EPA/600/R-94/204a
October 1992
HAZARD IDENTIFICATION IN
CARCINOGEN RISK ANALYSIS:
AN INTEGRATIVE APPROACH
by
Douglas J. Crawford-Brown
Kenneth G. Brown
October, 1992
This project report is part of the U. S. EPA's program of Research to Improve Health Risk
Assessment. It was funded by the Office of Health and Environmental Assessment, Office of
Research and Development, U. S. Environmental Protection Agency, Washington, D. C. 20460.
-------
DISCLAIMER
Although the information in this document has been funded by the U.S. Environmental
Protection Agency, under contract 68-C9-0009, Work Assignment S-l-56 (Kenneth G. Brown,
Ph.D., Inc.), it does not necessarily reflect the views of the Agency and no official endorsement
should be inferred. Mention of trade names or commercial products does not constitute
endorsement or recommendation for use.
-------
CONTENTS
Tables and Working Tables iv
Figures v
Preface vi
EXECUTIVE SUMMARY 1
FOREWORD 4
PART I RISK, REGULATORY SCIENCE, RATIONALITY AND SOCIETAL VALUES
1. BACKGROUND AND TASK OBJECTIVE 10
2. INTRODUCTION 12
3. CONSTRUCTING A COMPLETE VISION OF RATIONALITY 17
3.1. Seven Features Essential to a Rational System 17
3.2. Skills Required by the Rational Risk Analyst 19
3.3. Value Judgments in Rational Action 19
3.4. Broad Classes of Rationality Underlying Societal Debates 27
4. WHERE IS RATIONALITY FOUND? 33
4.1. Descriptive and Prescriptive Rationality 33
4.2. The Rationality of Beliefs, Means, and Ends 34
4.3. The Separation of Beliefs, Means, and Ends 39
4.4. Rationality of Parts and Rationality of Relationships 41
4.5. Summary Remarks on Rationality for Risk Analysis 42
5. CHARACTERISTICS OF REASONS AND RATIONALITY 43
5.1. The Classical Theory 44
5.2. The Skeptical Attack on the Classical Theory 47
5.3. Empiricist Rationality 47
5.4. Rationalist Rationality 47
5.5. Logical Positivism 48
5.6. Rational Skepticism 48
5.7. Probabilistic Rationality 50
5.8. The Problem of Incommensurability 51
5.9. Tests of Theories Other Than Empirical Tests 53
5.10. The Rationality of Crafting in Science .54
5.11. Sociological Views of Science 56
5.12. Foundations Versus Coherence 57
i i i
-------
Contents (continued)
5.13. Summary 58
6. CONCLUDING REMARKS 59
REFERENCES (PART I) 64
PART II ASSESSING EVIDENTIAL SUPPORT FOR CLAIMS OF CARCINOGENICITY:
ESSENTIAL ELEMENTS AND A FRAMEWORK FOR APPLICA TION
1. INTRODUCTION 68
1.1. Background and Objective 68
1.2. The Need for Rational Discourse 72
1.3. Elements of the Framework 74
2. CANCER MECHANISMS AND BIOLOGICALLY SIGNIFICANT DOSE 77
3. MEANING OF "CARCINOGENICITY" 83
3.1. Current Classifications of Carcinogenicity 83
3.2. A Taxonomy of Claims of Carcinogenicity 85
4. INFORMATIONAL BASE 90
4.1. Assembling the Observational Base 90
4.2. Assessing Data Quality by Observational Context 102
4.2.1. Completeness 105
4.2.2. Utility 107
4.2.2.1. Utility of Animal Studies 109
4.2.2.2. Utility of Epidemiologic Studies 112
4.2.2.3. Summarizing Utility 117
4.2.3. Observed Effect 118
4.2.4. Causality 123
5. WARRANTING CLAIMS TO CARCINOGENICITY 126
5.1. Relevance Strategies 129
5.2. Application of Relevance Strategies to Warrant Intra-Context Claims of
Carcinogenicity 133
5.2.1. Direct Empirical Warranting 138
5.2.2. Semi-Empirical Warranting 140
iv
-------
Contents (continued)
5.2.3. Warranting from Empirical Correlations 141
5.2.4. Warranting by Theory-Based Inference 150
5.2.5. Warranting by Existential Insight 156
5.3. Warranting Inter-Context Premises 158
5.4. Linking the Observation Categories/Items to Premises Required by Inter-Context
Relevance Strategies 161
5.4.1. Conversion from Exposure Conditions to BSDR 166
5.4.2. Host Factor and/or Concurrent Conditions Do Not
Differ Significantly 174
5.4.3. Extrapolation Across Doses and/or Dose-Rates 177
5.4.4. Consideration of Intrasubject and Intersubject Variability 181
5.5. Column/Row Summaries and the Issue of Coherence 182
6. THE SEVEN STEPS TO HAZARD IDENTIFICATION: AN OVERVIEW 185
REFERENCES (PART II) 195
v
-------
TABLES
0. A Summary of Ideas in Chapter 1 and Their Relationship
to the Framework of Analysis Provided in Chapter 2 61
1. Classification Scheme Utilized in the Present Report 88
2. Summary of the Role of Observation Statements
in Evidential Reasoning for Carcinogenicity Claims 92
3. Broad Theoretical Implications of Data Categories 96
4. Checklist for Methodological Critique of Epidemiologic Study 116
5. Developmental Processes that Enhance Susceptibility to
Environmental Pollutants 121
6. Prevelance of Subgroups Hypersusceptible to Effects of Common Pollutants 122
7. Rule of Data Categories in Carcinogenicity Claims 135
8. Relevance Strategies Available for Claims of Carcinogenicity 137
9. Role of Data Categories in Inter-Context Premises 159
10. Relevance Strategies Available for Extrapolation Premises 164
WORKING TABLES
1. Context Specification for Hazard Identification 100
2. Data Characteristics for Observational Contexts (Context No.) 103
3. Intra-Context Support for Claims of Carcinogenicity (Context No.) 128
4. Data Characterisitics for Observational Contexts
(From Context No. to Context No.) 162
5. Support for Inter-Context Extrapolation Premises
(From Context No. to Context No.) 165
6. Inter-Context Support for Claims of Carcinogenicity
(From Context No. to Context No.) 167
7. Summary of Overall Assessments for Claims of Carcinogenicity
by Context (Context No.) 168
vi
-------
FIGURES
l! Flow Diagram of Judgments for BSDR 78
2. Main Steps in the Carcinogenic Process 81
3. Levels of Taxonomic Claims of Carcinogenicity 89
4. The Seven Steps of Hazard Identification
in Carcinogen Risk Analysis 187
5. Flow Diagram for the Seven Steps of Hazard
Identification with Application of Working Tables 190
vi i
-------
PREFACE
This report was supported by the U.S. Environmental Protection Agency, Office of
Health and Environmental Exposure Assessment, as part of the Agency's program in "Research
to Improve Health Risk Assessment". Part I of the report describes the philosophical
foundations for rationality and reasoning available to the analyst for risk assessment. Part II
develops a framework to aid the analyst in systematically assembling and evaluating all
observational information potentially relevant to forming a judgment of carcinogenicity for an
agent of interest.
The strength and type of evidential support for warranting claims of carcinogenicity from
observational evidence in Part II derives from the foundational principles discussed in Part I.
By implementing these principles in a step-by-step procedure the analyst is systematically guided
toward developing a weight-of-evidence judgment of carcinogenicity. Human judgment is a key
element required throughout. The stepwise procedure is an aid to (not a substitute for) forming
judgments based on the available evidence. Individuals may make different judgments from the
same observational evidence, particularly if they differ in their "intellectual obligation" regarding
types of relevance strategies deemed necessary for warranting claims of carcinogenicity.
Differences of opinions, knowledge, and perspectives that individuals bring to bear on evaluating
evidence and making judgments will also contribute to diversity. Rational discourse between
persons who have formed their individual weight-of-evidence judgments on claims of
carcinogenicity is recommended to form a "belief' (e.g., an Agency classification regarding
carcinogenicity). If persons have followed the same procedural steps in arriving at different
judgments of carcinogenicity, then the sources of their differences are readily identifiable for
discussion directed toward conflict resolution.
viii
-------
The task of hazard identification is unique to each agent that might be considered. The
framework suggested is intended to contain essential features common to hazard identification
of any agent, but also to be sufficiently flexible to accommodate the wide diversity of situations
that may arise without excluding useful information. It is anticipated, however, that experience
with actual applications will indicate areas in which improvements can be made. In this sense
the procedure suggested should be viewed as a prototype. Perhaps the critical test will pertain
to the structural capability of the framework to accommodate changes and revisions that may be
helpful, particularly to correct any oversights in its development.
ix
-------
EXECUTIVE SUMMARY
This report is directed at improving health risk assessment, and deals exclusively with the
hazard identification step of risk analysis. It does not, however, begin with the guidelines of
EPA or another organization and attempt to make refinements that may result in reduced
uncertainty. Instead, it begins at a more fundamental level, first considering principles of
rationality by which to formulate judgments about carcinogenicity from the available information
base (the topic of Part I), then addressing what types and sources of information should be
assimilated into the informational base and how they may be systematically assimilated and
integrated into the formation of a judgment about carcinogenicity (the subject of Part II). The
term "carcinogenicity" is expanded from its usual meaning to distinguish within a taxonomy of
modes of action, such as initiator, promoter, risk modifier, etc. Additionally, carcinogenicity is
not considered to be a property of a chemical; it requires a reference context to be meaningful,
e.g., Chemical A is a carcinogen (in some particular sense from among the taxonomy of choices)
in Sprague-Dawley rats exposed for life at exposure concentration X. This is an important
concept since a chemical may be carcinogenic in one context but not another, and some of the
difficult questions to be addressed in hazard identification concern the inter-context
extrapolation of evidence, e.g., extrapolation across species or across exposure levels.
To expound further, the two parts of this report are conceptually distinct, yet practically
interrelated. Part I contains a review of philosophical ideas that have some bearing on the
practical matter of conducting and defending a risk analysis. In particular, discussion is directed
toward the issue of rationality with the philosophical literature review focused on a single
central question: Under what conditions, and in what sense, can an analyst assert that the
claims made in a risk analysis are rational? The significance of completeness of information and
1
-------
coherence of evaluation notwithstanding, the principle of rationality is considered fundamental to
provide an informed basis for identifying needed sources of information, processing and
evaluating it for credibility and relevance, and then making rational judgments. The authors
make no claim to having resolved philosophical questions related to the concept of rationality.
The review provided in Part I contains, instead, a summary of this topic from the primary
literature of philosophy, risk analysis, science, etc.
The discussion in Part I also forms the basis for the development of relevance strategies
for assessing evidence in the seven-step framework for performing hazard identification in Part
II. Relevance strategies are the bases by which observational information (such as
epidemiologic data, or results from chronic animal exposure, in vitro tests, or other sources) are
warranted as supporting evidence of carcinogenicity. For example, direct empirical observation
(I "saw" it) is a relevance strategy that some people may apply to human data demonstrating
that cancer incidence increased with exposure to the agent of interest. Empirical correlation
(Conditions correlate with situations where the agent was carcinogenic) and theory-based
inference (The outcome is consistent with theories of cancer mechanisms) are further examples
of relevance strategies. The intrinsic value of various relevance strategies for warranting claims
of carcinogenicity may differ between individuals within a rational framework for decision
making, thus creating one of the needs for discourse between decision makers.
Human judgment, however, and consequently human discourse to resolve conflicts and
work toward an agreement, are considered necessary to hazard identification. To assign a
decision rule or other form of computational or analytical scheme to avoid human decision
making and potential conflict in resolving differences would impose an artificial and limited
framework. It is not possible to foresee all possible contingencies and exigencies in advance,
2
-------
and even if it were, human decision making and discourse would still be required to formulate
the decision rule.
In general, the information available for consideration in forming individual judgments in
the hazard identification step may be widely heterogenous and difficult to assess without some
structure for guidance. The objective of Part II is to aid this process via a seven-step framework
that may be followed by an individual decision maker. An individual assessment of available
information guided by a common framework of decision making based on principles of
rationality, completeness, and coherence (as discussed in Part I), will lead to a logical conclusion
supportable by the individual's own perspective, opinion, and experience, as well as providing a
means for comparing the basis of conclusions drawn by different individuals. Assumptions and
"judgment calls" are made apparent, indicating areas and degrees of support/uncertainty. This
approach should (1) aid in the systematic evaluation of all sources of information that contribute
to hazard identification and the overall strength-of-evidence warranted; (2) identify sources of
divergence resulting from different perspectives and opinions of individuals, thus contributing to
conflict resolution; and (3) help to identify areas for research and assess their potential impact
on decision outcome(s) or their level of confidence. The overall objective this project is to
develop a logical assemblage of the diverse factors contributing to the decision making process
to accomplish these three goals.
3
-------
FOREWORD
"In the absence of direct human evidence of carcinogenicity, the conclusion that an agent
is liable to cause cancer in man is a matter of judgement...No hard and fast criteria can
be laid down that will automatically lead to an appropriate conclusion in all
circumstances. Only one rule is absolute: that all the available evidence must always be
taken into account." R. Doll (IARC Sci. Pub. No. 65, 1985)
Health risk assessment typically consists of four distinct steps, including (1) Hazard
Identification, (2) Dose-Response Assessment, (3) Exposure Assessment, and (4) Risk
Characterization. Of interest in this report is the task associated with the first step, hazard
identification, wherein the objective is to determine "whether a particular chemical is or is not
causally linked with a particular health effect" (NRC, 1983). With respect to cancer, the
particular health effect of interest in this report, the EPA guidelines define hazard identification
as "a qualitative risk assessment, dealing with the process of determining whether exposure to an
agent has the potential to increase the incidence of cancer...The hazard identification component
qualitatively answers the question of how likely an agent is to be a human carcinogen" (U.S.
EPA, 1986). EPA's approach is to make a judgment based on the weight-of-evidence and to
assign the chemical a group classification accordingly. The "evidence" used in the "weight-of-
evidence" for hazard identification should include a review of the following information to the
extent that it is available: physical-chemical properties and routes and patterns of exposure,
structure-activity relationships, metabolic and pharmacokinetic properties, toxicologic effects,
short-term tests, and long-term animal studies.
The EPA guidelines also note that there is a need for new methodology not addressed at
that time, e.g., the characterization of uncertainty. The present report is part of the agency's
subsequent EPA research effort directed at reducing uncertainty in risk analysis, which is
currently referred to as research to improve risk assessment. Broadly, there are three sources of
4
-------
uncertainty intrinsic to risk assessment in general, due to (1) limitations of observational
evidence, (2) incomplete knowledge on which to assess the significance of the evidence with
respect to predictions of risk, and (3) lack of rigorous development of the process by which
current knowledge is applied to available observational evidence to form a conclusion. This
report is concerned with the third category of uncertainty, as well as the manner in which the
first two categories are reflected in the process of justifying a conclusion of carcinogenicity.
The outcome of the hazard identification step for an agent depends on the evidence
available to decision makers and on their means or criteria for assessment of the evidence in
forming a judgment regarding carcinogenicity of the agent. It is somewhat analogous to a
courtroom trial in which the defendant is the chemical agent charged with causing cancer and
the jurors are the decision makers who hear the evidence. The jurors form individual judgments
initially and then a collective judgment following discourse. If every juror were expected to
weigh the merits of the evidence identically, a single juror would suffice. That, of course, is not
the case. Similarly, decision makers may evaluate the same evidence differently. For example,
the relative significance of the type of information, such as epidemiologic data, long term animal
studies, current theories of carcinogenesis, biological plausibility, physical and chemical
properties, and other categories of evidence, may differ between decision makers independent of
the strength of the evidence in each category. The axiom that reasonable men may disagree
applies to jurors and to decision makers in risk analysis, as elsewhere.
We would replace "reasonable" with "rational" as the objective for decision makers, i.e.,
the means or criteria for assessment of the evidence in forming a judgment regarding
carcinogenicity should be rational. While this objective may appear trivial (After all, who wants
decisions that are not rational?), the concept of rationality itself is not at all trivial and needs to
be understood if rationality is to be accepted as a guiding principle for judging evidence of
5
-------
carcinogenicity. This topic is central to Part I of this report entitled Risk, Regulatory Science,
Rationality, and Societal Values, which contains a review of philosophical ideas and principles
likely to have some bearing on the practical matter of conducting and defending a risk analysis.
In particular, the material in Part I focuses on the question: Under what conditions, and in
what sense, can an analyst assert that the claims made in a risk analysis are rational?
Within this framework of rationality, however, there is a great deal of disagreement as to
how the general features should be reflected in specific judgments made by individuals or
groups. This disagreement arises from philosophical differences concerning the nature of
evidence, how evidence is related to observations and experiments, when evidence is relevant to
a specific line of reasoning, when evidence is sufficiently strong to justify a claim, and so on. No
attempt is made in this report to resolve these issues, which probably are not resolvable except
in the sense of reaching a societal consensus. It is important to bear in mind that selection of a
particular view on rationality is as much a matter of human values as of logic, epistemology and
procedural rules. These values necessarily enter the discussion since a view on rationality is also
a view on human nature and on the form reasoning must take if a belief or claim is to be
thought of as having arisen from a process judged worthy of human decisions. In the language
of Section 4.1, rationality is both a descriptive and normative goal. The first chapter describes
competing conceptions of rationality, and the second chapter formalizes these into judgments
and decisions to be made by the analyst(s). The judgments remain an important task of the
analyst since the present authors have chosen to avoid imposing their own normative judgments.
If the principles of rationality guide the process of hazard identification, then the
informational base available fuels it. As noted in the EPA guidelines and evident in the several
categories of evidence listed above, a wide range of evidence is potentially available to the
decision maker. The goals of accuracy and minimal uncertainty are both consistent with the
6
-------
tenet that the best judgment is the most informed judgment. If less than complete information
is considered, then uncertainty can be further reduced by expanding the base of information.
With respect to accuracy of decisions, incomplete information has the potential to mislead or to
bias a decision. Identification of the information useful for hazard identification, a framework
to aid in its assessment and integration for decision making, and consideration of assumptions
required for extrapolation across contexts of observations (such as between species or dose
levels) are addressed in a seven-step guide in Part II of this report, entitled Assessing Evidential
Support for Claims of Carcinogenicity: Essential Elements and a Framework for Application. This
part of the report also addresses the question of what is meant by "carcinogenicity". The
concept that an agent is carcinogenic if exposure to it increases the incidence rate of cancer
does not cover all possibilities of interest. A taxonomy of carcinogenicity is defined for use
instead.
How should the report be read? Here, the answer depends upon the background and
interests of the reader. As described above, the report is divided into two parts: the first
containing a philosophical discussion concerning rationality and the second consisting of
practical matters relating the philosophical principles to the conduct of the hazard identification
stage of a risk analysis. Readers already familiar with the literature on rationality,
epistemology and philosophy of science might skip Part I entirely, moving directly to Part II for
the discussion of applications. Part I contains, however, a number of examples providing
insights into how the general philosophical ideas are related to the specific field of risk analysis.
Even readers familiar with the literature, therefore, might find it useful to at least scan Part I
for the overall flow of ideas and to read the examples provided there.
Those with little background in philosophy are left with two options. It is possible to
read Part II on its own since it contains (at least implicitly) consideration of many of the ideas
7
-------
discussed in Part I. Having mastered the quasi-formal framework of this second chapter, it then
would be useful to return to Part I to obtain a more detailed understanding of the various
philosophical positions that might be adopted in making the specific judgments called for in the
second part. An alternative approach (and probably the best approach) would be to read
Chapters 3 and 4 of Part I, which provide a broad discussion of rationality, followed by the
entirety of Part II, and finally by the more detailed discussion in Chapter 5 of Part I. In any
event, a firm grasp of the material in Part I (and of the primary literature reviewed there) is
essential to making full use of the framework in Part II. The need for such a grasp arises from
several considerations:
(1) While the quasi-formal framework of Part II contains a distinct set of procedures,
these explicit procedures call repeatedly for judgments related to the rationality
of collecting, analyzing, judging and employing evidence within lines of reasoning.
There are competing conceptions of how these explicit judgments should be
made, and these conceptions are reviewed in Part I.
(2) Part I contains material of potential importance in Part II, but not yet formally
incorporated into the framework of analysis reported there. The present
document is the first product of the authors' broader (and more long-term)
research program to link the subject matter of the two chapters. A review of
Part I may suggest to the reader new formal considerations to be introduced into
the analytic framework of Part II. Similarly, review of Part II may suggest
judgments requiring philosophical principles left out of the discussion in Part I.
To stimulate the interchange of ideas contained in the separate parts, the authors
have included a table (see Table 0) listing the central ideas in Part I and the
points at which they become important considerations within the framework
introduced in Part II. It is hoped that others will make this table more complete
through complimentary research.
(3) In an important sense, the explicit judgments called for in Part II must take place
within a context of rational discourse. A central thesis of the entire report is that
rationality is more than a formal rule-based procedure for making decisions. It
is, instead, a "frame of mind" in which the analyst attempts to determine the
quality of philosophical positions concerning beliefs, claims, evidence and
evidential reason. To be rational is to carry on a discourse concerning the many
potential views on what it means to have well-founded beliefs or claims (this
distinction between belief and claim is raised in Part II). Part I contains material
essential to such a discourse, even where there is no obvious judgment called for
in Part II. The authors have made a strong attempt to show explicitly how the
judgments of Part II are related to the principles of rationality set forth in Part I
8
-------
(See Table 0 mentioned above). Still, there will be points at which the links are
not explicit, calling upon the reader to provide creative input into the framework
of analysis. This creative input requires a constant and well-reasoned discourse
on the philosophical principles drawn from Chapter 1 and , ultimately, from the
primary literature.
The overall structure of the report, and of the research program, then is one of:
(a) Developing conceptions of rationality (Chapters 3 and 4 of Part I);
(b) Developing conceptions of evidential reason necessary within all conceptions of
rationality (Chapter of Part I);
(c) Formalizing results of (a) and (b) into explicit principles forming an axiomatic
base (or set of potential bases) of evidential reasoning (Table 0);
(d) Elucidating the framework of rational discourse concerning claims of hazard
identification (Chapters 1 and 5 of Part II);
(e) Developing the bodies of evidence potentially useful in providing the judgments
required by hazard identification, and
(f) Making explicit (to the degree possible) the link between (a, b) and (d, e)
(Table 0).
9
-------
PART ONE
RISK, REGULATORY SCIENCE, RATIONALITY
AND SOCIETAL VALUES
1. BACKGROUND AND TASK OBJECTIVE
The assessment of the risk of cancer and other adverse health effects associated with
exposure to toxic substances is a subject of much debate, posing intricate relations between
science and policy. That sentence describes the setting in 1983 when a committee of the
National Research Council (NRC) was mandated by a congressional directive to address some
fundamental questions regarding the feasibility and merits of alternative modes of conducting
risk assessment among all federal regulatory agencies.59 In particular, the committee was asked
to examine whether altered institutional arrangements or procedures can improve regulatory
performance. This request was largely motivated by criticisms of risk assessment that ranged
broadly from details of the process to administrative management to statutory authority. The
committee recommended that regulatory agencies take steps to establish and maintain a clear
conceptual distinction between assessment of risk and consideration of risk management
alternatives; that is, the scientific findings and policy judgments embodied in risk assessments
should be explicitly distinguished from the political, economic, and technical considerations that
influence the design and choice of regulatory strategies. The committee also recommended that
(1) uniform inference guidelines be developed for the use of federal regulatory agencies in the
risk assessment process that would be evaluated regularly for their usefulness and revised as
needed, (2) the evolving scientific basis of risk assessment should be critically assessed, and (3)
explicit underlying assumptions and policy ramifications of the inference options in each
component of the risk assessment process should be made explicit. The NRC committee
10
-------
differentiated between four components of risk assessment: hazard identification, dose-response
assessment, exposure assessment, and risk characterization.
The EPA had already formulated interim guidelines for risk assessment at the time of
the NRC committee's report in 1983. The Agency's current risk assessment guidelines adopt the
recommended distinction between risk management and risk assessment, and the categorization
of the latter into four component elements.60 The first of these, hazard identification, consists of
two elements (long-term animal studies and human studies) that provide the primary data and
five elements (physical-chemical properties and routes and patterns of exposure, structure
activity relationships, metabolic and pharmacokinetic properties, toxicologic effects, and short-
term tests) that may contribute useful information to the qualitative hazard evaluation as
available. Aside from at least one central data source from an animal or human study, the
information available on the hazard identification categories will vary from chemical-to-chemical
depending on the available research base. Consequently, there is no prescribed way of
integrating the available information in all cases, either in the evaluation of hazard or the
designation of weight-of-evidence classification.
Chemical agents are assessed on a case-by-case basis. Knowledge of metabolic and
pharmacokinetic properties, toxicologic effects, and the other categories of hazard identification
help to fill informational gaps otherwise bridged (sometimes implicitly) by assumptions and thus
reduce uncertainty. Essential to the process of hazard identification and subsequent dose-
response assessment when warranted, however, is not only completeness of information but
coherence and rational evaluation as well. An individual assessment of available information
guided by a common framework of decision making based on these factors will lead to logical
conclusion(s) supportable by the individual's own perspective, opinions, and experience, as well
as providing a means for comparing the basis of conclusions drawn by different individuals.
11
-------
Assumptions and "judgment calls" are made apparent, indicating areas and degrees of
support/uncertainty (to be replaced by the more definitive term "warrant" in the text). This
approach should (1) aid in the systematic evaluation of all sources of information that contribute
to hazard identification and the overall strength-of-evidence warranted; (2) identify sources of
divergence resulting from different perspectives and opinions of individuals, thus contributing to
conflict resolution; (3) help to identify areas for research and assess their potential impact on
decision outcome(s) or their level of confidence. The overall objective of this project is to
develop a logical assemblage of the diverse factors contributing to the decision making process
to accomplish these three goals.
2. INTRODUCTION
Identifying and assessing sources of uncertainty (levels of support) is essential to
evaluation of risk analysis for decision making. Numerous authors and workshops have raised
issues relevant to this topic, but there remains a need to knit together: (1) the diverse
components of information that may be useful, but variable in availability and quality from
chemical-to-chemical; (2) the process of inference from the available knowledge base, including
the role of primary data (animal and epidemiological data) and supporting informational
sources; and (3) human knowledge and judgment. Identification of sources of uncertainty is
necessary but not sufficient. The knitted components described above must provide a fabric of
rationality, based on completeness and coherence of assembling all available evidence into a risk
analysis.
What is rationality? Can scientists of different disciplines, administrators, decision
makers, industrialists potentially affected economically by regulatory decisions, health advisory
and other special interest groups, persons whose thoughts and perceptions may be molded by
12
-------
different backgrounds and interests, share a common conception of rationality? Are there
alternative conceptions of rationality? If so, are they compatible within the risk analysis
context? Is there a flexible conception of rationality that includes the others as specific cases?
These questions are fundamental to our overall objective.
Five fundamental rational bases for decision making need to be addressed before a
complete methodology for hazard identification can be created. These questions are:
(1) What does it mean to claim that a risk analysis is rational?
(2) Where would one look in a risk analysis to determine if a claim to rationality was
valid?
(3) What are the various views on rationality which might be adopted by competing
camps in a debate on risk, and how do these views fall out as specific cases of a
more general understanding of rationality?
(4) What kinds of information might appear in a risk analysis and what role does this
information play?
(5) What kinds of judgments must the analyst make concerning this information?
Part I of this report consists of several subtopics. Decisions are made by humans and
thus reflect human judgment. What is considered as a rational decision depends on many
factors, such as objectives (ends), alternatives (means), and beliefs (which include beliefs about
nature and about human values). Human judgment, however, with whatever background factors
and perspectives influence it, is the means by which decisions are made. No formula for
rational behavior is possible that could circumvent this element of human judgment.
Furthermore, rationality is a human conception with many variations, as apparent from the long
history of philosophical inquiry into that topic. This history of thought on rationality will serve
13
-------
r
to provide enlightenment on the breadth and depth of conception of rationality from which
decision making for risk analysis may be approached.
"Rationality" is not something that can be discovered and labeled, like a chemical
element. It must, instead, be defined consistent with principles of internal coherence and of
coherence with other words in the vocabulary. Consequently, as occurs in any attempt to discuss
any normative word (concept), one must sometimes tolerate circularity as experienced when one
word is described in terms of a second, leading to a third, etc., until the circle is completed to
referencing the original word. To illustrate, suppose we ask what minimal characteristics should
be included in a conception of rationality to which most people would agree, e.g., logical,
reasonable, making "common sense," all of which are well understood terms in frequent use.
One is immediately led to further questions of refinement, e.g., logical in what sense? Does it
follow deductively from some axiomatic truths? Does it attain a specific goal or objective? Can
two different opinions (decisions) based on the same evidence both be logical? Similarly, one
could replace "logical" with "reasonable" or "makes sense," or "rational" and the same outcome
would pertain.
A useful way for the reader to approach material in this report, perhaps ironically, is
simply to think of rationality initially in terms that are comfortable and meaningful, e.g.,
"logical," "reasonable," etc. This provides a useful starting point from which to expand and
perhaps modify one's perspective. The next higher stepping stone into this subject matter is the
"Seven Desiderata of Rationality" of Bunge, given in the next chapter under "constructing a
complete vision of rationality." Bunge describes seven factors that any claim to rationality must
consider, and these factors provide a useful framework for subsequent discussions in this report.
The next section (3.2) considers what human skills enhance rational decision making. The
intimate association of rationality and human judgment is also fundamental to the topic of the
14
-------
next chapter. To see this, one need only observe that the "meaning" of rationality is based on
human judgment, as is the assessment of the rationality of an action or decision. Twenty areas
in which an analyst or policy maker must make distinct judgments that reflect human values are
described. Human judgments may differ on assessment of an issue, or on the criteria for
assessment, making the relevance of the yardstick subjective. Just as people can often benefit
from the opinions of others, risk analysts and decision makers may benefit from the expressions
of diverse rationalities. Ten broad classes of rationality that may underlay societal debates, i.e.,
may produce alternative but rational differences of opinion (judgment), are given to conclude
Chapter 3.
Chapter 4 addresses some central issues regarding alternative perspectives on rationality.
Whether rationality is viewed as a descriptive or a normative (prescriptive) theory of human
behavior may depend on one's discipline (Section 4.1.) A risk analysis must be structured to
yield predictions related closely to an existing means for reaching a clear set of ends. This issue
is discussed under the rationality of beliefs, means, and ends, the subject of Section 4.4. The
most complete vision of rationality incorporates beliefs, means, and ends, judged both
intrinsically and through their relationship, which is significant to the risk assessment process.
These topics are addressed in Sections 4.3 and 4.4, and summarized in relation to other chapter
topics in Section 4.5.
We turn to the history of philosophy (particularly the history of science and analysis) in
Chapter 5 to discuss the manner and degree to which beliefs, means, and ends impact on each
other. Differences in perspective on the possibility of a logic of science lead to competing views
on the extent to which a risk analysis can be driven entirely by the internal goals of science.
Classical, empiricist, and rationalist rationality, logical positivism, rational skepticism, and
probabilistic rationality are compared and contrasted, and illustrated in the context of risk
15
-------
analysis. Several issues related to this topic also are addressed to complete the discussion of
rationality and beliefs as a basis for risk analysis and decision making.
Throughout Part I of this report, the example of a risk analysis for a carcinogen is
employed. In particular, the chosen example is radon in drinking water. This example was
selected not because it necessarily is more enlightening, but because one of the investigators
(Crawford-Brown) has been involved for several years in efforts by the U.S. EPA Office of
Drinking Water7, the Science Advisory Board, and the American Water Works Association to
reach agreement on a regulatory standard. The example is not developed in full detail, since
that detail is the subject of future research. It is intended, however, that the present report
summarize all of the considerations which will form the basis for that detail. As mentioned
earlier, reviewers are asked to read the present report for inconsistencies, missing principles of
reason, poorly or incorrectly defined terms, and (of most importance) a lack of clear relevance
to the specific issues which invariably arise in pushing a risk analysis through from conception to
justification.
Some of the sections of this part of the report contain fairly detailed analyses of
philosophical concepts. Readers may find it useful to skip those sections initially, focusing on
the summary sections to gain a broad overview of the kinds of issues raised and potential
solutions developed. A second reading might then be used to gain a more detailed
understanding of how the summary information arose.
Rationality is addressed in this project on uncertainty in risk analysis to provide an
informed basis for identifying needed sources of information, processing and evaluating it for
credibility and relevance, and then making rational judgments. What information is needed for
risk analysis? Which sources are preferable? Which are of greater significance for drawing
conclusions? What assumptions are implied by "default" guidelines when information is
16
-------
unavailable? How should expert opinions be weighed vis-a-vis experimental data? There are
many facets to the topic of uncertainty in risk analysis. Consequently, an approach that is
systematic and developmental, i.e., that can be expanded, revised, and refined as needed, may be
the most tractable approach. The basic components of such an approach are described briefly in
this part of the report.
3. CONSTRUCTING A COMPLETE VISION OF RATIONALITY
Decisions are made by people, some of whom are more skilled at making rational
decisions than others. What are these skills, i.e., what characteristics are found in a good
decision maker? This topic is treated in Section 3.2, following an introduction to the topic of
rationality via Bunge's "Seven Desiderata of Rationality." The discussion of Bunge's principles
in Section 3.1 is a useful departure into the topic because he describes characteristics that any
rational system must have, using common terminology. Having described the features of
rational decision making and the personal skills that enhance its implementation, twenty subject
areas are given in which the analyst/decision maker is called-on to make judgments that reflect
human values (Section 3.3). Disputes and disagreements will still arise in risk analysis, however,
which leads us to consider alternative classifications (or types) of rationality that may be
encountered in society, e.g., supporting adversarial positions (Section 3.4).
3.1. Seven Features Essential to a Rational System
Bunge45 describes seven characteristics essential to rationality which he calls the "seven
Desiderata of Rationality." they are included here as a framework for subsequent discussion.
1. Conceptual-a properly rational system minimizes fuzziness, vagueness and/or
ambiguity in its terms. The importance of this feature was highlighted by
Wittgenstein46, who felt that logic could not begin until language itself was placed
onto a well defined basis. All people using a term must mean the same thing,
17
-------
and this meaning must be chosen to reflect accurately the objective features of
the world. In his early work, Wittgenstein argued that the proper definition of a
word could be obtained through careful analysis of objective reality. His later
work47, however, emphasized the role of social agreement in choosing the
meaning of words. Regardless of the viewpoint chosen or the origin of a word's
meaning, rationality involves an attempt for conceptual clarity which allows a
comparison of the beliefs of individuals through a shared language.
2. Logical-a properly rational system strives for consistency and a lack of
contradiction. Beliefs (means, ends) are, to the degree possible, obtained through
the application of well defined rules of reason. This focus on logic is intended to
ensure that beliefs (means, ends) can be shown to be reasoned and not the result
of judgments which might be made as matters of convenience. As Nathanson48
writes, "Rationality, then, involves a striving to be objective, and objectivity
involves the attempt to discount those features of ourselves or our situation that
might involve our judgment but that are not relevant as evidence." The rules of
logic supposedly open choices to public discourse.
3. Methodological-a properly rational system produces a habit of questioning
beliefs (such as in Popper) and/or verifying beliefs (as in the Vienna Circle
writers such as Carnap49). Without the presence of these methodological rules,
people would have no shared approach to rooting out and changing false beliefs.
4. Epistemological~a properly rational system accepts evidence only .when that
evidence satisfies criteria of quality (such as empirical accuracy, relevance, etc.).
The epistemology instills in the rational person a love for particular kinds of
evidence (such as the logical positivist's desire for observation). Statements
which are pure conjectures, unsupported by "appropriate" evidence, are avoided
in the process of reasoning.
5. Ontological~a properly rational system uses only terms (such as force, cell, etc.)
which are believed to be descriptions of the world. Since Bunge believes strongly
in science, his own ontological criterion is that all terms used in reasoning must
be those used by science and technology. Others may disagree with him, but all
rational people will apply reason only to terms which describe entities and
relationships believed to exist in the world.
6. Valuational~a properly rational system strives for goals which have been
determined to be worth attaining. These goals should, moreover, be the highest
goals.
7. Practical~a properly rational system adopts means for action likely to yield the
desired ends. People may disagree as to what is to be meant by likely. In some
approaches, all of the available means are described in detail and the evidence
for their ability to yield the desired ends weighed. The means most likely to
reach those ends from amongst the competitors is selected. In other approaches,
only means which satisfy some minimal level of crafting form a rational basis for
action. Beliefs that fail to satisfy this minimal level are rejected even if they are
18
-------
the best amongst the competitors. In both approaches, however, the key to
practical rationality lies in demonstrating a clear link between means and ends.
Bunge moves from these seven desiderata to a distinction between various levels of
rationality. People who rely on only a subset of the seven are said to be semirational, a wording
similar to Simon's idea of bounded rationality.50 People who rely on all seven are said to be
fully rational. Anyone who rejects all of the criteria is said to be non-rational. Only a person
who accepts the criteria but goes deliberately against the outcome of rational analysis is said to
be irrational.
32. Skills Required by the Rational Risk Analyst
What skills enhance rationality? Rescher51 has identified a need for five distinct faculties
which must be brought to bear in rational discourse or action. There is a need for imagination,
since the rational person strives to look for alternative beliefs, means, and ends to be judged. A
firm grasp of information processing is required in order to form beliefs. Skills at evaluation of
alternatives (beliefs, means, ends) give the capacity to weigh the merits of those alternatives.
The rational person must also be able to select an option through informed choice. To some
views of rationality, such as the classical theory, selection is dictated by the methodology of
rationality. In other views, such as that by Berkson52, rational inquiry is a guide without
determining the final choice between options. In Berkson's words: "The basic idea is that the
individual, not the method, makes the choice; but the individual should be influenced in his
decision making process by rational argument." Finally, a rational person requires the skill of
agency, the capacity to actually carry out a choice.
33. Value Judgments in Rational Action
19
-------
The preceding discussion suggests twenty areas in which an analyst or policy maker must
make distinct judgments which reflect human values. Some of these areas will be clear only
after a reading of Chapter 5, which includes a discussion of the failure of philosophy to locate a
complete logic of science and evidential reason. These value judgments are listed here in no
particular order. Each judgment will be subject to dispute and complete rationality requires that
each be discussed in a setting which includes all affected parties, as required by game rationality
(see Section 3.4). At the highest level of judgment, it might also be asked how the specific
conceptions of rationality listed in Section 3.4 should be valued. The example of regulating
radon in drinking water is used throughout the following discussion.
(1) Relevance of evidence to the stated problem-evidence may be assembled from a
potentially infinite number of areas of human study. A radon risk analyst must
judge the relevance of data from experiments on the breakage of DNA by
radiation, from the transformation of cells in vitro, from small animal studies and
from human epidemiological studies. There also are experiences suggesting the
general reliability of the scientific community at transmitting results to various
groups in society. All of this evidence must be judged for its relevance to
attempts at predicting the effects of radon in the home.
(2) Epistemological status of evidence and beliefs-some people may look for direct
empirical evidence that removing radon from water has resulted in lowered risks
in practice, i.e., in historical cases. Others will be satisfied by deductions from
theories which predict the desired effects in the absence of any human experience
with mitigating radon. Still others may be satisfied by extrapolation from findings
at high levels of human exposure, using either theory-free curve fits to the
existing data or explicit theories about the role of radon in producing or reducing
-------
cancer. A judgment must be made as to the weight assigned to these varying
kinds of evidence and how that weight justifies particular courses of action such
as public warnings, expenditures for mitigation or the imposition of regulations.
Level of causality-all phenomena such as health have many levels at which
causality may be assigned. In the case of radon, the cause of cancer may be said
to be the deposition of radiation energy in cells. It may be said to be the
ingestion or inhalation of radon. It may be said to be the presence of radon in
the water. It may also be said to be the political system which allows such levels,
or the economic system which allows people to become poor, undernourished
and, hence, susceptible to the effects of radiation. Each of these levels of
causality suggests a different strategy for mitigating risks. A judgment must be
made of the appropriate level at which causality will be analyzed.
Clarity of terms required-some ideas, such as long-term frequency, are highly
technical and well defined. Others, such as human confidence, are less well
defined but potentially incorporate a wider range of considerations. A judgment
must be made of the clarity required before a term or idea qualifies as a valid
basis for analysis.
Degree of bounding-no risk analysis can include all of the factors related to even
a single end. If radon is removed from drinking water by aeration, it will enter
the air. If it is removed by absorption onto charcoal, it will enter the soil of a
landfill. From there, it may migrate to other parts of the environment. Given
the complexity of environmental systems, only a select sample of the potential
pathways of exposure can be considered. The model of risk must, therefore, be
-------
bounded. A judgment must be made as to where the boundaries are to be
imposed in light of the stated goal to protect the public health.
Specification of ends~the EPA desires to protect the public health. This end
may, however, include minimization of the average chance of fatal cancer,
minimization of effects in children, minimization of years of life lost, etc. In
addition, there may be the desired ends of lowest cost, degree of democratization
of choices, etc. A judgment must be made of the most significant ends and of
how the process of analysis will reflect those ends.
Dealing with uncertainty and ignorance-a problem may be viewed as a choice
between existing options. It might be argued that the risk analyst must choose
between existing theories for predicting the effects of radon on health. The
analyst then is in a state of uncertainty. It might also be argued that none of the
existing theories is well established, requiring an admission of ignorance. A
judgment must be made as to whether an analysis of uncertainties is required,
how that analysis should affect decisions, and whether the possibility of ignorance
should be factored into the decision in addition to uncertainty.
The will to believe specific predictions~as James has pointed out57, most
questions cannot be answered with certainty. If the question is not of great
importance, it may be acceptable to withhold belief until the evidence improves.
If the question is of great importance, however, it may be necessary to adopt a
belief despite great uncertainty and ignorance. This new belief might concern a
particular prediction of the effect of radon, or it might concern the more vague
belief that the effect lies within certain bounds (such as the belief that the chance
of cancer associated with a concentration of X is less than or equal to Y). A
-------
judgment must be made as to whether belief is warranted and what kind of belief
is warranted in light of the existing state of evidence.
Sufficiency of logical necessity-no beliefs follow necessarily from a firm
foundation of truths (see Chapter 5). And yet some beliefs are more firmly
rooted in observation and logic than others. There is a natural desire to see
beliefs reduced to logic, since this provides the easiest route to public scrutiny of
the link between evidence and belief, and avoids the need for judgment. But
logic ignores skills of crafting and intuitive insights gained from experience
(typically called engineering judgment). Toulmin58 conceives of the process of
reasoning as one of offering warrants for belief. Warrants do not guarantee the
truth of a conclusion and, therefore, do not lead to a strict logic. Yet they do
count as a kind of support for any conclusions. A judgment must be made of the
degree to which predictions of the effects of removing radon must follow the
rules of logic and the degree to which it may require the less stringent idea of
warranting. A judgment also must be made as to whether a particular warrant is
sufficiently weak to require strong public scrutiny or sufficiently strong to justify
acceptance without stringent requirements for debate.
Location of rationality--a judgment must be made as to whether an action is
rational if either the beliefs, means, ends, or the relationship between these parts
may be left unexamined in a claim to rationality (see Section 4.2).
Role of conceptual and empirical success-as mentioned in the discussion on
Laudan36 (see Chapter 5), theories may be assumed both to yield predictions of
the outcomes of experiments and to provide explanations which unite concepts
within a field of science. The former is a requirement of foundational truth and
-------
the latter is a requirement of coherence. In the case of theories concerning the
effects of radiation, some give explanations referring to a wide range of biological
factors (DNA damage, cell replication, etc.) while other refer only to physically
unspecified "thresholds" for cancer. Even if both predict the existing human data
equally, a judgment must be made of the importance of the former's conceptual
sophistication, its ability to unite biochemical, in vitro transformation, and
epidemiological data.
(12) Selection of humans-if writers such as Longino37 and Polanyi40 (see Chapter 5)
are correct, all rational activities, including science, require distinct judgments
and tacit skills. A question arises, therefore, as to how individuals possessing
such skills are to be located. Two features of the selected humans seem to be
essential. The first is an adequate base of experience from which skills may
develop. In the case of radon, this may require selection of scientists in the field
of radiation biophysics. The second is an adequate degree of reflection on the
quality of that experience and the resulting skills. This may require selection of
people familiar with rationality or legal argument. In addition, game rationality
requires that a range of humans with (perhaps unstated) differences in ends and
skills be incorporated into the process of decision.
(13) Completeness of the "hard look"—with limited time and resources to devote to an
activity such as the analysis and mitigation of radon risk, it may not be prudent to
suspend belief awaiting collection of all available data. At some point, it must be
judged that both the pool of data and the extent of critical reflection is adequate
for decisions. This requires a judgment of when an analysis constitutes a
sufficiently "hard look" to act as a rational basis for a particular course of action.
24
-------
If the course of action is publication of a pamphlet warning of radon risks, the
judgment of sufficiency may be different than would be true for setting regulatory
limits.
(14) "Knowing that" versus "knowing how"~two judgments will be required in this
regard (see Chapter 5). It is necessary to determine when a claim to
understanding, unsupported by a demonstration of the efficacy of a means in
practice, is an adequate base for rational action (such as may result if predictions
of the effect of environmental levels of radon have not been tested through tests
of mitigation). It must also be determined when a claim to limited success in
practice may form the basis for rational action if an understanding of that success
is not available.
(15) Interlocking of beliefs, means and ends-in an ideal case unconstrained by
contextual rationality (see Section 3.4), all beliefs, means, and ends might be
examined. In practice, however, only beliefs relevant to acceptable means, or
means relevant to acceptable ends, will be considered. For example, while
changes in nutritional status might aid in lowering susceptibility to the effects of
radon, the means for achieving this might lie outside those allowed to the EPA.
Similarly, only cancer risks typically are considered in specifying ends. A
judgment is necessary concerning the degree to which these three areas should be
explored completely and in isolation from each other, at least prior to the time
for decision.
(16) Extent of coherence-as proposed by Quine44, rationality implies a coherence
between the separate beliefs within a system of belief. It is not clear, however,
whether all beliefs must cohere when reaching for specific ends. Should, for
25
-------
example, religious beliefs be allowed to require a well developed consideration of
the influences of sin on the risk from radon? Should a belief in benign nature
modify the analysis? Must a theory concerning the risk from radon include
explicit consideration of oncogenes?
(17) Ontology-one criterion of rationality was that the system of belief employ
entities and relationships (such as cells, transformations, etc.) which are judged to
exist in the world. Some concepts such as states of emotion, confidence,
psychological causes, etc., may fall below a predetermined ontological status.
There is a debate in the modeling of health effects over whether the stages of a
multistage model of carcinogenesis satisfy this criterion. A judgment must be
made of the factors which will be allowed in any explanatory and/or predictive
theory.
(18) Choice of fides-if Fideism (see Section 3.4) is chosen as the basis for rational
action, a judgment must be made as to which beliefs are no longer open to
rational debate. These might, for instance, include the belief that a cancer may
be identified without question. Other beliefs, such as that particular patterns of
scattered light indicated DNA damage, may not satisfy the judgment of sufficient
belief to constitute a fide. They will require additional support from more well
established bodies of evidence.
(19) Ambiguous evidence-for all beliefs, there will be evidence in favor of, against, or
neutral to, that belief. The ability of radiation to produce DNA breaks acts as
support for the belief that radiation is an initiator. The cytotoxicity of radiation
supports the belief that it may be a promoter. A judgment must be made as to
26
-------
how these various instances of partial confirmation and partial falsification should
be weighed into an overall assessment of belief.
(20) Concept of the weight of evidence~the degree of belief in light of evidence may
be given by a purely subjective statement of confidence, a statement from
sampling theory, a statement deduced from a Bayesian perspective, etc. This
weight may be numerical or verbal, and may or may not refer to psychological
states. A judgment must be made as to how the concepts of weight of evidence,
probability, confidence, etc., are to be interwoven and defined.
It should be clear that a large number of judgments potentially appear in any claim to
rationality, even if attention is focused on the rationality of belief. All of these, in fact, underlie
such claims even if the judgments are not made explicit. Any framework for rational analysis,
decision and action should, therefore, make the above judgments explicit and the subject of
debate. It is important to bear in mind that the particular values brought to the forefront in
making those judgments may depend upon the activity in which one is engaged. An ordering or
assessment of values applied to the rationality of scientific research may be inappropriate for
regulatory action, given the potential for different ends. No attempt has been made here to
specify the values which should be adopted. The intent, instead, has been to reveal their
existence and to place them within a framework of rational thought and action.
3.4. Broad Classes of Rationality Underlying Societal Debates
This section closes with an attempt to catalogue various types of rationality which might
be found in society. Any person or group might employ more than one type of rationality in
forming beliefs or in carrying out a decision or action. Still, it is useful to establish a typology to
aid in locating the fundamental assumptions adopted by that person or group when a claim to
27
-------
rationality is made. The following summaries, then, represent a rough sketch of the various
stances which might be adopted in the search for, or in construction of, a rational system. An
example using the case of radon is given at the end of this section.
1. Classical rationality-this is the belief that rationality essentially is equal to truth,
and truth is obtained through the use of logic (see Chapter 5). This logic is
applied to a set of premises about the world which are taken to be absolutely
certain. The classical camp might usefully be divided into two schools with
different epistemologies. The empiricist school (Locke, the logical positivists)
choose their firm premises or foundations from the results of observation. The
rationalists (such as Descartes) chose their firm premises from "clearly and
distinctly perceived" items of introspection. In both cases, only beliefs following
deductively from these foundations are allowed in the quest for rationality.
2. Process rationality-this is the belief that no statement or action will satisfy the
desires of classical rationality. All beliefs are subject to question since there are
no firm foundations. Rationality implies, instead, a process of constant
questioning and a search for more evidence. An example is Popper's idea of
critical discourse with repeated attempts to falsify a theory (see Chapter 5). It is
the process, and not the current state of match between evidence and belief,
which justifies a claim to rationality.
3. Fideism-a "fide" is a faith, something taken without complete support in
evidence. The Fideist's (such as Polanyi) admit that the classical ideal cannot be
met since there is no completely firm foundation for belief. They argue, however,
that some beliefs (particularly those of well-tested science) are sufficiently strong
to justify acceptance as a matter of faith. People will, of course, disagree as to
which beliefs satisfy this criterion of being a proper fide. Once the fide is chosen,
however, this approach is similar to the classical approach to rationality.
4. Probabilistic rationality-the foundational assumptions required by the classical
approach may be said to be somewhat less than firm. They may, however, be
given a degree of support through probabilistic ideas. These probabilities may be
said to be an objective property of the world or of our methods for studying the
world (as in long-term frequency approaches to sampling theory), or a measure of
our state of confidence. As with Fideism, rationality then is obtained in a
manner similar to the classical approach once the probabilities are assigned to
the probabilistic premises. To be rational means to act on the belief possessing
the highest probability of being true.
5. Limited rationality-even scientific models fail to predict the entire functioning of
the world. Causal factors are left out because they are not understood. Others
are left out for computational convenience. It is not possible, therefore, to reach
all of the possible ends through the use of such models. Still, a rational person
may reach a few well prescribed ends through the use of bounded reason. There
28
-------
will, of course, be disagreement as to what constitutes proper and acceptable
bounds on the ends, means, and beliefs appearing in a rational discussion.
6. Contextual rationality-complete rationality might require a very large amount of
time to study a problem, particularly if the problem is complex. Humans,
however, have competing ends which require attention. It is important, therefore,
to allocate time so that a single problem does not attract an inordinate amount of
attention. A rational person analyzes a problem only to the degree feasible in
light of other demands on time. The goal is to limit analysis of specific problems
so that the full range of problems may be approached in at least an approximate
manner.
7. Game rationality-the classical theory assumes that all of the ends and methods
for judging means or beliefs can be ranked from highest to lowest (or best to
worst). While this still may prove to be the case, agreement on the appropriate
ranking has not been found in society. As a result, any societal debate will be
characterized by different groups adopting a different ranking. The rational
person treats this situation like a game in which the competing views must be
balanced through a process of interaction. In this sense, the view bears some
resemblance to process rationality (#2).
8. Adaptive rationality-it is a common feature of human understanding that beliefs
change with new evidence. A properly rational person, therefore, looks for a
process (implying a kind of process rationality) which mimics the growth of
knowledge. An example of this approach is Hesse's "learning machine" model of
science53, or the Bayesian methodology of updating beliefs in light of new
evidence.54
9. Selective rationality-if, as Longino emphasizes37, human judgment is essential to
assigning relevance to evidence (see Chapter 5), it would be best to let those
judgments be made by the most qualified people. Selective rationality then
involves the identification of groups or individuals with the best skills at
judgment. This is by no means a simple matter, since it is not clear how those
skills are to be identified. The problem is particularly difficult when ends, means,
. and beliefs become inseparable, since no individual or group is likely to possess
skills in all three areas of judgment.
10. Posterior rationality-the classical theory, as well as modern decision theory,
assumes that the ends are known before rational analysis begins. The means to
reach those ends then are assembled, beliefs assigned to those means, and the
appropriate means selected. At times, however, the ends may not be known
prior to analysis. They may, instead, be a product of an analysis, such as when
scientific research discovers the importance of species diversity, which becomes a
new end. A rational process should, therefore, be capable of generating new
ends or of changing the direction of analysis once new ends are identified in the
process or reasoning.
29
-------
These various conceptions of rationality can be illustrated by considering how they might cause a
person to view a risk analysis for radon. It should be emphasized that the conceptions are not
necessarily mutually exclusive. There should be a healthy regard given to each conception, since
each provides valuable insights into a complete picture of rationality.
Assume that the EPA wishes to regulate radon in air or water, and that a claim to
rationality is desired. The classical rationalist will focus attention on the set of beliefs
concerning how radon is related to a given end (here the production of cancer). Only
predictions which must follow logically from observations (empiricism) or "clearly and distinctly
perceived" statements (rationalism) will count as being true. Only these beliefs, therefore, will
count as rational products of the risk analysis. All other beliefs, regardless of their status as
being partially confirmed, will be rejected as non-rational. While the classical approach certainly
stands as an ideal, it is a simple matter to find potential flaws in any prediction concerning
radon and, hence, the classical skeptic could argue that all risk analyses were non-rational.
The fideist would search for some principles which could be taken as reasonable in
beginning the process of reasoning. This analyst might argue that an observation of an excess
incidence in an exposed population was a reasonable basis for belief, even if (as Hume argues)
it does not certify that belief. The analyst might also argue that a linear dose-response function
has been reasonably well established. Such fides then could be used to analyze the existing data
in analytic fashion. Given the fides, all of the predictions of the risk analysis would satisfy the
classical ideal. Different analysts probably would argue as to the appropriate fides, but at least
it would be clear what the argument concerned. Only beliefs following deductively from the
fides would be counted as rational products of the risk analysis.
The probabilistic rationalist would argue that it is incorrect to adopt any principle (such
as a linear model) as a fide. Instead, all assumptions must be given a probability of being true.
30
-------
Deductions from the assumptions would be developed and the calculus of probabilities used to
assign a probability to each deduction. The rational result of a risk analysis would be a set of
predictions concerning the effect of radon on lung cancer, the assumptions underlying those
predictions, and the probability assigned to each prediction. These probabilities might be
obtained either through statistical considerations or by specification of a state of confidence. In
either case, the search is for the prediction (of cancer attributable to radon) which possesses the
highest probability.
The above approaches focus on the quality of beliefs about the link between radon and
lung cancer. A second party might argue that morbidity is equally as important as mortality.
He might also point out that genetic makeup predisposes some people to lung cancer. This
would lead him to claim that the risk analysis by the first party was not rational since it lacked
consideration of morbidity and genetic predisposition. The first party would counter that the
EPA said to consider the average risk of lung cancer, and that while this is a limited goal, the
analysis still has limited rationality. The two parties disagree as to when the limitation is so
severe as to require withdrawal of the claim to rationality.
While this debate was going on, the first party might point to other concerns in society.
There are needs other than the control of radon. Pesticides, food additives, etc., must also be
regulated. Increasing the considerations taken into account in analyzing the risk of radon would
draw resources away from other analyses. This might increase the lives saved from radon
mitigation, but at the expense of ignoring other risks in the interim. An appeal is being made
here to contextual rationality. The limited rationality of the bounded study (which considers
only the average risk of lung cancer) is justified by referring to the larger societal context within
which the analysis occurs.
31
-------
The risk analysis for radon also presumes a set of ends. It might be assumed that the
end is to regulate radon effectively, or to do it at the least cost, or to set up an atmosphere
within which regulation itself is given historical precedent. The focus might be on lung cancer or
pneumonia, on old age groups or young. Different groups usually will not agree as to the
appropriate goal in dealing with radon. Game rationality requires that this difference be
acknowledged and a process of analysis put in place which allows the groups to interact. This
can be accomplished by opening the process of analysis to scrutiny, making it clear how different
assumptions concerning the goals are reflected in different results of the analysis.
The analyst would need to make judgments concerning the "quality" of beliefs generated
in the analysis. A hope of the classical rationalist is that this quality can be judged on the basis
of a logical relationship between the evidence and the belief. Different groups of researchers
would weight the evidence differently. For example, epidemiologists favor even poorly
controlled human studies in uranium mines over well controlled animal studies involving
exposure of rats and beagles in radon chambers, while the reverse judgment is made by those
conducting the animal experiments. The rationality of the final risk analysis depends heavily,
then, on the proper selection of experts chosen to give evidential weight. The claim to
rationality would rest on the ability to demonstrate that the analysts were chosen for appropriate
skills.
As a risk analysis for radon proceeds, new perspectives will develop. Information
deemed irrelevant prior to the study will gain relevance through the analysis. No predictive
models of the effects of radon might be developed, requiring a reassembling of relevant data.
This might require drawing upon experts not identified in originally structuring the analysis. A
procedure must be put in place to ensure that the boundaries of the analysis are not restricted
to those established prior to the analysis. This procedure, then, justifies the claim to adaptive
32
-------
rationality. Similarly, the analysis of radon might reveal that morbidity is a serious outcome
deserving of attention, despite a prior commitment to examining only lung cancer mortality.
The control of morbidity would constitute a new goal of the analysis. Procedures put in place to
ensure that the risk analysis was capable of generating new goals would satisfy the claim to
posterior rationality.
4. WHERE IS RATIONALITY FOUND?
The following section provides a sketch of the way in which people might differ in
speaking about rationality, at least to the degree that they differ over the aspect of the analysis
which might be called rational or irrational. A key feature to be noted is the possibility that the
rationality of a risk analysis cannot be judged without also understanding the purpose behind the
analysis (i.e., what the analysis is to be used for).
4.1. Descriptive and Prescriptive Rationality
One of the central topics in rationality in areas such as decision theory is whether it is a
descriptive or normative (prescriptive) theory of human behavior. Psychologists and
behaviorists8'9 tend to view rationality as a scientifically describable and discoverable process
which is found in actual human behavior or thinking. Their goal, then, is to demonstrate the
dynamics by which specific humans or groups come to be rational, irrational, or non-rational.
The intent is descriptive and/or predictive, and their approach may be used to help discover the
underlying goals and cognitive processes leading to specific risk analyses or decisions.10,11
Contrasting with this approach is the school of thought which links rationality to an ideal
of human reasoning and/or behavior, which may or may not reflect actual historical cases.
33
-------
Rationality, then, is a capacity of humans which must be defined and nourished. Lying primarily
in the realm of philosophy, this school searches for an ideal conception of rationality which will
he used as a yardstick for measuring the quality of human thought or behavior. The focus is not
on how people actually do risk analyses, but on how they ought to do them.
Both of these views are found necessary in the present paper. If an individual or
organization is to resolve disputes, or at least keep them under some measure of control, it is
necessary to understand the rationality of specific disputants. At the same time, it must be
admitted that people rarely meet an ideal of behavior, however well intentioned they might be.
The problem, then, is to design a systematic way of thinking which will help people become
rational or to improve on their rationality. Descriptive theories describe society. Prescriptive
theories contribute to the evolution of society. Both theories are needed in determining the
reasons for disputes and in establishing a framework for judging the relative merits of competing
positions in the dispute, or for resolving the dispute.
42. The Rationality of Beliefs, Means and Ends
Putting aside the issue of description and prescription, it is then necessary to find what is
being described and/or prescribed in discussions of rationality. What would it mean to suggest
that an EPA regulation limiting the concentration of radon in drinking water to level X (such as
300 pCi/je) is rational? To answer this, it is necessary to isolate the components of such a
regulatory decision. In the most general sense, it might be argued that EPA:
(1) has a goal of producing a world in which the chance of death from cancer caused
by radon is Y (where Y might be 10"6 or 10^);
(2) believes that a concentration of radon equal to X produces such a world; and
(3) chooses to limit the concentration of radon to X.
34
-------
The three components to this process of reasoning must be subject to the constraints of
rationality. It might be asked if the belief of the Agency (that a concentration of radon in water
equal to X will yield a chance of death Y) is rational. In this case, the intent is to show that the
belief follows in some reasonable sense from evidence about the nature of radon and its causal
link to cancer. To be rational, then, requires that the various beliefs available to the Agency be
subjected to scrutiny to determine their degree of support in the evidence. This situation will be
referred to as the rationality of belief.
By the same token, the decision to limit concentrations to X is only rational because the
goal (or end) is determined to be a world where the chance of dying is Y. This rationality thus
requires a reasonable Agency goal. Some other group, such as a water treatment association,
might argue that the proper goal is to minimize expenditures on water treatment. The EPA
might respond by stating that its primary (or highest) goal is protection of public health, and
that economic factors are less important. This examination of the ends of action will be referred
to as the rationality of ends.
Finally, the Agency must select a particular means to reach its stated ends. The means
might, for example, be to limit concentrations in water to X, alter the use of water supplies,
provide warnings, etc. Presumably, these means will be consistent with both the beliefs of the
Agency and the ends. It is assumed that the means will lead to the ends in a demonstrable
manner. Choosing means appropriate to the ends will be referred to here as the rationality of
means.
This suggests three areas in which the rationality of an Agency decision or action might
be challenged. Discussions might focus on the various beliefs available to the Agency, such as
whether 100 pCi/je, 300 pCi/i or 10,000 pCi/je yields a chance of death equal to Y. Rationality
then would require that each belief be supported by evidence concerning the fundamental
-------
physical laws governing radon and cancer. A choice must be made as to how this belief might
be characterized. In the field of risk, this step typically is though of as risk analysis.
Discussion might also focus on the stated ends, requiring a clear demonstration that a
world containing a chance of death from radon exposure equal to Y should be the ultimate goal.
Others might argue that the proper ultimate goal should be something else, e.g., human
happiness, and that the chance of death is only one factor in assessing this goal. They would
demand proof that reducing the chance of death to Y without consideration of cost truly leads
to the state of greatest human happiness. Their judgment of the rationality of an analysis would
be determined by assessing whether that analysis considered aspects of risk related to the most
significant ends. In the field of risk, this concern is loosely termed risk policy.
The discussion also might center around the appropriate means to reach the goal. The
EPA might argue that aeration of water is preferred to other means such as removal by
activated carbon.12 A water works association might argue that the EPA should first identify
individuals sensitive to radiation, and then limit concentrations only in their homes. The
disagreement here revolves around the most efficient means to reach a given end, conditional on
a specific belief about the world. The rationality of the analysis would be judged according to
the ability of the analysis to provide appropriate solutions. This area of risk usually is referred
to as risk mitigation.
Controversies in the philosophy of rationality focus on the relative importance of these
three aspects of an action in claims to rationality, and on the manner in which the three are
interrelated. Logicians and most "pure" scientists tend to focus on the state of belief. To be
rational means to base actions on beliefs which satisfy some criterion such as "minimal
confidence" or "being the best available belief." The rationality of a risk analysis could then be
judged without reference to any uses of the analysis. Technologists, "applied" scientists and
36
-------
policy analysts tend to focus attention on means. Their criterion for rationality is that the most
efficient means be adopted for reaching a given goal. Only analyses which focused on aspects of
risk related to known techniques of mitigation would count as being rational. Ethicists and
policy makers tend to focus attention on the stated ends. Their criterion of rationality is that
the chosen ends in some sense be the "highest" ends, the most noble expression of human needs
and desires. Only analyses which focused on those aspects of risk associated with the highest
ends of society would count as rational.
These three groups can disagree fundamentally about the ability to be rational in all
three senses. Logicians with a leaning towards empiricism tend to argue that ends have no
rationale. Taking their cue from David Hume13, they think of ends as being mere matters of
taste bereft of any reasonable support, and of means as being matters of crafting. To them
there is no sense in which either ends or means are the subject of reasoned debate. In this
classical theory of rationality (discussed in more detail in Chapter 5), rationality is equated with
truth, and truth consists of a perfect belief about reality. Evidence is gathered and shown to
support a belief, such as that a concentration of radon in water equal to X yields a chance of
fatal cancer equal to Y. The most rational belief is the one corresponding most closely to the
features of the world. Such people do not argue that means and ends are not important. They
simply claim that these aspects are matters of unreflected skills and tastes, respectively. The
rationality of dealing with risks is located in the choice of an appropriate predictive belief,
assumed to be the best scientific estimate.
Those concerned with the rationality of ends tend to be less concerned with the logical
relationship between evidence and beliefs. This is not to say that they deny the need for such a
relationship. They see the relationship as being an issue of logic, which they separate from
rationality. To be rational is to guide life according to the highest principles, which may include,
37
-------
but is not restricted to, consideration of the "truths" of science and logic. Evidence and belief
are useful only as an aid in reaching a higher goal. This focus is termed "aim-oriented
rationalism" by Maxwell14. He insists that to be rational it is necessary to ask why a particular
belief is needed. He speaks of the rationality of belief, with its focus on logic, as leading to
"neurotic science." Whereas those committed to logic search for ever greater precision (risk
estimates accurate to several decimal places), Maxwell argues that a focus on beliefs is rational
only to the degree it allows humanity to reach "higher" goals, e.g., relieving hunger, poverty, and
disease. The most rational enterprise begins by examining goals and tailoring the pursuit of
belief to satisfy those goals adequately.
Maxwell's approach shades over into the "rationality of means." The history of
philosophy prior to the 19th century tended to revolve around debates between rationalists and
empiricists. These two schools of thought agreed that beliefs could be placed on a firm
"foundation," giving rise to the term foundationalism. Scientific theories were considered true
once they rested on such a foundation of established facts. Rational people sought to base their
beliefs upon "real" features of the world, which were taken to be objective features unrelated to
human conceptual schemes. Rationalists and empiricists disagreed, however, as to how realism
was to be found. Empiricists relied on experience, particularly sight or observation.15
Rationalists relied on clearly perceived human insights much as in mathematics. Once these
foundations were obtained, all of the features of the world could be predicted. These
predictions would be analytic consequences (hence, the word analysis) of the foundational
beliefs. The only trick was to ensure that the foundations were real, objective properties of the
world.
The views of rationalists and empiricists can be contrasted with "instrumentalism" To
the instrumentalist, what matters is human action, not belief. A belief is "true" (and therefore a
38
-------
rational basis for action) only to the degree that it produces a means for reaching a desired end.
A scientific theory may or may not refer to real features of the world. This can never be
known; there is no foundation for truth. A theory gains its power instead by allowing humans to
do something in the world. A theory is nothing more than an instrument and must not be
confused with reality. Most scientific theories are, after all, eventually overthrown by a new
conception of the world. Nevertheless, the older theories were not considered completely
useless. They at least allowed selected actions to take place within limits of error; but they
provided limited means for reaching human goals.
Under instrumentalism, a given means to reach an end becomes rational because it
"works." The unexplained skill of the scientist or engineer, which embodies what Polanyi40 calls
"tacit knowledge," even can be a rational means for reaching an end. This is true even if it is
not understood why it reaches that end. In this sense, removing radon from water becomes a
rational means of lowering cancer if it has "worked" in the past, even in the absence of
understanding how radon produces cancer. As Scriven16 says, "the proof one knows how to do
something is doing it, not talking about it." He is contrasting here the ideas of "knowing that"
and "knowing how." To an instrumentalist, the latter world always takes precedence over the
former. A rational person must give evidence that a means leads to a specific goal in practice.
The distinction between "knowing that" and "knowing how" is roughly equivalent to Aristotle's
contrasting ideas of theoretical and practical knowledge.17
4.3. The Separation of Beliefs, Means, and Ends
In the area of risk, there is a merging of the rationalities of beliefs, means, and ends. It
certainly should be hoped that beliefs about the effects of radon on the chance of fatal cancer
are as well established as it is possible to obtain. Greater precision of prediction is, therefore,
39
-------
an important sign of rationality. But there are an infinite number of aspects to a risk. The
scientist can study the risk of various kinds of effects, of the interaction between radon and
other substances such as environmental tobacco smoke, of the influence of age or nutritional
status, of the role of social forces in causing people to use a specific amount of water, and so on.
The risk analyst must select what to study from the large number of choices; there are always
constraints of time, energy, and cost. This is where ends can modify a risk analysis, influencing
the direction of attempts to develop belief. The policy maker does not simply want well
established beliefs, which may be the goal of the analyst in a scientific research setting.
Rationality requires that these beliefs be researched with particular ends in mind. The
relationship may even be reversed. Ends may be adjusted to correspond to areas where beliefs
are strong and where there is, therefore, a reasonable expectation that the ends can be reached.
Means also may influence both beliefs and ends. A research project on the risk from
radon (i.e., on beliefs about radon) may be stopped well before the risk is fully understood, if
the existing understanding is sufficient to allow a selection of means. Scientific research such as
an experiment may be viewed as providing evidence for a belief. It may also be viewed,
however, as a concrete example of how to modify the world. An experiment in which radon is
lowered and the incidence of fatal cancer drops certainly helps in testing theories about how
radon yields cancer. But it also provides evidence that physically lowering the concentration of
radon is a means of lowering the incidence of cancer. This means is available even if it is not
understood why it works. Society may be unconcerned with looking for strong beliefs about
radon, i.e., well established theories, if means for reaching ends already are available. Similarly,
strong beliefs which are at present useless in suggesting means may be counted as irrelevant in
the quest for rationality. It may even be the case that some aspects of the risk, e.g., the role of
social forces, may be ignored in the analysis because the policy maker is not willing to consider
40
-------
mitigation measures based on those aspects. As a result, beliefs, means, and ends subtly interact
within the area of risk. This suggests that the relationship between component parts is an aspect
of a rational system. This topic will be addressed further in the next section.
4.4. Rationality of Parts and Rationality of Relationships
To complete this section, it will be useful to focus on a separate distinction which might
be drawn in thinking about rationality. It might be required that particular beliefs, means, or
ends be analyzed independently, such as occurs when risk analysis, mitigation, and policy are
addressed separately. Each endeavor focuses on a belief or a means or an end without
reference to the other two aspects. Consequently, rationality lies in the belief, or the means, or
the end itself, depending on the task. This view will be referred to as intrinsic rationality and
suggests that the risk analyst, the risk mitigator, and the risk policy maker can perform their
tasks without consideration of the others.
Alternatively, it might be claimed that rationality concerns the relationship between
beliefs, means, and ends. In this case, it is not the beliefs, means, or ends themselves that are
questioned. These become the givens of life. They may be true, useful, just, etc., but they are
not in themselves rational. Rationality requires, instead, a specific relationship between these
parts. It requires that society act to get what is desired given that the world is believed to be in
a particular state. In the words of Bertrand Russell18, rationality is the "just adaptation of means
to ends," an approach to be referred to as formal rationality. Given that the EPA (or another
group) believes that a concentration of X yields a chance of fatal cancer Y, that the EPA desires
to produce Y, and that lowering the concentration to X is possible, the rational action is to
control concentrations at X. It is rational because the means were consistent with both the
stated beliefs and the stated ends. Rationality, then, is located in the entire regulatory process
41
-------
and can only be judged by someone familiar with that entire process. This does not imply that
treating two aspects such as risk assessment and risk management independently is non-rational,
only that the rationality of the assessment begins once objectives have been assigned, e.g., the
task of risk assessment is to address specific health effects, such as cancer.
4.5. Summary Remarks on Rationality for Risk Analysis
In summary, rationality lies in the quality of the predictions generated by a risk analysis,
but also in the means and ends which the risk analysis will serve. The rationality lies in the
three separate parts and in the relationship between those parts. The latter view implies that a
risk analysis must be structured to yield predictions related closely to existing means for
reaching a clear set of ends.
Since logical rigor is associated most closely with questions of belief, practical action with
the development of means, and moral reasoning with the choice of ends, people will tend to
locate rationality differently depending upon which of these three activities they value most and
understand best. Increasing specialization tends to push people towards this intrinsic rationality,
since they may be unfamiliar with the relationship between the parts of action. Scientists, for
example, may tend to be more familiar with the state of scientific beliefs than with means or
ends acceptable to society. Technologists are apt to be more familiar with means than with
scientific belief or ends. Public policy practitioners may be familiar with the relationship
between beliefs, means, and ends, but not intimately knowledgeable about specific details
concerning the parts. They will, as a result, tend towards formal rationality.
The most complete vision of rationality incorporates beliefs, means, and ends, judged
both intrinsically and through their relationship. The "complete" rational person chooses
rational ends, selects means appropriate to those ends, and then justifies those means and ends
42
-------
(to the degree possible) by reasoned beliefs. The individual component parts and their
relationship are the subject of scrutiny and clear reasoning. Weakness within either a
component or in the relationship between components weakens the rationality of a decision and
associated action. Rationality is a kind of web where both the individual "nodes" (belief, means,
and ends) and the lines stretching between them (the relationships) must be constantly subjected
to reason, and, thereby, strengthened. Individuals or groups may be selected to work on specific
aspects of the web, but their work must be guided through a more complete conception of
rationality. The present research is intended to provide guidelines for understanding of the
quality of predictions generated by a risk analysis, and how that quality is related rationally to
means, ends, and beliefs.
In summary, intrinsic rationality refers to addressing risk analysis from a singular
perspective, such as the rationality of beliefs, the rationality of means, or the rationality of ends,
which may align most closely with the tasks and perspectives of the risk analyst, the risk
mitigator, and the risk policy maker, respectively. Although each rationality described is
important and of interest in itself, they are insufficient to the decision making process as a
whole, which must include the relationship between beliefs, means, and ends.
Having described basic features of rationality in a very general sense, it is necessary now
to provide more detail. Specifically, we ask how a particular belief, means, or end, or the
relationship between them, can be supported by reason. Since most of the literature on
rationality has been built around beliefs, and since beliefs are the product of a risk analysis, this
aspect of rationality will be used as a primary example in the following section. The general
ideas, however, apply to all aspects of rationality except as noted.
5. CHARACTERISTICS OF REASONS AND RATIONALITY
43
-------
This section of the report explores briefly the history of attempts to show why, and how,
a given belief or prediction may be claimed to justify adoption of a particular means to a stated
end. The various attempts will differ in the manner and degree to which beliefs, means, and
ends impact on each other. These differences lead to competing views on the extent to which a
risk analysis can be driven entirely by the internal goals of the scientific research community.
Sections 5.1 through 5.9 describe, from a historical perspective, philosophies on the
evidential support of belief. Section 5.10 moves to the question of whether belief is adequate.
Finally, some recent studies in the history and sociology of science are discussed that help to set
the stage for current thought on this topic. The concept of coherence as a goal for the rational
human is discussed. A summary section provides a brief listing of the various approaches to the
rationality of evidence and belief.
5.1. The Classical Theroy
Discussions of the history of rationality begin with Plato in his depiction of Socrates.19
This Platonic view has come to be known as the Classical Theory of Rationality, and much of
modern philosophy has arisen either in reaction to, or in support of, the classical theory. Most
practicing risk analysts presume the theory in supporting the use of risk analysis.
Plato conceived of rationality as the pursuit of Truth. When a person grasps the Truth,
he or she also has grasped Beauty, which is the highest end in life. To be rational, then, means
to have possession of the Truth. Moreover, it is necessary to realize that the Truth has been
located.20 In this sense, the Classical Theory of Rationality is similar to the rationality of belief
discussed earlier. Still, it involves the rationality of ends (since the highest end is truthful belief)
and the rationality of means (since it insists on logic as the proper route to truth).
44
-------
The classical theory has been summarized by Brown21 and by Agassi22. The theory
involves five primary attributes by which a belief may be judged rational.
(1) The belief must contain terms which are clearly defined an objective features of
the world. There is no room for ambiguity, since the world is not ambiguous.
(2) The belief must be formed on the basis of clearly defined rules of reason, which
are taken to be those of logic and mathematics. In short, the belief must be
deduced from what already is known by well established rules of reason.
(3) The rules must be universal and applied consistently. As stated by Kant23, "Act
only on that maxim through which you can at the same time will that it should
become a universal law." It must be very clear when the rules are to be applied.
This attribute is intended to prevent people from intentionally shifting allegiances
to beliefs when they are inconvenienced by them.
(4) A belief must be necessary in the logical sense. It must both rest on a firm
foundation of established truths and follow deductively from those truths. All
persons must come to the same conclusions if they are open to reason.
(5) There must be an algorithm which allows the rules to be applied in a finite
number of steps. An example of this is the use of syllogisms or of mathematical
formulae. Rules alone are not enough if they cannot be completed in a finite
time.
The net result of the classical theory of rationality is to remove human judgment from
beliefs. The human acts only as the vessel of belief. He or she has no choice in what to believe,
since the belief is dictated by firmly established truths and the laws of logic. There will,
therefore, be a complete consensus of belief between all rational people. In a very real sense,
45
-------
people are not responsible for their rational beliefs and have a guarantee that the beliefs will
lead to correct interactions with the world.
The most well developed conception of classical rationality is contained in set theory and
its application to a "covering law" hypothesis of science. The object to be analyzed (such as
human health and radon) is broken into distinct analytical categories. These categories contain
the entities of the world (lungs, radiation, radon, cells, cancer, etc., in our context) and constitute
a well defined and finite set. Each entity is said to possess a set of characteristics or attributes
(mass, energy, transformation, etc.) which explain the behavior of the entities. Explanations of a
particular phenomenon, such as radon yielding lung cancer, then are based on these attributes,
in all cases "covered by" the laws of the attributes. For example, it might be stated that all large
populations of people receiving X amount of radiation show a fraction, Y, which die of lung
cancer. This statement is a "covering law" which covers all phenomena satisfying the condition
of being a population receiving X amount of radiation. The rational person faced with radon in
the home is concerned with whether his case falls into the set of entities satisfying the above
condition. It either does or does not, i.e., each phenomenon is either in the set or out, and the
behavior of the phenomenon is certain if it falls into the set. The task of the rational person is
to be very clear as to how an entity qualifies to be in the set and to apply the appropriate rules
of logic in analyzing the properties and interactions between sets.
This classical view leads to a situation in which rationality and logic are essentially
identical. To Plato the world consists of universal, necessary, and eternal truths, so rationality
must also possess these attributes. Anything less may be a useful tool for action (Aristotle's
practical reasoning) but cannot count as being rational. These are very strong requirements, and
very little in life satisfies this classical model of rationality.
46
-------
52. The Skeptical Attack on the Classical Theory
Attacks on the classical view have tended to take the form of skepticism. Even in Plato's
time, there was a philosophical school, the Skeptics, who claimed that nothing satisfies Plato's
ideal. This led the Skeptics to the claim that no belief could be rational. It is important to note
here that this extreme pessimism followed from the Skeptics adherence to the classical view of
rationality.24
53. Empiricist Rationality
Locke25 hoped to bring rationality back into respect by focusing on experience as
providing a base for firm belief. Observations, particularly those from well controlled
experiments, would yield beliefs which were a necessary consequence of the observations. If this
were the case, the goal of Plato would be reached. Hume26, however, followed with a skeptical
response. As suggested by March27, rationality requires a decision as to the future actions of the
world. It is in the future, after all, that decisions will be carried out. For Hume, no prediction
of the future follows necessarily from past observations. Without necessity, skepticism was the
proper attitude and rationality was impossible. While Hume agreed with Locke that empirical
evidence, i.e., experience, was the best route to knowledge, he would not admit that necessary
beliefs ever could be found. Hence, Hume denied claims to rationality.
5.4. Rationalist Rationality
Descartes28 found Hume's reliance on experience untrustworthy. He rejected the
empiricism of Locke and Hume to rely on human insights instead (a bit like Plato). He
believed in the capacity of reason to discover truths which were "clearly and distinctly perceived,"
much as in logic or mathematics. Mathematics was, in fact, his model for rationality. The
47
-------
rational person discovered, through introspection, a set of axioms about the world. The rules of
mathematical logic then were applied to these axioms to yield necessary truths. Since these
rules are clearly defined, universal, and led to necessary truths, a rationalist in the Cartesian
sense satisfied the requirements of the classical view of rationality. All that was required was
the rather questionable belief that the axioms of the world could be obtained through a special
form of introspection.
5.5. Logical Positivism
At the beginning of the 20th century, logical positivism29 was introduced through a
philosophical circle in Vienna, known today as the Vienna Circle. Logical positivists returned to
Locke in believing that it was possible to begin reasoning with a set of "observation sentences"
concerning the objective properties of the world. These sentences would refer entirely to sense
impressions, but primarily sight, which was taken to be the most reliable sense. Experimentation
would provide those impressions. Being purely statements of observation, a rational person
could be "positive" about those beliefs. The rules of logic then could be applied to deduce any
new predictions about the world (hence, the name "logical positivism"). The only trick was to
produce observation statements which were necessarily true statements about the objective
features of the world. To be rational meant to believe a statement only if it was an observation
statement or followed deductively from observation statements. As such, logical positivism is a
form of the classical view of rationality, wedded to a strong idea of empiricism.
5.6. Rational Skepticism
Karl Popper was initially attracted to the Vienna Circle, but broke from that line of
reasoning. His primary complaint was that no statement about the future could follow
48
-------
necessarily from past observations (the contribution of Hume). Such statements were, instead,
mere hypotheses requiring further investigation.30 The proper attitude towards all statements,
including those of science, was skepticism. Still, Popper rejected the classical skeptic's claim that
people could not be rational. He argued that rationality was to be found in the process by which
hypotheses were subjected to tests. Central to his idea of rationality is criticism of an idea,
which included attempts to show that the idea is wrong. A rational person attempts to find
evidence which falsifies beliefs. As in logical positivism, this evidence was taken to be
observational experience. The rational person adopted the belief which had survived the most
stringent attempts to falsify it. Competing beliefs could not be shown to be true, but it was
possible to talk of the "truth content" of beliefs by referring to their success in avoiding
refutation or falsification. Rationality, in Popper's view, is a process of criticism in the form of
attempts at falsification and acceptance of the belief most resistant to criticism.
Another key point on which Popper and the logical positivists disagreed is the role of
"confirming" evidence in choosing beliefs. As mentioned previously, the positivists assumed that
observation of a phenomenon, such as a risk of Y, counted as a positive reason for believing any
theories which predicted Y. The observation in a sense confirmed these theories. A rational
person chose the belief which had been the most highly confirmed or "verified." This test of
verification was made against the full range of existing observations deemed to be relevant to a
belief. Popper countered that it was not enough for a belief to be confirmed by an observation,
since it might be the case that any belief would be consistent with that observation. It might
also be the case that a belief was so vague that any observation could be construed to confirm it
(this Popperian argument is often made against multistage theories of carcinogenesis). This led
him to his rationality of falsification, in which there must be a strong chance that a theory would
fail to be confirmed by an experimental test. Popperian epidemiologists have, for example,
49
-------
argued that the multistage theory of carcinogenesis is so flexible that it could fit any set of data.
There is no way, therefore, to say that the theory has the potential to be falsified, and it should
be rejected as unscientific.
5.7. Probabilistic Rationality
Regardless of whether verification or falsification is adopted as a rational process, there
is the problem of ambiguous, variable and conflicting evidence. All beliefs will have some
evidence in their favor and some against. In addition, observation will be limited to finite
samples which, as Popper correctly points out, may miss important features of reality.
Recognition of this situation gives rise to the probabilistic theory of rationality. Here, evidence
does not necessarily verify or falsify a given belief. It lends, instead, a "degree of belief' based
on some idea of the probability that a piece of evidence would have been produced if the belief
were correct.
This degree of belief has taken two main forms in the literature on rationality. The first
is classical statistics.31 Statistical properties of the method of observation are used to estimate
the probability that an observed result would be obtained if a belief were true. This probability
is taken to be an objective property of the method of observation, so the resulting assignment of
probability satisfies the classical requirements for objectivity, necessity, and universality. A
rational person chooses the belief which is assigned the largest probability based on the existing
observational evidence. All persons who agree with the statistical assumptions (which
themselves may be open to question) will assign the same degree of confirmation or belief.
Contrasting with this objective approach is a subjective approach to probability often
referred to as Bayesian confidence (although it is possible to establish a quasi-objective Bayesian
methodology32). Bayesians recognize that new beliefs follow from a mixture of prior beliefs and
50
-------
new observational data. In a methodology to be discussed in a later report, the confidence in a
belief is adjusted historically as new data are presented to the mind. This confidence is said to
be a measure of the state of mind of a person, rather than an objective property of the observed
phenomenon or the method of observation. Still, the Bayesian approach can be made to yield
consistent and necessary estimates of confidence based on a formal set of axioms32, thereby
satisfying some of the requirements of the classical view of rationality. The rational person then
adopts the belief associated with the highest confidence.
5.8. The Problem of Incommensurability
Common to all of the views discussed above is an assumption that beliefs, such as
scientific theories, can be compared simultaneously to a common set of observations. Once the
observations are specified, the degree of verification or falsification can be assigned and the
beliefs ranked accordingly. Kuhn33, however, denied this claim. Using scientific theories as the
main example of a potentially rational activity, he asserted that theories are basically
incommensurable. In other words, the theories themselves specified what to look for in the
world and how those observations would count as evidence. People working under different
theories observed the world in different ways and spoke differently about the world. As a result,
there was no fixed set of observations against which competing beliefs could be assigned
measures of confirmation or falsification. Beliefs must be seen instead as systems of belief, each
of which set up traditions of research but basically could not be compared. It was possible to be
rational within a scientific research tradition, but all such traditions were internally rational so
long as they remained open to change in light of the evidence they uncovered.
Kuhn's position on the rationality of science was misused by both his opponents and
supporters, prompting him to reply in a second book.34 The primary mistake made by both
51
-------
groups was the assertion that his position leads to complete relativism. If all scientific theories
choose the data which must be explained, and how those data are to be explained, why not say
that all areas of human knowledge are equally valid? After all, all subject areas, e.g., science,
religion, art, specify the experiences to be confronted, the methods of exploration, and the rules
of explanation. Since these features only make sense within the language of the theories, there
is no way to step outside of them all and compare them on a common basis. Science,
mythology, religion, and art simply become competing systems of thought. Each is internally
consistent and none may be rejected through appeal to higher criteria. The same may be said of
competing scientific theories.
This view found favor both in the social sciences and in the philosophy of Paul
Feyerabend35. Feyerabend argued that scientific theories were completely incommensurable and
internally coherent. There was not, therefore, any way to reject one in favor of any other.
Scientific choice between theories was profoundly irrational and the only reason scientists chose
one over the other was to be found in the exercise of power, prestige, etc. To be rational, a
society must give up the quest for selection between scientific theories. All theories, whether
scientific or not, must be encouraged to prosper on an equal footing. When the time came for
decision, democracy must be the method for choice. Kuhn's critics saw in Feyerabend the
natural culmination of the thesis of incommensurability, leading inexorably to irrationality, or
non-rationality, and ultimately to an anarchy of beliefs.
This turn of events was troubling to Kuhn. Within social science and anthropology, the
Kuhnian idea of a paradigm came to mean any model of the world. Since most ways of thinking
counted as a model of the world, most ways of thinking satisfied the (false) picture of a
paradigm and, therefore, constituted a perfectly acceptable theory of the world. But this misuse
of the idea of a paradigm greatly simplified and distorted Kuhn's original intent, although part
52
-------
of the blame lies in the vagueness of his original writings. Paradigms were intended to provide
a concrete means of confronting scientists with empirical tests of their theories. Different
theories might specify different tests to be performed, but all proper paradigms gave rise to
experiments which held the chance of verifying or falsifying the theory through tests against
experience. Not every theory of the world satisfied this criterion, so not every theory of the
world was rational. In addition, researchers within a theory were capable of admitting that their
theory was not working very well by their own rules. There even were many observations which
all theories agreed were important tests of any theory. Subjectivity and relativism were not
natural consequences of his view on rationality. Still, it is important to bear in mind his
important observation that researchers working within different theories may look for different
bodies of evidence in supporting those theories.
5.9. Tests of Theories Other Than Empirical Tests
Kuhn shared a common ground with Aristotle, Locke, Popper and the logical positivists
in adopting empiricism as the test of a belief. The proof of a theory was in the degree to which
its assumptions could be observed or in its ability to correctly predict the results of an
observation. Laudan36 agreed with this claim but added another criterion for rational
confidence. At times, a theory solves conceptual problems, which are quite different from
empirical problems. Consider, for example, a theory of radiation carcinogenesis which proposes
that radiation damages DNA and turns on an oncogene. This oncogene then produces cancer.
The theory provides an explanation which includes the role of oncogenes. If scientists had been
wondering how oncogenes fit into the picture of radiation and cancer, the new theory may be
said to solve a conceptual problem. This solution counts in favor of the theory even if it does
not yet lead to increased precision in predicting the incidence of cancer. With Laudan,
53
-------
therefore, rationality takes on a new measure. A rational theory explains, in a conceptual sense,
the most significant problems remaining in a field. While the ability of a prediction to "fit" data
is an empirical problem dictated by the rules of observation and statistical reason, the
conceptual ability of a theory is judged by humans. It is the human who assigns relevance and
importance to a conceptual problem, requiring an act of judgment. It is the human who decides
how well a theory explains a particular problem or puzzle.
A similar tact was taken by Longino37. She admitted that rules of reason might well
apply once a set of data were chosen as a test of a theory. Choosing the data which were
relevant as a test was, however, a process buried in human judgment. The cell biologist studying
the transformation of cells in vitro following irradiation might assume the results were highly
relevant to predictions of cancer in humans. Any theory of cancer would, therefore, be required
to explain and predict the in vitro experiments. Epidemiologists might argue the opposite,
asserting that the in vitro studies were irrelevant to theories of cancer. There is little hope of a
logical resolution to the decision, since assigning relevance to an observation presupposes the
theory which gives relevance to begin with. In this sense, Longino's insights are similar to those
of Kuhn. Far from being an abstract exercise in philosophy, the concerns of Longino were at
the heart of a debate on the effects of radiation on human cancer.38
5.10. The Rationality of Crafting in Science
All of the philosophies discussed to this point have held a common feature. They were
concerned with the degree to which evidence might support a belief. The belief was precise and
the only question was whether it was true or false. Laudan36 moved the focus slightly and asked
whether a belief is adequate. A key feature of modern philosophy of science is increased use of
the concept of approximation. No theory or belief is said to be completely true or false. Parts
54
-------
of the world are missing from models. The mathematical relationships between the parts are
understood partially. Even measurements of agreed upon relationships are imprecise. If truth
and rationality require complete empirical success for a theory, then no theory is true and no
one is rational.
And yet, scientific theories can be said to form a rational basis for belief. All that is
required is a change of emphasis on the need for complete predictive success. A successful
theory then predicts adequately within the guidelines set up by a community. Human judgment
is needed to define the term "adequate." This judgment cannot be avoided and usually will be
related to the ends to which the theory will be used. Still, once the idea of adequacy is given
clear expression, an imperfect theory may be said to provide a rational basis for belief.
A similar idea was advanced by Ravetz39, who likened science to a craft. Adequacy could
be thought of as being analogous to "tolerance" in engineering. If two parts must fit together
within an engine, there is no need for them to be perfect fits. They must, instead, fit together
within a certain margin of error (called the tolerance). Greater accuracy (beyond the desired
tolerance) does nothing to satisfy the goal of making a working engine. Lower accuracy may,
however, result in the engine shaking itself apart. As with Laudan, the rationality of belief for
Ravetz requires an idea of how a theory is to be used and how adequate a theory is to that use.
Ends, means, and beliefs converge, therefore, in the act of crafting.
This idea of crafting is carried even further by Polanyi40, Heidegger41 and Rouse42. Their
argument is that science is, first and foremost, a way of doing something. As mentioned earlier,
the proof of science (or of a scientific belief) might be said to lie in "knowing how," in practice,
rather than "knowing that." A proper belief is simply a conceptual representation of a method
for doing something. Experiments test the ability to produce a particular physical state in the
world, not the truth of beliefs. Beliefs (such as theories) are ways humans have of talking about
55
-------
successful actions. That, and only that, makes a belief rational or irrational. The rationality of
belief and of means shade over one into the other.
The focus on crafting and doing tends to negate some of the original concerns of Popper.
Popper discounted confirming evidence both because a theory or model might agree with reality
for the wrong reason and because they might be adjustable to fit almost any set of experiences
(such as experimental data). This is, indeed, a major problem if it is required that a theory or
model be a truthful explanation of the world. If, however, their role is to conform only to the
surface features of the world (i.e., to mimic the world), then truthful explanations are not so
important and confirmation implies that the theory or model at least functions like the world
(even if for the wrong reasons). This functional ability may be thought to make the theory or
model a perfectly rational basis for action, if not belief.
5.11. Scoiological Views of Science
This brings to a fairly complete end the discussion of rationality and belief. Two topics
remain to be addressed, however, if only briefly. The first concerns recent studies in the history
and sociology of science.42,43 The purpose of philosophy of science has been to explain why it
was rational for scientists to adopt a particular belief in light of evidence existing at some
moment. This might require what is known as a "rational reconstruction" of the actual history of
a science, but it still would be possible to show that a belief might have been rational according
to the criteria discussed in earlier paragraphs. Someone using the results of a risk analysis
typically is more concerned with whether a belief can be made rational, not whether a specific
analyst was in fact rational in developing that belief. Sociologists of science, however, denied
that any such rational reconstruction was either possible or relevant to the history of science (or
of risk analysis).
56
-------
These sociologists directed the focus away from the relationship between evidence and
beliefs and onto the social and physical setting of science. Acceptance or rejection of beliefs was
said to remain loosely based in evidence and mathematical reason, but the main driving forces
were prestige, social relationships, funding, strong personalities, etc. These factors influenced
the direction of research, the methods chosen for research, the assignment of relevance to
evidence, and so on. Since real risk analyses come out of actual historical circumstances, rather
than rational reconstructions of history, it seems wise to pay heed to the lessons from sociology
and history of science, if only to work harder at strengthening the influence of more classical
notions of rationality. The lesson from sociology is that risk estimates from "expert committees"
may be driven more by the dynamics of the committee than by the evidence itself.
5.12. Foundations Versus Coherence
Finally, there is an issue related to the proper metaphor to be used in picturing rational
belief. Philosophers in the classical vein tend to follow the idea of foundationalism. They
conceive of knowledge and belief as resting on a set of foundations or well established "facts' of
the world. All other beliefs are built on these foundations. Quine44 challenged this view of
belief. He noted that no philosophy had yet discovered a firm foundation beyond dispute. In
the place of foundations he inserted the idea of a "web of belief." Since no belief acted as a
foundation, all beliefs must be mutually supporting. Changes in one belief then would affect all
of the others. A rational person strives for coherence between beliefs. Conflicting beliefs are
rooted out and reconciled in a constant process of revision. No single belief is immune to this
revision, even (as Maxwell also asserts14) metaphysical beliefs about the ends and methods of
science.
57
-------
5.13. Summary
A very brief listing of the various approaches to the rationality of evidence and belief is
given below.
1. The Classical Theory--a risk analysis is rational if the predictions are necessarily
true. Truth is the only goal, regardless of whether that truth helps to meet other
goals (it is presumed that it does). This necessary truth may be founded on
empiricism or rationalism.
A. Empiricism-necessary truths in risk analysis are obtained by restricting
the predictions to those which have been observed. A variant of this
approach is logical empiricism or logical positivism.
B. Logical Positivism-necessary truths in risk analysis are obtained by
restricting the predictions to those that are deduced (logically) from
observations.
C. Rationalism-necessary truths in risk analysis are obtained by restricting
the predictions to those that are deduced (logically) from "clearly and
distinctly perceived" (i.e., intuited) insights, as in the axiomatic approach
to mathematics.
2. The Skeptical Approach-no belief is a necessary truth. Therefore, rationality is
not possible.
3. Critical Rationality-necessary truths cannot be attained from a risk analysis. The
proper attitude is one of skepticism. Still, rationality is possible by requiring that
beliefs constantly be questioned and debated. This debate occurs through the
testing of beliefs against evidence. The test may be one of falsification or
verification. The evidence usually is taken to be observational, and the test to be
correspondent to the observation. At any moment, it is possible to compare all
beliefs against a single body of evidence.
4. Paradigmatic Rationality-necessary truths cannot be attained from a risk
analysis. The judgments of belief (as in #3) must be made on the basis of
evidence, but the evidence differs between different beliefs (i.e., between
scientific theories). Rationality implies critical discussion, but the discussions
take place within research traditions, with these discussions being
incommensurable.
5. Conceptual Rationality-theories used in risk analyses are not judged purely on
correspondence to evidence, but on the degree to which the theory resolves
conceptual difficulties in a field. Empirical tests (as in #3 and # IB) are
important, but so is conceptual clarity and coherence.
58
-------
6. Rationality of Evidential Relevance-rationality in risk analysis requires that
evidence (regardless of whether it is empirical, conceptual, or the result of
insight) be judged relevant to a problem. This judgment is not a necessary truth
and must be made by researchers using human skills. The judgment of relevance
becomes part of the background assumptions (often hidden) adopted by a
particular analyst. Complete rationality requires a critical discussion of these
judgments.
7. Sociological Rationality-the goal (usually hidden) of the risk analyst is to
maximize power, prestige, etc. Rationality requires that these goals be
recognized and a social system adopted to modify their effect on the analysis.
8. Instrumental Rationality--a risk analysis is rational only if it produces explicit
means for mitigating the risk. Strong beliefs are not sufficient if they do not lead
to demonstrable solutions.
6. CONCLUDING REMARKS
The preceding chapters provide an overview of the nature of rationality and indicate
properties required of a framework for it to reflect this nature in risk analyses of environmental
carcinogens. Specifically, the framework (to be developed in the next part of this report) must
have the following properties:
(1) It must be capable of depicting how inferences of risk are related to premises
concerning probability, evidence and the physical world (satisfying the deductive
ideal of rational analysis).
(2) It must incorporate the full range of evidence typically brought to bear on a risk
analysis for carcinogens (satisfying the ideal of rational coherence).
(3) It must be capable of making explicit the relevance of particular bodies of
evidence to inferences of risk (satisfying the requirement of rational analysis of
evidential relevance).
(4) It must be capable of allowing an explicit incorporation of potential links between
beliefs, means, and ends (satisfying the requirement of formal rationality). These
links will depict how specific effects are selected for analysis contingent on
specified ends and how specific qualities of evidence are weighted according to
the degree to which means (or methods of mitigation) are considered.
(5) It must be capable of displaying the critical points at which human judgments
must be made of evidential strength and relevance, and of relating these
59
-------
judgments to specific assumptions concerning the nature of rational belief
(thereby aiding researchers in assessing their state of confidence).
(6) It must be capable of demonstrating how competing risk analyses arise from
differing explicit assumptions (aiding in the explication and resolution of
disputes).
(7) It must be susceptible to use as a tool for producing either qualitative or
quantitative judgments concerning the strength of inferences of risk (thereby
avoiding necessary commitment to specific schools of thinking about evidential
strength, the nature of probability, etc.).
(8) It must provide an explicit link between judgments made by researchers in a wide
range of scientific research, as well as judgments made by parties concerned with
the less specific issues of evidential reasoning, appropriate ends, and so forth
(satisfying the goal of game rationality).
The next part of this report develops a framework for discourse on hazard identification
with the eight properties described above. Additional useful concepts, such as a taxonomy of
claims of carcinogenicity, are also described and implemented in the framework. The result is a
seven-step guide to aid the individual decision maker in drawing inferences of carcinogenicity
and to facilitate discourse and aid conflict resolution between persons of differing opinions.
Table 0 has been constructed to aid the reader in making connections between this part
of the report and the next, and to make explicit where and how the concepts and principles of
rationality discussed in the preceding chapters are implemented in the seven-step guide for
applications. The table is located below to facilitate easy reference while reading Part II, which
follows.
60
-------
TABLE 0. A SUMMARY OF IDEAS IN CHAPTER 1 AND THEIR RELATIONSHIP
TO THE FRAMEWORK OF ANALYSIS PROVIDED IN CHAPTER 2.
Philosophical
Principle
Summary Description
Relevant Pages
in Chapter 1
Role within the Framework of Analysis in Chapter 2
Conceptual
Rationality
An attempt to provide consistent and clear
meanings to terms in an analysis.
12, 15, 39
Useful in establishing meaning of "claims" in Working
Tables 3, 6, 7; as well as all other terms in the analysis.
Logical
Rationality
An attempt to apply rules of deductive reason in
arriving at the results o f an analysis equivalent to
classical rationality.
12, 22, 38-41,
42
Useful in determining the assignment of epistemic
value to the relevance strategies in Working Tables 3,
6, 7; and to the column and overall summary entries in
these tables.
Methodological
Rationality
The habit of questioning beliefs used in an analysis.
Also termed process rationality.
12, 22, 43
Useful in determining the assignment of epistemic
value to the relevance strategies in Working Tables 3,
6, 7.
Epistemiological
Rationality
An exploration of the evidential support for beliefs,
and of the principles of by which the support is
judged.
12, 14
Useful in determining the assignment of epistemic
value to the relevance strategies in Working Tables 3,
6, 7.
Ontological
Rationality
The insistence that only entities demonstrated
convincingly to exist be employed in the reasoning
of an analysis.
12, 39
Useful in determining the data items to be
incorporated into Working Table 4.
Valuational
Rationality
The insistence that the process of analysis reflect
the most valued aspects of human intellectual
activity and of the ends of the analysis.
12, 16, 23
Useful in determining the degree of intellectual
obligation assigned to relevance strategies in Working
Tables 3, 6, 7; as required by column summary.
Practical
Rationality
The insistence that an analysis be a practical means
to action.
12
Useful in determining whether the different "claims of
carcinogenicity" categories are desired goals of the
analysis (Working Tables 3, 6, 7).
Evidential
Relevance
The ability of evidence, in the context of
background assumption to have bearing on a given
belief.
14
Useful in establishing Context Type (Working Table 1
and Step 2 of Fig. 4); Causality in Working Table 2;
and relevance strategies (Working Tables 3, 6, 7).
Empirical
Success
The ability of a theory to predict the observable
properties of a phenomenon. See also logical
positivism (p. 42).
18, 41, 42
Useful in determining the assignment of epistemic
value to the direct empirical, semi-empirical, and
theory-based-inference relevance strategies in Working
Tables 3, 6, 7.
(continued on following page)
61
-------
Table 0. (continued)
Philosophical
Principle
Summary Description
Relevant Pages
in Chapter 1
Role within the Framework of Analysis in Chapter 2
Conceptual
Success
The ability of a theory to provide a causal
explanation of observable properties of a
phenomenon.
18
Useful in determining the assignment of epistemic
value to the theory-based inference relevance strategy
in Working Tables 3, 6, 7. Also useful in assigning
causality in Working Table 2.
Hard look
The claim by an analyst that the assembled
evidence constitutes a reasonably complete body
reflecting the full body of evidence that would be
available given no restrictions of time or money.
19
Determination of the value of data completeness in
Working Tables 2, 4.
Correspondence
The philosophical principle that a theory is
established to be true by its ability to predict a
given observation.
18, 51-52
Useful in determining the assignment of epistemic
value to the theory-based-inference relevance strategy
in Working Tables 3, 6, 7.
Coherence
The philosophical principle that a theory or belief
is established to be true if it is not inconsistent
with other beliefs held by the analyst.
20, 51-52
Useful in determining the assignment of epistemic
value to the theory-based inference relevance strategy
in Working Tables 3, 6, 7. Also useful in determining
the assignment of epistemic value for column and
overall summaries in these tables. Also useful in
ensuring that value of intellectual obligation is
consistent across analyses.
Fides
Those beliefs taken by the analyst to require no
attempt at evidential justification.
20, 22
Useful in determining the strength of background
assumptions appearing in relevance strategies of
Working Tables 3, 6, 7.
Epistemic Status
The weight of the evidence brought to bear in
supporting a belief.
21
Identical to assignment of epistemic value to relevance
strategies in Working Tables 3, 6, 7.
Adaptive
Rationality
The belief that a rational system of analysis should
contain flexibility to accommodate new
insights/knowledge. See also posterior rationality
(p. 24).
23, 24
It is useful to reflect on this concept in determining the
degree to which categories of carcinogenicity claims
will be tied to specific theories of carcinogenesis.
(continued on following page)
62
-------
Table 0. (continued)
Philosophical
Principle
Summary Description
Relevant Pages
in Chapter 1
Role within the Framework of Analysis in Chapter 2
Selective
Rationality
The set of principles by which individuals are
selected for particular tasks and/or judgments.
23
Useful in determining the assignment of epistemic
value to the existential insight relevance strategy of
Working Tables 3, 6, 7.
Rationality of
Belief
The ability to demonstrate that beliefs are justified
by available evidence.
29-35
Useful in reflecting on the epistemic status of claims in
Working Tables 3, 6, 7.
Rationality of
Means
The demonstration, through reference to beliefs,
that a given means is likely to produce the desired
ends.
29-35
Useful in reflecting on whether a chosen category of
claims of carcinogenicity will be useful in reaching the
ends of the analyst and/or policy-maker.
Instrumentalism
The philosophical position that theories are
established to be true if they allow practical action.
33
Useful in determining the assignment of epistemic
status to the theory-based-inference relevance strategy
of Working Tables 3, 6. Also useful in assigning
degree of intellectual obligation to the relevance
strategies in these tables.
Realism
The philosophical position that theories are
established as true if they contain reference only to
entities and relationships that have ontological
status (see ontological rationality).
32
Useful in determining the assignment of epistemic
status to the theory-based inference relevance strategy
of Working Tables 3, 6. Also useful in determining
allowed data items in Table 2.
Rationalism
The philosophical position that beliefs are
established as true if they are clearly and distinctly
perceived by the human mind, or are obtained
from rationally justified beliefs using the rules of
deductive logic.
42
Useful in determining the assignment of epistemic
status to the various relevance strategies in Working
Tables 3, 6, 7.
Contextual
Rationality
The philosophical principle that focuses on a
particular aspect of an analysis should not interfere
with the overall goal of the analysis.
23
Useful in determining when examination of the
evidence in a data category is to be considered
reasonably complete (Working Tables 2, 4).
63
-------
REFERENCES (PART I)
1. B. Russell, "Philosophy and Politics," in Unpopular Essays, Simon and Schuster,
NY, 1950.
2. K. Popper, The Open Society and Its Enemies, Routledge and Kegan Paul, 1951.
3. This point is raised by W. Berkson, "Skeptical Rationalism" in Rationality: The
Critical View, ed. by J. Agassi and I. Jarvie, Martinus Nijhoff Publishers,
Dordrecht, p. 27, 1987.
4. M. Cohen, Reason and Nature, Free Press, Glencoe, Illinois, 1953.
5. W. Newton-Smith, The Rationality of Science, Routledge and Kegan Paul, London,
1981.
6. J. Rouse, Knowledge and Power, Cornell University Press, Ithaca, NY, 1987.
7. W. Mills and D. Egan, Jr., "Risk Assessment and Policy," in Environmental Radon,
ed. by C. Cothern and J. Smith, Jr., Plenum Press, NY, pp. 273-295, 1987.
8. M. Friedman, Rational Behavior, University of South Carolina Press, Columbia,
SC, 1975.
9. K. Irani, "Introduction: Modes of Rationality," in Rationality in Thought and
Action, ed. by M. Tamry and K. Irani, Greenwood Press, NY, pp. xi-xx, 1986.
10. R. Wollheim, "Wish-fulfillment," in Rational Action, ed. by R. Harrison,
Cambridge University Press, Cambridge, pp. 47-60, 1979.
11. B. Fischoff, P. Slovic, S. Lichtenstein, S. Read and B. Combs, "How Safe is Safe
Enough? A Psychometric Study of Attitudes Towards Technological Risks and
Benefits," Policy Sciences, 8, 127-152, 1978.
12. J. Cook and D. Egan, Jr., "Mitigation," in Environmental Radon, ed. by C.
Cothern and J. Smith, Jr., Plenum Press, NY, pp. 249-272, 1987.
13. D. Hume, A Treatise of Human Nature, ed. by L. Selby-Bigge, Clarendon Press,
Oxford, 1888.
14. N. Maxwell, From Knowledge to Wisdom, Basil Blackwell, Oxford, 1987.
15. D. Hamlyn, "Empiricism," in The Encyclopedia of Philosophy, MacMillan
Publishing Co., NY, pp. 499-505, 1967.
16. M. Scriven, Primary Philosophy, McGraw Hill Book Co., NY, pp. 10-49, 1966.
64
-------
REFERENCES (PART I) (continued)
17. See, for instance, the discussion in A. Maclntyre, Whose Justice? Which
Rationality?, University of Notre Dame Press, Notre Dame, Indiana, 1988.
18. See the discussion of Bertrand Russell in Encyclopedia of Philosophy, McGrawhill,
New York, pp. 417-430, 1974.
19. See, for instance, the discussion in S. Nathanson, "The Classical Ideal," in The
Ideal of Rationality, Humanities Press International, Atlantic Highlands, NJ, pp. 3-
14, 1985.
20. This point is brought out, for example, on p. 35 of H. Brown, Rationality,
Routledge, London, 1988.
21. H. Brown, Rationality, Routledge, London, 1988.
22. J. Agassi, "Theories of Rationality" in Rationality: The Critical View, ed. by J.
Agassi and I. Jarvie, Martinus Nijhoff Publishers, Dordrecht, pp. 249-263, 1987.
23. I. Kant, Groundwork of the Metaphysic of Morals, trans, by H. Paton, Harper and
Row, NY, 1956.
24. This point is raised in J. Agassi, "Theories of Rationality" in Rationality: The
Critical View, ed. by J. Agassi and I. Jarvie, Martinus Nijhoff Publishers,
Dordrecht, pp. 249-263, 1987.
25. J. Locke, An Essay Concerning Human Understanding, reprinted by Dover
Publications, Inc., NY, 1959.
26. D. Hume, An Enquiry Concerning Human Understanding, ed. by L. Selby-Bigge,
Clarendon Press, Oxford, 1902.
27. J. March, "Bounded Rationality, Ambiguity, and the Engineering of Choice," The
Bell Journal of Economics, pp. 587-608, 1978.
28. G. Anscombe and P. Geach, Descartes: Philosophical Writings, Edinburgh, 1954.
29. Discussions on this topic are numerous, and the reader might consult H. Putnam,
Reason, Truth and History, Cambridge University Press, Cambridge, 1981.
30. K. Popper, The Logic of Scientific Discovery, Hutchinson, London, 1959.
31. For basic review and references, see "Scientific Method and Statistics" in
Encyclopedia of Statistical Sciences, Vol. 8, John Wiley & Sons, NY, 1981.
32. For basic review and references, see "Bayesian Statistics" in Encyclopedia of
Statistical Sciences, Vol. 2, John Wiley & Sons, NY, 1981.
65
-------
REFERENCES (PART I) (continued) -
33. T. Kuhn, The Structure of Scientific Revolutions, University of Chicago Press,
Chicago, 1962.
34. T. Kuhn, The Essential Tension, University of Chicago Press, Chicago, 1977.
35. P. Feyerabend, Against Method, New Left Books, London, 1975.
36. L. Laudan, Progress and Its Problems, University of California Press, Berkeley,
1977.
37. H. Longino, Science as Social Knowledge, Princeton University Press, NJ, 1990.
38. National Academy of Sciences, "The Effects on Populations of Exposure to Low
Levels of Ionizing Radiation: 1980," National Academy Press, Washington, 1980.
39. J. Ravetz, Scientific Knowledge and Its Social Problems, Oxford University Press,
Oxford, 1971.
40. M. Polanyi, Personal Knowledge, University of Chicago Press, Chicago, 1958.
41. M. Heidegger, Being and Time, translation by Peter Hertz, Harper and Row, NY,
1962.
42. B. Barnes and D. Bloor, "Relativism, Rationalism and the Sociology of Science,"
in Rationality and Relativism, ed. by M. Hollis and S. Lukes, MIT Press,
Cambridge, MA, pp. 21-47, 1982.
43. B. Barnes and D. Edge, eds., Science in Context, MIT Press, Cambridge, MA,
1982.
44. W. Quine and J. Ullian, The Web of Belief, Random House, NY, 1970.
45. M. Bunge, "Seven Desiderata for Rationality," in Rationality: The Critical View,
ed. by J. Agassi and I. Jarvie, Martinus Nijhoff Publishers, Dordrecht, pp. 5-16,
1987.
46. L. Wittgenstein, Tractatus Logico-Philosophicus, Kegan Paul, Trench, Trubner &
Co., NY, 1922.
47. L. Wittgenstein, Philosophical Investigations, Basil Blackwell, Oxford, 1953.
48. S. Nathanson, The Ideal of Rationality, Humanities Press International, Atlantic
Highlands, NJ, 1985.
49. R. Carnap, "Testability and Meaning," in Readings in the Philosophy of Science, ed.
by H. Feigl and M. Brodbeck, Appleton-Century-Crofts, NY, pp. 47-92, 1953.
66
-------
REFERENCES' (PART I) (continued)
50. H. Simon, "Rational Choice and the Structure of the Environment," Psychological
Review, 63, 129-138, 1956.
51. N. Rescher, Rationality, Clarendon Press, Oxford, 1988.
52. W. Berkson, "Skeptical Rationalism," in Rationality: The Critical View, ed. by J.
Agassi and I. Jarvie, Martinus Nijhoff Publishers, Dordrecht, pp. 21-44, 1987.
53. M. Hesse, Revolutions and Reconstructions in the Philosophy of Science, University
of Indiana Press, Bloomington, 1980.
54. R. von Mises, Probability, Statistics and Truth, Dover Publications, NY, 1957.
55. D. Crawford-Brown and N. Pearce, "Sufficient Proof in the Scientific Justification
of Environmental Actions," Environmental Ethics, 11, 153-167, 1989.
56. A. Sen, "Informational Analysis of Moral Principles," in Rational Action, ed. by R.
Harrison, Cambridge University Press, Cambridge, pp. 115-132, 1979.
57. W. James, The Will to Believe, Dover Publications, Inc., NY, republished in 1956.
58. S. Toulmin, The Uses of Argument Analysis, Cambridge University Press,
Cambridge, 1958.
59. Committee on the Institutional Means for Assessment of Risks to Public Health,
Commission on Life Sciences, and the National Research Council, Risk
Assessment in the Federal Government: Managing the Process, National Academy
Press, Washington, D.C., 1983.
60. U.S. Environmental Protection Agency, The Risk Assessment Guidelines of 1986,
Washington, D.C., 1987.
67
-------
PART TWO
ASSESSING EVIDENTIAL SUPPORT FOR CLAIMS OF CARCINOGENICITY:
ESSENTIAL ELEMENTS AND A FRAMEWORK FOR APPLICATION
1. INTRODUCTION
1.1. Background and Objective
Scientific evidence of the last ten years increasingly suggests that carcinogens differ
according to their chemical or biological properties and that no single cancer mechanism is
universally applicable within all contexts or biological settings. Substances may differ in the
mechanism by which they bring about cellular conversions, the number of stages necessary for
producing cancer, and the particular conversions associated with a specific substance. For
example, it is widely accepted that some cancers arise from nongenotoxic origins (Barrett, 1987;
Hecker, 1984; Trosko and Chang, 1988) and that a substance may contribute to cancer
development without being a complete carcinogen itself (Barrett and Wiseman, 1987). This
suggests that the carcinogenicity of a substance may be a product of both the substance and the
conditions (context) under which it acts.
Consequently, the task of hazard identification has expanded as well, from simply
addressing whether a chemical is a carcinogen in some sense to a broader assessment of
carcinogenic potential, which includes conditions and mechanisms by which it may modify or
enhance cancer development either alone or in concert with other substances. The need for a
more detailed taxonomy is reflected in the ambiguity of the phrase "potential to increase the
incidence of cancer..." cited earlier (U.S. EPA, 1986). The term "potential" may refer to the fact
that a claim to carcinogenicity may not have been fully proven by the available evidence, or it
may refer to the fact that a substance yields an increase of cancer only under a prescribed set of
conditions. These two meanings of the term "potential" must be separated for clarity. The
68
-------
particular ends reached by more detailed taxonomies are (1) the ability to warrant a claim to
carcinogenicity through reference to conceptual understanding (Laudan, 1977), (2) the ability to
employ biophysical data other than direct observations of cancer incidence in warranting a
judgment of carcinogenicity, (3) the ability to cite the context (species, dosing regimen, etc.)
within which a substance will be capable of inducing carcinogenic effects and (4) the ability to
address the effect of simultaneous exposures to multiple substances possibly acting by different
routes in a non-additive manner (although this end requires considerations well beyond those
addressed in the present report).
The published literature related to biological aspects of cancer mechanisms is abundant
and much of it has potential implications for hazard identification or other aspects of risk
assessment. It is difficult, however, to conceptualize a methodologic framework that extracts
and integrates that potential in a rational and reasonably comprehensive manner without
imposing undue and contentious theories of carcinogenicity onto the analyst. Cancer research is
ongoing. Experimental evidence of varying strength and relevance (more on this term later) to
human carcinogenicity gives birth to new hypotheses and conjectures, resulting in turn in the
formation of new if incomplete paradigms of cancer mechanisms. But the task of identifying
environmental carcinogens cannot wait for perfect understanding. What principles, then, would
facilitate this task with maximal use of available information? Toward what objective should
those principles be directed? How is this objective related to a taxonomy of claims of
carcinogenicity?
It is assumed that the objective of hazard identification is to reach the most informed
judgment about the carcinogenic potential of a substance of interest vis-a-vis the observational
evidence and the current state of knowledge relevant to that judgment. An "informed judgment"
is taken to be one characterized by reflection on (1) the complete body of data available on a
69
-------
given substance, (2) the complete body of conceptual schemes by which the data are brought to
bear in justifying any of the possible claims, (3) the "strength" of the data and validity of
conceptual schemes and (4) the uncertainties introduced by the existence of contradictory data
and/or conceptual schemes. As discussed earlier, "carcinogenic potential" is taken to include
consideration of both the epistemic status of a claim to carcinogenicity (i.e. how well such a
claim may be supported) and the context within which a substance yields an increase in cancer.
This suggests several principles for implementation of a process leading to informed judgments
on claims to carcinogenicity: complete assembly of observational evidence, assessment of each
source of evidence for its quality (reliability) and relevance (to the task of justifying one or more
of the possible claims of carcinogenicity), and evaluation of the coherence of conclusions from
the available lines of inference.
The concepts of observational evidence and of its quality are well established in the
existing scientific literature and are not discussed at length here (more in Chapter 3). It simply
is noted that risk analysts must first establish an observational base (called the "set of
observation statements" in philosophy of science, but referred to as the "observations" or "data"
here) which the analysts determine to be sufficiently reliable to justify the observations' use in
subsequent lines of reasoning leading to claims of carcinogenicity. Clearly, no conclusion of
carcinogenicity can be stronger than the base of empirical data on which that conclusion must
ultimately rest unless non-empirical epistemic foundations are deemed an appropriate scientific
concept. The strength of this observational base increases as (1) the variety of relevant
empirical data increases, (2) the quality of each specific body of empirical data increases and (3)
the existing empirical data, taken as a whole, display coherence rather than contradictory
observations.
70
-------
The concept of relevance is less well established in the scientific literature, although it is
an important concept in logic and philosophy of science (Longino, 1990). In most cases,
observations available to the analyst are not direct observations of cancer in the human
population under conditions of interest (i.e. in the desired context). They might, instead, be
observations of DNA adducts, hyperplasia, increased cancer incidence in other species, etc. The
role of such data in supporting claims to carcinogenicity depends critically on the introduction of
premises into a line of reasoning leading from the data to the conclusion of carcinogenicity.
These premises constitute a set of background assumptions concerning the etiologic role of an
observed property that must be introduced into the analysis if the observations are to be taken
as support for any specific claim. The complete set of background assumptions (premises)
required to infer a given claim from a given observation is referred to as the relevance of the
observation statement to the task at hand. Relevance of an observation then increases as
support for the necessary set of background assumptions grows stronger. A set of observations
and the associated judgment of its relevance, taken as a whole, is referred to as the warrant
(Toulmin, 1958) for a claim to carcinogenicity. The strength of the warrant for a given claim
increases with (1) the support of the observational base, (2) the strength of the background
assumptions needed for relevance increases and (3) the degree to which the analyst explicitly
recognizes the necessary background assumptions and incorporates them into discourse and
reflection. The last requirement arises from the idea that premises must not only be true, but
must be recognized as such if an analyst is to make a claim to rationality (Alston, 1985).
Since the risk from carcinogens arises from an interaction between the carcinogen and a
biological system, the end of a complete hazard identification is to depict components of both
the carcinogen (such as structure, physical state or concentration) and the biological system
(such as physiological properties and mitotic rates) contributing to the existence of a hazard.
71
-------
This end is formalized in the potential claims to carcinogenicity noted previously. By
subdividing the claims into categories related to mode of action and context within which the
action takes place, the analyst aids the end of risk mitigation discussed earlier.
In summary, the current report assumes that a rational framework for hazard
identification (and, hence, an "informed judgment" of carcinogenicity) requires the following
components:
(1) Consideration of the total body of evidence to be used in warranting depictions
of risk. For example, the evidence might be data from in-vitro assays,
epidemiologic studies, measurements of mitotic rates, etc.
(2) A judgment that the evidence invoked by the analyst is an adequate
representation of the full body of available evidence (i.e. constitutes the "hard
look" as discussed in the second interim report).
(3) Assessment of the foundational quality (see the appendix) of each separate piece
of evidence used in the analysis. In other words, this is a determination of
whether a given piece of evidence is considered reliable as an observational claim
(an issue separate from that of the relevance of the evidence).
(4) Reflection on the relevance of each piece of evidence to each depiction of risk.
This includes consideration of all relevance strategies (such as scientific theories
constituted by etiologic premises) hypothesizing a role played by the measured
factor in the production of risk (here, the risk of cancer).
(5) Judgment of the epistemic status of a given claim of carcinogenicity. This status
reflects the foundational quality of the evidence; the relevance of that evidence to
the stated task as specified by each relevance strategy; the degree of coherence
between evidence that strengthens and weakens claims made in the analysis; the
evidential support for any relevance strategy and its body of background premises
used in the analysis; and philosophical reflection on the nature of evidence and
the relationship between this nature and assignment of epistemic status.
12. The Need for Rational Discourse
Formal (mathematical) decision rules (Haseman, 1990; Eddy, 1989; Eddy et al., 1990b)
may be helpful at interim steps in assembling and assessing observational evidence, but such
approaches do not eliminate the need for human judgment and discourse for rational decision
72
-------
making. First, the concept of probability typically employed in mathematical approaches can be
formalized best by consideration of a single piece of evidence (or homogenous body of evidence)
and its relationship to a single claim. While such considerations are useful in judging the
foundational epistemic status of each single piece of observational evidence (such as the
probability that the true mean of an underlying sampled population is X), probabilistic
approaches are not well developed for instances of judging the coherence between multiple,
dissimilar and potentially incommensurable bodies of evidence. Both foundational and
coherence issues underlie scientific discourse and justification (Laudan, 1977). The second
concern with relying on formal mathematical tools is the contention of the authors that rational
judgments neither require an algorithm of belief (Brown, 1988) nor are properly summarized
through mathematical probabilities.
A key assumption is that rationality is linked most intimately to discourse concerning
epistemic status, with this discourse being guided by (but not formalized by) principles of
evidential reason. As stated by Bernstein (1983), "central to this new understanding is a
dialogical model of rationality that stresses the practical, communal, character of this rationality
in which there is choice, deliberation, interpretation, judicious weighing and application of
universal criteria, and even rational disagreement about which criteria are relevant and most
important." This discourse is typified by the deliberations of a Science Advisory Board
Committee and the attending parties. Such discourse does not prevent mathematical
formalization when it captures the full quality of the discourse, but the burden is on formal tools
to display their utility rather than on the discourse to fit into the axiomatic base of the formal
tools. The intent of this report is to present an integrative framework for discourse on rational
justification of hazard identification against which all formal tools may (if so desired) be
compared.
73
-------
To summarize, this research is concerned with the epistemic link between evidence,
relevance strategies and claims about carcinogenicity, as well as the rational basis for discourse
concerning this link. The term "claim" is chosen deliberately to distinguish it from "belief'.
Having provided the claim and its associated epistemic status, there remains an important issue
as to when a claim, characterized by a given warrant, is to be elevated to the status of a belief by
some person or group. This latter issue, while important in completely justifying the assertion
that beliefs of an agency are rooted in rationality, is not addressed here since it involves
components of psychology, sociology and political responsibility falling outside the domain of the
present research. The present document focuses, therefore, on the evidential support for a
claim (such as the claim that a substance is or is not a carcinogen), but not on the process by
which claims are translated into beliefs. This presumes that it is possible to speak of epistemic
status in a non-normative manner (i.e. without reference to the manner in which epistemic
status should be related to the adoption of a belief). Complete rationality in policy matters, of
course, requires that this latter issue of translation be discussed by any "users" of the claims
generated in the risk analysis.
1.3. Elements of the Framework
A claim regarding carcinogenicity of an agent is the product of human judgment applied
to evaluation of the informational base. There are, however, multiple sources of observational
evidence, claims to carcinogenicity, and strategies of relevance for linking the two. The
procedure to be described in this report was developed around certain principles and premises,
such as completeness of evidence relevant to hazard identification, incorporation of qualitative
as well as quantitative characteristics of evidence, a rational basis for warranting claims of
74
-------
carcinogenicity from evidence, assessment of coherence of claims, and the centrality of informed
human judgment to decision-making.
These characteristics are derived from consideration of the epistemic status of a
depiction of risk (here, a claim of carcinogenicity), which is related to both the observational
data available to the analyst and the relevance of these data to each of the specific taxonomic
claims of carcinogenicity. This suggests that the epistemic status of a claim of carcinogenicity is
determined by three judgments related to each piece of observational information:
(1) What is it that has been observed and how well has this observation been
established? This judgment encompasses the first three of the five framework
components listed in Section 1.1.
(2) What is the relevance of the observation to each specific claim? This judgment is
given by the fourth component listed in Section 1.1.
(3) In what way does this observation support and/or detract from the claim that the
substance is a carcinogen in any sense? This judgment is given by the fifth
component listed in Section 1.1.
The first issue is related to the attempt to define the observational base on which any
claims made in the analysis will (in some manner) be warranted. The analyst must establish the
degree to which each claim to observation is considered well founded, regardless of the use to
which the observation subsequently is put in the analysis. In the terms of logical positivism, the
intent here is to produce "observation sentences" (Newton-Smith, 1981) from which the claims of
the final analysis may be constructed when employed in conjunction with background premises
required for relevance. These observation sentences should encompass all observations available
on a given substance (more on this below). For simplicity in this report, the term "observation
sentence" is replaced by the term "observation", although the two terms have distinctly different
meanings in epistemology.
75
-------
The second issue is related to the definition of relevance of the observations to the task
of warranting a claim of carcinogenicity (i.e. that the substance is a carcinogen in some sense).
As described in the appendix, observations gain relevance through the use of background
premises relating the observation to a claim of carcinogenicity. For example, observation of
DNA adducts gains relevance through the premise that adducts are a first step in the process of
neoplastic conversion (Perera, 1987; Craig et al., 1981; Belinsky et al., 1987c), or are correlated
with conversion (examples of relevance strategies, formalized in Section 5.1.). While multiple
observations may be used to warrant a given claim, it also is the case that multiple sets of
background assumptions may be used to assign relevance to a given observation. The analyst
must, therefore, establish the base of strategies for relevance by which observations are folded
into the analysis.
The framework to be described in the following chapters can be implemented on a
substance of interest by following the seven steps for its application given in Chapter 6. To
develop the framework, however, we need to consider: (1) determination of a taxonomy of
claims of carcinogenicity, (2) the collection and assessment (both qualitative and quantitative) of
the strength of observational evidence, (3) the assessment of the relevance and coherence of
evidence for warranting claims of carcinogenicity within the observational context of that
evidence, i.e. under the actual conditions of observation, particularly with respect to species and
exposure level for data on humans or animals ("context" is formerly defined in Section 4.1.), and
(4) the same as (3) except with extrapolation from one context to another, such as exposure to
humans instead of animals or exposure at environmental levels instead of at higher
concentrations. Stated briefly, we need to address four basic questions: What is meant by
"carcinogenicity"? What observational data should be initially collected and assessed for their
utility in judging carcinogenicity? How should the evidence be interpreted for hazard
76
-------
identification within the observed context (i.e., without extrapolation across doses or across
species, assuming the availability of data from at least one epidemiologic study or long-term
animal study)? What interpretation can be made with extrapolation (from high exposure to low
exposure or from exposure of animals to exposure of humans)?
These four questions are addressed in Chapters 3-5, following discussion of some
fundamentals regarding cancer mechanisms and a "biologically significant dose rate" in the next
chapter. The broad theoretical discussion of Chapter 2 constitutes a "minimal" set of etiologic
assumptions to be imposed on the analysis. This minimal set provides structure without forcing
judgments into the boundaries of a particular theory of carcinogenesis. In other words, existing
theories tend to share this common axiomatic base, allowing the resulting discourse to remain
fairly "theory-neutral" until explicit relevance strategies are invoked. It should be recognized,
however, that the existence of any "theory-neutral" framework is highly controversial in the
Literature on epistemology and philosophy of science (the classic text here is Kuhn, 1962).
2. CANCER MECHANISMS AND BIOLOGICALLY SIGNIFICANT DOSE
Two central concepts employed in causal theories of carcinogenesis are those of (1) a
biologically significant dose-rate (BSDR) within the body (Andersen, 1989; Gerlowski and Jain,
1983; Lutz and Dedrick, 1987) and (2) transitions between states of cancer brought about by the
BSDR (Moolgavkar, 1991). The BSDR is produced through a series of physical steps related to
the following conceptual categories (see Figure 1):
(1) Exposure to the substance in the environment produces an intake into the body
by various routes (via the lungs, G.I. tract, or skin).
(2) The intake results in an uptake into the body, if the substance is deposited in the
body following intake. This distinction between intake and uptake is required by
the fact that substances can, for example, be inhaled without depositing in the
lung (such as in the case of inert gases).
77
-------
Figure 1. Flow Diagram of Judgments for BSDR
Burden occurs
Uptake occurs
Exposure occurs
BSDR occurs
Intake occurs
Significant burden occurs
Necessary activation or
de-activation to biologically
significant form of the
substance occurs
Substance is retained in
the tissues, organs and/or
cells of the body
Biologically significant form
of substance is capable of
interacting with biological
structures (targets)
Inhalation, ingestion and/or
dermal absorption occurs
Substance is absorbed into
tissues, organs and/or
cells of the body
78
-------
(3) The uptake, in conjunction with retention of the substance in the body, yields a
burden (i.e. concentration of the substance in the body). Increases in either
uptake or retention generally yield increases in burden. The distinction between
uptake and burden is important because a substance may be taken into the body
but immediately removed, not allowing time for interaction with whatever tissue,
cell, organ, etc., constitutes the target for the effect. In addition, etiologic
theories of carcinogenesis may posit a threshold burden below which cancers do
not appear even if an uptake occurs.
(4) Burden refers to the amount of the original substance present in the body. For
many carcinogens, biotransformation (metabolism) changes the chemical form of
the original substance (Whitey, 1982; Vainio and Hietanen, 1980; Clayson, 1985;
Gehring et al., 1978; Farber, 1987; Miller and Miller, 1976). This new chemical
form may be either (1) the form responsible for cancer, in which case the
biotransformation results in activation or (2) a form incapable of inducing cancer,
in which case the biotransformation results in deactivation. In any event, theories
of carcinogenesis presuppose that there are some chemical or physical forms of a
substance capable of inducing cancer and other forms incapable of inducing
cancer. The former is termed here the biologically significant form. The burden
of this biologically significant form of the original substance in the body is termed
here the biologically significant burden. The distinction between this B-S-B and
the burden is important because a substance may be present in the body without
necessarily being present in the form capable of inducing cancer. This can occur
if enzyme systems required for activation are not present or if systems required
for deactivation transform all of the biologically significant substance into an
inactive form.
(5) Finally, the biologically significant burden (BSB) produces a biologically
significant dose rate (BSDR) if the biologically significant form of the substance
is capable of interacting with biological structures in the body (DNA, membranes,
etc.) believed to be the site of action for carcinogenesis (Andersen, 1989; Barrett
and Wiseman, 1987; Farber, 1987). The distinction between BSB and BSDR is
important because a substance may be present in the body but unable to interact
due to the presence of barriers to interaction or a lack of important sites of
interaction.
The above five considerations of the process leading to a biologically significant dose-rate
introduce premises (to be discussed in more detail in Chapter 5) necessary in ensuring that
exposures in the context of interest to the analyst result in at least some BSDR. We turn now
to the issue of changes in states of cancer as induced by this BSDR. In the broadest and least
theoretical sense, it might simply be premised that exposure to a substance (and, hence,
production of a BSDR) induces a change from "normal" health to cancer. The probability of
79
-------
this change presumably depends upon a background rate of change, the BSDR, and the length
of time over which the BSDR is maintained in the body. A more general way of stating this is
that the change depends upon the temporal pattern of the BSDR over the lifetime of an
exposed organism.
Current theories of carcinogenesis differentiate between a number of stages or states (by
"state" here, we do not necessarily mean a morphologically, physiologically, etc., identifiable
condition; different theories of carcinogenesis identify states differently and may even drop the
concept of a state) intermediate between "normal" and cancerous. It generally is accepted that
these states consist of at least neoplastic conversion and neoplastic development as depicted in
Figure 2 (Williams and Weisburger, 1991). Cells are presumed to pass from normal to
converted and finally from converted cells to developed tumors. Only cellular clusters
developing to a frank tumor possess the capability of inducing fatality as a result of the cancer.
Other cells or cellular clusters are considered to be precancerous. Changes from one state to
another typically are referred to loosely as transitions. (By "transition" here, we do not
necessarily mean a distinct stochastic change of state, but rather a general historical movement
towards one of the states of cancer; different theories of carcinogenesis treat transitions
differently, with individual theories focusing on transitions as stochastic events, alterations in
kinetics, etc.). Identification of the transitions as initiation, promotion and progression are
common in the literature (Pitot and Campbell, 1987; Barrett and Wiseman, 1987) but these
terms are more operational than mechanistically descriptive and are not included here as part of
the "minimal" theoretical base.
The above discussion of theories of carcinogenesis focuses on changes or transitions
without giving explicit mechanisms for those changes. While there is general agreement that
cancer is a multi-stage phenomenon, there is less (although still substantial) agreement that the
80
-------
Figure 2. Main Steps in the Carcinogenic Process
Neoplastic Development
Chemical Carcinogen
DNA Reaction
Epigenetic Effects
DNA Alteration
Expression
Neoplastic Cell
Promotion
Progression
Neoplasm
Adapted from Figure 5-1 of Williams and Weisburger
(1991).
81
-------
stages should be identified as neoplastic conversion and neoplastic development. There is even
less agreement as to the mechanisms by which these stages arise and whether there are
mechanisms common to carcinogens.
In the case of neoplastic conversion, the dominant etiologic theory relates conversion
either to changes in DNA (Lawley, 1987; Farber, 1987; Williams and Weisburger, 1991),
referred to as genotoxicity, or to epigenetic changes (Bartsch and Malaveille, 1990; Barrett,
1987; Butterworth, 1990; Perera, 1984), referred to (more generally) as non-genotoxicity. The
former changes are unclear at present, but candidates for important change are: (1) formation
of DNA adducts (Belinsky et al., 1987c; Farber, 1987), (2) changes in base-pair sequence, (3)
single or double-stranded DNA breaks (Slaga, 1988; Tennant et al., 1987a), (4) chromosomal
aberrations (Lewis and Adams, 1987; Hall and Freyer, 1991), and (v) activation of oncogenes
(including deactivation of repressor genes) (Stowers et al., 1987; Aaronson and Tronick, 1986;
Bos, 1988). Associated with the above four mechanisms are concepts of misrepair (through
improper insertion of bases) of adducts and DNA breaks (Curtis, 1991) and translocation of
broken DNA to allow expression of oncogenes (Della-Favera et al., 1988; IARC, 1986).
Epigenetic (non-genotoxic) changes refer to alterations in cellular structures other than DNA,
such as membranes and antigens, which might, in turn, yield changes to DNA (although DNA
changes are not necessarily part of neoplastic conversion).
In the case of neoplastic development, mechanistic premises have tended to focus on
intercellular communication (Trosko and Chang, 1988; Harper and Legator, 1987; Langenbach
et al., 1988; Slaga, 1984), hormonal control (Moolgavkar, 1986; Williams, 1990) and the kinetics
of growth and removal within organized cellular communities (Homburger and Treiger, 1969;
Loury et al., 1987; Swenberg and Short, 1987; Cohen and Ellwein, 1990; Short and Swenberg,
1990). Changes in these properties might arise from either genotoxic or epigenetic action,
82
-------
although the latter usually is cited as the route of action. There is little agreement as to
whether these changes are potentially reversible although it has been suggested that promotion
has both a reversible (Phase I) and irreversible (Phase II) component (Barrett and Wiseman,
1987; Hermo and Brandt-Rauf, 1987; Langenbach et al., 1988). As in neoplastic conversion, the
largest source of disagreement is over premises concerning mechanisms of action. Candidates
are (1) changes in DNA (Bailey et al., 1991), making neoplastic development similar to
conversion, although this aspect of promotion probably is limited to Phase II; (2) interference
with intercellular or intracellular communication through changes in messenger molecules, gap
junctions, microtubule structure, etc. (Trosko and Chang, 1988); (3) disruption of the histological
architecture of cellular communities as in the theory by (Tamplin and Cochran, 1974); (4)
induction of hyperplasia in the cellular community through stimulated mitosis or other (currently
unspecified) means (Ames and Gold, 1991a; Schulte-Hermann et al., 1991); and (v) changes in
the action of hormones on cells through changes in the amount of hormone present at the target
organ, the form of the hormone, the density of receptors for hormones and/or the specificity of
hormone receptors (Moolgavkar, 1986).
3. MEANING OF "CARCINOGENICITY"
3.1. Current Classifications of Carcinogenicity
The criteria for chemical group classifications currently are ranked according to the
supporting evidence on changes in tumor prevalence in humans and animals, with human
evidence considered more relevant. For example, in the classification system used by the EPA,
the taxonomy is: Group A: Human carcinogen, with sufficient evidence from epidemiologic
studies; Group Bl: Probable human carcinogen, with limited evidence from epidemiologic
studies; Group B2: Probable human carcinogen, with sufficient evidence from animal studies and
83
-------
inadequate evidence or no data from epidemiologic studies; Group C: Possible human
carcinogen, with limited evidence from animal studies in the absence of human data; etc. (U.S.
EPA, 1986). Classification schemes of the International Agency for Research on Cancer
(IARC), the National Toxicology Program (NTP), and the American Conference of
Governmental Industrial Hygienists (ACGIH) also have this characteristic; i.e., at the upper end,
the strength of the claims is dependent on the extent of evidence from human epidemiologic
studies, proceeding to weaker claims associated with evidence of changes in tumor prevalence
from animal studies. The single exception is an "indirect" classification of ACGIH which refers
to carcinogenicity activity that occurs primarily as secondary effects of some other toxic or
physiological action by the substance or its metabolites (Cohrssen and Covello, 1989), although
even here the evidence for such activity is change in tumor prevalence.
Several properties in common to these agency's classification criteria may be noted: (1)
The strength of the claim to human carcinogenicity is limited by the kinds of data available
rather than reflection upon the complete body of data on a suspect substance, e.g., an agent
could not be classified strongly as a "human carcinogen" (Group A or B) by EPA if there were
no epidemiologic data available, regardless of how convincing a case might be made on the basis
of other evidence and scientifically warranted lines of reasoning. This suggests the existence, at
least implicitly, of some conception of "minimal epistemic status" (see the discussion in the
appendix) associated with specific warrants. (2) The criteria do not state, but give the
impression, that if available in adequate quantity and quality, then animal data alone are
sufficient for classification of a substance as a "probable human carcinogen". The manner in
which animal data play such a role, however, is not specified, nor are the criteria specified by
which premises necessary for inter-species extrapolation are to be warranted. (3) "Human
carcinogen" suggests that if humans differ in susceptibility to a carcinogen, it is only by degree,
84
-------
i.e., that it is not possible for an agent to be a human carcinogen only to a subpopulation that
may be predisposed in some way. (4) There is no reference to the context within which a
substance acts as a carcinogen, requiring the (perhaps unwarranted) premise that carcinogenicity
is a property of the exposure factor alone rather than an interaction between exposure to the
factor of interest, biological properties of the exposed organism, and concurrent exposures.
These four features of existing classification schemes presume that a substance may be
categorized only as a "carcinogen" with epistemically-based modifiers (possible, probable, etc.)
depending upon the category of evidence available.
With its focus on carcinogenicity as a single claim and on epidemiologic and whole-
animal data, the existing classification scheme does not make explicit the etiologic differences
which underlay various carcinogens and the manner in which data other than whole-animal
carcinogenicity assays or epidemiologic studies may (1) elucidate these differences and (2)
strengthen or weaken claims to carcinogenicity within a given context. The framework of
analysis developed here extends the existing procedures for hazard identification to include the
above considerations.
32. A Taxonomy of Claims of Carcinogenicity
In the present report, three assumptions are made for a claim of carcinogenicity within a
framework for rational discourse on hazard identification: (1) the claim is manifold (i.e. has
different taxonomic forms useful for meeting different ends of risk mitigation); (2) the claim is
dependent on the nature and strength of the full body of supporting evidence on the agent
considered; (3) the claim should be formulated uniquely in each specific case. This suggests that
hazard identification for carcinogens might best be served by a taxonomic scheme giving explicit
recognition to the role of a substance in carcinogenesis and, hence, the context under which the
85
-------
substance increases the prevalence of cancer. Implicit to any claim of carcinogenicity is a
context, e.g., long-term animal exposures, humans exposed in an occupational setting, etc.
Consequently, the taxonomic scheme is applied to an agent/context combination, even when
explicit reference to the specific context is omitted. The uncertainty in claims of carcinogenicity
then will be a function of the context towards which those claims are directed.
The taxonomy employed in the present report is described in Table 1 and is depicted in
Figure 3. At the first and least informative level, the risk analyst might make the claim simply
that a substance increases the incidence of cancer (within a specified context). At the second
level, the analyst might make the claim that the substance is either a direct carcinogen {partial or
complete) or indirect carcinogen {mixing agent or helping agent). These terms are defined in
Table 1, which also describes conditions under which the claim would be preserved in another
context. The distinctions between "direct carcinogen", "mixing agent (mixer)", and "helping agent
(helper)" are important. A direct carcinogen (the parent agent or a biologically active product)
induces one or more transitions between states or "stages" in the cancer process by itself,
without requiring the presence of another chemical agent. If it induces all the necessary
transitions leading to a neoplasm, then it is called complete; otherwise it is called partial. For
example, formaldehyde, radiation, BaP, and most other substances that are commonly called
carcinogenic are complete carcinogens. Some examples of incomplete carcinogens are
urethane, which is an effective initiator but not a promotor (Berenblum and Haran-Ghera, 1957;
Barrett and Wiseman, 1987); phorbol esters, which are active promoters without initiating
activity (Boutwell, 1974; Barrett and Wiseman, 1987); and probably arsenic, which appears to act
primarily to increase progression (Barrett, 1984). A substance is a mixing agent if it is not a
direct carcinogen, but a direct carcinogen is produced when it is mixed with one or more
suitably chosen substances that also is not a direct carcinogen, i.e., if a mixture of substances,
86
-------
none of which is a direct carcinogen itself, interact biologically or chemically to effect transitions,
then the substances are called mixing agents. Note that by definition a mixing agent must be
able to produce a direct carcinogen from interaction with some other mixing agent, not with a
substance that is already a direct carcinogen (either complete or partial). If a substance is not a
direct carcinogen or a mixing agent, but it enhances one or more transitions induced by a
suitably chosen direct carcinogen, then it is called a helping agent. A helping agent is sometimes
referred to as a "risk modifier" in the literature. For examples of
87
-------
TABLE 1. CLASSIFICATION SCHEME UTILIZED IN THE PRESENT REPORT
Classification
Definition
Antecedent Conditions Under Which Substance Elevates
Cancer Incidence
Complete
Carcinogen
The substance, acting directly on components of the
organism, induces all changes required to elevate the
incidence of cancer in a population.
Any conditions in which the mechanism by which the
substance exerts its effect is present and operational.
Partial
Carcinogen
The substance, acting directly on components of the
organism, induces part of, but not all of, the changes
required to elevate the incidence of cancer in a
population.
Any conditions in which (i) the mechanism by which the
substance exerts its (partial) effect is present and operational
and (ii) the remaining changes required for cancer are
produced in the population by other means.
Mixer
The substance, only when acting in conjunction with a
second (mixer) substance, acts directly on components
of the organism to induce part of (partial mixer) or
all of (complete mixer) the changes required to
elevate the incidence of cancer in a population.
Any conditions in which (i) the necessary second substance is
present, (ii) the mechanism by which the two substances act in
conjunction to exert their effect is present and operational
and (iii), for partial mixers only, the remaining changes
required for cancer are produced in the population by other
means.
Helper
The substance, while not producing any of the
changes required to induce cancer, produces changes
in the antecedent conditions under which a second
(carcinogen) substance exerts its effect, thereby
magnifying the elevation in cancer incidence caused
by the second substance.
Any conditions in which (i) the second necessary substance is
present, (ii) the mechanism by which the original substance
exerts its magnifying effect is present and operational, and
(iii) the mechanism by which the second substance exerts its
(carcinogenic) effect is present and operational.
88
-------
Figure 3. Levels of Taxonomic Claims
of Carcinogenicity
Classification
Stage(s)
Mechanism(s)
Substance increases the
incidence of cancer.
Substance is a direct carcinogen
[Complete j j Partial
| Carcinogen | j Carcinogen
Substance induces
Neoplastic j j Neoplastic
Conversion ] } Development
Neoplastic
Conversion
Substance modifies
Neoplastic
Development j
Substance is an indirect carcinogen
Mixing
Agent
Agent
Genotoxic
Substance modifies
| Non-genotoxic
Genotoxic
Substance induces
transitions through
Non-genotoxic
Mechanisms
89
-------
helping agents, with a limited discussion of possible mixing actions, the reader should consult
Williams (1984), Berenblum (1985), Weisburger (1988), Hecker (1984) and Homburger and
Treiger (1969).
Returning to the description of the taxonomy, at the third level in the taxonomy the
analyst might differentiate between effects on neoplastic conversion, neoplastic development, or
both steps to cancer. At the fourth level, the analyst might make the claim that the substance
induces alterations in the organism through genotoxic and/or non-genotoxic routes of action.
4. INFORMATIONAL BASE
4.1. Assembling the Observational Base
Having established the taxonomy of claims that constitute the end of an analysis, the
initial step in analysis is establishment of the base of observations on which warrants of a claim
to carcinogenicity might be developed. This step establishes, to the extent possible, the
existential content on which all parties engaged in an analysis (or debate concerning an analysis)
might agree regardless of differences in interpretation of the observations. The existence of
such a "theory-neutral" base is controversial (Longino, 1990), but the step is included here since
there is substantial agreement within the field of carcinogenesis concerning the importance of a
number of specific observations to claims of carcinogenicity (even if the interpretation of the
observations varies between analysts). The present section describes the sources of information
potentially available to an analyst when faced with the task of hazard identification. The
collection and preliminary evaluation of all observational data of potential value in assessing
carcinogenicity of a specific agent is a major undertaking. To facilitate organization and
application in the suggested framework developed here, observational information is classified
by data category and data item, as illustrated in Table 2. A data category refers to a set of
90
-------
observations characterized by a common use in lines of reasoning (such as supporting a common
background premise). A data item refers to a specific observation within the data category.
The category "Related Substances Assessments" refers to hazard identification of other
agents, including mixtures, that might have some predictive value for the agent of interest
through similarities in the etiologic role in carcinogenesis. For example, ETS is largely the
product of sidestream smoke produced by a smoldering cigarette between "puffs". Sidestream
smoke has been shown to be qualitatively similar in chemical constituency to the mainstream
smoke that smokers inhale, containing several known or suspected carcinogens (U.S. EPA,
1990). Additionally, observational evidence strongly supports the conclusion that smoking
increases the risk of lung cancer. Thus, for hazard identification of ETS, prior assessments of
mainstream smoke provide information (from a related substance) useful in the line of
reasoning referred to in the literature on ETS as "cigarette equivalence" (Thorslund, 1990).
The broad theoretical implications of the data categories in Table 2 are described in
Table 3. The categorization shown in Table 2 was constructed to group data items supporting
similar lines of evidential reasoning for claims of carcinogenicity (to be discussed in Chapter 5).
Assessing the evidential support for carcinogenicity must also take into account (1) the context
in which data were observed (e.g., a long term animal study on mice or an epidemiologic study
of workers exposed at high concentrations) and (2) the context in which hazard identification is
of interest (e.g., persons exposed at typical environmental levels). This raises the issue of what
is to be meant by a context. In general, a context refers to a set of characteristics of an
exposure situation believed to affect the carcinogenicity of the substance (or exposure factor) of
interest. More formally, a context is an exposure-response scenario with specification of factors
present that may affect the assessment of hazard. Observations within a given context then are
assumed to share a common etiologic link between exposure and response, so that each
91
-------
TABLE 2. SUMMARY OF THE ROLE OF OBSERVATION STATEMENTS IN
EVIDENTIAL REASONING FOR CARCINOGENICITY CLAIMS
References Citing Role of
Specific
Data Category1
Data Item1
Data Item
Judgment3
Tumor
Incidence or Prevelance
15 120 221 247 248
C
Response2
Dose-Response
2 3 15 66 120 166 215 233 252
C
Time-to-Appearance
24 120
C
Multiplicity
120
C
Age-at-Appearance
120
C
Initiation
2 17 31 100 101 140 211 293
S
Promotion
2 17 30 31 53 67 100 211
S
Biophysical
Hyperplasia
6 8 66 173 175 247 248 252
PT
Effects
DNA Adducts
18 22 34 42 63 64 68 69 241 247
PT
Oncogene Activation
2 22 31 52 124
PT/T
Interference with
Intercellular
Communication
32 125 126 169 231 261
PT/T
Degree of Metastasis
81
T
Concentration of Tumor
Growth Factor (TGF)
50
PT
DNA Breakage
54 88 100 101 124 173 250 289
PT
Chromosomal Aberrations
54 69 88.100 101 124 250
9 69 88 100 101 124 250
PT/T
Site-Specific Mutation
9 69 88 101 107 124 202 214 226
PT/T
Mutagenicity
33 88 100 101 124
PT/T
Cellular Transformation
PT/T/S
Alterations in Membrane
Permeability
230
171
PT
RNA Alterations
PT
Appearance of Cancer
81 176 215 284
Marker Proteins
T/S/C
Alterations in Cellular
Antigens
81
T/S
Presence of Preneoplastic
29 34 151 231 284
T/S
Lesions
Alterations of Cellular
254
Architecture
PT
Alterations of Distribution in
Differentiation and/or
103
8 67 85 124 224 232 247 252 253
Histology
PT/T/S
Cytotoxicity
PT
Tumors Appearing in Hosts
after Injection of
Transformed Cells
81
S/C
Hormonal Alterations:
67
67
Hormone Production Rate
PT
Hormone Structural Form
67
PT
Hormone Binding Sites
PT
(continued on following page)
92
-------
Table 2. (continued)
References Citing Role of
Specific
Data Category1
Data Item1
Data Item
Judgment3
Host
DNA Repair Rates
85 88 96 223 241
PT
Characteristics
DNA Repair Specificity
Density of DNA Repair
223
PT
Enzymes
223
PT
Repair Kinetics
85 88 146 223
PT
Activation/Inactivation of
Repair Process
223
PT
Background Transition Rates
Initial State Vector
35 67 85 134
67 85 134
PT
PT
Presence of Target Organ
35 96 120
PT
Presence of Carcinogenicity
Mechanism
35 96
PT
Pharmaco-
Rate of Inhalation:
dynamics
Tidal Volume
3 11 105 106 122 182 208 222 247
EI
Minute Volume
Rate of Ingestion:
3 11 105 106 122 182 208 222 247
EI
Food
82
EI
Water
15 197
EI
Airway Diameters and Lengths
105 106 115 200
IU
Airway Branching Scheme
105 115 200
115 200
IU
Deposition Fractions
IU
Epithelial Integrity
7 149 183 245
IU
Air:Blood Partition
13 15 16 158 183 229
IU
Integrity of Mucus Layer
247
UB
Mucous Flow Rate
89 191 247 248
UB
Peristaltic Velocity
182 197
UB
Facilitated Transport:
Carrier Density
137 197
IU
Binding Coefficient
137 197
IU
Dumping Coefficient
137 197
IU
Specificity of Carrier
137 197
IU
Saturability of Carrier
137 197
IU
First Pass Excretion
18 281
UB/BSB
First Pass Metabolism
281
BSB
WatenOil Partition
197 208
IU
WatenLipid Partition
197 208
IU
Cardiac Output, Q
21 138 183 208 222
IU
Organ Perfusion
21 144 197
IU
Pore Size (membranes)
149
IU
(continued on following page)
93
-------
Table 2. (continued)
References Citing Role of
Specific
Data Category1
Data Item1
Data Item
Judgment3
Pharmaco-
Facilitated Transport Energy
dynamics
Source
137 149
IU
(continued)
Km for Metabolic Reaction
58 93 108 138 154 208 220 221 239
274
38 58 59 93 137 138 208 220 221
BSB
Vmax for Metabolic Reaction
BSB
Substrate Density
37 38 39 78 121
BSB
Presence of Competing
Metabolic Pathways
58 142 172 155 287
BSB
Renal Flow Rate
242 273
IU
Permeability of Renal Tubes
242
IU
Excretion Rates
18 161 221 242 273
UB
Identification of Active
Metabolite
20 37 49 93 95 120 206 220
BSB/BSDR
Identification of Target Cells
120 252
IU/BSB/BSDR
Identification of Site of
12 37 112 162 287
Metabolism
BSB
Substance Diffusion
Coefficient
149
IU/UB
Neutrophilicity
187
BSDR
Adduct Binding Coefficient
108 159
112
BSDR
Enzyme Concentration
BSB
Intake
15 65
EI
Uptake Fraction
15 65 245
IU/EI
Burden
15 65 225 245 246
UB/IU/EI
Dose Rate or Dose
15 246
UB/IU/EI
Biologically Significant Burden
15 42
BSB/UB/IU/
EI
BSD or BSDR
15 42 159 160
BSDR/BSB/
UB/IU/EI/UB
Retention Function
15 222
UB
Concurrent
State of Attachment to Other
Environmental
Substances
200 288
EI/IU/EB
Conditions
Presence of Oils in
288
Administered Dose
IU
Presence of Other Substances
in the Environment
79 114 130 156 194 215 228
PT
Structure
Bay Region Site
25 26 27 83 120 192 256
BSDR
Activity
Relationships
(continued on following page)
94
-------
Table 2. (continued)
Data Category1
Data Item1
References Citing Role of
Data Item
Specific
Judgment3
Related
see, e.g. 259
ALL
Substances
Assessments
Environmental
/Substance
Characteristics
Particle Size
Particle/Fiber Shape
Particle Hygroscopicity
Concentration Variability
Spatial
Temporal
Chemical Form
200 228
200 228
200
84
84
IU
IU
IU
EI
EI
ALL
See Working Table 2.
Quantitative measures are shown under "data items" for "tumor response" only. Quantitative
measures commonly used are not given in other data categories, only the endpoints of interest.
This refers to one or more of the specific judgments that must be made in using Figures 1
and/or 2 for supporting claims of carcinogenicity through theory-based inference. The key is as
follows:
EI: conversion from exposure to intake
IU: conversion from intake to uptake
UB: conversion from uptake to burden
BSB: conversion from burden to biologically significant burden
BSDR: conversion from biologically significant burden to biologically significant dose rate
S: indicates a state or stage of carcinogenesis
T: indicates a transition process between states or stages of carcinogenesis
C: indicates carcinogenesis directly
PT: partial transition; indicates an effect leading to (but not sufficient for) transitions
ALL: affects all of the conversions.
95
-------
TABLE 3. BROAD THEORETICAL IMPLICATIONS OF DATA CATEGROIES
Data Category1
Broad Theoretical Implications of Data Categories2
Tumor
Response2
Direct observation of the effect of interest (cancer). Intra-context claim
for carcinogenicity requires only the premise that any noted associations
are causal.
Biophysical
Effects
Provides evidence that a causal factor (such as exposure to the substance
of interest) yields effects deemed important in the etiology of cancer.
Since the observations are not of cancer directly, a theory of the etiology
of cancer is required. Such theories would support the contention either
that a given data item is (i) an indicator of cancer being present (e.g.
tumor growth factor, appearance of cancer marker proteins, alterations
in cellular antigens, degree of polyploidy), (ii) an indicator of transitions
between states of cancer (e.g. oncogene activation, interference with
cellular communication, mutagenicity, cellular transformation) or (iii) an
indicator of conditions (mechanisms) necessary for transitions (remaining
data items).
Host
Characteristics
These data provide evidence that there is not a unique pre-existing
characteristic of the organism that would preclude conclusions being
drawn with respect to the carcinogenicity context in which the organism
appears in the analysis. The intent is to ensure that an observed effect
being associated with the BSDR cannot be attributed to some unique
characteristic of the particular organism being observed. This category
does not refer to the characteristics affecting exposure and BSDR.
Pharmaco-
dynamics3
Provides evidence that the factor (such as exposure to the substance of
interest) results in a biologically significant dose-rate to the organism.
Since pharmacodynamic data do not constitute observations of effect,
they do not provide either direct empirical or semi-empirical warrants
for carcinogenicity claims. They can, however, provide conceptual
support for the claim that any effects observed in the 'Tumor Response"
and "Biophysical Effects" data category are causally connected with
exposure.
Concurrent
Environmental
Conditions
These data support the contention that there is nothing unique about the
exposure conditions that would call into question inclusion of a given
study into a given context, or extrapolation of an observed effect from
one context to another. These conditions occur prior to intake of the
substance and may affect the intake magnitude, intake route, uptake, etc.
of the substance. These data also support the contention that an
observed effect was (or was not) due entirely or partially to confounding
exposures.
Structure
Activity
Relationships
Certain chemical/physical structures have been determined to be
associated most strongly with carcinogenicity, particularly with respect to
initiation. These structural features presumably are an indication of the
ability of a substance to act on the organism by mechanisms governed
through general topological and/or chemical properties of molecules.
See Working Table 2.
(continued on following page) 96
-------
Table 3. (continued)
Quantitative measures are shown under "data items" for "tumor response" only. Quantitative
measures commonly used are not given in other data categories, only the endpoints of interest.
Quantitative data on pharmacodynamics are typically utilized in the dose-response step of risk
assessment, rather than in hazard identification. Pharmacodynamic data are included in Table
3, however, because they contribute qualitative information useful to the hazard identification
step. Specifically, it is useful to know that exposure to the substance of interest results in a
biologically significant dose-rate to the organism of interest if a claim of carcinogenicity is to be
supported.
97
-------
observation in the context may be taken as one sample from a potential population of samples sharing a
common relationship between exposure and response. This stage of specifying contexts involves
judgments founded in etiologic theories concerning the factors that might potentially exclude a given
observational setting from a given context. In the language of philosophy of science, a context is
defined by specifying the antecedent conditions under which a response is presumed to follow from
exposure (see the appendix). It may even be the case that in-vitro data are included within a context
otherwise defined by exposure to an animal or human, so long as the exposure-response characteristics
of the in-vitro system are deemed to share an etiologic link with characteristics of the organisms in the
context. This imprecise definition is intended to be more operational than descriptive, as the most one
can do is to include within the specification of a context all factors considered to be of potential
significance to the judgment of carcinogenicity.
Contexts are defined around actual or hypothetical exposures of animals or humans. If a
context describes a scenario within which study data are available, typically from either an animal study
(preferably, long-term) or epidemiologic study, then it is referred to as an observational context. A
target context describes an "endpoint" of hazard identification, including a non-observational context
requiring extrapolation of judgments of carcinogenicity across species or dose levels constituting the
observational contexts. For example, suppose there are two sets of study data on substance X, one a
controlled animal study and the other an epidemiologic study of persons exposed at atypically high
levels, perhaps due to their location or workplace environment (as in the case of exposures to airborne
radon (Cross, 1987) or formaldehyde (Graham et al., 1988)). A separate observational context may be
defined for each study (since the two studies are judged not to share a common etiologic link) and
perhaps a target context defined that is descriptive of typical environmental exposures. Alternatively
one might have one or more animal studies with no epidemiologic studies, requiring that target contexts
might be defined for human exposure (as in the case of acetaldehyde (Woutersen et al., 1986)).
98
-------
Working Table 1 is provided for a description and reference number of each observational context and
each target context.
Observational contexts may suffice without definition of a separate target context. For example,
epidemiologic studies of ETS tend to be conducted under typical living conditions, descriptive of the
context of concern for assessing hazard of lung cancer to adults or other endpoints, such as specified
respiratory effects in children. Judgment needs to be exercised in defining contexts. As a rule of
thumb, two contexts should differ in at least one characteristic of potential consequence for hazard
identification (i.e. by at least one difference in important antecedent conditions). This is an imprecise
procedure, on which judgments may differ. The conflict is between a need to summarize data for
manageability and decision-making without sacrificing useful information in the process. (Technical
note: This notion is somewhat analogous to the statistical concept of a sufficient statistic.) For example,
there are at least 26 case-control studies on ETS and lung cancer. Depending on one's judgment of the
homogeneity of the conditions under which the studies were conducted, from one to 26 contexts could
be defined.
If studies within each country are considered sufficiently homogeneous to pool the results by
methods of meta-analysis (DerSimonian and Laird, 1986; Greenland, 1987; Eddy, 1990; Eddy et al.,
1990a,b), but the same is not the case for studies in different countries, then a separate context might
be defined for each country. (Note: There are currently no firm guidelines on when, or even "if in the
minds of some analysts, meta-analysis is appropriate (Mann, 1990)). When there is more than one set
of study data on animals or humans within a context, then the separate sets are termed cases. For
example, if ETS studies are grouped into contexts by country (arguing essentially that subpopulations
within a country experience similar antecedent conditions, but that the same is not true across
99
-------
WORKING TABLE 1. CONTEXT SPECIFICATION FOR HAZARD IDENTIFICATION
Context
Number
(1, 2, ...)
Context
Type
(0,T)X
Description
O: Observational Context
T: Target Context
Observational contexts and target contexts are numbered separately.
-------
countries), then the individual studies within a country are cases. If studies between countries
are considered homogenous, a single context might be defined. Similarly, long-term animal
studies conducted by NTP generally include two species and two sexes. One to four contexts
could be defined. If it is reasonable to combine results for the two sexes within a species,
however, then one would have two contexts with two cases each. A context needs to include as
many cases as reasonable so as to facilitate comparisons and judgments of coherence across
studies, while still retaining as much specificity of detail as may be relevant to hazard
identification. Data categories in Table 2 that are exclusively context-specific include: (1) tumor
response, (2) biophysical effects, (3) concurrent environmental conditions, and (4)
environmental/substance characteristics. The remaining data categories are largely species-
specific or chemical-specific without reference to the context within which the given species is
exposed to the chemical (although there will be cases in which even pharmacodynamic data will
be context-dependent, as when absorption fractions, intake characteristics and/or retention
characteristics change with magnitude of intake (for examples of formaldehyde see Graham et
al., 1988; Burkhart et al., 1990)). Data items from these latter categories are included in
Working Table 2 (to be completed for each observational context in Working Table 1), as
applicable to the context. In other words, a data item (such as presence of the necessary
bioactivation enzymes) is included in the working table for a given context even if the item has
not been observed under the conditions of exposure defining the context, so long as it is judged
that the item is species-specific (the species also defining the context). Any use of the data item
to draw intra-context inferences of carcinogenicity will, of course, require introduction of the
premise that exposure (to the substance, as defining the context) would not have significantly
altered the observed data item. The contexts under which the analyst chooses to conduct the
101
-------
hazard identification are developed using Working Table 1. Completion of Working Table 2 is
discussed next.
42. Assessing Data Quality by Observational Context
The data available for each observational context need to be evaluated and summarized
for application to hazard identification. If there is more than one data source on an item (i.e.,
data from more than one case) then the assessment of the data item should represent the
coherence (or incoherence) of the composite information available. Discussion specific to
summarizing across cases within a context will not be included here. Aside, perhaps, from
application of meta-analysis for summarizing statistical outcomes across cases within a context,
the ideas employed are similar to those for summarizing across contexts that will be discussed.
Cases, however, are more homogeneous than contexts, which simplifies the judgments necessary.
Data characteristics to be assessed, shown along the top of Working Table 2, include (1)
"completeness" (all data of the item were found), (2) "utility" (data are of high quality, applicable
to hazard identification in the context observed, and statistically interpretable), (3) the observed
effect (there is statistical evidence of a relationship between exposure and the data item), and
(4) causality of the observed effect (the nature of the association between exposure and the
observed effect is best described as...). These four characteristics of data quality as summarizing
a data item will be discussed separately.
102
-------
WORKING TABLE 2. DATA CHARACTERISTICS FOR OBSERVATIONAL CONTEXTS
CONTEXT NO.
Data Category/Item
Completeness
(Hi/Me/Lo/No)
Utility1
(Hi/Me/Lo/No)
Observed Effect2
(Hi/Me/Lo/No)
Causality3
(AA/CC/EC/OC)
Exposure Context- Organism-
Effect Specific Specific
Measurement Measurement
Tumor Response
TRl
TR2
Biophysical Effects
BE1
BE2
Pharmacodynamics
PD1
PD2
Host Characteristics
HC1
HC2
Related Substances
Assessments
RSA1
RSA2
(continued on following page)
103
-------
Working Table 2 (continued)
Data Category/Item
Completeness
(Hi/Me/Lo/No)
Utility1
(Hi/Me/Lo/No)
Observed Effect^
(Hi/Me/Lo/No)
'J
Causality
(AA/CC/EC/OC)
Exposure
Effect
Context-
Specific
Measurement
Organism-
Specific
Measurement
Structure Activity
Relationships
SARI
SAR2
May be subdivided into additional categories when useful, e.g. "validity", "reliability", and "accuracy" may be judged separately for data
items from laboratory sources.
Refers to effect on data item of exposure to agent. Footnotes with explanatory comments may be needed. When statistical measures are
available, they may be more informative than a simple indication of Hi/Me/Lo/No, e.g., estimates or tests of statistical significance.
See Section 4.2.4 for explanation of the following choices available.
AA: Accidental Association
CC: Common Cause
EC: Empirical Causality
OC: Operational Causality
104
-------
4.2.1. Completeness
A term commonly employed in discussions of both law and rationality is the principle of
the "hard look" (see the appendix). The term is both descriptive and normative (i.e. implying
that the "hard look" is necessary for a proper claim to rationality), introducing the intellectual
obligation that rational discourse proceed from a base of observation and conceptual
frameworks that is as large as possible without sacrificing the overall end of completing an
analysis. The hard look may be taken to address three primary questions with respect to hazard
identification. These are:
(1) Have all published studies (within a data item) been collected?
(2) Are published studies an accurate representation of all conducted studies (raising
the issue of possible publication bias)?
(3) Do the collected studies represent accurately the phenomenon on which they
supposedly report?
In the present document, the first two questions are taken to be questions of the
completeness of data collection. Rationality then is weakened if the collected literature do not
constitute an adequate sample of available studies. The third question is referred to here as an
issue of utility. In this stage of the analysis, principles of study design and statistical sampling (a
quantitative source of uncertainty through variability) are examined to determine whether the
collected studies were capable of yielding an accurate representation of the empirical features of
the data item being considered.
The data items included in Working Table 2 should include those reporting some
outcome that may be a consequence of exposure to the agent of interest in the context of
interest, e.g., from data categories such as tumor response, biophysical effects, concurrent
environmental conditions, and pharmacodynamics (as applicable), as well as data items that are
not context-specific but may have some bearing on a judgment of carcinogenicity from that
context, e.g., data on structure activity relationships, host characteristics, and related substances
105
-------
assessments if applicable to the context. In Working Table 2, the generic notation for data items
(TR1, TR2, BE1, etc.) is used because the applicable data items will differ between contexts.
For example, under "tumor response" the notation "TR1, TR2, ..." is replaced by specific data
items from Table 2, such as "tumor incidence" and "age-at-appearance". The analyst then
determines whether published studies have been adequately collected and published studies are
an adequate representation of the full body of studies performed. For example, the study of
ETS would be weakened by failure to include all of the published human epidemiologic data, or
by demonstration that positive or negative studies had been consistently excluded from
publication for some reason. In making this judgment the analyst will find it useful to reflect on
several possible warrants for a claim to completeness:
(1) All studies (both published and unpublished) are known to have been collected.
(2) It is known that not all studies have been collected, but the collected studies
represent a properly constituted random sample of available studies.
(3) All published studies have been collected and there is no publication bias.
(4) It is not known whether all studies have been collected, but it is judged that
further searching would detract from the ability to carry out other tasks required
in the analysis (i.e. contextual rationality, as discussed in the appendix).
Completeness is judged on a scale of "high", "medium", "low" or "no", depending upon the
strength of the above warrants. An assignment of "high" implies that both published and
unpublished studies have been collected, or that published studies have been collected and there
is no publication bias. An assignment of "no" implies that there are no data available even after
having adequately examined the literature. Having assigned a judgment of completeness to each
data item, the analyst also assigns a judgment of completeness (on a scale of Hi/Med/Lo/No)
to the data category.
106
-------
422. Utility
Utility of a data item refers primarily to the quality of the process(es) under which
observational data were generated, e.g., by an epidemiologic study, long term animal study,
short term test, etc., independent of the observed outcome or of the completeness of the data
category/item. The utility of a body of data increases with increasing ability of the studies to
accurately and precisely define the underlying empirical properties under consideration in
collection of the data item. Qualitative characteristics of interest are those related to methods
and materials, including analysis and appropriateness of the conclusions drawn. For example, if
observational information on the data item is from an epidemiologic study then the quality of
the data item is judged vis-a-vis adherence of the study to principles of good epidemiologic
practice. Similarly, if the data are from a long term animal study or a short term test, or are
generated by some other means (as, for example, from consideration of structure activity
relationships), the methods used to generate and analyze the data item determine its quality.
The higher the quality, the more assurance that the observational information on the data item
and the reported conclusions are a valid depiction and summary of what was claimed to be
observed, aside from statistical sources of uncertainty or variability. Statistical uncertainty can
further reduce the utility of a data item, quality not withstanding. Statistical error decreases
with sample size, although size is not the only determinant. The smaller the sample size the
lower the power (i.e., the less likely) to detect an association between the substance of interest
and the data item. To summarize, utility should increase (1) as quality of the study (or other
data-generating activity) increases, because the uncertainty due to methods and materials
declines, and (2) sample size increases, because uncertainty due to statistical error declines.
Care needs to be taken to consider factors affecting utility that may not be foreseen or
not readily apparent in the discussion above, such as applicability to the context in which the
107
-------
data are being used in the analysis. For example, an epidemiologic study may be judged to be
of high quality under one set of circumstances when evaluated with respect to study design,
conduct, analysis, and conclusions, in the sense that all the recommended procedures for those
circumstances are exercised with care and rigor. The quality may suffer, however, from the
limitations of methodology under some other set of circumstances of interest to the analyst.
Epidemiologic studies often have more than one objective and data for multiple endpoints may
be collected simultaneously from a single questionnaire. Investigation of the substance of
interest for hazard identification may be among the secondary objectives of a study or the
available data may actually be a secondary data set from a previous study. The utility of the
data items for hazard identification in such a case may be compromised even though the utility
for the original end of the study was strong.
To illustrate, consider an epidemiologic study reporting tumor response from exposure to
environmental tobacco smoke (ETS) as part of a larger objective to investigate health risks
from indoor air pollution in general, of which the health effects from ETS is only one endpoint.
If the study was conducted in a region with unusually high indoor air pollution from related
substances, e.g., smoke from inadequate ventilation from indoor combustion of wood, coal, or
other materials used in cooking or heating, then the study may have low sensitivity to detect an
effect from ETS even though the quality for studying the effect of air pollution in general may
be high with respect to study design, conduct, analysis, etc. and the sample size may be
extremely large. What needs to be recognized in this case is the presence of factors that
methodology may not adequately address. A "noisy" background may make detection of an
effect from ETS very unlikely, contrary to the appearance of adequate sample size based on
power calculations. The application of statistical methods to adjust for confounding, and hence
to "correct" for the presence of other pollutants that may produce detrimental health effects
108
-------
similar to ETS, may be inadequate in such a situation. The difficulty does not lie in the
accuracy of the statistical methods, but in application to situations (contexts) that exceed their
limitations (i.e., the assumptions on which the validity of the methods are based are violated
sufficiently to produce misleading results). Consequently, an epidemiologic study might be
judged high in quality against the textbook criteria for good epidemiologic practice and be of
adequate sample size, but warrant a lower judgment for utility (i.e., usefulness, applicability,
value, contribution to the weight of evidence, etc.) within a given context defined by Working
Table 1 and employed in Working Table 2. Technically, this illustration is not a counterexample
to assessing quality, as discussed above. It does indicate, however, that care needs to be
exercised in making judgments of utility for data items, and this applies in varying degrees to all
the data items of Working Table 2.
The next two sections sketch some basic features of animal studies and epidemiologic
studies to consider in assessing utility. The discussion is largely summarized from the OSTP
guidelines for chemical carcinogens (Fed. Reg., 1985, p. 10372 - 10442), and interspersed with the
authors' opinions or material from other sources. There is considerable literature on both
topics, particularly related to the design and conduct of epidemiologic studies, to which the
reader is referred. The OSTP guidelines include numerous references that may be useful. Some
more recent sources on the design and conduct of animal studies are described in the following
section. Material on the design and conduct of epidemiologic studies is readily available in
textbooks, journals, and other sources.
4.2.2.1. Utility of Animal Studies Long-term animal exposure studies often provide much of the
data available for hazard identification. There are numerous references that discuss principles
of design and protocol for animal studies that may be useful in guiding a judgment of quality
109
-------
(utility) for data from animal studies (Code of Federal Regulations 1983a, 1983b; Hamm, 1985;
Prejean, 1985; Boorman et al., 1985; U.S.DHHS, 1984; Krewski et al., 1990; Haseman, 1985;
Portier and Hoel, 1983; IARC, 1986). Critical design criteria are summarized below under five
of the six headings used by the OSTP guidelines (Fed. Reg., 3/14/85, p. 10411-), and much of the
discussion is excerpted from that source. Only the OSTP guideline related to the selection of
species and strain of animal is omitted here since that factor is related to inter-context
extrapolation (see Chapter 5) rather than utility as developed here.
(1) Animal Care and Diet. Housing, purity of diet, water and air, proper housing and
care of animals, control of intercurrent diseases or parasites, and controlled exposure to
the test agent are critical to valid testing. Care must be taken to avoid bias in selection
and allocation of animals between controls and treated groups, or in placement of cages.
(2) Test and Control Groups. It has been recommended that each dose group and a
concurrent control contain at least 50 animals so that enough animals are available at the
end of the study for pathological examination. Fewer animals reduces statistical power
to detect an effect and reduces the expected number still alive at the end of the study,
reducing the number exposed for "life". With interim sacrifice of animals, the number
should be larger. The control group and treatment groups should be treated identically
aside from exposure to the agent for valid comparisons.
(3) Dose Levels, Frequency and Route of Exposure. Two or three dose levels in
addition to the concurrent control group are typically used. The highest dose currently
recommended is the maximum tolerated dose (MTD), a dose just high enough to elicit
110
-------
minimal toxicity without significantly altering animal survival rates under lifetime
exposure. Lower dose levels are typically at 1/2 or at 1/3 and 2/3 of the MTD. The
rationale for dose selection should be clearly stated. The rationale for use of the MTD
is to maximize the chance of detecting a carcinogenic effect. Lower doses provide
information on the shape of the dose response curve, for dose-response assessment, but
are also important for hazard identification. Multiple doses are needed to implement
tests for trend and to indicate nonlinear behavior that may influence extrapolation of
results across species or to low exposure levels. The route of administration used for the
bioassay ideally corresponds to the route of exposure of humans. However, so long as
the route of administration used on test animals results in absorption, distribution, and
metabolic activation (if required) of the substance, the test results are generally
regarded as relevant for hazard identification in humans.
(4) Duration of study. Test animals are normally exposed for most of their expected
lifespan, which means a study duration of about two years for rats and mice. Treatment
is typically begun soon after weaning. Termination of the study is ordinarily acceptable
when the number of survivors in the low dose or control group is reduced to 20-25%. If
high-dose toxicity affects survival in that exposure group, the study is still continued with
the remaining animals at lower exposure levels and in the control group. A negative test
is ordinarily accepted by regulatory agencies if: (1) no more than 10% of any group is
lost due to autolysis, cannibalism, or management problems; and (2) survival of all
groups (per sex per dose) is no less than 50% at 80 weeks for hamsters, 96 weeks for
mice, and 104 weeks for rats.
Ill
-------
(5) Data Collection and Reporting. Animals need to be observed on a regular basis for
indications of toxic effects or disease, and for autolysis or cannibalism. Body weights and
food intake should be recorded on a regular schedule. Clinical signs and mortality
should be recorded for all animals, with special attention paid to visible signs of tumor
development.
4.2.2.2. Utility of Epidemiologic Studies. The obvious advantage of epidemiologic data or
animal bioassay data is that observations are on human subjects, relieving the uncertainty of
extrapolation of results across species. On the other hand, the researcher has no control over
the exposure environment as in animal studies (completely removing the claim of operational
causality discussed in Section 4.2.4.) and often cannot completely specify "Concurrent
Environmental Conditions" (see Table 2). Exposure to the substance of interest may vary
between study subjects, and temporally for any given subject, and environmental conditions and
other factors of potential importance are not constant (see Section 5.4.4.). Those limitations not
withstanding, however, it is often difficult to obtain accurate and reliable data on the factors of
interest. Consequently, some types of epidemiologic studies are more useful for hazard
identification than others. For example, descriptive studies may utilize the correlational (or
ecological) approach, in which the rate of disease in a population is compared with the spatial or
temporal distribution of suspected risk factors. This type of study is helpful in developing or
refining hypotheses for further investigation, but is not very useful for hazard identification since
causality is unresolved. Data are collected on whole populations instead of individuals,
providing only correlational evidence that is too broad-based for inference of a causal
association between increased incidence of cancer (or other effect) in one population and
increased exposure to the substance of interest (again, see Section 4.2.4.).
112
-------
Analytical epidemiologic studies are the principle means for determining the human
health hazards of specific environmental agents. In contrast to a descriptive survey, data are
obtained on disease occurrence and putative risk factors for specific individuals, using mainly the
case-control or cohort method. Case-control studies start by identifying persons with a
particular disease (cases) and a group of similar persons without the disease (controls).
Information on past exposure to the agent of interest and other potential risk factors is collected
from which statistical comparisons are made between cases and controls. If the frequency of
exposure to the agent is higher in cases than in controls, after adjustment for other risk factors
that might produce the disease (for hazard identification, the disease is a cancer endpoint, such
as lung cancer in the studies of passive smoking, although it could be more specific, such as lung
adenocarcinoma), then the analyst must consider whether the observed outcome indicates an
association between the agent and the disease. The case-control approach is well suited to
studying relatively rare diseases, where either (1) exposure to the agent is common, as in studies
of environmental tobacco smoke or menopausal estrogens, or (2) exposure is rare but accounts
for a large portion of a particular cancer, such as liver angiosarcoma in the study of vinyl
chloride.
By contrast, cohort studies start by identifying a group of individuals with a particular
exposure and a similar group of unexposed persons, followed by examination of both groups
over time to determine subsequent health outcomes. The incidence of disease in the exposed
and unexposed groups are then compared. These investigations may be based on current
exposure and future health outcomes (prospective cohort study), but more commonly they utilize
past exposure information and disease occurrence (retrospective cohort study). Instead of an
unexposed comparison group, general population mortality or incidence rates (specific for age,
sex, race, and calendar time) are often used to determine the expected number of cases of
113
-------
diseases. Cohort studies are expensive, complex, require large numbers of exposed persons, and
long periods of follow-up. They are less subject to sources of bias (to be discussed further),
however, the risks attributed to a particular exposure can be estimated directly, multiple health
outcomes (including multiple cancer endpoints) can be assessed, and temporal relationships such
as latency period and duration of effect can be evaluated. The intervention study is a third type
of analytical epidemiologic study, especially useful in confirming causal relationships suggested
by case control or cohort studies (see particularly the discussion of operational causality in
Section 4.2.4.). This approach may be applied in programs designed, for example, to reduce
cigarette smoking.
The qualitative integrity of the design, conduct and methods of analysis in epidemiologic
studies is an essential determinant of the utility of the study results for hazard identification.
The capability to eliminate sources of bias and confounding as possible explanatory factors in
lieu of exposure to the agent being studied may depend heavily on characteristics related to
study quality. In both case-control and cohort studies, confounding cannot normally be removed
by appropriate study design alone, i.e., possible confounders need to be controlled in the
analysis, as possible. Since data are required, the collection of data on possible confounders
enhances the utility of a study (see "Concurrent Environmental Conditions" in Table 2 and the
discussion of causality in Section 4.2.4.). Bias is more of a concern in case-control studies than
cohort studies, arising largely from the case-control design. In particular, care should be taken
to avoid bias (to the extent possible) in the selection of cases and controls for study and in the
collection of data on exposure and related risk factors. In general, qualitative characteristics of
an epidemiologic study involve design, execution, and analysis. Methodologic criteria for
evaluation of quality are readily available in textbooks and other references on epidemiology.
Two IARC scientific publications specific to the application of case-control and cohort studies to
114
-------
cancer research that may be particularly useful are by Breslow and Day (1980, 1987), although
they focus on statistical methods. Additional sources that discuss the application of
epidemiology to cancer research include Cook (1982), Peto(1985), Hoel (1985), Day (1988),
Wald and Doll (1985), NRC(1986), Muir and Higginson (1985), Hoel and Landrigan (1987),
Mettlin (1987), Cone et al. (1987), Krewski et al. (1990), Morris (1990), and Sorsa et al. (1990).
Table 4 contains a checklist of items for methodological critique of studies as constructed
by Spitzer et al. (1990). Not shown in the table is the choice of responses to each item (yes,
uncertain/incomplete/substandard, no, don't know/not reported, N/A, N/C, and space for
comments). A worksheet for review of case-control studies of ETS was developed by Spitzer et
al (1990) around five categories, including: (1) General. Study objective, primary or secondary
data set, meaning of terms ("nonsmoker", "exposed to ETS", etc.), recall span (duration since
ETS exposure), type of exposure (cigarette, pipe, etc.), classification of ETS exposure in
unmarried women. (2) Data Collection. Inclusion/exclusion criteria for cases (and separately for
controls), source of subjects (hospitals, community, etc.), incident cases (Y/N), control sampling
(cumulative/density/matched/unmatched), method of collection (face-to-face, telephone, self-
administered questionnaire, medical records, etc.), verification of data with other sources (Y/N),
sample size, attrition in selection and follow-up, source of data on subject (subject or proxy),
exposure periods (adulthood, childhood, etc.), exposure sources (smoking by parents, spouse,
household members, etc.), age. (3) Clinical data. Method of verification of primary lung cancer,
airway location of lung cancer (central or peripheral), prevalence by tumor type (squamous cell,
adenocarcinoma, etc.). (4) Statistical Analysis. Raw data, unadjusted (crude) statistical analysis
(type and outcome of statistical tests for effect, for trend, etc.), adjusted statistical analysis
(same tests but adjusted for confounding factors). (5) Dependent variables used in matching (of
controls to cases), in analysis, and used or discussed otherwise. A checklist (such as Table 4) or
115
-------
TABLE 4. CHECKLIST FOR METHODOLOGICAL
CRITIQUE OF EPIDEMIOLOGIC STUDY
1.
Random assignment, properly done
2.
Suitable choice of reference group
3.
Similar methods of data collection for all groups
4.
Proper sampling or suitable assembly of comparison group
5.
Sample size
a. enables adequately precise estimates of priority variables found to be
significant
b. enables adequate precision in secondary variables reported (confounding
variables or incidental findings)
c. power reported for nonsignificant findings
d. power declared a priori
e. clinical or practical significance of statistically significant differences set forth
or justified
6.
Criteria for definition or measurement of the outcomes are objective or verifiable
7.
Definition of exposure; unambiguous and measurable
8.
Measurement of exposure; accurate and verifiable
9.
Blind assessment
10.
Observation bias minimized by design or accounted for in analysis
11.
Selection bias accounted for
12.
Objective criteria for eligibility of subjects (inclusion and exclusion)
13.
Attrition rates (%)
a. response rate
b. losses to follow-up
c. other
14.
Known confounders accounted for
a. by design
b. by analysis
15.
Any methods to attempt comparability between groups, other than randomization
16.
Comparability of groups under comparison demonstrated
17.
Appropriate statistical analytic plan
a. evidence that a priori hypotheses being tested
b. correct method used
c. adjustment made for
- multiple comparisons
- simultaneous multiple range testing
d. display of raw data permits assessment of actual measures and adjustments or
transformations made
18.
Conclusions supported by data presented
19.
Reproducibility of method(s)
20.
Generalizability of results
a. from sample(s) to parent population
b. from sample(s) to any relevant population
21.
Other, specify
Source: Spitzer et al. (1990)
116
-------
worksheet for items that may be important to judging utility of a study facilitates inclusion of
significant items and detection of items not addressed in a study that should be.
The utility of an epidemiologic study is also affected by its statistical power to detect an
effect of dose on the health endpoint of interest (this, and the following comment, also applies
to the utility of animal studies). If statistical significance is not achieved (typically a p-value <
0.05 is interpreted as "significant"), then it is of interest to know the power of the test to detect
the minimal effect level that would be considered consequential. It is not uncommon for an
estimate of relative risk to be "high", e.g., 2 or larger, but not significant because of inadequate
power. One cannot simply conclude, however, that if the sample size had been sufficiently large,
then the estimate would remain unchanged but become statistically significant. One could argue
that statistical power only affects the utility of a study when the outcome is nonsignificant. Since
the outcome is not known in advance, however, factors affecting statistical power must be taken
into account during the design phase. Factors affecting power include: (1) Size of the study
group and control group(s). (2) Variability of the background cancer rate. (3) Criteria for
significance (i.e., predetermined p-value to reject the hypothesis of no effect, technically called
the "test size"). (4) Magnitude of the expected association between exposure and effect. (5)
Design of the study and the statistical techniques used for analysis. (Marsh and Caplan, 1986, as
reproduced in Morris, 1990, p.271).
4.2.2.3. Summarizing Utility For the data item under consideration, the utility of the set of
available data is judged on a scale of "high", "medium", "low" or "no". An assignment of "high"
implies the data are characterized by an appropriate study within the prespecified context,
including both qualitative features (use of proper experimental and/or epidemiologic
techniques) and quantitative features (e.g. power). As either of these characteristics weaken,
117
-------
the assignment of utility is lowered due to decrease in the ability of the data to accurately
correspond to the property being measured. It is important to note here that the assignment of
utility does not include the issue of confidence intervals on point estimates, which are discussed
in the next section on observed effect.
423. Observed Effect
Observed effect for a data item indicates the level of empirical support for some form of
association between the data item and the agent of interest, independent of the judgment of utility
and completeness for the item. As discussed above, when the context for Working Table 2
includes more than one case a composite result for the data item in that context needs to be
formulated by the analyst for each data item. For example, suppose that for epidemiologic
studies of lung cancer and ETS, each country for which there are studies is defined as a context
and the intra-country studies are the cases for the context. Under data category "tumor
response" TR1 in Working Table 2 might denote "relative risk". A composite estimate,
confidence interval, and significance level (p-value) might be formulated for relative risk by
pooling the statistical results (not the data) on relative risk from the component studies. Data
items under the category "biophysical effects" would be treated similarly, although statistical
results from each study may be limited to tests of significance. Since estimates, confidence
intervals, and tests of significance contain different information, the analyst may choose to enter
these pooled statistical outcomes for a data item in place of a more qualitative choice from
(Hi/Me/Lo/No), or perhaps present the statistical summary in a footnote. Since the end of
hazard identification is to produce a claim that the substance does (or does not) induce a
change in cancer incidence, rather than an estimate of the magnitude of any change, loss or the
118
-------
quantitative information in summarizing the strength of association is not significant (unless
magnitude of change is deemed a measure of the existence of an association).
Only compatible statistical tests may be pooled, e.g., the outcome from a trend test
cannot be combined with the outcome of a pairwise comparison of cases and controls.
Technical deliberation may be required to form an overall judgment of observed effect. Data
items under "tumor response" may apply to an epidemiologic study or to an animal study,
depending on the context. Statistical methods, of course, differ accordingly. Several of the
references cited in the discussion on "utility" above (Sections 4.2.3.1 and 4.2.3.2) include material
on statistical methods and interpretation of results as well as aspects of design that may be
helpful. References for long term animal studies include Gart et al. (1979), Mantel (1980), Peto
et al.( 1980), Clayson et al. (1983), Haseman (1983, 1984), Haseman et al. (1984), Park and
Kociba (1985), Maronpot(1985), Huff et al. (1986), Farrar and Crump (1988), Krewski et al.
(1988), Archer and Ryan (1989), Bickis and Krewski (1989), Haseman et al. (1989), Haseman
(1989), Krewski et al. (1989), Portier and Bailer (1989), Haseman (1990), and Wahrendorf
(1990). For epidemiologic studies, the two IARC volumes by Breslow and Day (1980 & 1987)
are particularly oriented toward cancer studies, although there is a vast array of readily available
literature. Data on biophysical effects can generally be statistically analyzed with the effect of
interest replacing cancer as an endpoint, although categorical data may be more common since
observations may include severity rankings by a pathologist, e.g., classification hyperplasia as
severe, moderate, mild, or absent.
Data for relating the strength of association of a data item with exposure level are less
likely to be available (or particularly meaningful) for the six remaining data categories (host
characteristics, pharmacodynamics, concurrent environment conditions, related substance
assessments, and structure activity relationships). In particular, the data categories "related
119
-------
substances assessments" and "structure activity relationships" would not include data items for
which "exposure effect" has meaning for the factor under consideration in the analysis (although
it will have meaning for the "related substance") itself. Host characteristics may predispose
some populations to higher exposure effects than typical, which needs to be taken into
consideration (see examples in Tables 5 and 6). Consequently, there will be several forms of
"observed effect" summaries possible, as indicated in Working Table 2.
The first form of summary statement arises from the "tumor response" and "bioeffects"
data for which exposure-response observations are available. In this case, the desired judgment
is whether an association has been noted and, if so, the strength of that association. A judgment
of "No" indicates that there is no association, which this does not mean the data item will not
affect assessment of support for claims of carcinogenicity. A judgment of "Hi", "Med", or "Lo",
on the other hand, indicates some association (from "strong" to "weak") between exposure to the
agent and presence of the observed factor (tumor incidence, cellular transformation, etc.). This
form of statement is given in the column "Exposure-Effect" under "Observed Effect".
The second form of summary statement arises from other data categories in which the
observed factor (data item) has been measured in the presence of the agent of the level of
exposure characterizing the context. For example, the data item "bioactivation enzymes" may
contain data on the presence or absence of necessary metabolic enzymes under conditions of
exposure. This form of statement is given in the column "Context-Specific Measurement" under
"Observed Effect".
The third form of summary statement arises from data similar to those in the second
form with the exception that they (the former data) have not been obtained under conditions of
exposure. An example might be the presence of repair enzymes for DNA in unexposed
organisms (the organisms specifying the context). The analyst must already have judged, of
120
-------
TABLE 5. DEVELOPMENTAL PROCESSES THAT
ENHANCE SUSCEPTIBILITY TO ENVIRONMENTAL POLLUTANTS
High-Risk Groups
Immature Enzyme
Detoxification Systems
Immature Immune System
Deficient Immune System as
a Function of Age
Differential Absorption of
Pollutants as a Function of
Age
Retention of Pollutants as a
Function of Age
Pregnancy
Circadian Rhythms Including
Phase Shifts
Infant Stomach Acidity
Estimated Number of
Individuals in United States
Affected
Developmental Processes
Embryos, fetuses, and
neonates to the age of
approximately 2-3 months
Infants and children do not
reach adult levels of IgA until
the age of 10-12
Progressive degeneration
after adolescence
Infants and young children
Individuals above the age of
50
Approximately several million
females per year in the U.S.
All people have certain
periods of the day when they
are more susceptible to
challenge
Infants
Pollutant(s) to Which High-
Risk Group is (may be)
Increased Risk
Pesticides, polychlorinated
biphenyls (PCBs)
Respiratory irritants
Carcinogens, respiratory
irritants
Barium, lead, radium
Fluoride, heavy metals
Anticholinesterase
insecticides, carbon
monoxide, lead
Hydrocarbon carcinogens
and probably most other
pollutants
Nitrates
Source: Calabrese, E.J. (1984)
121
-------
TABLE 6. PREVELANCE OF SUBGROUPS
HYPERSUSCEPTIBLE TO EFFECTS OF COMMON POLLUTANTS
Hypersusceptible
Prevalence
Chemicals
Embryo, Fetus, Neonate
Pregnant Women:
Carcinogens, solvents, CO,
21/1000
mercury, lead, PCBs, pesticides
Young Children
Ages 1-4:
Hepatotoxins, PCBs, metals
70/1000
Lung Disease
Ozone, Cd, particulates, S02, N02
Coronary Heart Disease
Coronary Heart Disease:
Chlorinated solvents, fluorcarbons
16-27/1000
Liver Disease
Liver Condition:
Carbon tetrachloride, PCBs,
20/1000
insecticides, carcinogens
Source: U.S. EPA (1984)
122
-------
course, that such data items are applicable within the exposure conditions characterizing the
context. If the observed property is likely to be strongly affected by exposure to the agent, this
initial judgment of inclusion in the context would have been invalid. This third form of summary
statement is given in the column "Organism-Specific Measurement" under "Observed Effect".
For both the second and third forms of summary statement, the judgment is not one of
an association. It is, instead, a measurement of some property related to theories of the
etiology of cancer. The judgment is that the observed property is "strongly present" (Hi),
"moderately present" (Med), "weakly present" (Lo) or "not present" (No) for the substance
within the context under consideration. In each case, the "strength" of the "presence" is related
to the effect on a claim of carcinogenicity. For example, substances that are strongly absorbed
by the G.I. tract, tenaciously retained in the target organ, and readily activated to the
biologically significant metabolite through biotransformation in the organism would be assigned
a value of "Hi" in each of these three "observed effects" columns. Conversely, substances that
are not deposited readily in the lung following inhalation, that are not neutrophilic, or do not
partition from water to oil would be assigned a value of "Lo" or "No" in each of these three
"observed effects" columns. For reviews of the role of the various data items in judgments of
carcinogenicity, the reader should consult Andersen (1981), Andersen (1989), Anderson et al.
(1980), Barrett and Wiseman (1987), Clayson (1987), Clewell and Andersen (1989), Gehring et
al. (1978), Gerlowski and Jain (1983), Gillette (1976, 1984), Hattis (1990), Keck (1981), NRC
(1986, 1987), US EPA (1989) and US FDA (1982).
42.4. Causality
This section reviews the separate judgments to be made in warranting the claim that an
association noted in a given data item is or is not an indication of causality (again, see Working
123
-------
Table 2 and the "Exposure-Response" data items under the column "Observed Effect"). These
causal judgments take two primary forms. The first refers to properties of the particular study
from which the observation was obtained. This component focuses only on characteristics of the
study (data item) itself indicative of causality pertinent to the current discussion. The second
refers to expectations formed independent of the particular study, and is more properly
contained within the issue of theory-based inference raised in Section 5.2.4. These two forms of
causal warrant are discussed together here only due to their conceptual link. It should be borne
in mind that only the first mode of warranting causality is to be used in forming the judgment of
causality for Working Table 2.
The study results themselves may provide warrants for a judgment of causality. The
intent of the warrants is to distinguish between the following cases (proceeding from the weakest
to the strongest warrant):
(1) Accidental association. This is a judgment of non-causality. It is addressed
through reflection on statistical properties of the study and is summarized in
statistical measures of the significance of the association (see the discussion in
Section 4.2.2.). Such a judgment is warranted when the statistical properties are
insufficient to rule out the null hypothesis.
(2) Common cause. This is a judgment that the factor under consideration (such as
exposure to the substance of interest) does not cause the effect noted in the
study, but the factor of interest is caused by the same process that produces the
true causal factor. For example, particulates in ETS have been taken as a causal
factor in lung carcinogenesis (U.S. EPA, 1990). This claim to causality will be
weakened if both particulates and gases have a common cause in the burning of
cigarettes, and it is the gas which is the true carcinogen. Still, the use of
particulate measurements as an indicator of hazard in an environment will be
appropriate if it is premised that the factor under consideration (particulates) is
in common cause with the true factor (gas). To be in common cause implies that
the factor under consideration is present simultaneously with the true causal
factor, since whatever produces the former also produces the latter. Mitigation
strategies based on lowering of the former factor will produce the desired effect
since any mitigation of the former implies mitigation of the latter factor. The
utility of common cause arguments is greatly weakened, however, when it is
possible that the factor under consideration might in some cases exist
independently of the true causal factor. Common cause arguments, based on
strength of association only, is a form of epistemic instrumentalism as discussed
124
-------
in the appendix. A judgment of common cause is warranted when the factor of
interest is accompanied by other factors, as observed in the data category
"Concurrent Environmental Conditions", and where existence of the factor of
interest is caused by some process that also gives rise to one or more of the
"concurrent" factors.
(3) Empirical causality. In this instance, features of the study are used to warrant
the claim that the associations noted are physically causal rather than accidental
or the result of common cause. These features are given significance through
theories of carcinogenesis describing properties of study findings that are to be
expected in instances of a causal connection. The primary indications of causality
then are (1) appearance or disappearance of the effect when exposure is
historically present or absent (respectively), (2) the temporal pattern of
appearance of the effect (i.e. the latency period or the hazard function), (3)
appearance of the tumors at the site of the BSDR produced by the exposure, and
(4) there are no "Concurrent Environmental Conditions" judged capable of
explaining the observed change in prevalence.
(4) Operational causality. This is the strongest warrant for a judgment of causality.
The evidence invoked results from a proper experimental setting in which the
researcher has produced deliberate and intentional manipulations (operations) of
the factor of interest to yield the effect. The physicality of the manipulation is
essential in the warrant, and the strength of warrant increases as the analyst
increases confidence that only the intended manipulation occurred.
The four categories of causality above are to be warranted through reference to
empirical properties of the studies. For example, criteria for causal inference in epidemiologic
studies have been developed (Hill, 1965), qualified (Evans, 1978; Weiss, 1981) and discussed in a
context of inductive and deductive logic (Maclure, 1985). These empirical criteria, used in
assigning causality for Working Table 2, are not discussed here further due to their extensive
review in the existing literature. A causal connection may, however, also be supported through
arguments based on theory, often referred to as biological plausibility (in fact, this form of
causal argument must be excluded from consideration in generating Working Table 2, since it is
used in subsequent parts of the analysis discussed later). The reasoning here is not that the
study characteristics support the contention of causality (although this is not excluded), but
rather that prior understanding of the observed phenomenon leads to an expectation of a causal
125
-------
connection. Etiologic theories, and observational items from categories other than the category
under consideration, then might be invoked as further support for the claim to a causal
connection in the "tumor response" and "biophysical effects" data. For example, the causal
connection between exposure to ETS and lung cancer (as given in the data category "Tumor
Response") may be warranted partially by reference to observations on active smoking (in the
"Related Substances" category) and theories founded on the concept of cigarette equivalence
(Thorslund, 1990). This prior expectation of a causal connection must be assessed, of course,
both for the exposure factor of interest and for any other factors invoked as competing
explanations of the observed effect. It will be useful (as discussed in Chapter 5) to think of the
empirical and theory-based warrants as providing a system of warrants within which the
conceptual coherence of the claim to causality may be judged. In any event, theory-based
warrants for causality (i.e. biological plausibility arguments) constitute an instance of Theory-
Based Inference towards the claim of carcinogenicity, and discussion of this mode of warranting
is deferred until Section 5.2.4.
5. WARRANTING CLAIMS TO CARCINOGENICITY
The preceding two chapters have dealt with the first two of the four tasks for hazard
identification specified in Section 1.3., specifically the development of a taxonomy of claims of
carcinogenicity and the assemblage and assessment of observational evidence. The taxonomy
does not need to be reconsidered with each new agent evaluated for carcinogenicity, although it
could be altered as beneficial. Establishment and assessment of the informational base is, of
course, specific to the agent and/or context of interest. Working Tables 1 and 2 suggest formats
to assist with the organization and evaluation of observational evidence suitable for input to the
last two tasks concerned with warranting claims of carcinogenicity. The first of these two
126
-------
remaining tasks (i.e. Task 3) is the warranting of carcinogenicity claims within each
observational context, i.e., for each context in Working Table 1 for which there is observational
evidence. The completed Working Table 2 for each such context is utilized with the relevance
strategies to be described in this section to form weight-of-evidence judgments of the support for
claims of carcinogenicity shown in the taxonomy (Table 1). This third task is specific to intra-
context warranting of claims, i.e., the claims apply to an observational context listed in Working
Table 1. For example, suppose the basis for Context 1, as described in Working Table 1, is a
long-term animal study with Wistar rats exposed to controlled levels of the agent. All of the
relevant context-specific data items from that study would be included in Working Table 2 (for
Context 1), as well as data items from other sources that are not context-specific but considered
directly applicable, e.g., historical control rates of tumors in Wistar rats, chemical properties of
the exposure agent, pharmacodynamic information, etc. (as summarized in the "Organism-
Specific Measurement" column under "Observed Effect"). The objective of Task 3 of hazard
identification is to assess support for carcinogenicity of the substance of interest vis-a-vis the
species, exposure levels, and other relevant conditions specific to each context, without
extrapolation of evidence across contextual boundaries. Working Table 3 (to be described)
provides a format to facilitate Task 3 for each observational context.
These working tables (given by one Working Table 3 for each observational and target
context) serve as input, along with information from other sources that may be considered
applicable (described in Section 5.4.), to the fourth and final task of hazard identification
concerned with inter-context warranting of claims. Observational evidence is often unavailable
or inadequate for the target context of exposure to humans, or humans at low exposure levels
typical of environmental conditions. In that case support for claims of carcinogenicity within the
target context depends on the intra-context judgments formed in Task 3 for observational
127
-------
WORKING TABLE 3. INTRA-CONTEXT SUPPORT FOR CLAIMS OF CARCINOGENICITY1
CONTEXT NO.
Claims of Carcinogenicity2
Reference
Strategy
I.O.3
Increases
Incidence of
Cancer
Classification(s)
Stage(s)
Mechanism(s)
Complete
Partial
Mixer
Helper
Neo.
Conv.
Neo.
Devel.
Geno-
toxic
Non-
genotoxic
Direct Empirical
(D.E.)
Semi-Empirical
Extrapolation
(S.E.E.)
Empirical
Correlation
(EC.)
Theory-based
Inference
(T.B.I.)
Existential
Insight (E.I.)
Column Summary
Overall Summary
Top half of each entry is completed using dose-response data observed in the context. Bottom half of each entry includes
"floater-data" as well (see text).
Choices for cell entries are Hi/Me/Lo/No. See text of Section 5.1. This assignment is made independently of the assignment
of intellectual obligation.
"Intellectual Obligation" (Hi/Me/Lo/No). See text of Section 5.1. This assignment is made independently of the assignment
in the cells of the table (footnote 2).
128
-------
contexts, and the availability of supporting data needed to extrapolate claims of carcinogenicity
across contexts. The third and fourth tasks are described in detail in Sections 5.2., 5.3. and 5.4.
In the interim, the discussion turns to a general consideration of strategies by which
observational data are given relevance to tasks. Throughout the present chapter, a common
example of the carcinogenicity of inhaled radon is employed, with the observational base and
warrants being those discussed in the NRC (1988) and NRC (1991) reports on radon. The
former report is concerned primarily with a discussion of data in the categories of "tumor
response", "related substances" and "biophysical effects", and the latter report with a discussion
of the role of "pharmacodynamics", "host characteristics" and "concurrent environmental
conditions" data.
5.1. Relevance Strategies
Tasks 3 and 4 of hazard identification require the analyst to judge the levels of support
for claims of carcinogenicity. Fundamental to this warranting of claims, whether intra-context or
inter-context, is the epistemic basis on which observational information is judged to be
supportive evidence of a claim. The five alternative categories of evidential reasoning available
to the analyst, referred to here as relevance strategies, are described below. The characteristics
of the available observational data used in a warrant for a claim of carcinogenicity, as assessed
by context in Working Table 2, in conjunction with a judgment of the support for any
background premises necessary for use of the data in lines of reasoning, determine the strength
of the warrant; the relevance strategy employed is the type of warrant. More than one type of
warrant may be applied to data to support a claim of carcinogenicity and, as discussed in a later
section (Section 5.5.), the coherence of support across warrants is an important consideration in
the overall judgment of the epistemic status of a claim.
129
-------
(1) Direct Empirical fD.E.V This category involves the claim that a given
observation constitutes a direct observation of the property under debate (such as
carcinogenicity). For example, observational data on inorganic arsenic includes
epidemiologic studies relating skin cancer to arsenic levels in well water and
clinical observations linking treatment of acne with an arsenic solution to
increased occurrence of skin cancer. Interpretation of these studies alone as
evidence for a claim that arsenic increases the risk of skin cancer in humans (at
some exposure level) would be an example of warranting by direct empiricism.
Similarly, there exist studies relating exposure of human populations to radon in
the home and elevations in lung cancer incidence (see, for example, the review by
Samet, 1989), although the coherence of these studies is low since they do not
uniformly indicate an increase in lung cancer incidence within exposed groups.
The strength of the warrant would be based on the characteristics of the study
data. For example, the strength of a warrant would be increased by establishing
that the studies were of high quality and the elevated incidence was not the result
of confounding factors (such as the presence of other chemicals that may play a
role) or unusual antecedent conditions (such as atypical host characteristics). It
must also be demonstrated that the studies were conducted within the defining
context. Clearly, as these background premises are better satisfied, the strength
of the direct empirical warrant improves. In the case of the NRC (1988) report,
direct empirical studies are mentioned but their utility is judged to be too weak
to justify use of the direct empirical warrant.
(2) Semi-Empirical Extrapolation rS.E.E.V This category involves the claim that a
given observation was not obtained under the desired context of exposure, but
that any result of the observation may be extrapolated to the desired context.
Basically, the claim is to having observed a pattern in the available exposure-
response data which may be followed to the described context. For example,
dose-response curves obtained on mining populations and experimental animals
exposed to high concentrations of radon have shown no evidence of a threshold
down to exposures an order of magnitude above those occurring in homes (NRC,
1988). While not providing direct empirical evidence of effects in the home, the
data do provide a less convincing warrant for such effects, conditional on the
acceptance of the background premise that the mechanism of carcinogenicity is
fundamentally similar at high and low exposures and that the exposure-response
pattern is evident in the available data. Clearly, as any patterns in the data
become better established, and as the background premises already discussed in
(1) above are supported, the strength of the semi-empirical warrant improves.
Examination of Figures 2.A.I. through 2.A.4. in the NRC (1988) report suggests
that exposure-response patterns are not clear in the data but generally are
monotonic without an evident threshold. A more germane finding is that trend
tests are statistically significantly positive across the four studies examined.
(3) Empirical Correlations fE.C.1. This category involves the claim that a particular
type of observation (such as of in-vitro transformation) is correlated with the
desired property (such as carcinogenicity). There is no claim to understanding
why the correlation exists, this claim being the function of the fourth category of
relevance strategy discussed below. Clearly, as the strength and specificity of the
130
-------
correlation improves, so does the strength of the warrant for empirical
correlations. It is important to note that observational evidence on substances
other than the one of interest generally is used in warranting claims to
correlations (see the "Related Substances" data category in Working Table 2),
requiring a judgment as to how sets of substances are to be chosen in developing
correlations. The NRC (1988) report does not explicitly cite empirical
correlations as a warrant, although (as discussed in Section 5.2.3.) correlations
between induction of chromosomal aberrations and carcinogenicity are cited to
strengthen the claim that radon is expected to be a carcinogen or at least induces
a BSDR.
(4) Theory-based Inference (T.B.I.Y This category involves the use of data (such as
in-vitro cellular transformation or bio-activation processes) which do not yield
claims to observing cancer but do represent claims to observing an effect related
to the etiology of cancer. In this case, the role of background premises appearing
in theories of carcinogenesis becomes of central importance. Clearly, the
strength of theory-based inference improves as the evidence for the background
premises embodied in theories improves. It is important to note here that other
bodies of observational evidence, perhaps drawn from substances other than the
one of interest, might be invoked as warrants for a particular background premise
appearing in an etiologic (mechanistic) theory. For example, the premise
concerning a role of adduct formation in neoplastic conversion might be
warranted by data from other substances in which adduct formation has been
shown to play such a role. In the NRC (1988) report, information on cellular
radiobiology, particularly as this relates to the demonstrated ability of radon to
induce chromosomal damage and transformation of cells, is invoked to support
the claim that radon is a carcinogen. The invocation of theory-based warrants
justifies a claim to both empirical and conceptual success (see the appendix).
(5) Existential Insight fE.I.V As discussed in the appendix, there may be times when
a scientific expert judges that an observation suggests carcinogenicity, but the
expert is unable to explain how this judgment arose. The judgment is a product
of personal experience and a-logical (as opposed to illogical) reflection, as in the
case of "engineering judgment". While not strictly a logical warrant, the analyst
still may deem the judgment rational and wish to factor it into the analysis.
Clearly, the strength of this "warrant" improves as it is better shown that the
expert possesses (1) the necessary prior experience from which such judgments
might spring and (2) the skill of reflecting on experience and producing reliable
existential judgments (see the discussion in Section 5.2.). The only explicit use of
this mode of warrant within the NRC (1988) report is in the assignment of priors
for the Bayesian analysis of the mining data.
The preceding discussion of relevance leaves two issues unaddressed. First, some
measure of the quality of a given relevance strategy (for each of the five strategies above) must
be assigned. This measure, again on a scale of low to high, is determined through reflection on
131
-------
the quality of the observational data employed in using the strategy (from Working Table 2) and
by the strength of evidence for the background premises required by invocation of the strategy
(see the discussion of premises in Section 5.2.). The necessary background assumptions are
specific, in turn, to the particular taxonomic carcinogenicity claim, but are of the general form
noted in the discussions of the five relevance strategies above. For example, use of information
on in-vitro production of chromosomal aberrations by radon (Brandom et al., 1978) in
warranting a claim of carcinogenicity requires such background premises as: (1) the necessary
chromosomal damage occurs both in-vitro and in-vivo, (2) chromosomal aberrations result in
necessary DNA alterations, (3) the DNA alterations induce transitions, etc. An assignment of
"low" implies that the background premises have not been verified for the substance of interest
and the desired taxonomic claim, and an assignment of "high" implies strong verification of these
premises for the particular claim. These judgments are made in isolation from consideration of
the strength of the claim that chromosomal aberrations have, in fact, been observed.
The second issue of warranting the use of relevance strategies (as opposed to the
strength of those strategies irrespective of use) relates to the existence of five different strategies
for relevance. Regardless of the strength of warrant for the premises underlying a given
relevance strategy (such as the outcome of examining direct empirical evidence or existential
insight), the analyst must provide an indication of the degree to which the analyst deems it
rational (a priori) to base decisions on a particular mode of warranting claims to carcinogenicity.
Another way of putting this is that the analyst determines the degree to which availability of a
given strategy is an "intellectual obligation" in rationally justifying a claim to carcinogenicity.
Again, the scale is from low to high. An assignment of "low" implies that the analyst does not
find a particular relevance strategy to be reliable generally and/or a proper conception of the
"intellectual obligations" (Alston, 1985) required of a rational decision-maker (as might be the
132
-------
case in existential insight). An assignment of "high" implies that the analyst finds a particular
relevance strategy to be reliable generally and/or a proper conception of the intellectual
obligations required of a rational decision-maker (as might be the case in direct empirical
evidence). It is important to bear in mind that this assignment is independent of the evidence
available for the premises in each particular strategy in each particular case (i.e. for each
substance considered). It is formed, instead, on the basis of the epistemic status of a relevance
strategy in general as discussed in the appendix. It is a form of "intellectual obligation" imposed
consistently on the analyst during any act of warranting and requires coherent application of the
same obligation to each specific instance of warranting. In other words, broad principles of
epistemic reasoning should be applied consistently throughout the analysis in the absence of
counter-arguments as to why a relevance strategy might be more acceptable in one act of
reasoning than in another. This assignment is included here due to the fact that some
individuals may generally trust or distrust particular forms of reasoning (such as reliance on
existential insight or etiologic theories) and yet may apply this trust or distrust inconsistently
within an analysis or across analyses. While not discussed explicitly in the NRC (1988) report,
the degree of intellectual obligation implicitly assigned to "Direct Empirical" warrants must have
been low given the fact that the entire analysis of risk is based on semi-empirical evidence (i.e.
the evidence from high exposures in the mines) and theory-based inference (particularly through
use of the radiobiological experimental data).
5.2. Application of Relevance Strategies to Warrant Intra-Context Claims of Carcinogenicity
The role of data categories in carcinogenicity claims is described in Table 7, with
reference to the relevance strategies defined in the last section. The relevance strategies
available to each combination of data category and carcinogenicity claim are displayed in Table
133
-------
8 for reference. In this section the relevance strategies are applied to the data items in Working
Table 2 to formulate judgments of the level of support under each type of evidential reasoning
for each claim of carcinogenicity in the taxonomy. Working Table 3 is suggested as a format for
displaying the judgments. Cells in the bottom row of the table, labeled "overall assessment", are
for summary judgments from each column after the main part of the table is completed and
assessed for coherence across the five potential types of relevance strategy (see discussion in
Section 5.5.).
The reader will note that the cells of Working Table 3 (aside from the "summary" cells)
are divided by a dashed line. This division arises due to the presence in Working Table 2 of
some data specific to an exposure-context and others specific to the organism but not measured
in the context of exposure at the level of interest (contained under the sub-column "Organism-
Specific Measurement" in Working Table 2). The latter data may appear in several contexts (i.e.
several instances of Working Table 2), allowing the same data to play multiple roles in the
analysis perhaps out of proportion to the significance of the data (since they do not represent
empirical values obtained under exposure conditions). To prevent this, the judgments for cells
in Working Table 3 first are formed on the basis of only data from columns "Exposure-Effect"
and "Context-Specific Measurement" in Working Table 2. These judgments, placed into the cells
of Working Table 3 above the dashed lines, "root" the claims of the analyst in the primary data.
The secondary data, given by the "Organism-Specific Measurement" column in Working Table 2,
then are analyzed to "perturb" the primary judgments of the analyst, strengthening or weakening
the pre-existing assignments. These modified assignments then are placed into the cells of
Working Table 3 below the dashed lines. It is these modified assignments that will be used in
producing the summary assignments for the table (see Section 5.5.). The assignments
themselves are on a scale from "Hi" to "Lo" as discussed in Section 5.1.
134
-------
TABLE 7. RULE OF DATA CATEGORIES IN CARCINOGENICITY CLAIMS
Data Category1
Role of Data Categories in Carcinogenicity Claims
Tumor
Response2
Direct empirical support for "increases cancer", conditional upon causal
premise. Direct empirical support for "complete carcinogen" if
concurrent environmental conditions or previous/later
initiation/promotion conditions were not required to yield cancers.
Direct empirical support for "partial carcinogen" if previous/later
initiation/promotion conditions were required to yield cancers or if
observed effect is only an increase in multiplicity of tumors. Direct
empirical support for either "mixer" or "helper" claims if concurrent
exposure must be present (but incapable of distinguishing between the
two claims). Provides empirical correlation and/or existential insight
warrant for any of the above claims. This data category has no bearing
on claims to "stages" and "mechanisms".
Biophysical
Effects
Data items provide Theory-Based Inference warrants for the claims of
"increases incidence with cancer", "complete carcinogen" and "partial
carcinogen". The ability to distinguish between a "complete" and "partial"
carcinogen requires Biophysical Effects data indicating that the factor
induces all necessary transitions or only a subset of such transitions. The
data items can also provide empirical correlation and existential insight
warrants for the above claims, but direct empirical and semi-empirical
warrants are not available. If the data are an indicator of transitions,
they may provide direct empirical, theory-based inference, empirical
correlation and/or existential insight warrants for the "stage" claims. If
the data items are an indicator of conditions necessary for transitions,
they may provide the above warrants for the "mechanism" claims. If any
of the data items were obtained under conditions of concurrent
exposures, they may provide direct empirical, empirical correlation
and/or existential insight warrants for the classification as "mixer" or
"helper".
Host
Characteristics
This data category does not provide a warrant for claims to
carcinogenicity. The data affect only the decision to include a given
organism/study within a given context.
Pharmaco-
dynamics
These data support only the contention that a BSDR is produced in the
organism. Therefore, they do not directly warrant intra-context claims to
carcinogenicity. They can, however, provide a theory-based warrant for
the contention that any effects observed ("tumor response" and/or
"biophysical effects") are causally connected to exposure by warranting
the contention that at least a BSDR is produced in the organism.
Concurrent
Environmental
Conditions
These data support the contention that an observed effect (e.g. cancer
prevalence) was connected causally to exposure to the factor of interest.
The data also provide Theory-Based Inference support for the
"classification" of carcinogenicity claims if it is shown that concurrent
exposures are (mixer/helper) or are not (complete/partial) required to
yield the observed effect. The data have no implications for the other
claims to carcinogenicity.
(continued on following page)
135
-------
Table 7. (continued)
Data Category1
Role of Data Categories in Carcinogenicity Claims
Structure
Activity
Relationships
These data provide Theory-Based Inference, Empirical Correlation,
and/or Existential Insight warrants for the claims of "increases incidence
of cancer", "complete carcinogen" or "partial carcinogen". The specific
claim warranted depends upon the structural feature of the molecule and
the presence of the appropriate mechanism of action within the
organism. If, as is often the case, the structure indicates an initiating
agent, the data provide support for a partial carcinogen with no ability to
distinguish further between a partial and complete carcinogen. The data
do not pertain to the remaining classifications (mixer/helper). They do,
however, provide warrants for the claim that a substance acts on a given
stage or by a given mechanism, particularly with respect of initiation
(neoplastic conversion and genotoxicity).
See Working Table 2.
Quantitative measures are shown under "data items" for "tumor response" only.
Quantitative measures commonly used are not given in other data categories, only the
endpoints of interest.
136
-------
TABLE 8. RELEVANCE STRATEGIES AVAILABLE FOR CLAIMS OF CARCINOGENICITY
Claims of Carcinogenicity1
Data
Category
Increases
Incidence of
Cancer
Classification(s)
Stage(s)
Mechanism(s)
Complete
Partial
Mixer
Helper
Neo.
Conv.
Neo.
Devel.
Geno-
toxic
Non-
genotoxic
Tumor Response
D.E.
E.C.
E.I.
D.E.
E.C.
E.I.
D.E.
E.C.
E.I.
D.E.
E.C.
E.I.
D.E.
E.C.
E.I.
D.E.
E.C.
E.I.
D.E.
E.C.
E.I.
Biophysical
Effects
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
D.E.
E.C.
E.I.
D.E.
E.C.
E.I.
D.E.
T.B.I.
E.C.
E.I.
D.E.
T.B.I.
E.C.
E.I.
D.E.
T.B.I.
E.C.
E.I.
D.E.
T.B.I.
E.C.
E.I.
Host
Characteristics
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
N.A.
Pharmaco-
dynamics
N A.
N.A.
N.A.
N.A.
NA.
N.A.
N.A.
N.A.
N.A.
Concurrent
Environment
Conditions
N.A.
N.A.
N.A.
T.B.I.
T.B.I.
N.A.
N.A.
N.A.
N.A.
Related
Substances
Assessments
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
Structure Activity
Relationships
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
N.A.
N.A.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
T.B.I.
E.C.
E.I.
N.A.: None Available
D.E.: Direct Empirical
S.E.E.: Semi-empirical Extrapolations
E.C.: Empirical Correlations
T.B.I.: Theory-based Inference
E.I.: Existential Insight
137
-------
5.2.1. Direct Empirical Warranting
Direct empirical warrants are available at times when a data item may be taken as
directly indicative of carcinogenicity (i.e. as a warrant, in and of itself and with minimal
theoretical interpretation, for a carcinogenicity claim). As shown in Working Table 3, the
separate claims to be warranted are that:
(1) The substance (or exposure factor) increases the incidence and/or time of
appearance of cancer in a specific context (exposure level and species).
(2) The substance is a partial carcinogen, complete carcinogen, "mixer" or "helper".
(3) The substance acts on the organism through neoplastic conversion and/or
neoplastic development.
(4) The substance acts on the organism through genotoxic and/or non-genotoxic
mechanisms.
Turning first to the claim that a substance increases the incidence of cancer, the direct
empirical warrant requires observations on the incidence of cancer within the context of interest.
As a result, only observations in the category of tumor response may provide the basis for a
direct empirical warrant. The strength of the warrant then increases as (1) the completeness of
the observational base in a data item is improved, (2) the quality (utility) of the observational
studies increases, (3) the strength of the association (dose-effect) increases and (4) the claim to
a causal connection between exposure and the observed incidence is supported (see Working
Table 2). In the case of radon, direct empirical evidence was available (Samet, 1989), but the
utility, strength of association, causal connection and coherence was low to moderate.
For the distinction between the various classifications of carcinogens (complete, partial,
mixer and/or helper), direct empirical warrants require additional background premises and,
hence, additional observational evidence. For classification as a complete carcinogen (in the
row labeled "Direct Empirical" in Working Table 3), the necessary premise is that the substance
may increase cancer without the presence of concurrent exposures within the considered context.
138
-------
This requires tumor response data in which (1) the role of the substance as the sole causal agent
(aside from contributions from non-experimental factors) is well established, warranted by the
data category "concurrent environmental conditions" as summarized in the causality claim in
Working Table 2 and (2) cancer incidence was increased, warranted by the "tumor response"
data category. Classification as a partial carcinogen requires introduction of the premise that
deliberate (experimental) exposure to either an initiating, promoting or progressing agent must
accompany exposure to the substance of interest, requiring assays demonstrating empirically that
exposure to the substance must be accompanied by (previously, concurrently or subsequently)
exposure to either an initiating or promoting agent. Radon has been suggested to act primarily
through initiation at low doses and, hence, is to be considered primarily a partial carcinogen.
Classification as a mixer requires introduction of the premise that there be concurrent exposure
to another substance that is not itself a carcinogen, and empirical evidence that exposure to the
substance of interest must be accompanied by "concurrent environmental conditions" (see the
discussion of causality for Working Table 2) constituted by exposure to that other non-
carcinogenic substance. For classification as a helper, the same premises apply with the
exception that the "other" factor must be a carcinogen.
For the distinction between the stages (neoplastic conversion and/or neoplastic
development), the analyst must introduce the additional premise that the substance acts
explicitly to yield neoplastic conversion (as in the case of radon) and/or neoplastic development.
Direct empirical support for these claims may be taken as observation of biophysical effects (see
Working Table 2) such as cellular transformation (NRC, 1988; Woodruff, 1990b; Barrett et al.,
1986a, 1986b; Bartsch and Malaveille, 1990) and progression of preneoplastic lesions (Bannasch
et al., 1987), or from characteristics of the tumor response such as alterations in the time-to-
139
-------
appearance but not the incidence of tumors (indicating an effect on the "speed" of neoplastic
development) etc., following exposure within the desired context.
Finally, the mechanistic distinctions require premises that the substance acts through
changes in DNA (genotoxicity, as in the case of radon (NRC, 1988)) or changes in other
biological structures/functions (non-genotoxicity) (Barrett and Wiseman, 1987; Barrett, 1987;
Butterworth, 1990). Direct empirical support for premises is, again, given by observations of
biophysical effects, although of a more fundamental nature than those required in distinguishing
stages. The observational support for genotoxicity generally is taken as formation of DNA
adducts, DNA breakage, chromosomal aberrations, oncogene activation/deactivation and/or
base pair alterations (for reviews of the evidence for causal roles, see Barrett and Wiseman,
1987; Barrett, 1987; Butterworth, 1990; Belinsky et al., 1987c; Farber, 1987; Lawley, 1987;
Morris, 1990; Perera, 1987). Evidence for the role of chromosomal damage is radiation-induced
cancer is summarized in Hall and Freyer (1991). The observational support for non-genotoxic
mechanisms generally is taken as changes in inter-cellular communication, interference with
hormonal control, immune system destruction, hyperplasia induction, etc. (for reviews of
evidence for causal roles, see Langenbach et al., 1988; Lewis and Adams, 1987; Loury et al.,
1987; Moolgavkar, 1986; Perera, 1984; Scribner et al., 1987; Slaga, 1984; Swenberg and Short,
1987; Trosko and Chang, 1988). Crawford-Brown and Hofmann (1990) have proposed a non-
genotoxic mechanism by which cytotoxicity of radon results in promotional effects.
522. Semi-Empirical Warranting
Since the present section (5.2.) is directed towards intra-context claims, instances of
semi-empirical extrapolation are not appropriate at the second level of warrant. This form of
warrant will be described in Section 5.3., which deals with extrapolation across contexts
140
-------
(including extrapolation across exposure levels within another homogenous context, the subject
of semi-empirical warrants).
5.2.3. Warranting from Empirical Correlations
Correlational warrants require knowledge aside from the evidence on tumor response for
the agent of interest. Construction of sets within which correlation coefficients (strength and
specifity) between a data item and a carcinogenicity claim may be constructed requires premises
detailing how those sets should be constructed. In other words, it must first be specified how a
data item is to be chosen as a candidate for inclusion in the set of items for which the
correlation has been developed. These premises, in turn, are related to theories of
carcinogenesis. As a result, empirical correlations may be closely related to etiologic theories.
Although distinctly different in nature, it is convenient to discuss these two categories (theory
and correlation) together because empirical correlations have been used to support or question
cancer paradigms and thus have had some influence on the direction of their development. For
example, only a few years ago many analysts would have appealed to theory-based inference to
warrant a claim of "non-carcinogen" if an agent tested negative in short term tests (STTs) for
mutagenicity, such as the Ames salmonella bioassay. In this instance, the major premise of the
theory was that carcinogens acted through mechanisms of geonotoxicity to alter DNA
structure/function. But these predictions did not correlate well with observational evidence
from long-term animal studies, causing existing paradigms of cancer mechanisms to be modified
or refined. By the same token, theories of carcinogenesis (such as the influence of mitogenesis)
have altered the construction of sets within which correlation coefficients are developed. This
example will be discussed further, but it illustrates the inter-relationship of the predictive
capacity of empirical correlations and theory-based inference following from the partial
141
-------
warranting of cancer paradigms by empirical evidence, i.e., when theory-based predictions are
not consistent with the empirical record, then the theory is re-examined (and vice versa).
The introduction of bacterial and cell culture techniques to measure chemically-induced
mutations led to a widely held belief that mutagenicity and carcinogenicity were highly
correlated. In particular, the Salmonella/mammalian microsome test of Ames et al. (1975), took
center-stage as a predictor of carcinogenicity based on the short-term test for mutagenicity.
Somatic cell mutation was often considered a necessary initiating step in cancer. In a recent
review, Ashby (1990) summarizes "the simple paradigms of 1975" as follows: "Human
carcinogens are also carcinogenic to rodent; rodent carcinogens can be assumed to be human
carcinogens. The majority, if not all, organic rodent carcinogens are reactive to chromosomes or
their constituent DNA (genotoxic), usually following appropriate metabolic conversion. The
Salmonella mutation assay, coupled with a few other in vitro genotoxicity assays, is sufficient for
the detection of such genotoxic chemicals. With cancer, it is unwise to assume a safe-dose level
of human exposure. Thus, at its simplest level the activity of a chemical in the Salmonella assay
implies a cancer hazard to exposed humans." In contrast, in what he refers to as the "worst-case
scenario in 1990", Ashby notes that "No new human carcinogens have been defined since 1975:
many new rodent carcinogens are now known, many of which current data indicate to have no
cancer hazard for humans...a large and growing number of rodent carcinogens are non-
genotoxic,. i.e., they cannot be detected using current genotoxicity assays. Further, a large and
growing number of in vitro genotoxins appear to be devoid of rodent carcinogenicity."
The loss of simplicity in the last fifteen years is partially explainable by the increased
awareness that cancer does not develop by a single mechanism, common to all species and
routes of exposure wherein an agent may increase cancer incidence. Indeed, the primary
biological activity of an agent or a metabolite may not be genotoxic, as indicated by point
142
-------
mutations, insertions, deletions, or changes in chromosome structure or number. Chemicals that
exhibit such genotoxic activity can usually be detected by assays that measure reactivity with
DNA, induction of mutations or DNA repair, or cytogenetic effects (Butterworth, 1990). The
earlier conceptual simplicity of a high correlation between mutagenicity and carcinogenicity was
further discredited by evaluation of the results from long-term rodent assays conducted by NCI
and NIEHS. In an initial article results of four widely used in vitro assays for genetic toxicity
applied to 73 chemicals recently tested in the two-year NCI/NIEHS long term rodent studies
were compared (Tennant et al., 1987a, 1987b). The chemicals selected were well characterized
in rodents for carcinogenicity or non-carcinogenicity. Of the four assays (for induction of
mutations in Salmonella (SAL) and mouse lymphoma L5178Y cells (MLA), and induction of
sister chromatid exchanges (SCE) and chromosome aberrations (ABS) in Chinese hamster ovary
cells). It was found that SAL detected only about half of the carcinogens as mutagens and the
other three assays (MLA, SCE, and ABS) did not complement SAL, i.e., there was no gain from
using other any of the other three assays in addition to SAL. These results were confirmed by
examination of 41 additional chemicals (Zeiger et al., 1990).
The disappointing sensitivity of genetic toxicity tests to predict rodent carcinogenicity was
compounded by the results for specificity reported earlier by Shelby and Stasiewicz (1984) who
found that more than 60% of 80 rodent non-carcinogens had been found active in at least one
of the four in vitro tests. These results suggest that in vivo testing may be complementary to in
vitro tests. NTP is currently evaluating in vivo tests for chromosome aberrations and micronuclei
as a means of confirming in vitro mutagenicity (Zeiger et al., 1990). It may also be noted here
that the second Collaborative Study on the Assessment and Validation of Short-Term Tests for
Genotoxicity and Carcinogenicity (CSSTT/2) concludes that "in vivo tests have a vital role to
play in hazard assessment. This role is to define which chemicals, identified as genotoxic from
143
-------
in vitro tests, are active in vivo and, thus, are those most likely to present a
carcinogenic/mutagenic hazard to mammals, including humans" (WHO, 1990).
Further correlational analyses of NCI/NTP rodent studies have been conducted that are
informative for claims of relevance in hazard identification based on empirical correlations
and/or theory-based inference (Ashby and Tennant, 1988; Ashby, 1990; Tennant and Ashby,
1991; and Ashby and Tennant, 1991). Results reported in the last citation, entitled "Definitive
relationships among chemical structure, carcinogenicity, and mutagenicity for 301 chemicals
tested by the U.S. NTP", are excerpted in the following discussion. A high correlation was
observed between structural alerts to DNA reactivity and mutagenicity, but neither predicted
rodent carcinogenicity effectively. It appears that certain tissues are only sensitive to genotoxic
carcinogens, e.g., zymbal's gland and the lung. Other tissues in which cancers were observed are
subject to both genotoxic and non-genotoxic mechanisms. It is noted that "non-genotoxic rodent
carcinogens cannot be automatically neglected as possible human carcinogens. However, if non-
genotoxic carcinogens induce their effects via an interaction between the test agent and the
sensitive rodent tissue, the relevance of that interaction to human tissues should be studied,
particularly at low dose-levels, before a human carcinogenic hazard is assumed. Such questions
of extrapolation are distinct from those posed by genotoxic rodent carcinogens where
comparative metabolism and the intrinsic genotoxic potency of the test agent are the critical
parameters, DNA itself being common to rodents and man."
"Non-genotoxic" is a non-descript classification simply referring to any mechanism other
than genotoxic and, as discussed above, understanding the specific mechanism may be key to
predicting whether a rodent carcinogen is a cancer hazard to humans or whether increased
cancer incidence at a high exposure level constitutes a hazard at environmental concentrations
(a particular concern with respect to mitogenic action). As noted by Butterworth (1990) and
144
-------
other authors, a mechanism may appear to include both genotoxic and nongenotoxic elements.
For example, non-DNA reactive promoters that induce cell proliferation can yield mutagenic
events and chromosomal alterations secondary to that proliferation while mutagens given at
cytotoxic doses may induce cell proliferation. In practice, however, it may be useful to apply a
weight-of-evidence approach to classify a chemical as genotoxic or nongenotoxic, noting that the
latter category is a class of mechanisms that may not depend solely on chemical properties, but
the context of exposure (concentration, species, route of administration, etc.). Butterworth
(1990) suggests that such an approach might include "examination of constituents of the
molecule for prediction of DNA reactivity, overall results from cell culture assays, activity in the
whole animal, dose and target organ specificity, histopathological evaluation of tumor
development, and other obvious potential mechanisms of carcinogenicity (such as cell
proliferation) (Ashby and Tennant, 1988)". Cohen and Ellwein (1990) also recommend
classifying chemical carcinogens as genotoxic or nongenotoxic, with further division of the latter
category by mechanisms of action, if known. In particular, it is of interest to note whether the
chemical acts through a cellular receptor or a non-cellular mechanism. Cohen and Ellwein note
that agents acting through specific receptors tend to be active at low doses, making it unclear
whether there is a threshold level of exposure.
Ashby and Morrod (1991) summarize some current concerns on carcinogenicity testing
and their interpretation for hazard identification in light of recent developments discussed
above, principally the limited predictive capacity of short term tests for genetic toxicity and
heightened awareness of the prevalence and diversity of nongenetic mechanisms in long term
animal studies. Their suggested approach to testing for carcinogenicity has implications for the
type of data needed to warrant claims of carcinogenicity, of interest in the present report.
Ashby and Morod recommend an initial step of inspecting the chemical structure of an agent for
145
-------
sites of DNA-reactivity, followed by assessment of genotoxicity using in vitro tests and if
necessary, short-term rodent tests. This will identify potential genotoxic carcinogens. An issue
in genetic toxicology is the use of in vivo tests to separate in vitro genotoxins that are rodent
carcinogens from those that are noncarcinogenic to rodents, and the use of complementary
genotoxicity assays to detect the few Salmonella-negative genotoxins, such as benzene. The
authors note that the vast majority of classical human and rodent chemical carcinogens, together
with most of the NTP two-species carcinogens, are overtly genotoxic in vitro and in vivo. The
next step is to evaluate the remaining genotoxins for a range of toxicities associated with non-
genotoxic rodent carcinogens, noting that agents found to be inactive are probably not
carcinogenic. The authors provide a list of 14 potential non-genotoxic indicators, i.e., there is
some supportive evidence of association for the indicators but none has been established
definitively as a predictor. Without an explanatory mechanistic link with cancer at present, it is
suggested that the absence of each indicator is supportive of "probable non-carcinogenicity". It
is noted that many issues about non-genotoxic carcinogenesis remain to be understood. (Note:
The authors claim nothing new in their suggested scheme. Citations and credits have not been
included in the abbreviated discussion provided here.)
The discussion above is intended to provide an introduction to some of the issues that
need to be addressed in hazard identification and to give some idea of the type of mechanistic
information that may be usefully applied in an weight-of-evidence approach to hazard
identification. Attention is now directed to additional sources of empirical correlations for
warranting claims of carcinogenicity.
The 114 NCI/NTP studies used for comparison of four in vitro tests of genetic toxicity by
Zeiger et al. (1990) and Tennant et al. (1987a), described above, are evaluated for correlational
characteristics in Haseman and Clark (1990). The 114 chemicals are all those tested by
146
-------
NCI/NTP during a specified time period, thus eliminating the possibility of selection bias. The
NCI/NTP claims of carcinogenicity are as follows: 67 (59%) were carcinogens in at least one of
the four sex-species groups (male/female, rats/mice), the evidence for 17 (15%) was found to be
equivocal, and 30 (26%) were judged to be non-carcinogens. For interpretation of these figures
and the ones to follow, it is necessary to understand how NTP assigns a "category of evidence"
to a chemical based on evidence in rodents studies (usually of rats and mice of both sexes).
As described by Haseman and Clark, the final decision as to whether a study was positive,
equivocal, negative, or inadequate is a matter of scientific judgment. Biological factors are
considered as well as the statistical outcome, including: whether a dose-response was observed;
whether pre-neoplastic lesions were observed; the historical control rate, i.e., the accumulated
evidence from previous control groups regarding the spontaneous occurrence of observed
neoplastic lesions; biological characteristics of the lesions observed; the survival of dosed and
controlled animals; whether tumor latency was affected by dose; the multiplicity of site-specific
neoplasia; the adequacy of the experimental design and conduct of the study; and the
consistency of occurrence across sexes within a species, and across all four sex-species groups.
Consequently, "rodent carcinogenicity" refers to an assigned category based on human evaluation
of quantitative and qualitative characteristics of observational data from the long-term animal
studies. With awareness of factors considered in forming a claim of rodent carcinogenicity, the
correlations noted by Haseman and Clark in their review of the evidence from 114 chemical
studies will be described.
The inter-species concordance is 69% (compared to 74% for the whole NCI/NTP
database reported by Haseman and Huff (1987)). The concordance of 69% is comparable to
the concordance of 66% between the Salmonella-assay and rodent carcinogenicity for these
chemicals reported in Zeiger et al. (1990). The 67 carcinogens are unlikely to represent an
147
-------
equal risk to humans, according to Haseman and Clark, who conclude that carcinogens
producing effects at multiple sites and/or in multiple sex-species groups are probably more
important from a public health standpoint than single-sex and single-site carcinogens, or those
where carcinogenic sites were accompanied by organ toxicity. Haseman and Clark draw some
conclusions regarding the effect of chemicals that are toxic, but this does not refer to evidence
of organ toxicity. They refer to toxic potency of the chemical, i.e., chemicals with low maximum
tolerated doses (MTDs) are more "toxic" than those with high MTDs. The intended significance
is not clear, but it is reported that most (67-88%) of chemicals positive in STTs (referring to the
four in vitro tests described above that were applied to this set of chemicals, namely SAL, SCE,
ABS, and MLA) were also "toxic" and the majority (62-78%) of negative STT chemicals were
"nontoxic". Overall concordance between toxicity and rodent carcinogenicity was 65%, about the
same as the same as reported for STT outcomes and carcinogenicity in this set of chemicals. It
was found that all four STTs were correlated with chemical toxicity, particularly SCE and MLA
(P < 0.001). The Salmonella-assay is more predictive of carcinogenicity among the toxic
chemicals (65% positive SAL in carcinogens vs. 12% in noncarcinogens) than in non-toxic
chemicals (with corresponding percentages of 13 of 3). ABS demonstrated a similar pattern but
with less predictive capacity, while SCE and MLA results showed no evidence of association
with rodent carcinogenicity even within classes of toxic or non-toxic chemicals.
In another review of the NTP database, Hoel et al. (1988) address the issue of whether
organ toxicity, particularly at high doses, may be associated with judgments of rodent
carcinogenicity. Of principle concern is the extent to which a secondary mechanistic process,
such as cytotoxicity with resultant compensatory cell proliferation, may be responsible for tumor
induction in chemicals classified as rodent carcinogens. Since such a mechanism would be
unlikely to occur at environmental exposure concentrations, the classification of carcinogenicity
148
-------
is more nearly an artefact of exposure to high doses than an indicator of a potential cancer risk
to humans. A total of 378 two-year studies on 99 chemicals conducted by the NTP were studied,
53 (54%) of which were considered to be carcinogenic in one or more of the laboratory-animal
studies. The authors conclude that "only seven of the 53 positive carcinogenicity studies
exhibited the types of target organ toxicity that could have been the cause of all observed
carcinogenic effects. Furthermore, no apparent difference in mutagenicity as measured by the
Ames Salmonella assay was observed between 'high dose only' carcinogens and the entire set of
carcinogens." Some qualifications are contained within the article, however, relating to issues
that "certainly merit further investigation in order to make a more definitive statement about
toxicity and carcinogenicity". The reference stresses the importance of clearly defining what is
meant by "organ toxicity" and "preneoplastic lesion" in discussions of either.
Wilbourn et al. (1986) considered the validity of extrapolating results from long-term
carcinogenicity tests in animals to humans by reviewing the 41IARC Monographs on the
Evaluation of the Carcinogenic Risk of Chemicals to Humans (7-47) that had been published or
were in press. The IARC claims of carcinogenicity are the judgment of working groups of
experts who consider both epidemiologic and animal data. It was found that "84% of the 44
exposures with sufficient or limited evidence of carcinogenicity to humans also have some
carcinogenic activity in animals. The remaining chemicals and complex mixtures have not been
adequately tested for carcinogenicity in animals. In no case was there 'no evidence of
carcinogenicity'". It is also found that "for many exposures causally related to human cancer,
there is a target organ in common between humans and at least one animal species, despite
many inherent physiological differences."
Allen et al. (1988) conducted quantitative comparisons of carcinogenic potency in
animals and humans for 23 chemicals for which suitable animal and human data exist. These
149
-------
comparisons, based on the TD^, were found to be strongly correlated. The best prediction of
human results from animal data was achieved by utilizing data from several routes of exposure
and employing the assumption that animals and humans are equally sensitive to a carcinogen
when dose is measured in units of mg/kg body weight/day.
52.4. Warranting by Theory-Based. Inference
While Section 5.2.3. contained material related to etiologic theories, the invocation of
theories in that section was intended (1) as a tool for establishing sets within which empirical
correlations between a data item and carcinogenicity could be defined and (2) as a tool for
examining the role of existing correlations in testing axioms (premises) for theories. The current
section discusses the role of theories in directly warranting claims of carcinogenicity through
judgments of conceptual understanding (rather than strength of empirical correlation). This
warrant requires explicit use of both observational evidence for effects believed to constitute
part of the causal sequence leading to cancer (aside from tumor response) and etiologic theories
of cancer uniting the evidence and giving it relevance to the task of supporting a claim of
carcinogenicity. The two broad premises that must be introduced are (1) the substance induces
a biologically significant dose-rate (BSDR) following exposure and (2) the resulting BSDR
induces cancer in one of the taxonomic forms of claims of carcinogenicity (see Working Table
3). Both premises must, of course, be warranted within the desired context. For example, in
the case of radon (NRC, 1988; Hall and Freyer, 1991; NRC, 1991), the observation of
chromosomal aberrations, even at exposures below those for which direct empirical
(epidemiologic) evidence was available, is taken as evidence that at least the necessary first steps
in inducing cancer occur at these low exposures. This suggests that a BSDR is produced at
150
-------
these exposures and that this BSDR results in damage believed to be etiological^ significant in
the production of transitions towards cancer.
It is important to note here that theory-based inference does not employ direct empirical
evidence of carcinogenicity as contained in tumor response data (see Working Table 2),
although it may employ direct empirical evidence in separately warranting the two broad
premises above (as was done in the case of radon). The intent, instead, is to determine the
implications of other available data (biophysical effects, pharmacodynamics, etc.) within the
framework of etiologic theories. This isn't to say that the analyst ultimately ignores the tumor
response data. Those data appear, instead, as direct empirical warrants for the claims. In
addition, "Related Substances Assessments" data might provide observational warrants for the
premises underlying etiologic theories themselves, conditional upon acceptance of the further
premise that there is a common etiologic foundation to carcinogenicity in the contexts of the
"related substance" and the substance of interest to the analyst (as when active smoking data are
used to produce a theory of carcinogenicity for ETS, although it should be borne in mind that
the premise of a common etiologic basis for the action of different carcinogens is increasingly
being questioned in light of findings on the multiple mechanisms through which carcinogens may
act). For radon, the general finding of the carcinogenicity of all forms of ionizing radiation has
been invoked (NRC, 1988) as a warrant for the claim that radon (which yields ionizing
radiation) is a carcinogen. It might be useful to think of theory-based inference as providing a
prior expectation of carcinogenicity for a substance, prior to consideration of the direct
empirical evidence. Theory-based inference also warrants a claim to conceptual understanding,
allowing the analyst to assert not just that the substance has been observed to induce cancer
(summarized in the "Direct Empirical cells of Working Table 3) but that the etiologic role of the
substance is understood within the framework of existing theories. Such an argument is
151
-------
particularly strong in the case of radiation, where mechanistic understanding is much better
developed than for chemical carcinogens. The ability to assert scientific understanding is an
important component of scientific rationality and was raised initially in Section 4.2.4. and in
more detail in the appendix.
The first premise (that a BSDR is produced following exposure) is itself warranted by
observations contained under pharmacodynamics in Working Table 2. These data are
summarized for radon exposures in the latter NRC report (NRC, 1991). The role of each of
these observations within the reasoning of the analyst (i.e. in supporting the judgment that a
BSDR is produced) is displayed in Table 7 and is not repeated here. The warrant for the first
premise may take any of the five forms discussed in Section 5.1. For example, the analyst may
have observations of a BSDR being produced in the desired context, leading to a direct
empirical warrant for the first premise (but not, of course, for the claim to carcinogenicity, since
this discussion is contained within theory-based inference with respect to the carcinogenicity
claims). The analyst may have observed only that the necessary enzymes for bioactivation are
present, resulting in a theory-based warrant for the first premise or in a warrant of empirical
correlation or existential insight. Again, any of the five warrants (relevance strategies) may be
available for warranting the first premise. For radon (NRC, 1991), it has been experimentally
confirmed that radon progeny are deposited in the lung and produce a dose-rate of radiation to
sensitive cellular subpopulations. The form of warrant for the premise is not included formally
in the working tables, but is factored into the judgment of the analyst concerning the strength of
the warrant for "Theory-Based Inference" in Working Table 3.
An identical discussion applies to the second premise, i.e. that a BSDR (when present) is
capable of inducing cancer in one of the senses noted in the claims to carcinogenicity found in
Working Table 3. This second premise also may be warranted by any of the 5 relevance
152
-------
strategies and, as in the first premise, the form of warrant for premise 2 is factored into the
judgment of the strength of the "Theory-Based Inference" warrant of Working Table 3 (without
appearing explicitly in that table). For radon (NRC, 1988), this premise is warranted primarily
be experimental observations that irradiation of cells in-vitro yields transformed cells. The role
played by each potential observational category (and, hence, data item) in any line of reasoning
lending support to the second premise is displayed in Table 7.
Many of the effects potentially measured in a study of exposures to humans, animals or
cell lines are not of cancer induction directly but rather of biophysical phenomena believed to
play a causal role in carcinogenesis. Use of these data to support a claim of carcinogenicity
within theory-based inference requires introduction of a premise that the bioeffect at least is an
indicator of one or more of the transitions leading to a tumor. In other words, the etiologic role
of the observed bioeffect within the process of carcinogenesis appears as an explicit premise
associated with the etiologic theory used in drawing a "Theory-Based Inference" in the cells of
Working Table 3. Support for this premise, in conjunction with observation of the bioeffect
itself, then strengthens the second premise that the substance-induced BSDR yields transitions.
The role of the bioeffects data in supporting the premise that BSDR results in cancer
occurs on two conceptual levels. The first level relates to the specific stages of carcinogenesis,
taken here to be neoplastic conversion and development (see the discussion in Chapter 2). The
primary data of use at this level are measurements of preneoplastic lesions, initiation ability,
promotional ability, in-vitro cellular mutation and in-vitro cellular transformation, all of which
are discussed in the NRC 1988 report on radon. Initiation assays are taken to support a
contention that the substance of interest induces at least the conversion stage, requiring the
additional premise that the further transition (development) occurs with some non-zero
background rate or substance-induced rate. A similar comment applies to the results of
153
-------
promotional assays. Preneoplastic lesions are taken as an indication of both conversion and
promotion having taken place, requiring a premise that progression eventually would occur in
some fraction of the preneoplastic lesions if sufficient follow up time was available (a
particularly important premise in extrapolation from studies on short-lived animals to humans)
(Bannasch et al., 1987)). It is important here to bear in mind the specificity of transitions
induced by a given carcinogen, which might imply that a partial carcinogen can complete the
necessary transitions only for specific forms of the other transitions not induced by that partial
carcinogen. In other words, existing data on carcinogenicity indicate that it is not generally true
that a given promoter (for example) is capable of promoting all initiated cells. It may, instead,
be capable of promoting only specific forms of initiation-related damage. It also should be
noted that bioeffects data support claims that a substance induces some of the necessary
transitions, but not that it is a complete carcinogen (unless the bioeffect is taken to be the
entirety of effect necessary for cancer, with only a probabilistic component remaining in the
production of a frank tumor).
The second level at which bioeffects data may support the premise (that BSDR results in
cancer) relates to observations on processes leading to transitions between stages, but not
observations of the stages or transitions themselves. It typically is premised that neoplastic
conversion results from changes to DNA (see Chapter 2) with a less well-established link
between DNA changes and neoplastic development. As a result, the ability of a substance to
produce DNA adducts, DNA breakage, DNA gene mutations, activation of oncogenes, or de-
activation of repressor genes may be taken as evidence of neoplastic conversion if it is premised
that either (1) these bioeffects are correlated with conversion or (2) these bioeffects play a
causal role in conversion. The first premise is warranted by reference to correlational studies
using principles analogous to those as discussed in Section 5.2.3. The second is warranted by
154
-------
reference to (a) explicit mechanistic understanding of the process of conversion or (b) a claim to
existential insight using principles analogous to those discussed in Section 5.2.5. Structural
characteristics of molecules have also been correlated partially with the conversion activity of
those molecules (Ashby and Tennant, 1988, 1991; Ashby et al:, 1989). Again, the ability of
radon-associated radiation to produce chormosomal damage may be taken as support for the
premise that a BSDR results in transitions and, hence, cancer (NRC, 1988; Hall and Freyer,
1991).
A similar situation holds for links between the second level of causal evidence and the
stage of neoplastic development, with several important differences. First, the mechanistic
understanding of development is not as well developed as for conversion (see the discussions in
Butterworth et al., 1989), leading to a weakened warrant via theory-based inference. While
there is some limited agreement that development is related to changes in the kinetics of
cellular mitosis, differentiation and replacement (see Chapter 2 and Cohen and Ellwein, 1990),
there is much less agreement as to the particular biological structures (DNA membranes, etc.)
requiring damage to yield the changes in kinetics. At least promotion (and to a lesser degree
progression) has been associated with the onset of hyperplasia, with removal of hormonal
control on cellular growth and differentiation, with changes in intercellular communication, and
with extensive disruption of the histological architecture of cellular communities, but the steps
leading to these changes have not been determined. As a result, the second level of evidence
for neoplastic development will be limited to cases in which it is premised that development
results from hyperplasia and lowering of intercellular communication. Hyperplasia often is
indicated by observations on mitotic rates (Swenberg and Short, 1987) and/or labelling indices
(SS), while interference with intercellular communication is indicated by changes in growth
factor receptors and/or gap junction integrity (Trosko and Chang, 1988; Hartman and Rosen,
155
-------
1983; Bertram, 1990). Changes in hormonal control may be indicated by observation of changes
in hormone concentration, hormone receptor sites (structure or density on cell surface) or
hormone structure, although these are not fully understood.
A feature shared by claims of neoplastic conversion and development is the need for
premises embodying explicit mechanistic theories of carcinogenicity (unless the empirical
correlation or existential insight warrant is invoked). The analyst must warrant use of a
particular theory in linking the first level of evidence to claims of carcinogenicity, and the second
level of evidence to claims of neoplastic conversion and/or development (see Section 5.2.3.).
Considerations arising in the testing of theories were discussed in the appendix and are not
repeated here. It is noted simply that the warrant for a particular theory can be (1) conceptual
(the theory explains existing phenomena satisfactorily as judged by a suitably qualified expert),
(2) empirical (the theory predicts the results of experiments) or (3) instrumental (the theory has
allowed actions such as mitigation leading to desired outcomes such as lowered cancer
incidence). When reflecting on these three forms of warrant for a theory, the distinction
between verification and falsification made in the appendix is important. It also is necessary,
when supporting a Theory-Based Warrant in Working Table 3, to examine the observational
evidence under each of the existing theories and to determine the resulting coherence in the
various theory-based inferences (see Section 5.5.).
52.5. Warranting by Existential Insight
At times, a warrant given by expert judgment will be available for a specific claim. In
this case, there is no explicit line of reasoning leading from an observation to a specific claim. If
there is a specific line of reasoning, this is included in one of the other relevance strategies or
warrants of Working Table 3. This raises the issue as to whether expert judgments can be given
156
-------
rational support. If rational support for a given instance of expert judgment cannot be offered,
relevance strategies relying on existential insight will have weakened epistemic status.
The following premises constitute the rational basis of existential insight:
(1) It must be premised that the individual supplying the existential warrant has
experienced the base of observations employed in the relevance strategy. Under
existentialism, it is the physical setting of the observations which provides the
necessary insight. Without existing in the physical setting, the individual expert
cannot justify the necessary claim to experience from which insight (however
unstructured it might be) arises.
(2) It must be premised that the individual possesses the necessary theoretical
understanding to interpret experience. This is not to say that such theories are
used explicitly to interpret the experience (in which case theory-based inference is
the relevance strategy). Rather the theoretical understanding aids the expert to
be "receptive" to the implications of the experience (in the language of
existentialism, to "interpenetrate" with the observed phenomenon).
(3) It must be premised that the expert judgement has been elicited properly.
Support for this premise has two components. First, the expert must be capable
of retrieving the necessary insights from his or her psyche (meaning here
whatever acts as the psychological or physiological site of the insight within the
expert). The second condition is that the process of eliciting the judgment must
not bias that judgment through specific wording or through the
social/political/economic setting within which elicitations take place.
Finally, a brief note is in order concerning the strength of a warrant for existential
insight. This strength rises as each of the three premises above receives support for a given
expert. The first two premises (A and B) will be difficult to justify through reference to the
psychological properties of the expert. The third premise (C) might receive some support by
ensuring that the elicitation process satisfies established conditions of unbiased elicitation. Still,
in most cases, the best support will be provided by the reliability of the expert in past instances
(i.e. a form of correlation between past judgments and subsequent findings of carcinogenicity
when more detailed information becomes available). As in all correlations, it must be
established that a given instance of judgment is similar to the past instances on which the claim
to reliability is established. If past instances of existential insight were based on a different
157
-------
quality of observational evidence (see Chapter 3), or on a different theoretical grasp by the
given expert (as when experts move outside their field of expertise), or by a different setting
within which elicitations were obtained, then past reliability may not be an indication of present
reliability by the expert. The key premise is that the present instance is characterized by the
same conditions as underlie the set of past instances on which reliability is judged.
53. Warranting Inter-Context Premises
As described in the introduction to Chapter 5, the second role played by an observation
lies in warranting any premises necessary to the use of relevance strategies in inter-context
claims. These premises constitute the set of background assumptions necessary for assigning
relevance to a particular observation with respect to a particular carcinogenicity claim
extrapolated across contexts. The present section reviews briefly the role of specific
observations in warranting specific background assumptions or premises. For a more detailed
listing of the links between data categories and inter-context premises, the reader should consult
Table 9 and the separate discussions in Section 5.4.
The specific premises to be warranted are:
(1) Exposure to a substance results in a BSDR both within the first (i) and second (j)
contexts for which extrapolation is being attempted.
(2) Host factors and/or concurrent exposures (i.e. antecedent conditions) do not
differ so significantly between the two contexts so as to suggest that a substance
might be carcinogenic in one context but not the other.
(3) Cancers appearing at the first level of exposure are indicative of carcinogenicity
at the second.
(4) Variability in exposure, pharmacodynamics and/or host characteristics within the
first context (population) does not differ significantly from variability within the
second context.
158
-------
TABLE 9. ROLE OF DATA CATEGORIES IN INTER-CONTEXT PREMISES
Data Category1
Role of Data Categories in Inter-Context Premises
Tumor
Response2
Can provide semi-empirical support for the premise that the relationship
between BSDR and effect has a particular form, conditional upon
establishing the premise of "Exposure to BSDR Conversion" by other
data categories. If the above relationship has been established down to
the BSDR of interest, the warrant is direct empirical. Also provides
empirical correlation and existential insight warrants for the "BSDR to
Effect Conversion". Does not provide support for the other premises.
Biophysical
Effects
The data items have no bearing on the premises "Exposure to BSDR
Conversion", "Host Factors", "Environmental Conditions" and
"Intra(Inter) Subject Variability". They provide warrants for the "BSDR
to Effect Conversion" premise by supporting the premise that a given
BSDR of interest in extrapolation will be capable of inducing the effect.
As discussed in "Broad Theoretical Implications", support for this
premise depends upon whether the data item is (i) an indicator of
cancer, (ii) an indicator of transitions between states of cancer of (iii) an
indicator of conditions (mechanisms) necessary for transitions. Since
none of these 3 indicators is an observation of cancer directly, this data
category can provide only Theory-Based Inference, Existential Insight
and/or Empirical Correlation warrants for the "BSDR to Effect
Conversion" premise.
Host
Characteristics
The data do not pertain to the premises on "Exposure to BSDR
Conversion", "Environmental Conditions" and "Intra(Inter) Subject
Variability". They may provide Direct Empirical and/or Theory-Based
Inference support for the premise that "BSDR to Effect Conversion" is
the same in two contexts by supporting the contention that (i) the
target/mechanism is (or is not) present in both contexts, (ii) that
necessary background transition rates for partial carcinogens do not
differ significantly between the two contexts, and (iii) that repair of
sublesions does not differ significantly between the two contexts.
Pharmaco-
dynamics
The primary role of these data is in warranting the premise that there is
(or is not) a significant difference in the relationship between exposure
and BSDR for the two contexts, such that a BSDR might be produced in
one context but not the other. This support will be Direct Empirical if
the BSDR is measured directly in both species, Semi-Empirical if the
BSDR is measured in both species but at higher (or lower) exposures
than desired, or Theory-Based Inference if BSDR is not measured
directly but other data items related to the production of a BSDR have
been observed. If these "other" observations are correlated with
production of a BSDR, Empirical Correlation warrants may be available.
Concurrent
Environmental
Conditions
The data provide support for the premise that there are (or are not)
environmental conditions differing between 2 contexts that would call
into question the causal role of a factor in the 2 contexts. The data also
support the contention that Intra- and/or Intersubject Variability in
exposure conditions does not differ significantly between the 2 contexts.
(continued on following page)
159
-------
Table 9. (continued)
Data Category1
Role of Data Categories in Inter-Context Premises
Structure
Activity
Relationships
The data provide warrants (Theory-Based Inference, Empirical
Correlation and/or Existential Insight) for the premise that 2 contexts do
or do not differ significantly with respect to the ability of a BSDR to
yield the effect. They provide this warrant by demonstrating that the
substance is of a form capable of acting by a common mechanism in the
2 contexts, conditional upon the premise that a BSDR is produced in
both contexts. The data also provide a warrant for the "Exposure to
BSDR Conversion" premise by demonstrating that the substance is of a
form capable of yielding an interaction (BSDR) in the two contexts,
conditional upon the premise that a biologically significant burden is
produced in both contexts.
See Working Table 2.
Quantitative measures are shown under "data items" for "tumor response" only.
Quantitative measures commonly used are not given in other data categories, only the
endpoints of interest.
160
-------
These premises, and their warranting by specific data categories/items, are discussed in the
separate sections of Section 5.4.
5.4. Linking the Observation Categories/Items to Premises Required by Inter-Context
Relevance Strategies
The observational information within a context provides the most direct, and normally
the most useful, source of evidence for warranting claims of carcinogenicity (warranting intra-
context claims as discussed in Section 5.2.). For target contexts, however, one must resort to
extrapolation of evidence from the observational contexts. Additionally, extrapolation to an
observational context, from all other observational contexts, completes the utilization of all
information available. The evidence from other observational contexts may serve to either
strengthen or weaken the warrants for claims of carcinogenicity from the prior intra-context
assessment. This step of completing the assessment for each observational context should
precede extrapolation from observational contexts to target contexts. Working Table 4 is
provided for data items of value in extrapolation across contexts. It is identical in format to
Working Table 2, which is adequate for use in place of Working Table 4 if no additional data
items specific to extrapolation need to be added.
Table 9 summarizes briefly the roles of various data categories in warranting inter-
context premises. The relevance strategies available for extrapolation premises are displayed in
Table 10 for reference. The link between specific observational evidence and inter-context
premises is depicted in Working Table 5. The cells of this working table are discussed in
separate sub-sections which follow. Assignments for the cells proceed in a manner identical to
that employed in Working Table 3.
161
-------
WORKING TABLE 4. DATA CHARACTERISTICS FOR OBSERVATIONAL CONTEXTS
FROM CONTEXT NO. TO CONTEXT NO.
Data Category/Item
Completeness
(Hi/Me/Lo/No)
Utility1
(Hi/Me/Lo/No)
Observed Effect2
(Hi/Me/Lo/No)
Causality3
(AA/CC/EC/OC)
Exposure Context- Organism-
Effect Specific Specific
Measurement Measurement
Tumor Response
TR1
TR2
•
Biophysical Effects
BE1
BE2
Pharmacodynamics
PD1
PD2
Host Characteristics
HC1
HC2
Related Substances
Assessments
RSA1
RSA2
(continued on following page)
162
-------
Working Table 4 (continued)
Data Category/Item
Completeness
(Hi/Me/Lo/No)
Utility1
(Hi/Me/Lo/No)
Observed Effect2
(Hi/Me/Lo/No)
Causality3
(AA/CC/EC/OC)
Exposure
Effect
Context-
Specific
Measurement
Organism-
Specific
Measurement
Structure Activity
Relationships
SARI
SAR2
May be subdivided into additional categories when useful, e.g. "validity", "reliability", and "accuracy" may be judged separately for data
items from laboratory sources.
Refers to effect on data item of exposure to agent. Footnotes with explanatory comments may be needed. When statistical measures
are available, they may be more informative than a simple indication of Hi/Me/Lo/No, e.g., estimates or tests of statistical
significance.
See Section 4.2.4 for explanation of the following choices available.
AA: Accidental Association
CC: Common Cause
EC: Empirical Causality
OC: Operational Causality
163
-------
TABLE 10. RELEVANCE STRATEGIES AVAILABLE FOR EXTRAPOLATION PREMISES
Data
Exposure to BSDR
BSDR to Effect
Host Factors
Environmental
Intra(Inter)-
Category
Conversion
Conversion
Conditions
Subject Variability
Tumor
N.A.
D.E.
N.A.
N.A.
N.A.
Response
S.E.E.
EC.
E.I.
Biophysical Effects
N.A.
T.B.I.
N.A.
N.A.
N.A.
EC.
E.I.
Host
NA
N.A.
D.E.
N.A.
N.A.
Characteristics
T.B.I.
E.C.
E.I.
Pharmaco-
D.E.
NA.
N.A.
N.A.
D.E.
dynamics
S.E.E.
T.B.I.
T.B.I.
E.I.
E.C.
E.I.
Concurrent
N.A.
N.A.
N.A.
D.E.
D.E.
Environment
T.B.I.
S.E.E.
Conditions
S.E.E.
T.B.I.
E.C.
E.C.
E.I.
E.I.
Structure
T.B.I.
T.B.I.
N.A.
N.A.
N.A.
Activity
E.C.
E.C.
Relationships
E.I.
E.I.
N.A.: None Available
D.E.: Direct Empirical
S.E.E.: Semi-empirical Extrapolations
E.C.: Empirical Correlations
T.B.I.: Theory-based Inference
E.I.: Existential Insight
164
-------
WORKING TABLE 5. SUPPORT FOR INTER-CONTEXT EXTRAPOLATION PREMISES
CONTEXT NO. TO CONTEXT NO.
Reference
Strategy
Exposure to BSDR
Conversion
BSDR to Effect
Conversion
Host Factors
Environmental
Conditions
Intra(Inter)-
Subject Variability
Direct Empirical
(D.E.)
Semi-Empirical
Extrapolation
(S.E.E.)
Empirical
Correlation (E.C.)
Theory-based
Inference (T.B.I.)
Existential Insight
(E.I.)
Overall
Assessment
165
-------
Working Table 6 is provided for the assessment for inter-context support for claims of
carcinogenicity. That table is identical in format to Working Table 3 for assessment of intra-context
assessment. Working Table 4 (on data characteristics for inter-context extrapolation) is utilized in the
same way for completion of Working Table 6 as Working Table 2 (on data characteristics for
observational contexts) is utilized for completion of Working Table 3. The judgments in Working Table
6, however, should be based on the support for inter-context extrapolation premises depicted in
Working Table 5, as well as the data characteristics depicted in Working Table 4. The intra-context
assessment (applicable to observational contexts but not target contexts) and the inter-context
assessment (applicable to both observational and target contexts) are depicted and then summarized in
Working Table 7 for overall assessment from all sources.
5.4.1. Conversion from Exposure Conditions to BSDR
As stated in the broad theory of environmental carcinogenesis, it is assumed that a substance in
the environment must produce a dose-rate of the biologically significant form of the substance within a
target organ, tissue, cellular subpopulation, etc. It must be demonstrated, therefore, that there are no
important differences between the study population constituting the observational data and the
population for which claims to carcinogenicity will be made, at least with respect to the relationship
between exposure and BSDR. By "important" is meant differences which would strengthen (or weaken)
a claim that cancers in the study population might be due to pharmacodynamic factors present in that
population but not in the population for which claims of carcinogenicity will be made (presumably, the
target context). The inverse of this issue may also apply, in which case reflection on pharmacodynamic
factors might suggest that an observation of "no cancer" in the study population might be due to factors
present (or absent) in that population but not in the population for which claims to carcinogenicity will
166
-------
WORKING TABLE 6. INTER-CONTEXT SUPPORT FOR CLAIMS OF CARCINOGENICITY1
FROM CONTEXT NO. TO CONTEXT NO.
Claims of Carcinogenicity2
Reference
Strategy
I.O.3
Increases
Incidence of
Cancer
Classification(s)
Stage(s)
Mechanism(s)
Complete
Partial
Mixer
Helper
Neo.
Conv.
Neo.
Devel.
Geno-
toxic
Non-
genotoxic
Direct Empirical
(D.E.)
Semi-Empirical
Extrapolation
(S.E.E.)
Empirical
Correlation
(EC.)
Theory-based
Inference (T.B.I.)
Existential Insight
(EI)
Column Summary
Overall Summary
Top half of each entry is completed using dose-response data observed in the context. Bottom half of each entry includes "floater-
data" as well (see text).
Choices for cell entries are Hi/Me/Lo/No. See text of Section 5.1. This assignment is made independently of the assignment of
intellectual obligation.
"Intellectual Obligation" (Hi/Me/Lo/No). See text of Section 5.1. This assignment is made independently of the assignment in the
cells of the table (footnote 2).
167
-------
WORKING TABLE 7. SUMMARY OF OVERALL ASSESSMENTS
FOR CLAIMS OF CARCINOGENICITY BY CONTEXT
CONTEXT NO.
Claims of Carcinogenicity1
Context Number
Increases
Incidence of
Cancer
Classification(s)
Sta
ge(s)
Mechanism(s)
Complete
Partial
Mixer
Helper
Neo.
Conv.
Neo.
Devel.
Genotoxic
Non-
genotoxic
Intra-Context2
No. _
Inter-Context3
No. _
Inter-Context3
No. _
Inter-Context3
No. _
Inter-Context3
No. _
Column
Summary
Overall
Summary
Choices for cell entries are Hi/Me/Lo/No.
Context number should match context number in table heading.
Number of the context from which results were extrapolated.
168
-------
be made. In that case, observation of no carcinogenicity in the study population does not warrant a
claim that the substance is a non-carcinogen in the target context. These two possibilities of judgment
are not distinguished further in this report.
For the case of radon, differences in the relationship between exposure and BSDR within the
mining population (i) and home population (j) were explored at length in the second report on radon
(NRC, 1991). These differences resulted from population-specific differences in degree of attachment
of radon progeny to aerosols, aerosol size distribution, equilibrium fraction, breathing characteristics,
lung sizes, and mucus flow rates in the lung. The resulting analysis warranted the claim that the BSDR
generally would be lower in the present following exposure to airborne radon. This conclusion was
warranted regardless of the assumed location of target cells. The primary uncertainty, then, was
whether this lower BSDR was capable of inducing cancer (see Section 5.4.3.).
Radon is an extreme instance in the sense that pharmacodynamic data are plentiful. It must be
noted that complete understanding of the pharmacodynamics for a substance often is lacking. This may
arise if (1) pharmacodynamic properties are not understood (2) the target for action is unknown or (3)
the biologically significant form of the substance is unknown. Reference to Figure 1 is useful here in
determining how pharmacodynamic data may still be useful even when one of the above 3 cases of
weakened understanding applies. From this figure, it may be seen that the chain of reasoning in
pharmacodynamics proceeds from exposure to intake to uptake to organ burden to biologically
significant organ burden to biologically significant dose-rate (BSDR).
This introduces the following premises into the analysis in supporting the judgments on the
relationship between exposure and BSDR (the premises are given first in sketch, but are discussed in
more detail at the end of this section):
Premise 1: There are no differences between the study population (hereafter, Ps) and the
population of interest in making claims to carcinogenicity (hereafter, Pj), with
respect to the conversion from exposure to intake, which would detract from the
claim that cancers in Ps imply cancers in Pt at equal levels of exposure. The
-------
factors influencing this conversion are (a) inhalability of the physical form of the
substance in the environment, as when filters are present or particle sizes in air
are incapable of entering the nose and/or mouth; (b) breathing rates, ingestion
rates or degree of skin contact (for cases of inhalation, ingestion and dermal
exposures, respectively); (c) location or presence of Ps and Pj within the field of
exposure. Differences in these factors (discussed at length in NRC, 1991) may
arise either from inherent biological differences between Ps and Pr (as in
interspecies extrapolation) or from differences in exposure to other substances
that modify the relationship between exposure and intake. The same two sources
of difference apply to the other 4 premises that follow. The analyst reflects on
these issues and judges whether intake in Ps will be higher or lower than the
intake in Pr.
Premise 2: There are no differences between Ps and PI; with respect to the conversion from
intake to uptake, which would detract from the claim that cancers in Ps imply
cancers in Pr at equal levels of intake. The factors influencing this conversion are
(a) deposition fraction in the lung (for inhalation), (b) absorption from the lung
to the body fluids (blood, water, etc), (c) absorption from the body fluids to the
target organ, (d) first-pass excretion, (e) dermal absorption (for dermal
exposures) and (f) absorption from the G.I. tract to the body fluids (for
ingestion). If the target organ is the G.I. tract, then intake and uptake are
identical for the ingestion route of exposure. The same applies to the lung for
inhalation exposures. The analyst reflects on these issues and judges whether
organ uptake in Ps will be higher or lower than the uptake in P( (see NRC, 1991).
Premise 3: There are no differences between Ps and Pr, with respect to the conversion from
uptake to organ burden, which would detract from the claim that cancers in Ps
imply cancers in P! at equal levels of uptake. The factor influencing this
conversion is the retention of the substance in the organ. The analyst reflects on
this issue and judges whether organ burden in Ps will be higher or lower than the
organ burden in P! (see NRC, 1991).
Premise 4: There are no differences between Ps and P[; with respect to the conversion from
organ burden to biologically significant organ burden, which would detract from
the claim that cancers in Ps imply cancers in Pj at equal levels of organ burden.
The factor influencing this conversion is the fraction of the original substance
transformed into the biologically active form. The analyst reflects on this issue
and judges whether biologically significant organ burden in Ps will be higher or
lower than the biologically significant organ burden in P( (see NRC, 1991).
Premise 5: There are no differences between Ps and P[( with respect to the conversion from
biologically significant organ burden to BSDR within the target organ. Factors
influencing this are (a) the presence of target sites within the organ, (b) spatial
location of the substance molecules with respect to the target sites and (c) ability
of the substance molecules to interact with the target sites. It should be noted
that this fifth premise rarely can be supported due to a lack of information on
the target site (an exception being radon (NRC, 1991)). Still, the analyst reflects
170
-------
on this issue and iudees whether BSDR in Pc will be higher or lower than the
BSDR in P,.
Reflecting on each of the five premises, the analyst assigns an index of evidential support to
each premise (scale of low to high). If all five premises are well supported by evidence, with the
evidence being given by the factors discussed under each premise above, the judgment under the
"Exposure to BSDR Conversion" premise of Working Table 5 is deemed strongly supported and is
assigned an overall index of "high". As support for any of the premises is weakened, the index on the
judgment falls towards "low".
Support for a given premise may be in the form of any of the 5 relevance strategies of Section
5.1. For example, consider Premise 3 (concerning the relationship of uptake to organ burden). There
may be observational data available on the ratio of organ burden to organ uptake in both P, and Ps. In
that case, there is a direct empirical warrant for claims about this ratio, conditional on the observations
having been obtained at the uptakes of interest in the analysis. If the uptakes are larger than those of
interest in the analysis (such as when retention functions are obtained from acute and large uptakes),
the observed ratio must be extrapolated to lower levels of uptake. In that case, the warrant for premise
3 is semi-empirical. If information on the ratio is not available, but it is deemed that the ratio in Pj
generally correlates well with the ratio in Ps (perhaps in examinations of other substances for which
observational evidence is available), then a warrant of empirical correlation is made. If the analyst
simply judges that the ratio should be similar in P, and Ps, without giving explicit observational
warrants, the result is a warrant of existential insight. Finally, if direct observations of the ratio are
unavailable, but observations on the retention function are available, the analyst may judge the ratio to
be the same (or different) in Ps and P, based on the similarity (or difference) in retention. This
warrant is theory-based inference since it does not involve a direct measurement of organ burden in
either P, or Ps and requires a further premise that the identified important factors (here, retention) are
the only ones significantly affected the ratio (organ burden to uptake) in Pt and Ps.
171
-------
The preceding paragraph focused on the five premises necessary to judge the strength of the
warrant for the "Exposure to BSDR Conversion" Premise in Working Table 5. Having reflected on
these five premises, the analyst determines which of the five relevance strategies may be applied to the
"Exposure to BSDR Conversion" premises and assigns an indication of weight-of-evidence (Hi to Lo).
The forms of warrant for any of the inter-context premises appear explicitly in Working Table 5. It will
be noted, however, that the premises necessary to support a given inter-context premise (such as
premises 1 through 5 above) are not depicted explicitly in this working table. As a result, the form of
the warrant for premises 1 through 5 above is used in the judgment of the strength of the inter-context
premise (Working Table 5), but does not appear explicitly in Working Table 5.
In addition, it may be noted that it is not necessary to warrant each of the five premises from
this section separately in order to warrant the inter-context premise (concerning the relationship
between exposure and BSDR in the two contexts). Any of the five premises may be combined if direct
observations on the relationship between any step in the analysis of the judgment (exposure-intake,
intake-uptake, etc.) and any "earlier" step is available. For example, measurements of the relationship
between exposure and organ burden may preclude the need for separately warranting premises 1
through 3. The observational data warrant the three premises (1 through 3) simultaneously by showing
that the ratio of organ burden to exposure is similar (or different) in Ps and Pj. The strongest
conceptual strength of the warrant for the three premises remains, of course, one in which all three
premises are warranted separately, giving support to the claim that (1) the ratio is similar (or different)
in Ps and Pj and (2) the similarity (or difference) is understood in terms of the analytic steps leading to
an estimate of organ burden. Still, even if the reason for the similarity (or dissimilarity) is
unexplainable, there may exist direct measurements warranting premises 1 through 3 empirically.
Finally, a comment is in order concerning the role played by the above premises in the judgment
that the inter-context premise is warranted. Unless a threshold BSDR necessary for carcinogenicity or
172
-------
a non-monotonic relationship between BSDR and effect is premised, any value of BSDR observed to
produce cancer in Ps provides a warrant for the claim that the substance is a carcinogen in Pr
(conditional upon the "applicability" of the antecedent conditions in Pj). This applies regardless of the
numerical value of the BSDR in Pj. The judgment that consideration of factors concerning conversion
from exposure to BSDR does not alter the claim of carcinogenicity in the analysis, therefore, hinges on
the ability to demonstrate only that population P, will receive at least a non-zero BSDR. This will be
true so long as it can be shown that:
(1) Intake occurs in Pj, only requiring premises that (1) the population is exposed to the
substance in the environment, (2) does not have a filter of perfect efficiency placed over
the lung, G.I. tract or skin (depending upon the route of exposure), and (3) the lung, G.I.
tract or skin does not block movement of the substance into the body completely (as
when the rat nasal passages close down during exposure to high concentrations of
formaldehyde (Graham et al., 1988)).
(2) The uptake fraction is not zero in the target organ, only requiring premises that (1)
there is not complete first-pass excretion of the substance, (2) the substance is not
completely exhaled, regurgitated, or washed from the skin (depending upon the exposure
route) and (3) uptake into the organ does not require saturation of some mechanism of
absorption.
(3) The substance is not removed immediately from the target organ upon entry, requiring
only the premise that the retention function is non-zero.
(4) The substance is metabolized to some degree into the active form, requiring premises
that (1) this metabolic process be present in Pj and (2) that the process is not
characterized by saturation. If metabolism is from the inactive to the active form),
satisfying premise (2) requires that the activation process be operative at the organ
burden, expected in P, and not be set in motion only by higher burdens. If metabolism is
from the active to inactive form, satisfying premise (2) requires that the inactivation
process be less than 100% efficient at the organ burden expected in Pj.
(5) A target (organ, tissue, etc.) is present in both Ps and P,. This will be warranted most
strongly when the target has been identified, but in many cases the target is not known
and must be assumed to be present in both populations based on anatomical similarities.
173
-------
5.42. Host Factor and/or Concurrent Conditions Do Not Differ Significantly
From the judgment discussed in 5.4.1., the analyst determines that both Ps and Pt are receiving
some non-zero value of BSDR (with an associated index of evidential support). This supports the
judgment that targets of individuals in Pt are being dosed, to some degree, by the active form of the
substance in the target organ (tissue, cell, etc.). It might then be assumed that claims of an effect in Ps
(presumably due to the presence of a BSDR at the target site) is warrant for a claim that an effect will
occur in Pj (where a BSDR also has been judged present). This assumption, however, is based on the
premise that the causal association between BSDR and effect in Ps is not due to conditions other than
the BSDR that might be present in Ps but not Pt. These conditions are antecedent to exposure to the
substance under analysis in the sense that they establish the biological and exposure conditions within
which the substance under analysis exerts an effect. For example, in the case of ETS, it was argued that
the Chinese studies were not applicable to the U.S. population due to the presence of large background
levels of cooking and heating smoke in homes. A similar argument was made for radon, in which it is
was claimed that the large dust burden in mines allowed expression of the carcinogenic properties of
radon (see the discussion in NRC, 1988). It was claimed further that this antecedent condition related
to concurrent exposures (presence of large dust burden) was present in mines but not in homes, calling
into question the premise that cancer in the mining population (Ps) was due to the inherent action of
radon in the lung rather than to the antecedent condition (dust).
The discussion in this section is directed towards warranting a judgment that an effect produced
by the substance of interest in Ps did not require the simultaneous presence of antecedent conditions
(either concurrent exposures or host factors) in Ps unlikely to be present in Pj. This judgment clearly is
related to the judgment that a substance is (or is not) a "mixer" or "helper". The inverse of this
judgment is that a lack of effect produced by the substance of interest in Ps was not the result of lacking
the necessary antecedent conditions in Ps, with those antecedent conditions being present in Pj. The
174
-------
judgment under discussion here presupposes that the substance played a causal role in the effects noted
in Ps, and focuses on the judgment that this causal role was (or was not) conditional upon antecedent
conditions present in one population (Ps or P,) but not the other.
The role of antecedent conditions appears in the theory of carcinogenicity employed in an
analysis. It must be premised that there is nothing present in the exposure conditions and/or in bodies
of individuals in Ps, but missing in P, (or vice-versa), that would significantly alter the relationship
between BSDR and carcinogenicity. These antecedent conditions might relate to the following factors:
(1) Initial state vector (Crawford-Brown and Hofmann, 1990). Prior to exposure to the
substance of interest, Ps and P, might differ in the degree to which the target already has
been initiated and/or promoted. This could arise from differences in previous exposures
or from host factors. For example, exposures to high dust levels in mines have been
suggested to establish promotional action in lungs of miners (Ps) that would not be
present in homes (P,). If it is assumed that (1) radon does not induce both initiation
and promotion in P, and (2) that there are no sources of promotion in P! other than
exposure to dust, then carcinogenicity from radon in Ps might be claimed to be due
entirely to the antecedent conditions of promotion brought on by high dust exposures.
This weakens the claim that radon is a carcinogen in the home environment where dust
levels may be in sufficient to induce promotional action.
The initial state vector also is important when it is premised that a target (organ, tissue,
etc.) must possess a minimal number of cells in a given state (initiation, promotion
and/or progression) to yield a fatal tumor. If the number of cells in that state already is
sufficient, prior to exposure to a given substance, that substance will have no effect on
the number of cells in the considered state. If the prior number of cells is below, but
close to the threshold number, exposure to the substance might increase the incidence of
cancer by inducing transitions and raising the number of cells (in the given state) above
the threshold. If the prior number of cells in the considered state is well below the
threshold, the substance may yield an insufficient number of transitions to exceed the
required threshold. All of the above cases, of course, are significant if, and only if, a
threshold number of cells in a given state is required.
(2) Repair rates and/or efficiency. Particularly in the case of initiation (Hall and Freyer,
1991), it has been established that initial damage produced by a carcinogen can at times
be repaired. Two populations (Ps and Pj) may differ in the degree of repair for this
initial damage. For example, individuals with xeroderma pigmentosum lack the
necessary enzymes for operation of the DNA repair system, resulting in greatly increased
sensitivity to UV induced skin cancer.
Reflection on repair rates requires consideration of several other factors. If repair is
100% efficient in Ps but not in PI( then a finding of no carcinogenicity in Ps does not
fully warrant a claim of no carcinogenicity in Pt, conditional upon the premise that the
175
-------
substance of interest plays a role in carcinogenicity mediated by a process subject to
repair. If repair is 100% efficient in Pj but not in Ps, then a finding of carcinogenicity in
Ps does not fully warrant a claim of carcinogenicity in P1; again conditional upon the
premise that the substance of interest plays a role in carcinogenicity mediated by a
process subject to repair. If repair is less than 100% efficient in both populations, it
must be established that repair does not lower the number of unrepaired damage sites
below some threshold value. Observations relevant to the issue of repair are
measurements of repair rates.
(3) Background transition rates. These rates refer to the rates of transition between normal,
converted and developed cells without the presence of the substance of interest. This
consideration might be important due either to biological differences in Ps and Pr (i.e.
host factors) or differences in concurrent exposures. The significance of this factor
arises when it is possible to saturate processes leading to a given transition. For
example, consider the case of a substance that acts through stimulation of mitosis,
yielding a transition from initiated to promoted states. If the mitotic rate already is
saturated in a given population (Ps or Pj), exposure to the substance of interest might
prove incapable of increasing the incidence of tumors. In another population with an
unsaturated mitotic rate, exposure to the substance might increase the mitotic rate and,
hence, the incidence of cancer. It should be noted here that the issue being raised is not
one of the numerical value of the background transition rate but whether the transition
rate can be further modified by action of the substance of interest.
Another case where consideration of background transition rates are important is when
an etiologic process must exceed a threshold to yield the transition. For example, it was
suggested in the case of formaldehyde that promotion brought on by induction of
hyperplasia requires a threshold level of hyperplasia (Swenberg et al., 1983, 1987). A
given population (Ps or Pr) when exposed to a hyperplastic agent might possess a
sufficiently low background degree of hyperplasia that the added hyperplastic action of
the substance of interest does not increase hyperplasia above the threshold required for
a transition to the promoted state. A second population might possess a background
degree of hyperplasia close to the threshold, so that the added hyperplastic action of the
substance of interest does increase hyperplasia above the required threshold.
(4) Presence and/or absence of target in host.
As with all of the considerations in this section, the differences in background rates of transition
may be due to differences in the concurrent exposures or inherent biological properties of a population.
One further premise must be introduced in employing measured or estimated background rates
of transition. In its simplest form, the multistage theory of carcinogenesis employs stages, and
transitions between those stages, which are common to all carcinogens. For example, it might be
assumed that all instances of neoplastic conversion share common events leading to neoplastic
176
-------
development. This theoretical approach to carcinogenesis presupposes a general structure to
carcinogenesis that is invariant between specific carcinogens. To the degree that carcinogens act in
different ways to produce different forms of conversion and development, the premise of commonality
will be invalid. Measurements of background rates of transition may require selection of only those
forms of transition germane to the specific carcinogen under study. It must be premised, therefore, that
the background transition rates selected in an analysis represent the rates applicable to the specific
forms of conversion and development required by the specific carcinogen within the contexts under
consideration for extrapolation.
5.43. Extrapolation Across Doses and/or Dose-Rates
One of the most contentious areas of hazard identification concerns the judgment that
observation of tumor response carcinogenicity at high values of the dose (biologically significant dose)
and/or dose-rate (BSDR) provides a warrant for claims to carcinogenicity at values of interest in the
regulatory setting. If the task of hazard identification is to warrant the claim simply that a substance is
a carcinogen at some level of dose (BSDR, etc.), then observations of carcinogenicity under high levels
of dose (BSDR, etc.) serve as an appropriate warrant for that claim. If, however, the task is to
determine whether the substance is a carcinogen within a context of interest to the regulator (with these
conditions typically being characterized by relatively low values of dose, BSDR, etc.), then premises
concerning the effect of lowering doses (BSDR, etc.) below those contained in the "high exposure"
context must be introduced into the analysis and warranted. It is towards elucidation of these latter
premises and their warrants that the present section is directed. There may, of course, be observational
evidence available for the target context, but use of the evidence constituted the intra-context warrants
discussed previously.
177
-------
The strongest warrant for the premise that causal relationships between BSDR and tumor
prevalence are constant (but not numerical identical) is a direct empirical observation that the BSDR is
Ps and PT produces a response in both populations. This requires tumor response data at doses
(BSDRs) encompassing both Ps and PT contexts. Such data might arise, for example, from injection
(uptake) studies in which the relationship between exposure and uptake is not at issue. It must, of
course, be premised that route of administration (injection rather than environmental exposures) does
not alter the relationship between BSDR and response in a manner which differs between large and
small uptakes.
If the available tumor response data do not encompass dose-rates encountered in Pj, semi-
empirical warrants supplant the direct-empirical warrants for the premise under consideration in this
section. In this case, the claim of carcinogenicity at high dose or dose-rate is extrapolated to low dose
or dose-rate based on direct observation of a pattern in the relationship between BSDR and response in
Pj. There are two routes to semi-empirical warrants. The first is a warrant of observed patterns in a
plot of dose versus tumor response (dose-rate being held constant) and/or a plot of dose-rate versus
tumor response (dose being held constant) for the population (context) Ps. The analyst judges that the
pattern has been directly observed in the data on Ps. This pattern then is followed visually to the dose
and/or dose-rate of interest in P,, supporting (or weakening) the contention that the substance is a
carcinogen in Ps at low dose and/or dose-rate. The key premise here is that the pattern is
observationally evident in the data rather than being imposed on the data by a fitting equation. This
requires, of course, tumor response data of sufficiently quality to bring any underlying patterns into
view for the analyst, both for the case of dose and dose-rate. Such data were available for the case of
radon exposures, since the mining populations could be divided into groupings characterized by
different doses and dose-rates (NRC, 1988). It should be noted, however, that patterns in the data
were not clear, requiring the addition of a second form of warrant discussed in the following paragraph.
178
-------
The second warrant for semi-empirical extrapolation is theory-based semi-empirical
extrapolation. In this case, the same observational data discussed above (on dose and dose-rate versus
tumor response) are employed, but the requirement of direct observation of patterns in the data is
dropped. Instead, a fitting equation deduced from etiologic theory is employed in the extrapolation.
The NRC committee chose to employ a linear dose-response equation based on belief in a one-hit
model of radiation carcinogenicity (NRC, 1988). The extrapolation equation is not used in hazard
identification to estimate the actual risk at low exposures, but simply to determine whether
carcinogenicity is to be expected at lower doses and/or dose-rates. In addition, the premise that the
extrapolation equation itself is valid must be warranted by warranting the premises appearing in the
etiologic theory from which the equation is deduced (again, see the discussion in Section 5.2.). The
strongest case of semi-empirical extrapolation will hold when both (1) the pattern is evident in the data
and (2) the pattern is to be expected from the prior establishment of an etiologic theory and its
associated extrapolation equation.
Both the empirical correlation and existential insight warrants for extrapolation were discussed
previously (see Sections 5.2.3. and 5.2.5., respectively). That discussion will not be repeated here, other
than to comment that the correlation is between a finding of carcinogenicity at high doses and/or dose-
rates and carcinogenicity at low doses and/or dose-rates. This correlation might be expected to
improve as the dose and dose-rate in Ps approach the values in P,. The analyst may choose, therefore,
to construct the measure of correlation on sets of substances for which the magnitude of difference in
dose and/or dose-rate is similar to that of interest in the analysis.
Finally, the warrant for extrapolation may be in the form of theory-based inference. Only data
items from data categories other than "tumor response" may be employed here, since the latter were
employed in the direct empirical and semi-empirical warrants. In this case, the most explicit use of
etiologic theories is made in developing the necessary premises. Again, both dose and dose-rate must
179
-------
be considered. The remaining discussion in this section focuses on consideration of differences in dose
and dose-rate between Ps and Pj.
With regards to both dose and dose-rate, four primary premises must be warranted. These are:
(1) There is not a dose and/or dose-rate below which the transitions produced by the
substance of interest will not occur (or at least the dose and/or dose-rate in both Ps and
P, exceed the threshold). For example, it has been proposed that DNA repair
mechanisms can deal effectively with initiating damage so long as the rate of damage is
not sufficient to induce SOS repair. It also has been proposed that hyperplasia-induced
promotion by formaldehyde requires a threshold dose-rate. It also has been proposed
that a minimal level of organizational disruption in cellular communities is required to
induce cancer. This disruption presumably is a function of both total dose and dose-rate.
(2) There are not competing processes which change their "order" of impact on
carcinogenicity below some threshold level (or at least the doses and/or dose-rates in Ps
and P, are either both above or both below this threshold). For example, many
substances both induce transitions to states of cancer and are cytotoxic (for an
application to the case of radon, see Crawford-Brown and Hofmann, 1990). Substances
which induce transitions will tend to be carcinogenic. Substances which are cytotoxic
may kill cancerous cells, thereby lowering the incidence of cancer in a population. It
might be the case that a substance induces newly cancerous cells at a rate faster than it
kills previously cancerous cells, leading to a net observation of carcinogenicity at a given
dose and/or dose-rate in Ps. At lower (or higher) doses and/or dose-rates, however,
cytotoxicity to previously cancerous cells may exceed induced transitions, leading to a net
observation on non-carcinogenicity (or even "life-saving") at these lower (or higher) doses
and/or dose-rates.
(3) The latency period for cancer does not increase significantly at low doses and/or dose-
rates. For example Raabe et al. (discussed in NRC, 1988) have proposed that the
latency for bone sarcoma might exceed the expected lifetime at low doses and/or dose
rates. If this is the case, the latency for cancer in Ps might be less than the expected
lifetime (leading to an observation of carcinogenicity), but the latency for cancer at lower
doses and/or dose-rates in P[ might exceed the expected lifetime (leading to an
observation of non-carcinogenicity).
(4) There is not a mechanism for carcinogenesis present (or absent) at the level of dose
and/or dose-rate in Ps but not present (or absent) at the level of dose and/or dose-rate
in P,. If this mechanism simply requires exceedance of a threshold for one of the
transitions to cancer, then premise 4 and premise 1 are identical.
180
-------
5.4.4. Consideration of Intrasubject and Intersubject Variability
Epidemiologic and experimental studies of the effect of an environmental substance typically
employ estimates of the mean exposure, dose, etc. in defined groups (such as exposure groups in a
cohort study). The result is an estimate of the relationship between mean level of exposure (or dose,
etc.) and the response in a population.
Such approaches presuppose that response is predicted entirely by mean exposure rather than
by other properties of the distribution of exposure within a group characterized by inhomogeneity of (1)
exposure and (2) sensitivity to the action of a substance. In some cases, however, knowledge of the
mean exposure in a group is not sufficient to determine the response of the group (for a discussion
specific to radon, see Crawford-Brown and Hofmann, 1989) either qualitatively or quantitatively. In
such cases, other properties of the distribution such as variance may be of equal importance in
predicting response. This section summarizes the important consideration of variability within an
exposed population.
The variability has two primary components: intrasubject and intersubject variability.
Intrasubject variability refers to variations in exposure (intake, uptake, burden, dose, BSDR, etc.) both
spatially and temporally with respect to an individual organism such as a human or experimental
animal. This variation arises from the following factors:
(1) Exposure conditions may vary for the individual, as when radon concentrations in home
air fluctuate during the day. If it is premised that BSDR must exceed a threshold value
to produce cancer, and only if this is premised, the analyst must determine whether
temporal variation in exposure conditions affects the judgment of carcinogenicity. For
example, the study population (Ps) may have been exposed to a well controlled
environment where the mean exposure (and, hence, BSDR) was below the required
threshold for cancer. The population of interest (Pj) may be exposed to the same mean
exposure but with greater temporal variation. As a result, cancer may not be present in
Ps but might be expected in P, if the variation in the latter exposure produced intervals
of time during which the BSDR exceeded the threshold. Conversely, excessive variation
of exposures in Ps might produce cancer while none is expected in Pt due to lesser
variation. The key issue is whether the extent of variation differs between Ps and P, to
such a degree that thresholds of BSDR are exceeded in one population but not the
other.
181
-------
(2) The BSDR may vary spatially within the target organ of an individual due to
inhomogeneity of uptake or retention (see the information in Crawford-Brown and
Hofmann, 1989). Again, this spatial variation in BSDR may cause some fraction of the
organ to receive a BSDR above the threshold required for cancer (requiring, of course, a
premise that a threshold exists). If the inhomogeneity of BSDR within the target organ
differs dramatically between individuals in Ps and those in Pt, cancer might be caused in
one but not the other.
The second source of variability is intersubject variation. As in the case of intrasubject
variation, intersubject variability is significant to the task of hazard identification only if thresholds for
cancer are premised. The important sources of intersubject variability are as follows:
(3) Exposure conditions may vary between individuals in a group. This variability is
complicated by variability in pharmacodynamic properties for those individuals. The
result of these two factors is variability in BSDR between individuals in the group.
Again, premising a threshold BSDR necessary for cancer, the determinant of response in
the group is not mean exposure or mean BSDR but the fraction of individuals with a
BSDR above the threshold. A finding of cancer (or no cancer) in Ps does not, therefore,
warrant a claim of carcinogenicity in P, if the variability within Ps and P, differs to such a
degree that the fraction of individuals exceeding the threshold is zero in one population
but not the other.
(4) Individuals may vary with respect to sensitivity (Redmond, 1981). This variation is
significant for hazard identification if (and only if) sensitivity is characterized by a
threshold BSDR necessary for cancer. In that case, it must be premised that variability
of sensitivity in Ps and P, do not differ to such a degree that there are (1) individuals in
one population with a threshold below the delivered BSDR but (2) not in the other
population. This consideration gains significance when Ps is constituted by genetically
similar experimental animals (presumably with similar thresholds) and Pt is constituted
by a diverse human population (presumably with variation in thresholds, if thresholds
exist). It has been argued, for example, that a finding of a threshold for cancer in a
genetically homogeneous population does not warrant a claim of non-carcinogenicity in
human populations characterized by a wide range of sensitivities.
5.5. Column/Row Summaries and the Issue of Coherence
The various working tables display a number of points at which the issue of coherence must be
raised in order to provide the entries for specific cells. By coherence here, we mean the degree to
which a set of observations, theories, inferences, etc., present a unified and supporting pattern of
182
-------
warrant for a specific claim. Coherence is lost whenever there are important differences between
claims of carcinogenicity drawn on the basis of different:
(1) Cases within a data item.
(2) Data items within a data category.
(3) Theories within "Theory-Based Inference".
(4) Relevance strategies within a context.
(5) Contexts.
These issues, or bases of difference, are discussed separately in the present section.
First, a distinction between two aspects of coherence must be drawn. The analyst first should
determine whether the AVAILABLE evidential base displays coherence. This is referred to here as
extant coherence since it is a judgment that the existing evidential base leads to a consistent (non-
contradictory) judgment of carcinogenicity. Of equal importance, however, is a second form of
coherence referred to here as complete coherence. Complete coherence arises when all data items
potentially of use in analysis (in any manner of use) are (1) available and (2) present a consistent,
mutually supportive, pattern of evidence. For example, a number of epidemiologic studies might be
available (as in the direct empirical studies of radon in the home), leading to a claim of extant
coherence if their results are similar (this was not the case for the direct empirical studies of radon but
was the case for the semi-empirical mining data). Still, there may be greater support available if
pharmacodynamic data displayed the existence of a BSDR and if biophysical effects were observed (as
in the case of radon exposures). The existence of these latter data, if they yield Similar inferences of
carcinogenicity (as they do with radon), strengthen the coherence by increasing the claim to complete
coherence.
Examining Working Tables 2 through 7, then, it may be noted where issues of coherence arise
and must be factored into assignments for cells appearing in those tables. For Working Table 2, the
183
-------
coherence of the cases under a given data item is an essential component of the judgment of "observed
effect" and of "causality" (see the previous discussion in Section 4.2.). For Working Table 3, coherence
arises in several instances. Within a relevance strategy, different data items may be pertinent to a given
strategy due to the existence of several premises required for invocation of that strategy (see Section
5.2.). If the premises are supported consistently by the available data, extant coherence may be high.
Complete coherence will be high only if all data potentially of use in supporting the premises are
available and yield a consistent judgment concerning the premises. Coherence also enters in the use of
etiologic theories within the "Theory-Based Inference" cells. Here, the analyst must determine if a
similar claim of carcinogenicity is implied by each of the existing theories. Coherence in this instance
rises from "Lo" to "Med" to "Hi" as the various theories move the judgment form one of "inconsistent
inference between examined theories", to "consistent inference between examined theories" (extant
coherence) and, finally, to "complete coherence between a large body of theories" (complete coherence).
In this regard, it is of interest to note that etiologic theories of radon carcinogenicity do not uniformly
lead to claims that semi-empirical extrapolation suggests carcinogenicity at lower exposures in the home.
Within a claim in Working Table 3, the analyst then examines the coherence between inferences
drawn on the basis of different relevance strategies. If all of the available strategies yield a consistent
inference, an instance of extant coherence applies. If, in addition, all relevance strategies were available
(as is true for radon), complete coherence applies. When inconsistency between strategies exists, the
analyst must examine the claim to "intellectual obligation" to determine the impact of this incoherence
on the epistemic status of a claim. Incoherence then weakens the claim only if the incoherent
inferences arise from relevance strategies with high values of intellectual obligation. In any event, the
analyst reviews the coherence of the five relevance strategy-specific judgments in the cells of a column
and enters a composite judgment in the "column summary" cell for each claim in Working Table 3.
184
-------
Coherence then must be examined across the claims in the "Column Summary" row. The intent
here is to determine whether the separate claims present a consistent pattern. For example, a claim
that a substance is not genotoxic might be judged inconsistent with a claim that the substance induces
neoplastic conversion (if it is premised that conversion arises from genotoxicity). This consideration
might cause the analyst to adjust the claim to neoplastic conversion. This second level of summary,
taking into consideration coherence across the columns of the "Column Summary", is provided in the
"Overall Summary" of Working Table 3. The summary row in Working Tables 6 and 7 have the same
function as in Working Table 3. In Working Table 5 the single summary row labeled "Overall
Assessment" is like "Column Summary" in Working Table 3. There is no need for an additonal
summary row in Working Table 5 because coherence of column summaries is not an issue.
Working Table 7 examines the coherence of Target-Context claims (the product of one of the
forms of Working Table 3 as generated for the specific target context) and the inter-context
extrapolation claims from any observational context to the target context (the product of one of the
forms of Working Table 6). The analyst simply transcribes the results of the "Overall Summary" rows in
Working Tables 3 and 6. The analysis of coherence for Working Table 7 then proceeds as in the
discussion of Working Table 3. Coherence across contexts first is judged within a column (the "Column
Summary" of Working Table 7), followed by a judgment of coherence across the claims in the "Column
Summary" row. The result is the "Overall Summary" row of Working Table 7.
6. THE SEVEN STEPS OF HAZARD IDENTIFICATION: AN OVERVIEW
The seven steps indicated for hazard identification of a substance are shown schematically in
Figures 4 and 5. The elements within each step are described verbally in Figure 4. In Figure 5 the
steps are described in terms of the working tables to be completed and the inter-relationship between
steps is depicted. The following brief discussion of the seven steps supplements the descriptions given
185
-------
in the two figures. References are made to preceding sections of the report for discussion of topics
relevant to each step.
Steps 1 and 2 refer to the initial collection of all informational sources that may be potential
useful in assessing the support for claims of carcinogenicity of the substance of interest and definition of
contexts (Section 4.1). "Observational contexts" are defined around studies of dose-response data in
animals or humans while "target contexts" are defined for contexts of interest for hazard identification
in which no study data are available. Studies judged to be sufficiently similar to justify pooling their
results statistically are treated as "cases" belonging to the same observational context.
Step 3 is repeated for each observational context. Working Tables 2 and 3 (WT2 and WT3) are
completed for each context. The data items in Table 2 for which data are available are inserted in WT2
(Step 3.2) and assessments of the data characteristics (completeness, utility, observed effect, and
causality) are made (Step 3.3) (Sec. 4.2 and subsections). For contexts with more than one case, a
similar step must first be conducted for each case within the context to form a composite representation
of the context (Step 3.1). The completed WT2 for an observational context is utilized in completion of
WT3 for the same context, in which a judgment of support for each claim of carcinogenicity (c. of c.) is
made for each relevance strategy available (Step 3.4). The upper part of an entry in WT3 is for a
judgment based on data used only in that context (e.g., tumor response data items); the lower part of
the entry is for a judgment based on all data relevant to a judgment of carcinogenicity within the
context, including data that may be utilized in other contexts as well (e.g., biophysical effects or
pharmacodynamic effects obtained in vitro, structure activity relationships, etc.) (Sec. 5.2). After
completion of the entries of WT3 for an observational context, the support for each claim of
carcinogenicity is summarized (from the lower part of entries) in the "column summary" of WT3 (Step
3.4). Upon reflection of the assessments shown in the column summary for coherence, the "overall
summary" entries are completed in WT3 (Step 3.5).
186
-------
Figure 4. The Seven Steps of Hazard Identification in Carcinogen
Risk Analysis
Step 1
Assemble Informational Base
Step 2
Define Contexts and Cases Within Contexts
For each Observational Context
(3.1)
(3.2)
(3.3)
(3.4)
(3.5)
For each case within the context
Determine data categories/items with
data available for c. of c. (intra-case)
Assess data characteristics for c. of c.
(intra-case)
I
Determine data categories/items with
data available for c. of c. (intra-context)
Assess data characteristics for c. of c.
(intra-context)
Assess support for c. of c. by
relevance strategy (intra-context)
I
Assess overall support for c. of c.
(intra-context)
187
-------
Figure 4. (continued)
Step 4
For each ordered pair of observational contexts (i,j, i*j)
(4.1)
(4.2)
(4.3)
(4.4)
(4.5)
(4.6)
Assess data characteristics for
extrapolation premises (i-j)
Assess data characteristics for
extrapolation premises (i-j)
Determine data categories/items with
data available for extrapolation premises
-------
Figure 4. (continued)
Step 5
For each observational context
Assess (1) overall support for c. of c. from all
sources (intra-context and inter-context),
(2) coherence of support (across contexts and
across relevance strategies), and (3)
completeness of evidence.
Step 6
Assess support for extrapolation
premises by relevance strategy (i-j)
Assess data characteristics for
extrapolation premises (H)
Determine data categories/items with
data available for extrapolation premises
(HI ,
Assess overall support from
observational context i for c. of c. in
context j (inter-context)
Assess support from context i for c. of c.
in context j by relevance strategy
(inter-context)
For each observational context (i) paired with each target context 0)
(6.5)
(6.4)
(6.3)
(6.2)
(6.1)
Step 7
For each target context
Assess (1) overall support for c. of c. from all
sources (intra-context and inter-context),
(2) coherence of support (across contexts and
across relevance strategies), and (3)
completeness of evidence.
189
-------
Figure 5. Flow Diagram for the Seven Steps of Hazard
Identification with Application of Working Tables
Step 1
Step 2
Step 3
0 Informational Base
2) Contexts/Cases
Assemble Informational Base
Define Contexts and Cases within Contexts
Do for OC(i), i-1 N
®—~
2)—~
Complete
WT2, WT3
for OC(i)
Intra-context
support for
c. of c. in OC(i)
Step 4
Do for OC(i), OCO), i, j-1 N(i*j)
Inter-context support for
extrapolation premises
for OC(i)-OCG)
for OC(i)
Inter-context support for
c. of c. by OC(i)-OC(j)
Complete WT6
for OC(i)-OC(j)
Complete
WT5 for
OC(i)-OC(j)
190
-------
Figure 5. (continued)
Step 5 Do for OC(i), i=1 ,...N
All OC(j)-OC(i), j=1,...N (j*i)
for OC(i)
Complete
WT7 for
OC(i)
Overall support for c. of c., for
coherence, and for complete-
ness of evidence in OC(i)
Step 6 Do for OC(i), TCG), i-1 N, j-1,...M
©—*¦
©—~
Inter-context support for
extrapolation premises
for OC(i)-TCfl)
(D—~
for OC(i)
~
6) Inter-context support for c. of c.
^ by OC(i)-TCG)
Complete WT4,
WT5 for
OC(i)-TC(j)
Complete
WT6
191
-------
Figure 5. (continued)
Step 7
Do for TC(i), i=1 M
6) All OCOVTC(i), j=1 N
1
r
Complete
WT7
for TC(i)
r Overall support for c. of c., for
coherence, and for complete-
ness of evidence in TC(i)
Key to abbreviations:
c. of c.: claims of carcinogenicity
OC(i): observational context i
TC(i): target context i
(D,(E>,f,: output from Steps 1, 2,...
OC(i)-OC(j): "Extrapolation from observational context i to
observational context j"
OC(i)-TC(j): "Extrapolation from observational context i to
target context j"
N: Number of observational contexts
M: Number of target contexts
WT1: Working Table 1 (similarly for WT2, WT3,...)
192
-------
Step 4 deals with extrapolating claims of carcinogenicity from one observational context (e.g., the
jth observational context, denoted as OC(i) in Figure 4) to another observational context (e.g., the jth
observational context, denoted as OC(j)). The necessity for this step arises
from the need to base the judgment for claims of carcinogenicity in each observational context on all
the information that may contain some evidential basis. At this point we have an intra-context
assessment for each observational context, i.e., only the data from within the context itself has been
utilized. In Step 4, Working Tables 4, 5, and 6 are completed for each pair of observational contexts.
This step for inter-context support of carcinogenicity bears some similarity to Step 3 for intra-context
support. Instead of completing WT2 for data related to carcinogenicity within the context, however,
WT4 (with the same headings) is completed for data items needed to determine the support for
extrapolating the intra-context claims for carcinogenicity in OC(i) to OC(j). The data items for this
purpose are listed in Table 2 and the role of each data category is described in Table 9. Steps 4.1-4.4
are conceptually analogous to Steps 3.1-3.4 with the difference being that the endpoint is a judgment of
support for extrapolation premises from OC(i) to OC(j), instead of intra-context support for
carcinogenicity (Sec. 5.3). Continuing the analogy, WT4 and WT5 play the previous roles of WT2 and
WT3. The overall objective of Step 4, however, is a judgment of support for claims of carcinogenicity in
OC(j) based on extrapolation from OC(i). That assessment is made in Step 4.5 (which has no
counterpart in Step 3), wherein entries in WT6 are completed based on the support for extrapolation
premises (in WT5) and the intra-context support for carcinogenicity in OC(i) (in WT3). It may be
noted that WT6 is identical in format to WT3, aside from the heading. The remaining portions of WT6
are completed in the same way as WT3.
Step 5 summarizes the overall support for each observational context from the intra-context
assessment for the context (WT3) and the inter-context assessments from all other observational
193
-------
contexts (WT6), which are entered into WT7. It remains to assess support for claims of carcinogenicity
in target contexts.
Steps 6 and 7 accomplish the assessment of support for claims of carcinogenicity in target
contexts. Since there is no intra-context assessment (because there are no observational data, by
definition), judgments are based totally on extrapolation from observational contexts. For each
observational context, data items related to premises for extrapolation to the target context of interest
are assembled and evaluated in the same manner as in extrapolation between observational contexts.
WT4 and WT5 are completed as in Step 4. The judgment of the support for claims of carcinogenicity
in the target context from each observational context is based on the support for extrapolation premises
(in WT5) and the overall support for claims of carcinogenicity in the observational context (in WT7 for
the observational context), and is entered into WT6 for the target context. The overall support for
claims of carcinogenicity in the target context is based on the support determined by extrapolation from
each observational context, and is entered in WT7 for the target context.
194
-------
REFERENCES (PART II)
1. Aaronson, A.; Tronick, S. (1986) The Role of Oncogenes in Human Neoplasia. Important
Advances in Oncology. Philadelphia: Lippincott Publishers.
2. Alarie, Y. (1981) Dose-Response Analysis in Animal Studies: Prediction of Human Responses.
Environmental Health Perspectives 42:9.
3. Allen, B.; Crump, K.; Shipp, A. (1988) Is it possible to predict the carcinogenic potency of a
chemical in humans using animal data? Banbury Report 31: Carcinogen Risk Assessment:
New Directions in the Qualitative and Quantitative Aspects.
4. Alston, W. (1985) Concepts of Epistemic Justification. The Monist 68:57-89.
5. Ames, B.; Gold, L. (1991a) Mitogenesis, Mutagenesis and Animal Cancer Tests. Chemically
Induced Cell Proliferation: Implications for Risk Assessment. New York: Wiley and Sons,
pp. 1-20.
6. Ames, B.M.; Gold, L.S. (1991b) Animal Cancer Tests and the Prevention of Cancer.
Proceedings of the International Council for Coordinating Cancer Research Conference.
7. Ames, B.M.; Gold, L.S. (1990) Chemical carcinogenesis: Too many rodent carcinogens. Proc
Natl Acad Sci USA 87:7772-7776.
8. Ames, B.M.; McCann, J.; Yamasaki, E. (1975) Method for detecting carcinogens and mutagens
with the Salmonella/mammalian microsome test. Mutation Research 31:347-364.
9. Andersen, M. (1981) A Physiologically Based Toxicokinetic Description of the Metabolism of
Inhaled Gases and Vapors: Analysis of Steady State. Toxicology and Applied
Pharmacology 60:509.
10. Andersen, M.; Archer, R.; Clewell, H.; MacNaughton, M. (1984) A Physiological Model of the
Intravenous and Inhalation Pharmacokinetics of Three Dihalomethanes-CH2Cl2, CH2rCl
and CH2Br2-in the Rat. Toxicologist 4:111.
11. Andersen, M.; Garags, M.; Clewell, M. (1986) Suicide Inactivation of Microsomal Oxidation by
cis- and trans-dichloroethylene (C-DCE and T-DCE) in Male Fisher Rats In-vivo.
Toxicologist 6:12.
12. Andersen, M.; Gargas, M.; Ramsey, J. (1987) Inhalation Pharmacokinetics: Evaluating Systemic
Extraction, Total In-vivo Metabolism and the Time Course of Enzyme Induction for
Inhaled Styrene in Rats Based on Arterial Blood: Inhaled Air Concentration Ratios.
Toxicology and Applied Pharmacology 73:176.
13. Andersen, M.E. (1989) Tissue Dosimetry, Physiologically-Based Pharmacokinetic Modeling, and
Cancer Risk Assessment. Cell Biology and Toxicology 5(4):405-415.
195
-------
REFERENCES (PART II) (continued)
14. Andersen, M.E.; Clewell III, H.J.; Gargas, M.L.; Smith, F.A.; Reitz, R.H. (1987) Physiologically
Based Pharmacokinetics and the Risk Assessment Process for Methylene Chloride.
Toxicology and Applied Pharmacology 87:185-205.
15. Andersen, M.E.; Clewell III, H.J.; Reitz, R.H. (1987) Physiologically based pharmacokinetics and
methylene chloride cancer risk assessment considerations. Human Risk Assessment-The
Role of Animal Selection and Extrapolation. London: Taylor and Francis Publishers, pp.
129-139.
16. Anderson, M.W. (1987) Issues in Biochemical Applications to Risk Assessment: How Do We
Evaluate Individual Components of Multistage Models? Environmental Health
Perspectives 76:175-179.
17. Anderson, M.W.; Hoel, D.G.; Kaplan, N.L. (1980) A General Scheme for the Incorporation of
Pharmacokinetics in Low-Dose Risk Estimation for Chemical Carcinogenesis: Example-
Vinyl Chloride. Toxicology and Applied Pharmacology 55:154-161.
18. Archer, L.E.; Ryan, L.M. (1989) On the Role of Cause-of-Death Data in the Analysis of
Rodent Tumorgenicity Experiments. Appl Statist 38(l):81-93.
19. Argus, M.; Hoch-Ligetic, C.; Arcos, J.; Conney, A. (1978) Differential Effects of fi-napthoflavone
and pregnenolone-16-7-carbonitrile on dimethylnitrosamine Induced
Hepatocarcinogenesis. Journal of the National Cancer Institute 61:441.
20. Arms, A.D.; Travis, C.C. (1988) Reference Physiological Parameters in Pharmacokinetic
Modeling. United States Environmental Protection Agency. EPA/600/6-88/004.
21. Ashby, J. (1986) Genetic Toxicology in Industry: Perspectives and Initiatives. Risk and Reason:
Risk Assessment in Relation to Environmental Mutagens and Carcinogens. New York:
Alan R. Liss, Inc., pp. 89-94.
22. Ashby, J. (1990) Genotoxicity Testing: To What Extent Can It Recognize Mutagens and
Carcinogens? Progress in Predictive Toxicology. New York: Elsevier, pp. 185-205.
23. Ashby, J.; Morrod, R.S. (1991) Detection of Human Carcinogens. Nature 352:185-186.
24. Ashby, J.; Tennant, R.W. (1991) Definitive Relationships among Chemical Structure,
Carcinogenicity and Mutagencity for 301 Chemicals Tested by the US NTP. Mutation
Research 257:229-306.
25. Ashby, J.; Tennant, R.W. (1988) Chemical Structure, Salmonella Mutagenicity and Extent of
Carcinogenicity as Indicators of Genotoxic Carcinogenesis among 222 Chemicals tested
in Rodents by the US NTP. Mutation Research 204:17-115.
26. Ashby, J.; Tennant, R.W.; Zeiger, E.; Stasiewicz, S. (1989) Classification according to Chemical
Structure, Mutagenicity to Salmonella and level of Carcinogenicity of a further 42
Chemicals Tested by the US NTP. Mutation Research 223:73-103.
196
-------
REFERENCES (PART II) (continued)
27. Bailey, A.; Sanchez, J.; Rodu, B.; Shrestha, K.; Tran-Paterson, R.; Saliba, D.; Miller, D. (1991)
Induced Proto-Oncogene Expression Increases Susceptibility to Radiation-Induced
Carcinogenesis. Chemically Induced Cell Proliferation: Close Up Implications for Risk
Assessment. New York: Wiley and Sons, pp. 105-114.
28. Bannasch, P.; Enzmann, H.; Zerban, H. (1987) Preneoplastic Lesions as Indicators of the
Carcinogenic Risk Caused by Chemicals. Cancer Risks: Strategies for Elimination.
Springer Verlag, Berlin, pp. 47-64.
29. Barrett, J. (1984) Tumor Promotion and Tumor Progression. Carcinogenesis-A Comprehensive
Survey, vol. 8. New York: Raven Press, pp. 423-429.
30. Barrett, J. (1987) A Multistep Model for Neoplastic Development: Role of Genetic and
Epigenetic changes. Mechanisms of Environmental Carcinogenesis, Vol. II. Florida: CRC
Press, pp. 117-126.
31. Barrett, J.C.; Kakunaga, T.; Kuroki, T.; Neubert, D.; Trosko, J.E.; Vasiliev, J.M.; Williams,
G.M.; Yamasaki, H. (1986a) Report 10: In-Vitro Assays That May Be Predictive Of
Tumour-Promoting Agents. Long-Term and Short-Term Assays for Carcinogens: A Critical
Appraisal: IARC 83:287-302.
32. Barrett, J.C.; Kakunaga, T.; Kuroki, T.; Neubert, D.; Trosko, J.E.; Vasiliev, J.M.; Williams,
G.M.; Yamasaki, H. (1986b) Report 9: Mammalian Cell Transformation in Culture.
Long-Term and Short-Term Assays for Carcinogens: A Critical Appraisal: IARC 83:267-286.
33. Barrett, J.C.; Wiseman, R.W. (1987) Cellular and Molecular Mechanisms of Multistep
Carcinogenesis: Relevance to Carcinogen Risk Assessment. Environmental Health
Perspectives 76:65-70.
34. Bartsch, H.; Armstrong, B. (eds.) (1982) Host Factors in Human Carcinogenesis.
35. Bartsch, H.; Malaveille, C. (1990) Screening assays for carcinogenic agents and mixtures: an
appraisal based on data in the IARC Monograph series. Complex Mixtures and Cancer
Risk, pp.65-74.
36. Baty, D. (1979) Species, Strain and Sex Differences in Metabolism. Foreign Comp Metabolism in
Mammals 5:159.
37. Baty, D. (1981) Species, Strain and Sex Differences in Metabolism. Foreign Comp Metabolism in
Mammals 6:133.
38. Beaune, P.; Kremers, P.; Kaminsky, L.; DeGraeve, J.; Guengerich, F. (1986) Comparison of
Monooxygenase Activities and Cytochrome P-450 Isozyme Concentrations in Human
Liver Microsomes. Drug Metabolism and Disposition: The Biological Fate of Chemicals
14:437.
197
-------
REFERENCES (PART II) (continued)
39. Belinsky, S.; Walker, V.; Maronpot, R.; Swenberg, J.; Andersen, M. (1987a) Molecular
Dosimetry of DNA Adduct Formation and Cell Toxicity in Nasal Mucosa Following
Exposure to a Tobacco Specific Nitrosamine: Relationship to Induction of Neoplasia.
Cancer Research 47:6058.
40. Belinsky, S.; White, C.; Devereux, T.; Swenberg, J.; Andersen, M. (1987b) Cell Selective
Alkylation of DNA in Rat Lung Following Low Dose Exposure to the Tobacco Specific
Carcinogen 4-(n-methyl-n-nitrosamine)-l-(3-pyridyl)-l-butanone. Cancer Research
47:1143.
41. Belinsky, S.A.; White, C.M.; Devereux, T.R.; Anderson, M.W. (1987c) DNA Adducts as a
Dosimeter for Risk Estimation. Environmental Health Perspectives 76:3-8.
42. Bernblum, I. (1985) Challenging Problems in Cocarcinogenesis. Cancer Research 45:1917-1921.
43. Berenblum, I.; Haran-Ghera, N. (1957) A Quantitative Study of the Systemic Initiating Action
of Urethane (Ethyl Carbamate) in Mouse Skin Carcinogenesis. British Journal of Cancer
11:77-84.
44. Bernstein, R. (1983) Beyond Objectivism and Relativism. Philadelphia, PA: University of North
Carolina Press.
45. Bertram, J. (1990) Role of Gap Junctional Cell/Cell Communication in the Control of
Proliferation and Neoplastic Transformation. Radiation Research 123:252-256.
46. Bickis, M.; Krewski, D. (1985) Statistical Design and Analysis of the Long-Term Carcinogenicity
Bioassay. Toxicological Risk Assessment, Vol 1: Biological and Statistical Criteria.
Florida: CRC Press, pp. 125-147.
47. Bickis, M.; Krewski, D. (1989) Statistical Issues in the Analysis of the Long-Term
Carcinogenicity Bioassay in Small Rodents: An Empirical Evaluation of Statistical
Decision Rules. Fundamental and Applied Toxicology 12:202-221.
48. Boberg, E.; Miller, E.; Miller, J.; Poland, A.; Liem, A. (1983) Strong Evidence from Studies with
Brachymorphic Mice and Pentachlorphenol That l'-sulfooxysafrole is the Major
Electrophilic and Carcinogenic Metabolite of l'-hydroxysafrole in Mouse Liver. Cancer
Research 43:5163.
49. Bogen, K. (1989) Cell Proliferation Kinetics and Multi-Stage Cancer Risk Models. Journal of
National Cancer Institute 81:267-277.
50. Boorman, G.A.; Montgomery Jr., C.A.; Eustis, S.L.; Wolfe, M.J.; McConnell, E.E.; Hardisty, J.F.
(1985) Chapter 16. Quality Assurance in Pathology for Rodent Carcinogenicity Studies.
Handbook of Carcinogen Testing. New Jersey: Noyes Publications.
51. Bos, J. (1988) The ras Gene Family and Human Cancer. Mutation Research 195:255-271.
198
-------
REFERENCES (PART II) (continued)
52. Boutwell, R. (1974) The Function and Mechanism of Promoters of Carcinogenesis. CRC
Critical Reviews of Toxicology 2:419-443.
53. Brandom, W.; Scaaomanno, G.; Archer, V.; Arhcer, P.; Bloom, A. (1978) Chromosome
Aberrations as a Biological Dose-Response Indicator of Radiation Exposure in Uranium
Miners. Radiation Research 76:159-171.
54. Breslow, N.E.; Day, N.E. (1980) Statistical Methods in Cancer Research. Volume I. The Analysis
of Case-Control Studies.
55. Breslow, N.E.; Day, N.E. (1987) Statistical Methods in Cancer Research. Volume II. The Design
and Analysis of Cohort Studies.
56. Brown, I. (1988) Rationality. Routledge, London.
57. Burchell, B.; Jackson, M.; Fennedy, S.; McCarthy, L.; Barr, G. (1985) Characterization and
Regulation of Hepatic UDP-glucuronyltransferase. Microsomes and Drug Oxidations. A.
Boobis, et al., ed.; pp. 212-220.
58. Burke, M.; Upshall, D. (1976) Species and Phenobarbitone-induced Differences in the Kinetic
Constants of Liver Microsomal Harmine O-demethylation. Xenobiotica 6:321.
59. Burkhart, K.; Kulig, K; McMartin, K. (1990) Formate Levels Following a Formalin Ingestion.
Vet. Human Toxicology 3:157-184.
60. Butterworth, B.E. (1990) Consideration of both Genotoxic and Nongenotoxic Mechanisms in
Predicting Carcinogenic Potential. Mutation Research 239:117-132.
61. Butterworth, B.; Slaga, T.; Farland, W.; McClaim, M. (1989) Chemically Induced Cell
Proliferation: Implications for Risk Assessment. Progress in Clinical and Biological
Research, vol. 369.
62. Casanova-Schmitz, M.; Heck, H. (1983) Effects of Formaldehyde Exposure on the Extractability
of DNA from Proteins in the Rat Nasal Mucosa. Toxicology and Applied Pharmacology
70:121.
63. Casanova-Schmitz, M.; Starr, T.; Heck, H. (1984) Differentiation Between Metabolic
Incorporation and Covalent Binding in the Labeling of Macromolecules in the Rat Nasal
Mucosa and Bone Marrow by Inhaled C14 and H3 Formaldehyde. Toxicology and Applied
Pharmacology 76:26.
64. Chin, T.; Heck, H. (1981) Disposition of C14 Formaldehyde in Fisher-344 Rats Following
Inhalation Exposure. Toxicologist 1:78.
65. Clayson, D.B. (1985) Dose Relationships in Experimental Carcinogenesis: Dependence on
Multiple Factors Including Biotransformation. Toxicologic Pathology ISSN: 0192-6233
13(2): 119-127.
199
-------
REFERENCES (PART II) (continued)
66. Clayson, D.B. (1987) ICPEMC Publication No. 13: The Need for Biological Risk Assessment in
Reaching Decisions about Carcinogens. Mutation Research 185:243-369.
67. Clayson, D.B. (1988) Needs for Biological Risk Assessment in Interspecies Extrapolation.
Environmental Health Perspectives 77:93-97.
68. Clayson, D.B.; Arnold, D.L. (1991) ICPEMC Publication No. 19: The Classification of
Carcinogens Identified in the Rodent Bioassay as Potential Risks to Humans: What Type
of Substance Should be Tested Next? Mutation Research 257:91-106.
69. Clayson, D.B.; Krewski, D. (1986) ICPEMC Working Paper No. 3: The Concept of Negativity in
Experimental Carcinogenesis. Mutation Research 167:233-40.
70. Clayson, D.B.; Krewski, D.; Munro, I.C. (1983) The Power and Interpretation of the
Carcinogenicity Bioassay. Regulatory Toxicology and Pharmacology 3:329-348.
71. Clewell III, H.J.; Andersen, M.E. (1989) Biologically Motivated Models for Chemical Risk
Assessment. Health Physics 57(1): 129-137.
72. Code of Federal Regulations. (1983a) Toxic Substances Control: Good Laboratory Practice
Standards-Final Rule. C.F.R. §40:53922-53944.
73. Code of Federal Regulations. (1983b) Pesticide Programs: Good Laboratory Practice
Standards-Final Rule. C.F.R. §40:53946-53969.
74. Cohen, S.M.; Ellwein, L.B. (1990) Cell Proliferation in Carcinogenesis. Science 249:1007-1011.
75. Cohrssen, J.J.; Covello, V.T. (1989) Risk Analysis: A Guide to Principles and Methods for
Analyzing Health and Environmental Risks.
76. Cone, J.E.; Reeve, G.R.; Landrigan, P.J. (1987) Chapter 6. Clinical and Epidemiological
Studies. Toxic Substances and Human Risk. New York: Plenum Press.
77. Conney, A. (1982) Induction of Microsomal Enzymes by Foreign Chemicals and Carcinogenesis
by Polycyclic Aromatic Chemicals. Cancer Research 82:4975.
78. Conney, A.; Buening, M.; Pantuck, E.; Pantuck, C.; Fortner, J.; Anderson, K.; Kappas, A. (1980)
Regulation of Human Drug Metabolism by Dietary Factors. Environmental Chemicals,
Enzyme Function and Human Disease, Ciba Foundation Symposium Vol. 76.
79. Cook, R.R. (1982) The Role of Epidemiology in Risk Assessment. Drug Metabolism Reviews
13(5):913-923.
80. Cooper, G.M. (1990) Chapter 1: The Cancer Cell. Oncogenes. Boston: Jones and Bartlett
Publishers, pp. 3-18.
200
-------
REFERENCES (PART II) (continued)
81. Counts, C; Bruckner, J.; Feldman, S. (1982) Role of Dose Level, Food Intake and Diluent in
Toxicokinetics of Orally Administered 1,1-dichloroethylene (1,1 DCE). Pharmacologist
24:171.
82. Craig, P.; Enstein, K. (1981) Structure-Activity in Hazard Assessment. Hazard Assessment of
Chemicals: Current Development, Vol. 1. New York: National Academy Press, pp. 389-
420.
83. Crawford-Brown, D.; Hofmann, W. (1989) The Role of Variability of Dose in Dose-Response
Relationships for Alpha Emitting Radionuclides. Radiation Protection Dosimetry 28:283-
290.
84. Crawford-Brown, D.; Hofmann, W. (1990) A Generalized State-Vector Model for Radiation-
Induced Cellular Transformation. International Journal of Radiation Biology 57:407-423.
85. Cross, F. (1987) Health Effects. Environmental Radon. New York: Plenum Press, pp. 215-243.
86. Crump, K.; Allen, B. (1984) Quantitative Estimates of Risk from Occupational Exposure to
Benzene. Report to OSHA in May.
87. Curtis, S. (1991) Mechanisitic Models. Physical and Chemical Mechanisms in Molecular
Radiation Biology. New York: Plenum Press, pp. 367-385.
88. Dahlman, T.; Rhodin, J. (1956) Effects of Exposure to Lung-Irritating Gas on Transport of
Secretion and Fibrillation Frequency in Rat Trachea. Nordisk Hygienisk Tidskrist 37:82.
89. Day, N.E. (1988) Chapter 1. Epidemiological Methods for the Assessment of Human Cancer
Risk. Toxicological Risk Assessment, Vol. II. Florida: CRC Press.
90. Della-Favera, R.; Martinotti, S.; Gallo, R.; Erickson, J.; Croce, C. (1988) Translocation and
Rearrangements of the c-myc Oncogene Locus in Human Undifferentiated B-cell
Lymphomas. Science 219:963-997.
91. DerSimonian, R.; Laird, N. (1986) Meta-analysis in Clincal Trials. Controlled Clinical Trials
7:177.
92. Dietz, F.; Ramsey, J.; Watanabe, P. (1983) Relevance of Experimental Studies to Human Risk.
Environmental Health Perspectives 52:9.
93. Doll, R. (1985) General Conclusions. Interpretation of Negative Epidemiological Evidence for
Carcinogenicity, IARC Scientific Publications No. 65.
94. Dybing, E. (1986) Predictability of Human Carcinogenicity from Animal Studies. Regulatory
Toxicology and Pharmacology 6:399.
95. Eccles, S.A. (1987) Chapter 6: Animal models of cancer. Biology of Carcinogenesis. Lancaster:
MTP Press Limited, pp.109-139.
201
-------
REFERENCES (PART II) (continued)
96. Eddy, D.M. (1989) The Confidence Profile Method: A Bayesian Method for Assessing Health
Technologies. Oper Res 37:210.
97. Eddy, D.M.; Hasselblad, V.; Shachter, R. (1990a) An Introduction to a Bayesian Method for
Meta-analysis: The Confidence Profile Method. Med Decision Making 10:15.
98. Eddy, D.M.; Hasselblad, V.; Shachter, R. (1990b) A Bayesian Method for Synthesizing
Evidence: The Confidence Profile Method. Int. J. Technol. Asses. Health Care 6:31.
99. Environmental Health Criteria 109 (1990) Summary Report on the Evaluation of Short-term Tests
for Carcinogenesis (Collaborative Study on In-vivo Tests). World Health Organization.
100. European Chemical Industry Ecology and Toxicology Center (1981) The Mutagenic and
Carcinogenic Potential of Formaldehyde Technical Report No. 2. ECETOC, Brussels.
101. Evans, A. (1978) Casusation and Disease: A Chronological Journal. American Journal of
Epidemiology 108:249-258.
102. Farber, E. (1987) Possible Etiologic Mechanisms in Chemical Carcinogenesis. Environmental
Health Perspectives 75:65-70.
103. Farrar, D.B.; Crump, K.S. (1988) Exact Statistical Tests for Any Carcinogenic Effect in Animal
Bioassays. Fundamental and Applied Toxicology 11:652-663.
104. Fiserova-Bergerova, V. (1976) Mathematical Modeling of Inhalation Exposure. Journal of
Combustion Toxicology 3:201.
105. Fiserova-Bergerova, V.; Vlach, V.; Cassady, J. (1980) Predictable Individual Differences in
Uptake and Excretion of Gases and Lipid Soluble Vapors: Simulation Study. British
Journal of Industrial Medicine 37:42.
106. Friedman, M.; Wehr, C.; Schade, J.; MacGregor, J. (1982) Inactivation of Aflatoxin Bx
Mutagenicity By Thiols. Food and Chemical Toxicology 20:887.
107. Gargas, M.; Clewell, H.; Andersen (1986) Metabolism of Inhaled Dihalomethanes In-vivo:
Differentation of Kinetic Constants for Two Independent Pathways. Toxicology and
Applied Pharmacology 82:211.
108. Gart, J.J.; Chu, K.C.; Tarone, R.E. (1979) Statistical Issues in Interpretation of Chronic
Bioassay Tests for Carcinogenicity. Journal of the National Cancer Institute 62(4):957.
109. Gart, J.J.; Krewski, D.; Lee, P.N.; Tarone, R.E.; Wahrendorf, J. (1986) Statistical Methods in
Cancer Research. Volume III. The Design and Analysis of Long-Term Animal Experiments.
110. Gehring, P.; Watanabe, P.; Park, C. (1978) Resolution of Dose-Response Toxicity Data for
Chemicals Requiring Metabolic Activation-Example-Vinyl Chloride. Toxicology and
Applied Pharmacology 44:581.
202
-------
REFERENCES (PART II) (continued)
111. Gelboin, H.; PTsO (1978) Polycylic Hydrocarbons and Cancer, Vols. 1 and 2. New York:
Academic Press.
112. Gerlowski, L.; Jain, R. (1983) Physiologically Based Pharmacokinetic Modeling: Principles and
Applications. Journal of Pharmaceutical Science 72:1103.
113. Ghoshal, A.; Rushmore, T.; Farber, E. (1987) Induction of Cancer by Dietary Deficiency without
Added Carcinogens. Nongenotoxic Mechanisms in Carcinogenesis: Banbury Report 25:179-
187.
114. Gil, J. (1982) Comparative Morphology and Ultrastructure of the Airways. Mechanisms in
Respiratory Toxicology. Boca Raton, LA: CRC Press, pp. 3-25.
115. Gillette, J. (1976) Application of Pharmacokinetic Principles in the Extrapolation of Animal
Data to Humans. Clinical Toxicology 9:709.
116. Gillette, J. (1984) Problems in Correlating In-vitro and In-vivo Studies of Drug Metabolism.
Pharmacokinetics: A Modern View. New York: Plenum Press.
117. Graham, J.D.; Green, L.C.; Roberts, M.J. (1988) In Search of Safety: Chemicals and Cancer Risk.
Massachusetts: Harvard University Press.
118. Greenland, S. (1987) Quantitative Methods in the Review of Epidemiologic Literature.
Epidemiologic Reviews 9:1-30.
119. Gregory, A.R. (1988) Species Comparisons in Evaluating Carcinogenicity in Humans. Regulatory
Toxicology and Pharmacology 8:160-190.
120. Gustafsson, J.; MacGeoch, C.; Morgan, E. (1985) Isolation Characterization and Regulation of
Sex Specific Isozymes of Cytochrome P-450 Catalyzing 15 Beta-hydroxylation of Steroid
Sulfates and 16-alpha-hydroxylation of 4- and rostene-3, 17-dione. Microsomes and Drug
Oxidations. New York: Taylor & Francis.
121. Guyton, A. (1947) Measurement of the Respiration Volumes of Laboratory Animals. American
Journal of Physiology 150:70.
122. Hamm Jr., T.E. (1985) Chapter 12. Design of a Long-Term Animal Bioassay for
Carcinogenicity. Handbook of Carcinogen Testing. New Jersey: Noyes Publications, pp.
252-267.
123. Hall, E.; Freyer, G. (1991) The Molecular Biology of Radiation Carcinogenesis. Physical and
Chemical Mechanisms in Molecular Radiation Biology. New York: Plenum Press, pp. 3-
28.
124. Harper, B.L.; Legator, M.S. (1987) Tumor Promoters and Genotoxic Chemicals in Short-Term
Testing for Carcinogenicity. Advances in Veterinary Science and Comparative Medicine 8:1-
19.
203
-------
REFERENCES (PART II) (continued)
125. Hartman, T.; Rosen, J. (1983) Inhibition of Metabolic Cooperation by Cigarette Smoke
Condensate and Its Fractions in V-79 Chinese Hamster Lung Fibroblasts. Proceedings
of the National Academy of Science 80:5305-5309.
126. Haseman, J.K. (1983) Statistical Support of the Proposed National Toxicology Program
Protocol. Toxicologic Pathology 1(1):77.
127. Haseman, J.K. (1984) Statistical Issues in the Design, Analysis and Interpretation of Animal
Carcinogenicity Studies. Environmental Health Perspectives 58:385-392.
128. Haseman, J.K. (1985) Issues in Carcinogenicity Testing: Dose Selction. Fundamental and
Applied Toxicology 5:66-78.
129. Haseman, J.K. (1989) Commentary on "Are There Location/Cage/Systematic Nontreatment
Effects in Long-Term Rodent Studies? A Question Revisited". Fundamental and Applied
Toxicology 13:189-192.
130. Haseman, J.K. (1990) Use of Statistical Decision Rules for Evaluating Laboratory Animal
Carcinogenicity Studies. Fundamental and Applied Toxicology 14:637-648.
131. Haseman, J.K.; Clark, A. (1990) Carcinogenicity Results for 114 Laboratory Animal Studies
Used to Assess the Predictivity of Four In Vitro Genetic Toxicity Assays for Rodent
Carcinogenicity. Environmental and Molecular Mutagenesis 16(18): 15-31.
132. Haseman, J.K.; Huff, J.E. (1987) Species Correlation in Long-Term Carcinogenicity Studies.
Cancer Lettets 37:125-132.
133. Haseman, J.K,; Huff, J.; Boorman, G.A. (1984) Use of Historical Control Data in
Carcinogenicity Studies in Rodents. Toxicologic Pathology 12(2): 126-135.
134. Haseman, J.K.; Huff, J.E.; Rao, G.N.; Eustis, S.L. (1989) Sources of Variability in Rodent
Carcinogenicity Studies. Fundamental and Applied Toxicology 12:793-804.
135. Haseman, J.K.; Huff, J.E.; Zeiger, E.; McConnell, E.E. (1987) Comparative Results of 327
Chemical Carcinogenicity Studies. Environmental Health Perspectives 74:229-235.
136. Hattis, D. (1990) Pharmacokinetic Principles for Dose-Rate Extrapolation of Carcinogenic Risk
from Genetically Active Agents. Risk Analysis 10(2):303.
137. Hattis, D.; White, P.; Marmorstein, L.; Koch, P. (1990) Uncertainties in Pharmacokinetic
Modeling for Perchloroethylene. I. Comparison of Model Structure, Parameters, and
Predictions for Low-Dose Metabolism Rates for Models Derived by Different Authors.
Risk Analysis 10(3):449.
138. Hecker, E. (1984) Cocarcinogens of the Tumour-Promoter Type as Potential Risk Factors of
Cancer in Man. Models, Mechanisms and Etiology of Tumour Promotion: IARC 56:441-
463.
204
-------
REFERENCES (PART II) (continued)
139. Hennings, H.; Shores, R.; Wenk, M.; Spangler, E.; Tarone, R.; Yuspa, S. (1983) Malignant
Conversion of Mouse Skin Tumours is Increased by Tumour Initiators and Unaffected
by Tumour Promoters. Nature 304:67-69.
140. Hermo Jr., H.; Brandt-Rauf, P.W. (1987) Chemically Induced Oncogenesis. Seminars in
Occupational Medicine 2(4):235-43.
141. Heymann, E. (1980) Carboxylesterases and Amidase. Enzymatic Basis of Detoxification, Vol. II.
New York: Academic Press.
142. Hill, A. (1965) The Environment and Disease: Association and Causation. Proceedings of the
Royal Society of Medicine 58:295-300.
143. Himmelstein, K.; Lutz, R. (1979) A Review of the Application of Physiologically Based
Pharmacokinetic Modeling. Journal of Pharmacokinetics and Biopharmaceutics 7:127.
144. Hoel, D.G. (1985) Epidemiology of Cancer Mechanisms. National Cancer Inst Monogr 67:199-
203.
145. Hoel, D.; Kaplan, N.; Andersen, M. (1983) Implication of Nonlinear Kinetics on Risk Estimation
in Carcinogenesis. Science 219:1032.
146. Hoel, D.G.; Haseman, J.K.; Hogan, M.D.; Huff, J.; McConnell, E.E. (1988) The Impact of
Toxicity on Carcinogenicity Studies: Implications for Risk Assessment. Carcinogenesis
9(ll):2045-2052.
147. Hoel, D.G.; Landrigan, PJ. (1987) Chapter 7. Comprehensive Evaluation of Human Data.
Toxic Substances and Human Risk. New York: Plenum Press.
148. Hogben, C. (1971) Biological Membranes and Their Passage By Drugs. Concepts in Biochemical
Pharmacology, Vol. 28:1-8.
149. Homburger, F.; Tregier, A. (1969) Modifiers of Carcinogenesis. Progress in Experimental Tumor
Research, Vol II. New York: Karger, pp. 86-89.
150. Huff, J.E.; Eustis, S.L.; Haseman, J.K. (1989) Occurrence and Relevance of Chemically Induced
Benign Neoplasms in Long-term Carcinogenicity Studies. Cancer and Metastasis Reviews
8:1-21.
151. Huff, J.E.; McConnell, E.E.; Haseman, J.K.; Boorman, G.A.; Eustis, S.L.; Schwetz, B.A.; Rao,
G.N.; Jameson, C.W.; Hart, L.G.; Rail, D.P. (1986) Carcinogenesis Studies Results of
398 Experiments on 104 Chemicals. National Toxicology Program.
152. IARC, International Agency for Research on Cancer (1986) Long-Term and Short-Term Assays
for Carcinogenesis, IARC Scientific Publications, No. 83.
205
-------
REFERENCES (PART II) (continued)
153. Idle, J. (1980) Discussion. Environmental Chemicals, Enzyme Function and Human Disease, Ciba
Foundation Symposium Vol. 76, p. 284.
154. Jakoby, W.; Sekura, R.; Lyon, E.; Marco, C.; Wang, J. (1980) Sulfotransferases. Enzymatic Basis
of Detoxification. New York: Academic Press.
155. Kappas, A.; Alvarez, A.; Anderson, K.; Garland, W.; Pantuck, E.; Conney, A. (1977) The
Regulation of Human Drug Metabolism by Nutritional Factors. Drug Microsomes and
Drug Oxidations. New York: Pergamon Press.
156. Keck, G. (1981) Species Differences in Pharmacokinetics of Chemicals: Correlations with
Toxicity. Toxicology and Envrionmental Research 3:207.
157. Kety, S. (1951) The Theory and Applications of the Exchange of Inert Gas at the Lungs and
Tissues. Pharmacological Reviews 3:1.
158. Kinoshita, N.; Gelboin, G. (1972) Aryl Hydrocarbon Hydrolase and Polycyclic Hydrocarbon
Tumorigenesis: Effect of the Enzyme Inhibitor 7,8 benzoflavone on Tumorigenesis and
Macromolecular Binding. Proceedings of the National Academy of Sciences 69:824.
159. Kleinbloesem, C.; VanBrummeler, P.; Faber, H.; Danhof, M.; Vermuelen, N.; Breimer, D.
(1984) Variability and Dynamics: A Net Oxidation Polymorphism in Man. Biochemical
Pharmacology 33:3721.
160. Kopun, M.; Propping, P. (1977) The Kinetics of Ethanol Absorption and Elimination in Twins
and Supplementary Repetitive Experiments in Singleton Subjects. European Journal of
Clinical Pharmacology 11:337.
161. Krasovskii, G. (1976) Extrapolation of Experimental Data from Animals to Man. Environmental
Health Perspectives 13:51.
162. Krewski, D.; Colin, D.; Dewanji, A. (1990) Chapter 16. Statistical Considerations in the Design
of Toxicological Experiments with Chemical Mixtures. Progress in Predictive Toxicology.
New York: Elsevier Publishers.
163. Krewski, D.; Goddard, M.J.; Murdoch, D. (1989) Statistical Considerations in the
Interpretation of Negative Carcinogenicity Data. Regulatory Toxicology and
Pharmacology 9:5-22.
164. Krewski, D.; Murdoch, D.J.; Withey, J.R. (1986) Paper: The Application of Pharmacokinetic
Data in Carcinogenic Risk Assessment.
165. Krewski, D.; Smythe, R.T.; Dewanji, A. (1988) Tests for Trend in Carcinogen Bioassay with
Historical Controls.
166. Krewski, D.; Wigle, D.; Clayson, D.B.; Howe, G.R. (1990) Role of Epidemiology in Health
Risk Assessment. Occupational Cancer Epidemiology. New York: Springer-Verlad.
206
-------
REFERENCES (PART II) (continued)
167. Kuhn, T. (1962) The Structure of Scientific Revolutions. Chicago, IL: University of Chicago
Press.
168. Langenbach, R.; Tennant, R.W.; Barrett, J.C. (1988) Tumor Promotion: A Discussion of
Current Topics, Available Assay Systems and Research Needs. Tumor Promoters:
Biological Approaches for Mechanistic Studies and Assay Systems. New York: Raven Press,
pp. 441-448.
169. Laudan, L. (1977) Progress and Its Problems: Towards a Theory of Scientific Growth. Berkley,
CA: University of California Press.
170. Lawley, P.D. (1987) Concepts of Carcinogenesis. Biology of Carcinogenesis. Lancaster: MTP
Press Limited, pp. 1-22.
171. Levin, W.; Thomas, P.; Reitz, L.; Ryan, D.; Bandiera, S.; Haniu, M.; Shivley, J. (1985)
Immunochemical and Stuctural Characterization of Rat Hepatic Cytochrome P-450.
Microsomes and Drug Oxidations. New York: Taylor & Francis.
172. Lewis, J.G.; Adams, D.O. (1987) Inflammation, Oxidative DNA Damage, and Carcinogenesis.
Environmental Health Perspectives 76:19-27.
173. Longino, H. (1990) Science as Social Knowledge. Princeton, NJ: Princeton University Press.
174. Loury, D.J.; Goldsworthy, T.L.; Butterworth, B.E. (1987) The Value of Measuring Cell
Replication as a Predictive Index of Tissue-specific Tumorigenic Potential. Nongenotoxic
Mechanisms in Carcinogenesis: Banbury Report 25:119-136.
175. Lucier, G.W.; Thompson, C.L. (1987) Issues in Biochemical Applications to Risk Assessment:
When Can Lymphocytes be used as Surrogate Markers? Environmental Health
Perspectives 76:187-191.
176. Lutz, R.J.; Dedrick, R.L. (1987) Implications of Pharmacokinetic Modeling in Risk Assessment
Analysis. Environmental Health Perspectives 76:97-106.
177. Maclure, M. (1985) Popperian Refutation in Epidemiology. American Journal of Epidemiology
121:343-350.
178. Mann, C. (1990) Meta-Analysis in the Breech. Science 249:476-480.
179. Mantel, N. (1980) Assessing Laboratory Evidence for Neoplastic Activity. Biometrics 36:381-
399.
180. Maronpot, R.R. (1985) Chapter 18. Considerations in the Evaluation and Interpretation of
Long-Term Animal Bioassays for Carcinogenicity. Handbook of Carcinogen Testing.
New Jersey: Noyes Publications, pp. 372-383.
207
-------
REFERENCES (PART II) (continued)
181. Mattison, D.R., Jelovsek, F.R. (1987) Pharmacokinetics and Expert Systems as Aids for Risk
Assessment in Reproductive Toxicology. Environmental Health Perspectives 76:107-119.
182. McDougal, J.; Jepson, G.; Clewell, H.; MacNaughton, M.; Andersen, M. (1986) A Physiological
Pharmacokinetic Model for Dermal Absorption. Toxicology and Applied Pharmacology
85:286.
183. McNight, B. (1988) A Guide to the Statistical Analysis of Long-Term Carcinogenicity Assays.
Fundamental and Applied Toxicology 10:355-364.
184. Mettlin, C. (1987) Descriptive and Analytic Epidemiology. Cancer 62:1680-1687.
185. McBodile, M.; Holscher, M.; Neal, R. (1975) A Possible Protective Role for Reduced
Glutathione in Aflatoxin B1 Toxicity: Effects of Pretreatment of Rats with Phenobarbital
and 3-methylcholanthreneon Aflatoxin Toxicity. Toxicology and Applied Pharmacology
34:128.
186. Miller, E.; Miller, J. (1976) The Metabolism of Chemical Carcinogens to Reactive Electrophiles
and Their Possible Mechanisms of Action in Carcinogenesis. Chemical Carcinogens.
Washington, D.C.: American Chemical Society, p.737.
187. Miller, R.; Guengerich, F. (1983) Metabolism of Trichloroethylene in Isolated Hepatoxytes,
Microsomes and Reconstituted Enzyme Systems Containing Purified Cytochromes P-450.
Cancer Research 43:1145.
188. Moolgavkar, S. (1991) Carcinogenesis Models: An Overview. Physical and Chemical
Mechanisms in Molecular Radiation Biology. New York: Plenum Press, pp. 387-402.
189. Moolgavkar, S.H. (1986) Hormones and multistage carcinogenesis. Cancer Surveys 5:635-648.
190. Morgan, K.; Patterson, D.; Gross, E. (1983) Formaldehyde and the Nasal Mucociliary
Apparatus. Formaldehyde: Toxicology, Epidemiology and Mechanisms. New York: Marcel
Dekker Inc., pp. 193-209.
191. Morris, S.C. (1990) Cancer Risk Assessment. New York: Marcel Dekker, Inc.
192. Muir, C.S.; Higginson, J. (1985) Chapter 2: Role of Epidemiology in Identifying Chemical
Carcinogens. Handbook of Carcinogen Testing. New Jersey: Noyes Publications, pp. 28-
57.
193. National Academy of Sciences (1980) Principles of Toxicological Interactions Associated with
Multiple Chemical Exposures. Washington, D.C.: National Academy Press.
194. National Research Council (NRC) (1983) Risk Assessment in the Federal Government:
Managing the Process. Washington, D.C.; National Academy Press.
208
-------
REFERENCES (PART II) (continued)
195. National Research Council (NRC) (1986) Dose-Route Extrapolations: Using Inhalation Toxicity
Data to Set Drinking Water Limits. Drinking Water and Health, Vol 6. Washington,
D.C.: National Academy Press.
196. National Research Council (NRC) (1987) Pharmacokinetics in Risk Assessment: Drinking Water
and Health, Volume 8. Washington, D.C.: National Academy Press.
197. National Research Council (NRC) (1988) Health Risks of Radon and Other Internally Deposited
Alpha Emitters. BEIR IV, Washington, D.C.: National Academy Press.
198. National Research Council (NRC) (1989) Drinking Water and Health, Volume 9: Selected Issues
in Risk Assessment. Washington, D.C.: National Academy Press.
199. National Research Council (NRC) (1991) Comparative Dosimetry of Radon in Mines and Homes.
Washington, D.C.: National Academy Press.
200. National Toxicology Program (NTP). (1984) Report of the NTP ad hoc Panel on Chemical
Carcinogenesis Testing and Evaluation. U.S. Department of Health and Human Services,
Washington, D.C.
201. Nesnow, S.; Bergman, H. (1988) An Analysis of the Gene-Tox Carcinogen Data Base. Mutation
Research 205:237-253.
202. Newton-Smith, W. (1981) The Rationality of Science. London: Routledge and Kegan Paul.
203. OSHA (1980) Identification, Classification and Regulation of Potential Occupational
Carcinogens. Federal Register 45:5129.
204. OSHA (1980) Occupational Exposure to Formaldehyde. Federal Register 50:50458.
205. Office of Science and Technology Policy (OSTP) (1985) Chemical Carcinogens: Review of the
Science and its Associated Principles. Federal Register 50:10372-10442.
206. Park, C.N.; Kociba, R.J. (1985) Chapter 17. Statistical Evaluation of Long-Term Animal
Bioassays for Carcinogenicity. Handbook of Carcinogen Testing. New Jersey: Noyes
Publications, pp. 358-371.
207. Paustenbach, D.J.; Clewell III, H.J.; Gargas, M.L.; Andersen, M.E. (1988) A Physiologically
Based Pharmacokinetic Model for Inhaled Carbon Tetrachloride. Toxicology and Applied
Pharmacology 96:191-211.
208. Perera, F. (1987) The Potential Usefulness of Biological Markers in Risk Assessment.
Environmental Health Perspectives 76:141-145.
209. Perera, F.P. (1984) The Genotoxic/Epigenetic Distinction: Relevance to Cancer Policy.
Environmental Research 34:175-191.
209
-------
REFERENCES (PART II) (continued)
210. Peto, J. (1984) Early- and Late-Stage Carcinogenesis in Mouse Skin and in Man. Models,
Mechanisms and Etiology of Tumour Promotion: IARC 56:359.
211. Peto, R. (1985) Keynote Address: Epidemiological Reservations about Risk Assessment. Basic
Life Sciences, Vol. 33: Assessment of Risk from Low-Level Exposure to Radiation and
Chemicals.
212. Peto, R.; Pike, M.C.; Day, N.E.; Gray, R.G.; Lee, R.N.; Parish, S.; Peto, J.; Richards, S.;
Wahrendork, J. (1980) Long-Term and Short-Term Screening Assays for Carcinogens:
A Critical Appraisal. Supplement 2, IARC, Monographs on the Evaluation of the
Carcinogenic Risk of Chemicals to Man.
213. Piegorsch, W.W.; Hoel, D.G. (1988) Exploring Relationships Between Mutagenic and
Carcinogenic Potencies. Mutation Research 196:161-175.
214. Pitot, H.C.; Campbell, H.A. (1987) An Approach to the Determination of the Relative Potencies
of Chemical Agents During the Stages of Initiation and Promotion in Multistage
Hepatocarcinogenesis in the Rat. Environmental Health Perspectives 76:49-56.
215. Portier, C.; Hoel, D. (1983) Optimal Design of the Chronic Animal Bioassay. Journal of
Toxicology and Environmental Health 12:1-19.
216. Portier, C.J. (1988) Species Correlation of Chemical Carcinogens. Risk Analysis 8(4):551-553.
217. Portier, C.J.; Bailer, A.J. (1989) Testing for Increased Carcinogenicity Using a Survival-
Adjusted Quantal Response Test. Fundamental and Applied Toxicology 12:731-737.
218. Prejean, J.D. (1985) Chapter 13. Conduct of Long-Term Animal Bioassays. Handbook of
Carcinogen Testing. New Jersey: Noyes Publications, pp. 268-281.
219. Purchase, I. (1985) Carcinogenic Risk Assessment: A Toxicologists View. Risk Quantitation and
Regulatory Policy: Banbury Report 19:175-186.
220. Purchase, I.F.H.; Stafford, J.; Paddle, G.M. (1985) Chapter 8: Vinyl Chloride-A Cancer Case
Study. Toxicological Risk Assessment, Vol. 2. CRC Press, pp. 167-194.
221. Ramsey, J.; Andersen, M. (1984) A Physiologically Based Description of the Inhalation
Pharmacokinetics of Styrene in Rats and Humans. Toxicology and Applied Pharmacology
73:159.
222. Redmond, C.K. (1981) Sensitive Population Subsets in Relation to Effects of Low Doses.
Environmental Health Perspectives 42:137-140.
223. Reitz, R.; Fox, T.; Quast, J. (1982) Mechanistic Considerations for Carcinogenic Risk
Estimations: Chloroform. Environmental Health Perspectives 46:163.
210
-------
REFERENCES (PART II) (continued)
224. Rose, M.; Lock, E.; Smith, L.; Wyatt, I. (1976) Paraquat Accumulation: Tissues and Species
Specificity. Biochemical Pharmacology 25:419.
225. Rosin, M.; Stich, H. (1978) The Inhibitory Effect of Cysteine on the Mutagenic Activities of
Several Carcinogens. Mutation Research 54:73.
226. Samet, J. (1989) Radon and Lung Cancer. Journal of the National Cancer Institute 81:745-755.
227. Saracci, R. (1977) Asbestos and Lung Cancer: An Analysis of Epidemiological Evidence on the
Asbestos-smoking Interaction. International Journal of Cancer 20:323.
228. Sato, A.; Nakajima, T. (1979) Partition Coefficients of Some Aromatic Hydrocarbons and
Ketones in Water, Blood and Oil. British Journal of Industrial Medicine 36:231.
229. Scala, J.; McOsker, D.; Reller, H. (1968) The Percutaneous Absorption of Ionic Surfactants.
Journal of Investigative Dermatology 50:371.
230. Scherer, E.; Feringa, A.; Ennelot, P. (1984) Initiation-Promotion-Inititaion: Induction of
Neoplastic Foci within Islands of Precancerous Liver Cells in Rats. Models, Mechanisms
and Etiology of Tumour Promotion: IARC 56.
231. Schulte-Hermann, R.; Bursch, W.; Parzefall, W. (1991) Mitogenesis and Programmed Cell
Death as Determinants of Carcinogenicity of Nongenotoxic Chemicals. Chemically
Induced Cell Proliferation: Implications for Risk Assessment. New York: Wiley and Sons,
pp. 237-244.
232. Schwartz, J.; Riddiourgh, C.; Epste, S. (1982) Analysis of Carcinogenesis Dose-Response
Relations with Dichotomous Data. Teratogenesis, Carcinogenesis and Mutagenesis 2:179.
233. Schwetz, B.A.; Yang, R.S.H. (1990) Approaches used by the US National Toxicology Program in
assessing the toxicity of chemical mixtures. Complex Mixtures and Cancer Risk, pp. 113-
120.
234. Scribner, H.E.; McCarthy, K.L.; Doolittle, D.J. (1987) Practical Approaches to Evaluating
Nongenotoxic Carcinogens. Nongenotoxic Mechanisms in Carcinogenesis: Banbury Report
25:355-364.
235. Shelby, M.D.; Stasiewicz, S. (1984) Chemicals Showing No Evidence of Carcinogenicity in
Long-Term, Two-Species Rodent Studies: The Need for Short-Term Test Data.
Evnironmental Mutagen, in press.
236. Short, B.; Swenberg, J. (1990) Chemically Induced Cell Proliferation in Carcinogenesis in the
Male Rat Kidney. Chemically Induced Cell Proliferation: Implications for Risk Assessment.
New York: Wiley and Sons, pp. 357-368.
211
-------
REFERENCES (PART II) (continued)
237. Shubik, P.; Clayson, D.B. (1990) Chapter 8: Carcinogenesis in Animals: Which Agents Pose the
Greatest Problems for Humans? Progress in Predictive Toxicology. New York: Elsevier
Science Publishers, pp. 145-160.
238. Sjoqvist, F.; Borga, O.; Orme, M. (1976) Fundamentals of Clinical Pharmacology. Drug
Treatment: Principles and Practice of Clinical Pharmacology and Therapeutics. Australia:
AIDS.
239. Slaga, T.J. (1984) Can Tumour Promotion be Effectively Inhibited? Models, Mechanisms and
Etiology of Tumour Promotion: IARC 56:497-506.
240. Slaga, T.J. (1988) Interspecies Comparisons of Tissue DNA Damage, Repair, Fixation, and
Replication. Environmental Health Perspectives 77:73-82.
241. Smith, R. (1973) The Excretory Function of Bile: The Elimination of Drugs and Toxic Substances
in Bile. New York: Wiley & Sons.
242. Sorsa, M.; Vainio, H.; McMichael, A.J. (1990) Experimental and Epidemiologic Applications to
Cancer Risk Assessment of Complex Mixtures. Genetic Toxicology of Complex Mixtures.
New York: Plenum Press.
243. Spitzer, W.O.; Lawrence, V.; Dales, R.; Hill, G.; Archer, M.C.; Clark, P.; Abehnaim, L.; Hardy,
J.; Sampalis, J.; Pinfold, S.P.; Morgan, P.P. (1990) Links Between Passive Smoking and
Disease: A Best-Evidence Synthesis. Clinical and Investigative Medicine 13(1): 17-42.
244. Standaert, F.G. (1988) Absorption and Distribution of Xenobiotics. Environmental Health
Perspectives 76:63-71.
245. Starr, T.; Buck, R. (1984) The Importance of Delivered Dose in Estimating Low-dose Cancer
Risk from Inhalation Exposure to Formaldehyde. Fundamental and Applied Toxicology
4:740.
246. Starr, T.; Gibson, J.; Swenberg, J. (1985) An Integrated Approach to the Study of
Formaldehyde Carcinogenicity in Rats and Mice. Toxicological Risk Assessment, Vol II.
Florida: CRC Press, pp. 195-209.
247. Starr, T.B. (1990) Quantitative Cancer Risk Estimation for Formaldehyde. Risk Analysis
10(1):85-91.
248. Stowers, S.J.; Maronpot, R.R.; Reynolds, S.H.; Anderson, M.W. (1987) The Role of Oncogenes
in Chemical Carcinogenesis. Environmental Health, Perspectives 75:81-86.
249. Strom, S.C. (1987) Issues in Biochemical Applications to Risk Assessment: Can In Vitro Studies
Assist Us in Species Extrapplation? Environmental Health Perspectives 76:181-184.
250. Swenberg, J.; Gross, E.; Martin, J., Popp, J. (1983) Mechanisms of Formaldehyde Toxicity.
Formaldehyde Toxicity. Washington, D.C.: Hemisphere Publishers, pp. 132-137.
212
-------
REFERENCES (PART II) (continued)
251. Swenberg, J.A.; Richardson, F.C.; Boucheron, J.A.; Deal, F.H.; Belinsky, S.A.; Charbonneau, M.;
Short, B.G. (1987) High- to Low-Dose Extrapolation: Critical Determinants Involved in
the Dose Response of Carcinogenic Substances. Environmental Health Perspectives 76:57-
63.
252. Swenberg, J.A.; Short, B.G. (1987) Influence of Cytoxicity on the Induction of Tumors.
Nongenotoxic Mechanisms in Carcinogenesis: Banbury Report 25:151-161.
253. Tamplin, A.; Cochran, T. (1974) A Report on the Inadequacy of Existing Radiation Protection
Standards Related to Internal Exposure of Man to Insoluble Particles of Plutonium and
Other Alpha Emitting Hot Particles, Natural Defense Council, Washington, D.C.
254. Tennant, R.W. (1987) Issues in Biochemical Applications to Risk Assessment: Are Short-Term
Tests Predictive of In Vivo Tumorigenicity? Environmental Health Perspectives 76:163-
167.
255. Tennant, R.W.; Ashby, J. (1991) Classification According to Chemical Structure, Mutagenicity to
Salmonella and Level of Carcinogenicity of a Further 39 Chemicals Tested for
Carcinogenicity by the US NTP. Mutation Research 257:209-227.
256. Tennant, R.W.; Margolin, B.H.; Shelby, M.D.; Zeiger, E.; Haseman, J.K.; Spalding, J.; Caspary,
W.; Resnick, M.; Stasiewicz, S.; Anderson, B., Minor, R. (1987a) Prediction of Chemical
Carcinogenicity in Rodents from in Vitro Genetic Toxicity Assays. Science 236:933-941.
257. Tennant, R.W.; Spalding, J.W.; Stasiewicz, S.; Caspary, W.D.; Mason, J.M.; Resnick, M.A.
(1987b) Comparative Evaluation of Genetic Toxicity Patterns of Carcinogens and
Noncarcinogens: Strategies for Predictive Use of Short-Term Assays. Environmental
Health Perspectives 75:87-95.
258. Thorslund, T. (1990) Alternative Approaches for Estimating the Yearly Number of Lung
Cancer Deaths in Non-Smokers due to ETS Based on Dose-Response Modeling. Health
Effects of Passive Smoking: Assessment of Lung Cancer in Adulst and Respiratory Disorders
in Children. US EPA, EPA/600/6-90/006A, Appendix D.
259. Toulmin, S. (1958) The Uses of Argument Analysis. Cambridge: Cambridge University Press.
260. Trosko, J.E.; Chang, C. (1988) Nongenotoxic Mechanisms in Carcinogenesis: Role of Inhibited
Intercellular Communication. Carcinogen Risk Assessment: New Directions in the
Qualitative and Quantitative Aspects: Banbury Report 31:139-159.
261. United States Department of Health and Human Services (1984) Report of the NTP Ad Hoc
Panel on Chemical Carcinogenesis Testing and Evaluation.
262. United States Environmental Protection Agency. (1983a) Good Laboratory Practices Standards-
-Toxicology Testing. Federal Register 48:53922.
213
-------
REFERENCES (PART II) (continued)
263. United States Environmental Protection Agency. (1983b) Hazard Evaluations: Humans and
Domestic Animals. Subdivision F. Available from: NTIS, Springfield, VA. PB 83-153916.
264. United States Environmental Protection Agency. (1983c) Health Effects Test Guidelines.
Available from: NTIS, Springfield, VA. PB 83-232984.
265. United States Environmental Protection Agency (1984) Formaldehyde: Determination of
Significant Risk. Federal Register 49:21874.
266. United States Environmental Protection Agency. (1986) Guidelines for Estimating Exposures.
Federal Register 51(185):34042-34054.
267. United States Environmental Protection Agency. (1987) The Risk Assessment Guidelines of
1986. EPA/600/8-87/045.
268. United States Environmental Protection Agency. (1989) Biological Data for Pharmacokinetic
Modeling and Risk Assessment. Report of a Workshop Convened by the U.S. EPA and
ILSI Risk Science Institute. EPA/600/3-90/019.
269. United States Envrionmental Protection Agency (1990) Health Effects of Passive Smoking:
Assessment of Lung Cancer in Adults and Respiratory Disorders in Children.
USEPA/600/6-90/006A.
270. United States Food and Drug Admnistration (U.S. FDA) (1982) Toxicological Principles for
the Safety Assessment of Direct Food Additives and Color Additives Used in Food.
Available from: Bureau of Foods, U.S. FDA.
271. Vainio, H.; Hietanen, E. (1980) Role of Extrahepatic Metabolism in Drug Disposition and
Toxicity. Concepts in Drug Metabolism. New York: Marcel Dekker Inc., pp. 251-284.
272. VanGinnekan, C.; Russel, F. (1989) Saturable Pharmacokinetics in the Renal Excretion of
Drugs. Clinical Pharmacokinetics 16:38.
273. Vessell, E.; Penno, M. (1983) Intraindividual and Interindividual Variations. Biological Basis of
Detoxification. New York: Academic Press.
274. Wahrendorf, J. (1990) Statistical Aspects of Experimental Studies with Complex Mixtures.
Complex Mixtures and Cancer Risk, IARC 1990.
275. Wald, N.J.; Doll, R. (eds.) (1985) Interpretation of Negative Epidemiological Evidence for
Carcinogenicity.
276. Weisburger, J. (1988) Cancer Risk Assessment Strategies Based on Mechanisms of Action.
Journal of the American College of Toxicology 7:417-425.
214
-------
REFERENCES (PART II) (continued)
277. Weiss, N. (1981) Inferring Causal Relationships: Elaboration of the Criterion of "Dose-
Response". American Journal of Epidemiology 113:487-490.
278. Wheatley, D. (1968) Enhancement and Inhibition of the Induction by 7, 12-
dimethylbenz(a)anthracene of Mammary Tumors in Female Sprague-Dawley Rats.
British Journal of Cancer 22:787.
279. Whitey, J. (1982) Toxicodynamics and Biotransformation. Assessment of Multichemical
Contamination. Washington, D.C.: National Academy Press.
280. Whitey, J. (1985) Chapter 2: Pharmacokinetic Differences Between Species. Toxicological Risk
Assessment Volume I: Biological and Statistical Criteria. Florida: CRC Press, Inc., pp. 41-
52.
281. Wilbourn, J.; Haroun, L.; Heseltine, E.; Kaldor, J.; Partensky, C.; Vainio, H. (1986) Response of
Experimental Animals to Human Carcinogens: An Analysis Based upon the IARC
Monographs Programme. Carcinogenesis 7(11): 1853-1863.
282. Williams, G.M. (1984) Modulation of Chemical Carcinogenesis by Xenobiotics. Fundamental
and Applied Toxicology 4:325-344.
283. Williams, G.M. (1987) The Significance of Preneoplastic Liver Lesions in Experimental Animals.
Advances in Veterinary Science and Comparative Medicine 31:21-44.
284. Williams, G.M. (1990) Epigenetic Mechanisms of Liver Tumor Promotion. Mouse Liver
Carcinogenesis: Mechanisms and Species Comparisons, pp. 131-145.
285. Williams, G.M.; Weisburger, J.H. (1991) Chapter 5: Chemical Carcinogenesis. Toxicology.
New York: Pergamon Press, pp. 127-200.
286. Williams, R. (1974) Interspecies Variations in the Metabolism of Xenobiotics. Biochemical
Society Transactions 2:359.
287. Withey, J.; Collins, B.; Collins, P. (1983) Effect of Vehicle on the Pharmacokinetics and
Uptake of Four Halogenated Hydrocarbons form the Gastronintestinal Tract of the Rat.
Journal of Applied Toxicology 3:249-253.
288. Wogan, G.; Gorelick, N. (1985) Chemical and Biochemical Dosimetry of Exposure to Genotoxic
Chemicals. Environmental Health Perspectives 62:5.
289. Woodruff, M. (1990a) Chapter 1: Diversity of Cells in Tumours. Cellular Variation and
Adaptation in Cancer. New York: Oxford University Press, pp. 1-7.
290. Woodruff, M. (1990b) Chapter 2: Transformation, Carcinogenesis, and Progression. Cellular
Variation and Adaptation in Cancer. Oxford: University Press, pp. 8-27.
215
-------
REFERENCES (PART II) (continued)
291. Woutersen, R.; Appleman, L.; Van Garderen-Hoetmer, A.; Feron, V. (1986) Inhalation
Toxicity of Acetaldehyde in Rats. III. Carcinogenicity Study. Toxicology 41:213-231.
292. Zeiger, E.; Haseman, J.K.; Shelby, M.D.; Margolin, B.H.; Tennant, R.W. (1990) Evaluation of
Four In Vitro Genetic Toxicity Tests for Predicting Rodent Carcinogenicity:
Confirmation of Earlier Results With 41 Additional Chemicals. Environmental and
Molecular Mutagenesis 16( 18): 1-14.
216
-------
TECHNICAL REPORT DATA
(Please read Instructions on the reverse before completing)
1. REPORT NO.
EPA/600/R-94/204a
2.
3. RECIPIENT'S ACCESSION NO.
1 • 1
i
4. TITLE AND SUBTITLE
Hazard Identification in Carcinogen Risk Analysis:
An Integrative Approach
5. REPORT DATE
October 1992
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
8. PERFORMING ORGANIZATION REPORT NO
Douglas J. Crawford-Brown,
Kenneth G. Brown
OHEA-C-554
NCEA-W-013
9. PERFORMING ORGANIZATION NAME AND ADDRESS
Kenneth G. Brown, Ph.D. Inc.
10. PROGRAM ELEMENT NO.
P.O. Box 16608
Chapel Hill, NC 27516-6608
11. CONTRACT/GRANT NO. :
68-C9-0009; W.A. S-l-56
12. SPONSORING AGENCY NAME AND ADD&ESS
National Center for Environmental Assessment*
13. TYPE OF REPORT AND PERIOD COVERED
i ' i
Washington Laboratory
U.S. EPA
Washington, DC 20460
14. SPONSORING AGENCY CODE
EPA/600/021
15. SUPPLEMENTARY NOTES
This project report is part of the U.S. EPA's program, Research to Improve Health j
Risk Assessment (RIHRA). ^Formerly Ofc. of Health & Environmental Assessment j
16. ABSTRACT
¦ ,
The process of identifying carcinogenic agents, as the first stage in risk assessment, is examined
from the point of view of rationality in a two-part report. The report is designed to aid the risk assessor in
making rational judgments and to furnish a basis for discussion of these judgments with others, rather than
attempting to formulate decision rules to avoid human decisions and avoid conflicts. The first part
examines what is meant by the rationality of statements claiming that an agent is carcinogenic and
enumerates several strategies by which one can claim that observations made in the laboratory or in human
studies are relevant to human carcinogenicity. The second part of the report describes a seven-step
framework that can be followed by an analyst in formulating judgments about hazard identification. This
framework is based on the principles of rationality, coherence, and completeness discussed in the first part.
Using this framework should accomplish three things: 1) aid in the systematic evaluation of all sources of
information; 2) identify sources of divergence resulting from different perspectives and opinions of
individuals; 3) help identify areas for research and help assess the potential impact of research on the hazard
identification process.
17.
KEY WORDS AND DOCUMENT ANALYSIS
a. descriptors
b. IDENTIFIERS/OPEN ENDED TERMS
c. COS ATI Field/Gioup
18. distribution statement
Release to the Publie
19. SECURITY CLASS (This Report)
Unclassified
21. NO. OF PAGES
£ 2S
20. SECURITY CLASS (This page)
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77) previous
EDITION IS OBSOLETE
------- |