Chapter One
Principles of Cancer Risk Assessment: The Risk Assessment Paradigm
Vincent James Cogliano
International Agency for Research on Cancer
1.1 THE RISK ASSESSMENT PARADIGM
The United States National Research Council has outlined a series of steps that many agencies follow in analysing risk (National Research Council, 1983, 1994). Although this chapter is focused on cancer risk assessment, the risk assessment paradigm can apply equally to other adverse effects.
Hazard identification asks whether exposure to an agent can cause an increase in the incidence of an adverse effect. This involves consideration of the epidemiologic evidence in humans, the evidence in experimental animals, and other relevant data, including studies of toxicokinetics and mechanisms.
Dose-response assessment characterises the relationship between the dose of an agent and the incidence of an adverse health effect. The fundamental activity of dose-response assessment is modelling, which can range from simple linear (proportionality) models to more complex toxicokinetic and mechanism-based models. The purpose of the modelling is to extrapolate from the conditions observed in the epidemiologic and experimental studies to the conditions that are of interest in a human exposure situation. It is common for risk assessments to extrapolate from high doses to lower doses, from experimental animals to humans, from one route of exposure to another, or from one pattern of exposure to another. There are also other extrapolations that have received less discussion.
Exposure assessment determines the extent of human exposure to an agent. It identifies the pathways by which a population can be exposed, for example, through breathing contaminated air, eating contaminated fish, or swimming in a contaminated lake. For each pathway, an exposure assessment estimates how much of the agent the population is exposed to. This can depend on the magnitude, duration, and frequency of exposure.
Risk characterisation describes the nature and magnitude of human risk, including attendant uncertainty. It integrates the hazard identification, dose-response assessment, and exposure assessment. It often emphasises a quantitative component that estimates risk by determining where an exposure estimate falls on the dose-response curve. The risk characterisation also includes a qualitative component that describes the assessment's strengths and limitations and identifies some key research needs.
The hazard identification and dose-response assessment steps are generally regarded as descriptive of the properties of a toxic agent. For example, health agencies may identify a hazard by describing a chemical agent as "carcinogenic to humans" or "possibly carcinogenic to humans". They may also estimate a dose-response curve for each agent (US Environmental Protection Agency (US EPA), 2003a). These steps are generally performed centrally, as toxic properties and dose-response curves are not thought of as varying from one location to another. In contrast to these centralised steps of a risk assessment, the exposure assessment and risk characterisation steps address a specific human population. These latter steps consider the various exposure pathways that may be involved, estimate the level of human exposure to the toxic agent, and then characterise the risk by determining where that level of exposure falls on the dose-response curve. Exposure assessment and risk characterisation often depend on specific knowledge of local populations and exposure conditions; accordingly, they are often performed at a more decentralised level.
In the future, this distinction between centralised hazard identification and dose-response assessment and decentralised exposure assessment and risk characterisation may break down as more is known about the distribution of genetic polymorphisms in different ethnic groups, and it becomes possible to estimate different dose-response curves for different segments of the population. Similarly, as more is known about the effects of one agent on the absorption, metabolism, and elimination of another agent, it may become possible to estimate different dose-response curves for different populations based on their level of exposure to other agents that modify the effects of the agent being assessed. At this time, however, because not enough is known about polymorphisms and the interactions of multiple agents to be able to estimate different dose-response curves for different populations and exposure conditions, dose-response assessment has generally estimated a single dose-response curve for each agent.
Because this chapter is focused on the assessment of potential carcinogens, the hazard identification and dose-response assessment steps are treated in greater detail.
1.2 HAZARD IDENTIFICATION
The data considered for hazard identification include human epidemiologic studies, long-term bioassays in experimental animals, and other relevant data on toxicokinetics and cancer mechanisms. Each source of data has a role in the hazard assessment. Epidemiologic studies can provide unequivocal evidence of carcinogenicity, but often are not sufficiently sensitive to identify a carcinogenic hazard except when the risk is high or involves an unusual form of cancer. For this reason, animal studies generally provide the best means of assessing potential risks to humans. To answer questions about the similarity of response between animals and humans, studies of toxicokinetics and mechanisms have been employed. Toxicokinetic studies investigate similarities and differences in absorption, distribution, metabolism, and elimination across species. Mechanistic studies can elucidate the chemical species and cellular processes involved in cancer initiation and development, including short-term tests to identify the potential for genetic toxicity.
The conclusion is often a judgement that weighs the evidence that an agent may or may not cause a specific adverse effect. For example, the International Agency for Research on Cancer uses the following terms to describe a cancer hazard:
carcinogenic to humans (Group 1),
probably carcinogenic to humans (Group 2A),
possibly carcinogenic to humans (Group 2B),
not classifiable as to its carcinogenicity to humans (Group 3),
probably not carcinogenic to humans (Group 4).
Other agencies (for example, US EPA, 1986b, 2003a) have adopted similar descriptors for potential carcinogens.
A prominent trend in cancer hazard identification has been the greater availability and use of mechanistic information. Mechanistic studies attempt to identify the series of key precursor events that are involved in cancer development. This may permit a judgement about whether the mechanisms that cause cancer in experimental animals are likely to be operative in humans. Once the key precursor events involved in cancer development are identified, mechanistic studies also allow identification of sub-populations and life-stages that may be especially susceptible to cancer induced by the agent. In any use of mechanistic data, it is crucial to investigate whether more than one mechanism may be operating.
1.3 DOSE-RESPONSE ASSESSMENT
1.3.1 Different objectives, different data sets, different approaches
There have been two broad approaches to dose-response assessment. Safety assessment focuses on determining a safe dose for human exposure. This has generally involved identifying a dose that appears to be without adverse effects in the experimental studies and dividing that dose by several factors to arrive at a dose considered "safe" for human exposure. The other approach provides estimates of risk at low doses, by fitting dose-response models to data on the incidence of an adverse effect and using these models to estimate the incidence at lower levels of exposure. A large majority of cancer assessments have been of the latter variety, using models to estimate the cancer risk over a range of potential exposure levels.
The models used in dose-response assessment can be considered to belong to two broad classes. Empirical models are based on fitting a standard curve to data. (They are sometimes called curve-fitting models.) They are the least detailed models, describing the incidence of frankly observable adverse effects as a mathematical function of exposure to the agent. Such models are not based on specific knowledge about biological processes and mechanisms.
In contrast, physiologically based toxicokinetic models and mechanism-based dose-response models are based on specific knowledge about the biological processes and mechanisms leading from exposure to disease. (They are sometimes called biologically based models.) Toxicokinetic models simulate the relationship between external exposure and delivered dose at the target tissue. They can incorporate detailed information on an agent's absorption, distribution, metabolism, and elimination. Mechanism-based models simulate the relationship between cellular responses at the target tissue, precursor effects, and frankly observable adverse effects in the organism. These more detailed models require extensive knowledge of the disposition of a chemical in the body and the sequence of events leading to toxicity and disease.
Safety assessments, too, have begun to move toward increasing complexity. Formerly, dividing an apparently "safe" dose by a factor of 100 was believed to yield a dose fit for human exposure. This factor of 100 was later described as composed of a factor of 10 to account for differences between humans and experimental animals and another factor of 10 to account for differences between susceptible humans and the general population. More recently, these factors of 10 have in turn been described as being composed of a factor of 3 (being approximately the square root of 10) for toxicokinetic differences and another factor of 3 for toxicodynamic differences, or alternatively, a factor of 4 for toxicokinetic differences and a factor of 2.5 for toxicodynamic differences (assuming that these approximate factors can be refined to this level of precision). In addition, other factors have been used to account for differences between experimental conditions and human exposure conditions, for example, when a short-term animal study is used to infer a dose "safe" for lifetime human exposure, or when the experimental database lacks adequate studies of key adverse effects or studies that are pertinent to exposure during pre-natal and post-natal development. Another recent trend in safety assessment has been to replace the various factors of 10 by data-derived factors that attempt to reflect more precisely the degree of difference between humans and experimental animals.
Dose-response models can be developed for different objectives. For example, a dose-response model can serve as a framework for organising and synthesising the available data and identifying research needs. In this case, a rather detailed model might be appropriate, using default assumptions and parameter values as placeholders to indicate where further research is needed. A different objective of dose-response modelling would be to provide projections of the range of risks that populations can face under different actual or hypothetical exposure conditions. In this case, a model that is well supported by data and not overly detailed might be most useful when a government agency needs to reassure a community that they do not face an unsafe exposure condition. In this case, the more speculative components of a research-oriented model may not be useful in a governmental determination of the risk that may be present.
1.3.2 Extrapolations in dose-response assessment
The fundamental objective of risk assessment, as with modelling generally, is extrapolation. Experimental data observed under specified conditions are extrapolated to human exposure conditions that may be similar in some respects and dissimilar in other respects from those in the experiments. Similarly, epidemiologic data observed in one human population exposed to certain conditions may be extrapolated to other human populations and other exposure conditions. The models developed as part of a dose-response assessment describe how these extrapolations are made and provide a basis for evaluating how well the extrapolations are supported by data. Several kinds of extrapolations are made in dose-response assessment.
Extrapolation from high doses to lower doses is often a central objective of dose-response modelling. For example, studies in experimental animals generally use high doses to determine the effects that a specific agent can induce at some dose. This is often a practical necessity when fewer than, say, 100 animals are tested (and consequently a response rate of less than 1 percent cannot be observed), because government agencies are interested in ensuring that the risk to an exposed population is much less than 1 percent. This also happens when occupational studies are used to infer the potential for risk at environmental or occupational levels that are anticipated to be lower than those observed in prior occupational settings. Extrapolation from experimental animals to humans is also a common objective of dose-response modelling. Such extrapolation is often necessary just to compare modelling results from experiments involving different animal species. Although modelling can proceed using either animal dose metrics or human dose metrics, when the ultimate objective is a statement regarding risks to exposed humans, animal dose metrics must be converted into an equivalent human dose. Route-to-route extrapolation is the application of a dose-response relationship estimated from studies involving one exposure route (for example, ingestion, inhalation, or dermal exposure) to another exposure route. This form of extrapolation has both a qualitative and a quantitative component. Qualitatively, one must make a judgement about whether the effects observed following exposure by one route are pertinent to another exposure route. Such a judgement is generally warranted when effects are observed at a site distant from the site of entry and when absorption can occur by either exposure route to give an internal dose of the agent. Such a judgement is sometimes specified as a default option in the absence of adequate data to the contrary. Extrapolation from one pattern of exposure to another involves using Information derived from occupational exposure patterns or experimental exposure protocols (for example, a single daily dose five days a week for 24 months) to make inferences about other human exposure patterns that are likely to be different. This common extrapolation, used in most dose-response assessments, has typically been handled by the use of default dose metrics such as average daily dose or cumulative dose. Extrapolation from small samples to larger populations generally has not been explicitly discussed in risk assessments. The experimental uncertainty inherent in using small experimental samples can be described by the confidence bounds associated with the estimates from the experimental studies. An important qualitative concern, however, with the use of small samples in experimental studies is the need to discuss the experiment's power to detect adverse effects. For example, a response that is not statistically significant does not indicate a threshold; rather, it can be consistent with a small risk that falls below the experiment's power of detection. A similar concern is that small samples generally do not include adequate numbers to make reliable inferences about safe doses for susceptible individuals. Extrapolation from relatively homogeneous groups to more heterogeneous populations also generally has not been explicitly discussed in risk assessments. It can, however, strongly influence the shape of a dose-response curve. This is particularly important in view of the fact that many experimental studies use relatively homogeneous, genetically similar animals. Similar animals may have a tendency to respond at similar dose levels, while a more heterogeneous population would not necessarily all respond at the same dose level. This may be true even when the mechanism of action suggests a threshold dose below which adverse effects would not occur, because the threshold could vary across a heterogeneous population and the population dose-response curve could become indistinguishable from those associated with non-threshold models. In the human population, genetic and lifestyle risks to exposed humans, animal dose metrics must be converted into an equivalent human dose.
(Continues...)
Excerpted from Recent Advances in Quantitative Methods in Cancer and Human Health Risk Assessment Copyright © 2005 by John Wiley & Sons, Ltd . Excerpted by permission.
All rights reserved. No part of this excerpt may be reproduced or reprinted without permission in writing from the publisher.
Excerpts are provided by Dial-A-Book Inc. solely for the personal use of visitors to this web site.
Best-sellers | New | Upcoming | Calendars 2009 | 10% off Amazon Books Pricing on over 900,000 Titles
