negligible, particularly if the effects
supporting the RfDs and RfCs are minor. When the toxic effects are of major concern, such as
cancer or developmental toxicity, the estimated mixture risk should be judged in the context of
the effects, the shapes of the dose-response curves, and the characteristics of the exposed
population.
Whenever an assessment is based on component toxicity values, the risk characterization
must discuss the quality of the individual chemical estimates that are used. For example, RfDs
and RfCs differ in quality, as reflected by the variation in their uncertainty factors and the
confidence statements listed in the IRIS files. The cancer potency values also have uncertainty,
as reflected as subjective choices in modeling (e.g., significance levels for inclusion of model
terms, confidence levels for creating interval estimates, levels for deciding adequate goodness-of-
fit), as well as by qualitative descriptors of the weight of evidence that the chemical is a human
carcinogen. All these measures of uncertainty and unevenness of component estimates must be
described, at least in summary fashion, in the risk characterization.
4.2. HAZARD INDEX
4.2.1. Definition
The primary method for component-based risk assessment of lexicologically similar
chemicals is the Hazard Index (Teuschler and Hertzberg, 1995), which is derived from dose
addition (Svendsgaard and Hertzberg, 1994; also see Sections 2.6.1 and 4.1.1). In this guidance
document, dose addition is interpreted as simple similar action (Finney, 1971), where the
component chemicals act as if they were dilutions or concentrations of each other differing only
in relative toxicity. Dose additivity may not hold for all toxic effects. Further, the relative toxic
potency between chemicals may differ for different types of toxicity or toxicity by different
routes. To reflect these differences, the Hazard Index is then usually developed for each
exposure route of interest, and for a single specific toxic effect or for toxicity to a single target
organ. A mixture may then be assessed by several His, each representing one route and one toxic
effect or target organ.
The Hazard Index is defined as a weighted sum of the exposure measures for the mixture
component chemicals. The "weight" factor according to dose addition should be a measure of
the relative toxic strength, sometimes called "potency." Because the Hazard Index is tied to dose
addition, each weight factor should be based on an isotoxic dose. For example, if the preferred
isotoxic dose is the ED10, then the Hazard Index would equal the sum of each chemical's
exposure level divided by its ED10 estimate. The goal of a component-based quantitative mixture
assessment is to approximate what the mixture value would be if the whole mixture could be
-79-
-------
tested. For example, a Hazard Index for liver toxicity should approximate the concern for liver
toxicity that would have been assessed using actual toxicity results from exposure to the whole
mixture.
4.2.2. Information Requirements
Empirical evidence for dose addition includes similarly shaped dose-response curves of
the component chemicals, or identical dose-response curves when the doses are scaled for
relative potency as well as straight line isoboles (see Section 4.1.1 for other definitions and for
more background information). When the response involves quantal data on the number of
animals (people) responding, the evidence for dose addition can also include parallel log dose-
probit response curves of the component chemicals. Dose addition can also be demonstrated by
statistical comparisons of the observed mixture response with the estimated response derived
from dose addition, although this evidence may not apply to doses other than those tested. The
biological basis for dose addition is the similarity of chemical components regarding toxicologic
behavior, such as toxic mechanism, mode of action, or endpoint. When external exposure levels
are used in place of internal dose, then the similarity judgment also includes physiologic
disposition (uptake, metabolism, pharmacokinetics, etc.).
The Hazard Index method is specifically recommended only for groups of lexicologically
similar chemicals that all have dose-response data. In practice, because of the common lack of
information on mode of action and pharmacokinetics, the requirement of toxicologic similarity is
usually relaxed to that of similarity of target organs (U.S. EPA, 1989a). Additional information
on mode of action or on other factors that could affect tissue exposure (e.g., deposition pattern in
the nose) should be reviewed to ensure that dose additivity is appropriate. When evidence
indicates independence of action for low to moderate exposure levels, i.e., at doses near the
individual chemical NOAELs, response addition should be used (see Sections 2.6.2 and 4.5).
Any approach not based on dose addition must be clearly described, and the evidence for
applicability at low doses must be presented.
4.2.3. Alternative Formulas
The Hazard Index can be determined in several ways, depending on the available data and
on the interpretation of risks that is desired. The formula must represent dose addition as a sum
of exposures scaled by each chemical's relative toxicity. The only constraint is that the units of
exposure and relative toxicity should cancel, so that each term and the resultant index are
dimensionless. Clearly, all scaling factors in the same Hazard Index should reflect the same
toxicity measure (e.g., 1/ED10). There is no commonly accepted standard measure of toxicity for
exposure levels associated with minimal toxicity, in contrast to the slope factor for cancer (when
-80-
-------
nonthreshold, low-dose linearity is assumed) or the LD50 for lethal levels. To ensure consistency
with other EPA guidance on risk assessment, lethal dose data are not recommended for use in
mixture risk assessment. The approach taken in the 1986 mixture guidelines (U.S. EPA, 1986)
(Appendix A) for the scaling factors in the Hazard Index is to use the inverse of an acceptable
level (AL). The alternatives presented in this section use different toxicity-specific doses for AL.
The guidelines formula for the Hazard Index is then quite general:
HI = £ A (4-8)
t=iAL.
where
E = exposure level,
AL = acceptable level (both E and AL are in the same units), and
n = the number of chemicals in the mixture.
In practice, EPA risk assessors have usually calculated the Hazard Index by using the RfD or RfC
as the AL (U.S. EPA, 1989a). For example, for oral exposures:
(4-9)
where
E; = daily oral intake of the ith chemical, and
RfD; = EPA Reference Dose for the ith chemical.
Each term in Equation 4-9 is called a hazard quotient (U.S. EPA, 1989a) and represents
that chemical's contribution to the toxic endpoint of concern. This equation applies to oral
exposures. For the inhalation route, the exposure measure is the ambient air concentration and,
instead of the RfD, the AL is the RfC (U.S. EPA, 1994a).
By modifying the above formula, one can utilize other expressions for exposure and
relative toxicity that may be more appropriate for different situations. For example, for a Hazard
Index representing subchronic exposures, the appropriate subchronic data should be used, both
for the exposure estimate and the AL. To ensure clarity of interpretation, the scaling factors, AL,
-81-
-------
should be carefully documented and the resulting subchronic Hazard Index must be clearly
identified as representing the shorter term exposure.
The use of an acceptable level in the relative toxicity scaling factor (e.g., 1/RfD) may be
overly health protective in that the RfD (or RfC) is based on the critical effect, defined as the
toxic effect occurring at the lowest dose. When the Hazard Index is calculated for some
different, less sensitive effect, the RfD will be too low, so the factor (1/RfD) will overestimate
the relative toxicity and the Hazard Index will be too large. One alternative that avoids this
critical effect conservatism is to use a toxicity-based exposure level that is specific to the target
organ of interest and is derived similarly to an RfD (or RfC). For oral exposures, this value is
called the target organ toxicity dose or TTD (Mumtaz et al., 1997). The formula for the Hazard
Index would be identical to Equation 4-9, with the TTD replacing the RfD. For inhalation
exposures, a similarly defined target organ toxicity concentration (TTC) could be used. This
same approach can be applied to His for shorter exposures by using the effect-specific data
appropriate to the shorter exposure period of concern.
The TTD is not a commonly evaluated measure and currently there is no official EPA
activity deriving these values, as there is for the RfD and RfC. This alternative should be
considered when there is sufficient reason to believe that the overestimate of the Hazard Index
caused by use of RfDs is significant to the interpretation of the mixture assessment. In that case,
TTDs can be derived for the mixture components of interest by following the scientific steps
used in deriving an RfD. The evaluation of quality of the candidate toxicity studies and the
choice of uncertainty factors should parallel those steps in the RfD process. One difference in
the uncertainty factors concerns the factor for completeness of the database used for RfD
development. For example, if no two-generation study existed for a chemical, there could be an
additional uncertainty factor used to obtain the RfD because the RfD must protect against all
toxic effects. When developing a renal TTD, however, no additional factor would be used
because the data would only include renal effects (Mumtaz et al., 1997).
Any TTDs derived for a mixture assessment must be clearly documented, including the
array of studies considered, the study and dose selected for calculation purposes, and the
uncertainty factors chosen. When the critical effect of a chemical is the effect being described by
the HI, the RfD and TTD will apply to the same target organ and so should be the same unless
the TTD is based on newer information. When data for one or more components are not
sufficient for deriving their organ-specific TTDs, their RfDs should be used and noted as a
source of possible overestimation of the HI. This discussion and recommendations also apply to
His for shorter exposures, and to TTCs as replacements for RfCs in an Hazard Index for
inhalation exposures.
-82-
-------
Example. Consider a mixture of six chemicals, with data given in Table 4-1. When data
were not sufficient for deriving a TTD, the RfD was used as a surrogate. There were several
instances, however, where the critical effect of a component was the effect of concern, so the
TTD and RfD were the same. This example illustrates that, for some endpoints, the substitution
of the TTD will produce a Hazard Index value that is significantly less than the Hazard Index
based on RfDs alone, while for others the difference is minor. In this example, the Hazard Index
for reproductive effects changes from 3 to 1 by substituting the TTDs for the RfDs, whereas the
Hazard Index for renal effects only changes from 2 to 1. See Mumtaz et al. (1997) for more
complete discussion of this and other examples.
These two Hazard Index methods, by using a TTD or RfD, have a quantitative weakness.
The relative toxicity scaling factor (e.g., 1/RfD) is calculated from an experimental data point
(e.g., the highest NOAEL). As a result, the use of small experimental dose groups could produce
no significant response (the NOAEL) solely because of the low capability to detect the effects
(i.e., lack of statistical power), thereby overestimating the NOAEL and underestimating the
scaling factor. In addition, because the scaling factor is tied to actual experimental doses, wide
dose spacing limits the measure's precision.
A different approach to determining relative toxicity is to calculate a benchmark dose or
benchmark concentration (BMD/C) for the target organ of interest (U.S. EPA, 1996d). To
illustrate, consider oral exposures. The BMD approach entails identifying a dose (e.g., the ED10)
associated with a particular benchmark risk or magnitude of response (e.g., 10%) for the effect of
concern and involves statistically fitting a dose-response model to the toxicity data. For most
mixtures, however, the available dose-response data for the different component chemicals will
be based on different conditions, such as differences in exposure duration or test species. The
Hazard Index can use these BMDs only if some sort of standardization is applied so that the
1/BMD scaling factors describe a common scenario.
For example, if all component chemicals had chronic dose-response data on humans, then
the data are already consistent and the Hazard Index would use 1/BMD for each relative toxicity
scaling factor. The mixture risk could then be interpreted fairly precisely. When the HI=1, the
mixture is at its BMD. If the BMD is defined as the ED10, then when HI=1, the mixture exposure
should produce a 10% response (see Section 4.2.6, Equation 4-12).
When the chemical components do not have similar dose-response scenarios, some other
method must be used to standardize the BMDs. An obvious approach is to use uncertainty
factors and derive a TTD from each BMD, and then use 1/TTD for the scaling factor.
-83-
-------
Table 4-1. Example application of the target-organ toxicity dose
Chemical
Acetone
Chloroform
Dibutyl
phthalate
Diethyl
phthalate
Di(2-ethyl-
hexyl)
phthalate
Phenol
HI-RfD
HI-TTD
Hepatic TTD
l.OOE-01
RfD
l.E-02
RfD
NA
NA
2.E-02
RfD
NA
1.5
1.5
Renal TTD
l.OOE-01
RfD
l.E-01
TTD
NA
NA
2.E-02
RS
2.E+00
TTD
2.0
1.2
Reproductive
TTD
NA
NA
2.E-01
TTD
5.E+00
TTD
5.E-02
TTD
NA
2.7
0.8
Oral exposure
(mg/kg per day)
4.E-02
5.E-03
8.E-02
l.E+00
l.E-02
3.E-01
RfD (mg/kg
per day)
l.E-01
l.E-02
l.E-01
8.E-01
2.E-02
6.E-01
HQ
0.40
0.50
0.80
1.25
0.60
0.50
Critical effect
Renal, hepatic
Hepatic
Incr. mortality
Growth
Hepatic
Developmental
oo
In the TTD columns, the source of the value is coded as:
TTD: new TTD developed for this effect.
RfD: this is the critical effect, so the TTD=RfD.
RS: insufficient data for a TTD, so RfD used as a surrogate.
TTDs and RfDs are from Mumtaz et al. (1997). Exposure levels (dose) are set for illustration only.
-------
4.2.4. Comparison of the Hazard Index Formulas
The four approaches to calculating the Hazard Index can be compared by whether they
have various desirable characteristics. None of the approaches possesses all the desirable traits,
so the preferred method will need to be judged for every application.
One of the key desirable features is the constraint to use only data on the effect of
concern. Because the Hazard Index is tied to a specific effect, the underlying data should be on
that effect. Substituting data on the critical effect introduces an unknown degree of
conservatism, so that the Hazard Index is inflated by an unknown amount.
Another desirable characteristic is the use of statistical analysis on the entire dose-
response study data, e.g., to generate a BMD. Statistical analysis of the dose-response data
allows quantification of uncertainly and reflects more information by using the entire dose-
response data set. Restriction to an actual experimental dose, such as focusing on a single
NOAEL or LOAEL, ties the precision of the HI to the dose spacing used in the study. Also,
when only the actual exposure level is used, there is no reflection of its statistical uncertainty in
the HI calculation.
A third desirable characteristic is the constraint to use only data on humans for the
exposure scenario of concern. As more extrapolation is performed, such as using an uncertainty
factor to allow subchronic data to be used for a chronic risk assessment, the interpretation of the
HI becomes more vague. Uncertainty factors play an important role in standardizing the data so
that chemicals with different kinds of dose-response data can still be combined in the HI
calculation. Because uncertainty factors are judgmental, not statistically derived scaling factors,
their accuracy and precision are difficult to quantify.
Finally, it is important to have ready access to the data required for the particular
approach. Whereas direct human dose-response data are preferred, they are rarely available for
environmental chemicals. Similarly, although the TTD avoids the conservatism of the critical
effect, and may use fewer uncertainty factors than the RfD, there are no plans within EPA for
development of TTDs.
The four approaches can be summarized in Table 4-2. For easier comparison, only oral-
exposure nomenclature is used. For inhalation, each "D" (for oral dose) in the column headers
should be replaced by a "C" (for air concentration). BMD-hu refers to a BMD-based HI using
human data for the exposure scenario of concern. TTD-BMD refers to the TTD-based HI where
the TTDs use dissimilar BMDs that have been standardized by uncertainty factors.
The default procedure for the HI has traditionally been to use the RfD or RfC (U.S. EPA,
1989a). Because of their much wider availability than TTDs, standardized development process
including peer review, and official stature, the RfD and RfC are recommended for use in the
default procedure for the HI. When possible, the other methods should be employed, even if only
-85-
-------
Table 4-2. Comparison of HI methods
Feature
Toxic effect of
concern
Statistical
analysis of full
dose-response
data set
Species and
exposure
scenario of
concern
Easily available
data
BMD-hu
yes
yes
yes
no
TTD-BMD
yes
yes
no
not much
TTD
yes
no
no
some
RfD
not usually
no
no
yes
for some of the mixture components, to allow at least partial characterization of the uncertainty
and conservatism introduced by use of the RfD or RfC.
The mixture components to be included in the HI calculation are any chemical
components showing the effect described by the HI, regardless of the critical effect upon which
the RfD/C is based. If the effect of concern is different from the RfD's or RfC's critical effect,
the relative toxicity scaling factor for that chemical will be an overestimate, and the discussion of
the resulting HI must include a qualifying statement that notes the potential conservatism. For
shorter term exposures, the appropriate data and calculations should be used as described in the
previous sections. Other modifications, including development and use of ad hoc TTDs, are
possible but should be justified in each case and should clearly describe the underlying data used
in the determination.
A separate HI should be calculated for each toxic effect of concern (U.S. EPA, 1986,
1989a). The target organs to be addressed by the His should be decided for each particular
mixture assessment. The assessor should compare the dose-response curves for the different
toxic effects with the estimated exposure levels (and routes) to ensure that those effects most
relevant to the environmental exposure are addressed. When certain toxic effects are known to
occur, but at much higher exposure levels than those being assessed, then the HI for those effects
may not need to be evaluated, but an explanatory note should be included in the discussion of
assumptions and uncertainties for the mixture assessment.
-86-
-------
4.2.5. Interpretation
The HI is a quantitative decision aid that requires toxicity values as well as exposure
estimates; it is then part of the risk characterization. When each organ-specific HI for a mixture
is less than 1 and all relevant effects have been considered in the assessment, the exposure being
assessed for potential noncancer toxicity is to be considered unlikely to result in significant
toxicity. When each HI is less than 1 but important information is missing or highly uncertain,
then the conclusion of unlikely toxicity is weakened, and the discussion of uncertainties must be
expanded appropriately. When the applicability of dose addition is also questionable,
particularly if there is some evidence of synergism among some of the component chemicals,
then an HI less than 1 should be viewed cautiously and consideration should be given to
developing an interaction-based HI (see Section 4.4).
When any effect-specific HI exceeds 1, concern exists over potential toxicity. Some
research suggests that concordance across species of the sequence of target organs affected with
increasing dose (e.g., the critical effect) and concordance of the modes of action are variable and
should not be automatically assumed (Heywood, 1981, 1983). Some effects, such as hepatic
toxicity, are more consistent across species, but more research is needed in this regard. The
specific target organ or type of toxicity that is of greatest concern for humans may not be the
same as that for which the highest HI is calculated from animal studies, and so specific effects
should not be inferred unless considerable empirical or mechanistic information exists supporting
that cross-species concordance. As more His for different effects exceed 1, the potential for
human toxicity also increases. This potential for risk is not the same as probabilistic risk; a
doubling of the HI does not necessarily indicate a doubling of toxic risk. A specific numerical
value of the HI, however, is usually assumed to represent the same level of concern regardless of
the number of contributing chemical components or the particular toxic effect that is being
tracked.
When human BMD/Cs are available, then HI=1 will be easily understood as representing
the benchmark risk level of the specified effect. Because HI=1 is often used as a decision
threshold in risk assessment, this benchmark risk should be carefully selected to represent the
boundary below which the effect is deemed not to be of concern. The most recent EPA
benchmark dose guidance should be used in making that selection.
No specific decision threshold is proposed for general application of the HI. Because the
RfDs (and by inference the TTDs) are described as having precision no better than an order of
magnitude, the HI should be rounded to no more than one significant digit. Concern should
increase as the number of effect-specific His exceeding 1 increases. The numerical magnitude of
the HI must be interpreted in the context of the supporting information. For example, as a larger
number of effect-specific His exceed 1, concern over potential toxicity should increase. Both
-87-
-------
large and small His should be reviewed for large uncertainties. Small His can be caused by
incomplete characterization of the mixture composition, by missing RfDs, or by missing
exposure levels for some chemicals. A large HI can be caused by a few chemicals whose RfDs
(or TTDs) are based on large uncertainty factors, or because RfDs are used in place of TTDs and
are based on some effect other than the one addressed by the HI. Whenever an HI is included in
a risk assessment, its value must be accompanied by a description of the quality and contribution
of the supporting information and of any data gaps.
4.2.6. Reference Value for a Mixture
When only component toxicity data are available and dose or concentration addition can
be assumed, knowledge of individual chemical RfDs can be used to determine the mixture RfD
(Svendsgaard and Hertzberg, 1994). One example of this is human consumption offish
(Dourson and Clark, 1990). Assuming stable exposure conditions, the mixture intake is then
determined by the amount offish eaten (i.e., total mixture dose), while the relative proportions of
mixture components are constant. A fish RfD can then be calculated as the level that represents
the intake offish (e.g., kg offish flesh per day) associated with minimal risk.
The calculations are straightforward (Mumtaz and Hertzberg, 1993) and represent dose
addition applied to the chemical components that show similar toxicity. The easiest approach is
to start with the zero-interaction equation (Berenbaum, 1989), here given for a mixture of two
chemicals, and using 0.05 as the fixed response for scaling the component doses:
J=d/D1 + d/D2 (4-10)
where:
d; = dose of ith chemical, and
D; = dose of ith chemical that produce the response of 0.05.
In Berenbaum's equation, each dose is scaled according to "doses isoeffective with the
combination." In this example, the "effect" is defined as a small response value, say 0.05. Then
the D; values are the respective ED05 values for the two components when exposure is to one
chemical at a time. If the component doses are such that Equation 4-10 is true, then the mixture
dose, dm = (dj + d2), is at its ED05, denoted here by Dm . This is determined by representing the
joint exposure by fractions (f;) of total mixture dose (i.e., d; = f;*Dm):
7=/;*A/A + //A/^ (4-n)
-88-
-------
Dividing by Dm gives:
l/Dm =//A +f/D2 (4-12)
and inverting gives the mixture ED05, again valid only for fixed proportions fx and f2.
A similar procedure can be used to determine the reference dose for the mixture (RfDm)
by interpreting the isoeffective doses to be RfDs (i.e., doses producing negligible risk of adverse
effects). If we invert Equation 4-12 and substitute the component RfDs for the component
ED05s, then we obtain:
RfDm = J/(f/RJD1 +^RfD2) (4-13)
Example. Let the single chemical data be:
Chemical 1 Chemical 2
RfD 20 35
Fraction in mixture 0.7 0.3
Then application of Equation 4-13 gives the mixture RfD as:
RfDm = l/( 0.7/20 + 0.3/35) = I/ (.044) = 23
The reference value for a mixture, such as an RfD, is reasonable only when certain
conditions occur. Most critical is that the mixture composition must be fairly constant so that
total mixture intake is the only important variable. If this requirement cannot be assured, then
the mixture reference value should not be calculated. Another condition is that the component
chemicals are similar, so that dose addition can be applied. When toxicologic similarity cannot
be assured, then either another formula must be derived, or the mixture must be tested as a whole
(see Chapter 3). If any other formula is employed, then it must be justified. Further,
genotoxicity and other no-threshold, low-dose-linear toxicity must be ruled out. The other
cautions regarding component-based risk characterization also apply (see Section 4.1.4).
One of the main limitations to accuracy of this mixture reference value is the use of
component reference values. While individually they have a common definition, they do not
have a common database. As noted in the discussion of the HI (Section 4.2), RfDs (and RfCs)
for different chemicals are derived separately, and often represent differing degrees of quality
and relevance. Interpreting the overall quality of the mixture RfD as the composite of several
variable-quality individual RfDs is a difficult process. In the extreme, when one component's
-89-
-------
reference value is clearly of marginal quality, as reflected by a high uncertainty factor and few
studies, the assessor should discuss the uncertainty and should consider presenting two mixture
reference values: one that incorporates reference values for all chemicals and one that excludes
the highly uncertain reference value.
4.3. INTERACTION-BASED HI
In the method described in this section, the key assumption is that interactions in a
mixture can be adequately represented as departures from dose addition (Hertzberg et al., 1999).
The method follows an obvious approach: to begin with the dose-additive HI, and then modify
its calculation to reflect the interaction results, using plausible assumptions to fill in the data
gaps. A secondary assumption is that the influence of all the toxicologic interactions in the
mixture can be adequately approximated by some function of the pairwise interactions.
4.3.1. HI Definition
4.3.1.1. Background
Toxicologic interactions have been mostly studied with binary mixtures. One way to
include interactions in a mixture assessment is to modify the noninteractive assessment by
knowledge of these binary interactions; a tacit assumption is then that higher order interactions
are relatively minor compared to binary interactions. Few studies quantify interaction, and even
fewer quantitatively describe the dose-dependence of the interaction. Consequently, for an
approach to be able to use available data, some qualitative procedure is needed for judging the
impact of the potential toxicologic interactions.
EPA previously developed a weight-of-evidence procedure that uses binary interaction
data to modify the HI (Mumtaz and Durkin, 1992; Mumtaz et al., 1998). This procedure
reflected the strength of the available interaction studies as well as the amounts of each
component in the mixture. The first step entailed a review of relevant information on all of the
possible binary interactions in the mixture. Among the several factors considered are the degree
of understanding of the interaction, its relevance to toxicity, and the extent of extrapolation to the
exposure conditions of interest (e.g., route and species conversions). The strength and
consistency of this evidence was then assigned a numerical binary weight-of-evidence
(BINWOE) score. The BINWOE was then scaled to reflect the relative importance of the
component exposure levels. A main property of the Mumtaz and Durkin approach is that the
scaled BINWOE decreases with decreasing exposure levels, reflecting a common observation
that the significance of interactions in a mixture decreases as the exposure and likelihood of
response decreases. This scaled BINWOE is then used to modify the dose-additive HI as
follows:
-90-
-------
(4-14)
where HI^Q is the noninteractive HI based on dose addition, UFj is the uncertainty factor for
interactions, and WOF^j is the scaled BINWOE.
The procedure outlined by Mumtaz and Durkin (1992) has been a major advance in the
risk assessment of chemical mixtures. The approach is quite feasible: it uses available
information along with toxicological judgment and reflects many general concepts about
toxicologic interactions. When the approach is tested for consistency of application (Mumtaz et
al., 1995), individuals and groups tend to develop fairly similar scores, though sometimes with
different rationale.
The weaknesses in the approach are few, but important. The guidance on selecting the
uncertainty factor for interactions is not given, the steps in determining the BINWOE are fairly
complex, and the magnitude of the interaction is not included. The relative weights applied to
the various categories of information lack support from empirical assessments of the influence
that some key experimental variables have on the interaction consistency. Further, the formula
itself (Equation 4-14) may be overly simple in that the interactions and additivity components are
separable; i.e., the interactions information is completely represented by the multiplicative factor
UFWOE, which is applied to the entire additive HI.
The recommended procedure incorporates several changes from the original developed
by Mumtaz and Durkin (1992). The main difference is seen in the formula (Hertzberg et al.,
1999). Instead of the additive HI (Equation 4-9 in Section 4.2) being modified by a single
composite interaction factor, each term is modified according to the influence (interaction) of the
other components, and then these modified terms are summed.
Consider the example of a HI for liver toxicity. The Hazard Quotient (HQ;) for the ith
chemical (U.S. EPA, 1989a) reflects that chemical's individual contribution to hepatic toxicity.
The interactions approach then considers two contributions to toxicity: the hepatic toxicity
resulting from a single chemical by itself, indicated by the value of HQ;, and the influence of all
the other chemicals' interactions affecting the liver. In many cases, direct measurement of
changes in liver toxicity will not be available. General changes affecting internal dose, such as
the bioavailability or pharmacokinetics of the chemical, can then be substituted (Krishnan et al.,
1994).
The need to focus on a single chemical's toxicity is illustrated by studies showing
asymmetric interactions. For example, the influence of chemical A on chemical B's toxicity may
be synergistic, while the influence of B on A's toxicity may only be dose additive. By having
two separate terms in the interaction-based HI, these differences are incorporated.
-91-
-------
Component exposure levels also can affect the nature and magnitude of the interaction.
The high-to-low dose extrapolation is particularly problematic for mixtures. Many dramatic
interactions occur at high exposure levels, e.g., the substantial synergism between tobacco
smoking and radon exposure. Several publications note the expectation that most high-dose
interactions will be minimal at very low doses. Examples that include the dose dependence of
the interaction, however, are sparse. Feron et al. (1995) discuss some examples where
interactions occur at exposures near individual minimal-observed-effect levels while only dose-
addition is apparent near individual no-effect levels; they do not present a quantitative relation
between interaction and dose. The influence of the relative proportions is also of concern. For
example, with respect to the loss of righting reflex in mice (Gessner, 1995), the ED50
isobologram for the interaction between ethanol and chloral hydrate shows synergism at low
ethanol levels, but concentration additivity at higher ethanol levels. One suggestion is that the
interaction should become less important as one chemical begins to dominate the mixture
toxicity.
4.3.1.2. Formula
The interaction-based HI includes two evaluations of the weight of the evidence (WOE)
for interaction for each pair of component chemicals in the mixture: one WOE for the influence
of chemical A on the toxicity of chemical B, and one for the reverse. This qualitative judgment
is then changed into a numerical score. Some common assumptions and desirable properties
could also be included:
(1) The pairwise interactions capture most of the interaction effects in the mixture.
(2) The interaction is highest when both chemicals in the interacting pair are at
equally toxic doses (neither chemical is dominant).
(3) The interaction-based HI must reduce to the dose-additive HI as the interaction
magnitudes decrease.
(4) The main toxicologic effects from the mixture exposure are limited to those
effects induced by the individual component chemicals.
(5) The interaction magnitude is likely to decrease as mixture dose decreases.
The WOE procedure modifies each HQ in the formula for HI. For the ith chemical, the
modification means multiplying HQ; by the sum of all the pairwise interaction contributions from
the remaining chemicals (thus the summation index is for all i not equal to j). This multiplier is
(each term is described below):
-92-
-------
The full modified formula for the interaction-based HI, HI^ , is then:
t * f /, M?*) (4-15)
where:
HI^ = HI modified by binary interactions data,
HQ; = hazard quotient for chemical i (unitless, e.g., daily intake/RfD),
fjj = toxic hazard of the jth chemical relative to the total hazard from all chemicals
potentially interacting with chemical i (thus j cannot equal i),
My = interaction magnitude, the influence of chemical j on the toxicity of chemical i,
By = score for the strength of evidence that chemical j will influence the toxicity of
chemical i, and
By = degree to which chemicals i and j are present in equitoxic amounts.
Many formulas could be derived that reflect these ideas. The above formula is
recommended as an interim method that is also simple. Assumptions 1 and 4 are simplifications
in the data gathering stage. Assumption 2 can then be modeled by a simple symmetric function
that is maximal when HQ;=HQj. Assumption 5 has no quantitative empirical support we could
find, and may be more reflective of the reduction in toxicity as dose decreases, making detection
of an interaction more difficult. Consequently, assumption 5 will not be included here. Pairwise
interaction studies usually show the influence of one chemical on the toxicity of the other
chemical. If each HQ is used as the measure of that component chemical's toxicity, then we can
modify the HI by multiplying each HQ in the formula by a function of the following quantities:
the HQs of the other chemicals (to reflect the actual component exposure levels), the estimated
magnitude of each pairwise interaction, and the two WOE scores. In this way, we are
incorporating the interactions by modifying each HQ by the influences of all the other potentially
interacting chemicals. These modified HQs are then summed to get Equation 4-15, the
interaction-based HI for the mixture.
-93-
-------
4.3.1.3. Weight-of-Evidence Factor (B)
The binary weight-of-evidence factor By reflects the strength of evidence that chemical j
will influence the toxicity of chemical i, and that the influence will be relevant to human health
risk assessment. The factor need not be the same for the influence of chemical i on the toxicity
of chemical j; i.e., By ^ B^ . The weight-of-evidence determination begins with a classification
of the available information, followed by a conversion of that classification into a numerical
weight.
The current weight-of-evidence classification is given in Table 4-3. This scheme does
not focus specifically on the types of data available to support a WOE determination, but on the
interpretation of the data made by an analyst or a group of analysts. In this respect, the scheme is
less directive and more flexible than the BINWOE method originally developed by Mumtaz and
Durkin (1992). Further, to allow for future modification of this classification, the binary nature
is not mentioned; i.e., the "BINWOE" has been replaced by simply "WOE."
The scheme is based on the assessment of the direction of an interaction, the plausibility
that the interaction will occur, and the potential relevance of the interaction to human health.
Four levels of confidence in the assessment—Roman numerals /through W—are described. For
each category, the weight-of-evidence determination is not intended to consider the magnitude of
the interaction, the dose levels at which the interaction will occur, or the relative amounts of the
agents in the mixture. Similar to the original BINWOE method, these factors are considered at a
subsequent stage of the analysis, as detailed below. The WOE scheme is then defined as:
• Weight-of-Evidence Determination—A judgment reflecting the quality of the
available information that categorizes the most plausible nature of any potential
influence of one compound on the toxicity of another compound, for a given
exposure scenario.
As indicated in Table 4-3, the first category, /, is intended to reflect essentially complete
confidence that the interaction will occur in humans and, therefore, the interaction is assumed
relevant to human health. A classification of/ does not necessarily imply that the interaction has
been observed in humans, or even that the interaction has been demonstrated in vivo. Although
this might often be the case, it is not necessary. The classification does indicate that, in the
judgment of the analyst or group of analysts, an interaction will occur, the direction of the
interaction can be predicted with confidence, and the nature of the interaction has clear
toxicologic relevance for humans.
In this context, the term toxicologic relevance means both that the interaction clearly
affects the health of the whole animal and that the endpoint of concern for effects on human
-94-
-------
Table 4-3. Modified weight-of-evidence classification"
Categories
The interaction has been shown to be relevant to human health effects and the
direction of the interaction is unequivocal.
II
The direction of the interaction has been demonstrated in vivo in an appropriate
animal model, and relevance to potential human health effects is likely.
Ill
An interaction in a particular direction is plausible, but the evidence supporting the
interaction and its relevance to human health effects is weak.
IV
The information is:
A. Insufficient to determine the direction of any potential interaction.
B. Insufficient to determine whether any interaction would occur.
C. Adequate as evidence that no toxicologic interaction between/among the
compounds is plausible.
aSee text for more detailed descriptions of each category.
health will be affected by the interaction. For example, assume that two chemicals are under
consideration, both having RfDs based on liver damage. Also assume that a study is available
that demonstrates a synergistic interaction on the kidney. Depending on the nature of other
supporting evidence, the information about the kidney interaction might or might not be deemed
relevant to the assessment of potential interactions affecting the liver. If it is deemed relevant,
the kidney study could be used to support a categorization of/. Otherwise, a different category
would apply, as discussed below. In either case, the burden is placed on the analysts to provide
the rationale for the determination.
At the other extreme, the lowest classification level, IV, encompasses three very different
types of assessments. The first, IV. A, is that an interaction may occur, but the direction of the
interaction cannot be determined. This type of classification could be based on conflicting
experimental results or on mechanistic ambiguity. For example, suppose that two studies are
available on the effect of chemical A on chemical B. Both studies use essentially identical
experimental designs, but they yield conflicting information on the nature of the interaction. In
this case, concern that an interaction could occur might be high, but the direction of the
interaction could not be determined. Mechanistic ambiguity is a term used by Mumtaz and
Durkin (1992) to describe assessments in which considering information on the biological
activity of the components could lead to different interpretations. For example, if both agents are
conjugated by the same compound as part of the detoxification process, competition for the
conjugating compound could lead to a greater-than-additive interaction. If, however, both agents
-95-
-------
are also oxidized by the same enzyme system to more toxic intermediates prior to conjugation,
saturation of the enzyme system could lead to a less-than-additive interaction. In such a case,
concern for the interaction could be high, but again the direction of the interaction could not be
determined.
The second category in level IV, IV.B, is simply intended for cases in which no
information is available on how the compounds are likely to interact or even to indicate that any
interaction is likely. This may be considered the complete opposite of Category / : rather than
complete certainty, IV.B reflects the admission of complete uncertainty.
A classification oflV.C is almost identical to Category /in that there is complete
certainty. In this case, however, the certainty is that no interaction will occur. This type of
classification usually indicates that one of the additivity models has been demonstrated or is very
likely to apply.
These three very different states of knowledge are placed within a single category because
they all have the same effect on the risk assessment of a mixture. If the direction of the
interaction cannot be specified—either because of conflicting information or a lack of
information—or if the interaction is known to be additive, an additive model is used in the
mixtures risk assessment. Explicitly identifying these three very different states of knowledge,
however, is intended to highlight the need for reflecting these differences in the verbal narrative
that should accompany each risk assessment.
Any number of classifications could be constructed between the complete certainty that
an interaction will occur and the acceptance or demonstration of an additivity model. Only two
additional categories, // and ///, are defined in the recommended system. Category // is intended
for cases in which the data strongly support the determination that an interaction will occur in a
particular direction, but in which the relevance of the interaction to human health effects, while
plausible, cannot be demonstrated with a high level of assurance. Category // then reflects the
lowest extent of extrapolation, across species or target organ, but supported by some evidence of
the toxicologic similarity.
The above example of two chemicals with RfDs based on liver toxicity and available data
showing an interaction on renal toxicity could fit into this category if confidence were low in the
relevance of the kidney interaction to effects on the liver.
Category /// reflects more extrapolation and hence lower levels of confidence in the
assessment, either in terms of relevance to in vivo toxic effects or of uncertainties in the direction
of the interaction. This category is intended primarily for cases in which interactions have either
been demonstrated or seem plausible, but only under experimental conditions that do not
correspond to the exposure scenario of concern. For example, many studies are available on
interactions from sequential exposures: a group of animals is pretreated with one chemical and
-96-
-------
then dosed with a second chemical. Various control groups or different dose levels of the two
agents are used to determine if pretreatment with the first chemical has any influence on the
toxicity of the second chemical. These studies are usually designed to elucidate some aspect of
the mechanism of action or the metabolism of the second chemical. Depending on the specific
chemicals and the nature of any supporting information, the resulting data may or may not be
judged sufficiently relevant for a weight-of-evidence determination. If they are used, however, a
classification of ///will often be more appropriate than a classification of//.
Category ///will also encompass cases in which a toxicologic interaction has not been
demonstrated, but in which mechanistic data, while not compelling, are adequate evidence that
an interaction in a particular direction is more likely than an interaction in an opposite direction
and more likely than no interaction at all. In other words, mechanistic ambiguity may exist but
be resolvable to an extent that the case merits a score higher than IV.A.
The above descriptions of types of data that might fit each of the four basic categories in
the modified WOE classification are not intended to be restrictive. The nature of the data chosen
to support a particular classification is left to the discretion of the analyst. This relative lack of
structure is the major conceptual difference between this method and the BINWOE method
originally described in Mumtaz and Durkin (1992).
The term By is simply the quantitative weight assigned to the qualitative WOEs (Table
4-4). Positive values indicate synergism and negative values indicate antagonism. These
numerical assignments are only crude weighting factors, not specific measures of interaction. As
more information becomes available on toxicologic interactions, these assignments may change.
4.3.1.4. Exposure Factor (F)
The Hazard Quotient for a chemical is multiplied by a sum of terms that reflect the other
chemicals' interactions. This sum must reduce to unity (1) when dose addition is assumed, and
so must be normalized in some fashion to avoid double-counting the individual Hazard
Quotients. This is accomplished for each of the other components using the term f^:
HO.
L = - =L - (4-16)
where HIadd is the standard HI based on dose additivity. This factor then scales the interaction
contribution of chemical j by its importance relative to all the other chemicals interacting with
chemical i. The toxicologic importance here is represented by the Hazard Quotient.
-97-
-------
VO
oo
Table 4-4. Default weighting factors for the modified weight of evidence
Category
I
n
m
IV
Description
The interaction has been shown to be relevant to human health
effects and the direction of the interaction is unequivocal.
The direction of the interaction has been demonstrated in vivo
in an appropriate animal model, and the relevance to potential
human health effects is likely.
An interaction in a particular direction is plausible, but the
evidence supporting the interaction and its relevance to human
health effects is weak.
The assumption of additivity has been demonstrated or must be
accepted.
Direction
Greater than
additive
1.0
0.75
0.50
0.0
Less than
additive
-1.0
-0.5
0.0
0.0
-------
4.3.1.5. Interaction Magnitude (M)
The term My represents the maximum interaction effect, as defined below, that chemical j
can have on the toxicity of chemical i. As with the WOE score, B, the interaction magnitude
need not be symmetric; i.e., the magnitude of interactive influence of chemical i on the toxicity
of chemical j may be different than the corresponding magnitude of chemical j on the toxicity of
chemical i. The direction of the effect (synergism or antagonism) is not incorporated into My, co-
workers (1969, 1970) conducted a study on the joint action of all possible pairs of 27 chemicals
administered in equivolume combinations and 53 chemical pairs administered in equitoxic
concentrations. The range of predicted to observed LD50s was about 0.2-5. In other words, the
magnitude of the deviation from additivity for the mixtures tested was about a factor of 5 in
either direction (0.2 = 1/5). More extreme interactions have been noted, for example, the
interaction described by Mehendale for the effect of chlordecone on the toxicity of carbon
tetrachl oride.
The default interaction magnitude is set at 5 in this guidance to reflect the studies
described above. When the weight of evidence suggests an interaction but the magnitude of the
interaction cannot be quantified, this default value of 5 should be used for the interaction
parameter M. Because this value does not have strong empirical support, information specific to
the chemical components of concern should be used when available. Care should be taken to
ensure that the measured interactions are relevant to the low exposure levels usually involved in
environmental regulations, as well as to the health endpoints of concern.
4.3.1.6. Weighting Factor for Relative Proportions (0)
The term 0y reflects the degree to which components i and j are present in equitoxic
amounts. The definition of equitoxic is based on the relative magnitudes of the Hazard
Quotients. Thus, the 1th and jth components are said to be equitoxic if HQ; = HQj. A measure of
the deviation from equitoxic amounts for the ith and j* components is defined simply as the ratio
0y of the geometric mean to the arithmetic mean:
e =
* [(HQi+HQ) • 0.5]
Note that as HQ; approaches HQj, 0y approaches unity. As the difference between HQ; and
increases, 0^ approaches zero.
The term 0y is incorporated into the algorithm under the assumption that, for a given total
dose of two chemicals, the greatest deviation from additivity will occur when both of the
-99-
-------
components are present in equitoxic amounts. This assumption is also explicit in Finney's model
of a deviation from dose additivity (e.g., Finney, 1971, Equation 11.83, p. 262).
4.3.1.7. Example
The properties of the interaction-based HI and some sample calculations are presented in
this section, using hypothetical chemicals so that certain points can be illustrated. Consider the
following scenarios where high-quality information is known on the binary interactions of the
mixture components. In all three cases, the weight-of-evidence categories would be I and thus
the WOE scores would be 1.0.
Scenario 1
All binary combinations of three chemicals are known to synergize each other by a factor
of 5 for the route and duration of concern, with an interaction directly relevant to human
health.
Scenario 2
All binary combinations of three chemicals are known to be additive for the route and
duration of concern, with an interaction directly relevant to human health.
Scenario 3
All binary combinations of three chemicals are known to antagonize each other by a
factor of 5 for the route and duration of concern, with an interaction directly relevant to
human health.
In scenario 2, each B^ is equal to zero because the three chemicals are known to be
additive (category IV-C in Table 4-3). As a result, M is taken to the power of zero. Thus,
whatever default value is used for M, the value of M to the power of zero is unity. Also, from
Equation 4-16 we see that regardless of the ratios of the components in the mixture, the sum of
the f-jS will equal 1.
(HI
add
In other words, the HI will not change from one based on additivity. The HI modified for
interactions for scenario 2 is then:
-100-
-------
n
i-l i*i i-l
Scenarios 1 and 3 are not quite as simple. Because these scenarios are identical except
for the direction of the interaction (and hence the WOE weighting factors), only scenario 1 will
be examined in detail. If each of the chemicals in the mixture is present in equitoxic amounts,
then all the Hazard Quotients are equal. Equation 4-15 yields an adjusted HI five times greater
than the HI based on additivity. Note that in this simple case, both By = 1 and 0^ = 1. Assuming
that M is set to 5 (the proposed scenario says each chemical is known to potentiate the other by a
factor of 5), then Equation 4-15 reduces to:
Thus, if the HI based on additivity were 1, the HI considering interactions would be 5. The
counterpart, scenario 3, would give an interaction-based HI of 0.2.
Suppose, however, that the mixture of chemicals 1, 2, and 3 was such that the hazard
quotients of each chemical were 0.98, 0.01, and 0.01, respectively. For such a mixture, it would
not seem reasonable to assume as great an interaction as in the equitoxic mixture because the
relative amounts of chemicals 2 and 3 are much smaller than in the equitoxic mixture. For this
98:1:1 mixture of the three chemicals, 0y < 1 for pairs involving chemical 1, resulting in a
decrease in the interaction-based HI. For the effect of chemical 2 on chemical 1, using Equation
4-17 gives:
012 = (0.98*0.01>5 / (0.99/2) = 0.2, f12 = 0.01 / (1.00-0.98) = 0.5
Thus, the partial adjusted hazard quotient for just the effect of chemical 2 on chemical 1 is:
fu*MQu= 0.98 * 0.5* 5° 2=0.676
By symmetry, the effect of chemical 3 on chemical 1 would also be 0.676. Thus, the adjusted
hazard quotient for chemical 1 would be 1.35 [=0.676+0.676], a 38% increase over HQj.
By applying the same hazard quotients to the other terms in Equation 4-15, the adjusted
hazard quotients for chemicals 2 and 3 can be determined. The adjusted hazard quotient for
-101-
-------
chemical 2 is 0.014. Because chemical 3 is present in the same relative amount as chemical 2,
the adjusted hazard quotient for chemical 3 would also be 0.014. As a result, the interaction-
based HI is 1.37 [1.35+0.014+0.014] for this 98:1:1 mixture of the three chemicals. Rounding to
a single significant digit would yield a HI of 1, essentially the same as that under the assumption
of additivity. Any time one chemical dominates the mixture composition to this extent, a good
approximation is that the interaction-based HI will be close to the hazard quotient for that
chemical.
Other cases can be similarly calculated. For example, with the same assumptions and a
mixture composition of 8:1:1, a mixture having an additive HI = 1 would have an interaction-
based HI of 2.77, which would round off to 3. If the interactions evidence were only in a few
studies on animals, so that the WOE was level n and thus a score of 0.75, the interaction-based
HI would be 2.16, which rounds to 2.
Evidence of antagonism that is not of level I quality receives a lower score than its
counterpart for synergism (Table 4-4). The influence that this protective bias has on the
interaction-based HI can be seen by altering scenario 1 (equal hazard quotients, HI = 1) to have
interactions all of level II quality, so that antagonism yields B = 0.5 whereas synergism gives B =
0.75. The results are easily observed by the multiplicative (n-fold) increase or decrease in HI:
Synergism Antagonism
Interaction-Based HI 3.3 0.45
n-fold increase or decrease of HI 3.3 2.2
4.3.2. Information Requirements
Empirical evidence of toxicologic interaction is required only for interactions of pairs of
chemicals. Recall that one assumption of this procedure is that the mixture response can be
adequately approximated by the modification of each term in the additivity-based HI using only
pairwise interactions. The interaction-based HI, HIINT, applies to one type of toxicity, so the
interaction must influence that toxicity. For example, consider the case where metabolites of
chemical A cause liver toxicity, and chemical B potentiates that liver toxicity by enhancing the
metabolism of A. Then the interaction, the influence of B on A's toxicity, should be included.
Even if the primary toxicity of B, the interacting chemical, is different from the toxicity of
concern addressed by the index (e.g., chemical B causes kidney lesions), B is included because it
influences the toxicity addressed by HIINT. Contrast this procedure to the additivity-based HI
(Section 4.2), where only lexicologically similar chemicals are included. The consequence is
that an interaction-based HI can include more types of chemicals than would the additivity-based
HI.
-102-
-------
The inclusion of interacting chemicals that do not cause the toxicity of concern in the
calculation does not cause any difficulties. In the above example, if chemical B does not cause
liver toxicity, then its HQ is zero. Chemical B then only enters the calculation through its
influence on the toxicity of chemical A.
An improved HIINT would result if the default functions, f and g, could be replaced by
empirically derived models that reflect the dose-dependence of the interaction. Such information
is rare, and, although encouraged, is not required.
4.3.3. Interpretation
Algorithms are presented here for using qualitative weight-of-evidence determinations to
modify a risk assessment based on information on binary interactions. These algorithms are
somewhat more flexible than those originally proposed by Mumtaz and Durkin (1992) in that
information on the magnitude of the interaction can be explicitly incorporated, and that
modifications are made to each chemical's Hazard Quotient. In addition, if specific information
is available, the influence of mixture composition on magnitude of interaction can also be
incorporated, and the interaction can be asymmetric, i.e., the influence for chemical A on toxicity
of chemical B can be different than for chemical B on toxicity of chemical A.
The methods for modifying the HI are based on commonly discussed principles of
toxicologic interactions. The algorithms, however, do not attempt to directly model toxicologic
interactions. Instead, the method should be regarded as a method for modeling "concern" for
toxicologic interactions, which reflects issues of magnitude as well as likelihood. In this respect,
the scheme corresponds more closely with the current use of uncertainty factors in the risk
assessment of single chemicals than with an attempt to biologically model interactions. When
specific information is available to model the pairwise interactions as functions of component
dose, such information can be used in lieu of the default procedures outlined above. As more
interaction studies are completed and more interaction mechanisms and modes of interaction are
understood, these algorithms will be revised.
4.4. RELATIVE POTENCY FACTORS
4.4.1. Introduction
The toxicity (i.e., magnitude of toxic effect) of a chemical mixture is best determined by
direct toxicologic evaluation. When such studies are available for all of a mixture's component
chemicals, they may be used to develop a hazard index (see Section 4.2). Because of the
temporal and monetary constraints imposed by epidemiologic studies or direct toxicologic
evaluation of the components or the mixture as a whole, other approaches that rely more heavily
on scientific judgment have been developed to assess the special case of the toxicity of mixtures
-103-
-------
of related compounds. The use of existing data makes these approaches faster and less
expensive, but they are less certain because they employ simplifying assumptions and toxicity
inferences.
For the general case, evaluation of mixtures of related chemical compounds that are
assumed to be lexicologically similar can sometimes be made by using relative potency factors
(RPFs). The approach relies on both the existence of toxicologic dose-response data for at least
one component of the mixture (referred to as the index compound) and scientific judgment as to
the toxicity of the other individual compounds in the mixture and of the mixture as a whole. The
applicability of RPFs may be limited to certain types of effects or to a specific effect because of
data limitations; RPF application may also be limited to a specific route of exposure or exposure
duration. The toxicity of the related compounds is predicted from the index compound by
scaling the exposure level of each compound by its toxicity relative to the index compound. This
scaling factor or proportionality constant is based on an evaluation of the results of a (usually)
small set of toxicologic assays or analyses of the chemical structures. This constant is called the
RPF and represents the relative toxicity with respect to the index compound. For example, if
compound A is judged to be one-tenth as toxic as the index compound, i.e., it requires ten times
the exposure to cause the same toxicity, then the RPF for compound A is 0.1. If all components
of the mixture are assumed to be as toxic as the index compound, then all of RPFs would be 1.0;
conversely, if all of the related compounds have negligible toxicity, all of their RPFs could be
assigned a value of 0.
In the RPF approach, an exposure equivalent to the index compound is the product of the
measured concentration of the mixture component and the RPF. These dose equivalents are
summed to express the mixture exposure in terms of an equivalent exposure to the index
compound; risk can be quantified by comparing the mixture's equivalent dose in terms of the
index compound to the dose-response assessment of the index compound. This estimate of
equivalent index compound exposure should be considered an interim and approximate decision-
making tool. The RPFs must be defined as to the scope of toxicologic effects that are covered,
and the degree of similarity in chemical structure and mode of action that can be inferred from
the summation of the adjusted exposure levels. (Mode of action refers to a continuum that
describes the key events and processes starting from the point of toxicant-cell interaction and
leading to the onset of a health endpoint). In general, the mixture concentration expressed in
terms of the index compound for n compounds is,
(4-18)
-104-
-------
where
Cm = mixture concentration expressed as index compound,
Cj = concentration of the index compound in mixture,
Ck = concentration of the k* mixture component, and
RPFk = proportionality constant for toxicity of the kth mixture component relative to the
toxicity of the index compound.
Clearly, RPF^l, as k=l indicates the index chemical.
To date, the Agency has developed three examples of RPFs that estimate the toxicity of a
mixture of related compounds. Each of these examples has been developed as an interim
measure pending the development of more case-specific data. The three classes of compounds
for which relative potency approaches have been examined by EPA are the dioxins, the
polychlorinated biphenyls (PCBs), and the polycyclic aromatic hydrocarbons (PAHs). Because
the levels of current scientific understanding of the modes of action and the toxicologic databases
for these classes of compounds differ, these three attempts have not achieved the same level of
scientific acceptance.
4.4.1.1. Dioxins
In March of 1989, EPA released Interim Procedures for Estimating Risks Associated with
Exposures to Mixtures of Chlorinated Dibenzo-p-Dioxins and -Dibenzofurans (CDDs and CDFs)
and its 1989 update (EPA/625/3-89/016). These procedures also were discussed and adopted
internationally (Mukerjee and Cleverly, 1987; NATO/CCMS, 1988). In addition to describing
the regulatory need and the process of achieving scientific consensus, the 1989 EPA document
cautiously recommended comparing available toxicologic data and structure-activity relationship
information on dioxin class members with those of 2,3,7,8-TCDD, the index compound, to
estimate the significance of exposures to the other 209 compounds in this class, termed
congeners. The consequence of exposure to each compound was expressed in terms of an
equivalent exposure of 2,3,7,8-TCDD by multiplying the concentrations of the individual
congeners by their assigned toxicity equivalence factor (TEF), a specific type of RPF. The
resulting 2,3,7,8-TCDD toxicity equivalents (TEQ) were then summed to estimate the risk
associated with the mixture of these compounds. The TEFs were assigned on the basis of such
data as information regarding human carcinogenicity, carcinogenic potency based on animal
studies, reproductive effects data, in vitro test data, and structure-activity relations. Van
Leeuwen (1997) and van den Berg et al. (1998) identified each comparison of toxicity from an
individual experiment as a relative potency value, or REP. The term TEF was reserved for
consensus toxicity estimates where a single TEF is assigned to each dioxin congener. These
-105-
-------
TEFs were assumed to encompass and apply to all health endpoints and all exposure routes for
this class.
A number of toxicologic assumptions were associated with this approach; these included
the applicability of extrapolation from short-term to long-term health effects, similarities between
interspecies metabolism, appropriateness of high-dose to low-dose extrapolations, a common
mode of action for all members of the class, the constancy of TEF relationships for different
exposure routes and health endpoints, and the concept of dose additivity (U.S. EPA, 1989b). To
better capture the uncertainty in these assumptions, all TEFs were provided as order-of-
magnitude estimates, and the Agency regards the results of dioxin TEF application as interim.
The specific term TEF was applied to this class because of the wide acceptance of the approach
and the broad applications (i.e., across route and health endpoints) for which it was designed.
Similarly, use of the term TEQ implies the existence of a larger data set upon which to base
toxicologic comparisons than would be true for most RPFs, so that this term should not be used
for the general case.
After the TEFs were developed for dioxins, seven guiding criteria were developed for the
TEF approach (Barnes et al., 1991; U.S. EPA, 1991a). It must be noted that a key assumption for
the dioxins was that a single TEF could apply to all toxic endpoints, all routes of exposure, and
for all exposure durations. This means that, for example, for a given congener, the same TEF
would be used to assess cancer risk and to assess potential developmental effects. The criteria
were:
• Demonstrated need for an interim assessment
• A well-defined group of compounds that occur in environmental samples as
mixtures
TEF based on broad set of toxicity data covering many endpoints and many
congeners
• Relative congener toxicity generally consistent across many different endpoints
• Additivity of dose (i.e., dose addition)
• A presumed common mode for toxic endpoint of the components
• TEF are formed through a scientific consensus.
These criteria were developed for specific application to the dioxins and dioxin-like compounds.
The TEF is viewed as a specific type of application of the RPF. The criteria listed by Barnes et
al. reflect the specific nature of the application to the dioxins, and dioxin-like PCB as discussed
below in Section 4.4.1.2.
The assignment of consensus TEF for chlorinated dibenzo-p-Dioxin, Dibenzofurans, and
biphenyls has been reevaluated by a number of expert panels including a recent one organized by
the World Health Organization (WHO) in 1997 (Van den Berg et al., 1998). Based on the
-106-
-------
research into the toxicity of these compounds (e.g., Ahlborg et al., 1994), which occurred after
the early TEF work in the late 1980s and early 1990s, revisions were made to the TEFs that
reflected a consensus judgment of the expert panel. For REPs from a given scientific study to be
included in this TEF reevaluation effort, this expert panel developed explicit criteria; these were
the inclusion of a reference compound in the scientific study and demonstrated effects on the
relevant endpoint by both the reference compound and the study compound(s) in the scientific
study. The panel agreed upon a specific ranking scheme for weighting different types of
scientific studies. In this weighting scheme in vivo toxicity data were weighted more heavily than
in vitro data or assessments of toxicity based on structural elements of a compound (Structural
Activity Relationship (SAR) data). Within the in vivo toxicity data, results of chronic studies
were weighted most heavily followed by subchronic studies and acute studies. Toxic responses
were also weighted more heavily than adaptive responses.
The WHO expert panel (Van den Berg et al., 1998) also reevaluated the soundness of the
TEF approach for this group of compounds. They "...concluded that the TEF concept is still the
most plausible and feasible approach for risk assessment of..." this group of compounds. Studies
have been conducted that assess the toxicity of specific dioxin, furan and PCB mixtures in whole
mammals (or in cultured mammalian cell lines) and compare these measures with the TEF-
predicted toxicity. The TEF-predicted toxicity was found to generally agree with a range of
toxicity measures (e.g., Harris et al., 1993; Schrenk et al., 1994; Harper et al., 1995; Schmitz et
al., 1996; Smialowicz et al., 1997). However, for some toxicological responses, there appears to
be evidence for nonadditive interactions as well as antagonism and potentiation (e.g., Davis and
Safe, 1989; Safe, 1994; Birnbaum et al., 1985). This TEF approach and the TEF values
developed have been adapted and presented in the draft dioxin reassessment (U.S. EPA, 2000b).
Interestingly, the WHO expert panel (Van den Berg et al., 1998) extended the TEF
approach for this group of compounds to three classes of nonmammalian chordates, developing
consensus TEFs for two classes offish and birds. The expert panel also described studies in fish
and birds that test the validity of the TEF approach. The results of these efforts are described as
supportive of the general assumption of dose additivity, although deviations from this
assumption are identified.
4.4.1.2. PCBs
The Workshop Report on Toxicity Equivalency Factors for Polychlorinated Biphenyl
Congeners (U.S. EPA, 1991a) reported that certain groups of PCBs appear to share a common
mode of action with 2,3,7,8-TCDD. On this basis TEFs (this term was again applied rather than
RPF because of the specific application to this chemical subclass related to dioxins) were
proposed in that report and others (e.g., Ahlborg et al., 1994) that related the toxicity of exposure
-107-
-------
to members of these PCB subclasses to that of 2,3,7,8-TCDD. The same approach to estimating
TEQ was advanced for this group (U.S. EPA, 1991a). TEFs were proposed only for some
members of the class, and the TEFs proposed were considered applicable only to the health
endpoint of cancer through the common mode of action shared with the dioxins.
When assessing PCB mixtures, it is important to recognize that both dioxin-like and non-
dioxin-like modes of action contribute to overall PCB toxicity (Safe, 1994; McFarland and
Clarke, 1989; Birnbaum and DeVito, 1995). Because relatively few of the 209 PCB congeners
are dioxin-like, dioxin equivalence can explain only part of a PCB mixture's toxicity. RPFs
based on action similar to 2,3,7,8-TCDD have been developed for 13 dioxin-like PCB congeners
(Ahlborg et al., 1994), but no RPFs exist for the non-dioxin-like modes of action.
Because PCB cause cancer by both dioxin-like and non-dioxin-like modes of action, both
dioxin-like and non-dioxin-like portions of a mixture must be evaluated, either jointly or
separately. When environmental concentrations of the dioxin-like congeners are available, those
exposure estimates can be multiplied by the corresponding RPFs and then summed to yield the
equivalent 2,3,7,8-TCDD exposure level for the dioxin-like portion of the mixture. The
estimated cancer risk attributable to the dioxin-like portion of the mixture is then the cancer risk
for that exposure to 2,3,7,8-TCDD. For the non-dioxin-like portion, the total dose of the
remaining congeners (subtracting the 13 dioxin-like congeners) can be multiplied by the slope
factor that would otherwise be applied to the total PCB mixture. Then the cancer risk estimates
for those two portions of the mixture (dioxin-like and non-dioxin-like) can be added as an
estimate of the overall cancer risk posed by the mixture. U.S. EPA (1996a) provides an example
of this approach. (It should be noted that the cancer slope factor for PCBs in U.S. EPA 1996a
was developed at a time when the concentration of the dioxin-like PCB congeners in the tested
mixture had not been reported. This information has since become available [Cogliano, 1998]
and EPA is revising the procedure by which dioxin equivalence is estimated.)
4.4.1.3. PAHs
The Provisional Guidance for Quantitative Risk Assessment of Polycyclic Aromatic
Hydrocarbons (PAHs) (EPA/600/R-93/089) described an RPF approach for assessing the
carcinogenic risks posed by exposures to non-benzo(a)pyrene (B[a]P) PAHs that had been judged
by the Agency as B2 substances; i.e., probable human carcinogens. The results of mouse skin
carcinogen!city assays for these non-B[a]P B2 PAHs were compared with those of B[a]P to
estimate cancer potency. The approach assumed that the B2 PAHs had the same cancer slope
factor as B[a]P. The ability of these non-B[a]P B2 PAHs to elicit rodent skin tumors was
quantitatively compared to that of B[a]P; the results of this quantitative comparison were
expressed as an "estimated order of potency." Because this approach was limited to the cancer
-108-
-------
endpoint, based on B[a]P exposure from a single (oral) pathway (for the derivation of the slope
factor), and considered only a small subset of the PAHs, EPA has described it as an estimated
order of potency. This naming reflects the uncertainly EPA felt about the application of this
type of approach given the current state of science of PAHs. To estimate cancer risk for the B2
PAHs, the cancer slope factor for B[a]P was multiplied by the estimated order of potency and by
the concentration of the specific PAH.
4.4.2. Procedures for Developing a Relative Potency Factor (RPF) Approach
TEFs for dioxins were the first RPFs developed and reflect a chemical group with the
broadest database examined to date and an apparent uniform mode of action. The criteria for
developing TEFs are more rigorous than can be met by most classes of chemicals. However,
TEFs provide the background for the procedures for general development of the RPF. The RPF
may be less rigorous scientifically than the TEF and its application may be constrained by the
available data (Table 4-5). The RPF is viewed more broadly than the TEF and can be formulated
by the following procedures. Typically RPFs will be developed by a cross-disciplinary group of
scientists to address specific regulatory needs.
4.4.2.1. Demonstrate Need for the Use of RPF as an Interim Estimate of Exposure
The RPF approach should only be applied when dictated by a clear regulatory need.
When temporal or monetary issues preclude more thorough analyses of the chemical mixture of
concern, then a RPF approach may be appropriate. The RPF approach is considered to be an
interim method of dose-response assessment and its application may be more uncertain than
other methods.
4.4.2.2. Initiate the RPF Process
When developing an RPF, both the appropriate data and the relevant scientific expertise
needed to evaluate the data must be assembled. The minimum data needed for development of an
RPF approach include: (1) a known or suspected common mode of action shared by the class of
compounds; (2) a quantitative dose-response assessment for the index compound; and (3)
pertinent scientific data that allow the components to be meaningfully compared to the index
compound in terms of relative toxicity. The relevant toxicologic data for the individual
components may include short-term or chronic in vivo assays, in vitro assays, and quantitative
structure-activity relationship data. Because the RPF approach relies heavily on the judgment of
scientific data, it may be important to assemble a cross-disciplinary group composed of scientists
who have established expertise for the given chemical class or understand the relevance of the
various toxicologic assays to human health risks. This group can assemble, interpret, and
-109-
-------
Table 4-5. Differences Between TEFs and RTFs
Specific
Apply to
Apply to
Apply to
TEFs
type of RPF
all heath endpoints
all exposure routes
all exposure durations
Imply more abundant data and greater
certainty about mode of action
| RTFs
Generalized case
May be limited to specific health endpoints
May be limited to specific routes
May be limited to a specific exposure duration
May be based on lower quality/fewer data and
less certainty about the mode of action
integrate the relevant scientific data and may know of ongoing research activities that could be
brought to bear on the process. This scientific group may also be useful in the evaluation and
limitations of the final product(s) of the approach.
4.4.2.3. Define the Class of Compounds
The compounds included in the chemical class to be considered should be well-defined.
They should be described in terms of the commonalities that permit them to be combined in an
RPF approach. Included in the definition of the class should be the understanding of the
common mode of action leading to the observed toxicologic effects, the chemical similarity of
the compounds, and the identification of the spectrum of toxicologic impacts shared by the class.
The compounds should also be known to occur as mixtures in environmental samples. If
exposures to the class compounds are not simultaneous, the RPF approach may still be valid.
Sequential exposures could result in overlapping internal doses, or overlapping effects because of
persistence of single-chemical effects. In those cases, dose addition could be an appropriate
approximate characterization of the mixture exposure, and so the RPF approach may be adequate
for the mixture risk assessment. Example applications have not been located in the literature, so
each case must be considered on its own merits. Exposures to different chemicals in the class that
are widely separated in time, however, may be better characterized by separate assessments that
treat the chemicals independently.
While clearly it is important to know the compounds involved, it is also important to
describe what is not known about the chemical class of interest; this includes descriptions of the
limitations of current analytical techniques, fraction of unidentified material in typical
environmental mixtures, purity of the individual compounds when assayed, the costs related to
chemical analysis, the identification of toxicologic impacts not shared by the class of compounds,
-110-
-------
etc. In this step it is also important to identify which compounds or groups of compounds are not
being considered, the reasons for this, and the potential impact of this missing information on the
mixture risk assessment. The relative abundance of a compound should also be considered: if a
particular compound is relatively rare, then large uncertainties may not be a significant factor for
RPF development. The pertinent data include dose-response data over a relevant range of doses.
4.4.2.4. Develop the RPF
4.4.2.4.1. Select the index compound. All RPFs will be based on comparisons of toxicity with
that of an index compound. It is preferable to have a single index compound for the RPF
approach to promote consistency of application and interpretation. The index compound should
have a quantitative dose-response assessment of acceptable scientific quality. It is presumed that
typically the index chemical will be the best studied member of the class and have the largest
body of acceptable scientific data. The pertinent data include exposure data for the routes and
duration of interest and health assessment data for health endpoints of interest.
For most chemical classes the index compound will be obvious. When there is more than
one potential candidate for the index compound, a judgment must be made regarding which
candidate is most representative of the class and has the most extensive and best quality database.
Once the set of toxicologic assays has been chosen for determining the RPF values, the selection
of the index compound will not impact the calculation of the equivalent mixture exposure level
because the relative magnitudes of the RPFs compared to each other will be unchanged. The
index compound selection does change which dose-response function will be used in interpreting
the equivalent mixture exposure in terms of health risk. Consequently, when there are multiple
candidates for index chemical, the uncertainty or range in the resulting mixture risk estimate
should reflect the differences in the index chemical dose-response function, both regarding
overall quality as well as relevance to the exposure conditions being assessed. For example,
when exposure conditions represent more than one route, it may be more appropriate to select a
different index chemical for each exposure route, i.e., one with the best dose-response data for
that route. Because the index compound must also have (or be expected to have) similar toxic
effects to the rest of the members of the class, toxicologic information about the compounds not
selected could be used to assess confidence in the approach in at least a limited manner.
4.4.2.4.2. Describe the scientific basis for the RPF. The scientific criteria for RPF development
need to be clearly stated. The known or suspected common mode of action shared by members
of the class of compounds should be described. If the toxicologic assays used to develop the
RPFs were ranked, the justification for the ranking and its application should be described. For
example, some RPFs could be assigned based on evidence of deleterious health effects in
-111-
-------
humans or study animals, reproductive effects data, in vitro test data, or structure-activity
relations. Actual evidence of deleterious human effects or reproductive effects data for some
compounds is usually considered more certain than inferences based on the chemical structures
of compounds, and thus, the results of these in vivo studies may be weighted more heavily than
in vitro test data (see discussion of minimum criteria for RPF developed in Section 4.4.2.2).
If a single RPF is judged incapable of representing all toxic effects, then this must be
clearly noted. The effects that are encompassed by the approach and the scientific reasons they
are included should be described. The effects not included should also be described along with
the reasons for the decisions described.
4.4.2.4.3. Assign RPF. A description of the approach used to determine the RPF values should
be included. This description should include the qualitative and quantitative interpretations of
toxicologic analyses for the compounds included in the RPF. The assignment of numerical RPF
values should also be explained. For example, to convey better the uncertainty to potential end
users in the three examples presented in Section 4.4.1, RPFs were assigned only as order of
magnitude estimates. Clearly, the certainty or precision of the approach should not be overstated.
When two or more assays are available to compare the toxicity of a class of compounds
with the index compound, multiple assay results could be used. For example, three RPF values
could be derived for one compound by using data from three different studies. The body of
scientific data used to determine an RPF for a specific member of a chemical class may be
portrayed as a range or a distribution. The resulting RPF range or distribution would still require
justification, including interpretation and impact of the individual toxicologic studies from which
the RPFs were developed.
4.4.2.5. Characterize Uncertainty
The strongest recommendation expressed in the U.S. EPA Chemical Mixture Guidelines
(U.S. EPA, 1986) (Appendix A) is to describe the uncertainties in risk assessment. This step is
crucial to proper interpretation of the RPF approach and the resulting mixture risk assessment.
The areas of uncertainty described below are considered to be a minimum of what should be
discussed. Other uncertainties that arise during the application to a specific mixture should also
be addressed.
4.4.2.5.1. Define the health endpoints and exposure routes covered and not covered by the
approach. In this step the scientific support for including or excluding the various endpoints and
routes in the RPF approach should be carefully described. The applications of scientific judgment
in the process of RPF development should be identified and described.
-112-
-------
For the widest application, a data set encompassing a variety of animal species, exposure
durations, health endpoints, and exposure routes is needed. In the best cases it can be stated with
some confidence whether the effect on which the RPF is based is the most sensitive; the full
spectrum of health impacts may also be known. For those classes of compounds with less than
complete toxicologic endpoint data for all members, it may be necessary to limit the endpoints of
applicability of the proposed RPF approach. When only some endpoints are represented, it is
important to state what cannot be considered and why. A risk assessment applying the RPF
should still account for other types of adverse health effects that are not included in the RPF
approach. If different RPFs are developed for different toxic endpoints, and one or more effect-
specific RPFs for any class member cannot be developed, this limitation must be clearly noted as
a bias toward underestimating that toxicity.
4.4.2.5.2. Determine the consistency within the group of compounds considered. If multiple
health endpoints, exposure durations, or multiple exposure routes are covered by the RPF, the
issue of consistency across routes, durations, and endpoints should be addressed. For example, a
consistent approach may result in similar predicted RPF orderings across different health
endpoints and in vitro assay results. This type of consistency may strengthen the choice of a
single RPF for multiple health endpoints or exposure routes. Statistical procedures may also be
used in this determination. The significance of inconsistencies should also be indicated and
reconciled if a single RPF is adopted for multiple health endpoints or routes. These may indicate
uncertainty surrounding the common mode of action or uncertainty about the relationships
between the class members and the index compound. Uncertainty of no more than two orders of
magnitude across endpoints and a generally consistent trend across several endpoints or exposure
routes would permit the choice of single RPF for a class or subclass of compounds. This
criterion can be disregarded if the RPF is limited to a single endpoint and exposure route.
4.4.2.5.3. Assess mode of action. It is necessary to describe the mode of action of the class of
compounds underlying the health effects for which the RPF was developed. A common mode of
action for the class is the basis for the assumption of dose additivity. However, in some cases
the class may be linked by common effect with only suggestive or indirect information
concerning the underlying mode of action. The description of the RPF must answer the question,
"to what degree do the scientific data support the assumption of a common mode of action?"
4.4.2.5.4. Assess additivity of dose assumption. The RPF approach assumes an additivity of
dose. Clearly, there is a stronger basis for the RPF when dose additivity is scientifically
demonstrated by dose-response studies that examine simple mixtures of the chemical class. If
-113-
-------
these studies support the assumption of dose additivity, they increase the confidence in
applications of the approach. If they indicate that there are synergistic or antagonistic
interactions that are not being considered, then the final answer based on the RPFs may be
unrealistic, and so a different approach such as the interaction-based hazard index should be
considered. Interactions noted only at high exposures, however, should be viewed cautiously
because they may not occur at lower environmental exposures. Pharmacokinetic differences
among the class of compounds should be identified because differences in the pharmacokinetics
across species could substantially change RPFs developed from nonhuman data.
Chemical mixtures may exhibit dose additivity over certain combinations of dose ranges
for the individual chemicals but may not exhibit dose additivity over others. A methodology for
detecting regions of additivity and/or departure from additivity has been proposed (Gennings et
al., 1997; Gennings, 1995). A key feature of the methodology is that it enables generation of
experimental designs that are practical in size, being based only on dose-response data for each
component in the mixture.
4.4.2.5.5. Examine the issues related to application of the RPF. The purpose of this step is to
allow the developers of an RPF approach to describe their concerns that relate to application of
the RPF. The concerns of those that develop an RPF approach are viewed as related, but distinct,
from those of the end users that apply it. Their concerns may pertain to the overall confidence in
the application when most of the toxicity is based on a subset of components with weaker data
(e.g., this could be related to lower confidence in the common mode of action). They may also
have concerns about confidence for certain exposure routes or endpoints. The developers of the
RPF should note any differences in pharmacokinetics across the class. When PBPK models are
not available and external exposure levels must be used, the assumption for simultaneous
exposures is that the pharmacokinetics are similar across the class or that a rough proportionality
exists between the external exposure and the tissue dose. However, when the exposures to the
class compounds are sequential, then differences in the pharmacokinetics could result in overlap
of internal doses from the separate exposures. Such information should be described for
consideration by the end user of the RPF. (See previous discussion in Section 4.4.2.3.)
4.4.2.6. Evaluation of the RPF
The RPF approach should undergo scientific peer review. The review should evaluate the
scientific judgments employed in each step of RPF development as well as issues related to RPF
application. The review should assess the following:
judgment that a common mode of action is shared by members of class,
• assignment of class membership,
-114-
-------
scientific data supporting each of the RPFs,
• consistency of the RPFs across the class for multiple routes or endpoints,
• the appropriateness of specific limitations pertaining to exposure route or target
organ, and
application issues.
4.4.2.7. Research Needs
The RPF approach does not use direct toxicity data on every member of the chemical
class; it is considered to be an interim method, to be replaced by better approaches when the
required data are available. The method most often recommended to replace the RPF (or TEF)
approach is a component-based assessment using actual dose-response data on each chemical in
the class. The resulting approach could be a Hazard Index, which is also based on dose addition,
or a response addition estimate of probabilistic risk, as is common for cancer risk assessments
(Hertzberg et al., 1999; U.S. EPA, 1989).
4.4.3. Risk Characterization Using RPFs
4.4.3.1. TEF-BasedAssessments
When a mixture exposure is completely described by TEFs, then the mixture risk is
quantitatively determined as if the mixture were solely composed of the index chemical. Risk
assessments for the various endpoints or target organs are performed in the same manner as for
the index chemical by itself. The uncertainty characterization, however, will be different,
reflecting the quality of the additivity assumption and of the supporting data used in assigning the
TEFs.
4.4.3.2. General RPF-Eased Assessments
When all chemical class members are assigned single RPFs that represent all effects,
exposure routes, and durations, the mixture risk is based solely on the equivalent exposure level
for the index chemical and is handled similarly to the TEF-based assessment described above.
When multiple RPFs are deemed necessary for one or more mixture components, e.g., for
different exposure routes or toxic effects, a separate mixture assessment should be developed for
each exposure route or for each major effect or target organ, as appropriate. These evaluations
are similar to the separate assessments made in the usual HI procedure.
Quantitative mixture risk assessments based on RPFs, even those that satisfy the
requirements for TEFs, are weaker than those assessments based on direct toxicity data. The
uncertainty description is then a key part of the risk characterization. The discussion of
uncertainties and overall confidence in the risk assessment should characterize the contribution
-115-
-------
of the index chemical to the total predicted equivalent exposure estimate. Similarly, the fractions
of predicted equivalent exposure that result from components that exhibit direct evidence of
human health effects and from the components for which direct toxicity data are available should
also be quantified. When most of the mixture risk is based on inferred toxicity (e.g., the index
chemical is not present or its presence accounts for only a small fraction of the quantitative risk),
then the assessment should be presented both with and without the risk estimated by RPFs. (This
is particularly important if there is a large disparity between the index compound and other
members of the class with respect to the quantity, quality, and pertinence of the data set).
Confidence in this approach for a given chemical class must be characterized in the context of
the assessment in which it is utilized. In this way an assessor's scientific judgment of this
confidence will be factored into the final risk assessment.
The RPF should be carefully defined as to its underlying limitations, including the
notation that the value obtained is an estimate of exposure, and might not be extended to
quantitative risk assessment. Analysts applying the RPF should also evaluate evidence for dose
and route extrapolations, including the relevance of toxicologic assays to human health
endpoints. Of particular importance is that the RPF may not cover all risk or all endpoints, so
that other toxicology information is needed. In such cases, the discussion should clearly note the
limited coverage of the assessment if based only on such RPFs.
When the data are judged inadequate to use the above RPF procedures, an approach could
be adopted where all compounds in the class are assumed to be as toxic as the index chemical.
Adoption of this approach is the numerical equivalent of assigning all components an RPF of 1.
An opposite approach is to ignore the potential toxicity of the poorly studied chemicals when
assessing the mixture's toxicity (in which case their RPFs would be the numerical equivalent of
0). Some combination of these two extremes may be the most scientifically appropriate. For
example, a set of scientific criteria could be determined where some of these members of the
class could be assigned an RPF of 1 and the other members could be assigned an RPF of 0.
For some mixtures there are analytical limitations. Some members of the class that are
present in the chemical mixture may not be identified, but their presence may be inferred from
measures of mass balance. The procedure for including these compounds in the risk assessment
should be clearly defined.
4.4.4. Hypothetical Example of RPF Approach
The application of RPFs to the estimation of risk from a mixture of compounds that exert
the same toxic effect by similar mode of action can be demonstrated by the following
hypothetical example. A group of five structurally related chemicals is used as insecticides to
protect against infestations of insects on crops. This group of chemicals exhibits cholinesterase
-116-
-------
inhibition as its primary toxicologic endpoint of concern. The chemicals also exhibit a variety of
other effects, but these effects are not shared uniformly across the group and appear to be due
largely to other structural components of the chemicals than those conferring cholinesterase
inhibitory properties. In particular, one chemical is a carcinogen, another causes kidney lesions,
and three cause nonspecific hepatic hypertrophy at higher doses. Because of the commonality of
the cholinesterase inhibiting effects, but lack of commonality of other effects, an RPF approach is
judged to be appropriate for combining risk of cholinesterase inhibition from this group of
chemicals.
After determining that there is a regulating need, the first step in developing a set of RPFs
for a group of chemicals is to evaluate the data available for each and identify the chemical
whose data set appears to be the most extensive and that best describes the toxicologic propensity
of the chemicals in question. In Table 4-6, the information on the five chemicals in question is
summarized. From this data set, chlorophos was selected as the index compound to which the
other four will be standardized. This selection was made based upon the availability of an
extensive body of data defining the nature of the effects and dose response of the compound in a
Table 4-6. Characterization of the toxicologic
properties of five cholinesterase-inhibiting chemicals
Chemical
Alphaphos
Betaphos
Chlorophos
Ethaphos
Deltaphos
Study ED10
(mg/kg/day)
1.0
10.0
0.3
0.06
1.5
Test
species
Rat
Rat
Rat
Dog
Human
Duration
of critical
study
90 days
2 years
2 years
1 week
24 hours
Data set characteristics
Poor. Few poorly documented studies.
Good. Many well-conducted and
documented studies for a broad
spectrum of endpoints in multiple
species.
Extensive. Many well-conducted and
documented studies for a broad
spectrum of endpoints in multiple
species. Human confirmation of
relevance of effects.
Good. Many well-conducted and
documented studies for a broad
spectrum of endpoints.
Limited. Few studies but well-
conducted.
-117-
-------
number of species, and clearly relating the effects in test species to humans. The data sets for
the other compounds were not as extensive or well documented. In one case, only a few poor-
quality dose-response studies were available, although they provided an acceptable basis for
calculating an RPF. (Despite their limitations, these studies were judged to be useful for the
development of the RPF. They were judged to provide a better basis for assessing risk than the
use of other simple assumptions such as the toxicity of these compounds is equal to the index
compound, i.e., RPF = 1, or their toxicity is negligible, i.e., RPF = 0.) The data sets for each
compound must next be evaluated to determine the critical study and effect levels that will be
used for calculating the RPF. Often, this may be the same as the basis for the RfD.
Using chlorophos as the index compound, the RPF for each of the chemicals can be
calculated. This can be done by dividing the ED10 (U.S. EPA, 1996e) for the critical study of
chlorophos by the ED10 derived from the critical study for each compound. The results of this
calculation for the example data are presented in Table 4-7.
In the example provided, the goal of the assessment is to determine the total risk of
cholinesterase inhibition due to these five compounds in foods as result of their use as
insecticides on crops. Data on the concentrations of each of the chemicals in foods are available
and are also presented in Table 4-7. However, the information is compound specific and cannot
be directly combined. Using the calculated RPFs, the exposures for each of the chemicals are
Table 4-7. Relative potency factors and equivalent exposures for five
cholinesterase-inhibiting chemicals
Chemical Study ED10
||(mg/kg/day)
Alphaphos ||l.O
Betaphos 10.0
Chlorophos 0.3
Ethaphos ||o.06
Deltaphos ||o.l 5
Total |
Percentage of RPF - predicted
with the index compound
1 Relative
potency factor
|o.3
|o.03
Jl
J5
I2
1
toxicity associated
1 Exposure
(mg/kg/day)
|o.!5
|o.02
|o.25
|o.05
|o.!5
J|
J
1 Chlorophos equivalent
exposure (mg/kg/day)
|o.05
|6E-4
|o.3
|o.3
|o.3
|o.95
32%
-118-
-------
normalized to chlorophos-equivalent exposures. These exposures can then be combined and
compared to a chlorophos-based regulatory endpoint such as an RfD.
A number of simplifying assumptions and issues are evident in this example:
The first is that the points of departure (here, ED10) for the dose-response curves
of the five chemicals in question are the most significant in determining their
relative behavior. This assumes that the slope and shape of each curve will not be
of significance because exposures will generally be low, and the accompanying
effects will occur below or near the points of departure for each chemical.
Another issue is that the studies used in calculating the RPFs were conducted in
more than one species. The example provided combines these data assuming that
interspecies differences will not be of concern. This assumption should be
assessed in selecting appropriate data for calculating RPFs to ensure that
interspecies differences do not bias the outcome of the assessment. Where
interspecies variability is marked, all the RPFs should be calculated using data
from a single species to the extent possible.
The durations of the studies used in the example to calculate RPFs were different,
ranging from a single day to 2 years. This example assumes that the effects of
concern (or the exposures) are not cumulative over time. Where there is evidence
that effects are cumulative, studies used for calculating RPFs should be of similar
duration.
If the risk manager is interested in potential effects of exposures to these
compounds other than cholinesterase inhibition (e.g., carcinogenicity,
nephrotoxicity, and hepatotoxicity), then a separate assessment needs to be
developed.
4.5. RESPONSE ADDITION
4.5.1. Background
Response addition is usually applied when the mixture components are assumed to be
lexicologically independent (see Section 4.1.1.2) and when exposure to one chemical has no
influence on the likelihood or extent of toxicity caused by a second chemical. Such a condition
is highly dependent on the exposure levels and may also depend on the route of exposure. The
following discussion assumes that information supporting toxicological independence is
available for the exposure scenario being assessed or that the extrapolation is justified.
When two chemicals cause different kinds of toxicity, or induce effects in different
organs, they may be candidates for response addition, where the responses as probabilities of
toxic effects are combined. There are two applications of response addition with somewhat
-119-
-------
different calculations: likelihood of an individual showing toxic effects (see Section 4.5.2), and
the proportion of a population showing toxic effects (see Section 4.5.3). Because the population
aspect is different from the physiological independence discussed earlier, both issues need to be
addressed when assessing population risk. A key concept with both applications is functional
independence: whether exposure to chemical A has any influence on the toxicity produced by
exposure to chemical B.
For joint exposures to one individual, the concern is whether the two chemicals cause
toxicity by different processes, such as different target organs or different modes of action in the
same organ. The response measure must be the probability of a specific toxic effect. When
applied to an individual, the assumption for response addition is that the two chemicals produce
toxicity independently.
For joint exposures to a population, a different issue is whether the chemicals cause
toxicity to the same proportion of the population (U.S. EPA, 1990). The tolerance distribution
for chemical A shows the proportion of individuals responding as the exposure level of A
increases. For example, consider the simplest mixture of only two chemicals. If the two
chemicals' tolerance distributions are perfectly correlated, then the ordering of individual
sensitivities is the same for both chemicals, i.e., the individual most sensitive to chemical A is
also most sensitive to chemical B. The most toxic chemical then produces the toxic response
first in any of the individuals exposed. Although the severity of the toxicity may be exacerbated
by the second (less toxic) chemical, the number of individuals responding is determined only by
the most toxic chemical. This issue and the limitations in addressing population risk based on
correlations of tolerance distributions are discussed more fully in Section 4.5.3.
Few empirical studies have evaluated response addition in any depth, but the concepts
they address suggest possible research directions. Of the few studies at low exposure levels that
have modeled joint toxic effects as probabilities, most consider cancer, obviously influenced by
the much wider availability of response data for cancer when compared to other kinds of toxicity.
In a conceptual investigation of the performance of both the multistage model and the two-stage
clonal expansion model for carcinogenesis, assuming an experiment using a balanced 2x2
design with 50 animals per dose group and a strong synergistic interaction, NRC (1988, p. 193)
concluded that if the exposure to one or both agents is lowered by two orders of magnitude from
the experimental doses, the assumption of response additivity "is reasonably good" in predicting
the true mixture response.
Gibb and Chen (1986) also considered implications of the multistage model. They
showed that at low doses, the risks are additive for carcinogens acting on the same stage, whereas
the hazard functions are multiplied when calculating risks for carcinogens acting on different
stages. Brown and Chu (1988) show for the multistage model that partial lifetime exposures to
-120-
-------
two carcinogens lead to roughly additive relative risks. For the two-stage clonal expansion
model, Kodell et al. (1991) argue that". . .the mixture risk is roughly additive at low doses. . . ."
The primary requirements for response addition are the availability of data on population
fraction or percent response, and the assumption of functional independence. The other major
assumptions often used by EPA, the assumption of no threshold dose, low-dose linearity, and
interspecies scaling by body allometry, are not relevant to the premise of independence, although
they certainly may play a role in estimating the magnitude of an interaction. To simplify the
discussion, the following will address the case of the binary mixture, i.e., chemicals A and B.
4.5.2. Individual Toxicity
When an individual is exposed to two chemicals, A and B, there is the potential for A to
affect the toxicity of B, and vice versa. When the toxicity of each chemical is totally described
by its own exposure level, the two chemicals are said to be independent. The interpretation is that
chemicals A and B may cause some toxicity in the individual, but the presence of A (and its
toxicity) has no influence on the toxicity of B, and similarly, B has no influence on the toxicity of
A. In this context, the two concepts of functional (or physiological) independence and statistical
independence are consistent.
In the case where the toxicity of the two chemicals is the same type, say abnormal liver
function, then the estimated mixture response may be expressed in terms of general abnormal
liver function. At high doses, there may be physiological interactions between two different
toxicities. At low doses, especially when the affected tissues are physically separated and only a
small fraction of the tissue is damaged, the assumption of independence may hold. As shown by
Feron et al. (1995), toxicity within the same target organ but of different modes of action may
indicate independent processes (response addition) or similar processes (dose addition), or even
some intermediate characterization.
When the component effects are of minor severity, independence for different modes of
action seems plausible. One must be cautious about assuming independence in the same target
organ and then concluding that two minor effects are minor in the aggregate. If an organ is
compromised twice, its function may be worse than from exposure to either chemical alone.
When information is lacking on joint effects in the same organ, a conservative approach is to
assume dose addition.
Independence in quantitative risk assessment is often used when determining the
probability of an adverse effect from exposure to multiple chemicals. If the toxicity measure is
-121-
-------
the probability of an individual incurring toxic damage, then independence can be expressed by
the probabilistic definition:
Pm = 1 - (1 - Pl)*(l - Pi) (4-19)
where pm is the expected response from exposure to the mixture, and px and p2 are the responses
from exposure to chemicals A and B, respectively. This equation says that the response to the
mixture (caused by chemical A or B) is 1 minus the probability of not responding to either
chemical. Expanding the right-hand side and including the exposure levels dx and d2 for
chemicals A and B, respectively, one obtains:
pm(di, di) = pi(di) + pi(di) - pi(di) * p2(d2) (4-20)
In general, the formula is:
(4-21)
p. = 1 - (1 - pi)*(l - p*)*(l - pi)*...
or in more compact notation:
The product on the right-hand side is the probability under independence of not responding to
any of the chemicals. The second form of the formula (Equation 4-22) then clearly shows that
the probability of responding to the mixture is just 1 minus the probability of not responding to
any of the component chemicals.
Example. Applying this to a large number of chemicals (40), each posing a very small
risk (3 x 1Q-5),
No. of chemicals 40
Single-chemical risk 3 x 10"5
Mixture risk 1 - (1 - 3 x 10'5 )40 = 1 x 10'3
4.5.3. Population Toxicity
The dose-response assessment is different when considering the entire population of
exposed individuals. The risk is often then presented as the percent responding in the population.
Independence is not a matter of physiological interactions within an individual, but is based on
the correlation of tolerances for the two chemicals (see U.S. EPA, 1990, for an extended
-122-
-------
discussion). The tolerance distribution for any given chemical is the proportion of people
responding as the exposure level of that chemical increases.
For exposure to two chemicals, A and B, the ordering of the individual sensitivities to
chemical A is the same as the ordering for chemical B, then the tolerances for the two chemicals
are perfectly correlated (r = 1), and the most toxic chemical will elicit the response first:
(4-23)
pM = pi if r = 7 and pi < pi
If, on the other hand, the individual least sensitive to chemical A is most sensitive to chemical B,
and so on throughout the range of sensitivity, then the chemicals have perfect negative
correlation (r = -1) of tolerances, and the mixture response is:
pm = min(pi + pi, I) ifr= -I (4-24)
When the correlation is zero (r = 0), i.e., the ordering of the individuals showing toxic
effects from chemical A has no apparent relationship with the individuals showing toxic effects
from chemical B, then the two chemicals are said to act independently on the population. We
then have the familiar model for statistical independence:
pm = pi + pi - (pipi) iff = 0 (4_25)
Equation 4-25 is the same model described above for toxicologic independence in a single
exposed individual.
The response-addition formula for populations has limited use in risk assessment. First, it
is more complicated than the formula for the individual, because the tolerance correlation can be
any value from -1 to +1, and so requires more detailed data on the exposed population of
concern. In addition, the concepts of tolerance correlation only work well if there are two
chemicals in the mixture. For example, if a mixture has three chemicals, then the correlation of
tolerances must consider the three possible pairs of chemicals. No methods have been found for
using pairwise tolerance correlations in higher complexity mixtures. Also, some correlation
values cannot be applied to three or more chemicals. For example, tolerances of three chemicals
cannot all be negatively correlated with each other. The well-studied cases using tolerance
correlations are those discussed in this section. Consequently, response addition for populations
is not further developed in this guidance document.
-123-
-------
4.5.4. Application
Response addition is easily misinterpreted because of the appearance of accuracy and
precision given by the use of numbers to represent the risk of toxic effects. In contrast, it is
hoped that the HI is less likely to be over interpreted because it only indicates a rough level of
concern, not a probability or population count. Errors may arise from improper use of response
addition because of a lack of independence. With a mixture of a large number of chemicals, it is
particularly easy to overlook the influence of a poor-quality response estimate. Mixture
assessments based on response addition must include quality descriptions for each component's
response estimate.
For single-chemical responses Pi, p2, ..., response addition applied to the risk to an
individual is often approximated by the simple summation:
PSUM = '"' rf? J= n''rf'J + n?ia?i+... (4-2O)
For mixtures of a few chemicals and very small p, this approximation may be acceptable. For
mixtures with a large number of component chemicals or chemicals whose response is not small,
the full independence formula (Equation 4-21) should be used. For example, with a simple
mixture of only 16 chemicals, if each has a response of 0.02, the relative error is 16% (sum in
Equation 4-26 gives 0.32, true response from Equation 4-21 is 0.28). Because of the availability
of computers, the full formula (Equation 4-21) is easily implemented and should be used.
The other concern with a large number of chemicals in the mixture is that one poorly
studied chemical may dominate the response estimate. An excessive response estimate could
arise from improper statistical analysis or toxicological procedures employing highly sensitive
animal species. Similar factors could also lead to response estimates that are too low, often
caused by lack of statistical power in the study design. In all cases, the risk characterization
should highlight any chemicals whose supporting information is poor, and should attempt to
characterize the numerical uncertainty caused by the poor information. For example, if only one
chemical has a highly uncertain response estimate, the mixture assessment can be calculated with
and without the suspect chemical.
For minor toxic effects, the different effects are unlikely to interact, so the response
addition formula (Equation 4-21) is probably adequate. One mixture response could then be
estimated for all renal toxicity, with another estimated for all hepatic toxicity. The mixture
assessment could then result in several separate response addition estimates, one per effect or
target organ. For levels causing moderate toxicity, there is insufficient information to allow
predictions of the likelihood of physiological interactions between affected target organs. For
high-exposure estimates, additive formulas are not generally recommended because of the higher
-124-
-------
likelihood of toxicologic interactions (e.g., in the same tissue) among component chemicals in
the mixture as well as physiological interactions among the various affected target organs.
For low exposure levels, e.g., near the individual chemical NOAELs from well-designed
studies, lexicologically dissimilar chemicals are not generally expected to interact lexicologically
or physiologically, and can be assumed to be functionally independent. For the special case
where all component chemicals have RfDs or RfCs and the exposure levels of dissimilar
components are well below their respective RfDs or RfCs, the risk of toxicity can usually be
assumed to be negligible.
Example. Consider an oral exposure to three lexicologically independent chemicals, each
close to but below its RfD. The following calculations result:
Chemical
A
B
C
Exposure
13
7
22
RfD
16
8
24
Risk
0
0
0
Mixture risk = 0
In this example, 0 is used to denote a risk that is either subthreshold (a true zero risk) or small
enough to be generally considered virtually safe. In general, this kind of rough evaluation should
be limited to mixtures with a small number of chemical components. When the number of
chemicals in the mixture is large, even when all individual exposures are below their RfDs, the
toxicity data should be carefully examined to ensure that all effects and modes of action are being
considered when deciding functional independence. As the information becomes more uncertain,
such as with poor-quality RfDs or exposure estimates, any conclusion of negligible risk is
similarly uncertain and consideration should be given to obtaining better information.
4.5.5. Use of Upper Bound Response Estimates
The practice of assessing cancer risk for a mixture has usually involved applying response
addition to the lifetime excess cancer risk values available for the individual chemicals (U.S.
EPA, 1986). The common values generated by EPA are those available on the IRIS database.
Currently, most of the IRIS values for carcinogenic potency are for single chemicals and are
considered plausible upper bounds to the actual lifetime excess cancer risk. Use of such values
raises the concern that applying response addition to upper bounds will lead to unreasonably high
estimates of the actual upper bound on mixture risk. The available studies, summarized below,
suggest that for most mixtures of a few components, the risk estimates are not overly
conservative.
-125-
-------
Chen et al. (1990) and Kodell and Chen (1994) derive mathematical expressions for the
upper limit on mixture risk, but the procedures require intensive computations. Gaylor and Chen
(1996) extend this discussion and derive a simple approximation to the upper limit on the
mixture risk that can be more appropriate than the simple summing of component upper bounds.
The numerical consequences of Kodell and Chen (1994) suggest that the error in the simple
addition of component upper bounds is small compared to other uncertainties. For example, a
hypothetical example of four chemicals showed that the largest error from using the simple sum
of upper bounds occurred when all chemicals were roughly equal contributors to the mixture risk.
Their proposed method for the upper 95% confidence bound of a two-chemical mixture reduced
the conservatism, but only slightly. Their mixture upper bound was 4.3 x 10"7, whereas the
simple sum of the component upper bounds was 4.9 x 10"7.
Cogliano (1997) approached the question of summing upper bounds of mixture
components' risks in two ways: (1) whether the sum yields an improbable estimate of overall risk
(that is, is it only remotely possible for the true sum of risks to match the sum of upper bounds),
and (2) whether the sum gives a misleading estimate (that is, is the true sum of risks likely to be
very different from the sum of upper bounds). Analysis of several case studies showed that as
the number of mixture components increases, summing their upper bounds yields an improbable,
but not misleading, estimate of the overall risk. Thus, although the confidence attached to the
mixture bound may exceed the confidence levels for the component chemicals, the actual
mixture risk estimate (i.e., its magnitude) is not excessively high. Cogliano concludes that
simple sums of upper bounds are a good approximation of the overall risk and can be adjusted
downward (e.g., by dividing by 2) to give a more plausible upper bound, or even a central
estimate of overall risk.
These two measures of overconservatism, the estimate and the confidence level, are also
discussed in Cullen (1994). In contrast to Cogliano's results for sums of upper bounds, Cullen
showed substantial overconservatism for products of upper bounds.
4.5.6. Qualitative Judgments of Interaction Potential
Response addition may work well for many mixtures at very low doses with components
affecting different target organs. Other mixtures, even at low doses, may show evidence of
toxicologic interaction. In the example method described in this section, the key assumption is
that interactions in those mixtures can be adequately represented as departures from response
addition. The method follows an obvious approach: to begin with the response addition formula,
and then modify its estimate to reflect the interaction results. Although several studies describe
toxicologic interaction as a departure from response addition (e.g., changes from the predicted
LD50), few studies quantify interaction, and even fewer quantitatively describe the dose
-126-
-------
dependence of the interaction. Consequently, for an approach to be able to use available data,
some qualitative procedure is needed for judging the impact of the potential toxicologic
interactions.
Carcinogen interactions are the basis for the example method that follows (Section
4.5.6.1, Equation 4-27). The modeling of carcinogenic interactions is in an early stage of
development. Consequently, the following method is not currently recommended as a
quantitative method for adjusting the mixture risk estimate. It should be considered as a possible
approach to a qualitative description of the interactions in a mixture. Because of the dominance
of binary mixtures in interaction studies, only pairwise interactions are included in the example
method. A tacit assumption is that higher order interactions are relatively minor compared to
binary interactions.
Response addition of known carcinogens may give incorrect risk estimates for
multichemical exposure when toxicologic interactions are present. These interactions can
enhance or inhibit the cancer potency or the growth or progression of altered cells. Chemicals
with individually weak evidence of carcinogen!city may, in combination, show strong potential
to initiate tumors.
The best example of human data on carcinogen interactions can be found from
epidemiologic data on mortality from lung cancer in workers with exposure to cigarette smoke
and/or asbestos. Hammond et al. (1979) noted that in comparison with the lung cancer death
rates for nonsmokers who did not have occupational exposure to asbestos, the death rate was
5.17 times higher for asbestos workers who did not smoke, 10.85 times higher for smokers who
did not work with asbestos, and 53.24 times higher for smokers who worked with asbestos.
These data indicate that death rate from lung cancer is approximately 10 times higher for
asbestos workers who smoke than those who do not (Mukerjee and Stara, 1981). Under response
addition, where the two exposures are assumed to be independent causes of lung cancer, the
expected response from the joint exposure was 169.7 lung cancer deaths per 100,000 man-years
exposure, yet the observed response was 601.6 per 100,000. Note that the exposure levels in this
example are much higher than usual ambient environmental exposures, so other instances of
synergism between carcinogenic chemicals may be much less pronounced.
This synergism between asbestos and smoking is commonly described as an example of a
multiplicative interaction (Mukerjee and Stara, 1981). This term is used because when the
numerators in the single substance death rates are multiplied, the product is roughly equal to the
numerator in the death rate for the combination (i.e., 5.17 x 10.85 is roughly equal to 53.24).
The risks, however, are not multiplied, and there seems to be no biological process that can
motivate such a multiplication of death rate numerators. Similarly, Kodell and Pounds (1991)
note that the "multiplicative model of relative risk does not have a corresponding null model in
-127-
-------
pharmacology/toxicology studies." As discussed by Greenland and Rothman (1998), there are
several definitions of interaction used in toxicology, statistics, and epidemiology, and their
interpretations vary by use as well as by the scale of the effect measure. This variety of
definitions and their comparative analyses is beyond the scope of this document, but should be
addressed by future efforts.
When interactions have been noted, the goal of risk estimation is to include carcinogenic
interactions quantitatively in the mixture risk assessment. The currently available animal
database on carcinogen interactions, and in particular on promoters, is not sufficient for
recommending a general approach for their risk assessment. For example, the slope factor for a
carcinogen is estimated using cancer incidence data in an animal bioassay. The data on
promotion action suitable for estimating the slope factor are either incomplete or nonexistent.
Most of the animal data on promoters are on the increase in the number of papillomas or on
shortening of the time to tumor. Accordingly, in the absence of an adequate database, the
individual cancer response of various constituents present in the mixture should be combined
using response-addition to estimate the response of carcinogen mixtures with promotion
activities. This response-additive default approach can be followed by incorporation of a
correction for interaction effects if any deviation from additivity is noted. For the interim period
until the adequate database is available in the scientific literature, only qualitative approaches are
recommended. In the example method described below for estimating carcinogenic risk of
mixtures (Woo et al., 1995b), qualitative judgments of the interaction potential are used to
modify a relative ranking of the mixture based on carcinogenic risk.
4.5.6.1. Use of Interaction Data on Carcinogens
For known or suspected human carcinogens, past practice at EPA has been to assume
low-dose linearity in deriving quantitative risk estimates for environmental levels of materials.
This has involved the application of mathematical models to animal bioassay or human data and
the derivation of a slope factor, usually the upper bound on a low-dose linear term from a
multistage model. The recently proposed revisions to the Guidelines for Carcinogen Risk
Assessment (U.S. EPA, 1996a) substantially alter this procedure. Under the Proposed
Guidelines, dose-response assessment and hazard identification rely on consideration of the
likely mode of action of the agent in question. Data of various types relating to mode of action
are used to inform decisions as to the shape of dose-response curves and appropriate low-dose
extrapolation. In all cases a two-step approach is taken to dose-response assessment. In the first
step, data in the observed range are modeled using a biologically based model (if applicable) or
curve-fitting procedure. The observed range can be extended through use of appropriate
information, not limited to animal or human cancers from long-term studies. In the second step,
-128-
-------
decisions are taken as to type of low-dose extrapolation. For materials for which a hypothesis of
low-dose linearity can apply, a straight line is drawn from a reasonable point of departure from
the low end of the observed range through the origin (default approach); the slope of the line
serves as the slope factor or unit risk. If it is judged that the mode-of-action data supports low-
dose nonlinearity, a margin of exposure would be calculated using the lower end of the observed
range as the point of departure.
There are many opportunities for interactions among carcinogens and between
carcinogens and modifiers. There have been many reported instances of antagonism, inhibition,
synergism, and promotion/co-carcinogenesis. These cannot currently be incorporated
quantitatively into the cancer risk estimate for a mixture using any validated process. It is
recommended that the risk assessor provide a qualitative discussion of potential for interaction
among carcinogens or between carcinogens and noncarcinogens contributing to the overall
carcinogenic process of the mixture.
There are several databases that provide information on interactions for chemical pairs
tested in carcinogenicity or related bioassays. Information on binary mixtures of carcinogens can
be found in Arcos et al. (1988), on carcinogens and inhibitors in Bagheri et al. (1988/89), and on
carcinogens and promoters in Rao et al. (1989). Information from these three sources has been
combined into a computerized system called the Integral Search System (ISS).
This system, described in Woo et al. (1994), can be used to evaluate the potential for
interactions between members of chemical pairs to affect cancer risk. This paper also describes a
procedure for calculating an interaction weighting ratio or "hazard modification" component. An
outline of this approach is presented below as an example of a published methodology that seeks
to quantify the potential influence of interactions in carcinogenic mixtures. At this time, the
outline is not recommended for quantitative risk assessment but can be further explored as a tool
for qualitatively characterizing the potential influence of the interactions.
Woo et al. (1994) calculate (by response addition) a value by which they describe the
"inherent hazard" of the mixture, an estimate of its carcinogenic potential. They then generate all
possible binary pairs of chemicals in the mixture and search the databases for interaction "hits"
or reported instances of interactions, which may either enhance (synergism, promotion/cocar-
cinogenesis) or reduce (antagonism, inhibition) carcinogenic potential. The authors also infer
interactions for pairs not in their databases by using a mathematical procedure based on
association with chemical classes of structurally or functionally related chemicals. Information
on both inferred and reported interactions is used in the calculation of the weighting ratio (WR),
which is given by the following formula:
-129-
-------
WR - (4-27)
where p, q, r, and s are "hazard-modification effectiveness coefficients" that reflect the
effectiveness of the four types of combination effects to modify the carcinogenicity of chemicals:
HSyn = observed plus inferred instances of synergism between chemical pairs in
the mixture,
HPro = observed plus inferred instances of promotion between chemical pairs in
the mixture,
H^, = observed plus inferred instances of antagonism between chemical pairs in
the mixture, and
HInh = observed plus inferred instances of inhibition between chemical pairs in
the mixture.
The authors give numerical values for the "hazard-modification effectiveness coefficients" based
both on their scientific judgment and on inspection of the combination effects literature
encompassed in their databases. A WR of 1 would suggest that the additivity assumption is
reasonable. A high or low WR would suggest that the overall interaction tends to deviate from
additivity with a predominant hazard-enhancing or hazard-reducing interaction effect,
respectively.
This methodology does not have the full formality of the interaction-based HI approach
described in Section 4.3. Furthermore, it is not applied to the common unit risk or its
counterpart. It is based on a particular literature database and may not generalize to other
chemical classes. The potential of this and other approaches to risk assessment that incorporate
toxicologic interaction is discussed more fully in Section 2.7, Future Directions.
-130-
-------
REFERENCES
Ahlborg, UG; Becking, G; Birnbaum, L; et al. (1994) Toxic equivalency factors for dioxin-like
PCBs. Report on a WHO-ECEH and IPCS consultation, December, 1993. Chemosphere
28(6): 1049-1067.
Albert, RE. (1985) The comparative potency method: an approach to quantitative cancer risk
assessment. In: Methods for estimating risk of chemical injury: humans and non-human biota
and ecosystems, Vouk, VB; Butler, GC; Hoel, DG; Peakall, DB; eds. pp. 281-287. SCOPE
43:305-325.
Albert, RE; Chen, C. (1986) U.S. EPA diesel studies on inhalation hazards. In: Carcinogenic
and mutagenic effects of diesel engine exhaust, Ishinishi, N; Koizumi, A; McClellan, RO;
Stober, W; eds. New York:Elsevier, p. 411-419.
Albert, RE; Peterson, HT; Bohning, DE; Lippmann, M. (1975) Short-term effects of cigarette
smoking on bronchial clearance in humans. Arch Environ Health 30:361-367.
Albert, RE; Lewtas, J; Nesnow, S; Thorslund, TW; Anderson, E. (1983). Comparative potency
method for cancer risk assessment application to diesel particulate emissions. Risk Anal
3:101-117.
Amin, S; Huie, K; Hech, S. (1985) Mutagenicity and tumor initiating activity of methylated
benzo[b]fluoranthene. Carcinogenesis 6(7): 1023-1025.
Andersen, ME; Gargas, ML; Clewell, HJ, IE; Severyn, KM. (1987) Quantitative evaluation of
the metabolic interactions between trichloroethylene and 1,1-dichloroethylene in vivo using gas
uptake methods. Toxicol Appl Pharmacol 89:149-157.
Arcos, JC; Woo, YT; Lai, DY. (1988) Database on binary combination effects of chemical
carcinogens. Environ Carcinogen Revs (Part C of J Environ Sci Health) Vol. C6[l], p. 1-164.
Atkinson, R. (1994) Gas-phase tropospheric chemistry of organic compounds. J Phys Chem
RefData. Monograph No. 2.
Bagheri, D; Doeltz, MK; Fay, JR; Helmes, CT; Monasmith, LA; Sigman, CC. (1988/89)
Database of inhibitors of carcinogenesis. Environ Carcinogen Revs (Part C of J Environ Sci
Health: Vol. C6(3): 1-159.
Barnes, D; Alford-Stevens, A; Birnbaum, L; Kutz, F; Wood, W; Patton, D. (1991) Toxicity
equivalency factors for PCBs. Quality Assurance, Good Practice, and Law 1(1):70-81.
Berenbaum, MC. (1989) What is synergy? Pharmacol Rev 41:93-141.
-131-
-------
Bidleman, TF; Christensen, EJ; Harder, HW. (1981) Aerial deposition of organochlorines in
urban and coastal South Carolina. In: Atmospheric pollutants in natural waters. Eisenreich, SJ,
ed. Ann Arbor, MI: Ann Arbor Science Publishers Inc., pp. 481-508.
Birnbaum, L; DeVito, M. (1995) Use of toxicity equivalency factors for risk assessment for
dioxins and related compounds. Toxicology 105(2-3):391-401.
Birnbaum, LS; Harris, MW; Harris, JC; Lamb, JC; Mckinney, JD. (1985) Toxic interaction of
specific polychlorinated biphenyls and 2,3,7,8-tetrachlorodibenzo-p-dioxin: increased incidence
of cleft palate in mice. Toxicol Appl Pharmacol 77:292-302.
Blazka, ME; Shaikh, ZA. (1992) Cadmium and mercury accumulation in rat hepatocytes:
interactions with other metal ions. Toxicol Appl Pharmacol 113:118-125.
Bliss, CI. (1939) The toxicity of poisons applied jointly. Ann Appl Biol 26:585-615.
Bremner, I. (1993) Involvement of metallothionein in the regulation of mineral metabolism. In:
Metallothionein in. Suzuki, KT; Imura, N; Kimura, M, eds. Boston: Birkhauser Verlag, pp.
111-124.
Brown, CC; Chu, KC. (1988) Additive and multiplicative models and multistage carcinogenesis
theory. Risk Anal 9(1):99-105.
Brunner, MJ; Sullivan, TM; Singer, AW; et al. (1996) An assessment of the chronic toxicity
and oncogenicity of Aroclor-1016, Aroclor-1242, Aroclor-1254, and Aroclor-1260 administered
in diet to rats. Battelle Study No SC920192, Columbus, OH. Vol 1-2.
Cassee, FR; Groten, JP; van Bladeren, PJ; Feron, VJ. (1998) Toxicological evaluation and risk
assessment of chemical mixtures. CritRevs Toxicol 28(1):73-101.
Cerklewski, FL; Forbes, RM. (1976) Influence of dietary zinc on lead toxicity in rats. JNutr
106:689-696.
Chen, JJ;, Gaylor, DW; Kodell, RL. (1990) Estimation of the joint risk from multiple-
compound exposure based on single-compound experiments. Risk Anal 10(2):285-290.
Cogliano, VJ. (1997) Plausible upper bounds: Are their sums plausible? Risk Anal
17(l):77-84.
Cogliano, VJ. (1998) Assessing the cancer risk from environmental PCBs. Environ Health
Perspect 106(6):317-323.
Crawford-Brown, D. (1997) Chapter 2. Linking the population to the environment. In:
Theoretical and mathematical foundations of human health risk analysis: biophysical theory of
environmental health science. Norwell, MA: Kluwer Academic Publishers, pp. 25-52.
-132-
-------
Creasia, DA; Poggenburg, JK, Jr; Nettesheim, P. (1976) Elution of benzo(a)pyrene from carbon
particles in the respiratory tract of mice. J Toxicol Environ Health 1:967-975.
Cullen, A. (1994) Measures of compounding conservatism in probabilistic risk assessment.
Risk Anal 14:389-393.
Davis, D; Safe, S. (1989) Dose-response immunotoxicities of commercial polychlorinated
biphenyls (PCBs) and their interaction with 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol Lett
48:35-43.
Dourson, ML; Clark, JM. (1990) Fish consumption advisories: toward a unified, scientifically
credible approach. Regul Toxicol Pharmacol 12:165-172.
Durkin, PR; Hertzberg, RC; Stiteler, W; Mumtaz, M. (1995) The identification and testing of
interaction patterns. Toxicol Lett 79:251-264.
Eastern Research Group, Inc. (September 10, 1999) Report of the Peer Review Workshop on the
Guidance for Conducting Health Risk Assessments of Chemical Mixtures. Contract Number 68-
C6-0041. Lexington, MA
Eisenhans, B; Schumann, K; Forth, W. (1991) Toxic metals: interactions with essential metals.
In: Nutrition, toxicity, and cancer. Rowland, IR, ed. Boca Raton, FL: CRC Press, pp. 223-258.
Ekwall, P; Ermala, P; Setala, K; Sjoblom, L. (1951) Gastric absorption of 3,4-benzypyrene. n.
The significance of the solvent for the penetration of 3,4-benzpyrene into the stomach wall.
Cancer Res 11:758-763.
El-Masri, HA; Thomas, RS; Benjamin, SA; Yang, RS. (1995) Physiologically based
pharmacokinetic/pharmacodynamic modeling of chemical mixtures and possible applications in
risk assessment. Toxicology 105(2-3):275-282.
El-Masri, HA; Tessari, JD; Yang, RSH. (1996) Exploration of an interaction threshold for the
joint toxicity of trichloroethylene and 1,1-dichloroethylene: utilization of PB-PK model. Arch
Toxicol 70:527-539.
Enslein, K; Blake, BW; Borgstedt, HH. (1990) Prediction of probability of carcinogenicity for a
set of ongoing NTP bioassays. Mutagenesis 5:305-306.
Ferin, J; Leach, LJ. (1973) The effect of SO2 on lung clearance of TiO2 particles in rats. Am Ind
HygAssoc 134:260-263.
Feron, VJ; Groten, JP; Jonker, D; Cassess, FR; van Bladeren, PJ. (1995) Toxicology of
chemical mixtures: challenges for today and the future. Toxicology 105:415-427.
-133-
-------
Fingerling, G; Hertkorn, N; Parlar, H. (1996) Formation and spectroscopic investigation of two
hexachlorobornanes from six environmentally relevant toxaphene components by reductive
dechlorination in soil under anaerobic conditions. Environ Sci Technol 30(10):2984-2992.
Finney, DJ. (1971) Probit analysis. 3rd ed. Cambridge: Cambridge University Press.
Foulkes, EC. (1991) Role of metallothionein in epithelial transport and sequestration of
cadmium. In: Metallothionein in biology and medicine. Klassen, CD; Suzuki, KT, eds. Boca
Raton, FL: CRC Press, pp. 171-182.
Gandolfi, AJ; Brendel, IK; Fisher, RL; Michaud, J-P. (1995) Use of tissue slices in chemical
mixtures toxiciology and interspecies investigations. Toxicology 105:285-290.
Gaylor, DW; Chen, JJ. (1996) A simple upper limit for the sum of the risks of the components
in a mixture. Risk Anal 16(3):395-398.
Gaylor, DW; Axelrad, JA; Brown, RP; et al. (1997) Health risk assessment practices in the U.S.
Food and Drug Administration. Regul Toxicol Pharmacol 26:307-321.
Gennings, C. (1995) An efficient experimental design for detecting departure from additivity in
mixtures of many chemicals. Toxicology 105:189-197.
Gennings, C; Carter, WH, Jr. (1995) Utilizing concentration-response data from individual
components to detect statistically significant departures from additivity in chemical mixtures.
Biometrics 51:1264-1277.
Gennings, C; Schwartz, P; Carter, WH, Jr; Simmons, JE. (1997) Detection of departures from
additivity in mixtures of many chemicals with a threshold model. J Agric Biol Environ Stat
2(2): 198-211.
Gerrity, T; Henry, C, eds. (1990) Summary report of the workshops on principles of
route-to-route extrapolation for risk assessment. In: Principles of route-to-route extrapolation for
risk assessment. Proceedings of the workshops, March and July: Hilton Head, SC, and Durham,
NC. New York: Elsevier Science Publishing Co., Inc., pp. 1-12.
Gessner, PK. (1995) Isobolographic analysis of interactions: an update on applications and
utility. Toxicology 105(2-3): 161-179.
Gibb, HJ; Chen, CW. (1986) Multistage model interpretation of additive and multiplicative
carcinogenic effects. Risk Anal 6:167-170.
Hammond, EC; Selikoff, IJ; Seidman, H. (1979) Asbestos exposure, cigarette smoking and
death rates. Ann NY Acad Sci 330:473-490.
-134-
-------
Harper, N; Connor, K; Steinberg, M; Safe, S. (1995) Immunosuppressive activity of
polychorinated biphenyl mixtures and congeners: nonadditive (antagonistic) interactions.
Fundam Appl Toxicol 27(i): 131-139.
Harris, M; Zacharewski, T; Safe, S. (1993) Comparative potencies of Aroclors 1232, 1242,
1248, 1254, and 1260 in male Wistar rats: assessment of the toxic equivalency factor (TEF)
approach for polychlorinated biphenyls (PCBs). Fundam Appl Toxicol 20:456-463.
Henry, MC; Kaufman, DG. (1973) Clearance of benzo(a)pyrene from hamster lungs after
administration on coated particles. J Natl Cancer Inst 51:1961 -1964.
Hertzberg, RC. (1996) Risk assessment of truly complex mixtures using EPA's almost
proposed guidelines. Presented at the Health Effects Institute Annual Conference, Session on
"Vehicle Emissions as a Complex Mixture," April 28-30, 1996, Asheville, NC.
Hertzberg, RC; Rice, G; Teuschler, LK. (1999) Methods for health risk assessment of
combustion mixtures. In: Hazardous waste incineration: evaluating the human health and
environmental risks. Roberts, S; Teaf, C; Bean, J, eds. Boca Raton, FL: CRC Press, pp. 105-
148.
Heywood, R. (1981) Target organ toxicity. Toxicol Lett 8:349-358.
Heywood, R. (1983) Target organ toxi city II. Toxicol Lett 18:83-88.
Hoadley, JE; Johnson, DR. (1987) Effects of calcium on cadmium uptake and binding in the rat
intestine. Fundam Appl Toxicol 9:1-9.
Ikeda, M. (1988) Multiple exposure to chemicals. Regul Toxicol Pharmacol 8:414-421.
Jones, DP; Brown, LA; Sternberg, P. (1995) Variability in glutathione-dependent detoxication
in vivo and its relevance to detoxication of chemical mixtures. Toxicology 105:267-274.
Kedderis, GL. (1990) The role of the mixed-function oxidase system in the toxication and
detoxication of chemicals: relationship to chemical interactions. In: Toxic interactions.
Goldstein, RS; Hewitt, WR; Hook, JB, eds. New York: Academic Press, pp. 31-60.
Kodell, RL; Pounds, JG. (1991) Assessing the toxicity of mixtures of chemicals. In: Statistics
in toxicology. Krewski, D; Franklin, C, eds. New York: Gordon and Breach Science, pp.
559-591.
Kodell, RL; Chen, JJ. (1994) Reducing conservatism in risk estimation for mixtures of
carcinogens. Risk Anal 14(3):327-332.
Kodell, RL; Krewski, D; Zielinski, JM. (1991) Additive and multiplicative relative risk in the
two-stage clonal expansion model of carcinogenesis. Risk Anal 11(3):483-490.
-135-
-------
Krishnan, K; Brodeur, J. (1991) Toxicological consequences of combined exposure to
environmental pollutants. Arch Comp Environ Stud 3(3): 1-106.
Krishnan, K; Brodeur, J. (1994) Toxic interactions among solvents and environmental
pollutants: corroborating laboratory observations with human experience. Environ Health
Perspect 102(Suppl. 1):103-108.
Krishnan, K; Pelekis, M. (1995) Hematotoxic interactions: occurrence, mechanisms and
predictability. Toxicology 105:355-364.
Krishnan, K; Clewell, JH, IE; Anderson, ME. (1994) Physiologically-based pharmacokinetic
modeling of chemical mixtures. In: Toxicology of chemical mixtures. Yang, RSH, ed. New
York: Academic Press, pp. 399-437.
Kristensen, MB. (1976) Drug interactions and clinical pharmacokinetics. Clin Pharmacokinet
1:351-372.
Levin, ED; Schantz, SL; Bowman, RE. (1988) Delayed spatial alternation deficits resulting
from perinatal PCB exposure in monkeys. Arch Toxicol 62:267-273.
Lewtas, J. (1985) Development of a comparative potency method for cancer risk assessment of
complex mixtures using short-term in vivo and in vitro bioassays. Toxicol Ind Health 1:193-203.
Lewtas, J. (1988) Genotoxicity of complex mixtures: strategies for the identification and
comparative assessment of airborne mutagens and carcinogens from combustion sources.
Fundam Appl Toxicol 10:571-589.
Lloyd, JW. (1971) Long-term mortality study of steel workers. V. Respiratory cancer in coke
plant workers. J Occup Med 13(2):53-68.
Lloyd, JW; Lundin, FE, Jr.; Redmond, CK; Geiser, PB. (1970) Long-term mortality study of
steelworkers. IV. Mortality by work area. J Occup Med 12(5):151-157.
Lorber, M; Cleverly, D; Schaum, J. (1996) A screening level risk assessment of the indirect
impacts from the Columbus waste to energy facility in Columbus, Ohio. Proceedings of an
International Specialty Conference, sponsored by the Air and Waste Management Association
and the U.S. Environmental Protection Agency, held April 18-21, 1996, in Washington, DC. In:
Solid waste management: thermal treatment and waste-to-energy technologies, VIP - 53.
Pittsburgh: Air and Waste Management Association, pp. 262-278.
Lorber, M; Pinsky, P; Gehring, P; Braverman, C; Winters, D; Sovocool, W; (1998)
Relationships between dioxins in soil, air, ash, and emissions from a municipal solid waste
incinerator emitting large amounts of dioxins. Chemosphere 37:2173-2197.
Lyman, WJ; Reehl, WF; Rosenblatt, DH. (1990) Handbook of chemical property estimation
methods. Washington, DC: American Chemical Society.
-136-
-------
MacDonald, JR; Gandolfi, AJ; Sipes, IG. (1982) Acetone potentiation of 1,1,2-trichloroethane
hepatotoxicity. Toxicol Lett 13:57-69.
Marnicio, RJ; Hakkinen, PJ; Lutkenhoff, SD; Hertzberg, RC; Moskowitz, PD. (1991) Risk
analysis software and databases: review of Riskware '90 conference and exhibition. Risk Anal
11:545-560.
Mayes, BA; McConnell, EE; Neal, BH; et al. (1998) Comparative carcinogenicity in Sprague-
Dawley rats of the polychlorinated biphenyl mixtures, Aroclors 1016, 1242, 1254, and 1260.
Toxicol Sci 41(l):62-76.
Mazumdar, S; Redmond, C; Sollecito, W; Sussman, N. (1975) An epidemiological study of
exposure to coal tar pitch volatiles among coke oven workers. J Air Pollut Control Assoc
25(4):382-289.
McFarland, VA; Clarke, JU. (1989) Environmental occurrence, abundance, and potential
toxicity of polychlorinated biphenyl congeners: considerations for a congener-specific analysis.
Environ Health Perspect 81:225-239.
McKone, JE. (1990) Dermal uptake of organic chemicals from a soil matrix. Risk Anal 10:407-
419.
Mirsatari, SG; McChesney, MM; Craigmill, AC; Winterlin, WL; Seiber, JN. (1987) Anaerobic
microbial dechlorination: an approach to on-site treatment of toxaphene-contaminated soil. J
Environ Sci Health [B] 22(6):663-690.
Moller, BB; Ekelund, B. (1985) Toxicity of cyclosporine during treatment with androgens. N
Engl JMed313:1416.
Morgan, KT; Patterson, DL; Gross, EA. (1984) Frog plate mucillary apparatus: structure,
function and response to formaldehyde gas. Fundam Appl Toxicol 4:58-68.
Mukerjee, D; Stara, JF. (1981) Asbestos and its threats. Chem Week 129:5.
Mukerjee, D; Cleverly, D. (1987) Risk from exposure to polychlorinated dibenzo-p-dioxins and
dibenzofurans emitted from municipal incinerators. Waste Manage Res 5:269-283.
Mumtaz, MM; Durkin, PR. (1992) A weight-of-evidence scheme for assessing interactions in
chemical mixtures. Toxicol Ind Health 8:377-406.
Mumtaz, MM; Hertzberg, RC. (1993) The status of interactions data in risk assessment of
chemical mixtures. In: Hazard assessment of chemicals. Saxena, J, ed. Washington, DC:
Hemisphere Publishing Corp. Vol. 8, pp. 47-79.
-137-
-------
Mumtaz, MM; Knauf, LA; Reisman, DJ; et al. (1995) Assessment of effect levels of chemicals
from quantitative structure-activity relationship (QSAR) models. I. Chronic lowest-observed-
adverse-effect level (LOAEL). Toxicol Lett 79:131-143.
Mumtaz, MM; Poirier, KA; Coleman, JT. (1997) Risk assessment for chemical mixtures: fine-
tuning the hazard index approach. J Clean Technol, Environ Toxicol Occup Med 6(2): 189-204.
Mumtaz, MM; DeRosa, CT; Groten, J; et al. (1998) Estimation of toxicity of chemical mixtures
through modeling of chemical interactions. Environ Health Perspect 106(6): 1353-1360.
Murphy, SD. (1980) Assessment of the potential for toxic interactions among environmental
pollutants. In: The principles and methods in modern toxicology. Galli, CL; Murphy, SD;
Paoletti, R, eds. Amsterdam: Elsevier/North Holland Biomedical Press.
Muska, CF; Weber, LJ. (1977) An approach for studying the effects of mixtures of
environmental toxicants in whole organism performance. In: Recent advances in fish toxicology.
Taub, RA, ed. EPA/600/3-77/085, pp. 71-87.
Nash, RG; Woolson, EA. (1967) Persistence of chlorinated hydrocarbon insecticides in soils.
Science 157:924-927.
NATO/CCMS (North Atlantic Treaty Organization, Committee on the Challenges of Modern
Society). (1988) Scientific basis for the development of international toxicity equivalency factor
(I-TEF) method of risk assessment for complex mixtures of dioxins and related compounds.
Report No. 178.
Nesnow, S. (1990) Mouse skin tumours and human lung cancer: relationships with complex
environmental emissions. In: Complex mixtures and cancer risk. IARC Scientific Publ
104:44-54.
Nesnow, S. (1991) Multifactor potency scheme for comparing the carcinogenic activity of
chemicals. Environ Health Perspect 96:17-21.
Nesnow, S; Lewtas, J. (1991) Interspecies sensitivity to chemical carcinogens: relationship
between mouse skin tumors and human lung tumors. Environ Carcinogen Ecotoxicol Rev
C9(2):229-259.
Neuvonen, PJ. (1976) Interactions with the absorption of tetracyclines. Drugs 11:45-54.
Norback, DH; Weltman, RH. (1985) Polychlorinated biphenyl induction of hepatocellular
carcinoma in the Sprague-Dawley rat. Environ Health Perspect 60:97-105.
NRC (National Research Council). (1983) Risk assessment in the federal government:
managing the process. Committee on the Institutional Means for Assessment of Risks to Public
Health, Commission on Life Sciences, NRC. Washington, DC: National Academy Press.
-138-
-------
NRC (National Research Council). (1988) Complex mixtures: methods for in vivo toxicity
testing. Washington, DC: National Academy Press.
NRC (National Research Council). (1994) Science and judgment in risk assessment. Committee
on Risk Assessment of Hazardous Air Pollutants, Commission on Life Sciences, NRC.
Washington, DC: National Academy Press.
O'Flaherty, EJ. (1989) Application of pharmacokinetic principles to exposure to chemical
mixtures. Toxicol Ind Health 5:667-680.
Parr, JF; Smith, S. (1976) Degradation of toxaphene in selected anaerobic soil environments.
Soil Sci 121(l):52-57.
Plaa, GL; Hewitt, WR. (1982) In: Advances in pharmacology and therapeutics n. Yoshida, H;
Hagihara, Y; Ebashi, S; eds. Oxford: Pergamon. Vol. 5, pp. 65-75.
Plaa, GL; Venzina, M. (1990) Factors to consider in the design and evaluation of chemical
interaction studies in laboratory animals. In: Toxic interactions. Goldstein, RS; Hewitt, WR;
Hook, JB; eds. New York: Academic Press, pp. 3-30.
Pozzani, UC; Weil, CS; Carpenter, CP. (1959) The toxicological basis of threshold values: 5.
The experimental inhalation of vapor mixtures by rats, with notes upon the relationship between
single dose inhalation and single dose oral data. Am Ind Hyg Assoc J 20:364-369.
Prescott, LF. (1969) Pharmacokinetic drug interactions. Lancet 2:1239-1243.
Rao, VR; Woo, Y-T; Lai, DY; Arcos, JC. (1989) Database on promoters of chemical
carcinogenesis. Environ Carcinogen Revs (Part C of J Environ Sci Health) C7[2]: 1-278.
Richards, MP; Cousins, RJ. (1975) Mammalian Zn homeostasis: requirement for RNA and
metallothionein synthesis. Biochem Biophys Res Commun 64:1215-1223.
Rothman, KJ; Greenland, S. (1998) Concepts of interaction. In: Modern epidemiology. Chapter
18, 2nd ed. Philadelphia: Lippincott-Raven Publishers, pp. 329-342.
Safe, S. (1994) Polychlorinated biphenyls (PCBs): environmental impact, biochemical and toxic
responses, and implications for risk assessment. Crit Rev Toxicol 24(2):87-149.
Sauer, JM; Sipes, IG. (1995) Modulation of chemical-induced lung and liver toxicity by all-
trans-retinol in the male Sprague-Dawley rat. Toxicology 105:237-249.
Schantz, SL; Levin, ED; Bowman, RE, et al. (1989) Effects of perinatal PCB exposure on
discrimination-reversal learning in monkeys. Neurotoxicol Teratol 11:243-250.
Schantz, SL; Levin, ED; Bowman, RE. (1991) Long-term neurobehavioral effects of perinatal
polychlorinated biphenyl (PCB) exposure in monkeys. Environ Toxicol Chem 10:747-756.
-139-
-------
Schmitz, HJ; Behnisch, P; Hagenmaier, A; Hagenmaier, A; Bock, KW; Schrenk, D. (1996)
CYPlAl-inducing potency in H4IIE cells and chemical composition of technical mixtures of
polychlorinated biphenyls. Environ Toxicol Pharmacol l(l):73-79.
Schoeny, RS; Margosches, E. (1989) Evaluating comparative potencies: developing approaches
to risk assessment of chemical mixtures. Toxicol Ind Health 5:825-837.
Schrenk, D; Buclunann, A; Dietz, K; et al. (1994) Promotion of preneoplastic foci in rat liver
with 2,3,7,8-tetrachlorodibenzo-p-dioxin, 1,2,3,4,6,7,8-heptachlorodibenzo-p-dioxin and a
defined mixture of 49 polychlorinated dibenzo-p-dioxins. Carcinogenesis 15(3):509-515.
Seiber, JN; Winterlin, WL; McChesney, MM. (1979) Residues of toxaphene, DEF, and
paraquat in plant parts and gin waste from a treated cotton field. Arch Environ Contam Toxicol
8(2):125-137.
Simmons, JE; Yang, RSH; Svensgaard, DJ; Thompson, MB; Seely, JC; McDonald, A. (1994)
Toxicology studies of a chemical mixture of 25 groundwater contaminants: hepatic and renal
assessment, response to carbon tetrachloride challenge and influence of treatment-induced water
restriction. J Toxicol Environ Health 43:305-325.
Simmons, JE; Yang, RSH; Berman, E. (1995) Evaluation of the nephrotoxicity of complex
mixtures containing organics and metals: advantages and disadvantages of the use of real world
mixtures. Environ Health Perspect 103:67-71.
Sloan, KB; Koch, SAM; Siver, KG; Flowers, FP. (1986) Use of solubility parameters of drug
and vehicle to predict flux through skin. J Invest Dermatol 87:244-252.
Smialowicz, RJ; DeVito, MJ; Riddle, MM; Williams, WC; Birnbaum, LS. (1997) Opposite
effects of 2,2',4,4',5,5'-hexachlorobiphenyl and 2,3,7,8-tetrachlorodibenzo-p-dioxin on the
antibody response to sheep erythrocytes in mice. Fundam Appl Toxicol 37(2): 141-149.
Smith, S; Willis, GH. (1978) Disappearance of residual toxaphene in a Mississippi delta soil.
Soil Sci 126(2):87-93.
Smyth, HF; Weil, CS; West, JS; Carpenter, CP. (1969) An exploration of joint toxic action. I.
Twenty-seven industrial chemicals intubated in rats in all possible pairs. Toxicol Appl
Pharmacol 14:340-347.
Smyth, HF; Weil, CS; West, JS; Carpenter, CP. (1970) An exploration of joint toxic action. II.
Equitoxic versus equivolume mixtures. Toxicol Appl Pharmacol 17:498-503.
Suzuki, KT; Kawahara, S; Sunaga, H; Kobayashi, E. (1991) Discriminative uptake of cadmium,
copper and zinc by the liver. In: Metallothionein in biology and medicine. Klassen, CD; Suzuki,
KT, eds. Boca Raton, FL: CRC Press, pp. 197-208.
-140-
-------
Svendsgaard, DJ; Hertzberg, RC. (1994) Statistical methods for the toxicological evaluation of
the additivity assumption as used in the EPA chemical mixture risk assessment guidelines. In:
Toxicology of chemical mixtures: case studies, mechanisms, and novel approaches. Yang, RSH,
ed. New York: Academic Press, pp. 599-642.
Svendsgaard, DJ; Greco, WR. (1995) Session summary: experimental designs, analyses and
quantitative models. Toxicology 105:156-160.
Tardif, R; Charest-Tardif, G; Lapare, S; Brodeur, J; Krishnan, K. (1997) Physiologically-based
pharmacokinetic modeling of a ternary mixture of alkyl benzenes in rats and humans. Toxicol
Appl Pharmacol 143:120-134.
Teuschler, LK; Hertzberg, RC. (1995) Current and future risk assessment guidelines, policy,
and methods development for chemical mixtures. Toxicology 105:137-144.
Thompson, ML; Shuster, L; Shaw, K. (1979) Cocaine-induced hepatic necrosis in mice—The
role of cocaine metabolism. Biochem Pharmacol 28:2389-2395.
U.S. EPA (Environmental Protection Agency). (1984) Carcinogenic assessment of coke oven
emissions. Office of Health and Environmental Assessment, Washington, DC.
EP A/600/6-82/03 3F.
U.S. EPA. (1986) Guidelines for health risk assessment of chemical mixtures. Federal Register
51(185):34014-34025.
U.S. EPA. (1987) The risk assessment guidelines of 1986. EPA/600/8-87/045.
U.S. EPA. (1989a) Risk assessment guidance for Superfund. Vol. 1. Human Health Evaluation
Manual (Part A). EPA/540/1-89/002.
U.S. EPA. (1989b) Interim procedures for estimating risks associated with exposures to
mixtures of chlorinated dibenzo-p-dioxins and -dibenzofurans (CDDs and CDFs) and 1989
update. Risk Assessment Forum. EPA/625/3-89/016.
U.S. EPA. (1990) Technical support document on health risk assessment of chemical mixtures.
EPA/600/8-90/064.
U.S. EPA. (1991a) Guidelines for developmental toxicity risk assessment. Federal Register
56(234):63798-63826.
U.S. EPA. (1991b) Workshop report on toxicity equivalency factors for polychlorinated
biphenyl congeners. Risk Assessment Forum. EPA/625/3-91/020.
U.S. EPA. (1992) Guidelines for exposure assessment. Federal Register 57(104):22888-22938.
-141-
-------
U.S. EPA. (1994a) Methods for development of inhalation reference doses. Environmental
Criteria and Assessment Office, Office of Health and Environmental Assessment, Research
Triangle Park, NC. EPA/600/8-88/066F.
U.S. EPA. (1994b) Estimating exposures to dioxin-like compounds. Volumes I through HI,
EPA/600/6-88/005Ca,b,c.
U.S. EPA. (1995) Policy for risk characterization at the U.S. Environmental Protection Agency.
Memorandum from Carol Browner, March 21, 1995. (Version current on September 15, 1999.)
Online at .
U.S. EPA. (1996a) Proposed guidelines for carcinogen risk assessment. Federal Register
61(79):17960-18011.
U.S. EPA. (1996b) Reproductive toxicity risk assessment guidelines. Federal Register
61(212):56274-56322.
U.S. EPA. (1996c) PCBs: cancer dose-response assessment and application to environmental
mixtures. National Center for Environmental Assessment, Washington, DC.
EPA/600/P-96/001F.
U.S. EPA. (1996d) Benchmark dose technical guidance document. External Review Draft.
EPA/600/P-96/002A.
U.S. EPA. (1997a). Office of Science Policy (OSP). Online. Office of Research and
Development. Washington DC. http://www.epa.gov/ORD/spc/cumrisk2.htm.
U.S. EPA. (1997b) Research on risk assessment issues with commercial mixtures using
toxaphene as a case study. National Center for Environmental Assessment, Cincinnati, OH.
U.S. EPA. (1998a) Guidelines for ecological risk assessment. Federal Register 63(93):26846-
26924.
U.S. EPA. (1998b) Guidelines for neurotoxicity risk assessment. Federal Register
63(93):26926-26954.
U.S. EPA. (1998c) Methodology for assessing health risks associated with multiple pathways of
exposure to combustor emissions. EPA/600/R-98/137 (update to EPA/600/6-90/003).
U.S. EPA. (1998d) Comparative risk framework methodology and case study. SAB External
Review Draft. NCEA-C-0135.
U.S. EPA. (2000) Integrated Risk Information System (IRIS). Online. National Center for
Environmental Assessment, Washington, DC. http:www.epa.gov/iris.
-142-
-------
U.S. EPA. (2000a). Integrated Risk Information System (IRIS). Online. National Center for
Environmental Assessment, Washington, DC.
U.S. EPA. (2000b). Chapter 9. Toxicity equivalence factors for dioxin and related compounds.
In: Exposure and human health reassessment of 2,3,7,8-tetrachlorodibenzo-p-dioxin and related
compounds. External review draft, http://www.epa.gov/ncea/pdfs/dioxin/part2/drich9.pdf.
Van denBerg, M; Birnbaum, L; Bosveld, A; et al. (1998) Toxic equivalency factors (TEFs) for
PCBs, PCDDs, PCDFs for humans and wildlife. Environ Health Perspect 106(12):775-792.
Van Leeuwen, FXR. (1997) Derivation of toxic equivalence factors (TEFs) for PCBs, PCDDs,
and PCDFs for humans and wildlife. Organohalogen Compounds 34:237.
Waters, MD; Stack, HG; Brady, AL; Lohman, PHM; Haroun, L; Vainio, G. (1988a) Use of
computerized data listing and activity profiles of genetic and related effects in the review of 195
compounds. MutatRes 205:295-312. (Cited by Nesnow, 1990)
Waters, MD; Stack, HF; Rabinowitz, JR; Garrett, NE. (1988b) Genetic activity profiles and
pattern recognition in test battery selection. MutatRes 205:119-138. (Cited by Nesnow, 1990)
Woo, Y-T; DiCarlo, FJ; Arcos, JC; Argus, MF; Polansky, G; DuBose, J. (1994) Assessment of
carcinogenic hazard of chemical mixtures through analysis of binary chemical interaction data.
Environ Health Perspect 102(Suppl. 9): 113-118.
Woo, Y-T; Lai, DY; Argus, MF; Arcos, JC. (1995a) Development of structure-activity
relationship rules for predicting carcinogenic potential of chemicals. Toxicol Lett 79:219-228.
Woo, Y-T; et al. (1995b) Computerized data management as a tool to study combination effects
in carcinogenesis. In: Chemical induction of cancer—modulation and combination effects.
Arcos, JC, ed. Boston: Birkhauser, pp. 185-203.
York, RG; O'Flaherty, EJ; Scott, WJ, Jr; Shukla, R. (1987) Alteration of effective exposure of
dame and embryo to caffeine and its metabolites by treatment of mice with p-naphthoflavone.
Toxicol Appl Pharmacol 88:282-293.
-143-
-------
EPA/630/R-98/002
September 1986
APPENDIX A
Guidelines for the
Health Risk Assessment of
Chemical Mixtures
Published on September 24, 1986, Federal Register 51(185):34014-34025
Risk Assessment Forum
U.S. Environmental Protection Agency
Washington, DC
-------
DISCLAIMER
This document has been reviewed in accordance with U.S. Environmental Protection
Agency policy and approved for publication. Mention of trade names or commercial products
does not constitute endorsement or recommendation for use.
Note: This document represents the final guidelines. A number of editorial corrections have
been made during conversion and subsequent proofreading to ensure the accuracy of this
publication.
A-ii
-------
CONTENTS
List of Tables and Figures v
Federal Register Preamble vi
Part A: Guidelines for the Health Risk Assessment of Chemical Mixtures
1. Introduction 1
2. Proposed Approach 2
2.1. Data Available on the Mixture of Concern 2
2.2. Data Available on Similar Mixtures 6
2.3. Data Available Only on Mixture Components 7
2.3.1. Systemic Toxicants 8
2.3.2. Carcinogens 9
2.3.3. Interactions 10
2.3.4. Uncertainties 10
3. Assumptions and Limitations 12
3.1. Information on Interactions 12
3.2. AdditivityModels 13
4. Mathematical Models and the Measurement of Joint Action 14
4.1. Dose Addition 14
4.2. Response Addition 15
4.3. Interactions 16
5. References 19
Part B: Response to Public and Science Advisory Board Comments
1. Introduction 23
2. Recommended Procedures 23
2.1. Definitions 23
2.2. Mixtures of Carcinogens and Systemic Toxicants 24
3. Additivity Assumption 25
3.1. Complex Mixtures 25
3.2. Dose Additivity 25
3.3. Interpretation of the Hazard Index 26
3.4. Use of Interaction Data 26
4. Uncertainties and the Sufficiency of the Data Base 26
A-iii
-------
CONTENTS (continued)
5. Need for a Technical Support Document 28
A-iv
-------
LIST OF TABLES
Table 1. Risk assessment approach for chemical mixtures 3
Table 2. Classification scheme for the quality of the risk assessment of the mixture 5
LIST OF FIGURES
Figure 1. Flow chart of the risk assessment in Table 1 4
A-v
-------
GUIDELINES FOR THE HEALTH RISK ASSESSMENT OF CHEMICAL MIXTURES
[FRL-2984-2]
AGENCY: U. S. Environmental Protection Agency (EPA).
ACTION: Final Guidelines for the Health Risk Assessment of Chemical Mixtures.
SUMMARY: The U.S. Environmental Protection Agency is today issuing five guidelines for
assessing the health risks of environmental pollutants. These are:
Guidelines for Carcinogen Risk Assessment
Guidelines for Estimating Exposures
Guidelines for Mutagenicity Risk Assessment
Guidelines for the Health Assessment of Suspect Developmental Toxicants
Guidelines for the Health Risk Assessment of Chemical Mixtures
This notice contains the Guidelines for the Health Risk Assessment of Chemical
Mixtures; the other guidelines appear elsewhere in today's Federal Register.
The Guidelines for the Health Risk Assessment of Chemical Mixtures (hereafter
"Guidelines") are intended to guide Agency analysis of information relating to health effects data
on chemical mixtures in line with the policies and procedures established in the statutes
administered by the EPA. These Guidelines were developed as part of an interoffice guidelines
development program under the auspices of the Office of Health and Environmental Assessment
(OHEA) in the Agency's Office of Research and Development. They reflect Agency
consideration of public and Science Advisory Board (SAB) comments on the Proposed
Guidelines for the Health Risk Assessment of Chemical Mixtures published January 9, 1985 (50
FR1170).
This publication completes the first round of risk assessment guidelines development.
These Guidelines will be revised, and new guidelines will be developed, as appropriate.
EFFECTIVE DATE: The Guidelines will be effective September 24, 1986.
FOR FURTHER INFORMATION CONTACT: Dr. Richard Hertzberg, Waste Management
Division, U.S. Environmental Protection Agency, Atlanta Federal Center, 100 Alabama St., SW,
Atlanta, GA 30303-3104, TEL: 404-562-8663.
A-vi
-------
SUPPLEMENTARY INFORMATION: In 1983, the National Academy of Sciences (NAS)
published its book entitled Risk Assessment in the Federal Government: Managing the Process.
In that book, the NAS recommended that Federal regulatory agencies establish "inference
guidelines" to ensure consistency and technical quality in risk assessments and to ensure that the
risk assessment process was maintained as a scientific effort separate from risk management. A
task force within EPA accepted that recommendation and requested that Agency scientists begin
to develop such guidelines.
General
The guidelines published today are products of a two-year Agencywide effort, which has
included many scientists from the larger scientific community. These guidelines set forth
principles and procedures to guide EPA scientists in the conduct of Agency risk assessments, and
to inform Agency decision makers and the public about these procedures. In particular, the
guidelines emphasize that risk assessments will be conducted on a case-by-case basis, giving full
consideration to all relevant scientific information. This case-by-case approach means that
Agency experts review the scientific information on each agent and use the most scientifically
appropriate interpretation to assess risk. The guidelines also stress that this information will be
fully presented in Agency risk assessment documents, and that Agency scientists will identify the
strengths and weaknesses of each assessment by describing uncertainties, assumptions, and
limitations, as well as the scientific basis and rationale for each assessment.
Finally, the guidelines are formulated in part to bridge gaps in risk assessment
methodology and data. By identifying these gaps and the importance of the missing information
to the risk assessment process, EPA wishes to encourage research and analysis that will lead to
new risk assessment methods and data.
Guidelines for Health Risk Assessment of Chemical Mixtures
Work on the Guidelines for the Health Risk Assessment of Chemical Mixtures began in
January 1984. Draft guidelines were developed by Agency work groups composed of expert
scientists from throughout the Agency. The drafts were peer-reviewed by expert scientists in the
fields of toxicology, pharmacokinetics, and statistics from universities, environmental groups,
industry, labor, and other governmental agencies. They were then proposed for public comment
in the Federal Register (50 FR 1170). On November 9, 1984, the Administrator directed that
Agency offices use the proposed guidelines in performing risk assessments until final guidelines
became available.
After the close of the public comment period, Agency staff prepared summaries of the
comments, analyses of the major issues presented by the commentators, and preliminary Agency
A-vii
-------
responses to those comments. These analyses were presented to review panels of the SAB on
March 4 and April 22-23, 1985, and to the Executive Committee of the SAB on April 25-26,
1985. The SAB meetings were announced in the Federal Register as follows: February 12, 1985
(50 FR 5811), and April 4, 1985 (50 FR 13420 and 13421).
In a letter to the Administrator dated June 19, 1985, the Executive Committee generally
concurred on all five of the guidelines, but recommended certain revisions and requested that any
revised guidelines be submitted to the appropriate SAB review panel chairman for review and
concurrence on behalf of the Executive Committee. As described in the responses to comments
(see Part B: Response to the Public and Science Advisory Board Comments), each guidelines
document was revised, where appropriate, consistent with the SAB recommendations, and
revised draft guidelines were submitted to the panel chairmen. Revised draft Guidelines for the
Health Risk Assessment of Chemical Mixtures were concurred on in a letter dated August 16,
1985. Copies of the letters are available at the Public Information Reference Unit, EPA
Headquarters Library, as indicated elsewhere in this notice.
Following this Preamble are two parts: Part A contains the Guidelines and Part B the
Response to the Public and Science Advisory Board Comments (a summary of the major public
comments, SAB comments, and Agency responses to those comments).
The SAB requested that the Agency develop a technical support document for these
Guidelines. The SAB identified the need for this type of document due to the limited knowledge
on interactions of chemicals in biological systems. Because of this, the SAB commented that
progress in improving risk assessment will be particularly dependent upon progress in the science
of interactions.
Agency staff have begun preliminary work on the technical support document and expect
it to be completed by early 1987. The Agency is continuing to study the risk assessment issues
raised in the guidelines and will revise these Guidelines in line with new information as
appropriate.
References, supporting documents, and comments received on the proposed guidelines, as
well as copies of the final guidelines, are available for inspection and copying at the Public
Information Reference Unit (202-382-5926), EPA Headquarters Library, 401 M Street, SW,
Washington, DC, between the hours of 8:00 a.m. and 4:30 p.m.
I certify that these Guidelines are not major rules as defined by Executive Order 12291,
because they are nonbinding policy statements and have no direct effect on the regulated
A-viii
-------
community. Therefore, they will have no effect on costs or prices, and they will have no other
significant adverse effects on the economy. These Guidelines were reviewed by the Office of
Management and Budget under Executive Order 12291.
Dated: August 22, 1986 Signed by EPA Administrator
Lee M. Thomas
A-ix
-------
PART A: GUIDELINES FOR THE HEALTH RISK ASSESSMENT OF CHEMICAL
MIXTURES
1. INTRODUCTION
The primary purpose of this document is to generate a consistent Agency approach for
evaluating data on the chronic and subchronic effects of chemical mixtures. It is a procedural
guide that emphasizes broad underlying principles of the various science disciplines (toxicology,
pharmacology, statistics) necessary for assessing health risk from chemical mixture exposure.
Approaches to be used with respect to the analysis and evaluation of the various data are also
discussed.
It is not the intent of these Guidelines to regulate any social or economic aspects
concerning risk of injury to human health or the environment caused by exposure to a chemical
agent(s). All such action is addressed in specific statutes and federal legislation and is
independent of these Guidelines.
While some potential environmental hazards involve significant exposure to only a single
compound, most instances of environmental contamination involve concurrent or sequential
exposures to a mixture of compounds that may induce similar or dissimilar effects over exposure
periods ranging from short-term to lifetime. For the purposes of these Guidelines, mixtures will
be defined as any combination of two or more chemical substances regardless of source or of
spatial or temporal proximity. In some instances, the mixtures are highly complex, consisting of
scores of compounds that are generated simultaneously as byproducts from a single source or
process (e.g., coke oven emissions and diesel exhaust). In other cases, complex mixtures of
related compounds are produced as commercial products (e.g., PCBs, gasoline and pesticide
formulations) and eventually released to the environment. Another class of mixtures consists of
compounds, often unrelated chemically or commercially, which are placed in the same area for
disposal or storage, eventually come into contact with each other, and are released as a mixture to
the environment. The quality and quantity of pertinent information available for risk assessment
varies considerably for different mixtures. Occasionally, the chemical composition of a mixture
is well characterized, levels of exposure to the population are known, and detailed toxicologic
data on the mixture are available. Most frequently, not all components of the mixture are known,
exposure data are uncertain, and toxicologic data on the known components of the mixture are
limited. Nonetheless, the Agency may be required to take action because of the number of
individuals at potential risk or because of the known toxicologic effects of these compounds that
have been identified in the mixture.
A-l
-------
The prediction of how specific mixtures of toxicants will interact must be based on an
understanding of the mechanisms of such interactions. Most reviews and texts that discuss
toxicant interactions attempt to discuss the biological or chemical bases of the interactions (e.g.,
Klaassen and Doull, 1980; Levine, 1973; Goldstein et al., 1974; NRC, 1980a; Veldstra, 1956;
Withey, 1981). Although different authors use somewhat different classification schemes when
discussing the ways in which toxicants interact, it generally is recognized that toxicant
interactions may occur during any of the toxicologic processes that take place with a single
compound: absorption, distribution, metabolism, excretion, and activity at the receptor site(s).
Compounds may interact chemically, yielding a new toxic component or causing a change in the
biological availability of the existing component. They may also interact by causing different
effects at different receptor sites.
Because of the uncertainties inherent in predicting the magnitude and nature of toxicant
interactions, the assessment of health risk from chemical mixtures must include a thorough
discussion of all assumptions. No single approach is recommended in these Guidelines. Instead,
guidance is given for the use of several approaches depending on the nature and quality of the
data. Additional mathematical details are presented in Section 4.
In addition to these Guidelines, a supplemental technical support document is being
developed which will contain a thorough review of all available information on the toxicity of
chemical mixtures and a discussion of research needs.
2. PROPOSED APPROACH
No single approach can be recommended to risk assessments for multiple chemical
exposures. Nonetheless, general guidelines can be recommended depending on the type of
mixture, the known toxic effects of its components, the availability of toxicity data on the
mixture or similar mixtures, the known or anticipated interactions among components of the
mixture, and the quality of the exposure data. Given the complexity of this issue and the relative
paucity of empirical data from which sound generalizations can be constructed, emphasis must
be placed on flexibility, judgment, and a clear articulation of the assumptions and limitations in
any risk assessment that is developed. The proposed approach is summarized in Table 1 and
Figure 1 and is detailed below. An alphanumeric scheme for ranking the quality of the data used
in the risk assessment is given in Table 2.
2.1. DATA AVAILABLE ON THE MIXTURE OF CONCERN
For predicting the effects of subchronic or chronic exposure to mixtures, the preferred
approach usually will be to use subchronic or chronic health effects data on the mixture of
A-2
-------
Table 1. Risk assessment approach for chemical mixtures
1. Assess the quality of the data on interactions, health effects, and exposure (see Table 2).
a. If adequate, proceed to Step 2.
b. If inadequate, proceed to Step 14.
2. Health effects information is available on the chemical mixture of concern.
a. If yes, proceed to Step 3.
b. If no, proceed to Step 4.
3. Conduct risk assessment on the mixture of concern based on health effects data on the
mixture. Use the same procedures as those for single compounds. Proceed to Step 7
(optional) and Step 12.
4. Health effects information is available on a mixture that is similar to the mixture of concern.
a. If yes, proceed to Step 5.
b. If no, proceed to Step 7.
5. Assess the similarity of the mixture on which health effects data are available to the mixture
of concern, with emphasis on any differences in components or proportions of components,
as well as the effects that such differences would have on biological activity.
a. If sufficiently similar, proceed to Step 6.
b. If not sufficiently similar, proceed to Step 7.
6. Conduct risk assessment on the mixture of concern based on health effects data on the similar
mixture. Use the same procedures as those for single compounds. Proceed to Step 7
(optional) and Step 12.
7. Compile health effects and exposure information on the components of the mixture.
8. Derive appropriate indices of acceptable exposure and/or risk on the individual components
in the mixture. Proceed to Step 9.
9. Assess data on interactions of components in the mixtures.
a. If sufficient quantitative data are available on the interactions of two or more components
in the mixture, proceed to Step 10.
b. If sufficient quantitative data are not available, use whatever information is available to
qualitatively indicate the nature of potential interactions. Proceed to Step 11.
10. Use an appropriate interaction model to combine risk assessments on compounds for which
data are adequate, and use an additivity assumption for the remaining compounds. Proceed
to Step 11 (optional) and Step 12.
11. Develop a risk assessment based on an additivity approach for all compounds in the mixture.
Proceed to Step 12.
12. Compare risk assessments conducted in Steps 5, 8, and 9. Identify and justify the preferred
assessment, and quantify uncertainty, if possible. Proceed to Step 13.
13. Develop an integrated summary of the qualitative and quantitative assessments with special
emphasis on uncertainties and assumptions. Classify the overall quality of the risk
assessment, as indicated in Table 2. Stop.
14. No risk assessment can be conducted because of inadequate data on interactions, health
effects, or exposure. Qualitatively assess the nature of any potential hazard and detail the
types of additional data necessary to support a risk assessment. Stop.
Note—Several decisions used here, especially those concerning adequacy of data and similarity between two
mixtures, are not precisely characterized and will require considerable judgment. See text.
A-3
-------
1. Assess quality of data using Table 2.
I. Interactions
II. Health Effects
III. Exposure
Inadequate
14. Qualitatively assess hazard.
No quantitative risk assessment.
Adequate
2. Data on mixture of concern?
similar mixture? —
N
Data on mixture components.
(5. Mixtures sufficiently similar?)
N
3. Risk assessment using data on
mixture of concern.
>
6. Risk assessment using data on
similar mixtures.
12. Compare risk assessments from steps 3,
6, 10, 11 as appropriate. Identify
preferred assessment.
13. Develop integrated summary including
discussion on uncertainties.
8. Indices of acceptability and risk based on
component data.
9. Sufficient information to quantify
interactions?
10. Risk assessment with interactions
quantified where appropriate. Use
additivity for other components.
optional
11. Risk assessment using additivity for all
components.
Figure 1. Flow chart of the risk assessment in Table 1. Note that it may be desirable to conduct all three assessments when
possible (i.e., using data on the mixture, a similar mixture, or the components) in order to make the fullest use of the available
data. See text for further discussion.
-------
Table 2. Classification scheme for the quality of the risk assessment of the mixture"
Information on Interactions
I. Assessment is based on data on the mixture of concern.
II. Assessment is based on data on a sufficiently similar mixture.
IE. Quantitative interactions of components are well characterized.
IV. The assumption of additivity is justified based on the nature of the health effects and on the
number of component compounds.
V. An assumption of additivity cannot be justified, and no quantitative risk assessment can be
conducted.
Health Effects Information
A. Full health effects data are available and relatively minor extrapolation is required.
B. Full health effects data are available but extensive extrapolation is required for route or
duration of exposure or for species differences. These extrapolations are supported by
pharmacokinetic considerations, empirical observations, or other relevant information.
C. Full health effects data are available, but extensive extrapolation is required for route or
duration of exposure or for species differences. These extrapolations are not directly
supported by the information available.
D. Certain important health effects data are lacking and extensive extrapolations are required
for route or duration of exposure or for species differences.
E. A lack of health effects information on the mixture and its components in the mixture
precludes a quantitative risk assessment.
Exposure Information11
1. Monitoring information either alone or in combination with modeling information is
sufficient to accurately characterize human exposure to the mixture or its components.
2. Modeling information is sufficient to reasonably characterize human exposure to the mixture
or its components.
3. Exposure estimates for some components are lacking, uncertain, or variable. Information on
health effects or environmental chemistry suggests that this limitation is not likely to
substantially affect the risk assessment.
4. Not all components in the mixture have been identified, or levels of exposure are highly
uncertain or variable. Information on health effects or environmental chemistry is not
sufficient to assess the effect of this limitation on the risk assessment.
5. The available exposure information is insufficient for conducting a risk assessment.
aSee text for discussion of sufficient similarity, adequacy of data, and justification for additivity assumptions.
bSee the Agency's Guidelines for Estimating Exposures (U.S. EPA, 1986d) for more complete
information on performing exposure assessments and evaluating the quality of exposure data.
A-5
-------
concern and adopt procedures similar to those used for single compounds, either systemic
toxicants or carcinogens (see U.S. EPA, 1986a-c). The risk assessor must recognize, however,
that dose-response models used for single compounds are often based on biological mechanisms
of the toxicity of single compounds, and may not be as well justified when applied to the mixture
as a whole. Such data are most likely to be available on highly complex mixtures, such as coke
oven emissions or diesel exhaust, which are generated in large quantities and associated with or
suspected of causing adverse health effects. Attention should also be given to the persistence of
the mixture in the environment as well as to the variability of the mixture composition over time
or from different sources of emissions. If the components of the mixture are known to partition
into different environmental compartments or to degrade or transform at different rates in the
environment, then those factors must also be taken into account, or the confidence in and
applicability of the risk assessment are diminished.
2.2. DATA AVAILABLE ON SIMILAR MIXTURES
If the risk assessment is based on data from a single mixture that is known to be generated
with varying compositions depending on time or different emission sources, then the confidence
in the applicability of the data to a risk assessment also is diminished. This can be offset to some
degree if data are available on several mixtures of the same components that have different
component ratios which encompass the temporal or spatial differences in composition of the
mixture of concern. If such data are available, an attempt should be made to determine if
significant and systematic differences exist among the chemical mixtures. If significant
differences are noted, ranges of risk can be estimated based on the toxicologic data of the various
mixtures. If no significant differences are noted, then a single risk assessment may be adequate,
although the range of ratios of the components in the mixtures to which the risk assessment
applies should also be given.
If no data are available on the mixtures of concern, but health effects data are available an
a similar mixture (i.e., a mixture having the same components but in slightly different ratios, or
having several common components but lacking one or more components, or having one or more
additional components), a decision must be made whether the mixture on which health effects
data are available is or is not "sufficiently similar" to the mixture of concern to permit a risk
assessment. The determination of "sufficient similarity" must be made on a case-by-case basis,
considering not only the uncertainties associated with using data on a dissimilar mixture but also
the uncertainties of using other approaches such as additivity. In determining reasonable
similarity, consideration should be given to any information on the components that differ or are
contained in markedly different proportions between the mixture on which health effects data are
available and the mixture of concern. Particular emphasis should be placed on any toxicologic or
A-6
-------
pharmacokinetic data on the components or the mixtures which would be useful in assessing the
significance of any chemical difference between the similar mixture and the mixtures of concern.
Even if a risk assessment can be made using data on the mixtures of concern or a
reasonably similar mixture, it may be desirable to conduct a risk assessment based on toxicity
data on the components in the mixture using the procedure outlined in Section 2.B. In the case of
a mixture containing carcinogens and toxicants, an approach based on the mixture data alone
may not be sufficiently protective in all cases. For example, this approach for a two-component
mixture of one carcinogen and one toxicant would use toxicity data on the mixture of the two
compounds. However, in a chronic study of such a mixture, the presence of the toxicant could
mask the activity of the carcinogen. That is to say, at doses of the mixture sufficient to induce a
carcinogenic effect, the toxicant could induce mortality so that at the maximum tolerated dose of
the mixture, no carcinogenic effect could be observed. Since carcinogenicity is considered by the
Agency to be a nonthreshold effect, it may not be prudent to construe the negative results of such
a bioassay as indicating the absence of risk at lower doses. Consequently, the mixture approach
should be modified to allow the risk assessor to evaluate the potential for masking, of one effect
by another, on a case-by-case basis.
2.3. DATA AVAILABLE ONLY ON MIXTURE COMPONENTS
If data are not available on an identical or reasonably similar mixture, the risk assessment
may be based on the toxic or carcinogenic properties of the components in the mixture. When
little or no quantitative information is available on the potential interaction among the
components, additive models (defined in the next section) are recommended for systemic
toxicants. Several studies have demonstrated that dose additive models often predict reasonably
well the toxicities of mixtures composed of a substantial variety of both similar and dissimilar
compounds (Pozzani et al., 1959; Smyth et al., 1969, 1970; Murphy, 1980). The problem of
multiple toxicant exposure has been addressed by the American Conference of Governmental
Industrial Hygienists (ACGIH, 1983), the Occupational Safety and Health Administration
(OSHA, 1983), the World Health Organization (WHO, 1981), and the National Research Council
(NRC, 1980a,b). Although the focus and purpose of each group was somewhat different, all
groups that recommended an approach elected to adopt some type of dose additive model.
Nonetheless, as discussed in Section 4, dose additive models are not the most biologically
plausible approach if the compounds do not have the same mode of toxicologic action.
Consequently, depending on the nature of the risk assessment and the available information on
modes of action and patterns of joint action, the Federal Register most reasonable additive model
should be used.
A-7
-------
2.3.1. Systemic Toxicants
For systemic toxicants, the current risk assessment methodology used by the Agency for
single compounds most often results in the derivation of an exposure level which is not
anticipated to cause significant adverse effects. Depending on the route of exposure, media of
concern, and the legislative mandate guiding the risk assessments, these exposure levels may be
expressed in a variety of ways such as acceptable daily intakes (ADIs) or reference doses (RfDs),
levels associated with various margins of safety (MOS), or acceptable concentrations in various
media. For the purpose of this discussion, the term "acceptable level" (AL) will be used to
indicate any such criteria or advisories derived by the Agency. Levels of exposure (E) will be
estimates obtained following the most current Agency Guidelines for Estimating Exposures (U.S.
EPA, 1986d). For such estimates, the "hazard index" (HI) of a mixture based on the assumption
of dose addition may be defined as:
HI = Ej/ALj + E2/AL2 +. . . + E{/AL{ (2-1)
where:
E; = exposure level to the ith toxicant* and AL; = maximum acceptable level for the ith toxicant.
Since the assumption of dose addition is most properly applied to compounds that induce
the same effect by similar modes of action, a separate hazard index should be generated for each
end point of concern. Dose addition for dissimilar effects does not have strong scientific support,
and, if done, should be justified on a case-by-case basis in terms of biological plausibility.
The assumption of dose addition is most clearly justified when the mechanisms of action
of the compounds under consideration are known to be the same. Since the mechanisms of
action for most compounds are not well understood, the justification of the assumption of dose
addition will often be limited to similarities in pharmacokinetic and toxicologic characteristics.
In any event, if a hazard index is generated the quality of the experimental evidence supporting
the assumption of dose addition must be clearly articulated.
The hazard index provides a rough measure of likely toxicity and requires cautious
interpretation. The hazard index is only a numerical indication of the nearness to acceptable
limits of exposure or the degree to which acceptable exposure levels are exceeded. As this index
approaches unity, concern for the potential hazard of the mixture increases. If the index exceeds
unity, the concern is the same as if an individual chemical exposure exceeded its acceptable level
by the same proportion. The hazard index does not define dose-response relationships, and its
numerical value should not be construed to be a direct estimate of risk. Nonetheless, if sufficient
A-8
-------
data are available to derive individual acceptable levels for a spectrum of effects (e.g., MFO
induction, minimal effects in several organs, reproductive effects, and behavioral effects), the
hazard index may suggest what types of effects might be expected from the mixture exposure. If
the components' variabilities of the acceptable levels are known, or if the acceptable levels are
given as ranges (e.g., associated with different margins of safety), then the hazard index should
be presented with corresponding estimates of variation or range.
Most studies on systemic toxicity report only descriptions of the effects in each dose
group. If dose-response curves are estimated for systemic toxicants, however, dose-additive or
response-additive assumptions can be used, with preference given to the most biologically
plausible assumption (see Section 4 for the mathematical details).
2.3.2. Carcinogens
For carcinogens, whenever linearity of the individual dose-response curves has been
assumed (usually restricted to low doses), the increase in risk P (also called excess or incremental
risk), caused by exposure d, is related to carcinogenic potency B, as:
P = d B (2-2)
For multiple compounds, this equation may be generalized to:
P = £diBi (2-3)
This equation assumes independence of action by the several carcinogens and is
equivalent to the assumption of dose addition as well as to response addition with completely
negative correlation of tolerance, as long as P < 1 (see Section 4). Analogous to the procedure
used in Equation 2-1 for systemic toxicants, an index for n carcinogens can be developed by
dividing exposure levels (E) by doses (DR) associated with a set level of risk:
HI = Ej/DRj + E2/DR2 +. . .+ EJDR^ (2-4)
Note that the less linear the dose-response curve is, the less appropriate Equations 2-3 and
2-4 will be, perhaps even at low doses. It should be emphasized that because of the uncertainties
in estimating dose-response relationships for single compounds, and the additional uncertainties
in combining the individual estimate to assess response from exposure to mixtures, response
rates and hazard indices may have merit in comparing risks but should not be regarded as
measures of absolute risk.
A-9
-------
2.3.3. Interactions
None of the above equations incorporates any form of synergistic or antagonistic
interaction. Some types of information, however, may be available that suggest that two or more
components in the mixture may interact. Such information must be assessed in terms of both its
relevance to subchronic or chronic hazard and its suitability for quantitatively altering the risk
assessment.
For example, if chronic or subchronic toxicity or carcinogenicity studies have been
conducted that permit a quantitative estimation of interaction for two chemicals, then it may be
desirable to consider using equations detailed in Section 4, or modifications of these equations,
to treat the two compounds as a single toxicant with greater or lesser potency than would be
predicted from additivity. Other components of the mixture, on which no such interaction data
are available, could then be separately treated in an additive manner. Before such a procedure is
adopted, however, a discussion should be presented of the likelihood that other compounds in the
mixture may interfere with the interaction of the two toxicants on which quantitative interaction
data are available. If the weight of evidence suggests that interference is likely, then a
quantitative alteration of the risk assessment may not be justified. In such cases, the risk
assessment may only indicate the likely nature of interactions, either synergistic or antagonistic,
and not quantify their magnitudes.
Other types of information, such as those relating to mechanisms of toxicant interaction,
or quantitative estimates of interaction between two chemicals derived from acute studies, are
even less likely to be of use in the quantitative assessment of long-term health risks. Usually it
will be appropriate only to discuss these types of information, indicate the relevance of the
information to subchronic or chronic exposure, and indicate, if possible, the nature of potential
interactions, without attempting to quantify their magnitudes.
When the interactions are expected to have a minor influence on the mixture's toxicity,
the assessment should indicate, when possible, the compounds most responsible for the predicted
toxicity. This judgment should be based on predicted toxicity of each component, based on
exposure and toxic or carcinogenic potential. This potential alone should not be used as an
indicator of the chemicals posing the most hazard.
2.3.4. Uncertainties
For each risk assessment, the uncertainties should be clearly discussed and the overall
quality of the risk assessment should be characterized. The scheme outlined in Table 2 should be
used to express the degree of confidence in the quality of the data on interaction, health effects,
and exposure.
A-10
-------
a. Health Effects—In some cases, when health effects data are incomplete, it may be possible to
argue by analogy or quantitative structure-activity relationships that the compounds on which
no health effects data are available are not likely to significantly affect the toxicity of the
mixture. If a risk assessment includes such an argument, the limitations of the approach must
be clearly articulated. Since a methodology has not been adopted for estimating an
acceptable level (e.g., ADI) or carcinogenic potential for single compounds based either on
quantitative structure-activity relationships or on the results of short-term screening tests,
such methods are not at present recommended as the sole basis of a risk assessment on
chemical mixtures.
b. Exposure Uncertainties—The general uncertainties in exposure assessment have been
addressed in the Agency's Guidelines for Estimating Exposures (U.S. EPA, 1986d). The risk
assessor should discuss these exposure uncertainties in terms of the strength of the evidence
used to quantify the exposure. When appropriate, the assessor should also compare
monitoring and modeling data and discuss any inconsistencies as a source of uncertainty. For
mixtures, these uncertainties may be increased as the number of compounds of concern
increases.
If levels of exposure to certain compounds known to be in the mixture are not
available, but information on health effects and environmental persistence and transport
suggest that these compounds are not likely to be significant in affecting the toxicity of the
mixture, then a risk assessment can be conducted based on the remaining compounds in the
mixture, with appropriate caveats. If such an argument cannot be supported, no final risk
assessment can be performed until adequate monitoring data are available. As an interim
procedure, a risk assessment may be conducted for those components in the mixture for
which adequate exposure and health effects data are available. If the interim risk assessment
does not suggest a hazard, there is still concern about the risk from such a mixture because
not all components in the mixture have been considered.
c. Uncertainties Regarding Composition of the Mixture—In perhaps a worst-case scenario,
information may be lacking not only on health effects and levels of exposure, but also on the
identity of some components of the mixture. Analogous to the procedure described in the
previous paragraph, an interim risk assessment can be conducted on those components of the
mixture for which adequate health effects and exposure information are available. If the risk
is considered unacceptable, a conservative approach is to present the quantitative estimates of
risk, along with appropriate qualifications regarding the incompleteness of the data. If no
hazard is indicated by this partial assessment, the risk assessment should not be quantified
until better health effects and monitoring data are available to adequately characterize the
mixture exposure and potential hazards.
A-ll
-------
3. ASSUMPTIONS AND LIMITATIONS
3.1. INFORMATION ON INTERACTIONS
Most of the data available on toxicant interactions are derived from acute toxicity studies
using experimental animals in which mixtures of two compounds were tested, often in only a
single combination. Major areas of uncertainty with the use of such data involve the
appropriateness of interaction data from an acute toxicity study for quantitatively altering a risk
assessment for subchronic or chronic exposure, the appropriateness of interaction data on two
component mixtures for quantitatively altering a risk assessment on a mixture of several
compounds, and the accuracy of interaction data on experimental animals for quantitatively
predicting interactions in humans.
The use of interaction data from acute toxicity studies to assess the potential interactions
on chronic exposure is highly questionable unless the mechanisms of the interaction on acute
exposure were known to apply to low-dose chronic exposure. Most known biological
mechanisms for toxicant interactions, however, involve some form of competition between the
chemicals or phenomena involving saturation of a receptor site or metabolic pathway. As the
doses of the toxicants are decreased, it is likely that these mechanisms either no longer will exert
a significant effect or will be decreased to an extent that cannot be measured or approximated.
The use of information from two-component mixtures to assess the interactions in a
mixture containing more than two compounds also is questionable from a mechanistic
perspective. For example, if two compounds are known to interact, either synergistically or
antagonistically, because of the effects of one compound on the metabolism or excretion of the
other, the addition of a third compound which either chemically alters or affects the absorption of
one of the first two compounds could substantially alter the degree of the toxicologic interaction.
Usually, detailed studies quantifying toxicant interactions are not available on multicomponent
mixtures, and the few studies that are available on such mixtures (e.g., Gullino et al., 1956) do
not provide sufficient information to assess the effects of interactive interference. Concerns with
the use of interaction data on experimental mammals to assess interactions in humans is based on
the increasing appreciation for systematic differences among species in their response to
individual chemicals. If systematic differences in toxic sensitivity to single chemicals exist
among species, then it seems reasonable to suggest that the magnitude of toxicant interactions
among species also may vary in a systematic manner.
Consequently, even if excellent chronic data are available on the magnitude of toxicant
interactions in a species of experimental mammal, there is uncertainty that the magnitude of the
interaction will be the same in humans. Again, data are not available to properly assess the
significance of this uncertainty.
A-12
-------
Last, it should be emphasized that none of the models for toxicant interaction can predict
the magnitude of toxicant interactions in the absence of extensive data. If sufficient data are
available to estimate interaction coefficients as described in Section 4, then the magnitude of the
toxicant interactions for various proportions of the same components can be predicted. The
availability of an interaction ratio (observed response divided by predicted response) is useful
only in assessing the magnitude of the toxicant interaction for the specific proportions of the
mixture which was used to generate the interaction ratio.
The basic assumption in the recommended approach is that risk assessments on chemical
mixtures are best conducted using toxicologic data on the mixture of concern or a reasonably
similar mixture. While such risk assessments do not formally consider toxicologic interactions
as part of a mathematical model, it is assumed that responses in experimental mammals or
human populations noted after exposure to the chemical mixture can be used to conduct risk
assessments on human populations. In bioassays of chemical mixtures using experimental
mammals, the same limitations inherent in species-to-species extrapolation for single compounds
apply to mixtures. When using health effects data on chemical mixtures from studies on exposed
human populations, the limitations of epidemiologic studies in the risk assessment of single
compounds also apply to mixtures. Additional limitations may be involved when using health
effects data on chemical mixtures if the components in the mixture are not constant or if the
components partition in the environment.
3.2. ADDITIVITY MODELS
If sufficient data are not available on the effects of the chemical mixture of concern or a
reasonably similar mixture, the proposed approach is to assume additivity. Dose additivity is
based on the assumption that the components in the mixture have the same mode of action and
elicit the same effects. This assumption will not hold true in most cases, at least for mixtures of
systemic toxicants. For systemic toxicants, however, most single compound risk assessments
will result in the derivation of acceptable levels, which, as currently defined, cannot be adapted to
the different forms of response additivity as described in Section 4.
Additivity models can be modified to incorporate quantitative data on toxicant
interactions from subchronic or chronic studies using the models given in Section 4 or
modifications of these models. If this approach is taken, however, it will be under the
assumption that other components in the mixture do not interfere with the measured interaction.
In practice, such subchronic or chronic interactions data seldom will be available. Consequently,
most risk assessments (on mixtures) will be based on an assumption of additivity, as long as the
components elicit similar effects.
A-13
-------
Dose-additive and response-additive assumptions can lead to substantial errors in risk
estimates if synergistic or antagonistic interactions occur. Although dose additivity has been
shown to predict the acute toxicities of many mixtures of similar and dissimilar compounds (e.g.,
Pozzani et al., 1959; Smyth et al., 1969, 1970; Murphy, 1980), some marked exceptions have
been noted. For example, Smyth et al. (1970) tested the interaction of 53 pairs of industrial
chemicals based on acute lethality in rats. For most pairs of compounds, the ratio of the
predicted LD50 to observed LD50 did not vary by more than a factor of 2. The greatest variation
was seen with an equivolume mixture of morpholine and toluene, in which the observed LD50
was about five times less than the LD50 predicted by dose addition. In a study by Hammond et al.
(1979), the relative risk of lung cancer attributable to smoking was 11, while the relative risk
associated with asbestos exposure was 5. The relative risk of lung cancer from both smoking and
asbestos exposure was 53, indicating a substantial synergistic effect. Consequently, in some
cases, additivity assumptions may substantially underestimate risk. In other cases, risk may be
overestimated. While this is certainly an unsatisfactory situation, the available data on mixtures
are insufficient for estimating the magnitude of these errors. Based on current information,
additivity assumptions are expected to yield generally neutral risk estimates (i.e., neither
conservative nor lenient) and are plausible for component compounds that induce similar types of
effects at the same sites of action.
4. MATHEMATICAL MODELS AND THE MEASUREMENT OF JOINT ACTION
The simplest mathematical models for joint action assume no interaction in any
mathematical sense. They describe either dose addition or response addition and are motivated
by data on acute lethal effects of mixtures of two compounds.
4.1. DOSE ADDITION
Dose addition assumes that the toxicants in a mixture behave as if they were dilutions or
concentrations of each other, thus the true slopes of the dose-response curves for the individual
compounds are identical, and the response elicited by the mixture can be predicted by summing
the individual doses after adjusting for differences in potency; this is defined as the ratio of
equitoxic doses. Probit transformation typically makes this ratio constant at all doses when
parallel straight lines are obtained. Although this assumption can be applied to any model (e.g.,
the one-hit model in NRC, 1980b), it has been most often used in toxicology with the log-
dose probit response model, which will be used to illustrate the assumption of dose addition.
Suppose that two toxicants show the following log-dose probit response equations:
A-14
-------
Y! = 0.3 + 31ogZ1 (4-1)
Y2=1.2 + 31ogZ2 (4-2)
where Yt is the probit response associated with a dose of Zl (i = 1, 2). The potency, p, of
toxicant #2 with respect to toxicant #1 is defined by the quantity Zj/Z2 when Yt = Y2 (that is what
is meant by equitoxic doses). In this example, the potency, p, is approximately 2. Dose addition
assumes that the response, Y, to any mixture of these two toxicants can be predicted by
Y = 0.3 + 3 log (Zt + pZ2) (4-3)
Thus, since p is defined as Zl/Z2, Equation 4-3 essentially converts Z2 into an equivalent
dose of Zj by adjusting for the difference in potency. A more generalized form of this equation
for any number of toxicants is:
Y = E! + b log (fj + S f;pi) + b log Z (4-4)
where:
al = the y-intercept of the dose-response equation for toxicant #1
b = the slope of the dose-response lines for the toxicants
f; = the proportion of the 1th toxicant in the mixture
Pi = the potency of the ith toxicant with respect to toxicant #1 (i.e., Zj/Z;); and
Z = the sum of the individual doses in the mixture.
A more detailed discussion of the derivation of the equations for dose addition is
presented by Finney (1971).
4.2. RESPONSE ADDITION
The other form of additivity is referred to as response addition. As detailed by Bliss
(1939), this type of joint action assumes that the two toxicants act on different receptor systems
and that the correlation of individual tolerances may range from completely negative (r = -1) to
completely positive (r = + 1). Response addition assumes that the response to a given
concentration of a mixture of toxicants is completely determined by the responses to the
components and the pairwise correlation coefficient. Taking P as the proportion of organisms
responding to a mixture of two toxicants which evoke individual responses of Pj and P2, then.
A-15
-------
P = P1ifr=l andPj >P2 (4-5)
P = P2ifr=l andP!
-------
experimental design and the need for large numbers of animals, neither Equation 4-9 nor
Equation 4-10 has been generalized or applied to mixtures of more than two toxicants.
Modifications of response-additive models to include interactive terms have also been proposed,
along with appropriate statistical tests for the assumption of additivity (Korn and Liu, 1983;
Wahrendorfetal., 1981).
In the epidemiologic literature, measurements of the extent of toxicant interactions, S, can
be expressed as the ratio of observed relative risk to relative risk predicted by some form of
additivity assumption. Analogous to the ratio of interaction in classical toxicology studies, S = 1
indicates no interaction, S > 1 indicates synergism, and S < 1 indicates antagonism. Several
models for both additive and multiplicative risks have been proposed (e.g., Hogan et al., 1978;
NRC, 1980b; Walter, 1976). For instance, Rothman (1976) has discussed the use of the
following measurement of toxicant interaction based on the assumption of risk additivity:
S = (Rn-1)7(^ + ^-2) (4-11)
where R10 is the relative risk from compound #1 in the absence of compound #2, R01 is the
relative risk from compound #2 in the absence of compound #1, and Rn is the relative risk from
exposure to both compounds. A multiplicative risk model adapted from Walter and Holford
(1978, Equation 4) can be stated as:
S=Rn/(R10R01) (4-12)
As discussed by both Walter and Holford (1978) and Rothman (1976), the risk-additive
model is generally applied to agents causing diseases while the multiplicative model is more
appropriate to agents that prevent disease. The relative merits of these and other indices have
been the subject of considerable discussion in the epidemiologic literature (Hogan et al., 1978;
Kupper and Hogan, 1978; Rothman, 1978; Rothman et al., 1980; Walter and Holford, 1978).
There seems to be a consensus that for public health concerns regarding causative (toxic) agents,
the additive model is more appropriate.
Both the additive and multiplicative models assume statistical independence in that the
risk associated with exposure to both compounds in combination can be predicted by the risks
associated with separate exposure to the individual compounds. As illustrated by
Siemiatycki and Thomas (1981) for multistage carcinogenesis, the better fitting statistical model
will depend not only upon actual biological interactions, but also upon the stages of the disease
process which the compounds affect. Consequently, there is no a priori basis for selecting either
type of model in a risk assessment. As discussed by Stara et al. (1983), the concepts of
A-17
-------
multistage carcinogenesis and the effects of promoters and cocarcinogens on risk are extremely
complex issues. Although risk models for promoters have been proposed (e.g., Bums et al.,
1983), no single approach can be recommended at this time.
A-18
-------
5. REFERENCES
ACGIH (American Conference of Governmental Industrial Hygienists). 1983. TLVS: threshold
limit values for chemical substances and physical agents in the work environment with intended
changes for 1983-1984. Cincinnati, OH, p. 58.
Alatott, R.L, M.E. Tarrant, and R.B. Forney. 1973. The acute toxicities of 1-methylxanthine,
ethanol, and 1-methylxanthine/ethanol combinations in the mouse. Toxicol. Appl. Pharmacol.
24:393-404.
Bliss, C.I. 1939. The toxicity of poisons applied jointly. Ann. Appl. Biol. 26:585-615.
Bums, F., R. Albert, F. Altschuler, and E. Morris. 1983. Approach to risk assessment for
genotoxic carcinogens based on data from the mouse skin initiation-promotion model. Environ.
Health Perspect. 50:309-320.
Durkin, P.R. 1979. Spent chlorination liquor and chlorophenolics: a study in detoxication and
joint action using Daphnia magna. Ph.D. Thesis, Syracuse, NY: State University of New York
College of Environmental Science and Forestry, p. 145.
Durkin, P.R. 1981. An approach to the analysis of toxicant interactions in the aquatic
environment. Proceedings of the 4th Annual Symposium on Aquatic Toxicology. American
Society for Testing and Materials, p. 388-401.
Finney, DJ. 1942. The analysis of toxicity tests on mixtures of poisons. Ann. Appl. Biol. 29:82-
94.
Finney, DJ. 1971. Probit analysis. 3rd ed. Cambridge, Great Britain: Cambridge University
Press, 333 p.
Goldstein, A., L. Aronow, and S.M. Kalman. 1974. Principles of drug action: the basis of
pharmacology, 2nd ed. New York, NY: John Wiley and Sons, Inc., 854 p.
A-19
-------
Gullino, P., M. Winitz, S.M. Birnbaum, J. Cornfield, M.C. Otey, and J.P. Greenstein. 1956.
Studies on the metabolism of amino acids and related compounds in vivo. I. Toxicity of essential
amino acids, individually and in mixtures, and the protective effect of L-arginine. Arch.
Biochem. Biophys. 64:319-332.
Hammond, E.G., IV. Selikoff, and H. Seidman. 1979. Asbestos exposure, cigarette smoking and
death rates. Ann. NY Acad. Sci. 330:473-490.
Hewlett, P.S. 1969. Measurement of the potencies of drug mixtures. Biometrics 25:477-487.
Hogan, M.D., L. Kupper, B. Most, and J. Haseman. 1978. Alternative approaches to Rothman's
approach for assessing synergism (or antagonism) in cohort studies. Am J. Epidemiol. 108(1):60-
67.
Klaassen, C.D., and J. Doull. 1980. Evaluation of safety: toxicologic evaluation. In: J. Doull,
C.D. Klaassen, and M.O. Amdur, eds. Toxicology: the basic science of poisons. New York, NY:
Macmillan Publishing Co., Inc., p. 11-27.
Kom, E.L, and P-Y. Liu. 1963. Interactive effects of mixtures of stimuli in life table analysis.
Biometrika70:103-110.
Kupper, L., and M.D. Hogan. 1978. Interaction in epidemiologic studies. Am. J. Epidemiol.
108(6):447-453.
Levine, R.E. 1973. Pharmacology: drug actions and reactions. Boston, MA: Little, Brown and
Company, 412 p.
Murphy, S.D. 1980. Assessment of the potential for toxic interactions among environmental
pollutants. In: C.L. Galli, S.D. Murphy, and R. Paoletti, eds. The principles and methods in
modern toxicology. Amsterdam, The Netherlands: Elsevier/North Holland Biomedical Press.
NRC (National Research Council). 1980a. Drinking water and health, Vol. 3. Washington, DC:
National Academy Press, p. 27-28.
NRC (National Research Council). 1980b. Principles of toxicological interactions associated
with multiple chemical exposures. Washington, DC: National Academy Press, p. 204.
A-20
-------
OSHA (Occupational Safety and Health Administration). 1983. General Industry Standards,
Subpart2, Toxic and Hazardous Substances. Code of Federal Regulations. 40:1910.1000
(d)(2)(i). Chapter XVII — Occupational Safety and Health Administration, p. 667.
Plackett, R.L., and P.S. Hewlett. 1948. Statistical aspects of the independent joint action of
poisons. Ann. Appl. Biol. 35:347-358.
Pozzani, U.C., C.S. Weil, and C.P. Carpenter. 1959. The toxicological basis of threshold values:
5. The experimental inhalation of vapor mixtures by rats, with notes upon the relationship
between single dose inhalation and single dose oral data. Am. Ind. Hyg. Assoc. J. 20:364-369.
Rothman, K. 1976. The estimation of synergy or antagonism. Am. J. Epidemiol. 103(5):506-511.
Rothman, K. 1978. Estimation versus detection in the assessment of synergy. Am. J. Epidemiol.
Rothman. K., S. Greenland, and A. Walker. 1980. Concepts of interaction. Am. J. Epidemiol.
112(4):467-470.
Siemiatycki, J., and D.C. Thomas. 1981. Biological models and statistical interactions: An
example from multistage carcinogenesis. Int. J. Epidemiol. 10(4):383-387.
Smyth, H.F., C.S. Weil, IS. West, and C.P. Carpenter. 1969. An exploration of joint toxic action:
I. Twenty-seven industrial chemicals intubated in rats in all possible pairs. Toxicol. Appl.
Pharmacol. 14:340-347.
Smyth, H.F., C.S. Weil, J.S. West, and C.P. Carpenter. 1970. An exploration of joint toxic
action. II. Equitoxic versus equivolume mixtures. Toxicol. Appl. Pharmacol. 17:498-503.
Stara, J.F., D. Mukerjee, R. McGaughy, P. Durkin, and M.L. Dourson. 1983. The current use of
studies on promoters and cocarcinogens in quantitative risk assessment. Environ. Health
Perspect. 50:359-368.
U.S. EPA. 1986a. Guidelines for carcinogen risk assessment. Federal Register.
U.S. EPA. 1986b. Guidelines for mutagenicity risk assessment. Federal Register.
A-21
-------
U.S. EPA. 1986c. Guidelines for the health assessment of developmental toxicants. Federal
Register.
U.S. EPA. 1986d. Guidelines for estimating exposures. Federal Register.
Veldstra, H. 1956. Synergism and potentiation with special reference to the combination of
structural analogues. Pharmacol. Rev. 8:339-387.
Wahrendorf, J., R. Zentgrof, and C.C. Brown. 1981. Optimal designs for the analysis of
interactive effects of two carcinogens or other toxicants. Biometrics 37:45-54.
Walter, S.D. 1976. The estimation and interpretation of attributable risk in health research.
Biometrics 32:829-849.
Walter, S.D., and T.R. Holford. 1978. Additive, multiplicative, and other models for disease
risks. Am. J. Epidemiol. 108:341-346.
Withey, J.R. 1981. Toxicodynamics and biotransformation. In: International Workshop on the
Assessment of Multichemical Contamination. Milan, Italy. (Draft copy courtesy of J.R. Withey).
WHO (World Health Organization). 1981. Health effects of combined exposures in the work
environment. WHO Tech. Report Series No. 662.
A-22
-------
PART B: RESPONSE TO PUBLIC AND SCIENCE ADVISORY BOARD COMMENTS
1. INTRODUCTION
This section summarizes some of the major issues raised in public comments on the
Proposed Guidelines for the Health Risk Assessment of Chemical Mixtures published on January
9, 1985 (50 FR 1170). Comments were received from 14 individuals or organizations. An issue
paper reflecting public and external review comments was presented to the Chemical Mixtures
Guidelines Panel of the Science Advisory Board (SAB) on March 4, 1985. At its April 22-23,
1985, meeting, the SAB Panel provided the Agency with additional suggestions and
recommendations concerning the Guidelines. This section also summarizes the issues raised by
the SAB.
The SAB and public commentators expressed diverse opinions and addressed issues from
a variety of perspectives. In response to comments, the Agency has modified or clarified many
sections of the Guidelines, and is planning to develop a technical support document in line with
the SAB recommendations. The discussion that follows highlights significant issues raised in the
comments, and the Agency's response to them. Also, many minor recommendations, which do
not warrant discussion here, were adopted by the Agency.
2. RECOMMENDED PROCEDURES
2.1. DEFINITIONS
Several comments were received concerning the lack of definitions for certain key items
and the general understandability of certain sections. Definitions have been rewritten for several
terms and the text has been significantly rewritten to clarify the Agency's intent and meaning.
Several commentators noted the lack of a precise definition of "mixture," even though
several classes of mixtures are discussed. In the field of chemistry, the term "mixture" is usually
differentiated from true solutions, with the former defined as nonhomogeneous multicomponent
systems. For these Guidelines, the term "mixture" is defined as ". . any combination of two or
more chemicals regardless of spatial or temporal homogeneity of source" (Section 1). These
Guidelines are intended to cover risk assessments for any situation where the population is
exposed or potentially exposed to two or more compounds of concern. Consequently, the
introduction has been revised to clarify the intended breadth of application.
Several commentators expressed concern that "sufficient similarity" was difficult to
define and that the Guidelines should give more details concerning similar mixtures. The Agency
agrees and is planning research projects to improve on the definition. Characteristics such as
A-23
-------
composition and toxic end-effects are certainly important, but the best indicators of similarity in
terms of risk assessment have yet to be determined. The discussion in the Guidelines emphasizes
case-by-case judgment until the necessary research can be performed. The Agency considered but
rejected adding an example, because it is not likely that any single example would be adequate to
illustrate the variety in the data and types of judgments that will be required in applying this
concept. Inclusion of examples is being considered for the technical support document.
2.2. MIXTURES OF CARCINOGENS AND SYSTEMIC TOXICANTS
The applicability of the preferred approach for a mixture of carcinogens and systemic
(noncarcinogenic) toxicants was a concern of several public commentators as well as the SAB.
The Agency realizes that the preferred approach of using test data on the mixture itself may not
be sufficiently protective in all cases. For example, take a simple two-component mixture of one
carcinogen and one toxicant. The preferred approach would lead to using toxicity data on the
mixture of the two compounds. However, it is possible to set the proportions of each component
so that in a chronic bioassay of such a mixture, the presence of the toxicant could mask the
activity of the carcinogen. That is to say, at doses of the mixture sufficient for the carcinogen to
induce tumors in the small experimental group, the toxicant could induce mortality. At a lower
dose in the same study, no adverse effects would be observed, including no carcinogenic effects.
The data would then suggest use of a threshold approach. Since carcinogenicity is considered by
the Agency to be a nonthreshold effect, it may not be prudent to construe the negative results of
such a bioassay as indicating the absence of risk at lower doses. Consequently, the Agency has
revised the discussion of the preferred approach to allow the risk assessor to evaluate the
potential for masking of carcinogenicity or other effects on a case-by-case basis.
Another difficulty occurs with such a mixture when the risk assessment needs to be based
on data for the mixture components. Carcinogens and systemic toxicants are evaluated by the
Agency using different approaches and generally are described by different types of data:
response rates for carcinogens vs. effect descriptions for toxicants. The Agency recognizes this
difficulty and recommends research to develop a new assessment model for combining these
dissimilar data sets into one risk estimate. One suggestion in the interim is to present separate
risk estimates for the dissimilar end points, including carcinogenic, teratogenic, mutagenic, and
systemic toxicant components.
A-24
-------
3. ADDITIVITY ASSUMPTION
Numerous comments were received concerning the assumption of additivity, including:
a. the applicability of additivity to "complex" mixtures;
b. the use of dose additivity for compounds that induce different effects;
c. the interpretation of the Hazard Index; and
d. the use of interact!on data.
Parts of the discussion in the proposed guidelines concerning the use of additivity
assumptions were vague and have been revised in the final Guidelines to clarify the Agency's
intent and position.
3.1. COMPLEX MIXTURES
The issue of the applicability of an assumption of additivity to complex mixtures
containing tens or hundreds of components was raised in several of the public comments. The
Agency and its reviewers agree that as the number of compounds in the mixture increases, an
assumption of additivity will become less reliable in estimating risk. This is based on the fact that
each component estimate of risk or an acceptable level is associated with some error and
uncertainty. With current knowledge, the uncertainty will increase as the number of components
increases. In any event, little experimental data are available to determine the general change in
the error as the mixture contains more components. The Agency has decided that a limit to the
number of components should not be set in these Guidelines. However, the Guidelines do
explicitly state that as the number of compounds in the mixture increases, the uncertainty
associated with the risk assessment is also likely to increase.
3.2. DOSE ADDITIVITY
Commentators were concerned about what appeared to be a recommendation of the use of
dose additivity for compounds that induce different effects. The discussion following the dose
additivity equation was clarified to indicate that the act of combining all compounds, even if they
induce dissimilar effects, is a screening procedure and not the preferred procedure in developing
a hazard index. The Guidelines were further clarified to state that dose (or response) additivity is
theoretically sound, and therefore best applied for assessing mixtures of similar acting
components that do not interact.
A-25
-------
3.3. INTERPRETATION OF THE HAZARD INDEX
Several comments addressed the potential for misinterpretation of the hazard index, and
some questioned its validity, suggesting that it mixes science and value judgments by using
"acceptable" levels in the calculation. The Agency agrees with the possible confusion regarding
its use and has revised the Guidelines for clarification. The hazard index is an easily derived
restatement of dose additivity, and is, therefore, most accurate when used with mixture
components that have similar toxic action. When used with components of unknown or
dissimilar action, the hazard index is less accurate and should be interpreted only as a rough
indication of concern. As with dose addition, the uncertainty associated with the hazard index
increases as the number of components increases, so that it is less appropriate for evaluating the
toxicity of complex mixtures.
3.4. USE OF INTERACTION DATA
A few commentators suggested that any interaction data should be used to quantitatively
alter the risk assessment. The Agency disagrees. The current information on interactions is
meager, with only a few studies comparing response to the mixture with that predicted by studies
on components. Additional uncertainties include exposure variations due to changes in
composition, mixture dose, and species differences in the extent of the interaction. The Agency is
constructing an interaction data base in an attempt to answer some of these issues. Other
comments concerned the use of different types of interaction data. The Guidelines restrict the use
of interaction data to that obtained from whole animal bioassays of a duration appropriate to the
risk assessment. Since such data are frequently lacking, at least for chronic or subchronic effects,
the issue is whether to allow for the use of other information such as acute data, in vitro data, or
structure-activity relationships to quantitatively alter the risk assessment, perhaps by use of a
safety factor. The Agency believes that sufficient scientific upport does not exist for the use of
such data in any but a qualitative discussion of possible synergistic or antagonistic effects.
4. UNCERTAINTIES AND THE SUFFICIENCY OF THE DATA BASE
In the last two paragraphs of Section II of the Guidelines, situations are discussed in
which the risk assessor is presented with incomplete toxicity, monitoring, or exposure data. The
SAB, as well as several public commentors, recommended that the "risk management" tone of
this section be modified and that the option of the risk assessor to decline to conduct a risk
assessment be made more explicit.
This is a difficult issue that must consider not only the quality of the available data for
risk assessment, but also the needs of the Agency in risk management. Given the types of poor
A-26
-------
data often available, the risk assessor may indicate that the risk assessment is based on limited
information and thus contains no quantification of risk. Nonetheless, in any risk assessment,
substantial uncertainties exist. It is the obligation of the risk assessor to provide an assessment,
but also to ensure that all the assumptions and uncertainties are articulated clearly and quantified
whenever possible.
The SAB articulated several other recommendations related to uncertainties, all of which
have been followed in the revision of the Guidelines. One recommendation was that the summary
procedure table also be presented as a flow chart so that all options are clearly displayed. The
SAB further recommended the development of a system to express the level of confidence in the
various steps of the risk assessment.
The Agency has revised the summary table to present four major options: risk assessment
using data on the mixture itself, data on a similar mixture, data on the mixture's components, or
declining to quantify the risk when the data are inadequate. A flow chart of this table has also
been added to more clearly depict the various options and to suggest the combining of the several
options to indicate the variability and uncertainties in the risk assessment.
To determine the adequacy of the data, the SAB also recommended the development of a
system to express the level of confidence associated with various steps in the risk assessment
process. The Agency has developed a rating scheme to describe data quality in three areas:
interaction, health effects, and exposure. This classification provides a range of five levels of data
quality for each of the three areas. Choosing the last level in any area results in declining to
perform a quantitative risk assessment due to inadequate data. These last levels are described as
follows:
Interactions: An assumption of additivity cannot be justified, and no quantitative risk
assessment can be conducted.
Health effects: A lack of health effects information on the mixture and its components
precludes a quantitative risk assessment.
Exposure: The available exposure information is insufficient for conducting a risk
assessment.
Several commentors, including the SAB, emphasized the importance of not losing these
classifications and uncertainties farther along in the risk management process. The discussion of
uncertainties has been expanded in the final Guidelines and includes the recommendation that a
A-27
-------
discussion of uncertainties and assumptions be included at every step of the regulatory process
that uses risk assessment.
Another SAB comment was that the Guidelines should include additional procedures for
mixtures with more than one end point or effect. The Agency agrees that these are concerns and
revised the Guidelines to emphasize these as additional uncertainties worthy of further research.
5. NEED FOR A TECHNICAL SUPPORT DOCUMENT
The third major SAB comment concerned the necessity for a separate technical support
document for these Guidelines. The SAB pointed out that the scientific and technical background
from which these Guidelines must draw their validity is so broad and varied that it cannot
reasonably be synthesized within the framework of a brief set of guidelines. The Agency is
developing a technical support document that will summarize the available information on health
effects from chemical mixtures, and on interaction mechanisms, as well as identify and develop
mathematical models and statistical techniques to support these Guidelines. This document will
also identify critical gaps and research needs.
Several comments addressed the need for examples on the use of the Guidelines. The
Agency has decided to include examples in the technical support document.
Another issue raised by the SAB concerned the identification of research needs. Because
little emphasis has been placed on the toxicology of mixtures until recently, the information on
mixtures is limited. The SAB pointed out that identifying research needs is critical to the risk
assessment process, and the EPA should ensure that these needs are considered in the research
planning process. The Agency will include a section in the technical support document that
identifies research needs regarding both methodology and data.
A-28
-------
APPENDIX B
DEFINITIONS
Consistent and clear terminology is critical to the discussion of chemical mixtures risk
assessment methodology. Tables A-l and A-2 articulate the differences among the many terms
used to describe chemical mixtures and the types of interactions that may occur among
chemicals. Table A-l presents chemical mixtures definitions in terms of specific criteria
including the complexity of the mixture, similarity of biologic activity, similarity of chemical
structure or mixture composition, environmental source of the mixture, toxic endpoint, etc.
Table A-2 provides definitions for terms that describe various types of toxicologic interactions,
including forms of additivity, antagonism, synergism, and other toxicologic phenomena.
Tables A-l and A-2 can be used by the risk assessor to classify available toxicity and exposure
data in order to choose from among the risk assessment methods for chemical mixtures.
B-l
-------
Table B-l. Definitions of chemical mixtures
Chemical Mixture
Any set of multiple chemical substances that may or may not be identifiable, regardless of their
sources, that may jointly contribute to toxicity in the target population. May also be referred to as a
"whole mixture" or as the "mixture of concern."
Components
Single chemicals that make up a chemical mixture that may be further classified as systemic toxicants,
carcinogens, or both.
Simple Mixture
A mixture containing two or more identifiable components, but few enough that the mixture toxicity
can be adequately characterized by a combination of the components' toxicities and the components'
interactions.
Complex Mixture
A mixture containing so many components that any estimation of its toxicity based on its components'
toxicities contains too much uncertainty and error to be useful. The chemical composition may vary
over time or with different conditions under which the mixture is produced. Complex mixture
components may be generated simultaneously as by-products from a single source or process,
intentionally produced as a commercial product, or may coexist because of disposal practices. Risk
assessments of complex mixtures are preferably based on toxicity and exposure data on the complete
mixture. Gasoline is an example.
Similar Components
Single chemicals that cause the same biologic activity or are expected to cause a type of biologic
activity based on chemical structure. Evidence of similarity may include similarly shaped dose-
response curves, or parallel log dose-probit response curves for quantal data on the number of animals
(people) responding, and same mechanism of action or toxic endpoint. These components are
expected to have comparable characteristics for fate, transport, physiologic processes, and toxicity.
Similar Mixtures
Mixtures that are slightly different, but are expected to have comparable characteristics for fate,
transport, physiologic processes, and toxicity. These mixtures may have the same components but in
slightly different proportions, or have most components in nearly the same proportions with only a few
different (more or fewer) components. Similar mixtures cause the same biologic activity or are
expected to cause the same type of biologic activity due to chemical composition. Similar mixtures
act by the same mechanism of action or affect the same toxic endpoint. Diesel exhausts from different
engines are an example.
Chemical Classes
Groups of components that are similar in chemical structure and biologic activity, and that frequently
occur together in environmental samples, usually because they are generated by the same commercial
process. The composition of these mixtures is often well controlled, so that the mixture can be treated
as a single chemical. Dibenzo-dioxins are an example.
B-2
-------
Table B-2. Definitions of toxicologic interactions between chemicals"
Additivity
When the "effect" of the combination is estimated by the sum of the exposure levels or the effects of
the individual chemicals. The terms "effect" and "sum" must be explicitly defined. Effect may refer to
the measured response or the incidence of adversely affected animals. The sum may be a weighted
sum (see "dose addition") or a conditional sum (see "response addition").
Antagonism
When the effect of the combination is less than that suggested by the component toxic effects.
Antagonism must be defined in the context of the definition of "no interaction," which is usually dose
or response addition.
Chemical Antagonism
When a reaction between the chemicals has occurred and a new chemical is formed. The toxic effect
produced is less than that suggested by the component toxic effects.
Chemical Synergism
When a reaction between the chemicals has occurred and a different chemical is formed. The toxic
effect produced is greater than that suggested by the component toxic effects, and may be different
from effects produced by either chemical by itself.
Complex Interaction
When three or more compounds combined produce an interaction that cannot be assessed according to
the other interaction definitions.
Dose Additivity
When each chemical behaves as a concentration or dilution of every other chemical in the mixture.
The response of the combination is the response expected from the equivalent dose of an index
chemical. The equivalent dose is the sum of component doses scaled by their toxic potency relative to
the index chemical.
Index Chemical
The chemical selected as the basis for standardization of toxicity of components in a mixture. The
index chemical must have a clearly defined dose-response relationship.
Inhibition
When one substance does not have a toxic effect on a certain organ system, but when added to a toxic
chemical, it makes the latter less toxic.
Masking
When the compounds produce opposite or functionally competing effects at the same site or sites, so
that the effects produced by the combination are less than suggested by the component toxic effects.
No Apparent Influence
When one substance does not have a toxic effect on a certain organ or system, and when added to a
toxic chemical, it has no influence, positive or negative, on the toxicity of the latter chemical.
B-3
-------
Table B-2. Definitions of toxicologic interactions between chemicals" (continued)
No Observed Interaction
When neither compound by itself produces an effect, and no effect is seen when they are administered together.
Potentiation
When one substance does not have a toxic effect on a certain organ or system, but when added to a toxic
chemical, it makes the latter more toxic.
Response Additivity
When the toxic response (rate, incidence, risk, or probability of effects) from the combination is equal to the
conditional sum of component responses as defined by the formula for the sum of independent event probabilities.
For two chemical mixtures, the body's response to the first chemical is the same whether or not the second
chemical is present.
Synergism
When the effect of the combination is greater than that suggested by the component toxic effects. Synergism must
be defined in the context of the definition of "no interaction," which is usually dose or response addition.
Unable to Assess
Effect cannot be placed in one of the above classifications. Common reasons include lack of proper control
groups, lack of statistical significance, and poor, inconsistent, or inconclusive data.
aBased on definitions in U.S. EPA (1990). These definitions of interaction refer to the influence on observed
toxicity, without regard to the actual modes of interaction.
B-4
-------
APPENDIX C
PHARMACOKINETICS
The improvement in predictive estimates for mixture risk will likely follow the increase
in biological understanding and quantitative models of toxicologic interactions. The best studied
and modeled toxicologic interactions are those involving alterations in pharmacokinetics (e.g.,
see Krishnan et al., 1994). This section discusses the general concepts underlying toxicologic
interactions and more specific issues with pharmacokinetic models to provide background and
incentive for continued research in this area.
C.I. PHARMACOKINETIC/PHARMACODYNAMIC MODELING
The past two decades have seen great strides in our ability to assess the health risks of
chemicals present in our air, water, and food. Our ever-growing scientific databases are
increasing our understanding of the dose-response toxicity of individual chemicals and are
permitting better predictions of health effects. However, we are now reaching the point at which
we can, and must, increase the complexity of our calculations and incorporate chemical-chemical
interactions into our risk assessment analyses.
Although single-compound exposures are possible, in most instances contaminant chemi-
cals are present in our environment as mixtures. Some of these mixtures are relatively well
defined, such as coke oven emissions and diesel exhaust. Other mixtures, such as those released
from old disposal sites, are highly variable, complex, and largely undefined. As there is a
considerable body of literature indicating that chemical-chemical interactions occur, factors that
influence the toxicity of the chemicals in mixtures must be better understood if they are to be
effectively incorporated into our health risk assessments (U.S. EPA, 1986).
In theory, there are many ways in which one chemical could alter the toxicity of another.
Two chemicals could directly interact to form a new compound, or there might be changes in the
intestinal absorption of the chemicals. Absorption could be altered through competition for
membrane-binding sites or by the induction of a transport process. Plasma transport, tissue
accumulation, and elimination processes could also be altered through competition or
interference mechanisms, e.g., binding to metallothionein. Cellular metabolism and intracellular
effects may be modified either directly through competition for receptor- or enzyme-binding sites
or indirectly by the induction or depression of metabolizing enzymes and/or other detoxification
mechanisms, such as cellular glutathione levels.
Assessment of the health impacts of single chemicals or chemical mixtures present in our
environment is an important problem. Although we have made progress in recent years by
establishing "safe" concentrations and exposure conditions for many individual chemicals,
C-l
-------
related information for the same chemicals in mixtures is largely unavailable. Our challenge is to
accurately evaluate the risk posed by exposure to multiple chemicals as compared to exposures to
individual chemicals. This will occur only with a solid understanding of the modes of action or
mechanisms of toxicity of chemical agents and the factors that control their absorption,
metabolism, distribution, and elimination.
Chemical interactions can be divided into two major categories: those resulting from
toxicokinetic and those resulting from toxicodynamic modes of action. Toxicokinetic modes of
interaction involve alterations in metabolism or disposition of a toxic chemical. These
interactions can be mediated by the induction or inhibition of enzymes involved in xenobiotic
activation and detoxification. Toxicodynamic modes of interaction include interactions that do
not directly affect the metabolism or disposition of a xenobiotic, but affect a tissue's response or
susceptibility to toxic injury. Modes of toxicodynamic interactions include, among others,
depletion or induction of protective factors, alterations in tissue repair, changes in
hemodynamics, and immunomodulation. Sauer and Sipes (1995) have reported toxicodynamic
action between alWram'-retinol and other chemicals that involves the alteration of chemical-
induced tissue injury by the modulation of inflammatory cell activity.
Retinol pretreatment in this study provided protection against pulmonary toxicity induced
by 2-nitronaphthalene and paraquat by suppressing the inflammatory response. The investigators
looked at effects on liver for the combination of retinol and 2-nitronaphthalene. With this target
organ, they observed a potentiation of toxicity, rather than protection as seen in the lung. A
subsequent experiment indicated that retinol-induced activation of Kupffer cell function was a
major contributing factor in the lung. The selective destruction of Kupffer cells by gadolinium
pretreatment protected rats against the potentiation induced by retinol. From these studies, it is
clear that it can be difficult to predict interactions from one organ to another, let alone from
species to species. Likewise, results described indicate that in vitro studies alone would have
been of limited use in describing the range of effects observed in the intact animal with these
combinations.
Glutathione (GSH) plays a critical role in detoxifying many chemicals, and its depletion
within cells has long been known to increase the risk of chemical toxicity. Jones et al. (1995)
have provided information on factors that regulate GSH status in humans, including gender, age,
race, and dietary habits that could affect the risk of exposure. GSH levels in human plasma are
highly variable and potentially a marker of susceptibility. Because of GSH's central role in
detoxifying many chemicals, therapeutic manipulation of GSH levels may afford extra protection
that could reduce the risks of exposure to complex mixtures.
The utility of physiologically based pharmacokinetic (PBPK) modeling in predicting the
consequences of exposure to multiple solvents has been demonstrated by Krishnan and Pelekis
C-2
-------
(1995). The authors used PBPK models and existing data sets to predict the effect of multiple
solvent exposure on carboxyhemoglobin formation from dichloromethane. The interaction
involved the hepatic metabolism of the various solvents by one isozyme of cytochrome P450
(CYP2E1) and the effect of one metabolite, CO, on hemoglobin. Their predictions highlighted
the need to understand the disposition of chemicals and modes of action of toxicity in order to
effectively use PBPK in risk assessment.
This modeling exercise suggested that, with competitive metabolic inhibition mechanism,
the threshold for the appearance of binary chemical interactions will follow a downward trend
with increasing number of substrates or structurally similar substances in a mixture. The use of
this kind of mechanistic model, along with data from descriptive chemical interaction studies,
could form the very basis of mechanistic risk assessment methods for complex chemical
mixtures.
Several studies on toxic interactions have been published to date; the quantitative aspect
of the toxicokinetic/toxicodynamic mechanism of interactions, however, has only been
elucidated for a few chemical pairs (Krishnan and Brodeur, 1991). One approach to the problem
in assessing risk in the context of a complex mixture would be to develop biologically based
dosimetry and toxicity models, such that multiple interactions can be simultaneously
distinguished and systematically analyzed at any level of complexity. Physiologically based
pharmacokinetic and pharmacodynamic modeling (PBPK/PD) may therefore be considered a
viable approach. Tardif et al. (1997) developed a PBPK model for a ternary mixture of alkyl
benzenes in rats and humans. Model simulations and experimental data obtained in humans
indicated that exposure to atmospheric concentrations of the alkyl benzenes that remained within
the permissible concentrations (TLVs) for a mixture would not result in biologically significant
modifications of their pharmacokinetics. This study demonstrated the utility of PBPK models in
the prediction of the kinetics of components of chemical mixtures, by accounting for modes of
interaction of binary chemical mixtures.
The linkage of two of the most challenging areas in toxicology today, PBPK/PD and
statistical/mathematical modeling and experimental toxicology of chemical mixtures, will have
immense potential in application to risk assessment for chemical mixtures. Figure B-l represents
the possible application of combined PBPK/PD modeling to chemical mixtures and the
development of innovative risk assessment methodologies for chemical mixtures. El-Masri et al.
(1996) attempted to couple PBPK/PD and other experimental toxicology with isobolographic
analysis and/or response surface methodology for the modeling and analysis of toxicologic
interactions. With the aid of such techniques as Monte Carlo simulation, one may then estimate
tissue dosimetry at the pharmacokinetic and pharmacodynamic levels. Using these tissue values
-------
A Priori PBPK/PD Modeling
Model-Directed Focused Experiments/
Efficient Experimental Designs
PBPK/PD Isobolographic Analysis
Integrated + Response Surface Methodology
Toxicity Monte Carlo Simulation
Model
1
Predictive and Alternative Toxicology/
Target Tissue Dosimetry
Innovative Risk Assessment
Methodologies
Figure C-l.
The possible application of combined PBPK/PD modeling to chemical mixtures.
C-4
-------
as benchmark doses, human risk assessment of chemical mixtures may possibly be carried out
with quantification of the uncertainty.
C.2. PHARMACOKINETIC PRINCIPLES: CHEMICAL MIXTURES
Environmental exposures to naturally occurring and artificially produced substances
generally involve mixtures of chemicals. Exposure to single chemicals occurs in the context of
simultaneous exposure. When therapeutic agents are taken with the intent to produce a certain
pharmacological effect, other chemicals present at the time of their disposition may modulate
processes of absorption, tissue distribution, metabolism, or excretion so as to alter the shape of
the dose-effect relationship. Toxicokinetic interactions may influence the relationship between
administered dose and the dose delivered to the target site(s). This forces the distinction between
toxicokinetic interactions and toxicodynamic interactions. Toxicologic agents, or pharmacologic
agents administered at doses at which they exert other than their intended effects, more than
likely will interact with a variety of receptor sites, reversibly or irreversibly. Metabolites, in
particular, although they may be formed in very small amounts, may not move from the tissue or
even the intracellular site where they were produced. Given this broad spectrum of modes of
action, it is not surprising that toxicodynamic models of action and interaction are less fully
developed than toxicokinetic models. The interactions among chemicals may occur at any point
during absorption or disposition of the chemical components of the mixture. O'Flaherty (1989)
reviewed these modes of kinetic interaction during absorption and elimination; the following
discussions summarize this review and include other pertinent information available in the
current literature.
C.2.1. Absorption
C.2.1.1. Gastrointestinal
Gastrointestinal transit time may be affected by the constituents of a mixture. For
example, absorption may be higher or lower depending on transit time. Although some
lipophilic substances, such as paraffin oil and triglycerides, do not affect uptake, others such as
lipophilic substances possessing hydrophilic groups such as oleic acid and oleyl alcohol alter
absorption into the outermost layer of the glandular mucosa. When both hydrophilic and
lipophilic groups are present in the solvent with a dominant hydrophilic characteristic, an
administered compound readily penetrates into the stomach wall (Ekwall et al., 1951). Many
other factors, e.g. acid-base balance in the gastrointestinal lumen, gut mobility, and blood flow,
also affect the absorption of many xenobiotics. From a practical point of view, it is important to
differentiate between interactions that alter the rate of absorption from those that affect the
amount of xenobiotic absorbed. Kristensen (1976) has reported that a rate of absorption
C-5
-------
contributing to a longer plasma half-life may be needed to maintain a steady-state concentration
of certain drugs, e.g., antihypertensive drugs, whereas a shorter plasma half-life, or attainment of
higher unbound plasma levels of an active drug (e.g., digitoxin, oubain) because of rapid passage
across the gut may be important when a quick onset of drug effect is desired.
The competitive binding of metals to macromolecules can influence their intestinal
absorption, plasma transfer, tissue uptake, intracellular binding, and site-specific toxic effects.
The following discussion cites examples of such interactions. Although many have not been
studied in detail, it is possible that we have a lot to discover in this area.
The intestinal absorption and tissue accumulation of most toxic metals are influenced, to
a large extent, by the concentration of essential trace metals present in one's diet (Eisenhans et
al., 1991). The intestinal uptake of cadmium (Cd), for example, is significantly increased under
conditions of iron (Fe), zinc (Zn), and calcium (Ca) deficiency (Hoadley and Johnson, 1987).
Dietary Zn alters lead (Pb) toxicity, as evidenced by decreased Pb absorption, lower blood and
tissue Pb levels, and decreased inhibition of the Pb-sensitive enzyme aminolevulenic acid
dehydrase (ALAD) (Cerklewski and Forbes, 1976) under conditions of elevated dietary Zn
exposure.
The mechanisms underlying these effects undoubtedly involve multiple processes. Some
of these interactions occur through competition of the metal ions for membrane transport
systems, in a manner similar to that described by Blazka and Shaikh (1992) for Cd. These
investigators have found that Cd uptake by rat hepatocytes occurs through a sulfhydryl (SH)-
containing transport process that is inhibited by concomitant exposure to copper (Cu), iron, and
zinc. Thus, the relative extracellular concentrations of these ions will be an important
determinant of Cd uptake and accumulation. In vivo studies of hepatic Cd, Cu, and Zn uptake
and accumulation suggest that influx and efflux of metal ions are both important determinants of
final tissue metal concentrations (Suzuki et al., 1991).
In addition to mediating cellular toxicity in target organs, metallothionein (MT) in
intestinal cells alters the absorption of metals from dietary sources. Richards and Cousins (1975)
have proposed that MT regulates Zn absorption by chelating Zn ions in intestinal cells,
preventing their transfer across the basal membrane into the circulatory system. This proposed
function of MT is supported by the observation that intestinal MT concentrations are inversely
proportional to Zn absorption (Bremner, 1993). The binding of Cd ions to MT in the intestine
similarly decreases Cd absorption. Foulkes (1991) has demonstrated that pretreatment of animals
with Zn at levels that increase mucosal MT content causes a decrease in Cd transport across the
intestinal lumen.
Adsorption can reduce bioavailability from the gastrointestinal tract. Prescott (1969)
demonstrated that the salts of Ca, Fe, or magnesium (Mg) may interact with drugs in the intestine
C-6
-------
to produce insoluble and nonabsorbable complexes. For example, calcium phosphate filler
markedly reduces the absorption of tetracycline. In addition to calcium salts, Fe and aluminum
(Al) ions also form insoluble chelate complexes with tetracycline. These interactions, of
potential clinical significance, are avoidable if the drugs are given at properly spaced time
intervals (Neuvonen, 1976).
C.2.1.2. Pulmonary
Gaseous and particulate phases of an inhaled chemical mixture may play different
functional roles inducing or reducing pulmonary/systemic toxicity. For example, formaldehyde
can stimulate mucociliary function at low concentrations, but it inhibits mucociliary function
after prolonged exposure at high concentrations (Morgan et al., 1984). Gaseous and particulate
phases of cigarette smoke are cilia toxic and at sustained high levels can cause impairment of
tracheobronchial clearance. Low, brief exposures, however, actually appear to speed up lower
bronchial transport. In occupational settings, chronic exposures lower than those associated with
ambient air may significantly interfere with pulmonary clearance and may produce a variety of
toxicological events uncommon to the individual constituents of the mixture (Albert et al., 1975;
Ferin and Leach, 1973).
Airborne particulates, when adsorbed to chemical constituents of gases/vapors, may
influence the degree of absorption from the lung. Other factors, such as particle size, length, and
binding affinity, can also play a significant role in pulmonary absorption/retention. Henry and
Kaufman (1973) suggested that the ability of benzo[a]pyrene (B[a]P) to be eluted for its
particulate adsorption sites might be an important determinant of its biological activity. Creasia
et al. (1976) reported that B[a]P adsorbed to the larger carbon particles was cleared with the
particles themselves. Because the half-times of the large and small particles were similar, B[a]P
adsorbed to the smaller carbon particles was cleared about four times as fast as the particles from
the mouse lung.
C.2.1.3. Dermal
Despite lack of sufficient quantitative information, solvent effects on qualitative
absorption for the dermal route are well characterized. Within a limited range at least, partition
coefficients calculated for solubilities in skin and in various solvents appear to correlate with
permeability coefficients for penetration into the skin for those solvents (Sloan et al., 1986).
Although an adequate amount of information is known about the uptake of several classes
of neat chemicals (as liquids) through human skin, more needs to be known about the effects of
media on dermal uptake. In the workplace, employees are frequently exposed to liquid
chemicals, but environmental exposure almost never involves exposure to neat substances. For
C-7
-------
example, residents may be exposed to contaminated dust that has been transported through open
windows. Children are exposed to soils that have contaminants from particulate emissions from
cars, smelters, foundries, incinerators, or other processes, which have been deposited on yards
and playgrounds. Adults and children can also be exposed to organic contaminants in water
during showering or swimming.
Information on the neat chemical is helpful in understanding the dermal uptake of
chemicals bound to soil, dust, sludge, sediment, paint, etc., but there are other factors that should
also be considered. The best approach for mixtures assessments is to conduct specific tests with
the contaminated chemical on laboratory animals or use in vitro technologies. Since relatively
low concentrations of the chemical are typical in the environment and high concentrations are
used in laboratory studies, an extrapolation to environmental levels is often necessary. Other
factors such as the duration of contact, integrity of the skin, and the chemical properties of the
agent must ultimately be considered in the risk assessment.
Progress continues to be made to allow risk assessors to make fairly reasonable estimates
of the uptake of chemicals in soil. The development of models that can predict dermal
bioavailability and account for media effects would represent a signficant step forward. The role
of concentration on the rate of dermal uptake is an area that deserves further study. Work
conducted thus far suggests that the uptake will depend on the characteristics of the media (%
organics, particle size in soil, etc.) and the properties of the contaminant (lipophilicity,
temperature). These parameters need to be quantified and a general model developed. The work
of McKone (1990) represents an important step in this direction.
C.2.1.4. Elimination
Metabolism of one chemical may deplete reserves of a cofactor required for metabolism
of another chemical, reducing exposure to metabolites of the second chemical or shifting the
relative magnitudes of exposure to products of competing metabolic pathways. Induction of
metabolizing enzymes, often those of cytochrome P-450-dependent mixed-function oxidase
(MFO) systems, can alter the relative magnitudes of parallel pathways of metabolism as well as
increase the rate of magnitude of total metabolic production (O'Flaherty, 1989).
Andersen et al. (1987), while developing a PBPK model, considered the interaction
between 1,1-dichloroethylene (1,1-DCE) and trichloroethylene (TCE) metabolized by the same
enzyme system. In this study rats were exposed to these chemicals via inhalation. When the
chemicals reached dynamic steady states among the tissues and between blood and alveolar air,
the rate of loss of 1,1-DCE was found to be sharply reduced in the presence of TCE. Of the
several modeled mechanisms of interaction, competitive interaction gave the most successful
predictions. This led to the development of a co-exposure model with competitive interaction to
-------
predict the kinetic behavior of either compound in the presence of the other. The success with
which this was done was illustrated by a good concordance between predicted and observed
chamber concentrations of 1,1 -DCE without and with coexposure to TCE.
Induction of metabolizing enzymes may produce different effects on metabolic rates,
which could reduce integrated exposure to the parent chemical by increasing the rate of
metabolism. For instance, caffeine metabolism has been modeled as a capacity-limited process
giving rise to the three monitored metabolites (York et al., 1987). Elimination of the metabolites
was assumed to be first-order, an assumption justified by the observations that at no time did the
concentration of any metabolite exceed 1/10 of the maximum caffeine concentration and the
caffeine itself, indicating moderated capacity-limited behavior. Integrated exposure to caffeine,
as expected, decreased as a consequence of induction of caffeine metabolism; however,
integrated exposure to individual monodemethylated metabolites was also decreased by induction
of caffeine metabolism. This could probably be explained by consideration of a process of
caffeine elimination.
The toxicity of many organic chemicals is influenced by the action of mixed-function
oxidases (MFCs) and phase n biotransformation enzymes that catalyze their metabolism to more
hydrophilic forms in preparation for excretion. Because the synthesis of many of these enzymes
is affected by the chemicals they metabolize, multiple modes of action may be involved in the
chemical interactions involving these enzyme systems (Kedderis, 1990). For example, an
inhibition of toxicity can occur when the metabolism of one chemical to its more toxic form is
prevented by the preferential metabolism of another compound, or when one chemical induces an
MFO enzyme system that can catalyze the transformation of a second chemical to a less toxic
form. On the other hand, enhancement of toxicity can occur when the enzyme that bioactivates a
chemical has been previously induced in a cell by exposure to a second compound. Thus, the
toxicity of each individual chemical, in each situation, will depend on which biotransformation
enzymes have been induced, the relative affinity of each chemical for the available enzymes, and
the relative toxicity of the metabolized forms of the chemicals compared with the parent
compounds.
There are numerous examples of chemical interactions in experimental animals that have
their genesis in biotransformation. Chemicals such as piperonal butoxide and proadifen (SK&F
525 A), which inhibit MFO enzymes, decrease the hepatic toxicity of such compounds as
acetaminophen, bromobenzene, and cocaine, which require activation for toxicity (Thompson et
al., 1979). Increased toxicity can also occur when MFO enzymes are inhibited if a compound is
normally converted by these enzymes to a less toxic form. This appears to be the basis for the
increased nephrotoxicity of cyclosporine that occurs following cotreatment with compounds such
as ketoconazole, methyltestosterone, and erythromycin (Moller and Ekelund, 1985).
C-9
-------
In addition, the timing of the multiple-chemical exposures and the doses used can affect
the outcome of an interaction study (Plaa and Vezina, 1990). Plaa and Hewitt (1982), for
example, demonstrated that the magnitude of hepatotoxicity caused by chloroform varied more
than 100-fold when a second chemical, 2,5-hexanedione, was administered 10 versus 50 hours
before the chloroform. Also, MacDonald et al. (1982) have shown that whereas low doses of
acetone enhanced the toxicity of haloethanes such as trichloroethane, high doses reduced toxicity.
Thus, nonlinear or biphasic response curves for individual chemicals will lead to nonlinear and
sbiphasic interactive effects that must be considered in predictive studies.
C-10
------- |