United States     Science Advisory Board    EPA-SAB-EHC-97-010
       Environmental     Washington DC       September 1997
       Protection Agency
&EPA  AN SAB REPORT: GUIDE-
       LINES FOR CANCER RISK
       ASSESSMENT
       REVIEW OF THE OFFICE OF
       RESEARCH AND DEVELOPMENT'S
       DRAFT GUIDELINES FOR CANCER
       RISK ASSESSMENT BY THE
      ENVIRONMENTAL HEALTH
      COMMITTEE

-------
                              September 30, 1997
EPA-SAB-EHC-97-010
Honorable Carol M. Browner
Administrator
U.S. Environmental Protection Agency
401 M Street, SW
Washington, DC 20460

      Subject:     Science Advisory Board's review of the revised Guidelines for
                  Cancer Risk Assessment

Dear Ms. Browner:

      In 1983, the National Academy of Science/National Research Council (MAS/
NRC) published a report entitled Risk Assessment in the Federal Government:
Managing the Process.  In that report, the NRC recommended that Federal regulatory
agencies establish "inference guidelines" to insure consistency and technical quality in
risk assessments, and to ensure that the risk assessment process was maintained as a
scientific process separate from risk management. The Agency first issued cancer
guidelines in 1986.  In April 1996, the Agency released for public comment new
proposed Guidelines, incorporating significant changes in the approach assessing risk.
Following public comment, the Science Advisory Board's (SAB) Environmental Health
Committee (EHC) was asked to review the proposed revisions, and subsequently met
on February 13-14, 1997, in Washington, DC.

      In terms of a "global overview." the Committee regards the proposed Guidelines
as constituting a significant step forward in the "state-of-the-art" for carcinogen risk
assessment. The EHC particularly commends the Agency for addressing the
controversial aspects of the new Guidelines with a frank,  unbiased approach to all
points of view. The new Guidelines, when compared with the 1986 version, will cause
risk assessors to  place greater emphasis (than do the current Guidelines)  on the
utilization of all the available scientific information in characterizing cancer risks.

-------
      The above comments not withstanding, the EHC noted several areas in which
the proposed Guidelines would benefit from clarification or revision. These areas are:

      a)    The Committee endorses the Guidelines' emphasis on the use of
            narrative discussion describing the weight of evidence, but found
            problems the proposed categorization, particularly in the use of multiple
            terms (e.g., categories/descriptors/subdescriptors). Given the
            complexities involved,  the Committee could not come to a consensus as
            to how this problem should be addressed.

      b)    The majority of the Committee supported the Agency's handling of the
            issues raised when departure from the defaults are contemplated. A
            sizeable  minority of the Committee believed that the burden of proof
            should rest on showing that the defaults are implausible, and also that the
            Guidelines should be revised to include explicit and specific criteria for
            judging the validity of hypotheses invoked to depart from defaults.

      c)    The proposed Guidelines state that: the new default procedures are
            "public health conservative," but that they have been revised, when
            appropriate, to reflect the changes in the state-of-the-art since 1986. The
            EHC clearly understands that the primary goal of EPA actions is public
            health protection and recommends that clarification be added to the final
            Guidelines to alleviate potential concerns by readers or  users of the
            Guidelines, while at the same time promoting the use of good science for
            decision  making.

      d)    The EHC generally endorsed the  Guidelines' mode of action proposals,
            but suggested that the Guidelines contain specific criteria for judging that
            the data on mode of action are valid and adequate. This should include a
            discussion of mode of action  that  reflects the lack of a clear distinction
            between  direct DMA damage and  other mechanisms with respect to the
            low dose response relationship.

      e)    The proposed Guidelines provide for more flexibility in risk assessment
            and provide means for incorporating other types of biological data and
            information on mode of action into dose response assessment. However,
            some EHC Members noted that, in some areas, there may not be enough
            guidance, given the wide usage of the Guidelines, and in other areas
            additional clarification is  needed.  Although sympathetic to the Agency's

-------
      reluctance to develop additional criteria, the EHC believes that further
      guidance should be provided.

f)     The EHC understands the theoretical desirability of the proposed
      biologically based dose response model.  However, the proposed
      Guidelines may be misleading in this respect because no such model
      presently exists in a complete form, nor is it clear how to develop such in
      the foreseeable future.  The Agency's definition of a biologically based
      model thus seems unnecessarily narrow.  The Agency should define more
      clearly what is meant by a biologically based dose response model and to
      give guidance as to when such a model is preferred over the default
      linear or non-linear approaches. The enclosed report, in section 3.3.2,
      proposes a possible definition for the Agency's consideration.

g)    To mitigate against unnecessary inconsistencies and confusion in  the
      application of the Guidelines to determine the point of departure, the EHC
      suggests that further guidance be given in the Guidelines on this
      approach.

h)    There is no explicit statement in the proposal that statistical significance
      should be a  basic requirement for determining causality. This lack of an
      explicit statement has been interpreted as misleading and implying that
      there  is a hidden intent to eliminate statistical significance as a
      consideration in assessing causality.  Adding appropriate and specific
      language concerning statistical significance should rectify this problem.

i)     The developing infant and child should be recognized as a population
      subgroup that is particularly sensitive to the carcinogenicity of a number
      of agents. Risk assessments should explicitly account for the differential
      susceptibility of the young.  Other differences in susceptibility should also
      be taken into account when data permit.

j)     Vis-a-vis assessing tumors, the Committee endorses the procedure
      adopted by the National  Toxicology program of combining some closely
      related tumor types for statistical analyses, but otherwise conducting
      separate tests for different tumor types. The EHC also endorses the
      Agency's Guidelines that give additional significance to the finding of
      uncommon tumor types,  multiple sites, tumors by more than one route of
      administration, etc. (see pp. 55-56 of the Guidelines).  Lastly,  because of

-------
            data on the relationship of body weight and some tumor types, the EHC
            encourages the Agency to consider the role of diet in a given tumor
            response.  Consideration of this issue should be addressed when
            evaluating a given tumor data set.

      We appreciate having been given the opportunity to address these issues, and
look forward to receiving your response to our comments.

                              Sincerely,
                                   /signed/
                              Dr. Genevieve Matanoski, Chair
                              Science Advisory Board
                                   /signed/
                              Dr. Emil Pfitzer, Chair
                              Environmental Health Committee
Enclosure

-------
                                   NOTICE


      This report has been written as a part of the activities of the Science Advisory
Board, a public advisory group providing extramural scientific information and advice to
the Administrator and other officials of the Environmental Protection Agency.  The
Board is structured to provide balanced, expert assessment of scientific matters
relating to problems facing the Agency. This report has not been reviewed for approval
by the Agency and, therefore, the contents of this report do not necessarily represent
the views and policies of the Environmental Protection Agency, nor of other agencies in
the Executive Branch of the Federal government, nor does mention of trade names or
commercial products constitute a recommendation for use.

-------
                                   ERRATA
Copies of this report distributed prior to 11/5/97 contain an error on page ii, in
paragraph (h) of the Abstract.

The erroneous version stated:

h) there should  be an explicit statement that statistical significance should be a basic
requirement for determining causality.

The correct statement, incorporated in this document is:

g) there should  be an explicit statement in the Guidelines that statistical significance
should be considered in assessing causality.

-------
                                 ABSTRACT

      EPA first issued cancer guidelines in 1986. In April 1996, new draft Guidelines
were release. The Science Advisory Board's (SAB) Environmental Health Committee
(EHC) was asked to review the proposed revisions, and met on February 13-14, 1997,
in Washington, DC.

      The Committee found the proposed Guidelines to be significant step forward in
carcinogen risk assessment, and particularly commended the Agency for addressing
the controversial aspects of the new Guidelines with a frank, unbiased approach. The
new Guidelines will cause risk assessors to place greater emphasis on the utilization of
all the available scientific information in characterizing cancer risks. However, the also
Committee identified areas in which  improvement was possible.  These are:

      a) problems with the proposed categorization (The Committee could not come to
a consensus as to how this problem  should be addressed.)
      b) most Members supported the Agency's handling of the issues raised when
departure from the defaults are contemplated.  A sizeable minority of the Committee
believed that the burden of proof should rest on showing that the defaults are implausi-
ble.
      c) the Guidelines should be clarified to alleviate potential concerns by users
about the impacts of changes on the degree to which the document remains "public
health conservative."
      d) the EHC generally endorsed the Guidelines' mode of action proposals, but
suggested that the Guidelines contain specific criteria for judging that the data on mode
of action are valid and adequate.
      e) some Members noted that  there may not be enough guidance on incorporat-
ing other types of biological data and information on mode of action into dose response
assessment.
      f) the Agency's definition of a  biologically based model seems unnecessarily
narrow.
      g) the EHC suggests that further guidance be given in the Guidelines on
determine the point of departure.
      h) there should be an explicit  statement in the Guidelines that statistical signifi-
cance should be considered in assessing causality.
      I) The developing infant/child  should be recognized as a population subgroup
that is particularly sensitive to the carcinogenicity of a number of agents.
      j) The Committee endorses the procedure adopted by the National Toxicology
program of combining some closely  related tumor types for statistical analyses, but
otherwise conducting  separate tests for different tumor types.

KEYWORDS: cancer; risk assessment; guidelines; subpopuations; default values;
mode of action; cancer mechanisms.

-------
                U.S. ENVIRONMENTAL PROTECTION AGENCY
                        SCIENCE ADVISORY BOARD
                   ENVIRONMENTAL HEALTH COMMITTEE

                            February 13-14, 1997
CHAIR
Dr. Emil Pfitzer, Research Institute for Fragrance Materials, lnc.,Hackensack, NJ

MEMBERS
Dr. Charles Capen (SAP), Department of Veterinary Biosciences, The Ohio State
University, Columbus, OH

Dr. Adolfo Correa, The Johns Hopkins University, School of Hygiene and Public Health,
Baltimore, MD

Dr. Kenny S. Crump, ICF Kaiser, Ruston, LA

Dr. Ernest E. McConnell (SAP), Raleigh, NC

Dr. Frederica Perera, School of Public Health, Columbia University,  NYC, NY

Dr. Henry C. Pitot, McArdle Laboratory for Cancer Research, Madison, Wl

Dr. James A. Swenberg, University of North Carolina, Chapel Hill, NC

Dr. Mark J. Utell, University of Rochester Medical Center, Rochester, NY

Dr. Lauren Zeise, California Environmental Protection Agency, Berkeley, CA

CONSULTANT
Dr. Roger 0. McClellan, Chemical  Industry Institute of Toxicology, Research Triangle
Park, NC

SCIENCE ADVISORY BOARD STAFF
Mr. Samuel Rondberg, Designated Federal Official, US EPA, Science Advisory Board
(1400), 401 M Street, SW, Washington, D.C. 20460

Ms. Mary L. Winston, Staff Secretary US EPA, Science Advisory Board (1400) 401 M
Street, SW, Washington, DC 20460

-------
                         TABLE OF CONTENTS
1.  EXECUTIVE SUMMARY	  1

2 BACKGROUND AND CHARGE	  3
      2.1 Background 	  3
      2.2 Charge	  3

3.  DETAILED FINDINGS	  12
      3.1  Hazard Characterization: Descriptors/Narratives	  12
      3.2  Information Requirements to Depart from Defaults and to Use Mode of
           Action (MoA) Determinations 	  14
           3.2.1  Issues for departing from defaults	  15
           3.2.2 Precautionary views about risk assessment under the proposed
                 Guidelines  	  17
           3.2.3 Mode-of-action (MoA) Determinations  	  17
      3.3  Dose Response Assessment: Defining a Point of Departure	  18
           3.3.1  Dose-response Assessment	  18
           3.3.2 Biologically Based Dose-response Models	  20
           3.3.3 Defining  a point of departure	  22
      3.4  Margin of Exposure (MoE) Analysis	  24
      3.5  Use of Human Data  	  24
           3.5.1 Assessing adequacy of epidemiological studies	  25
           3.5.2 Criteria for Causality	  25
      3.6  NRC Recommendations for Tumor Data Analysis  	  27
           3.6.1  Other Considerations 	  28
      3.7  Susceptibility Factors for Human Variability	  29

4.  CONCLUSIONS  	  31

APPENDIX A  	  A -1

APPENDIX B  	  B -1
                                    IV

-------
                         1. EXECUTIVE SUMMARY

      The Environmental Health Committee (EHC) met on February 13-14, 1997, in
Washington, DC to review the EPA's Proposed Guidelines for Cancer Risk Assess-
ment. The Committee, in general, endorsed the proposal, and regard it as constituting
a significant step forward in carcinogen risk assessment. In particular, the new
Guidelines, when implemented, will cause risk assessors to place greater emphasis
(than do the current Guidelines) on the utilization of all the available scientific informa-
tion in characterizing cancer risks.

      As with all such reviews of a highly complex enterprise, the EHC identified areas
wherein clarifications and/or revisions are advised.  These areas, discussed in detail in
section 3 of this report, are outlined below:

      a)    The Committee endorses the Guidelines' emphasis on the use of narra-
            tive discussion describing the weight of evidence, but found problems with
            the resultant categorizations, particularly in the use of multiple terms such
            as categories/descriptors/subdescriptors). The Committee could not
            come to a consensus as to how this problem should be addressed.

      b)    The majority of the Committee supported the Agency's handling of the
            issues raised when departure from the defaults are contemplated.  A
            sizeable minority of the Committee believed that the burden of proof
            should rest on showing that the defaults are implausible,  and also that the
            Guidelines  should be revised to include  explicit and specific criteria for
            judging the validity of hypotheses invoked to depart from  defaults.

      c)    The proposed Guidelines state that: the  proposed default procedures are
            "public health conservative," but that they have been revised, when
            appropriate, to reflect the changes in the state-of-the-art since 1986.  The
            EHC clearly understands that the primary goal of EPA actions is public
            health protection and recommends that clarifications be included in the
            final Guidelines to alleviate potential concerns by readers or users of the
            Guidelines, while at the same time promoting the use of good science for
            decision making.

      d)    The EHC generally endorsed the  Guidelines' mode of action proposals,
            but suggested that the Guidelines contain specific criteria for judging that

-------
      the data on mode of action are valid and adequate. This should include a
      discussion of mode of action that reflects the lack of a clear distinction
      between direct DMA damage and other mechanisms with respect to the
      low dose response relationship.

e)    The proposed Guidelines provide for increased flexibility in risk assess-
      ment and provide means for incorporating other types of biological data
      and information on mode of action  into dose response assessment.
      Although sympathetic to the Agency's reluctance to develop additional
      criteria, the EHC believes that further guidance on determing the mode of
      action should be provided.

f)     The EHC understands the theoretical desirability of the proposed biologi-
      cally based dose response model.  However, the Agency's definition of a
      biologically based model seems to be unnecessarily narrow. The Agency
      should define more clearly what is  meant by a  biologically based dose
      response model and give guidance as to when such a model is preferred
      over the default linear or non-linear approaches.

g)    To mitigate against unnecessary inconsistencies and confusion in the
      application of the Guidelines to determine the point of departure, the EHC
      suggests that further guidance be given in the  Guidelines on this subject.

h)    There should be an explicit statement in the Guidelines that statistical
      significance should be considered  in assessing causality.

i)     The developing infant and child should be recognized as a subgroup
      which is particularly sensitive to the carcinogenicity of a number of
      agents. Risk assessments should  account explicitly for the differential
      susceptibility of the young.  Other differences in susceptibility should also
      be taken into account when data permit.

j)     Viis-a-vis assessing tumors, the Committee endorses the procedure
      adopted by the National Toxicology program of combining some closely
      related tumor types for statistical analyses, but otherwise conducting
      separate tests for different tumor types.  The EHC also endorses the
      Agency's Guidelines that give additional significance to the finding of
      uncommon tumor types, multiple sites, tumors  by more than one route of
      administration, etc. (see pp. 55-56  of the Guidelines). Lastly, because of

-------
data on the relationship of body weight and some tumor types, the EHC
encourages the Agency to consider the role of diet in a given tumor
response.

-------
                     2 BACKGROUND AND CHARGE

2.1 Background

      In 1983, the National Academy of Science/National Research Council
(NAS/NRC) published a report entitled Risk Assessment in the Federal Government:
Managing the Process. In that report, the NRC recommended that Federal regulatory
agencies establish "inference guidelines" to insure consistency and technical quality in
risk assessments, and to ensure that the risk assessment process was maintained as a
scientific process separate from risk management. The Agency first issued cancer
guidelines in 1986.  In April  1996, the Agency released for public comment new
proposed Guidelines, incorporating significant changes in the approach assessing risk.
Following public comment, the Science Advisory Board's (SAB) Environmental Health
Committee (EHC) was asked to review the proposed revisions.  The EHC subsequently
met on February 13-14, 1997, in Washington, DC.

2.2 Charge

      In addressing theCharge, the Committee took special note of the topics
highlighted during the public comment process.  In addition, under Clean Air Act
Amendments of 1990, section 112, the SAB is charged to review the responses the
Agency makes to recommendations of the National Academy report Science and
Judgment in Risk Assessment with respect to cancer guidelines issues. (The EPA
responses are in section 1.3 and Appendix B  of the proposed Guidelines).

      The detailed Charge  follows:

a)    Hazard Characterization: Descriptors and Narratives

      In its 1986 guidelines, the EPA adopted a classification system designating
agents in groups"A" through "E", Group A for  "known" human carcinogens through
Group E for agents with evidence of noncarcinogenicity. In its 1996 proposal, the
Agency has proposed to express its conclusions through a narrative with standard
descriptors.

      The proposal has discussed certain limitations in the Agency's 1986 carcinogen
classification system, and has sought to move from the  letter and number classification
to increase the information provided to risk managers.  To do so, the Agency has
proposed incorporating a limited number of descriptor phrases into a weight of

-------
evidence narrative summary of human, animal, and other key data. The proposal
includes categories of descriptors: known/likely, cannot be determined, and not likely,
with sub-descriptors.  Among other issues about descriptors, the proposed Guidelines
call specific attention  to the descriptor/sub-descriptor term "cannot-be-determined,
suggestive evidence," and seek input whether this should be made into a separate
descriptor category.

      Several public  comments received on the weight of evidence issues reveal
difficulty in getting past the category level to examine the use of the descriptors within
narratives.  For example, some commenters believed that "known/likely" was a single
descriptor and urged  separation of the terms in much the same way as the previous "A"
group was held to a higher standard than the "B" groups. Even though we used the
terms separately in example narratives, the  categories  seem to be paramount to some.
Since our intent is to discard simple categorization, we need to revisit the presentation.

      To clarify these points, the Risk Assessment Forum's technical panel is
considering revisions to reemphasize narrative descriptors, while de-emphasizing
categorization.

      The panel is considering a simple presentation of the descriptors without
categorizations, but with the explanation that the descriptors known, as if known, likely,
and not likely are used when the weight of evidence is  sufficient to support a conclusion
about human carcinogenic potential.  Other descriptors including suggestive,
conflicting, inadequate data, no data would be used when the weight of evidence is not
sufficient for this conclusion to be made.  Thus, we might present the "suggestive"
descriptor as "The weight of evidence is suggestive of  carcinogenicity, but  not sufficient
for a confident conclusion as to human carcinogenic potential."

      The original question of having a fourth, "suggestive" category may be moot if
we  delete the categories in favor of simply presenting the descriptors.  The technical
panel believes that some positive indications of carcinogenicity, in practice, fall  short of
making a weight of evidence to support a conclusion regarding human hazard.  Some
of the data bases we  would formerly have called "possible" human carcinogens (group
C) are of this kind. (Example:  Narrative #2 in Appendix A of the proposal) However,
some commenters are of the strong opinion that any indication of carcinogenicity, even
a very limited animal tumor response alone, should be  the basis of citing an agent as
potentially carcinogenic to humans. The panel would like to have the Committee's
views on these issues.

-------
b)    Information requirements necessary to depart from defaults

      The 1986 guidelines presented default assumptions to be used to continue
assessment when gaps in data or knowledge were encountered. These were
consistent with the 1985 Office of Science and Technology (OSTP) review of the
science. There was little discussion of data needs for departing from defaults. The
1996 proposal provides a framework of defaults, still consistent with the OSTP review.
The proposal  addresses the individual, major defaults and gives very general guidance
on their use and departing from them within the body of the guidance as well as in
response to NRC recommendations (section 1.3).

      Many respondents requested additional guidance on information requirements
considered by the Agency to be sufficient to reject default assumptions, particularly for
rejection of the linear default assumption in favor of a non-linear mode-of-action (MoA).
Concerns were expressed over:  the amount of information necessary to  reject the
linear default,  standards of proof  (plausible conservatism versus maximum use of
scientific information), administrative processes by which defaults could be over-ridden,
and the extent to which Agency internal peer review and determinations should be
supplemented by external peer review, either through the SAB or scientific meetings
and consensus.

      The proposal discusses the fact that different kinds of gap filling defaults are
inherently different in the amount of empirical data needed to replace them, e.g.,
moving from a scaling factor to toxicokinetic analysis of dose versus determining that
an animal tumor response is not relevant to humans. The proposal describes two
general criteria: 1) that the scientific principle underlying the use of the data  be
generally accepted and 2) that empirical data on the agent in question support the
conclusion made.

      The Committee is requested to provide views on issues of departing from
defaults generally, and on mode of action (MoA) conclusions in particular, see
below. Further criteria on the MoA point were discussed in a 1995 International  Life
Sciences Institute report titled "Low Dose Extrapolation of Cancer Risk".

Mode-of-action (MoA) determinations

      Public comments correctly focused  on the new (MoA) guidance as a critical
change from the Agency's 1986 policy. The MoA guidance allows for rejection of the
linear default  in favor of nonlinear or combined linear/nonlinear risk assessments. A

-------
variety of comments were elicited on the extent to which a detailed mechanistic
understanding of the chemical-specific cancer process is necessary in order to make a
mode-of-action determination. The spectrum of potential policy options reflected in
comments can be summarized as follows:

      1)    Detailed mechanistic knowledge: Comments favoring this option
            highlighted the need to have a detailed knowledge of all the steps to
            cancer formation before departing from the linear default assumption.
            Related to this is the concern that nonlinear increments to a background
            load, such as in multiple exposure situations, may increase risk in a linear
            fashion having exceeded the threshold level for that mechanistic pathway.

      2)    Weight-of-Evidence: This option allows for rejection of the linear default if
            there is confidence that the MoA is not through direct DMA damage,
            combined with sufficient evidence supporting a different, nonlinear, MoA.
            This is the preferred standard in the proposed Guidelines.

      3)    Primacy of DNA-reactivity. A number of comments favored the
            dichotomous approach that if a substance is DNA-reactive it should be
            presumed to follow a linear MoA , and if not DNA-reactive then the default
            should be to a nonlinear approach. This is a position commonly
            espoused by a number of European countries.

Related to this issue is the option to provide both a linear and nonlinear estimate of
risk, the situations in which such a dual assessment should occur, and how discrepant
estimates of risk should be treated. The proposed guideline supports use of both: 1)
when responses in different sites appear, with substantial supporting evidence, to have
different modes  of action,  2) in order to decouple consideration of the contribution of
two modes of action that operate at high versus low dose

c)    Dose response assessment: defining a point of departure

      The 1986 guidelines provided a basic rationale for linear dose response
assessment that considered the relationship to be a function of reaction of agents with
DMA in a stochastic  manner. While the guidelines said that other approaches might be
used if information supported them, the practice of linear assessment has been
predominant. The 1996 proposal bases dose response assessment on mode of action
information.  The process would  begin with analysis of observed data, followed by a
explicit decision about how to approach analysis of the range of extrapolation to low
doses, based on mode of action data.

-------
      The proposal would employ a "point of departure" to mark the beginning of
extrapolation. Moreover, the response data are proposed to include data on precursor
effects in appropriate cases,  in addition to, or in lieu of, tumor data.  To maintain
consistency with noncancer risk assessment, the proposal suggested using the lower
95% confidence limit on dose associated with 10% extra risk as the point of departure.
Comment was requested on alternatives, including using the central estimate, ED10

       As requested in the proposal, respondents focused on issues about determining
the point of departure, including:

Comments on the use of tumor data:

      1)    The choice of the point of departure, e.g., LED10, ED10,  ED01. The Agency
            proposed the LED10 point, and asked for comment on several alternatives.
            Respondents ventured a variety of preferences for lower limit and central
            estimates and for standard versus case by case choice of the response
            level 1.0% to 50%. In a recent peer consultation workshop on benchmark
            dose estimation for noncancer risk assessment, there seemed to be a
            general move toward support for using the central estimate.  This is
            consistent with a previous workshop report on a draft of the proposed
            cancer risk assessment guidelines.

      2)    The procedure by which this point of departure should be determined.  In
            the absence of biologically-based models, possible options include:
            choosing the 10% extra risk level, as proposed by the Agency,
            determining the lowest point in the data that can be statistically separated
            from the cancer rate variability in the control group, selecting the lowest
            significant data  result, or by using logistic regression models.

      3)    The data loss inherent in choosing a single point. Choosing a single point
            negates the  use of much of the additional bioassay or epidemiological
            data at higher exposures, particularly information on  the observed slope
            of  the cancer dose-response curve.

Comments on the use of precursor data:

      1)    Use of precursor data: The proposed Guidelines allow for three
            alternative uses of precursor data: 1) in lieu of tumor data if the precursor
            is found to be a better measure of risk, 2) to provide information about
            likely dose response behavior in the range of extrapolation, or 3) to link

                                       8

-------
            the dose response for the precursor with that for the tumor response.
            Respondents were universal in their concern that precursor data be
            clearly linked to the cancer of concern, and that adequate peer review be
            included in this process.  It also was noted by certain comments that
            application of the same MoE approach to precursor data as to tumor data
            would act to  reduce the margin below that resulting from use of tumor
            data.  This reduction in the margin could be problematic regarding risk
            presentation to the public and potential research efforts, a situation that
            would subtract from the potential gains from confirming nonlinearity
            through additional research.

d)    Margin of exposure (MoE) analysis

      The 1986 guidelines had no approach for nonlinear dose response assessment.
In the 1996 proposal, a margin-of-exposure (MoE) analysis would be used in such
circumstances.  This is because, without sufficient data to define model parameters,
there is not a way to choose a nonlinear model. If there were sufficient data then one
would have enough basis  for the preferred approach of using a biologically based or
case specific model.  Public comments revealed a number of concerns about MoE
analysis, as follows:

      1)    Support to risk managers in implementing the Guidelines. One concern is
            utility to risk managers.  For the MoE approach, the proposal attempted to
            strike  a balance between over-prescribing the risk managers' conclusions
            versus leaving them adrift with insufficient risk assessment guidance.
            Currently, the proposed Guidelines anticipate providing supporting
            analysis of factors for consideration by risk managers in determining an
            MoE that is adequate to protect public safety, but do not go so far as to
            prescribe what an adequate margin of exposure should be.  It is felt that
            the assessment can address the question of how risk is likely to decline
            with exposure, and issues of inter- and intra species variation in
            susceptibility, but the issue of adequacy of a margin is in the hands of the
            risk manager.

      2)     Some comments suggest that the risk assessment provide an opinion as
            to  "adequate margin of exposure" rather than a supporting analysis for the
            risk manager.

The Committee's view on this basic issue is requested.  Would addition of case
example for such analysis improve the guidance?

-------
e)    Human data

      In the 1986 guidelines, there was little discussion of the issues of evaluating
human studies.  In the 1996 proposal, more discussion of analysis has been included.
In 1986, criteria for "sufficient human data" as a basis for a "known" classification group
were given which said that a positive finding was not a result of bias, confounding, or
likely to be due to chance.  In the proposal, bias and confounding are addressed and
the  Bradford-Hill criteria for judging chance are incorporated as suggested in a
workshop  held by the EPA (1989).  For  dose response assessment, the guidance relies
on case-by-case judgment, providing no methodologies as data sets are so variable.
Two areas of comment stand out, among the others.

      1)    Dose response guidance.  The need for more guidance on dose response
            assessment was felt by some commenters, particularly with respect to the
            use of the range of observation/range of extrapolation approach and
            locating a point of departure.

      2)    Statistical significance. One commenter, followed by some others,
            expressed belief that by not retaining the 1986 list of criteria for
            "sufficient" human data as it was, in particular, the criterion that chance is
            unlikely to account for a result, the Agency has hidden an intent to
            eliminate statistical significance as a consideration.  The commenter
            apparently equates chance solely with statistical significance,  not with the
            several  Bradford-Hill criteria.

      Advice would be appreciated  on the issue whether the approaches for dose
response assessment of human data should be separated entirely from the  assessment
approaches for other  data, leaving human data to a case-by-case treatment.
Alternatively, should the two-step analysis of observed data and extrapolation be
retained for all data, with generous exceptions for human studies that are not amenable
to modeling in the observed range?

      On the point of statistical significance, since the proposal had no intention
of discarding statistical significance in judging human data, suggested
clarifications of the  existing discussions of the Bradford-Hill criteria (section
2.2.1.3.) and weighing evidence from human studies (section 2.6.1.) would be
welcome.

f)     NRC recommendations for tumor data analysis


                                      10

-------
      The 1986 guidelines discussed two methods for deciding how to measure
response.  One was to model a tumor response at a particular site. The other was to
count animals with tumor types that were individually statistically significant.  The
proposed Guidelines (section 3.2) provide a number of options for measuring response,
with the overall  aim of choosing a measure or measures of response that best
represent the biology.

      Several comments highlighted the proposal's failure to include among these
options the explicit NRC recommendation (NRC 1994, p. 241) that EPA should first
estimate cancer potencies for each tumor type and then sum tumor type-specific
potencies to reach an overall cancer potency estimate. The NRC-recommended
approach treats multiple tumor sites as independent events, whose effects should be
added separately. The Agency's proposed approach is otherwise in that, if summing
the total animal response is considered to be the best representation of response, the
approach would be to sum animals with cancer rather than their site-specific tumor
potencies. There are several factors to consider and a Committee perspective is
requested.

      1)     What is independence ? Independence of tumor expression at
            different sites in an animal as tested by probability? Independence
            of mode of action  for different tumor types? Which is the better
            basis for considering human risk potential?

      2)     Often one observes an increase in a high background tumor (e.g.
            mouse liver tumors) as well as an increase in an historically low
            background tumor (both statistically significant).  If the dose
            responses for these are summed, the latter will be swamped by the
            former.

      3)     What are the implications of different approaches with respect to
            assuming site concordance?

      4)     Whichever summing approach might be selected, ought it be a last
            resort, a standard approach, or be accompanied by separate
            analyses and discussions of different tumor types?

g)    Susceptibility Factor for  Human Variability

      The NRC (1994; p.219) report elaborates on human variability, recommending
that "EPA could choose to incorporate into its cancer risk assessment for individual risk

                                    11

-------
a 'default susceptibility factor' greater than the implicit factor of 1 that results from
treating all humans as identical." NRC (1994) also recommends continued research to
determine an appropriate default value for human variability.  The proposed Guidelines
do not incorporate a specific susceptibility factor in the linear extrapolation process.
The proposal's position is that sufficient conservatism is incorporated through the
assumption that humans have susceptibility similar to that of the of rodents in the
bioassay, or persons observed in epidemiological studies, combined with the inherent
conservatism of the linear extrapolation process. The Agency also notes that the
uncertainty resulting from human variability has been incorporated in the MoE
approach to nonlinear risk assessments. The NRC (1994) recommendation that
additional research be undertaken is being implemented; the Agency is funding
ongoing in house and grantee research on this issue. Several commenters suggest
that the guidelines incorporate an uncertainty factor across the board. What the
factor should be for linear extrapolation is unclear. A Committee perspective is
requested.
                                      12

-------
                          3. DETAILED FINDINGS

      Although there are far more positive, than negative, comments that could be
made about the Guidelines,  this report will be limited primarily to those areas (noted in
detail in the response to the  specific Charge issues below) wherein the Committee
believes that revisions are required before the proposed Guidelines are finalized; these
specific issues are discussed in detail below.

3.1  Hazard Characterization: Descriptors/Narratives

      In its 1986 Cancer Risk Assessment Guidelines, the Agency adopted a
carcinogen classification system assigning agents to groups "A" through "E" (Group A
for "known" human carcinogens through Group E for agents with evidence of non-
carcinogenicity). In its current (1996) draft revision, the Agency has proposed
assessing conclusions about an agent through the use of a narrative with standard
descriptors. The proposed Guidelines moved from the letter and number classification
with the declared intent of increasing the information provided to risk managers. This
revised classification system incorporates a  limited number of descriptor phrases into a
weight of evidence narrative summary of human, animal and other key data.  The
proposed Guidelines as published by the Agency provide three categories with 13
descriptors/subdescriptors as follows:

"known/likely"
      known, in humans from epidemiology and/or experimental evidence
            demonstrating  causality
      as if known, based on plausible causality and strong experimental evidence
      likely, due to tumors resulting from modes of action that are relevant to human
             carcinogenicity
                  likely, (high end of weight-of-evidence)
                  likely, (low end of weight-of-evidence)

"cannot be determined"
         cannot be determined, but suggestive
         cannot be determined, due to conflicting data
         cannot be determined, due to inadequate data
         cannot be determined, due to no data
                                      13

-------
"not likely"
         not likely, because of available animal studies
         not likely, because animal data shown to be not relevant to humans
         not likely, because of dose or route dependence
         not likely, because of extensive human experience.

      The EHC believes that the proposed Guidelines head in the right direction
by emphasizing the development of a narrative discussion describing the weight
of evidence and strongly endorses this approach. Users of the Guidelines should
be encouraged to support conclusions drawn by presenting in the narrative a full
discussion of the weight-of-evidence for carcinogenic hazard, including all critical
considerations made in reaching the conclusion.  In describing the weight of evidence,
the proposed Guidelines emphasize the importance of considering the relevance of the
route of exposure and the sufficiency of the evidence from animal studies, and the EHC
finds that this is appropriate and important.

      The use of the multiple terms, "categories/descriptors/subdescriptors," caused
occasional semantic confusion. The bottom-line is  that the EHC could not  come to
a consensus as to which, and how many, terms should be used. Some Members
found categories to be very important, and a few members indicated preference for the
categories used in the International Agency for Research on Cancer () classifications
that would harmonize the U.S. system with the rest of the world. It was noted  that
principles and procedures for IARC classifications include criteria for accepting
mechanistic data and the 'strength-of-evidence' approach (see Appendix A for an
extract of the relevant IARC document). Other Members felt that categories should be
eliminated, and that allowing the narrative to stand with selections made from  the 13
subdescriptors listed above provided the rich vocabulary that should be used to
describe what is known or not known with regard to carcinogenicity. Members
supporting this view found  that the IARC emphasis on hazard identification limited the
applicability of its criteria in US EPA risk assessments (see Appendix B for the IARC
Preamble statement). There was also support for the eight descriptors proposed by
Ashby, et al.,  which is basically an expanded IARC  classification.

      Although a majority of the Committee believes that some headings for the
narratives would be important, it was agreed that three is probably too few and
thirteen probably too many. If the Agency deems it necessary to use broad
categories to group subdescriptors, then the "known/likely" category should be
subdivided to prevent misunderstanding about the intent to use them as separate
terms, and 'suggestive' should not be  lumped with  'cannot be determined.' There was
                                      14

-------
also some expressed dislike for the term "likely" and for the possible inappropriateness
of the term "as if known." Another expressed concern was that the descriptor "not
likely, because animal data shown to be not relevant to humans" was illogical or
ambiguous unless reworded.  A possible change would be to make it similar to one of
the above descriptors, e.g.,"not likely, due to tumors resulting only from modes of
action that are not relevant to human carcinogenicity."

      The Committee recognizes that the ultimate choices of narrative descriptions will
not satisfy all users of the Guidelines, and hopes that the comments provided will help
EPA in making its choices.

3.2 Information Requirements to Depart from Defaults and to Use Mode of Action
(MoA) Determinations

      The 1986 Guidelines presented default assumptions to be used to continue
assessment when gaps  in data or knowledge were encountered. These were consistent
with the 1985 Office of Science and Technology (OSTP) review of the science.  There
was little discussion of data needs for departing from defaults. The 1996 proposal
provides a framework of defaults, still consistent with the OSTP review. The proposal
addresses the individual, major defaults and gives very general guidance on their use
and departures from them within the body of the guidance as well as in response to
NRC recommendations  (section 1.3).

      Many respondents to the proposed Guidelines requested additional  guidance on
information requirements considered by the Agency to be sufficient to reject default
assumptions, particularly for rejection of the linear default assumption in favor of a non-
linear mode-of-action (MoA).  The proposal discusses the fact that different kinds of
"gap filling" defaults are  inherently different in the amount of empirical data needed to
replace them, e.g.,  moving from a scaling factor to toxicokinetic analysis of dose versus
determining that an animal tumor response is not relevant to humans.

      The proposal describes two general criteria: a) that the scientific principle
underlying the use of the data be generally accepted; and b) that empirical data on the
agent in question support the conclusion made.

      Public comments to the proposed Guidelines correctly focused on the new
(MoA) guidance as a critical change from the Agency's 1986 policy. The MoA guidance
allows for rejection of the linear default in favor of nonlinear or combined
linear/nonlinear risk assessments. A variety of comments were elicited on  the extent to
                                      15

-------
which a detailed mechanistic understanding of the chemical-specific cancer process is
necessary in order to make a mode-of-action determination.

  3.2.1 Issues for departing from defaults

      The proposed Guidelines do not provide a list of formal decision criteria for
rejecting defaults.  Rather, they state that different amounts of data may be required for
different situations, i.e., screening assessments versus full assessments. The EHC
agreed that different types of assessments may have different data requirements
and levels of analysis, and that scientific peer review is important in evaluating
the adequacy of data. As stated in the proposed Guidelines:

      "If data are present, their evaluation may reveal inadequacies that also lead to
      use of the default.  If data support a plausible alternative to the default, but no
      more strongly than they support the default, both the default and its alternative
      are carried through the assessment and characterized for the risk manager.  If
      the data support an alternative to the default as the more reasonable judgment,
      the data are used."

      The majority of the Committee supported the Agency's approach  and
opined that it is likely that many of the new risk assessments will present both
linear and nonlinear assessments.

      A sizeable minority of the Committee believed that the burden of proof
should rest on showing that the defaults are implausible, and also that the
Guidelines should be revised to include explicit and specific criteria for judging
the validity of hypotheses invoked to  depart from defaults. The Members holding
this minority position also found  it ill-advised  to provide multiple assessments
and recommended as a matter of science policy that the Guidelines include the
following:

      a)    an explicit statement that, given the goal of public health  protection, where
           there is uncertainty regarding the mode of action of a chemical, the
           Agency will adopt the default assumptions that the agent in question is
           capable of acting linearly at low dose and that positive animal data are
           relevant to humans; and

      b)    an explicit statement that the burden of proof rests on showing the defaults
           (low dose linearity and relevance to humans) are not plausible (see

                                      16

-------
           section 3.4 of the  IARC document extract in Appendix A)  Some EHC
           Members felt that the statement should emphasize the point that the basis
           for departing from defaults was that the alternative was the most likely or
           most plausible choice, rather thandemonstrating that the default is not
           plausible; these Members felt that the proposed approach implied absolute
           knowledge or proof.

      Applying a weight-of-evidence approach to risk assessments may,  in some
instances, help reduce the uncertainty associated with the use of defaults.

      The new approach includes data on mode of action, toxicokinetics, cell
proliferation, species differences and interindividual differences in susceptibility. This
approach is consistent with the recommendations of the 1994 NRC report. The
EHC strongly endorses this change.  The proposed Guidelines clearly state that
the preferred method for developing the hazard characterization and risk
assessment for a chemical relies on a strong scientific database.  It is also
recognized that, for many chemicals, the assessor will not have a detailed
understanding of their mode of action and related material.  It is clear that regulatory
decisions must still be made in the absence of such data.  A linear default model will be
employed in such cases that will not differ greatly from the approach outlined in the
1986 Guidelines. The EHC agrees with the use of defaults in the absence of
scientific data and does not object to the simple extrapolation procedure outlined
in the proposed Guidelines.

      The proposed Guidelines indicate that the Agency  has carefully thought about
and decided against adopting formal criteria for judging when an analysis is sufficiently
plausible and persuasive to warrant deviation from defaults  (section 1.3.1 of the
proposed Guidelines).  Nonetheless two general criteria are given:

       "...that the underlying scientific principle has been generally accepted
      within the scientific community and that supportive experiments are
      available that test the application of the principle to the agent under
      review."

      These criteria are helpful, but given the wide use of the Guidelines by the
various programs within the Agency, more detailed guidance appears to be needed.
There were some expressions of dissatisfaction with the term "default." A suggested
alternative was to use "inference guidelines" (a term suggested in the National
Research Council "Red Book" in 1983).
                                      17

-------
  3.2.2 Precautionary views about risk assessment under the proposed
Guidelines

      The proposed Guidelines provide thoughtful background discussion on the
variety of complex issues on which policy positions are taken. The rationale provided
is well reasoned and transparent.  One area of discussion that deserves more attention
is in the characterization of the degree to which individual policy positions/defaults and
the Guidelines as a whole can be characterized as public health conservative.

      The proposed Guidelines state that:

            " Generally these defaults remain public health conservative, but in some
            instances they have been modified to reflect the evolution of knowledge
            since 1986."

      This statement leaves one with the general impression that those defaults that
have not changed are conservative and those that have changed now have greater
scientific justification. The general message appears to be that the overall process is
still public health conservative.  Because this overall impression is likely to be
translated to work products developed under these Guidelines and will thus
affect the public's and risk manager's view of the risk, it deserves greater
scrutiny. During the EHC meeting it was stated that it is possible that the
process does not always result in evaluations that are public health conservative
and examples of default as well as non-default procedures that might not be
public health conservative were raised. The EHC clearly understands that the
primary goal of EPA actions is public health  protection and recommends adding
clarifications that will  alleviate potential concerns by readers or users of the
Guidelines, while at the same time promoting the use of good science for
decision making.

  3.2.3 Mode-of-action (MoA) Determinations

      At the EHC public review meeting, there was considerable discussion regarding
MoA determinations. Most Members of the Committee agreed with the approach
outlined in the proposed Guidelines, a finding that absolute proof of a MoA is
neither attainable nor necessary.  Rather, an alternative should displace a default
when it is generally accepted  in peer review as the most reasonable judgment.
The EHC recommended that the Guidelines include the following:

                                     18

-------
      a)    specific criteria for judging that the data on mode of action are valid and
            adequate to support the alternative approach (see the IARC document
            exract in Appendix A, pages A-3 and A-4)

      b)    a discussion of mode of action that reflects the lack of a clear distinction
            between direct DMA damage and other mechanisms with respect to the
            low dose response relationship.  Most Members  felt that the Guidelines
            should include some specific examples of chemicals that are accepted as
            threshold carcinogens.

3.3 Dose Response Assessment: Defining a Point of Departure

      The 1986 Guidelines provided a basic rationale for linear dose response
assessment. This rationale considered the relationship to be a function of reaction of
agents with DMA in a stochastic manner.  While the Guidelines said that other
approaches might be used if information supported them, the practice of linear
assessment has been predominant.  The 1996 proposal bases dose response
assessment on mode  of action information. The process would begin with analysis of
observed data, followed by a explicit decision about how to  approach analysis of the
range of extrapolation to low doses, based on mode of action data.

      The proposal would employ a "point of departure" to  mark the beginning of
extrapolation. Moreover, the response data are proposed to include data on precursor
effects in appropriate cases, in addition to, or in lieu of, tumor  data.  To maintain
consistency with noncancer risk assessment, the proposal suggested using the lower
95% confidence limit on dose associated with 10% extra risk as the point of departure.

  3.3.1  Dose-response Assessment

      The Agency is to be commended for its efforts to develop a clear, well-reasoned
framework for dose response assessment.  The proposed Guidelines provide for more
flexibility in risk assessment and provide means for incorporating other types of
biological data and information on mode of action into dose  response assessment
However, some Members noted that, in some areas, there may not be enough
guidance, given the wide usage of the Guidelines, and in other areas additional
clarification is needed

      Although sympathetic to the Agency's reluctance to develop additional
criteria, the EHC believes that further guidance should be  provided. In the case of
                                      19

-------
dose response evaluation, the Agency is encouraged to develop criteria which lead to
judgments which consider the following: the reliability of predictions outside the
observable range; problematic data sets such as those with poor fits and extreme
curvature; the comprehensiveness of the supporting data; the degree to which
alternative hypotheses have been considered and investigated; and the adequacy of
the scientific peer review, considering such issues as the breadth of the review and the
degree to which the peer review group can be considered to represent the scientific
community at large.  To aid its deliberation, the Agency is referred to the 1996 NRC
report Understanding Risk and the 1994 NRC report Science and Judgment in Risk
Assessment.

      Unlike other parts of the Guidelines, no examples are provided in the dose
response section. However, a number of the approaches proposed in the dose
response section are fairly novel; consequently we have little experience on how they
will work in practice.  The Agency is encouraged to provide examples in the dose
response section to illustrate important features of the dose response methodology,
including:

      a)    the decision to use  a biologically based model for quantitative risk
            assessment

      b)    the decision to assume that a chemical has a (low-dose) non-linear mode
            of action

      c)    extrapolation to low doses using the linear mode  of action approach

      d)    calculation of an LED10 and ED10

      e)    use of precursor data - analyses based on non-cancer responses

      Generally a number of different models will provide an equivalent fit to the data.
However, some unrestricted models may give answers that are biologically unrealistic
in some cases and should be avoided. Since the issues regarding selection of a
model for use in the observable range are largely generic, the EHC recommends
that the Agency select a default procedure for routine use  in such calculations. A
requirement of biological feasibility leads to a dose response model that, at low doses,
is linear or sublinear, and does not assume an extreme superlinear shape (e.g., one
that predicts that the tumor response jumps abruptly from the background value to a
much higher value at an infinitesimally small dose).  It also calls for a model that is
monotonically increasing in dose. Appropriately constrained versions of the

                                      20

-------
polynomial, Weibull and logistic models all incorporate these features and have been
routinely used in the past for curve fitting. Consequently they are recommended to the
Agency for consideration.

      In cases where it is important to take into account early mortality in treated
animals, time dependent models should be considered. Additional guidance should be
provided to address cases where the dose response appears superlinear. In such
cases the analyst should be encouraged to explore the basis for the non-linearity (e.g.,
saturating  pharmacokinetic processes, differences in intercurrent mortality among
treatment groups) and if possible take that into account in the modeling exercise.
Finally under the Weibull model, unrealistic values of effect doses and confidence
bounds can result in cases of extreme sublinearity in which the shape parameter is
large and precautionary guidance should be given in this regard if the Weibull is
provided as a default model. The use of the unrestricted model can also lead to
unrealistic values when data are superlinear.

      With respect to the inferences made in performing dose response evaluations,
the assumption that all humans have the same sensitivity is the same as that made in
the 1986 Guidelines. As noted in the proposed Guidelines, human sensitivities  can
vary considerably. The Guidelines  also indicate that linear extrapolation is a public
health conservative assumption, but several examples suggest that in the absence of
information on pharmacokinetics and time to tumor the assumption may not be correct.
Other assumptions or procedures that are "public health neutral" or "not conservative"
include: dose averaging (see above); the treatment of data from standard two year
bioassay protocols as representative of full  life exposure (exposures before weaning
and in old age do not occur); the interspecies correction; presuming human tumor sites
are the same as animals in cases where PBPK analyses have been performed.

      The EHC supports the Agency's efforts, reflected in these and other
guidelines documents, to harmonize approaches to dose  response evaluations
for cancer and non-cancer endpoints, and encourages the Agency to do more.

  3.3.2 Biologically Based Dose-response Models

      Overall, the Guidelines provide reasonable approaches for the use of
biologically based models but are unclear in some critical areas, e.g., definitions of
such models, description of the "state of the art" in development of biologically based
dose response models, and characterization of the approach for applying biologically
based dose response models.  The document defines these models as ones whose
                                      21

-------
parameters are calculated independently of curve fitting of data. The EHC understands
the theoretical desirability of a biologically based dose response model and applauds
the encouragement to develop such approaches. However, the proposed Guidelines
may be misleading in this respect because no such model presently exists in a
complete form, nor is it clear how to develop such in the foreseeable future. The
Agency's definition of a biologically based model thus seems unnecessarily
narrow.  One can conceive of models in which non-tumor data are used to capture the
dose-response shape in the low dose range, and fitting the model to the tumor data is
appropriately used to scale the dose  response.  On the other hand, just because a
model developed independently of the tumor data predicts tumor responses at the
experimental doses is no indication that it will predict responses in the low dose range.
In fact, such a model would not necessarily even include a means for predicting low-
dose behavior.  The document also is not clear on what a "case-specific" model means,
and how such a model differs from a  biologically based model.

      The EHC believes that it is important for the Agency to define more  clearly
what is meant by a biologically based dose response  model and to give guidance
as to when such a model is preferred over the default linear or non-linear
approaches. Members of the Committee have developed the following definition for
the Agency's consideration:

      The dose response that relates the probability of tumor formation as a function of
      exposure to a xenobiotic is often expressed as the  composition of two processes:
      the toxicokinetic process that describes the concentration of the active agent at
      the target site as a function of exposure to the parent compound, and the
      toxicodynamic process that expresses the probability of tumor formation  as a
      function of the concentration of the active agent at  the target site.  A biologically
      based dose response model is one that uses either dose response data  for one
      or more intermediate steps in the toxicodynamic process or a biological
      understanding of the toxicodynamic process to develop a dose response model
      for predicting tumor response as a function of either exposure or concentration of
      the active agent in the target tissue.  Thus a biologically based dose response
      model of cancer incorporates biological information on the toxicodynamic
      process, and it may,  in addition, incorporate such information on the toxicokinetic
      process.

      If a definition such as this is adopted it appears that it may be unnecessary to
distinguish between a biologically based dose response model and a case-specific
model.
                                     22

-------
      The EHC also believes it would be helpful for the Agency to describe the "state
of the art" in development of biologically based dose response models. If possible, the
Agency should provide an example of an application of a biologically based dose
response model that would  be relied upon for low dose extrapolation.  If no such model
can presently be identified,  a statement to that effect would also be helpful.

      Given the general lack of experience in developing and applying biologically
based dose response models as well as the large amount of time and resources that
may be required in each instance, it seems unwise to hold up the use of such a model
as "the overriding preferred approach" for dose response evaluation (see section
1.3.2.5 of the  draft proposed Guidelines).  Instead, the Agency may wish to indicate
that the goal or ideal in cancer risk estimation is to predict risk reliably at levels of
human exposure based on  a clear understanding of the biology, even though this has
proven to be very difficult to achieve.  Nonetheless, when reliable predictions are
developed the Agency will use  them.

  3.3.3 Defining a point of departure

      The calculation of an LED10 for defining a  point of departure involves fitting a
statistical dose response model to data in the observable range.  This calculation
requires selecting a dose response model,  selecting a risk value (e.g., 10% or 1%), and
calculating statistical confidence intervals.  To mitigate against unnecessary
inconsistencies and confusion in the application of the Guidelines the EHC
suggests that further guidance be given in the Guidelines on this  approach. The
calculation of an LED10 is equivalent to the  benchmark calculation that is used by the
Agency with non-carcinogens.  The EHC sees little scientific justification for
separate approaches to LED10 and benchmark calculations, and therefore
recommends that the approaches used for non-carcinogens and (low dose) non-
linear carcinogens (e.g., choice of dose-response model, choice of point of
departure, and use of statistical confidence bounds) be harmonized to the extent
possible. There may be scientific justification for the use of different adjustment and
uncertainty factors for cancer and non-cancer. Non-cancer factors should not be
routinely  used for cancer without clear scientific justification for so doing.

      In  general, it is recommended that a single risk level, e.g., ED10, be selected
as the point of departure when the (low dose) non-linear approach is applied.
Selection of a common value will serve to standardize the calculation and facilitate
comparisons across chemicals. Use of different  values could complicate an MoE
evaluation since then a risk manager would somehow have to adjust for a  different

                                      23

-------
point of departure in different situations.  In some situations (e.g., large experiments), it
may be preferable to use a non-standard value. In those situations where it is
preferable to have a non-standard point of departure, the evaluation should provide
clear guidance to the risk manager regarding the evaluation of an MoE.

      The selection of the value of 10% extra risk proposed by the Agency as the
usual point of departure was supported by several members of the Committee. The
point of departure is supposed to represent the lower end of the range of risks that can
be measured with some reliability. A cancer bioassay typically involves 50 animals per
group.  Ten percent of 50 is five animals. Five responders out of 50 is only marginally
statistically different from zero responders out of 50 and with a higher background a
10% increase in extra risk would be even less significant.

      The consensus of the Committee was that both point estimates and
statistical bounds  can be useful in different circumstances, and recommended
that the Agency routinely calculate and present the point estimate of the ED10 and
the corresponding upper and lower 95% statistical bounds. It may be appropriate
to emphasize point  estimates in activities that involve ranking agents as to their
carcinogenic hazard.  On the other hand, it may be appropriate to emphasize lower
statistical bounds in activities designed to develop an appropriate human exposure
value, since such activities require accounting for various types of uncertainties and a
lower statistical bound on the ED10 (LED10)  is a scientifically-based approach for
accounting for uncertainty in the true value of the ED10.

      Other considerations impact this discussion. The special problem of our lack of
understanding of how to address early-in-life and late-in-life exposures should feature
more prominently in the section on dose averaging when appropriate.  Although the
approach of averaging cumulative dose over lifetime is used, biological models
presented over the  past 50 years as well as empirical studies (e.g., DEN, radiation,
saccharin) suggest  that age at  exposure is  important and more should be done to
address it. This is clearly an area where research is needed.

      Little guidance is given in the proposed Guidelines on how precursor data
should be used in dose response evaluation. As understood by the Committee, it was
considered useful for qualitative assessments, but there was no consensus on the
potential for quantitative assessment with precursor data. The EHC consensus was
that considerably more thought and work was needed in this area, with some
additional guidance andsome examples  provided in the document.
                                      24

-------
      Use of the terms "linear" and "non-linear" to distinguish the two procedures for
dose response evaluation can lead to some misunderstanding and confusion.  For
example, many well-studied agents associated with a DMA reactive mode of action are
observed to have non-linear dose response relationships (e.g., vinyl chloride and
diethylnitrosamine).  Non-linear tumor dose response relationships due to increased
mortality in high dose groups and dose dependent pharmacokinetics are frequently
observed.  Suggested alternative terminology to address this problem is "linear at
low doses" and "non-linear at low doses"; "low dose linear" and "low dose non-
linear"; or "non-threshold" and "presumed threshold-like."

3.4 Margin of Exposure (MoE) Analysis

      This term (MoE) seems to have evolved from the terms "margin of safety" and
"margin of error" where circumstances or exposures in everyday life  are compared to
those observed experimentally to have no effect. Regarding the use of the "margin of
exposure" term in the proposed Guidelines, an exposure estimate is  compared to a
dose associated with an undesirable outcome (i.e.,  excess 10% cancer incidence) and
not one for which lack of effect is presumed. Adoption of alternative terminology that
would leave less room for confusion and misinterpretation by a risk manager and the
public is recommended by the EHC.

      At times during the SAB public meeting there was confusion as to the proper
definition for MoE. Given the Committee's problems with the concept,  it was felt that
this topic needed the addition of case studies to illustrate its use by practitioners in the
field.  There was no consensus on the advice to be given to a risk assessor on the
magnitude of an appropriate MoE. Also, the EHC, as discussed with EPA staff in
attendance at the public review meeting, recommends that the Guidelines incorporate
additional information on how the Agency would use peer review in the process of
deciding on alternative approaches.
3.5 Use of Human Data

      In the 1986 Guidelines, there was little discussion of the issues of evaluating
human studies.  In the 1996 proposal, more discussion of analysis has been included.
In 1986, criteria for "sufficient human data" as a basis for a "known" classification group
were given which said that a positive finding was not a result of bias, confounding, or
likely to be due to chance.  In the proposed Guidelines, bias and confounding are
                                      25

-------
addressed and the Bradford Hill criteria for judging chance are incorporated as
suggested in a workshop held by the Agency (1989). For dose response assessment,
the guidance relies on case-by-case judgment, providing no methodologies, as data
sets are so variable.  The EHC endorses the approach and general concepts for
incorporating human data into risk assessments.

  3.5.1 Assessing adequacy of epidemiological studies

      The EHC finds that the section on epidemiological studies is both thoughtful and
cautious, given all the complexities of observational studies.  Observational studies,
such as cohort and case-control studies, represent the majority of studies on cancer
etiology in humans. In these studies, the investigator has no control on the assignment
of subjects to exposure groups, as  is the case with clinical trials.  Furthermore,
assessment of exposure to the agent of interest is subject to measurement error, and
the particular metric of exposure that may influence the risk of disease is often
unknown. Some of these methodological issues may result in spurious associations,
but others, particularly those relating to exposure, often result in a substantial
attenuation of the magnitude of a real association and a marked decrease in the
precision of the risk estimate. Therefore, in evaluating a particular association in an
observational study, several methodological issues, including possible selection bias,
confounding, and exposure misclassification need to be considered. It was noted that
it will often be important to have access to raw data for the best utilization of
human data for  risk assessment, and that the Agency should obtain and utilize
these data whenever possible.  To further this goal, the contractual agreement for all
studies receiving financial support from the Agency should be framed so that, within the
concerns of privacy for study participants, the raw data developed in the study will be
available to the Agency.

  3.5.2 Criteria for Causality

      There have been some criticisms (during the period for public comment) of the
manner in which  statistical significance has been treated in Sections 2.2.1.2 (Criteria
for Assessing Adequacy of Epidemiologic Studies, pp. 44-48), 2.2.1.3 (Criteria for
Causality, pp. 48-49), and 2.6.1 (Weighing Evidence from Human Studies, pp. 70-77)
of the proposed Guidelines for Carcinogen Risk Assessment.  These relate to the fact
that although the draft document makes reference to the need to consider statistical
significance in increasing confidence in a conclusion of causality (sections on
Statistical Considerations and Weighing the Evidence from Human Studies), there is no
explicit statement that statistical significance should be a basic requirement for
                                      26

-------
determining causality. The lack of such explicit wording has been interpreted as
misleading and implying that there is a hidden intent to eliminate statistical
significance as a consideration in  assessing causality. Adding appropriate and
specific language concerning statistical significance should rectify this problem.

      It is important, in evaluating the overall evidence for an effect from
epidemiological studies, to rule out chance as a possible explanation for an observed
association. The degree to which chance should account for an observed association
can be evaluated by calculating the statistical significance of the association.  In terms
of the Bradford  Hill criteria for causality, statistical significance is one means for
evaluating the strength of an association. Although the current version of the
Guidelines appears to appropriately  describe the role of statistical significance in
interpreting epidemiological studies,  the EHC believes, to avoid misinterpretation, the
section should be revised to reflect the need to consider the statistical significance of
an association.  However, in an observational study, factors other than chance may
affect the statistical significance of an association. These factors include selection,
confounding and exposure misclassification. In some instances these factors may
cause a spurious association to be statistically significant.  In other instances these
factors may cause a causal association not to achieve statistical significance. Also,  in
a data set in which numerous exploratory analyses may have been conducted,
interpretation of these tests should account for the multiplicity of tests.  Because of
these difficulties with the interpretation of observational studies, the EHC
believes that a weight of evidence  approach to evaluating causality is
appropriate.

      Although the Guidelines refer  to the Bradford Hill "criteria," the EHC believes
that the word "guidelines" may be more appropriate. A recent text on epidemiological
principles summarizes the position that most epidemiologists take with regard to causal
guidelines, and that the EHC supports:

      "Although causal guidelines are  often referred to as criteria, this term does  not
seem entirely appropriate. Although  it may be a desirable goal to place causal
inferences on a firm quantitative and structural foundation, at present we generally do
not have the information needed for doing so. The preceding list [i.e., Bradford Hill
guidelines for judging whether an association is causal] should therefore be considered
only guidelines which can be of most value when coupled with reasoned judgment in
making decisions about causation." (Epidemiology, Leon Gordis, W. B. Saunders,
1996: 181). Accordingly, the EHC recommends changing the word "criteria"  in
Sections 2.2.1.3 (Criteria for Causality), 2.2.1.4 (Assessment of Evidence of
                                       27

-------
Carcinogenicity from Human Data), and 2.6.1 (Weight of Evidence Analysis) to
"guidelines."

3.6 NRC Recommendations for Tumor Data Analysis

      Independence of multiple tumor sites may be viewed in terms of biological
independence or statistical independence. Independent biological action of multiple
tumor sites is supported by the known mechanisms of tumor causation (mode of
action).  It is  generally accepted that the molecular events (genetic damage)  that are
responsible for a given type of tumor cannot be assumed to be the same for other
tumor types,  even in the same animal and even though the offending chemical is the
same.  If they were the same, one would expect to see all tissue types qualitatively at
risk or not at risk in tandem.  This is not supportable by experimental evidence since a
given chemical usually causes tumors in only a few tissue sites.  Another reason for
treating different tumor types as biologically independent events is that the dose to one
tissue may be different from another based on its blood supply, fat content, etc.
Furthermore, the production of metabolites and the detoxification abilities varies from
tissue to tissue. As a result the mechanism of carcinogenesis in different tissues within
the same animals may differ, i.e., as a promoting agent in one tissue and  a complete
carcinogen in another.  On the other hand, statistical independence of tumor sites
means that the presence of a tumor in one organ does not affect the probability of there
being a tumor in a different organ. This is generally not  the case. Thus, tumors in
different sites are generally biologically independent, based on how the EHC is using
this term, but may be statistically dependent.

      Whether it is appropriate to combine different tumor types for analyses and, if
so, how, depends upon whether one is interested in conducting a statistical test for the
presence of a carcinogenic effect or in estimating the potency of a chemical that has
been identified as a carcinogen. In either case, combining all tumor types is  not
appropriate. In the former case, the procedure adopted by the National
Toxicology program of combining some closely related tumor types for statistical
analyses, but otherwise conducting separate tests for different tumor types is
endorsed by the EHC. In the latter case, estimating separate potencies for  different
tumor types and summing the potencies to estimate an overall carcinogenic potency
tacitly assumes that different tumors are statistically independent and therefore is not
recommended. An even less satisfactory approach would be to sum upper confidence
intervals on carcinogenic potency. Calculating a separate potency for each tumor type
is preferred to summing potencies.  However this approach may underestimate the
overall carcinogenic potency of a chemical that causes tumors at multiple sites.  To
                                     28

-------
account for this, the Agency should consider modeling the response in which an animal
is defined as a responder if it has one of the tumor types associated with exposure.
This approach does not require that tumors be either statistically or biologically
independent or dependent. A similar combining of responses was used recently by the
Agency in its risk assessment for methylmercury. Although the combined endpoints
were related to neurological deficit rather than cancer, the same principle applies to
both.  The Guidelines should be flexible enough to permit either of these two latter
approaches (calculating potencies for individual tumor types or defining an animal as a
responder if it has any tumor type associated with exposure) in individual cases, based
on what is dictated by the available data.

      As a general principle, chemicals that cause tumors at more than one site
and in multiple species, other things being equal, are most likely more hazardous
than those that produce tumors at a single site in only one species. This should
be reflected in cancer risk assessments by the Agency. The EHC endorses the
Agency's Guidelines that give additional significance to the finding of uncommon
tumor types, multiple sites, tumors by  more than one route of administration, etc.
(see pp. 55-56 of the Guidelines).

  3.6.1  Other Considerations

      The use of historical controls should be considered in the sense of
"reproducibility" of the bioassay.  To use data from only a single bioassay in isolation of
other control data, especially when the data are on controls conducted under the same
conditions in the same strain of animals in a reasonably similar time frame are
available, is not appropriate. Although concurrent controls should be given primary
weight in the evaluation of the results of a bioassay,  historical control data may also
provide valuable insight into whether the tumor response in concurrent controls was
atypical for that strain of animals.  However, before concluding, based on historical
control information, that the response in concurrent controls was atypical, one  should
carefully consider not only the average response in historical controls but also how
variable that response is from study to study and whether time trends are present, as
well as body weight differences between the two control groups. Other factors may also
need to be considered, such as whether animals were caged in the same way  in
historical controls as in concurrent controls.

      It is now well established that body weight is often correlated with tumor
response for several tumor types. Therefore, the EHC also encourages the
                                      29

-------
Agency to consider the role of diet in a given tumor response. Consideration of
this issue should be addressed when evaluating a given tumor data set.

3.7 Susceptibility Factors for Human Variability

      The proposed Guidelines do not incorporate a specific susceptibility factor in the
linear extrapolation process.  The proposal's position is that sufficient conservatism is
incorporated through the assumption that humans have susceptibility similar to that of
the rodents in the bioassay, or persons observed in epidemiological studies, combined
with the inherent conservatism of the linear extrapolation process.  The Agency also
notes that the uncertainty resulting from human variability has been incorporated in the
MoE approach to nonlinear risk assessments.

      There continues to be a  rapid growth in our knowledge of variability in both
susceptibility of humans to the development of cancer and the sensitivity of humans for
induction of cancer by exogenous agents.  As knowledge in the field increases, there
will be a better understanding of the distribution of susceptibility and sensitivity traits
across the population and for individuals within the population.  The introduction of a
"default susceptibility factor" must be closely linked with consideration of the "risk
target" for both individuals and  populations.

      The EHC recognized that human variabilities within populations are being
identified,  such as genetic polymorphisms, nutritional status, ethnicity, age,  stage of
development, gender,  preexisting diseases and ability for DMA repair. Each of these
and their potential interactions may be significant susceptibility factors, or even
potential resistance factors. Some of these factors have been estimated to  have an 85
to 500 fold variation within a human population. When the penetrance of such a factor
across a population  is  known, it should be taken into account. The developing infant
and child should be recognized as a subgroup which is particularly sensitive to
the carcinogenicity of a number agents. Risk assessments should explicitly
account for the differential susceptibility of the young.  Other differences in
susceptibility should also be taken into account when data permit.

      There was divided opinion with regard to the proposal that, when a specific
susceptibility factor is not known, an uncertainty factor should nevertheless  be used to
account for human variability during the extrapolation process. Some Members agreed
with the position in the proposed Guidelines that sufficient conservatism was already
built into the process and opined that the use of a conservative one in a million cancer
risk target for a population recognizes that there is a distribution of individual risks
                                      30

-------
within the population, including risks to individuals who may be more or less
susceptible than the population average. Several members questioned that sufficient
conservatism on human variability was present. The EHC finds that it is premature
to recommend a specific uncertainty factor for use during linear extrapolation.
The Agency should continue its research on the uncertainty of human variability
and the Guidelines should make the basis for conservatism in the process more
explicit with examples or case studies. The Agency should readdress the
question of an uncertainty factor for linear extrapolation at a later date.
                                     31

-------
                             4.  CONCLUSIONS
      The Committee's overall impression is that the proposed Guidelines
constitute a significant step forward in the "state-of-the-art" for carcinogen risk
assessment. The EHC particularly commends the Agency for addressing the
controversial aspects of the new Guidelines with a frank, unbiased approach to all
points of view.  The new Guidelines, when implemented, will cause risk assessors to
place greater emphasis (than do the current Guidelines) on the utilization of all the
available scientific information in characterizing cancer risks.

      The above comments not withstanding, the EHC  noted several areas in which
the Guidelines would benefit from clarification or revision. These areas are:

      a)          The Committee endorses the Guidelines' emphasis on narrative
                  descriptors, but found problems with its implementation,
                  particularly  in the use of multiple terms (i.e., categories;
                  descriptors, and subdescriptors). Given the complexities
                  involved, the Committee could not come to a consensus as to
                  how this problem should be addressed.  Some Members
                  suggested using the International Agency for Cancer Research
                  (IARC) classifications (see Appendix B); others advocated
                  eliminating categories in favor of a narrative with  selections made
                  from the proposed thirteen subdescriptors.  Still other Members
                  proposed use of the eight descriptors proposed by Ashby et al.,
                  which are related to the IARC scheme.

      b)          The majority of the Committee supported the Agency's handling of
                  the issues raised when departure from the defaults are
                  contemplated. A sizeable minority of the Committee believed that
                  the burden of proof should rest on showing that the defaults are
                  implausible, and also that the Guidelines should be revised to
                  include explicit and specific criteria  for judging the validity of
                  hypotheses invoked to  depart from defaults. TheMembers of this
                  minority group also found it ill-advised to provide multiple
                  assessments and recommended as a matter of science policy that
                  the Guidelines include  the following:
                                      32

-------
            1)    an explicit statement that, given the goal of public health
                  protection, where there is uncertainty regarding the mode of
                  action of a chemical, the Agency will adopt the default
                  assumptions that the agent in question is capable of acting
                  linearly at low dose and that positive animal data are
                  relevant to humans;  and

            2)    an explicit statement that the burden of proof rests on
                  showing the defaults (low dose linearity and relevance to
                  humans) are not plausible (see the IARC document extract
                  in Appendix A).  Some EHC Members felt that the
                  statement should emphasize the point that the basis for
                  departing from defaults was that the alternative was the
                  most likely or most plausible choice, rather than
                  demonstrating  that the default  is not plausible; these
                  Members felt that the proposed approach implied absolute
                  knowledge or proof.
c)           The proposed Guidelines state that: the proposed default
            procedures are "public health conservative," but that they have
            been revised, when appropriate, to reflect the changes in the state-
            of-the-art since 1986.   During the public review meeting, it was
            stated that, in some cases, the process could result in an
            assessment that is not specifically public health conservative;
            during the ensuing discussion, it was noted that both default as
            well as non-default procedures could yield results that might not be
            public health conservative. The  EHC clearly understands that the
            primary goal of EPA actions is public health protection and
            recommends adding clarifications that will alleviate potential
            concerns by readers or users of the Guidelines, while at the same
            time promoting the use of good science for decision making.

d)           The EHC generally endorsed the Guidelines' Mode of Action
            proposals, but suggested that the Guidelines contain specific
            criteria for judging that the data on mode of action are valid and
            adequate. This should include a discussion of  mode of action that
            reflects the lack of a clear distinction between direct DMA damage
            and other mechanisms with respect to the low dose response
            relationship. In fact, most Members felt that the Guidelines should

                                33

-------
            include some specific examples of chemicals that are accepted as
            threshold carcinogens.

e)          The proposed Guidelines provide for more flexibility in risk
            assessment and provide means for incorporating other types of
            biological data and information on mode of action into dose
            response assessment.  However, some EHC Members noted that,
            in some areas, there may not be enough guidance on mode of
            action determination and in other areas as well; given the wide
            usage of the Guidelines, these Members believe that additional
            clarification is needed.  Although sympathetic to the Agency's
            reluctance to develop additional criteria, the EHC recommends
            that further guidance should be provided. In addition, it was
            noted that, unlike other parts of the Guidelines, no examples are
            provided in the dose response section.  The Committee also noted
            that, often, a number of different models will provide an  equivalent
            fit to the data. However, some unrestricted models may give
            answers that are biologically unrealistic in some cases and should
            be avoided.  Since the issues regarding selection of a model
            for use in the observable range are largely generic, the EHC
            recommends that the Agency select a default procedure for
            routine use in such calculations. The EHC supports the
            Agency's efforts, reflected in these and other guidelines
            documents, to harmonize approaches to dose response
            evaluations for cancer and non-cancer endpoints, and
            encourages the Agency to do more..

f)           The EHC understands the theoretical desirability of the proposed
            biologically based  dose response model.  However, the proposed
            Guidelines may be misleading in this respect because no such
            model presently exists in a complete form, nor is it clear how
            to develop such in the foreseeable future. The Agency's
            definition of a biologically based model thus seems
            unnecessarily narrow.  The Agency should define more clearly
            what is meant by a  biologically based dose response model
            and to give guidance as to when such a model is preferred
            over the default linear or non-linear approaches. This report, in
            section 3.3.2, proposes a possible definition for the Agency's for
            the Agency's consideration.

                              34

-------
g)          To mitigate against unnecessary inconsistencies and
            confusion in the application of the Guidelines to determine the
            point of departure, the EHC suggests that further guidance be
            given in the Guidelines on this approach. In addition, the EHC
            sees little  scientific justification for separate approaches to LED10
            and benchmark calculations, and therefore recommends that the
            approaches used for non-carcinogens and (low dose) non-linear
            carcinogens (e.g., choice of dose-response model, choice of point
            of departure, and use of statistical confidence bounds) be
            harmonized to the extent possible.  In general, it is recommended
            that a single risk level, e.g., ED10, be selected as the point of
            departure when the (low  dose) non-linear approach is applied.
            The consensus of the Committee was that both point
            estimates and statistical bounds can  be useful in different
            circumstances, and recommended that the Agency routinely
            calculate and present the point estimate of the ED10 and the
            corresponding upper and lower 95% statistical bounds.

h)          There is no explicit statement in the proposal that statistical
            significance should be a basic requirement for determining
            causality.  This lack of an explicit statement has been interpreted
            as misleading and implying that there is a hidden intent to
            eliminate statistical significance as a consideration in assessing
            causality.  Adding appropriate and specific language
            concerning statistical significance should rectify this  problem.
            Also, because of the difficulties with the interpretation of
            observational studies, the EHC believes that a weight of
            evidence approach to evaluating causality is appropriate.

i)           Vis-a-vis  assessing tumors, the Committee endorses the
            procedure adopted by the National Toxicology program of
            combining some closely related tumor types for statistical
            analyses, but otherwise conducting separate tests for different
            tumor types.  The EHC also endorses the Agency's Guidelines
            that give  additional significance to the finding of uncommon
            tumor types, multiple sites, tumors by more than one  route of
            administration,  etc. (see pp. 55-56 of the Guidelines). Lastly,
            because of data  on the relationship of body weight and some tumor
            types, the EHC encourages the Agency to consider the role of diet

                               35

-------
            in a given tumor response.  Consideration of this issue should be
            addressed when evaluating a given tumor data set.

j)           The developing infant and child should be recognized as a
            subgroup which is particularly sensitive to the carcinogenicity of a
            number of agents. Risk assessments should explicitly account for
            the differential susceptibility of the young. Other differences in
            susceptibility should also be taken into account when data permit.
                                 36

-------
                               APPENDIX A

 Extract from The Consensus Report of an I ARC Monographs Working
 Group on Mechanisms of Carcinogenesis in Risk Identification (IARC
               Internal Technical Report No. 91/002,1991).
,™™«^^

      IV. PROPOSED PRINCIPLES AND PROCEDURES FOR USING information
ON MECHANISMS IN EVALUATING CARCINOGENIC RISKS TO HUMANS

1.  Introduction

The IARC Monographs programme is concerned with the evaluation of carcinogenic
hazard posed to humans by chemical and other agents.  In many cases,
epidemiological data adequate to permit definitive evaluations of human carcinogenicity
are not available.  In these circumstances, in addition to consideration of animal
bioassay data, assessment of data possibly relevant to the mechanism by which the
putative carcinogen acts may be helpful in making the overall evaluation.

2.  General principles

In view of the rapid and continuing development in understanding of fundamental
mechanisms of carcinogenesis, it would be unhelpful to define strict rules concerning
the need for epidemiological or other specific evidence to justify particular
classifications.  This raises the possibility that an agent might be placed in Group 1 in
The absence of sufficient epidemiological  evidence at least in the traditional sense.
'Sufficient' evidence of carcinogenicity in humans is in effect being redefined to
encompass its usual scientific meaning. To define what constitutes such evidence in
the current state of knowledge will be an important task of future Monographs working
groups.

      Many agents that might, in reality, cause substantial and/or widespread
increases in cancer risk in human populations are not found to increase risk in
conventional epidemiological research. The ability to measure exposures more
sensitively, and closer to the target tissue, along with the ability to identify tumors with
specific genetic markers, may greatly increase the capacity to detect cancer risks that
were previously undetectable.  (This is because the relative risks that are registered in
studies that incorporate better, more specific, biologically based measures are higher

                                   A-1

-------
and /or have lower statistical variance.) These developments will result in the
classification in Group 1 of an increased proportion of candidate agents.

      Mechanisms may be understood at many different levels (see pp. 40-44); e.g.,
for a genotoxic carcinogen: at the level of metabolism, DMA damage, repair, mutational
events, amino-acid changes in a proto-oncogene or tumor suppressor gene product,
changes in function of the protein, the effect of the altered protein on cellular function
or the stage in the carcinogenic process at which the change may be effective.
Because different agents act by different mechanisms, the relevance and importance of
each of these levels is dependent on which agent and which tumor site is being
considered.  It should always be made clear which levels are being considered and
which are not.  It should also be borne in mind that knowledge that an agent acts by
one mechanism does not exclude the operation of others.

      Conclusions about the operation of a particular mechanism should follow the
'strength-of- evidence' approach which is fundamental to the Agency's evaluations of
carcinogenicity per se. The strength of the evidence for a particular mechanism should
be assessed using terms such as 'weak', 'moderate' and 'strong'. The working groups
are responsible for assessing the available data with regard to the relevance,
reproducibility and sensitivity of the methods employed and other considerations of
good laboratory and scientific practice.

      The available evidence may show that similar mechanisms are acting in humans
and experimental animals.  Of particular concern are those situations in which the
possibility is considered of species-specific activity. One concern would be raised
when humans are the more affected or susceptible species. This could be evaluated
on the basis of knowledge of mechanisms and the comparative relevance  of a
mechanism to animal and human responses. Another concern is raised when the
putatively unaffected species is human beings.  Certain principles should be applied
before such species-specific activity can be concluded.  It should be established, (I) for
the tumor site in question, that the mechanism in question is the primary one in the
tumorigenesis in  that species; (ii) that the same or a similar mechanism  does not
operate in  humans; and (iii) whether the agent induces other types of tumors in
experimental animals.  If other types of tumors are induced, then (i) and (ii) would have
to be fulfilled for each of them. Qualitative differences, in which effects occurring in
one species are not expected to occur in another, should be distinguished from
quantitative differences (such as different rates of biotransformation), which may
influence only the degree of response rather than the presence or absence of a
response.
                                     A-2

-------
      In view of the considerations summarized above, the Group recommended that a
discussion of possible mechanisms be included in the monographs when appropriate
data are available.

3.  Use of data on mechanisms of carcinogenesis in the evaluation of
carcinogenic risk to humans

The following steps should be taken to determine the contribution made by specific
data on  (known or presumed) mechanisms of carcinogenesis to evaluation of the
carcinogenic risk to humans of a particular agent.

3.1  Summarization of the available data on mechanisms

For the agent, mixture or exposure circumstance being evaluated,  the available data on
mechanisms of carcinogenesis from studies in humans, animals and assorted tissue
and cell test systems should be considered within the following four descriptive
dimensions. These dimensions refer to the level at which the mechanism exists and
not necessarily to the level of the system in which the observations were made. An
agent may have effects on several or all dimensions.

The set  of dimensions constitutes a framework that should facilitate,  not rigidify, the
description  of the range of data available.  For each dimension, examples are given;
these are indicative only and would be subject to modification by each working group.

(a)    Evidence of genotoxicity (i.e., structural changes at the level of the gene), for
      example,
                  -structure-activity considerations
                  -toxicokinetic considerations
                  -adduct formation
                  -mutagenicity
                  -other genetic changes
(b)    Evidence of effects on the expression of relevant genes  (i.e., functional changes
      at the intracellular level), for example,
                  -alteration of the structure or quantity of the product of a proto-
                  oncogene or tumor suppressor gene
                  -other effect on gene expression
(c)    Evidence of relevant effects on cell behavior (i.e., morphological or behavioral
      changes at the cellular or tissue level), for example,
                  -induction of mitogenesis, compensatory cell proliferation,
                  preneoplasia and hyperplasia

                                     A-3

-------
                  -survival of premalignant or malignant cells
                  -metastatic potential
(d)    Evidence of time and dose relationships of carcinogenic effects and interactions
      between agents, for example,
                  -early/late stage, as inferred from multistage models
                  -initiation/promotion/progression/malignant conversion, as defined
                  in animal carcinogenicity experiments

      These dimensions are  not mutually exclusive.  Thus, the mechanisms by which
an agent acts on the expression of relevant genes would still be summarized under the
second dimension even if it were known with reasonable certainty that those effects
resulted from genotoxicity and would therefore also be described under the first
heading.

3.2 Evaluation of the strength of the evidence for action of certain mechanisms of
carcinogenesis

      It is proposed that the strength of the evidence for each effect be evaluated in
association with the framework proposed above.  The evaluation could be made using
terms such as 'strong', 'moderate' and 'weak' evidence for action of the agent via the
mechanism or, where appropriate, evidence of lack of effect of the mechanism.

      Ideally, operational guidelines for these evaluations would be worked out.  Such
guidelines may prove difficult to develop because of the diversity of the phenomena
being observed.  The provision of answers to the following general questions, however,
may  assist in evaluating the strength of the evidence for the action of particular
mechanisms:
      (1) Is the method valid, reliable and reproducible?
      (2) Is the time sequence between exposure to  the agent, measurement of the
      endpoint and cancer compatible with a cause-effect relationship?
      (3) Have the agent and the endpoint been evaluated in several different species
      and/or qualitatively different test systems?
      (4) Are there tenable alternatives to a cause-effect explanation for the
      association between exposure and the endpoint? Such alternatives would
      include chance, bias in design or conduct of the study and confounding of the
      exposure to the agent of interest with some exposure to other agent that can
      cause the endpoint.
      (5) Is the association between the agent and the mechanistic endpoint
      consistent with tumor response and other information on the carcinogenic
      process?

                                     A-4

-------
3.3.  Assessment of the relevance of the available data on mechanisms to the
evaluation of the carcinogenic risk of the agent to humans.

A number of issues are pertinent to assessing the relevance of data on mechanisms to
evaluation of the carcinogenic risk of the agent to humans, however strong the
evidence for an effect of the agent on the operation of these mechanisms might be.

      (1)     It would be desirable to have evidence that the effect lies in the chain
              of events that link carcinogenic agents with cancer. Such evidence
              need not necessarily be available for human, although evidence of the
              relevance of

      (2)     The  relevance of the test system to human responses must be
              considered.

      (3)     Generally, the closer the tissue, cells or cellular components of the
              system being studied are to the target cells in the mammon tissues in
              which the carcinogenic effect is thought to occur, the  more relevant
              data from the experimental system are to evaluation of carcinogenic
              risk.  Notwithstanding,  the validity of the endpoint examined is of
              greater relevance.

      (4)     A particular mechanistic result is more relevant to evaluation of
              carcinogenic risks in humans if it can be shown that toxicokinetic
              variables in the test system are similar to those in human beings.

      (5)     The  measured endpoint should be specific for the mechanism that is
              actually thought to be relevant to carcinogenesis. Thus, for example,
              when gene mutation is the relevant endpoint, greater weight would be
              given to results in which mutation had actually been observed than to
              those in which only less specific endpoints were attained.

3.4 Suggested guidelines

      When the available data on mechanisms are thought to be relevant to evaluation
of the carcinogenic risk of an agent to humans, they should be used in making the

                                     A-5

-------
overall evaluation, together with the combined evidence for animal and/or human
carcinogenicity.
      It is not possible to elaborate definitive guidelines for all possible situations in
which mechanistic data may influence evaluation of carcinogens.  The following
scenarios are illustrative of the range of options available.  First, information
concerning mechanisms of action may confirm  a particular level of carcinogen
classification as indicated on the basis of epidemiological and/or animal carcinogenicity
data.  Second, for a particular agent, strong evidence for a mechanism of action that is
relevant to carcinogenicity in humans could justify 'upgrading' its overall evaluation.
Third, an overall evaluation of human cancer hazard on the basis of animal
carcinogenicity data could be downgraded by strong evidence that the mechanism
responsible for tumor growth in experimental animals is not relevant to humans.  In
keeping with the goal of public health,  priority must be given to the demonstration that
the mechanism is  irrelevant to humans. Obviously, the absence of mechanistic data
should not influence the evaluation of human carcinogenic hazards on the basis of
epidemiological or animal tumor data.

      Despite these caveats, and in addition to the above-mentioned proposal for
systematic appraisal of mechanistic data (3.1, 3.2, 3.3), increased use of these data in
the overall evaluation may require modification  of the criteria for the IARC categories.

      The Working Group considered the impact of increased understanding of
mechanisms of carcinogenesis on the present guidelines for classification of
carcinogens into Groupl to 4.  It was noted that, as specified in the Preamble, the
guidelines for categorization are not definitive, and exceptions can be made.  Currently,
for all agents in Group 1  there  is sufficient evidence for cancer causation in humans
from studies that have established a link between levels of exposure and increased
cancer incidence.  A Group 1 categorization could also be achieved by:
                           'EXTRACT ENDS'
                                     A-6

-------
                               APPENDIX B

(From IARC Monographs Programme on the Evaluation of Carcinogenic Risks to
Humans. Preamble, pages 7-8, Volume 67,  1996)

      The Monographs represent the first step in carcinogenic risk assessment, which
involves examination of all relevant information in order to assess the strength of the
available evidence that certain exposures could alter the incidences of cancer in
humans. The second step is quantitative risk estimation.  Detailed, quantitative
evaluations of epidemiological data may be made in the Monographs, but without
extrapolation beyond the range of the data available. Quantitative extrapolation from
experimental data to the human situation is not undertaken.
                                    B-1

-------
                      DISTRIBUTION LIST
Administrator
Deputy Administrator
Assistant Administrators
Deputy Assistant Administrator for Science, ORD
Director, Office of Science Policy, ORD
Director, Office of Air Quality Planning and Standards, OAR
Director, National Center for Environmental Assessment, ORD, RTP, NC
Director, Office of Air Quality Planning and Standards, OAR,  RTP, NC
EPA Regional Administrators
EPA Laboratory Directors
EPA Headquarters Library
EPA Regional Libraries
EPA Laboratory Libraries
Library of Congress
National Technical Information Service
Congressional Research Service

-------