Notice: Meetings: Carcinogen Risk Assessment

Thursday,
April 7, 2005
Part II
Environmental
Protection Agency
Notice of Availability; Documents Entitled
Guidelines for Carcinogen Risk
Assessment and Supplemental Guidance
for Assessing Susceptibility From Early-
Life Exposure to Carcinogens; Notices
VerDate jul<14>2003 15:25 Apr 06, 2005 Jkt 205001 PO 00000 Frm 00001 Fmt 4717 Sfmt 4717 E:\FR\FM\07APN2.SGM 07APN2
17766 Federal Register / Vol. 70, No. 66 / Thursday, April 7, 2005 / Notices
ENVIRONMENTAL PROTECTION Government: Managing the Process, the comments, workshops and
AGENCY National Academy of Sciences recommendations of the SAB.
recommended that Federal regulatory CAA section 112(o)(7) provides ‘‘[t]he
[FRL–7895–2] Administrator shall consider, but need
agencies establish ‘‘inference
guidelines’’ to promote consistency and not adopt, the recommendations
Notice of Availability of the Document
technical quality in risk assessment, and contained in the report of the National
Entitled Guidelines for Carcinogen
to ensure that the risk assessment Academy of Sciences prepared pursuant
Risk Assessment
process is maintained as a scientific to this subsection and the views of the
AGENCY: U.S. Environmental Protection effort separate from risk management. A Science Advisory Board, with respect to
Agency (EPA). task force within EPA accepted that such report. Prior to the promulgation of
ACTION: Notice of availability of final recommendation and requested that any standard under [CAA section
document. EPA scientists begin to develop such 112(f)], and after notice and opportunity
guidelines. In 1984, EPA scientists for comment, the Administrator shall
SUMMARY: This Notice announces the publish revised Guidelines for
availability of the final document, began work on risk assessment
Carcinogenic Risk Assessment or a
Guidelines for Carcinogen Risk guidelines for carcinogenicity,
detailed explanation of the reasons that
Assessment (EPA/630/P–03/001F), mutagenicity, suspect developmental
any recommendations contained in the
hereafter referred to as the Guidelines. toxicants, chemical mixtures and report of the National Academy of
These Guidelines were developed as exposure assessment. Following Sciences will not be implemented.’’
part of an Agency-wide guidelines extensive scientific and public review, In response to CAA section 112(o)(7),
development program by a Technical these five guidelines were issued on the 1994 National Research Council
Panel of the U.S. EPA’s Risk Assessment September 24, 1986 (51 FR 33992– (NRC) report, and continuing
Forum, which was composed of 34054). Since 1986, additional risk developments in the science of cancer
scientists from throughout the Agency. assessment guidelines have been risk assessment, EPA began the process
Selected drafts were peer reviewed developed, revised and supplemented. of revising its Guidelines for Carcinogen
internally by the U.S. EPA’s Science EPA continues to revisit the Risk Assessment. Revisions to the
Advisory Board, and by experts from guidelines as experience and scientific Guidelines were intended to make
universities, environmental groups, consensus evolve. In 1996, the Agency greater use of the increasing scientific
industry and other governmental published proposed revisions to EPA’s understanding of the mechanisms that
agencies. The Guidelines were also 1986 cancer guidelines for public underlie the carcinogenic process.
subjected to several public comment comment. Since the 1996 proposal, the Several drafts of revisions to the
periods. Issuance of these final document has undergone extensive Guidelines have been subject to
Guidelines fulfills EPA’s obligations public comment and scientific peer extensive public comment and scientific
under section 112(o) (7) of the Clean Air review, including three reviews by peer review, including three reviews by
Act. EPA’s Science Advisory Board (SAB) in EPA’s SAB, as discussed above. EPA
DATES: The Guidelines are available for February 1997, January 1999 and July considered the 1994 recommendations
use by EPA risk assessors as March 29, 1999. The July 1999 review panel was of the NRC on the Guidelines. EPA’s
2005. supplemented by the EPA Children’s approach to those NRC
ADDRESSES: This Notice contains the Health Protection Advisory Committee. recommendations is reflected in the
full Guidelines document. The Public comments were received Guidelines. Draft EPA responses to the
Guidelines also are available NRC recommendations were presented
concurrent to each of these reviews. In
electronically through the EPA Web site in the preamble to the 1996 draft of
2001 (66 FR 59593, November 29, 2001)
at http://www.epa.gov/cancerguidelines. these revised Guidelines (61 FR 18003,
an additional public comment period
A limited number of paper and CDROM April 23, 1996). By issuing the final
was held requesting new information
copies will be available from the EPA’s Guidelines which address the
gained through the use of the July 1999
National Service Center for recommendations of the NRC, EPA has
draft final revised guidelines on issues
Environmental Publications (NSCEP), fulfilled its responsibilities under CAA
including, but not limited to, the nature
P.O. Box 42419, Cincinnati, OH 45242; section 112(o)(7).
and use of default assumptions;
telephone: (800) 490–9198 or (513) 489– definition and application of hazard Features of the Guidelines
8190; facsimile: (513) 489–8695. Please descriptors; identification of The Guidelines are intended to make
provide your name, mailing address and carcinogenic mode(s) of action and, in greater use of the increasing scientific
the title and number of the requested particular, consideration of relevancy understanding of the mechanisms that
EPA publication (EPA/630/P–03–001F). for children (e.g., the potential for underlie the carcinogenic process. The
Additionally, copies of the Guidelines differential life stage susceptibility); and final guidelines include discussions of
will be available for inspection at EPA guidance on the use of the margin of all of the four steps of the risk
headquarters and regional libraries, exposure analysis. The notice also assessment process and provide
through the U.S. Government announced that the July 1999 draft final guidance to risk assessors on these
Depository Library program. revised guidelines would serve as EPA’s steps. In applying these principles to the
FOR FURTHER INFORMATION CONTACT: Dr. interim guidance to EPA risk assessors development of these Guidelines, the
William P. Wood, Risk Assessment preparing cancer risk assessment, until following key issues were highlighted:
Forum, National Center for the issuance of final guidelines. In May use of default options, the consideration
Environmental Assessment (8601D), 2003 EPA made available for public of mode of action, understanding of
U.S. Environmental Protection Agency, comment a revised draft of the biological changes, fuller
Washington DC 20460, telephone: (202) guidelines, and in February 2005 the characterization of carcinogenic
564–3361; facsimile: (202) 565–0062; or guidelines underwent interagency potential, and consideration of
e-mail: risk.forum@epamail.epa.gov. review. The final Guidelines issued differences in susceptibility.
SUPPLEMENTARY INFORMATION: In the today are based, in part, upon the Use of default options—Default
1983 Risk Assessment in the Federal recommendations derived from public options are approaches that EPA can
Federal Register / Vol. 70, No. 66 / Thursday, April 7, 2005 / Notices 17767
apply in risk assessments when exist among people in their in Agency risk assessment documents,
scientific information about the effects susceptibility to carcinogens. Some and that Agency scientists identify the
of an agent on human health is subpopulations may experience strengths and weaknesses of each
unavailable, limited, or of insufficient increased susceptibility to carcinogens assessment by describing uncertainties,
quality. Under the final Guidelines, throughout their life, such as people assumptions and limitations, as well as
EPA’s approach begins with a critical who have inherited predisposition to the scientific basis and rationale for
analysis of available information, and certain cancer types or reduced capacity each assessment. The Guidelines are
then invokes defaults if needed to to repair genetic damage. Also, during formulated in part to bridge gaps in risk
address uncertainty or the absence of certain lifestages the entire population assessment methodology and data. By
critical information. may experience heightened identifying these gaps and the
Consideration of mode of action— susceptibility to carcinogens. In importance of the missing information
Cancer refers to a group of diseases particular, EPA notes that childhood to the risk assessment process, EPA
involving abnormal, malignant tissue may be a lifestage of greater wishes to encourage research and
growth. Research has revealed that the susceptibility for a number of reasons: analysis that will lead to new risk
development of cancer involves a rapid growth and development that assessment methods and data.
complex series of steps and that occurs prenatally and after birth, The Guidelines are guidance only.
carcinogens may operate in a number of differences related to an immature They do not establish any substantive
different ways. The final Guidelines metabolic system, and differences in ‘‘rules’’ under the Administrative
emphasize the value of understanding diet and behavior patterns that may Procedure Act or any other law and
the biological changes and how these increase exposure. have no binding effect on EPA or any
changes might lead to the development The final Guidelines explicitly call for regulated entity, but instead will
of cancer. They also discuss ways to consideration of possible sensitive represent a non-binding statement of
evaluate and use such information, subpopulations and/or lifestages (such policy. EPA believes that the Guidelines
including information about an agent’s as childhood). Therefore, concurrent represent a sound and up-to-date
postulated mode of action, or the series with release of the final Guidelines, EPA approach to cancer risk assessment and
of steps and processes that lead to published a separate guidance, entitled enhance the application of the best
cancer formation. Mode-of-action data, Supplemental Guidance for Assessing available science in EPA’s risk
when available and of sufficient quality, Susceptibility from Early-Life Exposure assessments. However, EPA cancer risk
may be used to draw conclusions about to Carcinogens (EPA/630/R–03/003F), assessments may be conducted
the potency of a chemical, its potential hereafter referred to as the differently than envisioned in the
effects at low doses, whether findings in Supplemental Guidance, describing Guidelines for many reasons, including
animals are relevant to humans, and possible approaches that could be used (but not limited to) new information,
which populations or lifestages may be to assess risks resulting from early life new scientific understanding or new
particularly susceptible. exposure to potential carcinogens. The science policy judgment. The science of
Fuller characterization of Supplemental Guidance is separate from risk assessment continues to develop
carcinogenic potential—In the final the Guidelines so that it may be more rapidly, and specific components of the
Guidelines, an agent’s human easily updated in a timely manner given Guidelines may become outdated or
carcinogenic potential is described in a the expected rapid evolution of may otherwise require modification in
weight-of-evidence narrative. The scientific understanding about the individual settings. Use of the
narrative summarizes the full range of effects of early-life exposures. Guidelines in future risk assessments
available evidence and describes any Availability of the Supplemental will be based on decisions by EPA that
conditions associated with conclusions Guidance is announced in a separate approaches from the Guidelines are
about an agent’s hazard potential. For notice, also published in today’s suitable and appropriate in the context
example, the narrative may explain that Federal Register. of those particular risk assessments.
a chemical appears to be carcinogenic These judgments will be tested through
by some routes of exposure but not by Risk Assessment Guidelines at EPA
peer review, and risk assessments will
others (e.g., by inhalation but not These Guidelines set forth principles be modified to use different approaches
ingestion). Similarly, a hazard may be and procedures to guide EPA scientists if appropriate.
attributed to exposures during sensitive in the conduct of cancer risk Even though the Guidelines are not
life-stages of development but not at assessments and to inform Agency binding rules, EPA is issuing them in a
other times. The narrative also decision makers and the public about manner consistent with the procedures
summarizes uncertainties and key these procedures. Policies in this in the Administrative Procedure Act
default options that have been invoked. document are intended as internal that are generally applicable to
To provide additional clarity and guidance for EPA. So risk assessors and rulemaking, including providing
consistency in weight-of-evidence risk managers at EPA are the primary opportunity for public comment. EPA
narratives, the Guidelines present a set audience. These Guidelines also provide considered and responded to all
of weight-of-evidence descriptors that basic information to the public about significant public comments as it
accompany the narratives. The EPA’s risk assessment methods. In prepared the Guidelines and will send
Guidelines emphasize that risk particular, the Guidelines emphasize a copy of the final Guidelines to
managers should consider the full range that risk assessments should be Congress. EPA certifies that the
of information in the narratives and not conducted on a case-by-case basis, Guidelines will not have a significant
focus exclusively on the descriptors. As giving full consideration to all relevant impact on a substantial number of small
in the case of the narratives, descriptors scientific information. This approach entities, because the Guidelines are for
may apply only to certain routes of means that Agency experts study the benefit of EPA and impose no
exposure, dose ranges and durations of scientific information on each agent requirements or costs on small entities.
exposure. under review and use the most
Consideration of differences in scientifically appropriate interpretation Implementation
susceptibility—The Guidelines to assess risk. The Guidelines also stress Beginning today, Guidelines and
explicitly recognize that variation may that this information be fully presented Supplemental Guidance serve as EPA’s
recommendation to Agency risk 2.3.4. Toxicological and Clinical Findings 5. Risk Characterization
assessors preparing cancer risk 2.3.5. Events Relevant to Mode of 5.1. Purpose
assessments. As EPA prepares cancer Carcinogenic Action 5.2. Application
2.3.5.1. Direct DNA-Reactive Effects 5.3. Presentation of the Risk
assessments under the Integrated Risk Characterization Summary
2.3.5.2. Indirect DNA Effects or Other
Information System (IRIS) program, as Effects on Genes/Gene Expression 5.4. Content of the Risk Characterization
well as in other EPA programs, the 2.3.5.3. Precursor Events and Biomarker Summary
Agency intends to begin to use the Information Appendix: Major Default Options
Guidelines and Supplemental Guidance. 2.3.5.4. Judging Data Appendix B: EPA’s Guidance for Data
EPA also intends to consider the 2.4. Mode of Action—General Quality Assessment
Guidelines and Supplemental Guidance Considerations and Framework for References
along with other selection factors when Analysis List of Figures
2.4.1. General Considerations
EPA selects agents for reassessment in Figure 1–1. Flow chart for early-life risk
2.4.2. Evaluating a Hypothesized Mode of
annual IRIS agendas (see for example, Action assessment using mode of action
70 FR 10616, March 4, 2005). 2.4.2.1. Peer Review framework
Dated: March 29, 2005. 2.4.2.2. Use of the Framework Figure 3–1. Compatibility of Alternative
2.4.3. Framework for Evaluating Each Points of Departure with Observed and
Stephen L. Johnson, Modeled Tumor Incidences
Hypothesized Carcinogenic Mode of
Acting Administrator. Action Figure 3–2. Crossing between 10% and 1%
2.4.3.1. Description of the Hypothesized Dose-Response Curves for Bladder
Contents
Mode of Action Carcinomas and Liver Carcinomas
1. Introduction Induced by 2–AAF
2.4.3.2. Discussion of the Experimental
1.1. Purpose and Scope of the Guidelines
Support for the Hypothesized Mode of 1. Introduction
1.2. Organization and Application of the
Action
Guidelines
2.4.3.3. Consideration of the Possibility of 1.1. Purpose and Scope of the
1.2.1. Organization
1.2.2. Application Other Modes of Action Guidelines
2.4.3.4. Conclusions About the
1.3. Key Features of the Cancer Guidelines These guidelines revise and replace
1.3.1. Critical Analysis of Available Hypothesized Mode of Action
2.4.4. Evolution with Experience the U.S. Environmental Protection
Information as the Starting Point for Agency’s (EPA’s, or the Agency’s)
Evaluation 2.5. Weight of Evidence Narrative
2.6. Hazard Characterization Guidelines for Carcinogen Risk
1.3.2. Mode of Action
1.3.3. Weight of Evidence Narrative 3. Dose-Response Assessment Assessment, published in 51 FR 33992,
1.3.4. Dose-response Assessment 3.1. Analysis of Dose September 24, 1986 (U.S. EPA, 1986a)
1.3.5. Susceptible Populations and 3.1.1. Standardizing Different Experimental and the 1999 interim final guidelines
Lifestages Dosing Regimens (U.S. EPA, 1999a; see U.S. EPA 2001b).
1.3.6. Evaluating Risks from Childhood 3.1.2. Toxicokinetic Data and Modeling They provide EPA staff with guidance
Exposures 3.1.3. Cross-species Scaling Procedures for developing and using risk
1.3.7. Emphasis on Characterization 3.1.3.1. Oral Exposures
assessments. They also provide basic
2. Hazard Assessment 3.1.3.2. Inhalation Exposures
3.1.4. Route Extrapolation information to the public about the
2.1. Overview of Hazard Assessment and Agency’s risk assessment methods.
Characterization 3.2. Analysis in the Range of Observation
3.2.1. Epidemiologic Studies These cancer guidelines are used with
2.1.1. Analyses of Data
2.1.2. Presentation of Results 3.2.2. Toxicodynamic (‘‘Biologically other risk assessment guidelines, such
2.2. Analysis of Tumor Data Based’’) Modeling as the Guidelines for Mutagenicity Risk
2.2.1. Human Data 3.2.3. Empirical Modeling (‘‘Curve Assessment (U.S. EPA, 1986b) and the
2.2.1.1. Assessment of Evidence of Fitting’’) Guidelines for Exposure Assessment
Carcinogenicity From Human Data 3.2.4. Point of Departure (POD) (U.S. EPA, 1992a). Consideration of
2.2.1.2. Types of Studies 3.2.5. Characterizing the POD: The POD other Agency guidance documents is
2.2.1.3. Exposure Issues Narrative also important in assessing cancer risks
2.2.1.4. Biological Markers 3.2.6. Relative Potency Factors
3.3. Extrapolation to Lower Doses
where procedures for evaluating specific
2.2.1.5. Confounding Actors
2.2.1.6. Statistical Considerations 3.3.1. Choosing an Extrapolation Approach target organ effects have been developed
2.2.1.6.1. Likelihood of Observing an Effect 3.3.2. Extrapolation Using a (e.g., assessment of thyroid follicular
2.2.1.6.2. Sampling and Other Bias Issues Toxicodynamic Model cell tumors, U.S. EPA, 1998a). All of
2.2.1.6.3. Combining Statistical Evidence 3.3.3. Extrapolation Using a Low-dose EPA’s guidelines should be consulted
Across Studies Linear Model when conducting a risk assessment in
2.2.1.7. Evidence for Causality 3.3.4. Nonlinear Extrapolation to Lower order to ensure that information from
2.2.2. Animal Data Doses studies on carcinogenesis and other
2.2.2.1. Long-term Carcinogenicity Studies 3.3.5. Comparing and Combining Multiple health effects are considered together in
2.2.2.1.1. Dosing Issues Extrapolations
3.4. Extrapolation to Different Human
the overall characterization of risk. This
2.2.2.1.2. Statistical Considerations
2.2.2.1.3. Concurrent and Historical Exposure Scenarios is particularly true in the case in which
Controls 3.5. Extrapolation to Susceptible a precursor effect for a tumor is also a
2.2.2.1.4. Assessment of Evidence of Populations and Lifestages precursor or endpoint of other health
Carcinogenicity From Long-term Animal 3.6. Uncertainty effects or when there is a concern for a
Studies 3.7. Dose-Response Characterization particular susceptible life-stage for
2.2.2.1.5. Site Concordance 4. Exposure Assessment which the Agency has developed
2.2.2.2. Perinatal Carcinogenicity Studies 4.1. Defining the Assessment Questions guidance, for example, Guidelines for
2.2.2.3. Other Studies 4.2. Selecting or Developing the Developmental Toxicity Risk
2.2.3. Structural Analogue Data Conceptual and Mathematical Models
2.3. Analysis of Other Key Data 4.3. Collecting Data or Selecting and
Assessment (U.S. EPA, 1991a). The
2.3.1. Physicochemical Properties Evaluating Available Data developmental guidelines discuss
2.3.2. Structure-Activity Relationships 4.3.1. Adjusting Unit Risks for Highly hazards to children that may result from
2.3.3. Comparative Metabolism and Exposed Populations and Lifestages exposures during preconception and
Toxicokinetics 4.4. Exposure Characterization prenatal or postnatal development to
sexual maturity. Similar guidelines exist with the purposes of these cancer dimension of characterization to the
for reproductive toxicant risk guidelines. hazard identification step: an evaluation
assessments (U.S. EPA, 1996a) and for These cancer guidelines are intended of the conditions under which its
neurotoxicity risk assessment (U.S. EPA, as guidance only. They do not establish expression is anticipated. Risk
1998b). The overall characterization of any substantive ‘‘rules’’ under the assessment questions addressed in these
risk is conducted within the context of Administrative Procedure Act or any cancer guidelines are as follows.
broader policies and guidance such as other law and have no binding effect on • For hazard—Can the identified
Executive Order 13045, ‘‘Protection of EPA or any regulated entity, but instead agent present a carcinogenic hazard to
Children From Environmental Health represent a non-binding statement of humans and, if so, under what
Risks and Safety Risks’’ (Executive policy. EPA believes that the cancer circumstances?
Order 13045, 1997) which is the guidelines represent a sound and up-to- • For dose response—At what levels
primary directive to Federal agencies date approach to cancer risk assessment, of exposure might effects occur?
and departments to identify and assess and the cancer guidelines enhance the • For exposure—What are the
environmental health risks and safety application of the best available science conditions of human exposure?
risks that may disproportionately affect in EPA’s risk assessments. However, • For risk—What is the character of
children. EPA cancer risk assessments may be the risk? How well do data support
conducted differently than envisioned conclusions about the nature and extent
The cancer guidelines encourage both of the risk from various exposures?
consistency in the procedures that in the cancer guidelines for many
reasons, including (but not limited to) The risk characterization process first
support scientific components of summarizes findings on hazard, dose
new information, new scientific
Agency decision making and flexibility response, and exposure
understanding, or new science policy
to allow incorporation of innovations characterizations and then develops an
judgment. The science of risk
and contemporaneous scientific integrative analysis of the whole risk
assessment continues to develop
concepts. In balancing these goals, the case. It ends in the writing of a technical
rapidly, and specific components of the
Agency relies on established scientific risk characterization. Other documents,
cancer guidelines may become outdated
peer review processes (U.S. EPA, 2000a; such as summaries for the risk managers
or may otherwise require modification
OMB 2004). The cancer guidelines and the public, reflecting the key points
in individual settings. Use of the cancer
incorporate basic principles and science of the risk characterization are usually
guidelines in future risk assessments
policies based on evaluation of the will be based on decisions by EPA that written. A summary for managers is a
currently available information. The the approaches are suitable and presentation for those who may or may
Agency intends to revise these cancer appropriate in the context of those not be familiar with the scientific details
guidelines when substantial changes are particular risk assessments. These of cancer assessment. It also provides
necessary. As more information about judgments will be tested through peer information for other interested readers.
carcinogenesis develops, the need may review, and risk assessments will be The initial steps in the risk
arise to make appropriate changes in modified to use different approaches if characterization process are to make
risk assessment guidance. In the appropriate. building blocks in the form of
interim, the Agency intends to issue characterizations of the assessments of
special reports, after appropriate peer 1.2. Organization and Application of the hazard, dose response, and exposure.
review, to supplement and update Cancer Guidelines The individual assessments and
guidance on single topics (e.g., U.S. 1.2.1. Organization characterizations are then integrated to
EPA, 1991b). One such guidance arrive at risk estimates for exposure
document, Supplemental Guidance for Publications by the Office of Science
scenarios of interest. As part of the
Assessing Susceptibility from Early-Life and Technology (OSTP, 1985) and the
characterization process, explicit
Exposure to Carcinogens National Research Council (NRC) (NRC,
evaluations are made of the hazard and
(‘‘Supplemental Guidance’’), was 1983, 1994) provide information and
risk potential for susceptible lifestages,
developed in conjunction with these general principles about risk
including children (U.S. EPA, 1995,
cancer guidelines (U.S. EPA., 2005). assessment. Risk assessment uses
2000b).
Because both the methodology and the available scientific information on the The 1994 NRC document also
data in the Supplemental Guidance (see properties of an agent 1 and its effects in explicitly called attention to the role of
Section 1.3.6) are expected to evolve biological systems to provide an the risk assessment process in
more rapidly than the issues addressed evaluation of the potential for harm as identifying scientific uncertainties that,
in these cancer guidelines, the two were a consequence of environmental if addressed, could serve to reduce their
developed as separate documents. The exposure. The 1983 and 1994 NRC uncertainty in future iterations of the
Supplemental Guidance, however, as documents organize risk assessment risk assessment. NRC recommended that
well as any other relevant (including information into four areas: Hazard when the Agency ‘‘reports estimates of
subsequent) guidance documents, identification, dose-response risk to decisions-makers and the public,
should be considered along with these assessment, exposure assessment, and it should present not only point
cancer guidelines as risk assessments for risk characterization. This structure estimates of risk, but also the sources
carcinogens are generated. The use of appears in these cancer guidelines, with and magnitudes of uncertainty
supplemental guidance, such as the additional emphasis placed on associated with these estimates’’ (p. 15).
Supplemental Guidance for Assessing characterization of evidence and Thus, the identified uncertainties serve
Cancer Susceptibility from Early-life conclusions in each area of the as a feedback loop to the research
Exposure to Carcinogens, has the assessment. In particular, the cancer community and decisionmakers,
advantage of allowing the Supplemental guidelines adopt the approach of the specifying areas and types of
Guidance to be modified as more data NRC’s 1994 report in adding a information that would be particularly
become available. Thus, the 1 The term ‘‘agent’’ refers generally to any
useful.
consideration of new, peer-reviewed chemical substance, mixture, or physical or
There are several reasons for
scientific understanding and data in an biological entity being assessed, unless otherwise individually characterizing the hazard,
assessment can always be consistent noted (See Section 1.2.2 for a note on radiation.). dose response, and exposure
assessments. One is that they are often The assessment of risk from radiation scientific knowledge of the phenomena
done by different people than those who sources is informed by the continuing in question and are also matters of
do the integrative analyses. The second examination of human data by the policy concerning the appropriate way
is that there is very often a lapse of time National Academy of Sciences/NRC in to bridge uncertainties that concern
between the conduct of hazard and its series of numbered reports: potential risk to human health.
dose-response analyses and the conduct ‘‘Biological Effects of Ionizing These cancer guidelines do not
of exposure assessment and integrative Radiation.’’ Although some of the suggest that all of the kinds of data
analysis. Thus, it is important to capture general principles of these cancer covered here will need to be available
characterizations of assessments as the guidelines may also apply to radiation or used for either assessment or decision
assessments are done to avoid the need risk assessments, some of the details of making. The level of detail of an
to go back and reconstruct them. their risk assessment procedures may assessment is a matter of Agency
Finally, frequently a single hazard not, as they are most focused on other management discretion regarding
assessment is used by several programs kinds of agents. Therefore, these cancer applicable decision-making needs. The
for several different exposure scenarios. guidelines are not intended to provide Agency generally presumes that key
There may be one or several documents the primary source of, or guidance for, cancer information (e.g., assessments
involved. ‘‘Integrative analysis’’ is a the Agency’s evaluation of the contained in the Agency’s Integrated
generic term; and many documents that carcinogenic risks of radiation. Risk Information System) is ‘‘influential
have other titles may contain integrative Not every EPA assessment has the information’’ as defined by the EPA
analyses. In the following sections, the same scope or depth, a factor recognized Information Quality Guidelines and
elements of these characterizations are by the National Academy of Sciences ‘‘highly influential’’ as defined by
discussed. (NRC, 1996). For example, EPA’s OMB’s Information Quality Bulletin for
1.2.2. Application Information Quality Guidelines (U.S. Peer Review (OMB 2004).
The cancer guidelines apply within EPA, 2002a, see Appendix B) discuss 1.3. Key Features of the Cancer
the framework of policies provided by influential information that ‘‘will have Guidelines
applicable EPA statutes and do not alter or does have a clear and substantial
impact * * * on important public 1.3.1. Critical Analysis of Available
such policies. Information as the Starting Point for
• The cancer guidelines cover the policies or private sector decisions
* * * that should adhere to a rigorous Evaluation
assessment of available data. They do
not imply that one kind of data or standard of quality.’’ It is often difficult As an increasing understanding of
another is prerequisite for regulatory to know a priori how the results of a risk carcinogenesis is becoming available,
action concerning any agent. It is assessment are likely to be used by the these cancer guidelines adopt a view of
important that, when evaluating and Agency. Some risk assessments may be default options that is consistent with
considering the use of any data, EPA used by Agency economists and policy EPA’s mission to protect human health
analysts incorporate the basic standards analysts, and the necessary information while adhering to the tenets of sound
of quality, as defined by the EPA for such analyses, as discussed in detail science. Rather than viewing default
Information Quality Guidelines (U.S. later in this document, should be options as the starting point from which
EPA, 2002a see Appendix B) and other included when practicable (U.S. EPA, departures may be justified by new
Agency guidance on data quality such 2002a). On the other hand, Agency staff scientific information, these cancer
as the EPA Quality Manual for often conduct screening-level guidelines view a critical analysis of all
Environmental Programs (U.S. EPA, assessments for priority setting or of the available information that is
2000e), as well as OMB Guidelines for separate assessments of hazard or relevant to assessing the carcinogenic
Ensuring and Maximizing the Quality, exposure for ranking purposes or to risk as the starting point from which a
Utility, and Integrity of Information decide whether to invest resources in default option may be invoked if needed
Disseminated by Federal Agencies collecting data for a full assessment. to address uncertainty or the absence of
(OMB, 2002). It is very important that Moreover, a given assessment of hazard critical information. Preference is given
all analyses consider the basic standards and dose response may be used with to using information that has been peer
of quality, including objectivity, utility, more than one exposure assessment that reviewed, e.g., reported in peer-
and integrity. A summary of the factors may be conducted separately and at reviewed scientific journals. The
and considerations generally used by different times as the need arises in primary goal of EPA actions is
the Agency when evaluating and studying environmental problems protection of human health;
considering the use of scientific and related to various exposure media. The accordingly, as an Agency policy, risk
technical information is contained in cancer guidelines apply to these various assessment procedures, including
EPA’s A Summary of General situations in appropriate detail, given default options that are used in the
Assessment Factors for Evaluating the the scope and depth of the particular absence of scientific data to the
Quality of Scientific and Technical assessment. For example, a screening contrary, should be health protective
Information (U.S. EPA, 2003). assessment may be based almost (U.S. EPA, 1999b).
• Risk management applies directives entirely on structure-activity Use of health protective risk
in statutes, which may require relationships (SARs) and default assessment procedures as described in
consideration of potential risk or solely options, when other data are not readily these cancer guidelines means that
hazard or exposure potential, along with available. When more data and estimates, while uncertain, are more
social, economic, technical, and other resources are readily available, likely to overstate than understate
factors in decision making. Risk assessments can use a critical analysis hazard and/or risk. NRC (1994)
assessments may be used to support of all of the available data as the starting reaffirmed the use of default options as
decisions, but in order to maintain their point of the risk assessment. Under ‘‘a reasonable way to cope with
integrity as decision-making tools, they these conditions, default options would uncertainty about the choice of
are not influenced by consideration of only be used to address uncertainties or appropriate models or theory’’ (p. 104).
the social or economic consequences of the absence of critical data. Default NRC saw the need to treat uncertainty
regulatory action. options are inferences based on general in a predictable way that is
‘‘scientifically defensible, consistent information is present but critical significant risk management decisions
with the agency’s statutory mission, and analysis reveals inadequacies, a default will often benefit from a more
responsive to the needs of decision- option may also be used. If critical comprehensive assessment, including
makers’’ (p. 86). The extent of health analysis of agent-specific information is alternative risk models having
protection provided to the public consistent with one or more biologically significant biological support. To the
ultimately depends upon what risk based models as well as with the default extent practicable, such assessments
managers decide is the appropriate option, the alternative models and the should provide central estimates of
course of regulatory action. When risk default option are both carried through potential risks in conjunction with
assessments are performed using only the assessment and characterized for the lower and upper bounds (e.g.,
one set of procedures, it may be difficult risk manager. In this case, the default confidence limits) and a clear statement
for risk managers to determine how model not only fits the data, but also of the uncertainty associated with these
much health protectiveness is built into serves as a benchmark for comparison estimates.
a particular hazard determination or risk with other analyses. This case also In the absence of sufficient data or
characterization. When there are highlights the importance of extensive understanding to develop of a robust,
alternative procedures having experimentation to support a conclusion biologically based model, an
significant biological support, the about mode of action, including appropriate policy choice is to have a
Agency encourages assessments to be addressing the issue of whether single preferred curve-fitting model for
performed using these alternative alternative modes of action are also each type of data set. Many different
procedures, if feasible, in order to shed plausible. Section 2.4 provides a curve-fitting models have been
light on the uncertainties in the framework for critical analysis of mode developed, and those that fit the
assessment, recognizing that the Agency of action information to address the observed data reasonably well may lead
may decide to give greater weight to one extent to which the available to several-fold differences in estimated
set of procedures than another in a information supports the hypothesized risk at the lower end of the observed
specific assessment or management mode of action, whether alternative range. In addition, goodness-of-fit to the
decision. modes of action are also plausible, and experimental observations is not by
Encouraging risk assessors to be whether there is confidence that the itself an effective means of
receptive to new scientific information, same inferences can be extended to discriminating among models that
NRC discussed the need for departures populations and lifestages that are not adequately fit the data (OSTP, 1985). To
from default options when a ‘‘sufficient represented among the experimental provide some measure of consistency
showing’’ is made. It called on EPA to data. across different carcinogen assessments,
articulate clearly its criteria for a Generally, cancer risk decisions strive EPA uses a standard curve-fitting
departure so that decisions to depart to be ‘‘scientifically defensible, procedure for tumor incidence data.
from default options would be consistent with the agency’s statutory Assessments that include a different
‘‘scientifically credible and receive mission, and responsive to the needs of approach should provide an adequate
public acceptance’’ (p. 91). It was decision-makers’’ (NRC, 1994). justification and compare their results
concerned that ad hoc departures would Scientific defensibility would be with those from the standard procedure.
undercut the scientific credibility of a evaluated through use of EPA’s Science Application of models to data should be
risk assessment. NRC envisioned that Advisory Board, EPA’s Office of conducted in an open and transparent
principles for choosing and departing Pesticide Programs’ Scientific Advisory manner.
from default options would balance Panel, or other independent expert peer
1.3.2. Mode of Action
several objectives, including ‘‘protecting review panels to determine whether a
the public health, ensuring scientific consensus among scientific experts The use of mode of action 2 in the
validity, minimizing serious errors in exists. Consistency with the Agency’s assessment of potential carcinogens is a
estimating risks, maximizing incentives statutory mission would consider main focus of these cancer guidelines.
for research, creating an orderly and whether the risk assessment overall This area of emphasis arose because of
predictable process, and fostering supports EPA’s mission to protect the significant scientific advances that
openness and trustworthiness’’ (p. 81). human health and safeguard the natural have developed concerning the causes
Appendices N–1 and N–2 of NRC environment. Responsiveness to the of cancer induction. Elucidation of a
(1994) discussed two competing needs of decisionmakers would take mode of action for a particular cancer
standards for choosing default options into account pragmatic considerations response in animals or humans is a
articulated by members of the such as the nature of the decision; the data-rich determination. Significant
committee. One suggested approach required depth of analysis; the utility,
would evaluate a departure in terms of time, and cost of generating new 2 The term ‘‘mode of action’’ is defined as a
whether ‘‘it is scientifically plausible’’ sequence of key events and processes, starting with
scientific data; and the time, personnel, interaction of an agent with a cell, proceeding
and whether it ‘‘tends to protect public and resources allotted to the risk through operational and anatomical changes, and
health in the face of scientific assessment. resulting in cancer formation. A ‘‘key event’’ is an
uncertainty’’ (p. 601). An alternative With a multitude of types of data, empirically observable precursor step that is itself
approach ‘‘emphasizes scientific analyses, and risk assessments, as well a necessary element of the mode of action or is a
biologically based marker for such an element.
plausibility with regard to the use of as the diversity of needs of Mode of action is contrasted with ‘‘mechanism of
alternative models’’ (p. 631). Reaching decisionmakers, it is neither possible action,’’ which implies a more detailed
no consensus on a single approach, NRC nor desirable to specify step-by-step understanding and description of events, often at
recognized that developing criteria for criteria for decisions to invoke a default the molecular level, than is meant by mode of
action. The toxicokinetic processes that lead to
departures is an EPA policy matter. option. A discussion of major default formation or distribution of the active agent to the
The basis for invoking a default options appears in the Appendix. target tissue are considered in estimating dose but
option depends on the circumstances. Screening-level assessments may more are not part of the mode of action as the term is
Generally, if a gap in basic readily use default parameters, even used here. There are many examples of possible
modes of carcinogenic action, such as mutagenicity,
understanding exists or if agent-specific worst-case assumptions, that would not mitogenesis, inhibition of cell death, cytotoxicity
information is missing, a default option be appropriate in a full-scale with reparative cell proliferation, and immune
may be used. If agent-specific assessment. On the other hand, suppression.
information should be developed to epidemiologic studies are generally action for each tumor type. Because an
ensure that a scientifically justifiable preferred for characterizing human agent may induce multiple tumor types,
mode of action underlies the process cancer hazard and risk. However, all of the dose-response assessment includes
leading to cancer at a given site. In the the information discussed above could an analysis of all tumor types, followed
absence of sufficiently, scientifically provide valuable insights into the by an overall synthesis that includes a
justifiable mode of action information, possible mode(s) of action and characterization of the risk estimates
EPA generally takes public health- likelihood of human cancer hazard and across tumor types, the strength of the
protective, default positions regarding risk. The cancer guidelines recognize mode of action information of each
the interpretation of toxicologic and the growing sophistication of research tumor type, and the anticipated
epidemiologic data: Animal tumor methods, particularly in their ability to relevance of each tumor type to humans,
findings are judged to be relevant to reveal the modes of action of including susceptible populations and
humans, and cancer risks are assumed carcinogenic agents at cellular and lifestages (e.g., childhood).
to conform with low dose linearity. subcellular levels as well as Dose-response assessment for each
Understanding of mode of action can toxicokinetic processes. tumor type is performed in two steps:
be a key to identifying processes that Weighing of the evidence includes assessment of observed data to derive a
may cause chemical exposures to addressing not only the likelihood of point of departure (POD),4 followed by
differentially affect a particular human carcinogenic effects of the agent extrapolation to lower exposures to the
population segment or lifestage. Some but also the conditions under which extent that is necessary. Data from
modes of action are anticipated to be such effects may be expressed, to the epidemiologic studies, of sufficient
mutagenic and are assessed with a extent that these are revealed in the quality, are generally preferred for
linear approach. This is the mode of toxicological and other biologically estimating risks. When animal studies
action of radiation and several other important features of the agent. are the basis of the analysis, the
agents that are known carcinogens. The weight of evidence narrative to estimation of a human-equivalent dose
Other modes of action may be modeled characterize hazard summarizes the should utilize toxicokinetic data to
with either linear or nonlinear 3 results of the hazard assessment and inform cross-species dose scaling if
approaches after a rigorous analysis of provides a conclusion with regard to appropriate and if adequate data are
available data under the guidance human carcinogenic potential. The available. Otherwise, default procedures
provided in the framework for mode of narrative explains the kinds of evidence should be applied. For oral dose, based
action analysis (see Section 2.4.3). available and how they fit together in on current science, an appropriate
drawing conclusions, and it points out default option is to scale daily applied
1.3.3. Weight of Evidence Narrative significant issues/strengths/limitations doses experienced for a lifetime in
The cancer guidelines emphasize the of the data and conclusions. Because the proportion to body weight raised to the
importance of weighing all of the narrative also summarizes the mode of 3⁄4 power (U.S. EPA, 1992b). For
evidence in reaching conclusions about action information, it sets the stage for inhalation dose, based on current
the human carcinogenic potential of the discussion of the rationale science, an appropriate default
agents. This is accomplished in a single underlying a recommended approach to methodology estimates respiratory
integrative step after assessing all of the dose-response assessment. deposition of particles and gases and
individual lines of evidence, which is in In order to provide some measure of estimates internal doses of gases with
contrast to the step-wise approach in the clarity and consistency in an otherwise different absorption characteristics.
1986 cancer guidelines. Evidence free-form, narrative characterization, When toxicokinetic modeling (see
considered includes tumor findings, or standard descriptors are used as part of Section 3.1.2) is used without
lack thereof, in humans and laboratory the hazard narrative to express the toxicodynamic modeling (see Section
animals; an agent’s chemical and conclusion regarding the weight of 3.2.2), the dose-response assessment
physical properties; its structure-activity evidence for carcinogenic hazard develops and supports an approach for
relationships (SARs) as compared with potential. There are five recommended addressing toxicodynamic equivalence,
other carcinogenic agents; and studies standard hazard descriptors: perhaps by retaining some of the cross-
addressing potential carcinogenic ‘‘Carcinogenic to Humans,’’ ‘‘Likely to species scaling factor (see Section 3.1.3).
processes and mode(s) of action, either Be Carcinogenic to Humans,’’ Guidance is also provided for
in vivo or in vitro. Data from ‘‘Suggestive Evidence of Carcinogenic adjustment of dose from adults to
Potential,’’ ‘‘Inadequate Information to children (see Section 4.3.1).
3 The term ‘‘nonlinear’’ is used here in a narrower Assess Carcinogenic Potential,’’ and Response data on effects of the agent
sense than its usual meaning in the field of ‘‘Not Likely to Be Carcinogenic to on carcinogenic processes are analyzed
mathematical modeling. In these cancer guidelines, Humans.’’ Each standard descriptor may (nontumor data) in addition to data on
the term ‘‘nonlinear’’ refers to threshold models tumor incidence. If appropriate, the
(which show no response over a range of low doses
be applicable to a wide variety of data
that include zero) and some nonthreshold models sets and weights of evidence and is analyses of data on tumor incidence and
(e.g., a quadractic model, which shows some presented only in the context of a on precursor effects may be used in
response at all doses above zero). In these cancer weight of evidence narrative. combination. To the extent the
guidelines, a nonlinear model is one whose slope relationship between precursor effects
is zero at (and perhaps above) a dose of zero. A low-
Furthermore, as described in Section 2.5
dose-linear model is one whose slope is greater than of these cancer guidelines, more than and tumor incidence are known,
zero at a dose of zero. A low-dose-linear model one conclusion may be reached for an precursor data may be used to estimate
approximates a straight line only at very low doses; agent. a dose-response function below the
at higher doses near the observed data, a low-dose- observable tumor data. Study of the
linear model can display curvature. The term ‘‘low- 1.3.4. Dose-Response Assessment dose-response function for effects
dose-linear’’ is often abbreviated ‘‘linear,’’ although
a low-dose-linear model is not linear at all doses. Dose-response assessment evaluates
Use of nonlinear approaches does not imply a potential risks to humans at particular 4 A ‘‘point of departure’’ (POD) marks the
biological threshold dose below which the response exposure levels. The approach to dose- beginning of extrapolation to lower doses. The POD
is zero. Estimating thresholds can be problematic; is an estimated dose (usually expressed in human-
for example, a response that is not statistically
response assessment for a particular equivalent terms) near the lower end of the
significant can be consistent with a small risk that agent is based on the conclusion observed range, without significant extrapolation to
falls below an experiment’s power of detection. reached as to its potential mode(s) of lower doses.
believed to be part of the carcinogenic Otherwise, default approaches can be These cancer guidelines view
process influenced by the agent may applied that are consistent with current childhood as a sequence of lifestages
also assist in evaluating the relationship understanding of mode(s) of action of rather than viewing children as a
of exposure and response in the range the agent, including approaches that subpopulation, the distinction being
of observation and at exposure levels assume linearity or nonlinearity of the that a subpopulation refers to a portion
below the range of observation. dose-response relationship, or both. A of the population, whereas a lifestage is
The first step of dose-response default approach for linearity extends a inclusive of the entire population.
assessment is evaluation within the straight line from the POD to zero dose/ Exposures that are of concern extend
range of observation. Approaches to zero response (see Section 3.3.3). The from conception through adolescence
analysis of the range of observation of linear approach is used when: (1) There and also include pre-conception
epidemiologic studies are determined is an absence of sufficient information exposures of both parents. These cancer
by the type of study and how dose and on modes of action or (2) the mode of guidelines use the term ‘‘childhood’’ in
response are measured in the study. In action information indicates that the this more inclusive sense.
the absence of adequate human data for dose-response curve at low dose is or is Rarely are there studies that directly
dose-response analysis, animal data are expected to be linear. Where alternative evaluate risks following early-life
generally used. If there are sufficient approaches have significant biological exposure. Epidemiologic studies of
quantitative data and adequate support, and no scientific consensus early-life exposure to environmental
understanding of the carcinogenic favors a single approach, an assessment agents are seldom available. Standard
process, a biologically based model may may present results using alternative animal bioassays generally begin dosing
be developed to relate dose and approaches. A nonlinear approach can after the animals are several weeks old,
response data on an agent-specific basis. be used to develop a reference dose or when many organ systems are mature.
Otherwise, as a default procedure, a a reference concentration (see Section This could lead to an understatement of
standard model can be used to curve-fit 3.3.4). risk, because an accepted concept in the
the data. science of carcinogenesis is that young
The POD for extrapolating the 1.3.5. Susceptible Populations and animals are usually more susceptible to
relationship to environmental exposure Lifestages the carcinogenic activity of a chemical
levels of interest, when the latter are An important use of mode of action than are mature animals (McConnell,
outside the range of observed data, is information is to identify susceptible 1992).
generally the lower 95% confidence populations and lifestages. It is rare to At this time, there is some evidence
limit on the lowest dose level that can have epidemiologic studies or animal of higher cancer risks following early-
be supported for modeling by the data. bioassays conducted in susceptible life exposure. For radiation
SAB (1997) suggested that, ‘‘it may be individuals. This information need can carcinogenesis, data indicate that risks
appropriate to emphasize lower be filled by identifying the key events of for several forms of cancer are highest
statistical bounds in screening analyses the mode of action and then identifying following childhood exposure (NRC,
and in activities designed to develop an risk factors, such as differences due to 1990; Miller, 1995; U.S. EPA, 1999c).
appropriate human exposure value, genetic polymorphisms, disease, altered These human results are supported by
since such activities require accounting organ function, lifestyle, and lifestage, the few animal bioassays that include
for various types of uncertainties and a that can augment these key events. To perinatal (prenatal or early postnatal)
lower bound on the central estimate is do this, the information about the key exposure. Perinatal exposure to some
a scientifically-based approach precursor events is reviewed to identify agents can induce higher incidences of
accounting for the uncertainty in the particular populations or lifestages that the tumors seen in standard bioassays;
true value of the ED10 [or central can be particularly susceptible to their some examples include vinyl chloride
estimate].’’ However, the consensus of occurrence (see Section 2.4.3.4). Any (Maltoni et al., 1981),
the SAB (1997) was that, ‘‘both point information suggesting quantitative diethylnitrosamine (Peto et al., 1984),
estimates and statistical bounds can be differences between populations or benzidine, DDT, dieldrin, and safrole
useful in different circumstances, and lifestages is flagged for consideration in (Vesselinovitch et al., 1979). Moreover,
recommended that the Agency routinely the dose-response assessment (see perinatal exposure to some agents,
calculate and present the point estimate Section 3.5 and U.S. EPA 2002b). including vinyl chloride (Maltoni et al.,
of the ED10 [or central estimate] and the 1981) and saccharin (Cohen, 1995;
1.3.6. Evaluating Risks From Childhood
corresponding upper and lower 95% Whysner and Williams, 1996), can
Exposures
statistical bounds.’’ For example, it may induce different tumors that are not
be appropriate to emphasize the central NRC (1994) recommended that ‘‘EPA seen in standard bioassays. Surveys
estimate in activities that involve formal should assess risks to infants and comparing perinatal carcinogenesis
uncertainty analysis that are required by children whenever it appears that their bioassays with standard bioassays for a
OMB Circular A–4 (OMB, 2003) as well risks might be greater than those of limited number of chemicals
as ranking agents as to their adults.’’ Executive Order 13045 (1997) (McConnell, 1992; U.S. EPA, 1996b)
carcinogenic hazard. Thus, risk requires that ‘‘each Federal Agency shall have concluded that
assessors should calculate, to the extent make it a high priority to identify and • The same tumor sites are usually
practicable, and present the central assess environmental health and safety observed following either perinatal or
estimate and the corresponding upper risks that may disproportionately affect adult exposure, and
and lower statistical bounds (such as children, and shall ensure that their • Perinatal exposure in conjunction
confidence limits) to inform policies, programs, and standards with adult exposure usually increases
decisionmakers. address disproportionate risks that the incidence of tumors or reduces the
The second step of dose-response result from environmental health risks latent period before tumors are
assessment is extrapolation to lower or safety risks.’’ In assessing risks to observed.
dose levels, if needed. This children, EPA considers both effects The risk attributable to early-life
extrapolation is based on extension of a manifest during childhood and early-life exposure often appears modest
biologically based model if supported exposures that can contribute to effects compared with the risk from lifetime
by substantial data (see Section 3.3.2). at any time later in life. exposure, but it can be about 10-fold
higher than the risk from an exposure of aspects of development listed above. Supplemental Guidance recommends,
similar duration occurring later in life Examples of such data include for such chemicals when no chemical-
(Ginsberg, 2003). Further research is toxicokinetics that predict a sufficiently specific data exist, a default approach
warranted to investigate the extent to large internal dose in children or a using estimates from chronic studies
which these findings apply to specific mode of action where a key precursor (i.e., cancer slope factors) with
agents, chemical classes, and modes of event is more likely to occur during appropriate modifications to address the
action or in general. childhood. There is no recommended potential for differential risk of early-
These empirical results are consistent default to settle the question of whether lifestage exposure.
with current understanding of the tumors arising through a mode of action The Agency considered both the
biological processes involved in are relevant during childhood; and advantages and disadvantages to
carcinogenesis, which leads to a adequate understanding the mode of extending the recommended, age
reasonable expectation that children can action implies that there are sufficient dependent adjustment factors for
be more susceptible to many data (on either the specific agent or the carcinogenic potency to carcinogenic
carcinogenic agents (Anderson et al., general mode of action) to form a agents for which the mode of action
2000; Birnbaum and Fenton, 2003; confident conclusion about relevance remains unknown. EPA decided to
Ginsberg, 2003; Miller et al., 2002; during childhood (see Section 2.4.3.4). recommend these factors only for
Scheuplein et al., 2002). Some aspects In the dose-response assessment, the carcinogens acting through a mutagenic
potentially leading to childhood potential for susceptibility during mode of action based on a combination
susceptibility are listed below. childhood warrants explicit of analysis of available data and long-
• Differences in the capacity to consideration in each assessment. These standing science policy positions which
metabolize and clear chemicals can cancer guidelines encourage developing govern the Agency’s overall approach to
result in larger or smaller internal doses separate risk estimates for children carcinogen risk assessment. In general,
of the active agent(s). according to a tiered approach that the Agency prefers to rely on analyses
• More frequent cell division during considers what pertinent data are of data, rather than general defaults.
development can result in enhanced available (see Section 3.5). Childhood When data are available for a sensitive
expression of mutations due to the may be a susceptible period; moreover, lifestage, they would be used directly to
reduced time available for repair of exposures during childhood generally evaluate risks for that chemical and that
DNA lesions (Slikker et al., 2004). are not equivalent to exposures at other lifestage on a case-by-case basis. In the
• Some embryonic cells, such as times and may be treated differently case of nonmutagenic carcinogens,
brain cells, lack key DNA repair from exposures occurring later in life when the mode of action is unknown,
enzymes. (see Section 3.5). In addition, the data were judged by EPA to be too
• More frequent cell division during adjustment of unit risk estimates may be limited and the modes of action too
development can result in clonal warranted when used to estimate risks diverse to use this as a category for
expansion of cells with mutations from from childhood exposure (see Section which a general default adjustment
prior unrepaired DNA damage (Slikker 4.4). factor approach can be applied. In this
et al., 2004). At this time, several limitations situation, a linear low-dose
• Some components of the immune preclude a full assessment of children’s extrapolation methodology (without
system are not fully functional during risk. There are no generally used testing further adjustment) is recommended. It
development (Holladay and protocols to identify potential is the Agency’s long-standing science
Smialowicz, 2000; Holsapple et al., environmental causes of cancers that are policy position that use of the linear
2003). unique to children, including several low-dose extrapolation approach
• Hormonal systems operate at forms of childhood cancer and cancers provides adequate public health
different levels during different that develop from parental exposures, conservatism in the absence of
lifestages. and cases where developmental chemical-specific data indicating
• Induction of developmental exposure may alter susceptibility to differential early-life sensitivity or when
abnormalities can result in a carcinogen exposure in the adult the mode of action is not mutagenic.
predisposition to carcinogenic effects (Birnbaum and Fenton, 2003). Dose- The Agency expects to produce
later in life (Anderson et al., 2000; response assessment is limited by an additional supplemental guidance for
Birnbaum and Fenton, 2003; Fenton and inability to observe how developmental other modes of action, as data from new
Davis, 2002). exposure can modify incidence and research and toxicity testing indicate it
To evaluate risks from early-life latency and an inability to estimate the is warranted. EPA intends to focus its
exposure, these cancer guidelines ultimate tumor response resulting from research, and work collaboratively with
emphasize the role of toxicokinetic induced susceptibility to later its federal partners, to improve
information to estimate levels of the carcinogen exposures. understanding of the implications of
active agent in children and To partially address the limitations early life exposure to carcinogens.
toxicodynamic information to identify identified above, EPA developed in Development of guidance for estrogenic
whether any key events of the mode of conjunction with these cancer agents and chemicals acting through
action are of increased concern early in guidelines, Supplemental Guidance for other processes resulting in endocrine
life. Developmental toxicity studies can Assessing Susceptibility from Early-Life disruption and subsequent
provide information on critical periods Exposure to Carcinogens carcinogenesis, for example, might be a
of exposure for particular targets of (‘‘Supplemental Guidance’’). The reasonable priority in light of the human
toxicity. Supplemental Guidance addresses a experience with diethylstilbesterol and
An approach to assessing risks from number of issues pertaining to cancer the existing early life animal studies. It
early-life exposure is presented in risks associated with early-life is worth noting that each mode of action
Figure 1–1. In the hazard assessment, exposures generally, but provides for endocrine disruption will probably
when there are mode of action data, the specific guidance on procedures for require separate analysis.
assessment considers whether these adjusting cancer potency estimates only As the Agency examines additional
data have special relevance during for carcinogens acting through a carcinogenic agents, the age groupings
childhood, considering the various mutagenic mode of action. This may differ from those recommended for
assessing cancer risks from early-life Guidance. For example, the age 1.3.7. Emphasis on Characterization
exposure to chemicals with a mutagenic groupings selected for the age- The cancer guidelines emphasize the
mode of action. Puberty and its dependent adjustments for carcinogens importance of a clear and useful
associated biological changes, for acting through a mutagenic mode of characterization narrative that
example, involve many biological action were initially selected based on summarizes the analyses of hazard,
processes that could lead to changes in the available data, i.e., for the laboratory dose-response, and exposure
sensitivity to the effects of some animal age range representative of birth assessment. These characterizations
carcinogens, depending on their mode to < 2 years in humans. More limited summarize the assessments to explain
of action. The Agency is interested in data and information on human biology the extent and weight of evidence, major
identifying lifestages that may be were used to determine a science- points of interpretation and rationale for
particularly sensitive or refractory for informed policy regarding 2 to < 16 their selection, strengths and
carcinogenesis, and believes that the years. Data were not available to refine weaknesses of the evidence and the
mode of action framework described in the latter age group. If more data become analysis, and discuss alternative
these cancer guidelines is an available regarding carcinogens with a conclusions and uncertainties that
appropriate mechanism for elucidating mutagenic mode of action, deserve serious consideration (U.S.
these lifestages. For each additional consideration may be given to further EPA, 2000b). They serve as starting
mode of action evaluated, the various refinement of these age groups. materials for the overall risk
age groupings determined to be at characterization process that completes
differential risk may differ from those the risk assessment.
proposed in the Supplemental BILLING CODE 6560–50–P
BILLING CODE 6560–50–C

EN07AP05.000</GPH>
2. Hazard Assessment which the conclusion rests. This populations and the factors that may
narrative is a brief summary that in toto influence that distribution. The goals of
2.1. Overview of Hazard Assessment
replaces the alphanumerical cancer epidemiology are to identify
and Characterization
classification system used in EPA’s 1986 distribution of cancer risk and
2.1.1. Analyses of Data cancer guidelines (U.S. EPA, 1986a). determine the extent to which the risk
The purpose of hazard assessment is 2.2. Analysis of Tumor Data can be attributed causally to specific
to review and evaluate data pertinent to exposures to exogenous or endogenous
Evidence of carcinogenicity comes factors (see Centers for Disease Control
two questions: (1) Whether an agent from finding tumor increases in humans
may pose a carcinogenic hazard to and Prevention [CDC, 2004]).
or laboratory animals exposed to a given
human beings, and (2) under what Epidemiologic data are extremely
agent or from finding tumors following
circumstances an identified hazard may exposure to structural analogues to the valuable in risk assessment because they
be expressed (NRC, 1994). Hazard compound under review. The provide direct evidence on whether a
assessment involves analyses of a significance of observed or anticipated substance is likely to produce cancer in
variety of data that may range from tumor effects is evaluated in reference to humans, thereby avoiding issues such
observations of tumor responses to all the other key data on the agent. This as: species-to-species inference,
analysis of structure-activity section contains guidance for analyzing extrapolation to exposures relevant to
relationships (SARs). The purpose of the human and animal studies to decide people, effects of concomitant exposures
assessment is not simply to assemble whether there is an association between due to lifestyles. Thus, epidemiologic
these separate evaluations; its purpose exposure to an agent or a structural studies typically evaluate agents under
is to construct a total analysis analogue and occurrence of tumors. more relevant conditions. When human
examining what the biological data Note that the use of the term ‘‘tumor’’ data of high quality and adequate
reveal as a whole about carcinogenic in these cancer guidelines is defined as statistical power are available, they are
effects and mode of action of the agent, malignant neoplasms or a combination generally preferable over animal data
and their implications for human hazard of malignant and corresponding benign and should be given greater weight in
and dose-response evaluation. neoplasms. hazard characterization and dose-
Conclusions are drawn from weight-of- Observation of only benign neoplasia response assessment, although both can
evidence evaluations based on the may or may not have significance for be used.
combined strength and coherence of evaluation under these cancer
inferences appropriately drawn from all guidelines. Benign tumors that are not Null results from epidemiologic
of the available information. To the observed to progress to malignancy are studies alone generally do not prove the
extent that data permit, hazard assessed on a case-by-case basis. There absence of carcinogenic effects because
assessment addresses the question of is a range of possibilities for their such results can arise either from an
mode of action of an agent as both an overall significance. They may deserve agent being truly not carcinogenic or
initial step in identifying human hazard attention because they are serious health from other factors such as: inadequate
potential and as a component in problems even though they are not statistical power, inadequate study
considering appropriate approaches to malignant; for instance, benign tumors design, imprecise estimates, or
dose-response assessment. may be a health risk because of their confounding factors. Moreover, null
The topics in this chapter include effect on the function of a target tissue results from a well-designed and well-
analysis of tumor data, both human and such as the brain. They may be conducted epidemiologic study that
animal, and analysis of other key significant indicators of the need for contains usable exposure data can help
information about properties and effects further testing of an agent if they are to define upper limits for the estimated
that relate to carcinogenic potential. The observed in a short-term test protocol, or dose of concern for human exposure in
chapter addresses how information can such an observation may add to the cases where the overall weight of the
be used to evaluate potential modes of overall weight of evidence if the same evidence indicates that the agent is
action. It also provides guidance on agent causes malignancies in a long- potentially carcinogenic in humans.
performing a weight of evidence term study. Knowledge of the mode of Furthermore, data from a well designed
evaluation. action associated with a benign tumor and well conducted epidemiologic
response may aid in the interpretation study that does not show positive
2.1.2. Presentation of Results
of other tumor responses associated results, in conjunction with compelling
Presentation of the results of hazard with the same agent. In other cases, mechanistic information, can lend
assessment should be informed by observation of a benign tumor response support to a conclusion that animal
Agency guidance as discussed in alone may have no significant health responses may not be predictive of a
Section 2.6. The results are presented in hazard implications when other sources human cancer hazard.
a technical hazard characterization that of evidence show no suggestion of
serves as a support to later risk Epidemiology can also complement
carcinogenicity.
characterization. It includes: experimental evidence in corroborating
• A summary of the evaluations of 2.2.1. Human Data or clarifying the carcinogenic potential
hazard data, Human data may come from of the agent in question. For example,
• The rationales for its conclusions, epidemiologic studies or case reports. epidemiologic studies that show
and (Clinical human studies, which involve elevated cancer risk for tumor sites
• An explanation of the significant intentional exposures to substances, corresponding to those at which
strengths or limitations of the may provide toxicokinetic data, but laboratory animals experience increased
conclusions. generally not data on carcinogenicity.) tumor incidence can strengthen the
Another presentation feature is the The most common sources of human weight of evidence of human
use of a weight of evidence narrative data for cancer risk assessment are carcinogenicity. Furthermore,
that includes both a conclusion about epidemiologic investigations. biochemical or molecular epidemiology
the weight of evidence of carcinogenic Epidemiology is the study of the may help improve understanding of the
potential and a summary of the data on distribution of disease in human mechanisms of human carcinogenesis.
2.2.1.1. Assessment of Evidence of 2.2.1.2. Types of Studies causality can rarely be inferred from
Carcinogenicity From Human Data The major types of cancer case reports alone. Investigative follow-
epidemiologic study designs used for up may or may not accompany such
All studies that are considered to be reports. For cancer, the most common
examining environmental causes of
of acceptable quality, whether yielding types of case series are associated with
cancer are analytical studies and
positive or null results, or even occupational and childhood exposures.
descriptive studies. Each study type has
suggesting protective carcinogenic Case reports can be particularly valuable
well-known strengths and weaknesses
effects, should be considered in that affect interpretation of results, as for identifying unique features, such as
assessing the totality of the human summarized below (Lilienfeld and an association with an uncommon
evidence. Conclusions about the overall Lilienfeld, 1979; Mausner and Kramer, tumor (e.g., inhalation of vinyl chloride
evidence for carcinogenicity from and hepatic angiosarcoma in workers or
1985; Kelsey et al., 1996; Rothman and
available studies in humans should be ingestion of diethylstilbestrol by
Greenland, 1998).
summarized along with a discussion of Analytical epidemiologic studies, mothers and clear-cell carcinoma of the
uncertainties and gaps in knowledge. which include case-control and cohort vagina in offspring).
Conclusions regarding the strength of designs, are generally relied on for 2.2.1.3. Exposure Issues
the evidence for positive or negative identifying a causal association between For epidemiologic data to be useful in
associations observed, as well as human exposure and adverse health determining whether there is an
evidence supporting judgments of effects. In case-control studies, groups of association between health effects and
causality, should be clearly described. individuals with (cases) and without exposure to an agent, there should be
In assessing the human data within the (controls) a particular disease are adequate characterization of exposure
overall weight of evidence, identified and compared to determine information. In general, greater weight
determination about the strength of the differences in exposure. In cohort should be given to studies with more
epidemiologic evidence should clearly studies, a group of ‘‘exposed’’ and precise and specific exposure estimates.
identify the degree to which the ‘‘nonexposed’’ individuals are identified Questions to address about exposure
observed associations may be explained and studied over time to determine are: What can one reliably conclude
by other factors, including bias or differences in disease occurrence. about the exposure parameters
confounding. Cohort studies can be performed either including (but not limited to) the level,
prospectively or retrospectively from duration, route, and frequency of
Characteristics that are generally historical records. The type of study exposure of individuals in one
desirable in epidemiologic studies chosen may depend on the hypothesis population as compared with another?
include (1) Clear articulation of study to be evaluated. For example, case- How sensitive are study results to
objectives or hypothesis; (2) proper control studies may be more appropriate uncertainties in these parameters?
selection and characterization of for rare cancers while cohort studies Actual exposure measurements are
comparison groups (exposed and may be more appropriate for more not available for many retrospective
unexposed groups or case and control commonly occurring cancers. studies. Therefore, surrogates are often
groups); (3) adequate characterization of On the other hand, descriptive used to reconstruct exposure
exposure; (4) sufficient length of follow- epidemiologic studies examine parameters. These may involve
up for disease occurrence; (5) valid symptom or disease rates among attributing exposures to job
ascertainment of the causes of cancer populations in relation to personal classifications in a workplace or to
morbidity and mortality; (6) proper characteristics such as age, gender, race, broader occupational or geographic
consideration of bias and confounding and temporal or environmental groupings. Use of surrogates carries a
factors; (7) adequate sample size to conditions. Descriptive studies are most potential for misclassification, i.e.,
detect an effect; (8) clear, well- frequently used to generate hypotheses individuals may be placed in an
documented, and appropriate about exposure factors, but subsequent incorrect exposure group.
methodology for data collection and analytical designs are necessary to infer Misclassification generally leads to
analysis; (9) adequate response rate and causality. For example, cross-sectional reduced ability of a study to detect
methodology for handling missing data; designs might be used to compare the differences between study and referent
and (10) complete and clear prevalence of cancer between areas near populations.
documentation of results. No single and far from a Superfund site. However, When either current or historical
criterion determines the overall in studies where exposure and disease monitoring data are available, the
information applies only to the current exposure evaluation includes
adequacy of a study. Practical and
conditions, it is not possible to infer that consideration of the error bounds of the
resource constraints may limit the
the exposure actually caused the monitoring and analytic methods and
ability to address all of these
disease. Therefore, these studies are whether the data are from routine or
characteristics in a study. The risk used to identify patterns or trends in
assessor is encouraged to consider how accidental exposures. The potential for
disease occurrence over time or in misclassification and for measurement
the limitations of the available studies different geographical locations, but errors is amenable to both qualitative
might influence the conclusions. While typical limitations in the and quantitative analysis. These are
positive biases may be due, for example, characterization of populations in these essential analyses for judging a study’s
to a healthy worker effect, it is also studies make it difficult to infer the results, because exposure estimation is
important to consider negative biases, causal agent or degree of exposure. the most critical part of a retrospective
for example, workers who may leave the Case reports describe a particular study.
workforce due to illness caused either effect in an individual or group of
by high exposures to the agent or to individuals who were exposed to a 2.2.1.4. Biological Markers
effects of confounders such as smoking. substance. These reports are often Biological markers potentially offer
The following discussions highlight the anecdotal or highly selective in nature excellent measures of exposure (Hulka
major factors included in an analysis of and generally are of limited use for and Margolin, 1992; Peto and Darby,
epidemiologic studies. hazard assessment. Specifically, cancer 1994). In some cases, molecular or
cellular effects (e.g., DNA or protein the results (stratification or direct or 2.2.1.6. Statistical Considerations
adducts, mutation, chromosomal indirect adjustment). Direct adjustment The analysis should apply
aberrations, levels of thyroid- in the statistical analysis may not be appropriate statistical methods to
stimulating hormone) can be measured possible owing to the presentation of the ascertain whether the observed
in blood, body fluids, cells, and tissues data or because needed information was association between exposure and
to serve as biomarkers of exposure in not collected during the study. In this effects would be expected by chance. A
humans and animals (Callemen et al., case, indirect comparisons may be description of the method or methods
1978; Birner et al., 1990). As such, they possible. For example, in the absence of used should include the reasons for
can act as an internal surrogate measure data on smoking status among their selection. Statistical analyses of
of chemical dose, representing, as individuals in the study population, an the bias, confounding, and interaction
appropriate, either recent exposure (e.g., examination of the possible contribution are part of addressing the significance of
serum concentration) or accumulated of cigarette smoking to increased lung an association and the power of a study
exposure over some period (e.g., cancer risk may be based on information to detect an effect.
hemoglobin adducts). Validated markers from other sources, such as the The analysis augments examination of
of exposure such as alkylated American Cancer Society’s longitudinal the results for the whole population
hemoglobin from exposure to ethylene studies (Hammand, 1966; Garfinkel and with exploration of the results for
oxide (Van Sittert et al., 1985) or urinary Silverberg, 1991). The effectiveness of groups with comparatively greater
arsenic (Enterline et al., 1987) can adjustments contributes to the ability to exposure or time since first exposure.
improve estimates of dose over the draw inferences from a study. This may support identifying an
relevant time periods for the markers.
Different studies involving exposure association or establishing a dose-
Markers closely identified with effects
to an agent may have different response trend. When studies show no
promise to greatly increase the ability of
confounding factors. If consistent association, such exploration may apply
studies to distinguish real effects from
bias at low levels of relative risk increases in cancer risk are observed to determining an upper limit on
between populations (Taylor et al., across a collection of studies with potential human risk for consideration
1994; Biggs et al., 1993) and to resolve different confounding factors, the alongside results of animal tumor effects
problems of confounding risk factors. inference that the agent under studies.
However, when using molecular or investigation was the etiologic factor is 2.2.1.6.1. Likelihood of Observing an
cellular effects as biomarkers of strengthened. Effect
exposure, since many of these changes There may also be instances where The power of a study—the likelihood
are often not specific to just one type of the agent of interest is a risk factor in of observing an effect if one exists—
exposure, it is important to be aware conjunction with another agent. For increases with sample size, i.e., the
that changes may be due to exposures instance, interaction as well as effect- number of subjects studied from a
unrelated to the exposure of interest and measure modification are sometimes population. (For example, a quadrupling
attention must be paid to controlling for construed to be confounding, but they of a background rate in the 1 per 10,000
potential confounders. are different than confounding.
Biochemical or molecular range would require more subjects who
Interaction is described as a situation in have experienced greater or longer
epidemiologic studies may use which two or more risk factors modify
biological markers of effect as indicators exposure or lengthier follow-up, than a
the effect of each other with regard to doubling of a background rate in the 1
of disease or its precursors. The
the occurrence of a given effect. This per 100 range.) If the size of the effect
application of techniques for measuring
phenomenon is sometimes described as is expected to be very small at low
cellular and molecular alterations due to
effect-measure modification or doses, higher doses or longer durations
exposure to specific environmental
heterogeneity of effect (Szklo and Nieto, of exposure may be needed to have an
agents may allow conclusions to be
2000). Effect-measure modification appreciable likelihood of observing an
drawn about the mechanisms of
carcinogenesis (see section 2.4 for more refers to variation in the magnitude of effect with a given sample size. Because
information on this topic). measure exposure effect across levels of of the often long latency period in
another variable (Rothman and cancer development, the likelihood of
2.2.1.5. Confounding Factors Greenland, 1998). The variable across observing an effect also depends on
Control for potential confounding which the effect measure varies and is whether adequate time has elapsed
factors is an important consideration in called an effect modifier (e.g., hepatitis since exposure began for effects to
the evaluation of the design and in the virus B and aflatoxin in hepatic cancer). occur. Since the design of the study and
analysis of observational epidemiologic Interaction, on the other hand, means the choice of analysis, as well as the
studies. A confounder is a variable that effect of the exposure on the outcome design level of certainty in the results
is related to both the health outcome of differs, depending on the presence of and the magnitude of response in an
concern (cancer) and exposure. another variable (the effect modifier). unexposed population also affect the
Common examples include age, When the effect of the exposure of likelihood of observing an effect, it is
socioeconomic status, smoking habits, interest is accentuated by another important to carefully interpret the
and diet. For instance, if older people variable, it is said to be synergistic absence of an observed effect. A unique
are more likely to be exposed to a given interaction. Synergistic interaction can feature that can be ascribed to the effects
contaminant as well as more likely to be additive (e.g., hepatitis virus B and of a particular agent (such as a tumor
have cancer because of their age, age is aflatoxin in hepatic cancer) or type that is seen only rarely in the
considered a confounder. Adjustment multiplicative (e.g., asbestos and absence of the agent) can increase
for potentially confounding factors smoking in lung cancer). If the effect of sensitivity by permitting separation of
(from a statistical as contrasted with an exposure is diminished or eliminated by bias and confounding factors from real
epidemiologic point of view) can occur another variable, it said to be effects. Similarly, a biomarker particular
either in the design of the study (e.g., antagonistic interaction (e.g., intake of to the agent can permit these
individual or group matching on critical vitamin E and lower occurrence of lung distinctions. Statistical re-analyses of
factors) or in the statistical analysis of cancer). data, particularly an examination of
different exposure indices, can give the effect of publication bias. Meta- robustness of the effects associations.
insight into potential exposure-response analysis may not be advantageous in Consideration of the robustness of the
relationships. These are all factors to some circumstances. These include associations takes into account a
explore in statistical analysis of the when the relationship between exposure number of factors, including in
data. and disease is obvious from the particular the impact of alternative
individual studies; when there are only models and model specifications and
2.2.1.6.2. Sampling and Other Bias
a few studies of the key health potential confounding factors, as well
Issues
outcomes; when there is insufficient issues related to the consequences of
When comparing cases and controls information from available studies measurement error. Consideration of the
or exposed and non-exposed related to disease, risk estimate, or consistency of the effects associations
populations, it would be preferable for exposure classification to insure involves looking across the results of
the two populations to differ only in comparability; or when there are studies conducted by different
exposure to the agent in question. substantial confounding or other biases investigators in different places and
Because this is seldom the case, it is that cannot be adjusted for in the times. Particular weight may be given,
important to identify sources of analysis (Blair et al., 1995; Greenland, consistent with Hill’s views, to the
sampling and other potential biases 1987; Peto, 1992). presence of ‘‘similar results reached in
inherent in a study design or data quite different ways, e.g., prospectively
collection methods. 2.2.1.7. Evidence for Causality
and retrospectively’’ (Hill, 1965).
Bias is a systematic error. In Determining whether an observed Looking beyond the epidemiological
epidemiologic studies, bias can occur in association (risk) is causal rather than evidence, evaluation of the biological
the selection of cases and controls or spurious involves consideration of a plausibility of the associations observed
exposed and non-exposed populations, number of factors. Sir Bradford Hill in epidemiologic studies reflects
as well as the follow up of the groups, (Hill, 1965) developed a set of consideration of both exposure-related
or the classification of disease or guidelines for evaluating epidemiologic factors and toxicological evidence
exposure. The size of the risks observed associations that can be used in relevant to identification of potential
can be affected by noncomparability conjunction with the discussion of modes of action (MOAs). Similarly,
between populations of factors such as causality such as the 2004 Surgeon consideration of the coherence of health
general health, diet, lifestyle, or General’s report on smoking (CDC, effects associations reported in the
geographic location; differences in the 2004) and in other documents (e.g., epidemiologic literature reflects broad
way case and control individuals recall Rothman and Greenland 1998; IPCS, consideration of information pertaining
past events; differences in data 1999). The critical assessment of to the nature of the biological markers
collection that result in unequal epidemiologic evidence is conceptually evaluated in toxicologic and
ascertainment of health effects in the based upon consideration of salient epidemiologic studies.
populations; and unequal follow-up of aspects of the evidence of associations In identifying these aspects as being
individuals (Rothman and Greenland, so as to reach fundamental judgments as particularly salient in this assessment, it
1998). Other factors worth consideration to the likely causal significance of the is also important to recognize that no
can be inherent in the available cohorts, observed associations. In so doing, it is one aspect is either necessary or
e.g., use of occupational studies (the appropriate to draw from those aspects sufficient for drawing inferences of
healthy worker effect), absence of one initially presented in Hill’s classic causality. As Hill (1965) emphasized:
sex, or limitations in sample size for one monograph (Hill, 1965) and widely used
None of my nine viewpoints can bring
or more ethnicities. by the scientific community in indisputable evidence for or against the
The mere presence of biases does not conducting such evidence-based cause-and-effect hypothesis and none can be
invalidate a study, but should be reviews. A number of these aspects are required as a sine qua non. What they can
reflected in the judgment of its strengths judged to be particularly salient in do, with greater or less strength, is to help
or weaknesses. Acceptance of studies evaluating the body of evidence us to make up our minds on the fundamental
for assessment depends on identifying available in this review, including the question—is there any other way of
their sources of bias and the possible aspects described by Hill as strength, explaining the set of facts before us, is there
effects on study results. experiment, consistency, plausibility, any other answer equally, or more, likely
than cause and effect?
and coherence. Other aspects identified
2.2.1.6.3. Combining Statistical While these aspects frame
by Hill, including temporality and
Evidence Across Studies biological gradient, are also relevant and considerations weighed in assessing the
Meta-analysis is a means of considered here (e.g., in characterizing epidemiologic evidence, they do not
integrating the results of multiple lag structures and concentration- lend themselves to being considered in
studies of similar health effects and risk response relationships), but are more terms of simple formulas or hard-and-
factors. This technique is particularly directly addressed in the design and fast rules of evidence leading to answers
useful when various studies yield analyses of the individual about causality (Hill, 1965). One, for
varying degrees of risk or even epidemiologic studies included in this example, cannot simply count up the
conflicting associations (negative and assessment. As discussed below, these numbers of studies reporting
positive). It is intended to introduce salient aspects are interrelated and statistically significant results or
consistency and comprehensiveness considered throughout the evaluation of statistically non-significant results for
into what otherwise might be a more the epidemiologic evidence generally carcinogenesis and related MOAs and
subjective review of the literature. The reflected in the integrative synthesis of reach credible conclusions about the
value of such an analysis is dependent the mode of action framework. relative strength of the evidence and the
upon a systematic review of the The general evaluation of the strength likelihood of causality. Rather, these
literature that uses transparent criteria of the epidemiological evidence reflects important considerations are taken into
of inclusion and exclusion. In consideration not only of the magnitude account throughout the assessment with
interpreting such analyses, it is of reported effects estimates and their a goal of producing an objective
important to consider the effects of statistical significance, but also of the appraisal of the evidence (informed by
differences in study quality, as well as precision of the effects estimates and the peer and public comment and advice),
which includes the weighing of cause and effect, especially when such 1990). Other studies of special design
alternative views on controversial relationships are also observed for are useful for observing formation of
issues. Thus, although these guidelines duration of exposure (e.g., increasing preneoplastic lesions or tumors or
have become known as ‘‘causal effects observed following longer investigating specific modes of action.
criteria,’’ it is important to note that exposure times). There are many Their applicability is determined on a
they cannot be used as a strictly possible reasons that an epidemiologic case-by-case basis.
quantitative checklist. Rather, these study may fail to detect an exposure-
2.2.2.1. Long-Term Carcinogenicity
‘‘criteria’’ should be used to determine response relationship. For example, an
Studies
the strength of the evidence for analysis that included decreasing
concluding causality. In particular, the exposures due to improved technology The objective of long-term
absence of one or more of the ‘‘criteria’’ that is combined with higher prior carcinogenesis bioassays is to determine
does not automatically exclude a study exposure in an initial analysis can the potential carcinogenic hazard and
from consideration (e.g., see discussion require a segmented analysis to dose-response relationships of the test
in CDC, 2004). The list below has been apportion exposure. Other reasons for agent. Carcinogenicity rodent studies
adapted from Hill’s guidelines as an aid failure to detect a relationship may are designed to examine the production
in judging causality. include a small range of exposures. of tumors as well as preneoplastic
(a) Consistency of the observed Thus, the absence of an exposure- lesions and other indications of chronic
association. An inference of causality is response relationship does not exclude toxicity that may provide evidence of
strengthened when a pattern of elevated a causal relationship. treatment-related effects and insights
risks is observed across several (f) Biological plausibility. An into the way the test agent produces
independent studies. The inference of causality tends to be tumors. Current standardized
reproducibility of findings constitutes strengthened by consistency with data carcinogenicity studies in rodents test at
one of the strongest arguments for from experimental studies or other least 50 animals per sex per dose group
causality. If there are discordant results sources demonstrating plausible in each of three treatment groups and in
among investigations, possible reasons biological mechanisms. A lack of a concurrent control group, usually for
such as differences in exposure, mechanistic data, however, is not a 18 to 24 months, depending on the
confounding factors, and the power of reason to reject causality. rodent species tested (OECD, 1981; U.S.
the study are considered. (g) Coherence. An inference of EPA, 1998c). The high dose in long-term
(b) Strength of the observed causality may be strengthened by other studies is generally selected to provide
association. The finding of large, precise lines of evidence that support a cause- the maximum ability to detect
risks increases confidence that the and-effect interpretation of the treatment-related carcinogenic effects
association is not likely due to chance, association. Information is considered while not compromising the outcome of
bias, or other factors. A modest risk, from animal bioassays, toxicokinetic the study through excessive toxicity or
however, does not preclude a causal studies, and short-term studies. The inducing inappropriate toxicokinetics
association and may reflect a lower level absence of other lines of evidence, (e.g., overwhelming absorption or
of exposure, an agent of lower potency, however, is not a reason to reject detoxification mechanisms). The
or a common disease with a high causality. purpose of two or more lower doses is
background level. (h) Experimental evidence (from to provide some information on the
(c) Specificity of the observed human populations). Experimental shape of the dose-response curve.
association. As originally intended, this evidence is seldom available from Similar protocols have been and
refers to increased inference of causality human populations and exists only continue to be used by many
if one cause is associated with a single when conditions of human exposure laboratories worldwide.
effect or disease (Hill, 1965). Based on have occurred to create a ‘‘natural All available studies of tumor effects
our current understanding that many experiment’’ at different levels of in whole animals should be considered,
agents cause cancer at multiple sites, exposure. Strong evidence for causality at least preliminarily. The analysis
and many cancers have multiple causes, can be provided when a change in should discard studies judged to be
this is now considered one of the exposure brings about a change in wholly inadequate in protocol, conduct,
weaker guidelines for causality. Thus, disease frequency, for example, the or results. Criteria for the technical
although the presence of specificity may decrease in the risk of lung cancer that adequacy of animal carcinogenicity
support causality, its absence does not follows cessation of smoking. studies have been published and should
exclude it. (i) Analogy. SARs and information on be used as guidance to judge the
(d) Temporal relationship of the the agent’s structural analogues can acceptability of individual studies (e.g.,
observed association. A causal provide insight into whether an NTP, 1984; OSTP, 1985; Chhabra et al.,
interpretation is strengthened when association is causal. Similarly, 1990). As these criteria, in whole or in
exposure is known to precede information on mode of action for a part, may be updated by the National
development of the disease. Because a chemical, as one of many structural Toxicology Program (NTP) and others,
latent period of up to 20 years or longer analogues, can inform decisions the analyst should consult the
is often associated with cancer regarding likely causality. appropriate sources to determine both
development in adults, the study should the current standards as well as those
consider whether exposures occurred 2.2.2. Animal Data that were contemporaneous with the
sufficiently long ago to produce an Various whole-animal test systems are study. Care should be taken to include
effect at the time the cancer is assessed. currently used or are under studies that provide some evidence
This is among the strongest criteria for development for evaluating potential bearing on carcinogenicity or that help
an inference of causality. carcinogenicity. Cancer studies interpret effects noted in other studies,
(e) Biological gradient (exposure- involving chronic exposure for most of even if these studies have some
response relationship). A clear the lifespan of an animal are generally limitations of protocol or conduct. Such
exposure-response relationship (e.g., accepted for evaluation of tumor effects limited, but not wholly inadequate,
increasing effects associated with (Tomatis et al., 1989; Rall, 1991; Allen studies can contribute as their
greater exposure) strongly suggests et al., 1988; but see Ames and Gold, deficiencies permit. The findings of
long-term rodent bioassays should be indicate that an adequate high dose has case, the information at hand should be
interpreted in conjunction with results been exceeded. evaluated and a rationale should be
of prechronic studies along with Other signs of treatment-related given for the position taken.
toxicokinetic studies and other toxicity associated with an excessive • Adequately high dose. If an
pertinent information, if available. high dose may include (a) Significant adequately high dose has been used,
Evaluation of tumor effects takes into reduction of body weight gain (e.g., tumor effects are judged positive or
consideration both biological and greater than 10%), (b) significant negative depending on the presence or
statistical significance of the findings increases in abnormal behavioral and absence of significant tumor incidence
(Haseman, 1984, 1985, 1990, 1995). The clinical signs, (c) significant changes in increases, respectively.
following sections highlight the major hematology or clinical chemistry, (d) • Excessively high dose. If toxicity or
issues in the evaluation of long-term saturation of absorption and mortality is excessive at the high dose,
carcinogenicity studies. detoxification mechanisms, or (e) interpretation depends on whether or
marked changes in organ weight, not tumors are found.
2.2.2.1.1. Dosing Issues morphology, and histopathology. It —Studies that show tumor effects only
Among the many criteria for technical should be noted that practical upper at excessive doses may be
adequacy of animal carcinogenicity limits have been established to avoid compromised and may or may not
studies is the appropriateness of dose the use of excessively high doses in carry weight, depending on the
selection. The selection of doses for long-term carcinogenicity studies of interpretation in the context of other
chronic bioassays is based on scientific environmental chemicals (e.g., 5% of study results and other lines of
judgments and sound toxicologic the test substance in the feed for dietary evidence. Results of such studies,
principles. Dose selection should be studies or 1 g/kg body weight for oral however, are generally not considered
made on the basis of relevant gavage studies [OECD, 1981]). suitable for dose-response
toxicologic information from For dietary studies, weight gain
extrapolation if it is determined that
prechronic, mechanistic, and reductions should be evaluated as to
the mode(s) of action underlying the
toxicokinetic and mechanistic studies. whether there is a palatability problem
tumorigenic responses at high doses is
A scientific rationale for dose selection or an issue with food efficiency;
not operative at lower doses.
should be clearly articulated (e.g., NTP, certainly, the latter is a toxic
—Studies that show tumors at lower
1984; ILSI, 1997). How well the dose manifestation. In the case of inhalation
studies with respirable particles, doses, even though the high dose is
selection is made is evaluated after the excessive and may be discounted,
completion of the bioassay. evidence of impairment of normal
clearance of particles from the lung should be evaluated on their own
Interpretation of carcinogenicity study should be considered along with other merits.
results is profoundly affected by study signs of toxicity to the respiratory —If a study does not show an increase
exposure conditions, especially by airways to determine whether the high in tumor incidence at a toxic high
inappropriate dose selection. This is exposure concentration has been dose and appropriately spaced lower
particularly important in studies that do appropriately selected (U.S. EPA, doses are used without such toxicity
not show positive results for 2001a). For dermal studies, evidence of or tumors, the study is generally
carcinogenicity, because failure to use a skin irritation may indicate that an judged as negative for carcinogenicity.
sufficiently high dose reduces the adequate high dose has been reached • Inadequately high dose. Studies of
sensitivity of the studies. A lack of (U.S. EPA, 1989). inadequate sensitivity where an
tumorigenic responses at exposure In order to obtain the most relevant adequately high dose has not been
levels that cause significant impairment information from a long-term reached may be used to bound the dose
of animal survival may also not be carcinogenicity study, it is important to range where carcinogenic effects might
acceptable. In addition, overt toxicity or maximize exposure conditions to the be expected.
qualitatively altered toxicokinetics due test material. At the same time, caution
to excessively high doses may result in 2.2.2.1.2. Statistical Considerations
is appropriate in using excessive high-
tumor effects that are secondary to the dose levels that would confound the The main aim of statistical evaluation
toxicity rather than directly attributable interpretation of study results to is to determine whether exposure to the
to the agent. humans. The middle and lowest doses test agent is associated with an increase
With regard to the appropriateness of should be selected to characterize the of tumor development. Statistical
the high dose, an adequate high dose shape of the dose-response curve as analysis of a long-term study should be
would generally be one that produces much as possible. It is important that performed for each tumor type
some toxic effects without unduly the doses be adequately spaced so that separately. The incidence of benign and
affecting mortality from effects other the study can provide relevant dose- malignant lesions of the same cell type,
than cancer or producing significant response data for assessing human usually within a single tissue or organ,
adverse effects on the nutrition and hazard and risk. If the testing of are considered separately but may be
health of the test animals (OECD, 1981; potential carcinogenicity is being combined when scientifically defensible
NRC, 1993a). If the test agent does not combined with an evaluation of (McConnell et al., 1986).
appear to cause any specific target organ noncancer chronic toxicity, the study Trend tests and pairwise comparison
toxicity or perturbation of physiological should be designed to include one dose tests are the recommended tests for
function, an adequate high dose can be in addition to the control(s) that is not determining whether chance, rather
specified in terms of a percentage expected to elicit adverse effects. than a treatment-related effect, is a
reduction of body weight gain over the There are several possible outcomes plausible explanation for an apparent
lifespan of the animals. The high dose regarding the study interpretation of the increase in tumor incidence. A trend
would generally be considered significance and relevance of test such as the Cochran-Armitage test
inadequate if neither toxicity nor change tumorigenic effects associated with (Snedecor and Cochran, 1967) asks
in weight gain is observed. On the other exposure or dose levels below, at, or whether the results in all dose groups
hand, significant increases in mortality above an adequate high dose. The together increase as dose increases. A
from effects other than cancer generally general guidance is given here; for each pairwise comparison test such as the
Fisher exact test (Fisher, 1950) asks situations prompts further thought 2.2.2.1.4. Assessment of Evidence of
whether an incidence in one dose group about the meaning of the response in the Carcinogenicity From Long-term Animal
is increased over that of the control current study in context with other Studies
group. By convention, for both tests a observations in animal studies and with In general, observation of tumors
statistically significant comparison is other evidence about the carcinogenic under different circumstances lends
one for which p is less than 0.05 that the potential of the agent. These other support to the significance of the
increased incidence is due to chance. sources of information may reinforce or
findings for animal carcinogenicity.
Significance in either kind of test is weaken the significance given to the
Significance is generally increased by
sufficient to reject the hypothesis that response in the hazard assessment.
the observation of more of the factors
chance accounts for the result. Caution should be exercised in simply
A statistically significant response listed below. For a factor such as
looking at the ranges of historical
may or may not be biologically malignancy, the severity of the observed
responses, because the range ignores
significant and vice versa. The selection pathology can also affect the
differences in survival of animals among
of a significance level is a policy choice significance. The following observations
studies and is related to the number of
based on a trade-off between the risks of add significance to the tumor findings:
studies in the database.
false positives and false negatives. A In analyzing results for uncommon • Uncommon tumor types;
result with a significance level of greater tumors in a treated group that are not • Tumors at multiple sites;
or less than 5% (the most common statistically significant in comparison • Tumors by more than one route of
significance level) is examined to see if with concurrent controls, the analyst administration;
the result confirms other scientific may be informed by the experience of • Tumors in multiple species, strains,
information. When the assessment historical controls to conclude that the or both sexes;
departs from a simple 5% level, this result is in fact unlikely to be due to • Progression of lesions from
should be highlighted in the risk chance. However, caution should be preneoplastic to benign to malignant;
characterization. A two-tailed test or a used in interpreting results. In analyzing • Reduced latency of neoplastic
one-tailed test can be used. In either results for common tumors, a different lesions;
case a rationale is provided. set of considerations comes into play. • Metastases;
Statistical power can affect the Generally speaking, statistically • Unusual magnitude of tumor
likelihood that a statistically significant significant increases in tumors should response;
result could reasonably be expected. not be discounted simply because • Proportion of malignant tumors;
This is especially important in studies incidence rates in the treated groups are and
or dose groups with small sample sizes within the range of historical controls or • Dose-related increases.
or low dose rates. Reporting the because incidence rates in the In these cancer guidelines, tumors
statistical power can be useful for concurrent controls are somewhat lower observed in animals are generally
comparing and reconciling positive and than average. Random assignment of assumed to indicate that an agent may
negative results from different studies. animals to groups and proper statistical produce tumors in humans. Mode of
Considerations of multiple procedures provide assurance that action may help inform this assumption
comparisons should also be taken into statistically significant results are on a chemical-specific basis. Moreover,
account. Haseman (1983) analyzed unlikely to be due to chance alone. the absence of tumors in well-
typical animal bioassays that tested both However, caution should be used in conducted, long-term animal studies in
sexes of two species and concluded that, interpreting results that are barely at least two species provides reasonable
because of multiple comparisons, a statistically significant or in which assurance that an agent may not be a
single tumor increase for a species-sex- incidence rates in concurrent controls carcinogenic concern for humans.
site combination that is statistically are unusually low in comparison with 2.2.2.1.5. Site Concordance
significant at the 1% level for common historical controls.
tumors or 5% for rare tumors In cases where there may be reason to Site concordance of tumor effects
corresponds to a 7–8% significance discount the biological relevance to between animals and humans should be
level for the study as a whole. humans of increases in common animal considered in each case. Thus far, there
Therefore, animal bioassays presenting tumors, such considerations should be is evidence that growth control
only one significant result that falls weighed on their own merits and clearly mechanisms at the level of the cell are
short of the 1% level for a common distinguished from statistical concerns. homologous among mammals, but there
tumor should be treated with caution. When historical control data are used, is no evidence that these mechanisms
the discussion should address several are site concordant. Moreover, agents
2.2.2.1.3. Concurrent and Historical issues that affect comparability of observed to produce tumors in both
Controls historical and concurrent control data, humans and animals have produced
The standard for determining such as genetic drift in the laboratory tumors either at the same site (e.g., vinyl
statistical significance of tumor strains, differences in pathology chloride) or different sites (e.g.,
incidence comes from a comparison of examination at different times and in benzene)(NRC, 1994). Hence, site
tumors in dosed animals with those in different laboratories (e.g., in criteria for concordance is not always assumed
concurrent control animals. Additional evaluating lesions; variations in the between animals and humans. On the
insights about both statistical and techniques for the preparation or other hand, certain modes of action with
biological significance can come from reading of tissue samples among consequences for particular tissue sites
an examination of historical control data laboratories), and comparability of (e.g., disruption of thyroid function)
(Tarone, 1982; Haseman, 1995). animals from different suppliers. The may lead to an anticipation of site
Historical control data can add to the most relevant historical data come from concordance.
analysis, particularly by enabling the same laboratory and the same
identification of uncommon tumor types supplier and are gathered within 2 or 3 2.2.2.2. Perinatal Carcinogenicity
or high spontaneous incidence of a years one way or the other of the study Studies
tumor in a given animal strain. under review; other data should be used The objective of perinatal
Identification of common or uncommon only with extreme caution. carcinogenesis studies is to determine
the carcinogenic potential and dose- (Goldsworthy et al., 1986). Others use identification (Spalding et al., 2000;
response relationships of the test agent tumor endpoints, such as the induction Gulezian et al., 2000; ILSI, 2001).
in the developing organism. Some of lung adenomas in the sensitive strain
2.2.3. Structural Analogue Data
investigators have hypothesized that the A mouse (Maronpot et al., 1986) or
age of initial exposure to a chemical tumor induction in initiation-promotion For some chemical classes, there is
carcinogen may influence the studies using various organs such as the significant available information, largely
carcinogenic response (Vesselinovitch et bladder, intestine, liver, lung, mammary from rodent bioassays, on the
al., 1979; Rice, 1979; McConnell, 1992). gland, and thyroid (Ito et al., 1992). In carcinogenicity of analogues. Analogue
Current standardized long-term these tests, the selected tissue rather effects are instructive in investigating
carcinogenesis bioassays generally begin than the whole animal is, in a sense, the carcinogenic potential of an agent as
dosing animals at 6–8 weeks of age and test system. Important information well as in identifying potential target
continue dosing for the lifespan of the concerning the steps in the carcinogenic organs, exposures associated with
animal (18–24 months). This protocol process and mode of action can be effects, and potential functional class
has been modified in some cases to obtained from ‘‘start/stop’’ experiments. effects or modes of action. All
investigate the potential of the test agent In these protocols, an agent is given for appropriate studies should be included
to induce transplacental carcinogenesis a period of time to induce particular and analyzed, whether indicative of a
or to investigate the potential lesions or effects and then stopped in positive effect or not. Evaluation
differences following perinatal and order to evaluate the progression or includes tests in various animal species,
adult exposures, but currently there is reversibility of processes (Todd, 1986; strains, and sexes; with different routes
not a standardized protocol for testing Marsman and Popp, 1994). of administration; and at various doses,
agents for carcinogenic effects following Assays in genetically engineered as data are available. Confidence in
prenatal or early postnatal exposure. rodents may provide insight into the conclusions is a function of how similar
Several cancer bioassay studies have chemical and gene interactions involved the analogues are to the agent under
compared adult and perinatal exposures in carcinogenesis (Tennant et al., 1995). review in structure, metabolism, and
(see McConnell, 1992; U.S. EPA, 1996b). These mechanistically based approaches biological activity. It is important to
A review of these studies reveals that involve activated oncogenes that are consider this confidence to ensure a
perinatal exposure rarely identifies introduced (transgenic) or tumor balanced position.
carcinogens that are not found in suppressor genes that are deleted
(knocked out). If appropriate genes are 2.3. Analysis of Other Key Data
standard animal bioassays. Exposure
that is perinatal can increase the selected, not only may these systems The physical, chemical, and structural
incidence of a given type of tumor. The provide information on mechanisms, properties of an agent, as well as data on
increase may reflect an increased length but the rodents typically show tumor endpoints that are thought to be critical
of exposure and a higher dose for the development earlier than in the elements of the carcinogenic process,
developing organism relative to the standard bioassay. Transgenic provide valuable insights into the
adult or an increase in susceptibility in mutagenesis assays also represent a likelihood of human cancer risk. The
some cases. Additionally, exposure that mechanistic approach for assessing the following sections provide guidance for
is perinatal through adulthood mutagenic properties of agents as well analyses of these data.
sometimes reduces the latency period as developing quantitative linkages
between exposure, internal dose, and 2.3.1. Physicochemical Properties
for tumors to develop in the growing
organism (U.S. EPA, 1996b). EPA mutation related to tumor induction Physicochemical properties affect an
evaluates the usefulness of perinatal (Morrison and Ashby, 1994; Sisk et al., agent’s absorption, tissue distribution
studies on an agent-by-agent basis (e.g., 1994; Hayward et al., 1995). (bioavailability), biotransformation, and
U.S. EPA, 1997a, b). The support that these studies give to degradation in the body and are
Perinatal study data analysis generally a determination of carcinogenicity rests important determinants of hazard
follows the principles discussed above on their contribution to the consistency potential (and dose-response analysis).
for evaluating other long-term of other evidence about an agent. For Properties that should be analyzed
carcinogenicity studies. When instance, benzoyl peroxide has promoter include, but are not limited to,
differences in responses between activity on the skin, but the overall molecular weight, size, and shape;
perinatal animals and adult animals evidence may be less supportive (Kraus valence state; physical state (gas, liquid,
suggest an increased susceptibility of et al., 1995). These studies also may solid); water or lipid solubility, which
perinatal or postnatal animals, such as contribute information about mode of can influence retention and tissue
the ones below, a separate evaluation of action. It is important to recognize the distribution; and potential for chemical
the response should be prepared: limitations of these experimental degradation or stabilization in the body.
• A difference in dose-response protocols, such as short duration, An agent’s potential for chemical
relationship, limited histology, lack of complete reaction with cellular components,
• The presence of different tumor development of tumors, or experimental particularly with DNA and proteins, is
types, manipulation of the carcinogenic also important. The agent’s molecular
• An earlier onset of tumors, or process, that may limit their size and shape, electrophilicity, and
• An increase in the incidence of contribution to the overall assessment. charge distribution are considered in
tumors. Generally, their results are appropriate order to decide whether they would
as aids in the interpretation of other facilitate such reactions.
2.2.2.3. Other Studies toxicological evidence (e.g., rodent
Intermediate-term and acute dosing chronic bioassays), especially regarding 2.3.2. Structure-Activity Relationships
studies often use protocols that screen potential modes of action. On the basis (SARs)
for carcinogenic or preneoplastic effects, of currently available information, it is SAR analyses and models can be used
sometimes in a single tissue. Some unlikely that any of these assays, which to predict molecular properties,
protocols involve the development of are conducted for 6 months with 15 surrogate biological endpoints, and
various proliferative lesions, such as animals per group, will replace all carcinogenicity (see, e.g., Richard,
foci of alteration in the liver chronic bioassays for hazard 1998a, b; Richard and Williams, 2002;
Contrera et al., 2003). Overall, these identification of active metabolites, issues in comparing species and
analyses provide valuable initial identifying changes in distribution and formulating dose-response assessment
information on agents, they may metabolic pathway or pathways over a approaches.
strengthen or weaken concern, and they dose range, and making comparisons • Identifying changes in
are part of the weight of evidence. among different routes of exposure. toxicokinetics and metabolic pathways
Currently, SAR analysis is most useful If extensive data are available (e.g., with increases in dose. These changes
for chemicals and metabolites that are blood/tissue partition coefficients and may result in important differences
believed to initiate carcinogenesis pertinent physiological parameters of between high and low dose levels in
through covalent interaction with DNA the species of interest), physiologically disposition of the agent or generation of
(i.e., DNA-reactive, mutagenic, based toxicokinetic models can be its active forms. These studies play an
electrophilic, or proelectrophilic constructed to assist in a determination important role in providing a rationale
chemicals) (Ashby and Tennant, 1991). of tissue dosimetry, species-to-species for dose selection in carcinogenicity
For organic chemicals, the predictive extrapolation of dose, and route-to-route studies.
capability of SAR analysis combined extrapolation (Conolly and Andersen, • Identifying and comparing
with other toxicity information has been 1991; see Section 3.1.2). If sufficient metabolic process differences by age,
demonstrated (Ashby and Tennant, data are not available, it may be sex, or other characteristic so that
1994). The following parameters are assumed as a default that toxicokinetic susceptible subpopulations can be
useful in comparing an agent to its and metabolic processes are recognized. For example, metabolic
structural analogues and congeners that qualitatively comparable among species. capacity with respect to P450 enzymes
produce tumors and affect related Discussion of appropriate procedures in newborn children is extremely
biological processes such as receptor for quantitative, interspecies limited compared to that in adults, so
binding and activation, mutagenicity, comparisons appears in Chapter 3. that a carcinogenic metabolite formed
and general toxicity (Woo and Arcos, The qualitative question of whether through P450 activity will have limited
1989): an agent is absorbed by a particular effect in the young, whereas a
• Nature and reactivity of the route of exposure is important for carcinogenic agent deactivated through
electrophilic moiety or moieties present; weight of evidence classification, P450 activity will result in increased
• Potential to form electrophilic discussed in Section 2.5. Decisions susceptibility of this lifestage (Cresteil,
reactive intermediate(s) through about whether route of exposure is a 1998). A variety of changes in
chemical, photochemical, or metabolic limiting factor on expression of any toxicokinetics and physiology occur
activation; hazard, e.g., absorption does not occur from the fetal stage to post-weaning to
• Contribution of the carrier molecule by a specified route, are generally based young child. Any of these changes may
to which the electrophilic moiety(ies) is on studies in which effects of the agent make a difference for risk (Renwick,
attached; or its structural analogues have been 1998).
• Physicochemical properties (e.g., observed by different routes, on • Determining bioavailability via
physical state, solubility, octanol/water physical-chemical properties, or on different routes of exposure by
partition coefficient, half-life in aqueous toxicokinetics studies. analyzing uptake processes under
solution); Adequate metabolism and various exposure conditions. This
• Structural and substructural toxicokinetic data can be applied analysis supports identification of
features (e.g., electronic, stearic, toward the following, as data permit. hazards for untested routes. In addition,
molecular geometric); Confidence in conclusions is enhanced use of physicochemical data (e.g.,
• Metabolic pattern (e.g., metabolic when in vivo data are available. octanol-water partition coefficient
pathways and activation and • Identifying metabolites and reactive information) can support an inference
detoxification ratio); and intermediates of metabolism and about the likelihood of dermal
• Possible exposure route(s) of the determining whether one or more of absorption (Flynn, 1990).
agent. these intermediates is likely to be Attempts should be made in all of
Suitable SAR analysis of non-DNA- responsible for the observed effects. these areas to clarify and describe as
reactive chemicals and of DNA-reactive Information on the reactive much as possible the variability to be
chemicals that do not appear to bind intermediates focuses on SAR analysis, expected because of differences in
covalently to DNA should be based on analysis of potential modes of action, species, sex, age, and route of exposure.
knowledge or postulation of the and estimation of internal dose in dose- The analysis takes into account the
probable mode(s) of action of closely response assessment (D’Souza et al., presence of subpopulations of
related carcinogenic structural 1987; Krewski et al., 1987). individuals who are particularly
analogues (e.g., receptor mediated, • Identifying and comparing the vulnerable to the effects of an agent
cytotoxicity related). Examination of the relative activities of metabolic pathways because of toxicokinetic or metabolic
physicochemical and biochemical in animals and in humans, and at differences (genetically or
properties of the agent may then provide different ages. This analysis can provide environmentally determined) (Bois et
the rest of the information needed in insights for extrapolating results of al., 1995) and is a special emphasis for
order to make an assessment of the animal studies to humans. assessment of risks to children.
likelihood of the agent’s activity by that • Describing anticipated distribution
within the body and possibly identifying 2.3.4. Toxicological and Clinical
mode of action.
target organs. Use of water solubility, Findings
2.3.3. Comparative Metabolism and molecular weight, and structure analysis Toxicological findings in
Toxicokinetics can support qualitative inferences about experimental animals and clinical
Studies of the absorption, anticipated distribution and excretion. observations in humans are important
distribution, biotransformation, and In addition, describing whether the resources for the cancer hazard
excretion of agents permit comparisons agent or metabolite of concern will be assessment. Such findings provide
among species to assist in determining excreted rapidly or slowly or whether it information on physiological effects and
the implications of animal responses for will be stored in a particular tissue or effects on enzymes, hormones, and
human hazard assessment, supporting tissues to be mobilized later can identify other important macromolecules as well
as on target organs for toxicity. For structure-activity relationship (SAR) Endpoints indicative of DNA damage
example, given that the cancer process analyses in a weight-of-evidence but not measures of mutation per se,
represents defects in processes such as approach (Dearfield et al., 1991; U.S. such as DNA adducts or strand
terminal differentiation, growth control, EPA, 1986b; Waters et al., 1999). Key breakage, may be detected in relevant
and cell death, developmental studies of data for a mutagenic mode of action may target tissues and thus contribute to
agents may provide an understanding of be evidence that the carcinogen or a evaluating an agent’s mutagenic
the activity of an agent that carries over metabolite is DNA-reactive and/or has potential. Evidence of chemical-specific
to cancer assessment. Toxicity studies the ability to bind to DNA. Also, DNA adducts (e.g., reactions at oxygen
in animals by different routes of mutagenic carcinogens usually produce sites in DNA bases or with ring
administration support comparison of positive effects in multiple test systems nitrogens of guanine and adenine)
absorption and metabolism by those for different genetic endpoints, provides information on a mutagen’s
routes. Data on human variability in particularly gene mutations and ability to directly interact with DNA (La
standard clinical tests may also provide structural chromosome aberrations, and and Swenberg, 1996). Some planar
insight into the range of human in tests performed in vivo which molecules (e.g., 9-aminoacridine)
susceptibility and the common generally are supported by positive tests intercalate between base pairs of DNA,
mechanisms of agents that affect the in vitro. Additionally, carcinogens may which results in a physical distortion in
tested parameters. be identified as operating via a DNA that may lead to mutations when
mutagenic mode of action if they have DNA replicates. As discussed below,
2.3.5. Events Relevant to Mode of
similar properties and SAR to some carcinogens do not interact
Carcinogenic Action
mutagenic carcinogens. Endpoints that directly with DNA, but they can
Knowledge of the biochemical and provide insight into an agent’s ability to produce increases in endogenous levels
biological changes that precede tumor alter gene products and gene expression, of DNA adducts (e.g., 8-
development (which include, but are together with other features of an agent’s hydroxyguanine) by indirect
not limited to, mutagenesis, increased potential mode of carcinogenic action, mechanisms.
cell proliferation, inhibition of are discussed below.
programmed cell death, and receptor 2.3.5.2. Indirect DNA Effects or Other
activation) may provide important 2.3.5.1. Direct DNA-Reactive Effects Effects on Genes/Gene Expression
insight for determining whether a It is well known that many Although some carcinogens may
cancer hazard exists and may help carcinogens are electrophiles that result in an elevation of mutations or
inform appropriate consideration of the interact with DNA, resulting in DNA cytogenetic anomalies, as detected in
dose-response relationship below the adducts and breakage (referred to in standard assays, they may do so by
range of observable tumor response. these cancer guidelines as direct DNA indirect mechanisms. These effects may
Because cancer can result from a series effects). Usually during the process of be brought about by chemical-cell
of genetic alterations in the genes that DNA replication, these DNA lesions can interactions rather than by the chemical
control cell growth, division, and be converted into and fixed as (or its metabolite) directly interacting
differentiation (Vogelstein et al., 1988; mutations and chromosomal alterations with DNA. An increase in mutations
Hanahan and Weinberg, 2000; Kinzler that then may initiate and otherwise might be due to cytotoxic exposures
and Vogelstein, 2002), the ability of an contribute to the carcinogenic process causing regenerative proliferation or to
agent to affect genotype (and hence gene (Shelby and Zeiger, 1990; Tinwell and mitogenic influences (Cohen and
products) or gene expression is of Ashby, 1991; IARC, 1999). Thus, studies Ellwein, 1990). Increased cell division
obvious importance in evaluating its of mutations and other genetic lesions may elevate mutation by clonal
influence on the carcinogenic process. continue to inform the assessment of expansion of initiated cells or by
Initial and key questions to examine are: potential human cancer hazard and in increasing the number of genetic errors
Does the agent (or its metabolite) the understanding of an agent’s mode of by rapid cell division and reduced time
interact directly with DNA, leading to carcinogenic action. for DNA repair. Some agents might
mutations that bring about changes in EPA has published testing guidelines result in an elevation of mutations by
gene products or gene expression? Does for detecting the ability of an agent to interfering with the enzymes involved
the agent bring about effects on gene damage DNA and produce mutations in DNA repair and recombination
expression via other nondirect DNA and chromosomal alterations (as (Barrett and Lee, 1992). Damage to
interaction processes? discussed in Dearfield et al., 1991). certain critical DNA repair genes or
Furthermore, carcinogenesis involves Briefly, standard tests for gene other genes (e.g., the p53 gene) may
a complex series and interplay of events mutations in bacteria and mammalian result in genomic instability, which
that alter the signals a cell receives from cells in vitro and in vivo and for predisposes cells to further genetic
its extracellular environment, thereby structural chromosomal aberrations in alterations and increases the probability
promoting uncontrolled growth. Many, vitro and in vivo are important examples of neoplastic progression (Harris and
but not all, mutagens are carcinogens, of relevant methods. New molecular Hollstein, 1993; Levine et al., 1994;
and some, but not all, agents that induce approaches, such as mouse mutations Rouse and Jackson, 2002). Likewise,
cell proliferation lead to tumor and cancer transgenic models, are DNA repair processes may be saturated
development. Thus, understanding the providing a means to examine mutation at certain doses of a chemical, leading
range of key steps in the carcinogenic at tissue sites where the tumor response to an elevation of genetic alterations.
process upon which an agent might act is observed (Heddle and Swiger, 1996; The initiation of programmed cell
is essential for evaluating its mode of Tennant et al., 1999). Additionally, death (apoptosis) can potentially be
action. Determination of carcinogens continued improvements in fluorescent- blocked by an agent, thereby permitting
that are operating by a mutagenic mode based chromosome staining methods replication of cells carrying genetic
of action, for example, entails (fluorescent in situ hybridization errors that would normally be removed
evaluation of in vivo or in vitro short- [FISH]) will allow the detection of from the proliferative pool. At certain
term testing results for genetic specific chromosomal abnormalities in doses an agent may also generate
endpoints, metabolic profiles, relevant target tissues (Tucker and reactive oxygen species that produce
physicochemical properties, and Preston, 1998). oxidative damage to DNA and other
macromolecules (Chang et al. 1988; involved, to measure effects in both • The spectrum of genetic changes in
Kehrer, 1993; Clayson et al., 1994). The normal and neoplastic tissue, to proliferative lesions and tumors
role of cellular alterations that are distinguish between apoptosis and following chemical administration to
attributable to oxidative damage in necrosis, and to determine the dose that experimental animals can be
tumorigenesis (e.g., 8-hydroxyguanine) affects these processes. Gap-junctional determined and compared with that in
is currently unclear. intercellular communication is believed spontaneous tumors in control animals,
Several carcinogens have been shown to play a role in tissue and organ in animals exposed to other agents of
to induce aneuploidy (the loss or gain development and in the maintenance of varying structural and functional
of chromosomes) (Barrett, 1992; Gibson a normal cellular phenotype within activities, and in persons exposed to the
et al., 1995). Aneuploidy can result in tissues. A growing body of evidence agent under study.
the loss of heterozygosity or genomic suggests that chemical interference with • Biomarkers of effect and/or
instability (Cavenee et al., 1986; Fearon gap-junctional intercellular precursors may help to identify
and Vogelstein, 1990). Agents that cause communication is a contributing factor subpopulations of individuals who may
aneuploidy typically interfere with the in tumor development (Swierenga and be at an elevated risk for a certain
normal process of chromosome Yamasaki, 1992; Yamasaki, 1995). cancer or exposure to a certain agent,
segregation by interacting with non- e.g., cytochrome P450 2D6/debrisoquine
DNA targets such as the proteins needed 2.3.5.3. Precursor Events and Biomarker sensitivity for lung cancer (Caporaso et
for chromosome segregation and Information al., 1989) or inherited colon cancer
chromosome movement. Whether this Most testing schemes for mutagenicity syndromes (Kinzler et al., 1991;
chromosome imbalance is the cause or and other short-term assays were Peltomäki et al., 1993).
the effect of tumorigenesis is not clear. designed for hazard identification • As with biomarkers of exposure, it
Thus, it is important to understand if purposes; thus, these assays are may be justified in some cases to use
the agent induces aneuploidy as a key generally conducted using acute biomarkers of effect and/or precursors
early event in the carcinogenic process. exposures. For data on ‘‘precursor for dose-response assessment or to
It is possible for an agent to alter gene steps’’ to be useful in informing the provide insight into the potential shape
expression by transcriptional, dose-response curve for tumor of the dose-response curve at doses
translational, or post-translational induction below the level of below those at which tumors are
modifications. For example, observation, it is often useful for data to induced experimentally.
perturbation of DNA methylation come from in vivo studies and from In applying biomarker data to cancer
patterns may cause effects that studies where exposure is repeated or assessment an assessment should
contribute to carcinogenesis (Jones, given over an extended period of time. consider:
1986; Holliday, 1987; Goodman and Although consistency of results across • Analytical methodology,
Counts, 1993; Chuang et al., 1996; different assays and animal models • Routes of exposure,
Baylin and Bestor, 2002). provides a stronger basis for drawing • Exposure to mixtures,
Overexpression of genes by DNA conclusions, it is desirable to have data • Time after exposure,
amplification has been observed in on the precursor event in the same • Sensitivity and specificity of
certain tumors (Vainio et al., 1992). target organ, sex, animal strain, and biomarkers, and
species as the tumor data. In evaluating • Dose-response relationships.
Mechanisms of altering gene expression
may involve cellular reprogramming an agent’s mode of action, it is usually 2.3.5.4. Judging Data
through hormonal or receptor-mediated not sufficient to determine that some Criteria that are generally applicable
mechanisms (Barrett, 1992; Ashby et al., event commences upon dosing. It is for judging the adequacy of
1994). important to understand whether it is a mechanistically based data include:
Both cell proliferation and necessary event that plays a key role in • Mechanistic relevance of the data to
programmed cell death can be part of the process that leads to tumor carcinogenicity,
the maintenance of homeostasis in development versus an effect of the • Number of studies of each
many normal tissues, and alterations in cancer process itself or simply an endpoint,
the level or rate of either can be associated event. • Consistency of results in different
important elements of the carcinogenic Various endpoints can serve as test systems and different species,
process. The balance between the two biological markers of effects in • Similar dose-response relationships
can directly affect the survival and biological systems or samples. These for tumor and mode of action-related
growth of initiated cells as well as may help identify doses at which effects,
preneoplastic and tumor cell elements of the carcinogenic process are • Conduct of the tests in accordance
populations (i.e., increase in cell operating; aid in interspecies with generally accepted protocols, and
proliferation or decrease in cell death) extrapolations when data are available • Degree of consensus and general
(Moolgavkar, 1986; Cohen and Ellwein, from both experimental animal and acceptance among scientists regarding
1990, 1991; Cohen et al., 1991; Bellamy human cells; and under certain interpretation of the significance and
et al., 1995). Thus, measurements of circumstances, provide insights into the specificity of the tests.
these events can contribute to the possible shape of the dose-response Although important information can
weight of the evidence for cancer hazard curve below levels where tumor be gained from in vitro test systems, a
prediction and to mode of action incidences are observed (e.g., Choy, higher level of confidence is generally
understanding. In studies of 1993). given to data that are derived from in
proliferative effects, distinctions should Genetic and other findings (such as vivo systems, particularly those results
be made between mitogenesis and changes in proto-oncogenes and tumor that show a site concordance with the
regenerative proliferation (Cohen and suppressor genes in preneoplastic and tumor data.
Ellwein, 1990, 1991; Cohen et al., 1991). neoplastic tissue or, possibly, measures It is important to remember that when
In applying information from studies of endocrine disruption) can indicate judging and considering the use of any
on cell proliferation and apoptosis to the potential for disease and, as such, data, the basic standard of quality, as
risk assessment, it is important to serve as biomarkers of effect. They, too, defined by the EPA Information Quality
identify the tissues and target cells can be used in different ways. Guidelines, should be satisfied.
2.4. Mode of Action—General • Similarity of metabolic activation nonmutagenic influences (Ashby and
Considerations and Framework for and detoxification for a specific Tennant, 1991; Huff et al., 1991).
Analysis chemical between humans and tested The types of data and their influence
species; on judgments regarding mode of action
2.4.1. General Considerations • Influence of route of exposure on are expected to evolve, both as science
The interaction between the biology the spectrum of tumors and whether advances and as the risk assessment
of the organism and the chemical they occur at point of exposure or community gains more experience with
properties of the agent determine systemic sites; these analyses. This section contains a
whether there is an adverse effect. Thus, • Effect of high dose exposures on the framework for evaluating hypothesized
mode of action analysis is based on target organ or systemic toxicity that mode(s) of action. This framework has
physical, chemical, and biological may not reflect typical physiological similarities to and differences with the
information that helps to explain key conditions, for example, urinary concepts presented in other MOA
events in an agent’s influence on chemical changes associated with stone frameworks (e.g., IPCS, 1999; Sonich-
development of tumors. The entire formation, effects on immune Mullin et al., 2001; Meek et al., 2003).
range of information developed in the surveillance; Differences are often due to the context
assessment is reviewed to arrive at a • Presence of proliferative lesions, for of the use for the framework. For
reasoned judgment. An agent may work example, hepatic foci, or hyperplasia; example, the Meek et al. (2003) presents
by more than one mode of action, both • Effect of dose and time on the a stand-alone document for addressing
at different sites and at the same tumor progression of lesions from mode of action issues; thus, it
site. Thus the mode of action and preneoplastic to benign tumors, then to recommends that conclusions
human relevance cannot necessarily be malignant; concerning MOA be rendered
generalized to other toxic endpoints or • Ratio of malignant to benign tumors separately. In these cancer guidelines,
tissues or cell types without additional as a function of dose and time; however, they are incorporated into the
analyses (IPCS, 1999; Meek et al., 2003). • Time of appearance of tumors after context of all of the data regarding
At least some information bearing on commencing exposure; weight of the evidence for
mode of action (e.g., SAR, screening • Development of tumors that invade carcinogenicity.
tests for mutagenicity) is present for locally or systemically, or lead to death;
most agents undergoing assessment of • Tumors at organ sites with high or 2.4.2. Evaluating an Hypothesized Mode
low background historical incidence in of Action
carcinogenicity, even though certainty
about exact molecular mechanisms may laboratory animals; 2.4.2.1. Peer Review
be rare. • Biomarkers in tumor cells, both
induced and spontaneous, for example, In reaching conclusions, the question
Information for mode of action of ‘‘general acceptance’’ of a mode of
DNA or protein adducts, mutation
analysis generally includes tumor data action should be tested as part of the
spectra, chromosome changes, oncogene
in humans and animals and among independent peer review that EPA
activation; and/or
structural analogues, as well as the other obtains for its assessment and
• Shape of the dose-response curve in
key data. The more complete the data conclusions. In some cases the mode of
the range of tumor observation, for
package and the generic knowledge action may already have been
example, linear versus nonlinear.
about a given mode of action, the more Some of the myriad ways in which established by development of a large
confidence one has and the more one information from chronic animal studies body of research information and
can rely on assessment of available data influences mode of action judgments characterization of the phenomenon
rather than reverting to default options include, but are not limited to, the over time. In some cases there will have
to address the absence of information on following: been development of an Agency policy
mode of action. Reasoned judgments are • Multisite and multispecies tumor (e.g., mode of action involving alpha-2u-
generally based on a data-rich source of effects that are often associated with globulin in the male rat [U.S. EPA,
chemical, chemical class, and tumor mutagenic agents; 1991b]) or a series of previous
type-specific information. Many times • Tumors restricted to one sex or assessments in which both the mode of
there will be conflicting data and gaps species suggesting an influence action and its applicability to particular
in the information base; it is important restricted to gender, strain, or species; cases has been explored. If so, the
to carefully evaluate these uncertainties • Late onset of tumors that are assessment and its peer review can
before reaching any conclusion. primarily benign, are at sites with a high focus on the evidence that a particular
In making decisions about potential historical background incidence, or agent acts in this mode. The peer review
modes of action and the relevance of show reversal of lesions on cessation of should also evaluate the strengths and
animal tumor findings to humans exposure suggesting a growth-promoting weaknesses of competing modes of
(Ashby et al., 1990; Ashby and Tennant, mode of action; action.
1991; Tennant, 1993; IPCS 1999; • The possibility that an agent acting In other cases, the mode of action may
Sonich-Mullin et al., 2001; Meek et al., differently in different tissues; or not have previously been the subject of
2003), very often the results of chronic • The possibility that has more than an Agency document. If so, the data to
animal studies may give important one mode of action in a single tissue. support both the mode of action and the
clues. Some of the important factors to Simple knowledge of sites of tumor associated activity of the agent should
review include: increase in rodent studies can give undergo EPA assessment and
• Tumor types, for example, those preliminary clues as to mode of action. subsequent peer review.
responsive to endocrine influence or Experience at the National Toxicology
those produced by DNA-reactive Program (NTP) indicates that substances 2.4.2.2. Use of the Framework
carcinogens; that are DNA reactive and that produce The framework supports a full
• Number of studies and of tumor gene mutations may be unique in analysis of mode of action information,
sites, sexes, and species affected or producing tumors in certain anatomical but it can also be used as a screen to
unaffected in those studies and if the sites, whereas tumors at other sites may decide whether sufficient information is
data present a coherent story; arise from both mutagenic or available to evaluate or whether the data
gaps are too substantial to justify further critical for induction of tumors. It is not hypothesized mode(s) of action in
analysis. Mode of action conclusions are generally expected that the complete animals and humans (see Section
used to address the question of human sequence will be known at the 2.4.3.2). The possibility of other modes
relevance of animal tumor responses, to molecular level. Instead, empirical of action also should be considered and
address differences in anticipated observations made at different levels of discussed (see Section 2.4.3.3); if there
response among humans, such as biological organization—biochemical, is evidence for more than one mode of
between children and adults or men and cellular, physiological, tissue, organ, action, each should receive a separate
women; and as the basis of decisions and system—are analyzed. analysis. Conclusions about each
about the anticipated shape of the dose- Several important points should be hypothesized mode of action should
response relationship. Guidance on the considered when working with the address whether the mode of action is
latter appears in Section 3. framework: supported in animals and is relevant to
This framework is intended to • The topics listed for analysis should humans and which populations or
provide an analytical approach for not be regarded as a checklist of lifestages can be particularly susceptible
evaluating the mode of action. It is necessary ‘‘proofs.’’ The judgment of (see Section 2.4.3.4). In a risk
neither a checklist nor a list of required whether an hypothesized mode of assessment document, the analysis of an
criteria. As the type and amount of action is supported by available data hypothesized mode of action can be
information will depend on the mode of takes account of the analysis as a whole. presented before or with the
action postulated, scientific judgment is • The framework provides a structure characterization of an agent’s potential
important to determine if the weight of for organizing the facts upon which hazard to humans.
evidence is sufficient. conclusions as to mode of action rest.
The purpose of using the framework is 2.4.3.1. Description of the Hypothesized
2.4.3. Framework for Evaluating Each to make analysis transparent and to Mode of Action
Hypothesized Carcinogenic Mode of allow the reader to understand the facts
Action Summary description of the
and reasoning behind a conclusion. hypothesized mode of action. For each
This framework is intended to be an • The framework does not dictate an tumor site, the mode of action analysis
analytic tool for judging whether answer. The weight of evidence that is begins with a description of the
available data support a mode of sufficient to support a decision about a hypothesized mode of action and its
carcinogenic action hypothesized for an mode of action may be less or more, sequence of key events. If there is
agent. It is based upon considerations depending on the purpose of the evidence for more than one mode of
for causality in epidemiologic analysis, for example, screening, action, each receives a separate analysis.
investigations originally articulated by research needs identification, or full risk
Hill (1965) but later modified by others Identification of key events. In order
assessment. To make the reasoning
and extended to experimental studies. to judge how well data support
transparent, the purpose of the analysis
The original Hill criteria were applied to involvement of a key event in
should be made apparent to the reader.
epidemiologic data, whereas this • Toxicokinetic studies may carcinogenic processes, the
framework is applied to a much wider contribute to mode of action analysis by experimental definition of the event or
assortment of experimental data, so it contributing to identifying the active events should be clear and reproducible.
retains the basic principles of Hill but form(s) of an agent that is central to the To support an association, experiments
is much modified in content. mode of action. Apart from contributing should define and measure an event
The modified Hill criteria can be in this way, toxicokinetics studies may consistently.
useful for organizing thinking about reveal effects of saturation of metabolic • Can a list of events be identified
aspects of causation, and they are processes. These may not be considered that are key to the carcinogenic process?
consistent with the scientific method of key events in a mode of action, but they • Are the events well defined?
developing hypotheses and testing those are given separate consideration in Pertinent observations may include,
hypotheses experimentally. During assessing dose metrics and potential but are not limited to, receptor-ligand
analysis by EPA, and as guidance for nonlinearity of the dose-response changes, cytotoxicity, cell cycle effects,
peer review, a key question is whether relationship. increased cell growth, organ weight
the data to support a mode of action • Generally, ‘‘sufficient’’ support is a differences, histological changes,
meet the standards generally applied in matter of scientific judgment in the hormone or other protein perturbations,
experimental biology regarding context of the requirements of the or DNA and chromosome effects.
inference of causation. decisionmaker or in the context of 2.4.3.2. Discussion of the Experimental
All pertinent studies are reviewed in science policy guidance regarding a Support for the Hypothesized Mode of
analyzing a mode of action, and an certain mode of action. Action
overall weighing of evidence is • Even when an hypothesized mode
performed, laying out the strengths, of action is supported for a described The experimental support for the
weaknesses, and uncertainties of the response in a specific tissue, it may not hypothesized mode of action should be
case as well as potential alternative explain other tumor responses observed, discussed from several viewpoints
positions and rationales. Identifying which should get separate consideration patterned after the Hill criteria (see
data gaps and research needs is also part in hazard and dose-response Section 2.2.1.7). For illustration, the
of the assessment. assessment. explanation of each topic includes
To evaluate whether an hypothesized For each tumor site being evaluated, typical questions to be addressed to the
mode of action is operative, an analysis the mode of action analysis should available empirical data and
starts with an outline of the scientific begin with a description of the relevant experimental observations anticipated
findings regarding the hypothesized key data and key events that may be to be pertinent. The latter will vary from
events leading to cancer, and then associated with an hypothesized mode case to case. For a particular mode of
weighing information to determine of action and its sequence of key events action, certain observations may be
whether there is a causal relationship (see Section 2.4.3.1). This can be established as essential in practice or
between these events and cancer followed by a discussion of various policy, for example, measures of thyroid
formation, i.e., that the effects are aspects of the experimental support for hormone levels in supporting thyroid
hormone elevation as a key event in Temporal relationship. If an event is high doses and through mutagenicity at
carcinogenesis. shown to be causally linked to lower doses where cytotoxicity may not
Strength, consistency, specificity of tumorigenesis, it will precede tumor occur.
association. A statistically significant appearance. An event may also be If there is evidence for more than one
association between events and a tumor observed contemporaneously or after mode of action, each should receive a
response observed in well-conducted tumor appearance; these observations separate analysis. There may be an
studies is generally supportive of may add to the strength of association uneven level of experimental support
causation. Consistent observations in a but not to the temporal association. for the different modes of action.
number of such studies with differing • What is the ordering of events that Sometimes this can reflect
experimental designs increase that underlie the carcinogenic process? disproportionate resources spent on
support, because different designs may • Is this ordering consistent among investigating one particular mode of
reduce unknown biases. Studies independent studies? action and not the validity or relative
showing ‘‘recovery,’’ i.e., absence or Pertinent observations include studies importance of the other possible modes
reduction of carcinogenicity when the of varying duration observing the of action. Ultimately, however, the
event is blocked or diminished, are temporal sequence of events and information on all of the modes of
particularly useful tests of the development of tumors. action should be integrated to better
association. Specificity of the Biological plausibility and coherence. understand how and when each mode
association, without evidence of other It is important that the hypothesized acts, and which mode(s) may be of
modes of action, strengthens a causal mode of action and the events that are interest for exposure levels relevant to
conclusion. A lack of strength, part of it be based on contemporaneous human exposures of interest.
consistency, and specificity of understanding of the biology of cancer
2.4.3.4. Conclusions About the
association weakens the causal to be accepted. If the body of
Hypothesized Mode of Action
conclusions for a particular mode of information under scrutiny is consistent
action. with other examples (including Conclusions about the hypothesized
• What is the level of statistical and structurally related agents) for which mode of action should address the
biological significance for each event the hypothesized mode of action is issues listed below. For those agents for
and for cancer? accepted, the case is strengthened. which the mode of action is considered
• Do independent studies and Because some modes of action can be useful for the risk assessment, the
different experimental hypothesis- anticipated to evoke effects other than weight of the evidence concerning mode
testing approaches produce the same cancer, the available toxicity database of action in animals as well as its
associations? on noncancer effects, for example, relevance for humans would be
• Does the agent produce effects other reproductive effects of certain hormonal incorporated into the weight of evidence
than those hypothesized? disturbances, can contribute to this narrative (Section 2.5).
• Is the key event associated with evaluation. (a) Is the hypothesized mode of action
precursor lesions? • Is the mode of action consistent sufficiently supported in the test
Pertinent observations include tumor with what is known about animals? Associations observed
response associated with events (site of carcinogenesis in general and for the between key events and tumors may or
action logically relates to event[s]), case specifically? may not support an inference of
precursor lesions associated with • Are carcinogenic effects and events causation. The conclusion that the agent
events, initiation-promotion studies, consistent across structural analogues? causes one or more key events that
and stop/recovery studies. • Is the database on the agent results in tumors is strengthened as
Dose-response concordance. If a key internally consistent in supporting the more aspects of causation are satisfied
event and tumor endpoints increase purported mode of action, including and weakened as fewer are satisfied.
with dose such that the key events relevant noncancer toxicities? Consistent results in different
forecast the appearance of tumors at a Pertinent observations include the experiments that test the hypothesized
later time or higher dose, a causal scientific basis for considering an mode of action build support for that
association can be strengthened. Dose- hypothesized mode of action generally, mode of action. Replicating results in a
response associations of the key event given the contemporaneous state of similar experiment does not generally
with other precursor events can add knowledge of carcinogenic processes; meaningfully strengthen the original
further strength. Difficulty arises when previous examples of data sets showing evidence, and discordant results
an event is not causal but accompanies the mode of action; data sets on generally weaken that support.
the process generally. For example, if analogues; and coherence of data in this Experimental challenge to the
tumors and the hypothesized precursor case from cancer and noncancer toxicity hypothesized mode of action, where
both increase with dose, the two studies. interrupting the sequence of key events
responses will be correlated regardless suppresses the tumor response or
2.4.3.3. Consideration of the Possibility enhancement of key events increases the
of whether a causal relationship exists.
of Other Modes of Action tumor response, creates very strong
This is similar to the issue of
confounding in epidemiologic studies. The possible involvement of more support for the mode of action.
Dose-response studies coupled with than one mode of action at the tumor (b) Is the hypothesized mode of action
mechanistic studies can assist in site should be considered. Pertinent relevant to humans? If an hypothesized
clarifying these relationships. observations that are not consistent with mode of action is sufficiently supported
• What are the correlations among the hypothesized mode of action can in the test animals, the sequence of key
doses producing events and cancer? suggest the possibility of other modes of precursor events should be reviewed to
Pertinent observations include, but action. Some pertinent observations can identify critical similarities and
are not limited to, 2-year bioassay be consistent with more than one mode differences between the test animals and
observation of lesions correlated with of action. Furthermore, different modes humans. The question of concordance
observations of hormone changes and of action can operate in different dose can be complicated by cross-species
the same lesions in shorter term studies ranges; for example, an agent can act differences in toxicokinetics or
or in interim sacrifice. predominantly through cytotoxicity at toxicodynamics. For example, the active
agent can be formed through different assessment (see Section 3.5). This cancer. For example, the narrative can
metabolic pathways in animals and includes the potential for a higher clearly state to what extent the
humans. Any information suggesting internal dose of the active agent or for determination was based on data from
quantitative differences between an increased occurrence of a key human exposure, from animal
animals and humans is flagged for precursor event. Quantitative experiments, from some combination of
consideration in the dose-response differences may result in separate risk the two, or from other data. Similarly,
assessment. This includes the potential estimates for susceptible populations or information on mode of action can
for different internal doses of the active lifestages. specify to what extent the data are from
agent or for differential occurrence of a The possibility that childhood is a in vivo or in vitro exposures or based on
key precursor event. susceptible period for exposure should similarities to other chemicals. The
‘‘Relevance’’ of a potential mode of be explicitly addressed. Generic extent to which an agent’s mode of
action is considered in the context of understanding of the mode of action can action occurs only on reaching a
characterization of hazard, not level of be used to gauge childhood minimum dose or a minimum duration
risk. Anticipated levels of human susceptibility, and this determination should also be presented. A hazard
exposure are not used to determine can be refined through analysis of agent- might also be expressed
whether the hypothesized mode of specific data. disproportionately in individuals
action is relevant to humans. Exposure possessing a specific gene; such
2.4.4. Evolution With Experience
information is integrated into the overall characterizations may follow from a
risk characterization. Several groups have proposed or better understanding of the human
The question of relevance considers incorporated mode of action into their genome. Furthermore, route of exposure
all populations and lifestages. It is risk assessments (see, e.g., U.S. EPA, should be used to qualify a hazard if, for
possible that the conditions under 1991b; Sonich-Mullin et al., 2001; Meek example, an agent is not absorbed by
which a mode of action operates exist et al., 2003). As the frameworks and
some routes. Similarly, a hazard can be
primarily in a particular population or mandates under which these
attributable to exposures during a
lifestage, for example, in those with a evaluations were produced differ, the
susceptible lifestage on the basis of our
pre-existing hormonal imbalance. Other specific procedures described in and
understanding of human development.
populations or lifestages may not be conclusions drawn may also differ. The weight of evidence-of-evidence
analogous to the test animals, in which Nevertheless, the number of case studies narrative should highlight:
case the question of relevance would be from all venues remains limited. More • The quality and quantity of the
decided by inference. experience with differing modes of data;
Special attention should be paid to action are expected to highlight and • All key decisions and the basis for
whether tumors can arise from illustrate the strengths and limitations these major decisions; and
childhood exposure, considering of the general framework proposed in • Any data, analyses, or assumptions
various aspects of development during these cancer guidelines. Moreover, that are unusual for or new to EPA.
these lifestages. Because the studies that additional toxicological techniques may To capture this complexity, a weight
support a mode of action are typically expand or change scientific judgments of evidence narrative generally includes:
conducted in mature animals, regarding which information is useful • Conclusions about human
conclusions about relevance during for mode of action determinations. As carcinogenic potential (choice of
childhood generally rely on inference. warranted, additional guidance may be descriptor(s), described below),
There is currently no standard Agency proposed as experience is gained and/or • A summary of the key evidence
position regarding the issue of whether as toxicological knowledge advances. supporting these conclusions (for each
tumors arising through the hypothesized descriptor used), including information
mode of action are relevant during 2.5. Weight of Evidence Narrative
on the type(s) of data (human and/or
childhood; understanding the mode of The weight of evidence narrative is a animal, in vivo and/or in vitro) used to
action implies that there are sufficient short summary (one to two pages) that support the conclusion(s),
data (on either the specific agent or the explains an agent’s human carcinogenic • Available information on the
general mode of action) to form a potential and the conditions that epidemiologic or experimental
confident conclusion about relevance characterize its expression. It should be conditions that characterize expression
during childhood. sufficiently complete to be able to stand of carcinogenicity (e.g., if
(c) Which populations or lifestages alone, highlighting the key issues and carcinogenicity is possible only by one
can be particularly susceptible to the decisions that were the basis for the exposure route or only above a certain
hypothesized mode of action? If an evaluation of the agent’s potential human exposure level),
hypothesized mode of action is judged hazard. It should be sufficiently clear • A summary of potential modes of
relevant to humans, information about and transparent to be useful to risk action and how they reinforce the
the key precursor event(s) is reviewed to managers and non-expert readers. It may conclusions,
identify populations or lifestages that be useful to summarize all of the • Indications of any susceptible
might reasonably expected to be significant components and conclusions populations or lifestages, when
particularly susceptible to their in the first paragraph of the narrative available, and
occurrence. Although agent-specific and to explain complex issues in more • A summary of the key default
data would provide the strongest depth in the rest of the narrative. options invoked when the available
indication of susceptibility, this review The weight of the evidence should be information is inconclusive.
may also rely on general knowledge presented as a narrative laying out the To provide some measure of clarity
about the precursor events and complexity of information that is and consistency in an otherwise free-
characteristics of individuals essential to understanding the hazard form narrative, the weight of evidence
susceptible to these events. Any and its dependence on the quality, descriptors are included in the first
information suggesting quantitative quantity, and type(s) of data available, sentence of the narrative. Choosing a
differences between populations or as well as the circumstances of exposure descriptor is a matter of judgment and
lifestages should be flagged for or the traits of an exposed population cannot be reduced to a formula. Each
consideration in the dose-response that may be required for expression of descriptor may be applicable to a wide
variety of potential data sets and to be caused by a human metabolite, in Although the term ‘‘likely’’ can have
weights of evidence. These descriptors which case in the narrative the untested a probabilistic connotation in other
and narratives are intended to permit agent could have the same descriptor as contexts, its use as a weight of evidence
sufficient flexibility to accommodate the metabolite. As new testing methods descriptor does not correspond to a
new scientific understanding and new are developed and used, assessments quantifiable probability of whether the
testing methods as they are developed may increasingly be based on inferences chemical is carcinogenic. This is
and accepted by the scientific from toxicokinetic and mode of action because the data that support cancer
community and the public. Descriptors information in the absence of tumor assessments generally are not suitable
represent points along a continuum of studies in animals or humans. for numerical calculations of the
evidence; consequently, there are When a well-studied agent produces probability that an agent is a carcinogen.
gradations and borderline cases that are tumors only at a point of initial contact, Other health agencies have expressed a
clarified by the full narrative. the descriptor generally applies only to comparable weight of evidence using
Descriptors, as well as an introductory the exposure route producing tumors terms such as ‘‘Reasonably Anticipated
paragraph, are a short summary of the unless the mode of action is relevant to to Be a Human Carcinogen’’ (NTP) or
complete narrative that preserves the other routes. The rationale for this ‘‘Probably Carcinogenic to Humans’’
complexity that is an essential part of conclusion would be explained in the (International Agency for Research on
the hazard characterization. Users of narrative. Cancer).
these cancer guidelines and of the risk When tumors occur at a site other The following descriptors can be used
assessments that result from the use of than the point of initial contact, the as an introduction to the weight of
these cancer guidelines should consider descriptor generally applies to all evidence narrative. The examples
the entire range of information included exposure routes that have not been presented in the discussion of the
in the narrative rather than focusing adequately tested at sufficient doses. An descriptors are illustrative. The
simply on the descriptor. exception occurs when there is examples are neither a checklist nor a
In borderline cases, the narrative convincing information, e.g., limitation for the descriptor. The
explains the case for choosing one toxicokinetic data that absorption does complete weight of evidence narrative,
descriptor and discusses the arguments not occur by another route. rather than the descriptor alone,
for considering but not choosing When the response differs provides the conclusions and the basis
another. For example, between qualitatively as well as quantitatively for them.
‘‘suggestive’’ and ‘‘likely’’ or between with dose, this information should be ‘‘Carcinogenic to Humans.’’ This
‘‘suggestive’’ and ‘‘inadequate,’’ the part of the characterization of the descriptor indicates strong evidence of
explanation clearly communicates the hazard. In some cases reaching a certain human carcinogenicity. It covers
information needed to consider dose range can be a precondition for different combinations of evidence.
appropriately the agent’s carcinogenic effects to occur, as when cancer is • This descriptor is appropriate when
potential in subsequent decisions. secondary to another toxic effect that there is convincing epidemiologic
Multiple descriptors can be used for appears only above a certain dose. In evidence of a causal association
a single agent, for example, when other cases exposure duration can be a between human exposure and cancer.
carcinogenesis is dose-or route- precondition for hazard if effects occur • Exceptionally, this descriptor may
dependent. For example, if an agent only after exposure is sustained for a be equally appropriate with a lesser
causes point-of-contact tumors by one certain duration. These considerations weight of epidemiologic evidence that is
exposure route but adequate testing is differ from the issues of relative strengthened by other lines of evidence.
negative by another route, then the absorption or potency at different dose It can be used when all of the following
agent could be described as likely to be levels because they may represent a conditions are met: (a) There is strong
carcinogenic by the first route but not discontinuity in a dose-response evidence of an association between
likely to be carcinogenic by the second. function. human exposure and either cancer or
Another example is when the mode of When multiple bioassays are the key precursor events of the agent’s
action is sufficiently understood to inconclusive, mode of action data are mode of action but not enough for a
conclude that a key event in tumor likely to hold the key to resolution of causal association, and (b) there is
development would not occur below a the more appropriate descriptor. When extensive evidence of carcinogenicity in
certain dose range. In this case, the bioassays are few, further bioassays to animals, and (c) the mode(s) of
agent could be described as likely to be replicate a study’s results or to carcinogenic action and associated key
carcinogenic above a certain dose range investigate the potential for effects in precursor events have been identified in
but not likely to be carcinogenic below another sex, strain, or species may be animals, and (d) there is strong evidence
that range. useful. that the key precursor events that
Descriptors can be selected for an When there are few pertinent data, the precede the cancer response in animals
agent that has not been tested in a descriptor makes a statement about the are anticipated to occur in humans and
cancer bioassay if sufficient other database, for example, ‘‘Inadequate progress to tumors, based on available
information, e.g., toxicokinetic and Information to Assess Carcinogenic biological information. In this case, the
mode of action information, is available Potential,’’ or a database that provides narrative includes a summary of both
to make a strong, convincing, and ‘‘Suggestive Evidence of Carcinogenic the experimental and epidemiologic
logical case through scientific inference. Potential.’’ With more information, the information on mode of action and also
For example, if an agent is one of a well- descriptor expresses a conclusion about an indication of the relative weight that
defined class of agents that are the agent’s carcinogenic potential to each source of information carries, e.g.,
understood to operate through a humans. If the conclusion is positive, based on human information, based on
common mode of action and if that the agent could be described as ‘‘Likely limited human and extensive animal
agent has the same mode of action, then to Be Carcinogenic to Humans’’ or, with experiments.
in the narrative the untested agent strong evidence, ‘‘Carcinogenic to ‘‘Likely to Be Carcinogenic to
would have the same descriptor as the Humans.’’ If the conclusion is negative, Humans.’’ This descriptor is appropriate
class. Another example is when an the agent could be described as ‘‘Not when the weight of the evidence is
untested agent’s effects are understood Likely to Be Carcinogenic to Humans.’’ adequate to demonstrate carcinogenic
potential to humans but does not reach not provide further insights. Some ‘‘Not Likely to Be Carcinogenic to
the weight of evidence for the descriptor examples include: Humans.’’
‘‘Carcinogenic to Humans.’’ Adequate • A small, and possibly not ‘‘Not Likely to Be Carcinogenic to
evidence consistent with this descriptor statistically significant, increase in Humans.’’ This descriptor is appropriate
covers a broad spectrum. As stated tumor incidence observed in a single when the available data are considered
previously, the use of the term ‘‘likely’’ animal or human study that does not robust for deciding that there is no basis
as a weight of evidence descriptor does reach the weight of evidence for the for human hazard concern. In some
not correspond to a quantifiable descriptor ‘‘Likely to Be Carcinogenic to instances, there can be positive results
probability. The examples below are Humans.’’ The study generally would in experimental animals when there is
meant to represent the broad range of not be contradicted by other studies of strong, consistent evidence that each
data combinations that are covered by equal quality in the same population mode of action in experimental animals
this descriptor; they are illustrative and group or experimental system (see does not operate in humans. In other
provide neither a checklist nor a discussions of conflicting evidence and cases, there can be convincing evidence
limitation for the data that might differing results, below); in both humans and animals that the
support use of this descriptor. • A small increase in a tumor with a agent is not carcinogenic. The judgment
Moreover, additional information, e.g., high background rate in that sex and may be based on data such as:
on mode of action, might change the strain, when there is some but • Animal evidence that demonstrates
choice of descriptor for the illustrated insufficient evidence that the observed lack of carcinogenic effect in both sexes
examples. Supporting data for this tumors may be due to intrinsic factors in well-designed and well-conducted
descriptor may include: that cause background tumors and not studies in at least two appropriate
• An agent demonstrating a plausible due to the agent being assessed. (When animal species (in the absence of other
(but not definitively causal) association there is a high background rate of a animal or human data suggesting a
between human exposure and cancer, in specific tumor in animals of a particular potential for cancer effects),
most cases with some supporting sex and strain, then there may be • Convincing and extensive
biological, experimental evidence, biological factors operating experimental evidence showing that the
though not necessarily carcinogenicity independently of the agent being only carcinogenic effects observed in
data from animal experiments; assessed that could be responsible for animals are not relevant to humans,
• An agent that has tested positive in • Convincing evidence that
the development of the observed
animal experiments in more than one carcinogenic effects are not likely by a
tumors.) In this case, the reasons for
species, sex, strain, site, or exposure particular exposure route (see Section
determining that the tumors are not due
route, with or without evidence of 2.3), or
to the agent are explained;
carcinogenicity in humans; • Convincing evidence that
• A positive tumor study that raises • Evidence of a positive response in
carcinogenic effects are not likely below
additional biological concerns beyond a study whose power, design, or
a defined dose range.
that of a statistically significant result, conduct limits the ability to draw a A descriptor of ‘‘not likely’’ applies
for example, a high degree of confident conclusion (but does not only to the circumstances supported by
malignancy, or an early age at onset; make the study fatally flawed), but the data. For example, an agent may be
• A rare animal tumor response in a where the carcinogenic potential is ‘‘Not Likely to Be Carcinogenic’’ by one
single experiment that is assumed to be strengthened by other lines of evidence route but not necessarily by another. In
relevant to humans; or (such as structure-activity those cases that have positive animal
• A positive tumor study that is relationships); or experiment(s) but the results are judged
strengthened by other lines of evidence, • A statistically significant increase at to be not relevant to humans, the
for example, either plausible (but not one dose only, but no significant narrative discusses why the results are
definitively causal) association between response at the other doses and no not relevant.
human exposure and cancer or evidence overall trend. Multiple Descriptors. More than one
that the agent or an important ‘‘Inadequate Information to Assess descriptor can be used when an agent’s
metabolite causes events generally Carcinogenic Potential.’’ This descriptor effects differ by dose or exposure route.
known to be associated with tumor of the database is appropriate when For example, an agent may be
formation (such as DNA reactivity or available data are judged inadequate for ‘‘Carcinogenic to Humans’’ by one
effects on cell growth control) likely to applying one of the other descriptors. exposure route but ‘‘Not Likely to Be
be related to the tumor response in this Additional studies generally would be Carcinogenic’’ by a route by which it is
case. expected to provide further insights. not absorbed. Also, an agent could be
‘‘Suggestive Evidence of Carcinogenic Some examples include: ‘‘Likely to Be Carcinogenic’’ above a
Potential.’’ This descriptor of the • Little or no pertinent information; specified dose but ‘‘Not Likely to Be
database is appropriate when the weight • Conflicting evidence, that is, some Carcinogenic’’ below that dose because
of evidence is suggestive of studies provide evidence of a key event in tumor formation does not
carcinogenicity; a concern for potential carcinogenicity but other studies of occur below that dose.
carcinogenic effects in humans is raised, equal quality in the same sex and strain
but the data are judged not sufficient for are negative. Differing results, that is, 2.6. Hazard Characterization
a stronger conclusion. This descriptor positive results in some studies and The hazard characterization contains
covers a spectrum of evidence negative results in one or more different the hazard information needed for a full
associated with varying levels of experimental systems, do not constitute risk characterization (U.S. EPA, 2000b).
concern for carcinogenicity, ranging conflicting evidence, as the term is used It presents the results of the hazard
from a positive cancer result in the only here. Depending on the overall weight assessment and explains how the weight
study on an agent to a single positive of evidence, differing results can be of evidence conclusion was reached.
cancer result in an extensive database considered either suggestive evidence or The hazard characterization
that includes negative studies in other likely evidence; or summarizes, in plain language,
species. Depending on the extent of the • Negative results that are not conclusions about the agent’s potential
database, additional studies may or may sufficiently robust for the descriptor, effects, whether they can be expected to
depend qualitatively on the levels of interest. Dose-response Humans.’’ When there is suggestive
circumstances of exposure, and if assessments are useful in many evidence, the Agency generally would
anyone can be expected to be especially applications: Estimating risk at different not attempt a dose-response assessment,
susceptible. It discusses the extent to exposure levels, estimating the risk as the nature of the data generally
which these conclusions are supported reduction for different decision options, would not support one; however, when
by data or are the result of default estimating the risk remaining after an the evidence includes a well-conducted
options invoked because the data are action is taken, providing the risk study, quantitative analyses may be
inconclusive. It explains how complex information needed for benefit-cost useful for some purposes, for example,
cases with differing results in different analyses of different decision options, providing a sense of the magnitude and
studies were resolved. The hazard comparing risks across different agents uncertainty of potential risks, ranking
characterization highlights the major or health effects, and setting research potential hazards, or setting research
issues addressed in the hazard priorities. The purpose of the priorities. In each case, the rationale for
assessment and discusses alternative assessment should consider the quality
the quantitative analysis is explained,
interpretations of the data and the of the data available, which will vary
considering the uncertainty in the data
degree to which they are supportable from case to case.
A dose-response analysis is generally and the suggestive nature of the weight
scientifically and are consistent with
developed from each study that reports of evidence. These analyses generally
EPA guidelines.
When the conclusion is supported by quantitative data on dose and response. would not be considered Agency
mode of action information, the hazard Alternative measures of dose are consensus estimates. Dose-response
characterization also provides a clear available for analyzing human and assessments are generally not done
summary of the mode of action animal studies (see Section 3.1). A two- when there is inadequate evidence,
conclusions (see Section 2.4.3.4), step approach distinguishes analysis of although calculating a bounding
including the completeness of the data, the dose-response data from inferences estimate from an epidemiologic or
the strengths and limitations of the made about lower doses. The first step experimental study that does not show
inferences made, the potential for other is an analysis of dose and response in positive results can indicate the study’s
modes of action, and the implications of the range of observation of the level of sensitivity and capacity to
the mode of action for selecting viable experimental or epidemiologic studies detect risk levels of concern.
approaches to the dose-response (see Section 3.2). Modeling is Cancer is a collection of several
assessment. The hazard characterization encouraged to incorporate a wide range diseases that develop through cell and
also discusses the extent to which mode of experimental data into the dose-
tissue changes over time. Dose-response
of action information is available to response assessment (see Sections 3.1.2,
assessment procedures based on tumor
address the potential for 3.2.1, 3.2.2, 3.2.3). The modeling yields
a point of departure (POD) near the incidence have seldom taken into
disproportionate risks in specific account the effects of key precursor
populations or lifestages or the potential lower end of the observed range,
without significant extrapolation to events within the whole biological
for enhanced risks on the basis of process due to lack of empirical data
interactions with other agents or lower doses (see Sections 3.2.4, 3.2.5).
The second step is extrapolation to and understanding about these events.
stressors, if anticipated. In this discussion, response data
Topics that can be addressed in a lower doses (see Section 3.3). The
extrapolation approach considers what include measures of key precursor
hazard characterization include:
• Summary of the results of the is known about the agent’s mode of events considered integral to the
hazard assessment; action (see Section 3.3.1). Both linear carcinogenic process in addition to
• Identification of any likely and nonlinear approaches are available tumor incidence. These responses may
susceptible populations and lifestages, (see Sections 3.3.3, 3.3.4). When include changes in DNA, chromosomes,
especially attending to children, infants, multiple estimates can be developed, or other key macromolecules; effects on
and fetuses; the strengths and weaknesses of each growth signal transduction, including
• Conclusions about the agent’s mode are presented. In some cases, they may induction of hormonal changes; or
of action, and implications for selecting be combined in a way that best physiological or toxic effects that
approaches to the dose-response represents human cancer risk (see include proliferative events diagnosed
assessment; Section 3.3.5). Special consideration is as precancerous but not pathology that
• Identification of the available lines given to describing dose-response is judged to be cancer. Analysis of such
of evidence (e.g., animal bioassays, differences attributable to different responses may be done along with that
epidemiologic studies, toxicokinetic human exposure scenarios (see Section of tumor incidence to enhance the
information, mode of action studies, and 3.4) and to susceptible populations and tumor dose-response analysis. If dose-
information about structural analogues lifestages (see Section 3.5). It is response analysis of nontumor key
or metabolites), highlighting data important to discuss significant events is more informative about the
quality and coherence of results from uncertainties encountered in the carcinogenic process for an agent, it can
different lines of evidence; and analysis (see Section 3.6) and to be used in lieu of, or in conjunction
• Strengths and limitations of the characterize other important aspects of with, tumor incidence analysis for the
hazard assessment, highlighting the dose-response assessment (see overall dose-response assessment.
significant issues in interpreting the Section 3.7).
data, alternative interpretations that are The scope, depth, and use of a dose- As understanding of mode of action
considered equally plausible, critical response assessment vary in different improves and new types of data become
data gaps, and default options invoked circumstances. Although the quality of available, dose-response assessment will
when the available information is dose-response data is not necessarily continue to evolve. These cancer
inconclusive. related to the weight of evidence guidelines encourage the development
descriptor, dose-response assessments and application of new methods that
3. Dose-Response Assessment are generally completed for agents improve dose-response assessment by
Dose-response assessment estimates considered ‘‘Carcinogenic to Humans’’ reflecting new scientific understanding
potential risks to humans at exposure and ‘‘Likely to Be Carcinogenic to and new sources of information.
3.1. Analysis of Dose 3.1.1. Standardizing Different may be averaged over the week, where
Experimental Exposure Regimens appropriate.
For each effect observed, dose- Doses in studies at the cellular or
response assessment should begin by Complex exposure or dosing regimens
molecular level can be difficult to relate
determining an appropriate dose metric. are often present in experimental and
to organ- or organism-level dose metrics.
Several dose metrics have been used, epidemiologic studies. The resulting
Toxicokinetic modeling can sometimes
e.g., delivered dose, body burden, and internal dose depends on many
be used to relate doses at the cellular or
area under the curve, and others may be variables, including concentration,
molecular level to doses or exposures at
appropriate depending on the data and duration, frequency of administration,
higher levels of organization.
mode of action. and duration of recovery periods
Selection of an appropriate dose between administrations. Internal dose 3.1.2. Toxicokinetic Data and Modeling
metric considers what data are available also depends on variables that are In the absence of chemical-specific
and what is known about the agent’s intrinsic to the exposed individual, such data, physiologically based
mode of action at the target site, and as lifestage and rates of metabolism and toxicokinetic modeling is potentially the
uncertainties involved in estimation and clearance. To facilitate comparing most comprehensive way to account for
application of alternative metrics. The results from different study designs and biological processes that determine
dose metric specifies: to make inferences about human internal dose. Physiologically based
• The agent measured, preferably the exposures, a summary estimate of the models commonly describe blood flow
active agent (administered agent or a dose metric, whether the administered between physiological compartments
metabolite); dose or inhalation exposure and simulate the relationship between
concentration or an internal metric, may applied dose and internal dose.
• Proximity to the target site
be derived for a complex exposure Toxicokinetic models generally need
(exposure concentration, potential dose,
regimen. data on absorption, distribution,
internal dose, or delivered dose,5
Toxicokinetic modeling is the metabolism, and elimination of the
reflecting increasing proximity); and
preferred approach for estimating dose administered agent and its metabolites.
• The time component of the effective metrics from exposure. Toxicokinetic Additionally, in the case of inhalation
dose (cumulative dose, average dose, models generally describe the exposures, models can explicitly
peak dose, or body burden). relationship between exposure and characterize the geometry of the
Analyses can be based on estimates of measures of internal dose over time. respiratory tract and the airflow through
animal dose metrics or human dose More complex models can reflect it, as well as the interaction of this
metrics. The assessment should describe sources of intrinsic variation, such as airflow with the entrained particles or
the approach used to select a dose polymorphisms in metabolism and fibers and gases (Kimbell et al., 2001;
metric and the reasons for this clearance rates. When a robust model is Subramaniam et al., 2003). Because of
approach. The final analysis, however, not available, or when the purpose of large interspecies differences in airway
should determine a human equivalent the assessment does not warrant morphometry such models can be
dose metric. This facilitates comparing developing a model, simpler approaches particularly useful in interspecies
results from different datasets and may be used. extrapolations. When employed,
effects by using human equivalent dose/ For chronic exposure studies, the however, the potential for large inter-
concentrations as common metrics. cumulative exposure or dose individual differences in airway
When appropriate, it may be necessary administered often is expressed as an morphometry, are considered to ensure
to convert dose metrics across exposure average over the duration of the study, that the models provide information
routes. When route-to-route as one consistent dose metric. This representative of human populations.
extrapolations are made, the underlying approach implies that a higher dose Toxicokinetic models can improve
data, algorithms, and assumptions are administered over a short duration is dose-response assessment by revealing
clearly described. equivalent to a commensurately lower and describing nonlinear relationships
Timing of exposure can also be dose administered over a longer between applied and internal dose.
important. When there is a susceptible duration. Uncertainty usually increases Nonlinearity observed in a dose-
lifestage, doses during the susceptible as the duration becomes shorter relative response curve often can be attributed to
period are not equivalent to doses at to the averaging duration or the toxicokinetics (Hoel et al., 1983; Gaylor
other times, and they would be analyzed intermittent doses become more intense et al., 1994), involving, for example,
separately. than the averaged dose. Moreover, doses saturation or induction of enzymatic
during any specific susceptible or processes at high doses. In some cases,
5 Exposure is contact of an agent with the outer
refractory period would not be toxicokinetic processes tend to become
boundary of an organism. Exposure concentration equivalent to doses at other times. For linear at sufficiently low doses (Hattis,
is the concentration of a chemical in its transport
or carrier medium at the point of contact. Dose is these reasons, cumulative exposure or 1990).
the amount of a substance available for interaction potential dose may be replaced by a A discussion of confidence should
with metabolic processes or biologically significant more appropriate dose metric when accompany the presentation of model
receptors after crossing the outer boundary of an indicated by the data. results and include consideration of
organism. Potential dose is the amount ingested,
inhaled, or applied to the skin. Applied dose is the
For mode of action studies, the dose model validation and sensitivity
amount of a substance presented to an absorption metric should be calculated over a analysis, stressing the predictive
barrier and available for absorption (although not duration that reflects the time to performance of the model and whether
necessarily having yet crossed the outer boundary occurrence of the key precursor effects. the model is sufficient to support
of the organism). Absorbed dose is the amount
crossing a specific absorption barrier (e.g., the
Mode of action studies are often of decision-making. Quantitative
exchange boundaries of skin, lung, and digestive limited duration, as the precursors can uncertainty analysis is important for
tract) through uptake processes. Internal dose is a be observed after less-than-chronic evaluating the performance of a model,
more general term, used without respect to specific exposures. When the experimental whether the model is based primarily on
absorption barriers or exchange boundaries.
Delivered dose is the amount of the chemical
exposure regimen is specified on a default assumptions or chemical-
available for interaction by any particular organ or weekly basis (for example, 4 hours a specific data. The uncertainty analysis
cell (U.S. EPA, 1992a). day, 5 days a week), the daily exposure covers questions of model uncertainty
(e.g., Is the model based on the equivalence, which determines tissue respiratory tract) or remotely. For
appropriate biology and how does that doses in animals and humans that yield example, current default models (U.S.
affect estimates of dose metrics?) and equal lifetime risks (U.S. EPA, 1992b). EPA, 1994) use parameters such as:
parameter uncertainty (e.g., Do the data Toxicokinetic modeling (see Section • Inhalation rate and surface area of
support unbiased and stable estimates of 3.1.2) addresses factors associated with the affected part of the respiratory tract
the model parameters?). When a toxicokinetic equivalence, and for gases eliciting the response locally,
delivered dose measure is used in toxicodynamic modeling (see Section • Blood:gas partition coefficients for
animal-to-human extrapolation, the 3.2.2) addresses factors associated with remote acting gases,
assessment discusses the confidence of toxicodynamic equivalence. When • Fractional deposition with
the target tissue and its toxicodynamics toxicokinetic modeling is used without inhalation rate and surface area of the
being the same in both species (see toxicodynamic modeling, the dose- affected part of the respiratory tract for
Section 3.6). Toxicokinetic modeling response assessment develops and particles eliciting the response locally,
results may be presented alone as the supports an approach for addressing and
preferred method of estimating human toxicodynamic equivalence, perhaps by • Fractional deposition with
equivalent exposures or doses, or these retaining some of the cross-species inhalation rate and body weight for
results may be presented in parallel scaling factor (e.g., using the square root particles eliciting the response remotely.
with default procedures (see Section of the cross-species scaling factor or The current default values for some
3.1.3), depending on the confidence in using a factor of 3 to cover parameters used in the default models
the modeling. toxicodynamic differences between (e.g., breathing rate and respiratory tract
animals and humans, as is currently surface area) are based on data from
3.1.3. Cross-Species Scaling Procedures
done in deriving inhalation reference adults (U.S. EPA, 1994). The human
Standard cross-species scaling concentrations [U.S. EPA, 1994]).
procedures are available when the data respiratory system passes through
When assessing risks from childhood
are not sufficient to support a several distinct stages of maturation and
exposure, the mg/kg3/4-d scaling factor
toxicokinetic model or when the growth during the first several years of
does not use the child’s body weight
purpose of the assessment does not life and into adolescence (Pinkerton and
(U.S. EPA, 1992b). This reflects several
warrant developing one. The aim is to Joad, 2000), during which
uncertainties in extrapolating risks to
define exposure levels for humans and characteristics important to disposition
children:
animals that are expected to produce the • The data supporting the mg/kg3/4-d of inhaled toxicants may vary. Children
same degree of effect (U.S. EPA, 1992b), scaling factor were derived for and adults breathing the same
taking into account differences in scale differences across species and may not concentration of an agent may receive
between test animals and humans, such apply as well to differently sized different doses to the body or lungs
as size and lifespan. individuals of the same species or to (U.S. EPA, 2002b). Consequently, it may
different lifestages. be appropriate to evaluate the default
3.1.3.1. Oral Exposures models by considering physiologic and
• In addition to metabolic differences,
For oral exposures, administered there are also important toxicodynamic anatomic factors representative of early
doses should be scaled from animals to differences; for example, children have lifestages, for example through the
humans on the basis of equivalence of faster rates of cell division than do substitution of child-specific parameters
mg/kg3/4-d (milligrams of the agent adults, so scaling across different (U.S. EPA, 2002b). Such evaluation uses
normalized by the 3⁄4 power of body lifestages and species simultaneously the default model and dosimetric
weight per day) (U.S. EPA, 1992b). The may be particularly uncertain. adjustment in use at the time of the
3⁄4 power is consistent with current assessment coupled with the best
science, including empirical data that 3.1.3.2. Inhalation Exposures understanding of child-specific
allow comparison of potencies in For inhalation exposures parameters at that time (e.g., drawn from
humans and animals, and it is also experimental exposure concentrations the scientific literature). This analysis is
supported by analysis of the allometric are replaced with human equivalent undertaken with caution: (1) because of
variation of key physiological concentrations calculated using EPA’s the correlations between activity level,
parameters across mammalian species. methods for deriving inhalation breathing rate, respiratory tract
It is generally more appropriate at low reference concentrations (U.S. EPA, dimensions, and body weight and (2) to
doses, where sources of nonlinearity 1994), which give preference to the use avoid the possibility of mismatching the
such as saturation of enzyme activity are of toxicokinetic modeling. When type of agent (gas or particle) and its site
less likely to occur. This scaling is toxicokinetic models are unavailable, of response (within the respiratory tract
intended as an unbiased estimate rather default dosimetry models are employed or remote from the respiratory tract)
than a conservative one. Equating to extrapolate from experimental with the relevant dosimetry factors in
exposure concentrations in food or exposure concentrations to human use at the time of the assessment.
water is an alternative version of the equivalent concentrations. When Analyses of children’s inhalation
same approach, because daily intakes of toxicokinetic modeling or dosimetry dosimetry are also considered when
food or water are approximately modeling is used without using model structures beyond the
proportional to the 3⁄4 power of body toxicodynamic modeling, the dose- default models (e.g., physiologically
weight. response assessment develops and based toxicokinetic models).
The aim of these cross-species scaling supports an approach for addressing When using dosimetry modeling, the
procedures is to estimate administered toxicodynamic equivalence. comparison of human-equivalent
doses in animals and humans that result The default dosimetry models concentrations for different lifestages
in equal lifetime risks. It is useful to typically involve the use of species- (e.g., for an adult and a child) can
recognize two components of this specific physiologic and anatomic indicate whether it is important to carry
equivalence: toxicokinetic equivalence, factors relevant to the form of the agent both concentrations forward in the dose-
which determines administered doses in (e.g., particle or gas) and categorized response assessment or whether a verbal
animals and humans that yield equal with regard to whether the response characterization of any findings will
tissue doses, and toxicodynamic occurs either locally (i.e., within the suffice.
3.1.4. Route Extrapolation internal dose for effects distant from the data are in the range of the exposures of
In certain situations, an assessment point of contact. interest. In other cases, as with data
based on studies of one exposure route Route extrapolation can be used to from animal experiments, information
understand how internal dose and from the observable range is
may be applied to another exposure
subsequent effects depend on exposure extrapolated to exposures of interest.
route. Route-to-route extrapolation has
route. If testing by different exposure Analysis of effects raises additional
both qualitative and quantitative
routes is available, the observation of issues:
aspects. For the qualitative aspect, the • Many studies collect information
similar or dissimilar internal doses can
assessor should weigh the degree to from death certificates, which leads to
be important in determining whether
which positive results by one exposure estimates of mortality rather than
and what conclusions can be made
route support a judgment that similar incidence. Because survival rates vary
concerning the dose-response
results would be expected by another for different cancers, the analysis may
function(s) for different routes of
route. In general, confidence in making be improved by adjusting mortality
exposure.
such a judgment is strengthened when figures to reflect the relationship
tumors are observed at a site distant 3.2. Analysis in the Range of between incidence and mortality.
from the portal of entry and when Observation • Epidemiologic studies, by their
absorption is similar through both The principle underlying these cancer nature, are limited in the extent to
routes. In the absence of contrary data, guidelines is to use approaches that which they can control for effects due
a qualitative default option can be used: include as much information as to exposures from other agents. In some
If the agent is absorbed through an possible. Quantitative information about cases, the agent can have discernible
exposure route to give an internal dose, key precursor events can be used to interactive effects with another agent,
it may be carcinogenic by that route. develop a toxicodynamic model. making it possible to estimate the
When a qualitative extrapolation can Alternatively, such information can be contribution of each agent as a risk
be supported, quantitative extrapolation fitted by empirical models to extend the factor for the effects of the other. For
may still be problematic due to the dose-response analysis of tumor example, competing risks in a study
absence of adequate data. The incidence to lower doses and response population can limit the observed
differences in biological processes levels. The analysis in the range of occurrence of cancer, while additive
among routes of exposure (oral, observation is used to establish a POD effects may lead to an increase
inhalation, dermal) can be great because near the lower end of the observed range occurrence of cancer. In the case of rates
of, for example, first-pass effects and (see Section 3.3). not already so adjusted, the analysis can
different results from different exposure be improved by correcting for
patterns. There is no generally 3.2.1. Epidemiologic Studies
competing or additive risks that are not
applicable method for accounting for Ideally, epidemiologic data would be similar in exposed and comparison
these differences in uptake processes in used to select the dose-response groups.
a quantitative route-to-route function for human exposures. Because • Comparison groups that are not free
extrapolation of dose-response data in epidemiologic data are usually limited from exposure to the agent can bias the
the absence of good data on the agent of and many models may fit the data risk estimates toward zero. The analysis
interest. Therefore, route-to-route (Samet et al.,1998), other factors may can be improved by considering
extrapolation of dose data relies on a influence model choice. For background exposures in the exposed
case-by-case analysis of available data. epidemiologic studies, including those and comparison groups.
When good data on the agent itself are with grouped data, analysis by linear • The latent period for most cancers
limited, an extrapolation analysis can be models in the range of observation is implies that exposures immediately
based on expectations from physical generally appropriate unless the fit is preceding the detection of a tumor
and chemical properties of the agent, poor. The relatively small exposure would be less likely to have contributed
properties and route-specific data on range observed in many epidemiologic to its development and, therefore, may
structurally analogous compounds, or in studies, for example, makes it difficult count less in the analysis. Study
vitro or in vivo uptake data on the agent. to discern the shape of the exposure-or subjects who were first exposed near the
Route-to-route uptake models may be dose-response curve. Exposure end of the study may not have had
applied if model parameters are suitable misclassification and errors in exposure adequate time since exposure for cancer
for the compound of interest. Such estimation also obscure the shape of the to develop; therefore, analysis of their
models are currently considered interim dose-response curve. When these errors data may be similar to analysis of data
methods; further model development are unsystematic or random, the result for those who were not exposed.
and validation is awaiting the is frequently to bias the risk estimates However, for carcinogens that act on
development of more extensive data. toward zero. When a linear model fits multiple stages of the carcinogenic
For screening or hazard ranking, route- poorly, more flexible models that allow process, especially the later stages, all
to-route extrapolation may be based on for low-dose linearity, for example, a periods of exposure. including recent
assumed quantitative comparability as a linear-quadratic model or a Hill model exposures, may be important.
default, as long as it is reasonable to (Murrell et al., 1998), are often Some study designs can yield only a
assume absorption by compared routes. considered next. partial characterization of the overall
When route-to-route extrapolation is Analysis of epidemiologic studies hazard and therefore risk as, for
used, the assessor’s degree of confidence depends on the type of study and example, in studies that: (1) investigate
in both the qualitative and quantitative quality of the data, particularly the only one effect (typical of many case-
extrapolation is discussed in the availability of quantitative measures of control studies), (2) include only one
assessment and highlighted in the dose- exposure. The objective is to develop a population segment (e.g., male workers
response characterization. dose-response curve that estimates the or workers of one socioeconomic class),
Toxicokinetic modeling can be used incidence of cancer attributable to the or (3) include only one lifestage (e.g.,
to compare results of studies by dose (as estimated from the exposure) to childhood leukemia following maternal
different exposure routes. Results can the agent. In some cases, e.g., tobacco exposure to contaminated drinking
also be compared on the basis of smoke or occupational exposures, the water). To obtain a more complete
characterization that includes risks of agent-specific data to estimate the models and not biological information
other cancers, estimates from these model’s parameters. An example is the about the agent being assessed or about
studies can be supplemented with two-stage clonal expansion model carcinogenesis in general. In cases
estimates from other studies that developed by Moolgavkar and Knudson where curve-fitting models are used
investigated other cancers, population (1981) and Chen and Farland (1991). because the data are not adequate to
segments, or lifestages (see Section 3.5). These models continue to be improved support a toxicodynamic model, there
When several studies are available for as more information becomes available. generally would be no biological basis
dose-response analysis, meta-analysis It is possible for different models to to choose among alternative curve-
can provide a systematic approach to provide equivalent fits to the observed fitting models. However, in situations
weighing positive studies and those data but to diverge substantially in their where there are alternative models with
studies that do not show positive projections at lower doses. When model significant biological support, the
results, and calculating an overall risk parameters are estimated from tumor decisionmaker can be informed by the
estimate with greater precision. Issues incidence data, it is often the case that presentation of these alternatives along
considered include the comparability of different combinations of parameter with their strengths and uncertainties.
studies, heterogeneity across studies, estimates can yield similar results in the Quantitative data on precursors can
and the potential for a single large study observed range. For this reason, critical be used in conjunction with, or in lieu
to dominate the analysis. Confidence in parameters (e.g., mutation rates and cell of, data on tumor incidence to extend
a meta-analysis is increased when it birth and death rates) are estimated from the dose-response curve to lower doses.
considers study quality, including laboratory studies and not by curve- Caution is used with rates of molecular
definition of the study population and fitting to tumor incidence data (Portier, events such as mutation or cell
comparison group, measurement of 1987). This approach reduces model proliferation or signal transduction.
exposure, potential for exposure uncertainty (see Section 3.6) and Such rates can be difficult to relate to
misclassification, adequacy of follow-up ensures that the model does not give cell or tissue changes overall. The
period, and analysis of confounders (see answers that are biologically unrealistic. timing of observations of these
Section 2.2.1.3). This approach also provides a phenomena, as well as the cell type
robustness of results, where the results involved, is linked to other precursor
3.2.2. Toxicodynamic (‘‘Biologically
are not likely to change substantially if events to ensure that the measurement
Based’’) Modeling
fitted to slightly different data. is truly a key event (Section 2.4).
Toxicodynamic modeling can be used Toxicodynamic modeling can provide For incidence data on either tumors
when there are sufficient data to insight into the relationship between or a precursor, an established empirical
ascertain the mode of action (see tumors and key precursor events. For procedure is used to provide objectivity
Section 2.4) and quantitatively support example, a model that includes cell and consistency among assessments.
model parameters that represent rates proliferation can be used to explore the The procedure models incidence,
and other quantities associated with the extent to which small increases in the corrected for background, as an
key precursor events of the mode of cell proliferation rate can lead to large increasing function of dose. The models
action. Toxicodynamic modeling is lifetime tumor incidences (Gaylor and are sufficiently flexible in the observed
potentially the most comprehensive way Zheng, 1996). In this way, range to fit linear and nonlinear
to account for the biological processes toxicodynamic modeling can be used to datasets. Additional judgments and
involved in a response. Such models select and characterize an appropriate perhaps alternative analyses are used
seek to reflect the sequence of key precursor response level (see Section when the procedure fails to yield
precursor events that lead to cancer. 3.2.2, 3.2.5). reliable results. For example, when a
Toxicodynamic models can contribute model’s fit is poor, the highest dose is
to dose-response assessment by 3.2.3. Empirical Modeling (‘‘Curve
often omitted in cases where it is judged
revealing and describing nonlinear Fitting’’)
that the highest dose reflects competing
relationships between internal dose and When a toxicodynamic model is not toxicity that is more relevant at high
cancer response. Such models may available or when the purpose of the doses than at lower doses. Another
provide a useful approach for analysis assessment does not warrant developing example is when there are large
in the range of observation, provided the such a model, empirical modeling differences in survival across dose
purpose of the assessment justifies the (sometimes called ‘‘curve fitting’’) groups; here, models that includes time-
effort involved. should be used in the range of to-tumor or time-to-event information
If a new model is developed for a observation. A model can be fitted to may be useful.
specific agent, extensive data on the data on either tumor incidence or a key For continuous data on key precursor
agent are important for identifying the precursor event. Goodness-of-fit to the effects, empirical models can be chosen
form of the model, estimating its experimental observations is not by on the basis of the structure of the data.
parameters, and building confidence in itself an effective means of The rationale for the choice of model,
its results. Conformance to the observed discriminating among models that the alternatives considered and rejected,
tumor incidence data alone does not adequately fit the data (OSTP, 1985). and a discussion of model uncertainty
establish a model’s validity, as a model Many different curve-fitting models are included in the dose-response
can be designed with a sufficiently large have been developed, and those that fit characterization.
number of parameters so as to fit any the observed data reasonably well may
given dataset. Peer review, including lead to several-fold differences in 3.2.4. Point of Departure (POD)
both an examination of the scientific estimated risk at the lower end of the For each tumor response, a POD from
basis supporting the model and an observed range. Another problem occurs the observed data should be estimated
independent evaluation of the model’s when a multitude of alternatives are to mark the beginning of extrapolation
performance, is an essential part of presented without sufficient context to to lower doses. The POD is an estimated
evaluating the new model. make a reasoned judgment about the dose (expressed in human-equivalent
If a standard model already exists for alternatives. This form of model terms) near the lower end of the
the agent’s mode of action, the model uncertainty reflects primarily the observed range without significant
can be adapted for the agent by using availability of different computer extrapolation to lower doses.
The POD is used as the starting point calculate and present the point estimate tumor data will provide a more reliable
for subsequent extrapolations and of the ED10 [or central estimate] and the POD. Precursor effects may or may not
analyses. For linear extrapolation, the corresponding upper and lower 95% be biologically adverse in themselves;
POD is used to calculate a slope factor statistical bounds.’’ For example, it may the intent is to consider not only tumors
(see Section 3.3.3), and for nonlinear be appropriate to emphasize the central but also damage that can lead to
extrapolation the POD is used in the estimate in activities that involve formal subsequent tumor development by the
calculation of a reference dose or uncertainty analysis that are required by agent. Analysis of continuous data may
reference concentration (see Section OMB Circular A–4 (OMB, 2003) as well differ from discrete data; Murrell et al.
3.3.4). In a risk characterization, the as ranking agents as to their (1998) discuss alternative approaches to
POD is part of the determination of a carcinogenic hazard. Thus, risk deriving a POD from continuous data.
margin of exposure (see Section 5.4). assessors should calculate, to the extent
3.2.5. Characterizing the POD: The POD
With appropriate adjustments, it can practicable, and present the central
Narrative
also be used as the basis for hazard estimate and the corresponding upper
rankings that compare different agents and lower statistical bounds (such as As a single-point summary of a single
or health effects. confidence limits) to inform dose-response curve, the POD alone
The lowest POD is used that is decisionmakers. does not convey all the critical
adequately supported by the data. If the When tumor data are used, a POD is information present in the data from
POD is above some data points, it can obtained from the modeled tumor which it is derived. To convey a
fail to reflect the shape of the dose- incidences. Conventional cancer measure of uncertainty, the POD should
response curve at the lowest doses and bioassays, with approximately 50 be presented as a central estimate with
can introduce bias into subsequent animals per group, generally can upper and lower bounds. A POD
extrapolations (see Figure 3–1). On the support modeling down to an increased narrative summarizes other important
other hand, if the POD is far below all incidence of 1–10%; epidemiologic features of the database and the POD
observed data points, it can introduce studies, with larger sample sizes, below that are important to account for in low-
model uncertainty and parameter 1%. Various models commonly used for dose extrapolations or other analyses.
uncertainty (see Section 3.6) that carcinogens yield similar estimates of (a) Nature of the response. Is the POD
increase with the distance between the the POD at response levels as low as 1% based on tumors or a precursor? If on
data and the POD. Use of a POD at the (Krewski and Van Ryzin, 1981; Gaylor et tumors, does the POD measure
lowest level supported by the data seeks al., 1994). Consequently, response levels incidence or mortality? Is it a lifetime
to balance these considerations. It uses at or below 10% can often be used as measure or was the study terminated
information from the model(s) a small the POD. As a modeling convention, the early? The relationships between
distance below the observed range lower bound on the doses associated precursors and tumors, incidence and
rather than discarding this information with standard response levels of 1, 5, mortality, and lifetime and early-
and using extrapolation procedures in a and 10% can be analyzed, presented, termination results vary from case to
range where the model(s) can provide and considered. For making case. Modeling can provide quantitative
some useful information. Statistical tests comparisons at doses within the insight into these relationships, for
involving the ratio of the central observed range, the ED10 and LED10 are example, linking a change in a precursor
estimate and its lower bound (i.e., EDxx/ also reported and can be used, with response to a tumor incidence (see
LEDxx) can be useful for evaluating how appropriate adjustments, in hazard Section 3.2.2). This can aid in
well the data support a model’s rankings that compare different agents evaluating the significance of the
estimates at a particular response level. or health effects (U.S. EPA, 2002c). A response at the POD and adjusting
(Note that the ability to model at a no-observed-adverse-effect level different PODs to make them
particular response level is not the same (NOAEL) generally is not used for comparable.
as the study’s ability to identify an assessing the potential for carcinogenic (b) Level of the response. What level
increase at that response level as response when one or more models can of response is associated with the POD,
statistically significant.) be fitted to the data. for example, 1% cancer risk, 10%
The POD for extrapolating the When good quality precursor data are cancer risk, or 10% change in a
relationship to environmental exposure available and are clearly tied to the precursor measure?
levels of interest, when the latter are mode of action of the compound of (c) Nature of the study population. Is
outside the range of observed data, is interest, models that include both the POD based on humans or animals?
generally the lower 95% confidence tumors and their precursors may be How large is the effective sample size?
limit on the lowest dose level that can advantageous for deriving a POD. Such Is the study group representative of the
be supported for modeling by the data. models can provide insight into general population, of healthy adult
SAB (1997) suggested that, ‘‘it may be quantitative relationships between workers, or of a susceptible group? Are
appropriate to emphasize lower tumors and precursors (see Section both sexes represented? Did exposure
statistical bounds in screening analyses 3.2.2), possibly suggesting the precursor occur during a susceptible lifestage?
and in activities designed to develop an response level that is associated with a (d) Slope of the dose-response curve
appropriate human exposure value, particular tumor response level. The at the POD. How does response change
since such activities require accounting goal is to use precursor data to extend as dose is reduced below the POD? A
for various types of uncertainties and a the observed range below what can be steep slope indicates that risk decreases
lower bound on the central estimate is observed in tumor studies. EPA is rapidly as dose decreases. On the other
a scientifically-based approach continuing to examine this issue and hand, a steep slope also indicates that
accounting for the uncertainty in the anticipates that findings and errors in an exposure assessment can
true value of the ED10 [or central conclusions may result in supplemental lead to large errors in estimating risk.
estimate].’’ However, the consensus of guidance to these cancer guidelines. If Both aspects of the slope are important.
the SAB (1997) was that, ‘‘both point the precursor data are drawn from small The slope also indicates whether dose-
estimates and statistical bounds can be samples or if the quantitative response curves for different effects are
useful in different circumstances, and relationship between tumors and likely to cross below the POD. For
recommended that the Agency routinely precursors is not well defined, then the example, in the ED01 study where 2-
acetylaminofluorene caused bladder that characterize risk as a probability When the weight of evidence
carcinomas and liver carcinomas in over a range of environmental exposure evaluation of all available data are
mice (Littlefield et al., 1980), the dose- levels. These risk probabilities allow insufficient to establish the mode of
response curves for these tumors cross estimates of the risk reduction under action for a tumor site and when
between 10% and 1% response (see different decision options and estimates scientifically plausible based on the
Figure 3–2). This crossing, which can be of the risk remaining after an action is available data, linear extrapolation is
inferred from the slopes of the curves at taken and provide the risk information used as a default approach, because
a 10% response, shows how considering needed for benefit-cost analyses of linear extrapolation generally is
the slope can lead to better inferences different decision options. considered to be a health-protective
about the predominant effects expected When a dose-response model is not approach. Nonlinear approaches
at lower doses. Mode of action data can developed for lower doses, another form generally should not be used in cases
also be useful; quantitative information of low-dose extrapolation is a safety where the mode of action has not been
about key precursor events can be used assessment that characterizes the safety ascertained. Where alternative
to describe how risk decreases as dose of one lower dose, with no explicit approaches with significant biological
decreases below the POD. characterization of risks above or below support are available for the same tumor
(e) Relationship of the POD with other that dose. Although this type of response and no scientific consensus
cancers. How does the POD for this extrapolation may be adequate for favors a single approach, an assessment
cancer relate to PODs for other cancers evaluation of some decision options, it may present results based on more than
observed in the database? For example, may not be adequate for other purposes one approach.
a POD based on male workers would not (e.g., benefit-cost analyses) that require A nonlinear approach should be
reflect the implications of mammary a quantitative characterization of risks selected when there are sufficient data
tumors in female rats or mice. across a range of doses. At this time, to ascertain the mode of action and
(f) Extent of the overall cancer safety assessment is the default conclude that it is not linear at low
database. Have potential cancer approach for tumors that arise through doses and the agent does not
responses been adequately studied (e.g., a nonlinear mode of action; however, demonstrate mutagenic or other activity
were all tissues examined), or is the EPA continues to explore methods for consistent with linearity at low doses.
database limited to particular effects, Special attention is important when the
quantifying dose-response relationships
population segments, or lifestages? Do data support a nonlinear mode of action
over a range of environmental exposure
the mode of action data suggest a but there is also a suggestion of
levels for tumors that arise through a
potential for cancers not observed in the mutagenicity. Depending on the
nonlinear mode of action (U.S. EPA,
database (e.g., disruption of particular strength of the suggestion of
2002c). EPA program offices that need
endocrine pathways leading to related mutagenicity, the assessment may
this more explicit dose-response
cancers)? justify a conclusion that mutagenicity is
information may develop and apply
not operative at low doses and focus on
3.2.6. Relative Potency Factors methods that are informed by the
a nonlinear approach, or alternatively,
Relative potency factors (of which methods described in these cancer
the assessment may use both linear and
toxicity equivalence factors are a special guidelines.
nonlinear approaches.
case) can be used for a well-defined 3.3.1. Choosing an Extrapolation Both linear and nonlinear approaches
class of agents that operate through a Approach may be used when there are multiple
common mode of action for the same modes of action. If there are multiple
toxic endpoint. A complete dose- The approach for extrapolation below tumor sites, one with a linear and
response assessment is conducted for the observed data considers the another with a nonlinear mode of
one well-studied member of the class understanding of the agent’s mode of action, then the corresponding approach
that serves as the index chemical for the action at each tumor site (see Section is used at each site. If there are multiple
class. The other members of the class 2.4). Mode of action information can modes of action at a single tumor site,
are tied to the index chemical by suggest the likely shape of the dose- one linear and another nonlinear, then
relative potency factors that are based response curve at lower doses. The both approaches are used to decouple
on characteristics such as relative extent of inter-individual variation is and consider the respective
toxicological outcomes, relative also considered, with greater variation contributions of each mode of action in
metabolic rates, relative absorption spreading the response over a wider different dose ranges. For example, an
rates, quantitative SARs, or receptor range of doses. agent can act predominantly through
binding characteristics (U.S. EPA, Linear extrapolation should be used cytotoxicity at high doses and through
2000c). Examples of this approach are when there are MOA data to indicate mutagenicity at lower doses where
the toxicity equivalence factors for that the dose-response curve is expected cytotoxicity does not occur. Modeling to
dioxin-like compounds and the relative to have a linear component below the a low response level can be useful for
potency factors for some carcinogenic POD. Agents that are generally estimating the response at doses where
polycyclic aromatic hydrocarbons. considered to be linear in this region the high-dose mode of action would be
Whenever practicable, toxicity include: less important.
equivalence factors should be validated • Agents that are DNA-reactive and
have direct mutagenic activity, or 3.3.2. Extrapolation Using a
and accompanied by quantitative
• Agents for which human exposures Toxicodynamic Model
uncertainty analysis.
or body burdens are high and near doses The preferred approach is to develop
3.3. Extrapolation to Lower Doses associated with key precursor events in a toxicodynamic model of the agent’s
The purpose of low-dose the carcinogenic process, so that mode of action and use that model for
extrapolation is to provide as much background exposures to this and other extrapolation to lower doses (see
information as possible about risk in the agents operating through a common Section 3.2.2). The extent of
range of doses below the observed data. mode of action are in the increasing, extrapolation is governed by an analysis
The most versatile forms of low-dose approximately linear, portion of the of model uncertainty, where alternative
extrapolation are dose-response models dose-response curve. models that fit similarly in the observed
range can diverge below that range (see 3.3.4. Nonlinear Extrapolation to Lower considered and a judgment made about
Section 3.6). Substantial divergence is Doses how best to represent the human cancer
likely when model parameters are A nonlinear extrapolation method can risk. Some options for presenting results
estimated from tumor incidence data, so be used for cases with sufficient data to include:
that different combinations of parameter • Adding risk estimates derived from
ascertain the mode of action and to
estimates yield similar fits in the different tumor sites (NRC, 1994),
conclude that it is not linear at low
observed range but have different • Combining data from different
doses but with not enough data to
implications at lower doses. An analysis datasets in a joint analysis (Putzrath and
support a toxicodynamic model that
of model uncertainty can be used to Ginevan, 1991; Stiteler et al., 1993;
may be either nonlinear or linear at low
determine the range where extrapolation Vater et al., 1993),
doses. Nonlinear extrapolation having a • Combining responses that operate
using the toxicodynamic model is
significant biological support may be through a common mode of action,
supported and where further
presented in addition to a linear • Representing the overall response
extrapolation would be based on either
approach when the available data and a in each experiment by counting animals
a linear or a nonlinear default, as
weight of evidence evaluation support a with any tumor showing a statistically
appropriate (see Sections 3.3.3, 3.3.4).
nonlinear approach, but the data are not significant increase,
3.3.3. Extrapolation Using a Low-Dose, strong enough to ascertain the mode of • Presenting a range of results from
Linear Model action applying the Agency’s mode of multiple datasets (in this case, the dose-
Linear extrapolation should be used action framework. If the mode of action response assessment includes guidance
in two distinct circumstances: (1) When and other information can support on how to choose an appropriate value
there are data to indicate that the dose- chemical-specific modeling at low from the range),
response curve has a linear component doses, it is preferable to default • Choosing a single dataset if it can be
below the POD, or (2) as a default for a procedures. justified as most representative of the
tumor site where the mode of action is For cases where the tumors arise overall response in humans, or
not established (see Section 3.3.1). For through a nonlinear mode of action, an • A combination of these options.
linear extrapolation, a line should be oral reference dose or an inhalation Cross-comparison of estimates from
drawn from the POD to the origin, reference concentration, or both, should human and animal studies can provide
corrected for background. This implies be developed in accordance with EPA’s a valuable risk perspective.
a proportional (linear) relationship established practice for developing such • Calculating an animal-derived slope
between risk and dose at low doses. values, taking into consideration the factor and using it to estimate the risk
(Note that the dose-response curve factors summarized in the expected in a human study can provide
generally is not linear at higher doses.) characterization of the POD (see Section information with which to evaluate the
The slope of this line, known as the 3.2.5). This approach expands the past human study design, for example,
slope factor, is an upper-bound estimate focus of such reference values adequacy of exposure level and sample
of risk per increment of dose that can be (previously reserved for effects other size.
used to estimate risk probabilities for than cancer) to include carcinogenic • Calculating an upper-bound slope
different exposure levels. The slope effects determined to have a nonlinear factor from a human study that does not
factor is equal to 0.01/LED01 if the LED01 mode of action. As with other health show positive results but that has good
is used as the POD. effects of concern, it is important to put exposure information, and comparing it
Unit risk estimates express the slope cancer in perspective with the overall to an animal-derived slope factor can
in terms of µg/L drinking water or µg/ health impact of an exposure by indicate whether the animal and
m3 or ppm air. In general, the drinking comparing reference value calculations humans studies are consistent.
water unit risk is derived by converting for cancer with those for other health
a slope factor from units of mg/kg-d to 3.4. Extrapolation to Different Human
effects.
units of µg/L, whereas an inhalation Exposure Scenarios
For effects other than cancer,
unit risk is developed directly from a reference values have been described as As described in the previous cancer
dose-response analysis using equivalent being based on the assumption of guidelines, special problems arise when
human concentrations already biological thresholds. The Agency’s the human exposure situation of
expressed in units of µg/m3. Unit risk more current guidelines for these effects concern suggests exposure regimens,
estimates often assume a standard (U.S. EPA, 1996a, 1998b), however, do e.g., route and dosing schedule, that are
intake rate (L/day drinking water or m3/ not use this assumption, citing the substantially different from those used
day air) and body weight (kg), which difficulty of empirically distinguishing a in the relevant animal studies. Unless
may need to be reconciled with the true threshold from a dose-response there is evidence to the contrary in a
exposure factors for the population of curve that is nonlinear at low doses. particular case, the cumulative dose
interest in an exposure assessment (see Economic and policy analysts need to received over a lifetime, expressed as
Section 4.4). Alternatively, when the know how the probability of cancer average daily exposure prorated over a
slope factor for inhalation is in units of varies at exposures above the reference lifetime, is recommended as an
ppm, it may sometimes be termed the value and whether, and to what extent, appropriate measure of exposure to a
inhalation unit risk. Although unit risks there are health benefits from reducing carcinogen. That is, the assumption is
have not been calculated in the past for exposures below the reference value. made that a high dose of a carcinogen
dermal exposures, both exposures that The risk assessment community is received over a short period of time is
are absorbed into the systemic working to develop better methods to equivalent to a corresponding low dose
circulation and those that remain in provide more useful information to spread over a lifetime. This approach
contact with the skin are also important. economic and policy analysts. becomes more problematical as the
Risk-specific doses are derived from exposures in question become more
the slope factor or unit risk to estimate 3.3.5. Comparing and Combining intense but less frequent, especially
the dose associated with a specific risk Multiple Extrapolations when there is evidence that the agent
level, for example, a one-in-a-million When multiple estimates can be has shown dose-rate effects (U.S. EPA
increased lifetime risk. developed, all datasets should be 1986a).
Accordingly, for lifetime human from less-than-lifetime exposure, • In the absence of such agent-
exposure scenarios that involve bringing consistency to the methods specific data, there is some general
intermittent or varying levels of used for dose-response assessment and information to indicate that childhood
exposure, the prevailing practice has exposure assessment in such cases. The can be a susceptible lifestage for
been to assess exposure by calculating a dose-response assessment can provide a exposure to some carcinogens (U.S.
lifetime average daily exposure or dose recommendation to exposure assessors EPA, 2005); this warrants explicit
(U.S. EPA, 1992a). about the averaging time that is consideration in each assessment. The
For less-than-lifetime human appropriate to the mode of action and to potential for susceptibility from early-
exposure scenarios, too, the lifetime the exposure duration of the scenario. life exposure is expected to vary among
average daily exposure or dose has often specific agents and chemical classes. In
been used. The use of these lifetime 3.5. Extrapolation to Susceptible
addition, the concern that the dose-
average exposure metrics was adopted Populations and Lifestages
averaging generally used for assessing
with low-dose linear cancer assessments The dose-response assessment strives less-than-lifetime exposure is more
in mind. The lifetime averaging implies to derive separate estimates for likely to understate than overstate risk
that less-than-lifetime exposure is susceptible populations and lifestages (see Section 3.4) contributes to the
associated with a linearly proportional so that these risks can be explicitly suggestion that alternative approaches
reduction of the lifetime risk, regardless characterized. For a susceptible be considered for assessing risks from
of when exposures occur. Such population, higher risks can be expected less-than-lifetime exposure that occurs
averaging may be problematic in some from exposures anytime during life, but during childhood. Accompanying these
situations. This can be illustrated using this applies to only a portion of the cancer guidelines is the Supplemental
both the multistage model and the two- general population (e.g., those bearing a Guidance that the Agency will use to
stage clonal expansion model that particular genetic susceptibility). In assess risks from early-life exposure to
predict that short-duration risks are not contrast, for a susceptible lifestage, potential carcinogens (U.S. EPA, 2005).
necessarily proportional to exposure higher risks can be expected from The Supplemental Guidance may be
duration and can depend on the nature exposures during only a portion of a updated to reflect new data and new
of the carcinogen and the timing of lifetime, but everyone in the population understanding that may become
exposure (Goddard et al., 1995; may pass through those lifestages. available in the future.
Murdoch et al., 1992). These examples Effects of exposures during a susceptible
indicate some circumstances in which period are not equivalent to effects of 3.6. Uncertainty
use of a lifetime average daily dose exposures at other times; consequently, The NRC (1983, 1994, 1996, 2002) has
(LADD) would underestimate cancer it is useful to estimate the risk repeatedly advised that proper
risk by two-to fivefold, and others in attributable to exposures during each characterization of uncertainty is
which it might overestimate risk period. essential in risk assessment. An
(Murdoch et al., 1992). Thus, averaging Depending on the data available, a assessment that omits or underestimates
over the duration of a lifestage or a tiered approach should be used to uncertainty can leave decisionmakers
critical window of exposure may be address susceptible populations and with a false sense of confidence in
appropriate. As methodological research lifestages. estimates of risk. On the other hand, a
focuses on new approaches for • When there is an epidemiologic high level of uncertainty does not imply
estimating risks from less-than-lifetime study or an animal bioassay that reports that a risk assessment or a risk
exposures, methods and defaults can be quantitative results for susceptible management action should be delayed
expected to change. individuals, the data should be analyzed (NRC, 2002). Uncertainty in dose-
This highlights the importance for to provide a separate risk estimate for response assessment can be classified as
each dose-response assessment to those who are susceptible. If either model uncertainty or parameter
critically evaluate all information susceptibility pertains to a lifestage, it is uncertainty. A related concept, human
pertaining to less-than-lifetime useful to characterize the portion of the variation, is discussed below.
exposure. For example, detailed stop- lifetime risk that can be attributed to the Assessments should discuss the
exposure studies can provide susceptible lifestage. significant uncertainties encountered in
information about the relationship • When there are data on some risk- the analysis, distinguishing, if possible,
between exposure duration, precursor related parameters that allow between model uncertainty, parameter
effects, potential for reversibility, and comparison of the general population uncertainty, and human variation.
tumor development. Toxicokinetic and susceptible individuals, the data Origins of these uncertainties can span
modeling can investigate differences in should be analyzed with an eye toward a range, from a single causal thread
internal dose between short-term and adjusting the general population supported by sparse data, to abundant
long-term exposure or between estimate for susceptible individuals. information that presents multiple
intermittent and constant exposure. This analysis can range from possible conclusions or that does not
Persistence in the body can be useful in toxicokinetic modeling that uses coalesce. As described in Section 2.6
explaining long-term effects resulting parameter values representative of and in Section 5.1, all contributing
from shorter-term exposures. susceptible individuals to more simply features should be noted.
For nonlinear cancer analyses, it may adjusting a general population estimate Model uncertainty refers to a lack of
be appropriate to assess exposure by to reflect differences in important rate- knowledge needed to determine which
calculating a daily dose that is averaged governing parameters. Care is taken to is the correct scientific theory on which
over the exposure duration for the study not make parameter adjustments in to base a model. In risk assessment,
(see Section 3.1.1). For example, when isolation, as the appropriate adjustment model uncertainty is reflected in
the analysis is based on precursor can depend on the interactions of alternative choices for model structure,
effects that result from less than a several parameters; for example, the dose metrics, and extrapolation
lifetime exposure, that exposure period ratio of metabolic activation and approaches. Other sources of model
may be used. This reflects an clearance rates can be more appropriate uncertainty concern whether surrogate
expectation that the precursor effects on than the activation rate alone (U.S. EPA, data are appropriate, for example, using
which the analysis is based can result 1992b). data on adults to make inferences about
children. The full extent of model characterized, but not reduced, by cancer guidelines are intended to be
uncertainty usually cannot be further research. Fields other than risk flexible enough to incorporate
quantified; a partial characterization can assessment use ‘‘variation’’ or additional approaches for characterizing
be obtained by comparing the results of ‘‘variability’’ to mean dispersion about a uncertainty that have less commonly
alternative models. Model uncertainty is central value, including measurement been used by regulatory agencies. In all
expressed through comparison of errors and other random errors that risk scientific and engineering fields, data
separate analyses from each model, assessors address as uncertainty. and research limitations often limit the
coupled with a subjective probability Probabilistic risk assessment, application of established methods. A
statement, where feasible and informed by expert judgment, has been dearth of data is a particular problem
appropriate, of the likelihood that each used in exposure assessment to estimate when quantifying the probability
model might be correct (NRC, 1994). human variation and uncertainty in distribution of model outputs. In many
Some aspects of model uncertainty lifetime average daily exposure of these scientific and engineering
that should be addressed in an concentration or dose. Probabilistic disciplines, researchers have used
assessment include the use of animal methods can be used in this exposure rigorous expert elicitation methods to
models as a surrogate for humans, the assessment application because the overcome the lack of peer-reviewed
influence of cross-species differences in pertinent variables (for example, methods and data. Although expert
metabolism and physiology, the use of concentration, intake rate, exposure elicitation has not been widely used in
effects observed at high doses as an duration, and body weight) have been environmental risk assessment, several
indicator of the potential for effects at identified, their distributions can be studies have applied this methodology
lower doses, the effect of using linear or observed, and the formula for as a tool for understanding quantitative
nonlinear extrapolation to estimate combining the variables to estimate the risk. For example, expert elicitation has
risks, the use of using small samples lifetime average daily dose is well been used in chemical risk assessment
and subgroups to make inferences about defined (see U.S. EPA, 1992a). and its associated uncertainty (e.g.,
entire human populations or Similarly, probabilistic methods can be Richmond, 1981; Renn, 1999; Florig et
subpopulations with differential applied in dose-response assessment al., 2001; Morgan et al., 2001; Willis et
susceptibilities, and the use of when there is an understanding of the al., 2004), components of risk
experimental exposure regimens to important parameters and their assessment such as hazard assessment
make inferences about different human relationships, such as identification of and dose-response evaluation (e.g.,
exposure scenarios (NRC, 2002). the key determinants of human Hawkins and Graham 1988; Jelovsek et
Toxicokinetic and toxicodynamic variation (for example, metabolic al., 1990; Evans et al., 1994; IEc, 2004;
models are generally premised on site polymorphisms, hormone levels, and U.S. EPA 2004) and exposure
concordance across species, modeling, cell replication rates), observation of the assessment (e.g., Whitfield and
for example, the relationship between distributions of these variables, and Wallsten, 1989; Hawkins and Evans,
administered dose and liver tissue valid models for combining these 1989; Winkler et al., 1995; Stiber et al.,
concentrations to predict increased variables. With appropriate data and 1999; Walker et al., 2001, 2003; Van Der
incidences of liver cancer. This expert judgment, formal approaches to Fels-Klerx et al., 2002), and for
relationship, which can be observed in probabilistic risk assessment can be evaluating other types of risks (e.g.,
animals, is typically only inferred for applied to provide insight into the North and Merkhofer, 1976; Fos and
humans. There are, however, numerous overall extent and dominant sources of McLin, 1990). These cancer guidelines
examples of an agent causing different human variation and uncertainty. In
are flexible enough to accommodate the
cancers in different species. The doing this, it is important to note that
use of expert elicitation to characterize
assessment should discuss the relevant analyses that omit or underestimate
data that bear on this form of model cancer risks, as a complement to the
some principal sources of variation or
uncertainty. methods presented in the cancer
uncertainty could provide a
Parameter uncertainty refers to a lack guidelines. According to NRC (NRC,
misleadingly narrow description of the
of knowledge about the values of a 2002), the rigorous use of expert
true extent of variation and uncertainty
model’s parameters. This leads to a elicitation for the analyses of risks is
and give decisionmakers a false sense of
distribution of values for each considered to be quality science.
confidence in estimates of risk.
parameter. Common sources of Specification of joint probability 3.7. Dose-Response Characterization
parameter uncertainty include random distributions is appropriate when
measurement errors, systematic A dose-response characterization
variables are not independent of each
measurement errors, use of surrogate extracts the dose-response information
other. In each case, the assessment
data instead of direct measurements, needed in a full risk characterization
should carefully consider the questions
misclassification of exposure status, (U.S. EPA, 2000b), including:
of uncertainty and human variation and
random sampling errors, and use of an discuss the extent to which there are • Presentation of the recommended
unrepresentative sample. Most types of data to address them. estimates (slope factors, reference doses,
parameter uncertainty can be quantified Probabilistic risk assessment has also reference concentrations) and
by statistical analysis. been used in dose-response assessment alternatives with significant biological
Human variation refers to person-to- to determine and distinguish the degree support,
person differences in biological of uncertainty and variability in • A summary of the data supporting
susceptibility or in exposure. Although toxicokinetic and toxicodynamic these estimates,
both human variation and uncertainty modeling. Although this field is less • A summary and explanation of the
can be characterized as ranges or advanced that probabilistic exposure modeling approaches used,
distributions, they are fundamentally assessment, progress is being made and • A description of any special
different concepts. Uncertainty can be these cancer guidelines are flexible features such as the development and
reduced by further research that enough to accommodate continuing consolidation of multiple estimates as
supports a model or improves a advances in these approaches. detailed in Section 3.3.5,
parameter estimate, but human variation Advances in uncertainty analysis are • The POD narrative (see Section
is a reality that can be better expected as the field develops. The 3.2.5),
• A summary of the key defaults individuals, although the calculation of • The study design does not include
invoked, upper bounds is not based on exposure during a susceptible lifestage,
• Identification of susceptible susceptibility data. Similarly, exposure for example, perinatal exposure.
populations or lifestages and during some lifestages can contribute • The study population is of less-
quantification of their differential more or less to the total lifetime risk than-average susceptibility, for example,
susceptibility, and than do similar exposures at other healthy adult workers.
• A discussion of the strengths and times. The dose-response assessment • There is random exposure
limitations of the dose-response characterizes, to the extent possible, the misclassification or random exposure
assessment, highlighting significant extent of these variations. measurement error in the study from
issues in developing risk estimates, Depending on the supporting data and
which the slope factor is derived.
alternative approaches considered modeling approach, a slope factor can
have a mix of traits that tend to either Some examples of traits that
equally plausible, and how these issues inherently neither overestimate nor
were resolved. estimate, overestimate, or underestimate
risk. underestimate risk include the
All estimates should be accompanied following.
Some examples of traits that tend to
by the weight of evidence descriptor and • The slope factor is derived from
overestimate risk include the following.
its narrative (see Section 2.5) to convey • The slope factor is derived from data in humans or in an animal strain
a sense of the qualitative uncertainty data on a highly susceptible animal that responds like humans.
about whether the agent may or may not
be carcinogenic.
strain. • Linear extrapolation is appropriate
• Linear extrapolation is used as a for the agent’s mode of action.
Slope factors generally represent an default and extends over several orders
upper bound on the average risk in a • Environmental exposures are close
of magnitude. to the observed data.
population or the risk for a randomly • The largest of several slope factors
selected individual but not the risk for is chosen. • Several slope factors for the same
a highly susceptible individual or Some examples of traits that tend to tumor are averaged or a slope factor is
group. Some individuals face a higher underestimate risk include the derived from pooled data from several
risk and some face a lower risk. The use following. studies.
of upper bounds generally is considered • Several tumor types were observed, • The slope factor is derived from the
to be a health-protective approach for but the slope factor is based on a subset only suitable study.
covering the risk to susceptible of them. BILLING CODE 6560–50–P
BILLING CODE 6560–50–C

EN07AP05.001</GPH>
4. Exposure Assessment important differences in exposures. may need to be assessed as a special

Exposure assessment is the See, for example, the discussion in subgroup. For further guidance, see
determination (qualitative and Exposure Factors Handbook (U.S. Guidelines for Exposure Assessment
quantitative) of the magnitude, EPA, 1997c, Appendix 1A). (U.S. EPA, 1992a, § 5.3.5.2 ). As
—Pregnant and lactating women may methodological research focuses on new
frequency, and duration of exposure and
have exposures that differ from the approaches for estimating risks from
internal dose (U.S. EPA, 1992a). This
general population (e.g., slightly less-than-lifetime exposures, methods
section provides a brief overview of
higher water consumption) (U.S. EPA, and defaults can be expected to change.
exposure assessment principles, with an
1997c). Further, exposure to pregnant There may be cases where the mode
emphasis on issues related to
women may result in exposure to the of action indicates that dose rates are
carcinogenic risk assessment. The
developing fetus (NRC, 1993b). important in the carcinogenic process.
information presented here should be —Children consume more food per In these cases, short-term, less-than-
used in conjunction with other guidance body weight than do adults while lifetime exposure estimates may be
documents, including Guidelines for consuming fewer types of foods, i.e., more appropriate than the LADD for risk
Exposure Assessment (U.S. EPA, 1992a), have a more limited diet (ILSI, 1992; assessment. This may be the case when
Science Policy Council Handbook: Risk NRC, 1993b; U.S. EPA, 1997c). In a nonlinear dose-response approach is
Characterization (U.S. EPA, 2000b), addition, children engage in crawling used (see Section 3.3.4).
Exposure Factors Handbook (U.S. EPA, and mouthing (i.e., putting hands and
1997c), the 1997 Policy for Use of objects in the mouth) behaviors, 4.3. Collecting Data or Selecting and
Probabilistic Analysis in Risk which can increase their exposures. Evaluating Available Data
Assessments (U.S. EPA, 1997d), and the —The elderly and disabled may have After the assessment questions have
1997 Guiding Principles for Monte Carlo important differences in their been defined and the conceptual and
Analysis (U.S. EPA, 1997e). In addition, exposures due to a more sedentary mathematical models have been
program-specific guidelines for lifestyle (U.S. EPA, 1997c). In developed, it is important to compile
exposure assessment should be addition, the health status of this and evaluate existing data or, if
consulted. group may affect their susceptibility necessary, to collect new data.
Exposure assessment generally to the detrimental effects of exposure. Depending on the exposure scenario
consists of four major steps: defining the For further guidance, see Guidelines under consideration, data on a wide
assessment questions, selecting or for Exposure Assessment (U.S. EPA, variety of exposure factors may be
developing the conceptual and 1992a, § 3). needed. EPA’s Exposure Factors
mathematical models, collecting data or Handbook (U.S. EPA, 1997c) contains a
selecting and evaluating available data, 4.2. Selecting or Developing the large compilation of exposure data, with
and exposure characterization. Each of Conceptual and Mathematical Models some analysis and recommendations.
these steps is briefly described below. Carcinogen risk assessment models Some of these data are organized by age
4.1. Defining the Assessment Questions have generally been based on the groups to assist with assessing such
premise that risk is proportional to subgroups as children. See, for example,
In providing a clear and unambiguous cumulative lifetime dose. For lifetime Exposure Factors Handbook (U.S. EPA,
statement of the purpose and scope of human exposure scenarios, therefore, 1997c, Volume 1, Chapter 3). When
the exposure assessment (U.S. EPA, the exposure metric used for using these existing data, it is important
1997e), consider the following. carcinogenic risk assessment has been to evaluate the quality of the data and
• The management objectives of the the lifetime average daily dose (LADD) the extent to which the data are
assessment will determine whether or, in the case of inhalation exposure, representative of the population under
deterministic screening level analyses the lifetime average exposure consideration. EPA’s (U.S. EPA, 2000d)
are adequate or whether full concentration. These metrics are and OMB’s (OMB 2002) guidance on
probabilistic exposure characterization typically used in conjunction with the information quality, as well as program-
is needed. corresponding slope factor to calculate specific guidances can provide further
• Identify and include all important individual excess cancer risk. The assistance for evaluating existing data.
sources (e.g., pesticide applications), LADD is typically an estimate of the When existing data fail to provide an
pathways (e.g., food or water), and daily intake of a carcinogenic agent adequate surrogate for the needs of a
routes (e.g., ingestion, inhalation, and throughout the entire life of an particular assessment, it is important to
dermal) of exposure in the assessment. individual, while the lifetime average collect new data. Such data collection
If a particular source, pathway, or route exposure concentration is the efforts should be guided by the
is omitted, a clear and transparent corresponding estimate of average references listed above (e.g., Guidance
explanation should be provided. exposure concentration for the for Data Quality Assessment and
• Separate analyses should be carcinogenic agent over the entire life of program-specific guidance). Once again,
conducted for each definable subgroup an individual. Depending on the subpopulations or lifestages of concern
within the population of interest. In objectives of the assessment, the LADD are an important consideration in any
particular, subpopulations or lifestages or lifetime average exposure data collection effort.
that are believed to be highly exposed concentration may be calculated
or susceptible to a particular health deterministically (using point estimates 4.3.1. Adjusting Unit Risks for Highly
effect should be studied. These include for each factor to derive a point estimate Exposed Populations and Lifestages
people with certain diseases or genetic of the exposure) or stochastically (using Unit risk estimates that have been
susceptibilities and others whose probability distributions to represent developed in the dose-response
behavior or physiology may lead to each factor and such techniques as assessment often assumed standard
higher exposure or susceptibility. Monte Carlo analysis to derive a adult intake rates. When an exposure
Consider the following examples: distribution of the LADD) (U.S. EPA, assessment focuses on a population or
—Physiological differences between 1997e). Stochastic analyses may help to lifestage with differential exposure,
men and women (e.g., body weight identify certain population segments or good exposure assessment practice
and inhalation rate) may lead to lifestages that are highly exposed and would replace the standard intake rates
with values representative of the It identifies and compares the 5. Risk Characterization
exposed population. Small changes in contribution of different sources,
5.1. Purpose
exposure assessments can be pathways, and routes of exposure. In
approximated by using linearly particular, a qualitative discussion of EPA has developed general guidance
proportional adjustments of exposure the strengths and limitations on risk characterization for use in its
parameters, but a more accurate (uncertainties) of the data and models risk assessment activities. The core of
integrative analysis may require an are presented. EPA’s risk characterization policy (U.S.
analysis stratified by exposure duration The discussion of uncertainties is a EPA, 2000b, 1995) includes the
(see Section 5.1). critical component of the exposure following.
characterization. Uncertainties can arise Each risk assessment prepared in support
For example, to adjust the drinking water
unit risk for an active population that drinks out of problems with the conceptual and of decision making at EPA should include a
4 L/day (instead of 2 L/day), multiply the mathematical models. Uncertainties can risk characterization that follows the
unit risk by 2. also arise from poor data quality and principles and reflects the values outlined in
data that are not quite representative of this policy. A risk characterization should be
Because children drink more water prepared in a manner that is clear,
the population or scenario of interest.
relative to their body weight than do transparent, reasonable, and consistent with
Consider the following examples of
adults (U.S. EPA, 2002d), adjustments to other risk characterizations of similar scope
uncertainties.
unit risk estimates are warranted • National data (i.e., data collected to prepared across programs in the Agency.
whenever they are applied in an represent the entire U.S. population) Further, discussion of risk in all EPA reports,
assessment of childhood exposure. presentations, decision packages, and other
may not be representative of exposures documents should be substantively
For example, to adjust the drinking water occurring within a regional or local consistent with the risk characterization. The
unit risk for a 9-kg infant who drinks 1 L/ population. nature of the risk characterization will
day (instead of a 70-kg adult who drinks 2 • Use of short-term data to infer depend upon the information available, the
L/day), multiply the unit risk by [(1 L/day) chronic, lifetime exposures should be regulatory application of the risk
/ (9 kg)] / [(2 L/day) / (70 kg)] = 3.9.
done with caution. Use of short-term information, and the resources (including
Inhalation dosimetry is employed to data to estimate long-term exposures has time) available. In all cases, however, the
derive the human equivalent exposure the tendency to underestimate the assessment should identify and discuss all
concentrations on which inhalation unit number of people exposed while the major issues associated with determining
risks, and reference concentrations, are overestimating the exposure levels the nature and extent of the risk and provide
commentary on any constraints limiting
based (U.S. EPA, 1994). As described experienced by those in the upper end
fuller exposition.
previously (see Sections 3.1.2, 3.1.3), (i.e., above the 90th percentile) of the
different dosimetry methods may be exposure distribution. For further Risk characterization should be
employed depending on the availability guidance, refer to Guidelines for carried out in accordance with the EPA
of relevant data and chemical-specific Exposure Assessment (U.S. EPA, 1992a, (U.S. EPA, 2002a) and OMB (2002)
characteristics of the pollutant. § 5.3.1). information quality guidelines. EPA’s
Consideration of lifestage-particular • Children’s behavior, including their risk characterization handbook (U.S.
physiological characteristics in the more limited diet, may lead to relatively EPA, 2000b) provides detailed guidance
dosimetry analysis may result in a high but intermittent exposures. This to Agency staff. The discussion below
refinement to the human equivalent pattern of exposure, ‘‘one that gradually does not attempt to duplicate this
concentration (HEC) to insure relevance declines over the developmental period material, but it summarizes its
in risk assessment across lifestages, or and which remains relatively constant applicability to carcinogen risk
might conceivably conclude with thereafter’’ is not accounted for in the assessment.
multiple HECs, and corresponding LADD model (ILSI, 1992). Further, the The risk characterization includes a
inhalation unit risk values (e.g., separate physiological characteristics of children summary for the risk manager in a
for childhood and adulthood). may lead to important differences in nontechnical discussion that minimizes
The dose-response assessment exposure. Some of these differences can the use of technical terms. It is an
discusses the key sources of uncertainty be accounted for in the LADD model. appraisal of the science that informs the
in estimating dosimetry, including any For further guidance, see Guidelines for risk manager in public health decisions,
related to lifestage. Review of this Exposure Assessment (U.S. EPA, 1992a, as do other decision-making analyses of
discussion and of the dosimetric § 5.3.5.2). economic, social, or technology issues.
analysis performed in deriving the HEC Overall, the exposure characterization It also serves the needs of other
and resultant unit risk will assist in the should provide a full description of the interested readers. The summary is an
appropriate application of inhalation sources, pathways, and routes of information resource for preparing risk
unit risk values to exposure across exposure. The characterization also communication information, but being
lifestages. should include a full description of the somewhat more technical than desired
populations assessed. In particular, for communication with the general
4.4. Exposure Characterization highly exposed or susceptible public, is not itself the usual vehicle for
The exposure characterization is a subpopulation or lifestage should be communication with every audience.
technical characterization that presents discussed. For further guidance on the The risk characterization also brings
the assessment results and supports the exposure characterization, consult together the assessments of hazard, dose
risk characterization. It provides a Guidelines for Exposure Assessment response, and exposure to make risk
statement of the purpose, scope, and (U.S. EPA, 1992a), the Policy and estimates for the exposure scenarios of
approach used in the assessment, Guidance for Risk Characterization interest. This analysis that follows the
identifying the exposure scenarios and (U.S. EPA, 2000b,1995) and EPA’s Rule summary is generally much more
population subgroups covered. It Writer’s Guide to Executive Order 13045 extensive. It typically will identify
provides estimates of the magnitude, (especially Attachment C: Technical exposure scenarios of interest in
frequency, duration, and distribution of Support for Risk Assessors— decision making and present risk
exposures among members of the Suggestions for Characterizing Risks to analyses associated with them. Some of
exposed population as the data permit. Children [U.S. EPA, 1998d]). the analyses may concern scenarios in
several media; others may examine, for if a choice is a significant issue, it is typically rely on animal data because
example, only drinking water risks. As highlighted in the summary. In human data are rarely available. The
these cancer guidelines allow different situations where there are alternative objective of characterization of the use
hazard characterizations and different approaches for a risk assessment that of animal data is not to recount generic
potencies for specified conditions, e.g., have significant biological support, the issues about interpreting and using
exposure level, route of exposure, or decisionmaker can be informed by the animal data; Agency guidance
lifestage, some of the integrative presentation of these alternatives along documents cover these issues. Rather,
analyses may need to be stratified to with their strengths and uncertainties. the objective is to highlight any
accommodate the appropriate significant issues that arose within the
5.2. Application
combinations of parameters across particular assessment being
relevant exposure durations. Risk characterization is a necessary characterized and inform the reader
In constructing high end estimates of part of generating any Agency report on about significant uncertainties that
risk, the assessor should bear in mind risk, whether the report is preliminary— affect conclusions.
that the high-end risk is a plausible to support allocation of resources
estimate of the risk for those persons at toward further study—or 5.3. Presentation of the Risk
the upper end of the risk distribution comprehensive—to support regulatory Characterization Summary
(U.S. EPA, 1992a). The intent of this decisions. In the former case, the detail The presentation is a nontechnical
approach is to convey an estimate of and sophistication of the discussion of important conclusions,
risk in the upper range of the characterization are appropriately small issues, and uncertainties that uses the
distribution, but to avoid estimates that in scale; in the latter case, appropriately hazard, dose response, exposure, and
are beyond the true distribution. Overly extensive. Even if a document covers integrative analyses for technical
conservative assumptions, when only parts of a risk assessment (hazard support. The primary technical supports
combined, can lead to unrealistic and dose-response analyses, for within the risk assessment are the
estimates of risk. This means that when instance), the results of these are hazard characterization, dose-response
constructing estimates from a series of characterized. characterization, and exposure
factors (e.g., emissions, exposure, and Risk assessment is an iterative process characterization described in these
unit risk estimates) not all factors that grows in depth and scope in stages cancer guidelines. The risk
should be set to values that maximize from screening for priority making to characterization is derived from these.
exposure, dose, or effect, since this will preliminary estimation to fuller The presentation should fulfill the aims
almost always lead to an estimate that examination in support of complex outlined in the purpose section above.
is above the 99th-percentile confidence regulatory decision making. Default
options may be used at any stage, but 5.4. Content of the Risk Characterization
level and may be of limited use to
they are predominant at screening stages Summary
decisionmakers. This is particularly
problematic when using unbounded and are used less as more data are Specific guidance on hazard, dose-
lognormal factor distributions. gathered and incorporated at later response, and exposure characterization
While it is an appropriate aim to stages. Various provisions in EPA- appears in previous sections. Overall,
assure protection of health and the administered statutes require decisions the risk characterization routinely
environment in the face of scientific based on differing findings for which includes the following, capturing the
uncertainty, common sense, reasonable differing degrees of analysis are important items covered in hazard, dose
applications of assumptions and policy, appropriate. There are close to 30 response, and exposure
and transparency are essential to avoid provisions within the major statutes that characterization:
unrealistically high estimates. It is also require decisions based on risk, hazard, • Primary conclusions about hazard,
important to inform risk managers of the or exposure assessment. For example, dose response, and exposure, including
final distribution of risk estimates (U.S. Agency review of pre-manufacture alternatives with significant biological
EPA, 2000b; 1995). Otherwise, risk notices under Section 5 of the Toxic support;
management decisions may be made on Substances Control Act relies on • Nature of key supporting
varying levels of conservatism, leading screening analyses, whereas information and analytic methods;
to misplaced risk priorities and requirements for industry testing under • Risk estimates and their attendant
potentially higher overall risks. (Nichols Section 4 of that Act rely on preliminary uncertainties, including key uses of
and Zeckhauser,1986; Zeckhauser and analyses of risk or simply of exposure. default options when data are missing
Viscusi,1990). In comparison, air quality criteria under or uncertain.
The risk characterization presents an the Clean Air Act rest on a rich data —With linear extrapolations, risk below
integrated and balanced picture of the collection and are required by statute to the POD is typically approximated by
analysis of the hazard, dose-response, undergo periodic reassessment. There multiplying the slope factor by an
and exposure. The risk analyst should are provisions that require ranking of estimate of exposure, i.e., Risk = Slope
provide summaries of the evidence and hazards of numerous pollutants—which Factor × Exposure. For exposure
results and describe the quality of may be addressed through a screening levels above the POD, the dose-
available data and the degree of level of analysis—and other provisions response model is used instead of this
confidence to be placed in the risk for which a full assessment of risk is approximation.
estimates. Important features include more appropriate. —With nonlinear extrapolations, the
the constraints of available data and the Given this range in the scope and method of risk assessment depends on
state of knowledge, significant scientific depth of analyses, not all risk the procedure used. If a nonlinear
issues, and significant science and characterizations can or should be equal dose-response function has been
science policy choices that were made in coverage or depth. The risk assessor determined, it can be used with the
when alternative interpretations of data should carefully decide which issues in expected exposure to estimate a risk.
exist (U.S. EPA, 1995, 2000b). Choices a particular assessment are important to If an RfD or RfC was calculated, the
made about using data or default present, choosing those that are hazard can be expressed as a hazard
options in the assessment are explicitly noteworthy in their impact on results. quotient (HQ), defined as the ratio of
discussed in the course of analysis, and For example, health effect assessments an exposure estimate over the
reference dose (RfD) or reference include, where practicable, expected or resulting data are predictive of cancer in
concentration (RfC), i.e., HQ = central estimates of risk, as well as any other exposed human population.
Exposure / (RfD or RfC). From the upper and lower bounds, e.g., Most studies investigating cancer
hazard quotient, it can generally be confidence limits, based on the POD, if outcomes in humans from exposure to
inferred whether the nonlinear mode not a full characterization of uncertainty agents are often studies of
of action is relevant at the of the risk. As discussed in EPA’s occupationally exposed humans. By sex,
environmental exposure level in Guidelines for Ensuring and Maximizing age, and general health, workers may
question; the Quality, Objectivity, Utility, and not be representative of the general
• Statement of the extent of Integrity of Information Disseminated by population exposed environmentally to
extrapolation of risk estimates from the Environmental Protection Agency the same agents. In such studies there is
observed data to exposure levels of (Appendix B), statutory mandates, such no opportunity to observe
interest and its implications for as the Safe Drinking Water Act, the subpopulations who are likely to be
certainty or uncertainty in quantifying Food Quality Protection Act, and the
under represented, such as fetuses,
risk. The extent of extrapolation can be Clean Air Act, call for the Agency to
infants and children, women, or people
expressed as a margin of exposure generate specific kinds of risk
(MOE), defined as the ratio of the POD information, and thus these updated in poor health, who may respond
over an exposure estimate (MOE = POD cancer assessment guidelines should be differently from healthy workers.
/ Exposure); read in conjunction with the Agency’s Therefore, it is understood that this
• Significant strengths and statutory mandates regarding risk option could still underestimate the
limitations of the data and analyses, assessment. response of certain human
including any major peer review issues; subpopulations (NRC, 1993b, 1994).
• Appropriate comparison with Appendix A: Major Default Options
When cancer effects are not found in
similar EPA risk analyses or common This discussion covers the major an exposed human population, this
risks with which people may be default options commonly employed information by itself is not generally
familiar; and when data are missing or sufficiently sufficient to conclude that the agent
• Comparison with all appropriate uncertain in a cancer risk assessment, as poses no carcinogenic hazard to this or
assessments of the same problem by adopted in these cancer guidelines.
others. other populations of potentially exposed
These options are predominantly
It is often difficult to know a priori humans, including susceptible
inferences that help use the data
when or how different results of a observed under empirical conditions in subpopulations or lifestages. This is
cancer risk assessment are likely to be order to estimate events and outcomes because epidemiologic studies often
used by Agency economists, policy under environmental conditions. have low power to detect and attribute
analysts, and decisionmakers, so it is Several inferential issues arise when responses and typically evaluate cancer
important that the resulting effects seen in a subpopulation of potential in a restricted population (e.g.,
characterizations include the necessary humans or animals are used to infer by age, healthy workers). The topic of
information for these analyses to the potential effects in the population of susceptibility and variation is addressed
extent practicable. OMB and EPA environmentally exposed humans. further in the discussion below of
guidelines for benefit-cost analysis Several more inferential issues arise in quantitative default options about dose-
require expected or central estimates of extrapolating the exposure-effect response relationships. Well-conducted
risk and information on the uncertainty relationship observed empirically to studies that fail to detect a statistically
of the estimate when it is possible or lower-exposure environmental significant positive association,
practicable. The extent of the conditions. The following issues cover however, may have value and should be
uncertainty information needed for the major default areas. judged on their merits, including
analysis depends, in part, on the scale • Is the presence or absence of effects population size, duration of the study,
of the policy being considered, with observed in a human population the quality of the exposure
formal quantitative analysis of predictive of effects in another exposed characterization and measures of
uncertainty being required in some human population? outcome, and the magnitude and
cases.6 OMB Circular A–4 (OMB, 2003) • Is the presence or absence of effects duration of the exposure.
emphasizes that agencies ‘‘should try to observed in an animal population
provide some estimate of the probability predictive of effects in exposed There is not yet enough knowledge to
distribution of regulatory benefits and humans? form a basis for any generally applicable
costs.’’ These OMB guidelines note, • How do metabolic pathways relate qualitative or quantitative inference to
‘‘Whenever it is possible to characterize across species and among different age compensate for the gap in knowledge
quantitatively the probability groups and between sexes in humans? concerning other populations. In these
distribution, some estimates of expected • How do toxicokinetic processes cancer guidelines, this problem is left to
value * * * must be provided in relate across species and among analysis in individual cases, to be
addition to ranges, variances, specified different age groups and between sexes attended to with further general
low-end and high-end percentile in humans? guidance as future research and
estimates, and other characteristics of • What is the relationship between information allow. When information
the distribution.’’ The risk the observed dose-response relationship on a susceptible subpopulation or
characterization should therefore to the relationship at lower doses? lifestage exists, it will be used. For
Is the Presence or Absence of Effects example, an agent such as
6 Specifically, OMB guidelines state: ‘‘For rules Observed in a Human Population diethylstilbestrol (DES) causes a rare
that exceed the $1 billion annual [economic effects]
threshold, a formal quantitative analysis of Predictive of Effects in Another Exposed form of vaginal cancer (clear-cell
uncertainty is required. For rules with annual Human Population? adenocarcinoma) (Herbst et al., 1971) in
benefits and/or costs in the range from 100 million about 1 per 1000 of adult women whose
to $1 billion, you should seek to use more rigorous
When cancer effects in exposed
approaches with higher consequence rules.’’ (OMB, humans are attributed to exposure to an mothers were exposed during pregnancy
2003, page 158) agent, the default option is that the (Hatch et al., 1998).
Is the Presence or Absence of Effects the highest dose may be a consequence Target organ concordance is not a
Observed in an Animal Population of cell killing with compensatory cell prerequisite for evaluating the
Predictive of Effects in Exposed replication or of general physiological implications of animal study results for
Humans? disruption rather than inherent humans. Target organs of carcinogenesis
The default option is that positive carcinogenicity of the tested agent. for agents that cause cancer in both
effects in animal cancer studies indicate There is little doubt that this may animals and humans are most often
that the agent under study can have happen in some cases, but skepticism concordant at one or more sites
carcinogenic potential in humans. Thus, exists among some scientists that it is a (Tomatis et al., 1989; Huff, 1994).
if no adequate human or mode of action pervasive problem (Ames and Gold, However, concordance by site is not
data are present, positive effects in 1990; Melnick et al., 1993; Barrett, uniform. The mechanisms of control of
animal cancer studies are a basis for 1993). If adequate data demonstrate that cell growth and differentiation are
assessing the carcinogenic hazard to the effects are solely the result of concordant among species, but there are
excessive toxicity rather than marked differences among species in the
humans. This option is a public health-
carcinogenicity of the tested agent per way control is managed in various
protective policy, and it is both
se, then the effects may be regarded as tissues. For example, in humans,
appropriate and necessary, given that
not appropriate to include in assessment mutations of the tumor suppressor genes
we do not test for carcinogenicity in
of the potential for human p53 and retinoblastoma are frequently
humans. The option is supported by the
carcinogenicity of the agent. This is a observed genetic changes in tumors.
fact that nearly all of the agents known
matter of expert judgment, with These tumor-suppressor genes are also
to cause cancer in humans are
consideration given to all of the data observed to be operating in some rodent
carcinogenic in animals in tests that
available about the agent, including tissues, but other growth control
have adequate protocols (IARC, 1994;
effects in other toxicity studies, mechanisms predominate in other
Tomatis et al., 1989; Huff, 1994). structure-activity relationships, and rodent tissues. Thus, an animal
Moreover, almost one-third of human effects on growth control and response may be due to changes in a
carcinogens were identified subsequent differentiation. control that are relevant to humans but
to animal testing (Huff, 1993). Further When cancer effects are not found in appear in animals in a different way.
support is provided by research on the well-conducted animal cancer studies in However, it is appropriate under these
molecular biology of cancer processes, two or more appropriate species and cancer guidelines to consider the
which has shown that the mechanisms other information does not support the influences of route of exposure,
of control of cell growth and carcinogenic potential of the agent, metabolism, and, particularly, some
differentiation are remarkably these data provide a basis for modes of action that may either support
homologous among species and highly concluding that the agent is not likely to or not support target organ concordance
conserved in evolution. Nevertheless, possess human carcinogenic potential, between animals and humans. When
the same research tools that have in the absence of human data to the data allow, these influences are
enabled recognition of the nature and contrary. This default option about lack considered in deciding whether agent-,
commonality of cancer processes at the of cancer effects has limitations. It is species-, or organ-specific situations are
molecular level also have the power to recognized that animal studies (and appropriate to use in preference to this
reveal differences and instances in epidemiologic studies as well) have very default assumption (NRC, 1994). In
which animal responses are not relevant low power to detect cancer effects. contrast, use of toxicokinetic modeling
to humans (Lijinsky, 1993; U.S. EPA, Detection of a 10% tumor incidence is inherently assumes site concordance, as
1991b). Under these cancer guidelines, generally the limit of power with these models are used to estimate
available mode of action information is standard protocols for animal studies delivered dose to a particular tissue or
studied for its implications in both (with the exception of rare tumors that organ in humans on the basis of the
hazard and dose-response assessment are virtually markers for a particular same tissue or organ from animal data.
and its ability to obviate default options. agent, e.g., angiosarcoma caused by The default is to include benign
There may be instances in which the vinyl chloride). In some situations, the tumors observed in animal studies in
use of an animal model would identify tested animal species may not be the assessment of animal tumor
a hazard in animals that is not truly a predictive of effects in humans; for incidence, if such tumors have the
hazard in humans (e.g., the alpha-2u- example, arsenic shows only minimal or capacity to progress to the malignancies
globulin association with renal no effect in animals, whereas it is with which they are associated. This
neoplasia in male rats [U.S. EPA, clearly positive in humans. Therefore, it default is consistent with the approach
1991b]). The extent to which animal is important to consider other of the National Toxicology Program and
studies may yield false positive information as well; absence of the International Agency for Research
indications for humans is a matter of mutagenic activity or absence of on Cancer and is more protective of
scientific debate. To demonstrate that a carcinogenic activity among structural public health than not including benign
response in animals is not relevant to analogues can increase the confidence tumors in the assessment; benign and
any human situation, adequate data to that negative results in animal studies malignant tumors are treated as
assess the relevancy issue are important. indicate a lack of human hazard. representative of related responses to
In general, while effects seen at the Another limitation is that standard the test agent (McConnell et al., 1986),
highest dose tested are assumed to be animal study protocols are not yet which is scientifically appropriate.
appropriate for assessment, it is available for effectively studying Nonetheless, in assessing findings from
necessary that the experimental perinatal effects. The potential for animal studies, a greater proportion of
conditions be scrutinized. Animal effects on the very young generally malignancy is weighed more heavily
studies are conducted at high doses in should be considered separately. Under than is a response with a greater
order to provide statistical power, the existing Agency policy (U.S. EPA, proportion of benign tumors. Greater
highest dose being one that is minimally 1997a, b), perinatal studies frequency of malignancy of a particular
toxic (maximum tolerated dose or accomplished by modification of tumor type in comparison with other
MTD). Consequently, the question often existing adult bioassay protocols are tumor responses observed in an animal
arises of whether a carcinogenic effect at important in special circumstances. study is also a factor to be considered
in selecting the response to be used in by a scaling factor based on body weight for internal tumors an internal dose is
dose-response assessment. to the 3⁄4 power. The same factor is used significant no matter what the route of
Benign tumors that are not observed for children because it is slightly more exposure. Additionally, the metabolism
to progress to malignancy are assessed protective than using children’s body of the agent will be qualitatively the
on a case-by-case basis. There is a range weight (see Section 3.1.3). This same for an internal dose. The issue of
of possibilities for the overall adjustment factor is used because it quantitative extrapolation of the dose-
significance of benign tumors. They may represents scaling of metabolic rate response relationship from one route to
deserve attention because they are across animals of different size. Because another is addressed case by case.
serious health problems even though the factor adjusts for a parameter that Quantitative extrapolation is
they are not malignant; for instance, can be improved on and brought into complicated by considerations such as
benign tumors may be a health risk more sophisticated toxicokinetic first-pass metabolism.
because of their effect on the function of modeling when such data become
a target tissue, such as the brain. They What Is the Correlation of the Observed
available, they are usually preferable to
may be significant indicators of the need Dose-Response Relationship to the
the default option.
for further testing of an agent if they are For inhalation exposure, a human Relationship at Lower Doses?
observed in a short-term test protocol, or equivalent dose for adults is estimated If sufficient data are available, a
such an observation may add to the by default methodologies that provide biologically based model for both the
overall weight of evidence if the same estimates of lung deposition and observed range and extrapolation below
agent causes malignancies in a long- internal dose (U.S. EPA, 1994). The that range may be used. Although no
term study. Knowledge of the mode of methodologies can be refined to more standard biologically based models are
action associated with a benign tumor sophisticated forms with data on in existence, an agent-specific model
response may aid in the interpretation toxicokinetic and metabolic parameters may be developed if extensive data exist
of other tumor responses associated of the specific agent. This default in a particular case and the purpose of
with the same agent. option, like the one for oral exposure, is the assessment justifies the investment
selected in part because it lays a of the resources needed. The default
How Do Metabolic Pathways Relate procedure for the observed range of data
foundation for incorporating better data.
Across Species and Among Different when a biologically based model is not
The use of information to improve dose
Age Groups and Between Sexes in used is to use a curve-fitting model for
estimation from applied to internal to
Humans? incidence data.
delivered dose is encouraged, including
The default option is that there is a use of toxicokinetic modeling instead of In the absence of data supporting a
similarity of the basic pathways of any default, where data are available. biologically based model for
metabolism and the occurrence of There are important differences extrapolation outside of the observed
metabolites in tissues in regard to the between infants, adults, and older range, the choice of approach is based
species-to-species extrapolation of adults in the processes of absorption, on the view of mode of action of the
cancer hazard and risk. If comparative distribution, and elimination; for agent arrived at in the hazard
metabolism studies were to show no example, infants tend to absorb metals assessment. If more than one approach
similarity between the tested species through the gut more rapidly and more (e.g., both a nonlinear and linear
and humans and a metabolite(s) was the efficiently than do older children or approach) are supported by the data,
active form, there would be less support adults (Calabrese, 1986). Renal they should be used and presented to
for an inference that the animal elimination is also not as efficient in the decisionmaker.
response(s) relates to humans. In other infants. Although these processes reach A linear extrapolation approach is
cases, parameters of metabolism may adult competency at about the time of used when the mode of action
vary quantitatively between species; this weaning, they may have important information is supportive of linearity or
becomes a factor in deciding on an implications, particularly when the mode of action is not understood. The
appropriate human-equivalent dose dose-response relationship for an agent linear approach is used when a view of
based on animal studies, optimally in is considered to be nonlinear and there the mode of action indicates a linear
the context of a toxicokinetic model. is an exposure scenario response, for example, when a
Although the basic pathways are disproportionately affecting infants, conclusion is made that an agent
assumed to be the same among humans, because in these cases the magnitude of directly causes alterations in DNA, a
the presence of polymorphisms in the dose is more pertinent than the usual kind of interaction that not only
general population and factors such as approach in linear extrapolation of theoretically requires one reaction but
the maturation of the pathways in averaging dose across a lifetime. also is likely to be additive to ongoing,
infants should be considered. The active Efficiency of intestinal absorption in spontaneous gene mutation. Other kinds
form of an agent may be present to older adults tends to be generally less of activity may have linear implications,
differing degrees, or it may be overall for most chemicals. Another for example, linear rate-limiting steps
completely absent, which may result in notable difference is that, post-weaning would also support a linear procedure.
greater or lesser risk for subpopulations. (about 1 year), children have a higher The linear approach is to draw a straight
metabolic rate than do adults (Renwick, line between a point of departure from
How Do Toxicokinetic Processes Relate 1998), and they may toxify or detoxify observed data, generally as a default, an
Across Species and Among Different agents at a correspondingly higher rate. LED chosen to be representative of the
Age Groups and Between Sexes in For a route-to-route exposure lower end of the observed range, and the
Humans? extrapolation, the default option is that origin (zero incremental dose, zero
A major issue is how to estimate an agent that causes internal tumors by incremental response). This approach is
human-equivalent doses in one route of exposure will be generally considered to be public-health
extrapolating from animal studies. As a carcinogenic by another route if it is protective.
default for oral exposure, a human absorbed by the second route to give an The linear default is thought to
equivalent dose for adults is estimated internal dose. This is a qualitative generally provide an upper-bound
from data on another species by an option and is considered to be public- calculation of potential risk at low
adjustment of animal applied oral dose health protective. The rationale is that doses, for example, a
1/100,000 to 1/1,000,000 risk. This dose or exposure. This assumes that a assessment. Environ Health Perspect
upper bound is thought to be public- high dose of such an agent received over 100:9–20.
health protective at low doses for the a shorter period of time is equivalent to Barrett, JC; Lee, TC. (1992) Mechanisms of
arsenic-induced gene amplification. In:
range of human variation, considering a low dose spread over a lifetime. This
Kellems, RE, ed. Gene amplification in
the typical Agency target range for risk is thought to be a relatively public- mammalian cells: a comprehensive
management of 1/1,000,000 to 1/10,000, health-protective option and has some guide. New York: Marcel Dekker.
although it may not completely be so empirical support (Monro, 1992). A Baylin, S; Bestor, TH. (2002) Altered
(Bois et al., 1995) if pre-existing disease counter example, i.e., effects of short- methylation patterns in cancer cell
or genetic constitution place a term, high exposure levels that result in genomes: causes or consequence? Cancer
percentage of the population at greater subsequent cancer development, is Cell 1:299–305.
risk from exposure to carcinogens. The treatment of cancer patients with certain Bellamy, CO; Malcomson, RD; Harrison, DJ;
question of what may be the actual et al. (1995) Cell death in health and
chemotherapeutic agents. When
disease: the biology and regulation of
variation in human susceptibility is one sufficient information is available to apoptosis. Seminars in Cancer Biology,
that was discussed in general in the support a different approach, it can be Apoptosis in Oncogenesis and
NRC (1994) report, as well as the NRC used. For example, short-term exposure Chemotherapy 6:3–16.
report on pesticides in children and estimates (several days to several Biggs, PJ; Warren, W; Venitt, S; et al. (1993)
infants (NRC, 1993b). NRC has months) may be more appropriate than Does a genotoxic carcinogen contribute
recommended research on the question, the lifetime average daily dose. In these to human breast cancer? The value of
and EPA and other agencies are cases, both agent concentration and mutational spectra in unraveling the
conducting such research. Given the duration are likely to be important, etiology of cancer. Mutagenesis 8:275–
current state of knowledge, EPA will 283.
because such effects may be reversible Birnbaum, LS; Fenton, SE. (2003) Cancer and
assume that the linear default procedure at cessation of very short-term developmental exposure to endocrine
adequately accounts for human exposures. disruptors. Environ Health Perspect
variation unless there is case-specific 111:389–394.
information for a given agent or mode of Appendix B: EPA’s Guidance for Data
Birner, G; Albrecht, W; Neumann, HG. (1990)
action that indicates a particularly Quality Assessment Biomonitoring of aromatic amines. III:
susceptible subpopulation or lifestage, U.S. EPA (U.S. Environmental Protection hemoglobin binding and benzidine and
in which case the special information Agency). (2000d) Guidance for data quality some benzidine congeners. Arch Toxicol
will be used. assessment: practical methods for data 64(2):97–102.
When adequate data on mode of analysis. Office of Environmental Blair, A; Burg, J; Foran, J; et al. (1995)
action provide sufficient evidence to Information, Washington, DC. EPA/600/R– Guidelines for application of meta-
support a nonlinear mode of action for 96/084. Available from: http://www.epa.gov/ analysis in environmental epidemiology.
quality/qs-docs/g9-final.pdf. Regul Toxicol Pharmacol 22:189–197.
the general population and/or any
Bois, FY; Krowech, G; Zeise, L. (1995)
subpopulations of concern, a different References Modeling human interindividual
approach—a reference dose/reference Allen, BC; Crump, KS; Shipp, AM. (1988) variability in metabolism and risk: the
concentration that assumes that Correlation between carcinogenic example of 4-aminobiphenyl. Risk Anal
nonlinearity—is used. The POD is again potency of chemicals in animals and 15:205–213.
generally an BMDL when incidence data humans. Risk Anal 8:531–544. Calabrese, EJ. (1986) Age and susceptibility
are modeled. A sufficient basis to Ames, BN; Gold, LS. (1990) Too many rodent to toxic substances. New York: Winter-
support this nonlinear procedure is carcinogens: mitogenesis increases Interscience Publication, John Wiley and
likely to include data on responses that mutagenesis. Science 249:970–971. Sons, Inc.
are key events integral to the Anderson, LM; Diwan, BA; Fear, NT; et al. Callemen, CJ; Ehrenberg, L; Jansson, B; et al.
(2000) Critical windows of exposure for (1978) Monitoring and risk assessment
carcinogenic process. This means that
children’s health: cancer in human by means of alkyl groups in hemoglobin
the POD may be based on these epidemiological studies and neoplasms in persons occupationally exposed to
precursor response data, for example, in experimental animal models. Environ ethylene oxide. J Environ Pathol Toxicol
hormone levels or mitogenic effects Health Perspect 108(Suppl 3):573–594. 2:427–442.
rather than tumor incidence data. Ashby, J; Tennant, RW. (1991) Definitive Caporaso, N; Hayes, RB; Dosemeci, M; et al.
When the mode of action information relationships among chemical structure, (1989) Lung cancer risk, occupational
indicates that the dose-response carcinogenicity and mutagenicity for 301 exposure, and the debrisoquine
function may be adequately described chemicals tested by the U.S. NTP. Mutat metabolic phenotype. Cancer Res
by both a linear and a nonlinear Res 257:229–306. 49:3675–3679.
approach, then the results of both the Ashby, J; Tennant, RW. (1994) Prediction of Cavenee, WK; Koufos, A; Hansen, MF. (1986)
linear and the nonlinear analyses are rodent carcinogenicity for 44 chemicals: Recessive mutant genes predisposing to
results. Mutagenesis 9:7–15. human cancer. Mutat Res 168:3–14.
presented. An assessment may use both
Ashby, J; Doerrer, NG; Flamm, FG; et al. CDC (Centers for Disease Control and
linear and nonlinear approaches if (1990) A scheme for classifying Prevention). (2004) The health
different responses are thought to result carcinogens. Regul Toxicol Pharmacol consequences of smoking: a report of the
from different modes of action or a 12:270–295. surgeon general. Dept. of Health and
response appears to be very different at Ashby, J; Brady, A; Elcombe, CR; et al. (1994) Human Services, Washington, DC.
high and low doses due to influence of Mechanistically based human hazard Available from: http://www.cdc.gov/
separate modes of action. The results assessment of peroxisome proliferator- tobacco/sgr/sgr_2004/index.htm.
may be needed for assessment of induced hepatocarcinogenesis. Hum Chang, CC; Jone, C; Trosko, JE; et al. (1988)
combined risk from agents that have Exper Toxicol 13:1–117. Effect of cholesterol epoxides on the
common modes of action. Barrett, JC. (1992) Mechanisms of action of inhibition of intercellular
Absent data to the contrary, the known human carcinogens. In: communication and on mutation
Mechanisms of carcinogenesis in risk induction in Chinese hamster V79 cells.
default assumption is that the identification. IARC Sci Pubs No. 116, Mutat Res 206:471–478.
cumulative dose received over a 115–134. International Agency for Chuang, LS; Ng, HH; Chia, JN; Li, BF. (1996)
lifetime, expressed as a lifetime average Research on Cancer, Lyon, France. Characterisation of independent DNA
daily dose or lifetime average daily Barrett, JC. (1993) Mechanisms of multistep and multiple Zn-binding domains at the
exposure, is an appropriate measure of carcinogenesis and carcinogen risk N terminus of human DNA-(cytosine-5)
methyltransferase: modulating the ) Use of probabilistic expert judgment in assessment. Toxicol Pathol 28:482–499.
property of a DNA-binding domain by uncertainty analysis of carcinogenic Hammand, EC. (1966) Smoking in relation to
contiguous Zn-binding motifs. J Mol Biol potency. Regul Toxicol Pharmacol. 2:15– the death rates of one million men and
257(5):935–48. 36. women. In: Haenxzel, W, ed.
Chen, C; Farland, W. (1991) Incorporating Executive Order 13045 (1997) Protection of Epidemiological approaches to the study
cell proliferation in quantitative cancer children from environmental health risks of cancer and other chronic diseases.
risk assessment: approaches, issues, and and safety risks, issued April 21, 1997. National Cancer Institute Monograph No.
uncertainties. In: Butterworth, B., Slaga, Fearon, E; Vogelstein, B. (1990) A genetic 19. Washington, DC.
T., Farland, W., et al., eds. Chemical model for colorectal tumorigenesis. Cell Hanahan, D; Weinberg, RA. (2000) The
induced cell proliferation: implications 61:959–967. hallmarks of cancer. Cell 100:57–70.
for risk assessment. New York: Wiley- Fenton, SE; Davis, CC. (2002) Atrazine Harris, CC; Hollstein, M. (1993) Clinical
Liss, pp. 481–499. exposure in utero increases implications of the p53 tumor suppressor
Chhabra, RE; Huff, JE; Schwetz, BS; Selkirk, dimethylbenz a anthracene-induced gene. N Engl J Med 329:1318–1327.
J. (1990) An overview of prechronic and mammary tumor incidence in long evans Haseman, JK. (1983) Issues: a reexamination
chronic toxicity/carcinogenicity offspring. Toxicol Sci 66(1–2):185. ‘‘The of false-positive rates for carcinogenesis
experimental study designs and criteria Toxicologist, Abstracts of the 41st studies. Fundam Appl Toxicol 3:334–
used by the National Toxicology Annual Meeting of the Society of 339.
Program. Environ. Health Perspect. Toxicology.’’ (Abstract 903) Haseman, JK. (1984) Statistical issues in the
86:313–321. Fisher, RA. (1950) Statistical methods for design, analysis and interpretation of
Choy, WN. (1993) A review of the dose- research workers. Edinburgh, Scotland: animal carcinogenicity studies. Environ
response induction of DNA adducts by Oliver and Boyd. Health Perspect 58:385–392.
aflatoxin B2 and its implications to Florig, HK; Morgan, MG; Morgan, KM; Jenni, Haseman, JK. (1985) Issues in carcinogenicity
quantitative cancer-risk assessment. KE; Fischhoff, B; Fischbeck, PS; DeKay, testing: dose selection. Fundam Appl
Mutat Res 296:181–198. ML. (2001) A deliberative method for Toxicol 5:66–78.
Clayson, DB; Mehta, R; Iverson, F. (1994) ranking risks (I): Overview and test bed Haseman, JK. (1990) Use of statistical
Oxidative DNA damage—the effects of development. Risk Anal. 21:913–21. decision rules for evaluating laboratory
certain genotoxic and operationally non- Flynn, GL. (1990) Physicochemical animal carcinogenicity studies. Fundam
genotoxic carcinogens. Mutat Res determinants of skin absorption. In: Appl Toxicol 14:637–648.
317:25–42. Gerrity, TR, Henry, CJ, eds. Principles of Haseman, JK. (1995) Data analysis: Statistical
Cohen, SM. (1995) Role of urinary physiology route to route extrapolation for risk analysis and use of historical control
and chemistry in bladder carcinogenesis. assessment. New York: Elsevier Science; data. Regul Toxicol Pharmacol 21:52–59.
Fd Chem Toxicol 33:715–30. pp. 93–127. Hatch, EE; Palmer, JR; Titus-Ernstoff, L;
Cohen, SW; Ellwein, LB. (1990) Cell Fos, PJ; McLin, CL. (1990) The risk of falling Noller, KL et al. (1998) Cancer risk in
proliferation in carcinogenesis. Science in the elderly: a subjective approach. women exposed to diethylstilbestrol in
249:1007–1011. Med Decis Making 10:195–200. utero. JAMA 280:630–634.
Cohen, SM; Ellwein, LB. (1991) Genetic Garfinkel, L; Silverberg, E. (1991) Lung Hattis, D. (1990) Pharmacokinetic principles
errors, cell proliferation and cancer and smoking trends in the United for dose-rate extrapolation of
carcinogenesis. Cancer Res 51:6493– States over the past 25 years. Cancer carcinogenic risk from genetically active
6505. 41:137–145. agents. Risk Anal 10:303–316.
Cohen, SM; Purtilo, DT; Ellwein, LB. (1991) Gaylor, DW; Zheng, Q. (1996) Risk Hawkins, NC; Evans, JS. (1989) Subjective
Pivotal role of increased cell assessment of nongenotoxic carcinogens estimation of toluene exposures: a
proliferation in human carcinogenesis. based on cell proliferation/death rates in calibration study of industrial hygienists,
Mod Pathol 4:371–375. rodents. Risk Anal 16(2):221–225. Applied Industrial Hygiene, 4:61–68.
Conolly, RB; Andersen, ME. (1991) Gaylor, DW; Kodell, RL; Chen, JJ; et al. (1994) Hawkins, NC; Graham, JD. (1988) Expert
Biologically based pharmacodynamic Point estimates of cancer risk at low scientific judgment and cancer risk
models: tools for toxicological research doses. Risk Anal 14(5):843–850. assessment: a pilot study of
and risk assessment. Ann Rev Pharmacol Gibson, DP; Aardema, MJ; Kerckaert, GA; et pharmacokinetic data, Risk Anal. 8:615–
Toxicol 31:503–523. al. (1995) Detection of aneuploidy- 25.
Contrera, JF; Matthews, EJ; Benz, RD. (2003) inducing carcinogens in the Syrian Hayward, JJ; Shane, BS; Tindall, KR; et al.
Predicting the carcinogenic potential of hamster embryo (SHE) cell (1995) Differential in vivo mutagenicity
pharmaceuticals in rodents using transformation assay. Mutat Res 343:7– of the carcinogen-noncarcinogen pair
molecular structural similarity and E- 24. 2,4- and 2,6-diaminotoluene.
state indices. Regul. Toxicol. Pharmacol. Ginsberg, GL. (2003) Assessing cancer risks Carcinogenesis 10:2429–2433.
38:243–259. from short-term exposures in children. Heddle, JA; Swiger, RR. (1996) Risk
Cresteil, T. (1998) Onset of xenobiotic Risk Anal 23(1):19–34. estimation from somatic mutation assays.
metabolism in children: toxicological Goddard, MJ; Murdoch, DJ; Krewski, D. Mutat Res 365(1–3):107–17.
implications. Food Addit Contam 15, (1995). Temporal aspects of risk Herbst, AL, Ulfelder, H, Poskanzer, DC.
Supplement 45–51. characterization. Inhal Toxicol 7:1005– (1971) Adenocarcinoma of the vagina:
Dearfield, K. L.; Auletta, A. E.; Cimino, M. C., 1018. association of maternal stilbestrol
et al. (1991) Considerations in the U.S. Goldsworthy, TL; Hanigan, MH; Pitot, HC. therapy with tumor appearance in young
Environmental Protection Agency’s (1986) Models of hepatocarcinogenesis women. N Engl J Med 284:878–881.
testing approach for mutagenicity. Mutat. in the rat-contrasts and comparisons. Hill, AB. (1965) The environment and
Res. 258:259–283. CRC Crit Rev Toxicol 17:61–89. disease: association or causation? Proc R
D’Souza, RW; Francis, WR; Bruce, RD; et al. Goodman, JI; Counts, JL. (1993) Soc Med 58:295–300.
(1987) Physiologically based Hypomethylation of DNA: A possible Hoel, DG; Kaplan, NL; Anderson, MW. (1983)
pharmacokinetic model for ethylene nongenotoxic mechanism underlying the Implication of nonlinear kinetics on risk
chloride and its application in risk role of cell proliferation in estimation in carcinogenesis. Science
assessment. In: Pharmacokinetics in risk carcinogenesis. Environ Health Perspect 219:1032–1037.
assessment: drinking water and health. 101 Supp. 5:169–172. Holliday, R. (1987) DNA methylation and
Vol. 8. Washington, DC: National Greenland, S. (1987) Quantitative methods in epigenetic defects in carcinogenesis.
Academy Press. the review of epidemiologic literature. Mutat Res 181:215–217.
Enterline, PE; Henderson, VL; Marsh, GM. Epidemiol Rev 9:1–29. Holladay SD, Smialowicz RJ. 2000.
(1987) Exposure to arsenic. Amer J Gulezian, D; Jacobson-Kram, D; McCullough, Development of the murine and human
Epidemiol 125:929–938. CB; et al. (2000) Use of transgenic immune system: differential effects of
Evans, JS; Gray, GM; Sielken, RL Jr; Smith, animals for carcinogenicity testing: immunotoxicants depend on time of
AE; Valdez-Flores, C; Graham, JD. (1994 considerations and implications for risk exposure. Environ Health Perspect 108
Suppl 3:463–473. cancer. Cancer Res 46:461–466. of age at first exposure. In: Guzelian, P;
Holsapple MP, West LJ, Landreth KS. 2003. Kehrer, JP. (1993) Free radicals as mediators Henry, CJ; Olin, SS, eds. Similarities and
Species comparison of anatomical and of tissue injury and disease. Crit Rev difference between children and adults:
functional immune system development. Toxicol 23:21–48. implications for risk assessment.
Birth Defects Res B Dev Reprod Toxicol Kelsey, JL; Whittemore, AS; Evans, AS; Washington, DC: ILSI Press; pp. 66–78.
68(4):321–334. Thompson, WD. (1996) Methods in McConnell, EE; Solleveld, HA; Swenberg, JA;
Huff, JE. (1993) Chemicals and cancer in observational epidemiology. New York: et al. (1986) Guidelines for combining
humans: first evidence in experimental Oxford University Press. neoplasms for evaluation of rodent
animals. Environ Health Perspect Kimbell, JS; Subramaniam, RP; Gross, EA; carcinogenesis studies. J Natl Cancer Inst
100:201–210. Schlosser, PM; Morgan, KT. (2001) 76:283–289.
Huff, JE. (1994) Chemicals causally Dosimetry modeling of inhaled Meek, ME; Bucher, JR; Chohen, SM; Dellarco,
associated with cancers in humans and formaldehyde: comparisons of local flux V; Hill, RN; Lehman-McKeeman, LD;
laboratory animals. A perfect predictions in the rat, monkey and Longfellow, DG; Pastoor, T; Seed, J.; and
concordance. In: Carcinogenesis. human nasal passages. Toxicol Sci Patton, DE. (2003) A framework for
Waalkes, MP, Ward, JM, eds., New York: 64(1):100–110. humand relevance analysis of
Raven Press; pp. 25–37. Kinzler, KW; Vogelstein, B. (2002) Colorectal information on carcinogenic modes of
Huff J, Cirvello J, Haseman J, Bucher J (1991) tumors. In: Vogelstein, B; Kinzler, KW, action. Crit Rev Toxicol 33:591–653.
Chemicals associated with site-specific eds. The genetic basis of human cancer. Melnick, RL, Huff, JE, Barrett, JC, Maronpot,
neoplasia in 1394 long-term New York: McGraw-Hill. RR, Lucier, G, Portier, CJ. (1993) Cell
carcinogenesis experiments in laboratory Kinzler, KW; Nilbert, MC; Su, L–K; et al. proliferation and chemical
rodents. Environ Health Perspect (1991) Identification of FAP locus genes carcinogenesis: A symposium overview.
93:247–70. Erratum in: Environ Health from chromosome 5q21. Science Mol Carcinog 7:135–138.
Perspect 1991 Nov; 95:213. 253:661–665. Miller, RW. (1995) Special susceptibility of
Hulka, BS; Margolin, BH. (1992) Kraus, AL; Munro, IC; Orr, JC; et al. (1995) the child to certain radiation-induced
Methodological issues in epidemiologic Benzoyl peroxide: an integrated human cancers. Environ Health Perspect
studies using biological markers. Am J safety assessment for carcinogenicity. 103(suppl 6):41–44.
Epidemiol 135:122–129. Regul Toxicol Pharmacol 21:87–107. Miller, MD; Marty, MA; Arcus, A; et al.
IARC (International Agency for Research on Krewski, D; Van Ryzin, J. (1981) Dose (2002) Differences between children and
Cancer). (1994) IARC monographs on the response models for quantal response adults: implications for risk assessment
evaluation of carcinogenic risks to toxicity data. In: Csorgo; Dawson; Rao; et at California EPA. Int J Toxicol 21:403–
humans. Vol. 60. Some industrial al., eds. Statistics and related topics. 418.
chemicals. Lyon, France: IARC; pp. 13– Amsterdam: North-Holland, pp. 201– Monro, A. (1992) What is an appropriate
33. 231. measure of exposure when testing drugs
IARC. (International Agency for Research on Krewski, D; Murdoch, DJ; Withey, JR. (1987) for carcinogenicity in rodents? Toxicol
Cancer) (1999) The use of short- and The application of pharmacokinetic data Appl Pharmacol 112:171–181.
medium-term tests for carcinogens and in carcinogenic risk assessment. In: Moolgavkar, SH. (1986) Carcinogenesis
data on genetic effects in carcinogenic Pharmacokinetics in risk assessment: Modelin: From Molecular Biology to
hazard evaluation. Lyon, France. drinking water and health. Vol. 8. Epidemiology. Am Rev Public Health
IEc (Industrial Economics, Incorporated). Washington, DC: National Academy 7:151–169.
2004. ‘‘An Expert Judgment Study of the Press; pp. 441–468. Moolgavkar, SH; Knudson, AG. (1981)
Concentration-Response Relationship La, DK; Swenberg, JA. (1996) DNA adducts: Mutation and cancer: a model for human
Between PM2.5 Exposure and biological markers of exposure and carcinogenesis. J Natl Cancer Inst
Mortality,’’ Available at: http:// potential applications to risk assessment. 66:1037–1052.
www.epa.gov/ttn/ecas/benefits.html. Mutat Res 365(1–3):129–46. Morgan, KM; DeKay, ML; Fischbeck, PS;
ILSI (International Life Sciences Institute). Levine, AJ; Perry, ME; Chang, A; et al. (1994) Morgan, MG; Fischhoff, B; Florig, HK.
(1992) Similarities and differences The 1993 Walter Hubert lecture: the role (2001) A deliberative method for ranking
between children and adults; of the p53 tumor-suppressor gene in risks (II): Evaluation of validity and
implications for risk assessment. tumorigenesis. Br J Cancer 69:409–416. agreement among risk managers. Risk
Washington, DC: ILSI Press. Lijinsky, W. (1993) Species differences in Anal. 21:923–37.
ILSI (International Life Sciences Institute). carcinogenesis. In Vivo 7:65–72. Morrison, V; Ashby, J. (1994) A preliminary
(1997) Principles for the selection of Lilienfeld, AM; Lilienfeld, D. (1979) evaluation of the performance of the
doses in chronic rodent bioassays. Foran, Foundations of epidemiology, 2nd ed. mutaTM mouse (lacZ) and Big BlueTM
JA, ed. Washington, DC: ILSI Press. New York: Oxford University Press. (lacI) transgenic mouse mutation assays.
ILSI (International Life Sciences Institute). Littlefield, NA; Farmer, JH; Gaylor, DW. Mutagenesis 9:367–375.
(2001) Proceedings of workshop on the (1980) ED01 study. J Environ Pathol Murdoch, DJ; Krewski, D; Wargo, J. (1992)
evaluation of alternative methods for Toxicol 3:17. Cancer risk assessment with intermittent
carcinogenesis testing. Toxicol Pathol Maltoni, C; Lefemine, G; Ciliberti, A; et al. exposure. Risk Anal 12(4):569–577.
29:1–351. (1981) Carcinogenicity bioassay of vinyl Murrell, JA; Portier, CJ; Morris, RW. (1998)
IPCS (International Programme on Chemical chloride monomer: a model of risk Characterizing dose-response I: critical
Safety). (1999) IPCS workshop on assessment on an experimental basis. assessment of the benchmark dose
developing a conceptual framework for Environ Health Perspect 41:3–29. concept. Risk Anal 18(1):13–25.
cancer risk assessment, February 16–18, Maronpot, RR; Shimkin, MB; Witschi, HP; et NRC (National Research Council). (1983) Risk
1999, Lyon, France. IPCS/99.6. World al. (1986) Strain A mouse pulmonary assessment in the federal government:
Health Organization, Geneva. tumor test results for chemicals managing the process. Committee on the
Ito, N; Shirai, T; Hasegawa, R. (1992) previously tested in National Cancer Institutional Means for Assessment of
Medium-term bioassays for carcinogens. Institute carcinogenicity test. J Natl Risks to Public Health, Commission on
In: Vainio, H, Magee, PN, McGregor, DB, Cancer Inst 76:1101–1112. Life Sciences, NRC. Washington, DC:
et al., eds. Mechanisms of carcinogenesis Marsman, DS; Popp, JA. (1994) Biological National Academy Press.
in risk identifications. International potential of basophilic hepatocellular NRC (National Research Council). (1990)
Agency for Research on Cancer, Lyon, foci and hepatic adenoma induced by the Health effects of exposure to low levels
France; pp. 353–388. peroxisome proliferator, Wy-14,643. of ionizing radiation (BEIR V).
Jelovsek, FR; Mattison, DR; Young, JF. (1990) Carcinogenesis 15:111–117. Washington, DC: National Academy
Eliciting principles of hazard Mausner, JS; Kramer, S. (1985) Epidemiology, Press.
identification from experts. Teratology 2nd ed. Philadelphia: W.B. Saunders. NRC (National Research Council). (1993a)
42:521–533. McConnell, EE. (1992) Comparative response Issues in risk assessment. Committee on
Jones, PA. (1986) DNA methylation and in carcinogenesis bioassay as a function Risk Assessment Methodology.
Washington, DC: National Academy 571–577. Shelby, MD; Zeiger, E. (1990) Activity of
Press. Peto, J; Darby, S. (1994) Radon risk human carcinogens in the Salmonella
NRC (National Research Council). (1993b) reassessed. Nature 368:97–98. and rodent bone-marrow cytogenetics
Pesticides in the diets of infants and Peto, R; Gray, R; Brantom, P; et al. (1984) tests. Mutat Res 234:257–261.
children. Washington, DC: National Nitrosamine carcinogenesis in 5120 Silberstein, GB. (2001) Tumour-stromal
Academy Press. rodents: chronic administration of interactions: role of the stroma in
NRC (National Research Council). (1994) sixteen different concentrations of mammary development. Breast Cancer
Science and judgment in risk NDEA, NDMA, NPYR and NPIP in the Res 3:218–223.
assessment. Washington, DC: National water of 4440 inbred rats, with parallel Slikker W, 3rd, Mei N, Chen T. 2004. N-
Academy Press. studies on NDEA alone of the effect of ethyl-N-nitrosourea (ENU) increased
NRC (National Research Council). (1996) age of starting (3,6, or 20 weeks) and of brain mutations in prenatal and neonatal
Understanding risk: informing decisions species (rats, mice or hamsters). IARC mice but not in the adults. Toxicol Sci
in a democratic society. Washington, DC: Sci Publ 57:627–665. 81(1):112–120. Sisk, SC; Pluta, LJ; Bond,
National Academy Press. Pinkerton, KE; Joad, J. (2000) The JA; et al. (1994) Molecular analysis of
NRC (National Research Council). (2002) mammalian respiratory system and lacI mutants from bone marrow of
Estimating the public health benefits of critical windows of exposure for B6C3F1 transgenic mice following
proposed air pollution regulations. children’s health. Environ Health inhalation exposure to 1,3-butadiene.
Washington, DC: National Academy Perspect 108(suppl):457–462. Carcinogenesis 15(3):471–477.
Press. Portier, C. (1987) Statistical properties of a Snedecor, GW; Cochran, WG. (1967)
NTP (National Toxicology Program). (1984) two-stage model of carcinogenesis. Statistical methods, 6th ed. Ames, Iowa:
Report of the ad hoc panel on chemical Environ Health Perspect 76:125–131. Iowa State University Press.
carcinogenesis testing and evaluation of Putzrath, RM; Ginevan, ME (1991) Meta- Sonich-Mullin, C; Fielder, R; Wiltse, J;
the National Toxicology Program, Board analysis: Methods for combining data to Baetcke, K; Dempsey. K; Fenner-Crisp, P;
of Scientific Counselors. Washington, improve quantitative risk assessment. Grant, D; Hartley, M; Knaap, A; Kroese,
DC: U.S. Government Printing Office. Regul Toxicol Pharmacol 14:178–188. D; Mangelsdorf, I; Meek, E; Rice, JM; and
1984–421–132:4726. Rall, DP. (1991) Carcinogens and human Yones, M. (2001) IPCS conceptual
Nichols, AL; Zeckhauser, RJ. (1986). The health: part 2. Science 251:10–11. framework for evaluating a mode of
dangers of caution: Conservatism in the Renn, O. (1999) Model for an analytic- action for chemical carcinogenesis. Regul
assessment and the mismanagement of deliberative process in risk management. Toxicol Phamacol 34:146–152.
risk. In: Smith, VK, ed., Advances in Environ Sci Technol 33:3049–3055. Spalding, JW; French, JE; Stasiewicz, S;
Applied Micro-Economics: Risk, Renwick, AG. (1998) Toxicokinetics in Furedi-Machacek, M; Conner, F; Tice,
Uncertainty, and the Valuation of infants and children in relation to the RR; Tennant, RW. (2000) Responses of
Benefits and Costs, Vol. 4, Greenwich, ADI and TDI. Food Addit Contam 15, transgenic mouse lines p53(+/¥) and
Conn.: JAI Press, pp. 55–82. Suppl 17–35. Tg.AC to agents tested in conventional
North, DW ; Merkhofer, MW. (1976). A Rice, JM. (1979) Problems and perspective in carcinogenicity bioassays. Toxicol Sci
methodology for analyzing emission perinatal carcinogenesis: a summary of 53(2)213–223.
control strategies. Comput Oper Res the conference. NCI Monogr 51:271–278. Stiber, NA; Pantazidou, M; Small, MJ. (1999)
3:187–207. Richard, AM. (1998a) Structure-based Expert system methodology for
OECD (Organization for Economic methods for predicting mutagenicity and evaluating reductive dechlorination at
Cooperation and Development). (1981) carcinogenicity: are we there yet? Mutat TCE sites. Environ Sci Technol 33:3012–
Guidelines for testing of chemicals. Res 400:493–507. 3020.
Carcinogenicity studies. No. 451. Paris, Richard, AM, (1998b) Commercial toxicology Stiteler, WH; Knauf, LA; Hertzberg, RC; et al.
France. prediction systems: A regulatory (1993) A statistical test of compatibility
OMB (Office of Management and Budget). perspective. Toxicol. Lett. 102–103:611– of data sets to a common dose-response
(2002) Guidelines for ensuring and 616. model. Regul Toxicol Pharmacol 18:392–
maximizing the quality, objectivity, Richard, AM; Williams, CR. (2002) 402.
utility, and integrity of information Distributed structure-searchable toxicity Subramaniam, RP; Asgharian, B; Freijer, JI;
disseminated by federal agencies. (DSSTox) public database network: a Miller, FJ; Anjilvel, S. (2003) Analysis of
Federal Register 67(36):8451–8460. proposal: Mutat. Res. 499:27–52. differences in particle deposition in the
Available from: http://www.epa.gov/oei/ Richmond, HM. (1981). A framework for human lung. Inhal Toxicol 15:1–21.
qualityguidelines/fr22fe02–117.htm. assessment of health risks associated Swierenga, SHH; Yamasaki, H. (1992)
OMB (Office of Management and Budget). with national ambient air quality Performance of tests for cell
(2003) Circular A–4: Regulatory standards. Environ Prof 3:225–234. transformation and gap junction
Analysis. September 17. Available from: Rothman, KJ; Greenland, S. (1998) Modern intercellular communication for
http://www.whitehouse.gov/omb/ Epidemiology. Philadelphia: Lippincott detecting nongenotoxic carcinogenic
circulars/a004/a-4.pdf. Williams and Wilkins Publishers. activity. In: Mechanisms of
OMB (Office of Management and Budget). Rouse, J; Jackson, SP. (2002) Interfaces carcinogenesis in risk identification.
(2004) Revised information quality between the detection, signaling, and IARC Sci. Pubs. No. 116, Lyon, France;
bulletin for peer review. April 15. repair of DNA damage. Science 297:547– pp. 165–193.
Available from: http:// 551. Szklo, M; Nieto, FJ. (2000): Epidemiology
www.whitehouse.gov/omb/inforeg/ Samet, JM; Schnatter, R; Gibb, H. (1998) Beyond the Basics. Gaithersburg, MD:
peer_review041404.pdf. Invited Commentary: Epidemiology and Aspen Publishers, Inc.
OSTP (Office of Science and Technology risk assessment. Am J Epidemiol Tarone, RE. (1982) The use of historical
Policy). (1985) Chemical carcinogens: 148:929–936. control information in testing for a trend
review of the science and its associated Scheuplein, R; Charnley, G; Dourson, M. in proportions. Biometrics 38:215–220.
principles. Federal Register 50:10372– (2002) Differential sensitivity of children Taylor, JH; Watson, MA; Devereux, TR; et al.
10442. and adults to chemical toxicity. I: (1994) p53 mutation hotspot in radon-
Peltom̈aki, P; Aaltonen, LA; Sisonen, P; et al. biological basis. Regul Toxicol associated lung cancer. Lancet 343:86–
(1993) Genetic mapping of a locus Pharmacol 35:429–447. 87.
predisposing human colorectal cancer. SAB (Science Advisory Board). (1997) An Tennant, RW. (1993) Stratification of rodent
Science 260:810–812. SAB report: guidelines for cancer risk carcinogenicity bioassay results to reflect
Peto, J. (1992) Meta-analysis of assessment. Washington DC: U.S. relative human hazard. Mutat Res
epidemiological studies of Environmental Protection Agency, 286:111–118.
carcinogenesis. In: Mechanisms of September. EPA–SAB–EHC–97–010. Tennant, RW; French, JE; Spalding, JW.
carcinogenesis in risk assessment. IARC Available from: http://www.epa.gov/sab/ (1995) Identifying chemical carcinogens
Sci. Pubs. No. 116, Lyon, France; pp. pdf/ehc9710.pdf. and assessing potential risk in short-term
bioassays using transgenic mouse Assessment Office, Research Triangle of Prevention, Pesticides and Toxic
models. Environ Health Perspect Park, NC. EPA/600/8–90/066F. Substances, Washington, DC. EPA/712/
103:942–950. U.S. EPA (U.S. Environmental Protection C–98/212. Available from: http://
Tennant, RW; Stasiewicz, S; Mennear, J; et al. Agency). (1995) Policy for risk www.epa.gov/opptsfrs/
(1999) Genetically altered mouse models characterization. Memorandum of Carol OPPTS_Harmonized/
for identifying carcinogens. In: M. Browner, Administrator, March 21, 870_Health_Effects_Test_Guidelines/
McGregor, DB; Rice, JM; Venitt, S, eds. 1995, Washington, DC. Available from: Series/.
The use of short- and medium-term tests http://www.epa.gov/osp/spc/ U.S. EPA (U.S. Environmental Protection
for carcinogens and data on genetic 2riskchr.htm. Agency). (1998d) EPA’s rule writer’s
effects in carcinogenic hazard U.S. EPA (U.S. Environmental Protection guide to Executive Order 13045.
evaluation. Lyon, France: International Agency). (1996a) Guidelines for Available from: http://yosemite.epa.gov/
Agency for Research on Cancer. reproductive toxicity risk assessment. ochp/ochpweb.nsf/content/
Tinwell, H; Ashby, J. (1991) Activity of the Federal Register 61(212):56274–56322. whatwe_regulate.htm.
human carcinogen MeCCNU in the Available from: http://cfpub.epa.gov/ U.S. EPA (U.S. Environmental Protection
mouse bone marrow mironucleus test. ncea/raf/recordisplay.cfm?deid=2838. Agency). (1999a) Guidelines for
Environ Molec Mutagen 17:152–154. U.S. EPA (U.S. Environmental Protection carcinogen risk assessment (review
Todd, GC. (1986) Induction of reversibility of Agency). (1996b) Comparison of the draft). Risk Assessment Forum,
thyroid proliferative changes in rats effects of chemicals with combined Washington, DC. NCEA–F–0644.
given an antithyroid compound. Vet perinatal and adult exposure vs. Adult Available from: http://www.epa.gov/
Pathol 23:110–117. only exposure in carcinogenesis studies. ncea/raf/cancer.htm.
Tomatis, L; Aitio, A; Wilbourn, J; et al. (1989) Office of Pesticide Programs. U.S. EPA (U.S. Environmental Protection
Human carcinogens so far identified. Jpn Washington, DC. Agency). (1999b) Review of revised
J Cancer Res 80:795–807. U.S. EPA (U.S. Environmental Protection sections of the proposed guidelines for
Tucker, JD; Preston, RJ. (1996) Chromosome Agency). (1997a) A proposed OPP policy carcinogen risk assessment. Science
aberrations, micronuclei, aneuploidy, on determining the need for perinatal Advisory Board, Washington, DC. EPA/
sister chromatid exchanges, and cancer carcinogenicity testing on a pesticide. SAB/EC–99/015. Available from: http://
risk assessment. Mutat Res 365(1– Office of Pesticide Programs. www.epa.gov/ncea/raf/cancer.htm.
3):147–59. Review. Washington, DC. U.S. EPA (U.S. Environmental Protection
U.S. EPA (U.S. Environmental Protection U.S. EPA (U.S. Environmental Protection Agency). (1999c) Cancer risk coefficients
Agency). (1986a) Guidelines for Agency). (1997b) A set of scientific for environmental exposure to
carcinogen risk assessment. Federal issues being considered by the Agency in radionuclides: Federal guidance report
Register 51(185):33992–34003. Available connection with the criteria for requiring no. 13. Office of Air and Radiation. EPA/
from: http://www.epa.gov/ncea/raf/. in-utero cancer studies. Office of 402/R–99/001. Available from: http://
U.S. EPA (U.S. Environmental Protection Pesticide Programs. FIFRA Scientific www.epa.gov/radiation/federal.
Agency). (1986b) Guidelines for Advisory Panel. September 1997 meeting U.S. EPA (U.S. Environmental Protection
mutagenicity risk assessment. Federal report. Available from: http:// Agency). (2000a) Science Policy Council
Register 51(185):34006–34012. Available www.epa.gov/pesticides/SAP/archive/ handbook: peer review. Office of
from: http://cfpub.epa.gov/ncea/raf/ september/finalsep.htm. Research and Development, Office of
recordisplay.cfm?deid=23160. U.S. EPA (U.S. Environmental Protection Science Policy, Washington, DC. EPA/
U.S. EPA (U.S. Environmental Protection Agency). (1997c) Exposure factors 100/B–98/001. Available from: http://
Agency). (1989) Summary of the second handbook. National Center for www.epa.gov/osp/spc/prhandbk.pdf.
workshop carcinogenesis bioassay with Environmental Assessment, Washington, U.S. EPA (U.S. Environmental Protection
the dermal route. May 18–19, 1988, DC. EPA/600/P–95/002F. Available from: Agency). (2000b) U.S. EPA. Science
Research Triangle Park, NC. EPA/560/6– http://cfpub.epa.gov/ncea/cfm/ Policy Council handbook: risk
89/003. Available from NTIS, recordisplay.cfm?deid=12464. characterization. EPA Science Policy
Springfield, VA 22161. U.S. EPA (U.S. Environmental Protection Council, Washington, DC. EPA/100/B–
U.S. EPA (U.S. Environmental Protection Agency). (1997d) Policy for use of 00/002. Available from: http://
Agency). (1991a) Guidelines for probabilistic analysis in risk assessment. www.epa.gov/osp/spc/rchandbk.pdf.
developmental toxicity risk assessment. Memorandum of Fred Hansen, Deputy U.S. EPA (U.S. Environmental Protection
Federal Register 56(234):63798–63826. Administrator, May 15, 1997. Available Agency). (2000c) Supplementary
Available from: http://cfpub.epa.gov/ from: http://www.epa.gov/osp/spc/ guidance for conducting health risk
ncea/raf/recordisplay.cfm?deid=23162. probpol.htm. assessments of chemical mixtures. Risk
U.S. EPA (U.S. Environmental Protection U.S. EPA (U.S. Environmental Protection Assessment Forum, Washington, DC.
Agency). (1991b) Alpha-2u-globulin: Agency). (1997e) Guiding principles for EPA/630/R–00/002. Available from:
association with chemically induced Monte Carlo analysis. Risk Assessment http://cfpub.epa.gov/ncea/raf/
renal toxicity and neoplasia in the male Forum, Washington, DC. EPA/630/R–97/ recordisplay.cfm?deid=20533.
rat. Risk Assessment Forum, 001. Available from: http:// U.S. EPA (U.S. Environmental Protection
Washington, DC. EPA/625/3–91/019F. cfpub.epa.gov/ncea/raf/ Agency). (2000d) Guidance for data
U.S. EPA (U.S. Environmental Protection recordisplay.cfm?deid=29596. quality assessment: practical methods for
Agency). (1992a) Guidelines for exposure U.S. EPA (U.S. Environmental Protection data analysis. Office of Environmental
assessment. Federal Register Agency). (1998a) Assessment of thyroid Information, Washington, DC. EPA/600/
57(104):22888–22938. Available from: follicular cell tumors. Risk Assessment R–96/084. Available from: http://
http://cfpub.epa.gov/ncea/raf/ Forum, Washington, DC. EPA/630/R–97/ www.epa.gov/quality/qs-docs/g9-
recordisplay.cfm?deid=15263. 002. Available from: http:// final.pdf.
U.S. EPA (U.S. Environmental Protection cfpub.epa.gov/ncea/raf/ U.S. EPA (U.S. Environmental Protection
Agency). (1992b) Draft report: a cross- recordisplay.cfm?deid=13102. Agency). (2000e) EPA quality manual for
species scaling factor for carcinogen risk U.S. EPA (U.S. Environmental Protection environmental programs 5360 A1.
assessment based on equivalence of mg/ Agency). (1998b) Guidelines for Available from: http://www.epa.gov/
kg3⁄4/day. Federal Register neurotoxicity risk assessment. Federal quality/qs-docs/5360.pdf.
57(109):24152–24173. Register 63(93):26926–26954. Available U.S. EPA (U.S. Environmental Protection
U.S. EPA (U.S. Environmental Protection from: http://cfpub.epa.gov/ncea/raf/ Agency). (2001a) Health effects test
Agency). (1994) Methods for derivation recordisplay.cfm?deid=12479. guidelines. Combined chronic toxicity/
of inhalation reference concentrations U.S. EPA (U.S. Environmental Protection carcinogenicty testing of respirable
and application of inhalation dosimetry. Agency). (1998c) Health effects test fibrous particles. OPPTS 870.8355.
Office of Health and Environmental guidelines: OPPTS 870.4300 combined Available from: http://www.epa.gov/
Assessment, Environmental Criteria and chronic toxicity/carcinogenicity. Office opptsfrs/home/guidelin.htm.
U.S. EPA (U.S. Environmental Protection health. Risk Anal.22:67–81. Within Reason, Science 248:559–564.
Agency). (2001b) Notice of opportunity Van Sittert, NJ; De Jong, G; Clare, MG; et al.
to provide additional information and (1985) Cytogenetic, immunological, and [FR Doc. 05–6642 Filed 4–6–05; 8:45 am]
comment. FR 66:59593–59594. Available hematological effects in workers in an BILLING CODE 6560–50–C
from: http://cfpub.epa.gov/ncea/raf/ ethylene oxide manufacturing plant. Br J
recordisplay.cfm?deid=55868. Indust Med 42:19–26.
U.S. EPA (U.S. Environmental Protection Vater, ST; McGinnis, PM; Schoeny, RS; et al. ENVIRONMENTAL PROTECTION
Agency). (2002a) Guidelines for ensuring (1993) Biological considerations for AGENCY
and maximizing the quality, objectivity, combining carcinogenicity data for
utility and integrity for information quantitative risk assessment. Regul. [FRL–7895–1]
disseminated by the Environmental Toxicol Pharmacol 18:403–418.
Protection Agency. Office of Vesselinovitch, SD; Rao, KVN; Mihailovich, Notice of Availability of the Document
Environmental Information, Washington, N. (1979) Neoplastic response of mouse Entitled Supplemental Guidance for
DC. EPA/260/R–02/008. Available from: tissues during perinatal age periods and Assessing Susceptibility From Early-
http://www.epa.gov/oei/ its significance in chemical Life Exposure to Carcinogens
qualityguidelines/index.html. carcinogenesis. NCI Monogr 51:239.
U.S. EPA (U.S. Environmental Protection Vogelstein, B; Fearon, ER; Hamilton, SR; et AGENCY: U.S. Environmental Protection
Agency). (2002b) A review of the al. (1988) Genetic alterations during Agency (EPA).
reference dose and reference colorectal-tumor development. N Eng J ACTION: Notice of availability of final
concentration process. Risk Assessment Med 319:525–532. document.
Forum, Washington, DC. EPA/630/P–02/ Walker, KD; MacIntosh, D; Evans, JS. (2001)
002F. Available from: http:// Use of expert judgment in exposure SUMMARY: This Notice announces the
cfpub.epa.gov/ncea/raf/ assessment. Part I. Characterization of availability of the final document
recordisplay.cfm?deid=55365. personal exposure to benzene. J entitled Supplemental Guidance for
U.S. EPA (U.S. Environmental Protection Exposure Environ Epidemiol 11:308–
Agency). (2002c) Workshop on the 322.
Assessing Susceptibility from Early-Life
benefits of reductions in exposure to Walker, KD; Catalano, P; Hammitt, JK; Evans, Exposure to Carcinogens, hereafter
hazardous air pollutants: developing best JS. (2003) Use of expert judgment in referred to as Supplemental Guidance.
estimates of dose-response functions. exposure assessment: part 2. Calibration DATES: The Supplemental Guidance is
Science Advisory Board, Washington, of expert judgments about personal available for use by EPA risk assessors
DC. EPA/SAB–EC/WKSHP/02/001. exposures to benzene. J Expo Anal as of March 29, 2005.
Available from: http://www.epa.gov/ Environ Epidemiol. 13:1–16.
ADDRESSES: The Supplemental
science1/fiscal02.htm. Waters, MD; Stack, H; F. Jackson, MA. (1999)
U.S. EPA (U.S. Environmental Protection Short-term tests for defining mutagenic Guidance document is available
Agency). (2002d) Child-specific exposure carcinogens. In: McGregor, DB; Rice, JM; electronically through the EPA Web site
factors handbook (interim report). EPA/ Venitt, S, eds. The use of short term tests at http://www.epa.gov/cancerguidelines.
600/P–00/002B. Office of Research and for carcinogens and data on genetic A limited number of paper and CDROM
Development, National Center for effects in carcinogenic hazard copies will be available from the EPA’s
Environmental Assessment, Washington, evaluation. Lyon, France: International National Service Center for
DC, 448 pp. Available from: http:// Agency for Research on Cancer. IARC Environmental Publications (NSCEP),
cfpub.epa.gov/ncea/cfm/ Sci. Publ. No. 146, pp.499–536. P.O. Box 42419, Cincinnati, OH 45242;
recordisplay.cfm?deid=55145. Whitfield, RG; Wallsten, TS. (1989). A risk
U.S. EPA (U.S. Environmental Protection assessment for selected lead-induced
telephone: (800) 490–9198 or (513) 489–
Agency). (2003) A summary of general health effects: an example of a general 8190; facsimile: (513) 489–8695. Please
assessment factors for evaluating the methodology. Risk Anal. 9:197–208. provide your name, mailing address, the
quality of scientific and technical Whysner, J; Williams, GM. (1996) Saccharin title and the EPA number of the
information. Science Policy Council, mechanistic data and risk assessment: requested publication (EPA/630/R–03/
Washington, DC. EPA 100/B–03/001. urine composition, enhanced cell 003F). Additionally, copies of the
Available from: http://www.epa.gov/osa/ proliferation, and tumor promotion. Supplemental Guidance will be
spc/htm/assess2.pdf. Pharmacol Ther 71:225:252. available for inspection at EPA
U.S. EPA (U.S. Environmental Protection Willis, HH; DeKay, ML; Morgan, MG; Florig, headquarters and regional libraries,
Agency). (2004). Final Regulatory HK; Fischbeck, PS. (2004) Ecological risk
Analysis: Control of Emissions from ranking: development and evaluation of
through the U.S. Government
Nonroad Diesel Engines. Prepared by a method for improving public Depository Library program.
U.S. EPA, Office of Transportation and participation in environmental decision FOR FURTHER INFORMATION CONTACT: Dr.
Air Quality, Washington, DC, May; EPA making, Risk Anal. 24:363–78. William P. Wood, Risk Assessment
report no. EPA420–R–04–007. See Winkler, RL; Wallsten, TS; Whitfield, RG; Forum, National Center for
chapter 9 and Appendix B. Available Richmond, HM; Rosenbaum, AS. (1995). Environmental Assessment (8601D),
from: http://www.epa.gov/nonroad- An assessment of the risk of chronic lung U.S. Environmental Protection Agency,
diesel/2004fr.htm#ria. injury attributable to long-term ozone
1200 Pennsylvania Avenue NW.,
U.S. EPA (U.S. Environmental Protection exposure. Operations Research 43:19–28.
Agency). (2005) Supplemental guidance Woo, YT; Arcos, JC. (1989) Role of structure- Washington, DC 20460; telephone: (202)
for assessing cancer susceptibility from activity relationship analysis in 564–3361; facsimile: (202) 565–0062; or
early-life exposure to carcinogens. Risk evaluation of pesticides for potential e-mail: risk.forum@epa.gov.
Assessment Forum, Washington, DC. carcinogenicity. In: Ragsdale, NN; SUPPLEMENTARY INFORMATION:
Available from: http://www.epa.gov/ Menzer, RE, eds. Carcinogenicity and
ncea/raf. pesticides: principles, issues, and Background
Vainio, H; Magee, P; McGregor, D; et al. relationship. ACS Symposium Series No. In another notice in today’s Federal
(1992) Mechanisms of carcinogenesis in 414. San Diego: Academic Press; pp. Register, EPA has announced the
risk identification. IARC Sci. Pubs. No. 175–200. availability of final Guidelines for
116. Lyon, France: IARC. Yamasaki, H. (1995) Non-genotoxic
Van Der Fels-Klerx, IHJ; Goossens, LHJ; mechanisms of carcinogenesis: Studies
Carcinogen Risk Assessment (EPA/630/
Saatkamp, HW; Horst, SHS. (2002) of cell transformation and gap junctional P–03/001F), hereafter referred to as the
Elicitation of quantitative data from a intercellular communication. Toxicol Guidelines. The background and scope
heterogeneous expert panel: formal Lett 77:55–61. of the Guidelines are explained in that
process and application in animal Zeckhauser, RJ; Viscusi, WK. (1990). Risk notice. The Guidelines explicitly call for

Notice: Meetings: Carcinogen Risk Assessment

Încărcat de

Informații document

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Notice: Meetings: Carcinogen Risk Assessment

Încărcat de

Drepturi de autor:

Formate disponibile

Thursday,

BILLING CODE 6560–50–C

BILLING CODE 6560–50–C

4. Exposure Assessment important differences in exposures. may need to be assessed as a special

S-ar putea să vă placă și