Risk assessment is the systematic, scientific characterization of potential adverse effects of human exposures to hazardous agents or activities. Risk assessment as an organized activity of the federal agencies began in the 1970s. Earlier, the American Conference of Governmental Industrial Hygienists had set threshold limit values for exposures of workers, and the Food and Drug Administration (FDA) had set acceptable daily intakes of pesticide residues and food additives in the diet. In the middle 1970s, the Environmental Protection Agency (EPA) and FDA issued guidance for estimating risks associated with small exposures to potentially carcinogenic chemicals. Their guidance made estimated risks of one extra cancer over the lifetime of 100,000 people (EPA) or 1 million people (FDA) action levels for regulatory attention. Estimated risks below those levels are considered negligible because they add individually so little to the background rate of about 240,000 cancer deaths out of every 1 million people who die every year in the United States. The ultimate goal is, of course, to lower the background rate itself, part of which can be attributed to an array of pollution-generating activities.

During 1977-1980, an interagency regulatory liaison group was actively engaged in bridging scientific, statutory, and policy responsibilities and activities of EPA, FDA, the Occupational Safety and Health Administration, the Consumer Product Safety Commission, and the Food Safety and Quality Service of the Department of Agriculture. The White House Office of Science and Technology Policy (OSTP) participated in the scientific discussions supporting risk assessment and risk management and published a scheme for identifying potential hazards, characterizing risks, and managing the risks, usually by reduction of use, of emissions, or of exposures (Calkins et al. 1980) (see table 3.1).

Table 3.1 Framework for Regulatory Decision-Making.

That scheme makes clear that information about potential hazards can come from epidemiologic studies of workers and other people who are exposed to hazards, from direct experimental tests in animals and in cells in the laboratory, and from comparisons of chemical structures. The next stage involves the potency of the chemical (dose-response relationship), detailed understanding of exposure pathways, and the reasons for variation in responses among exposed people. Risk, then, is characterized both qualitatively (the nature of effects, the strength of evidence, and the reversibility or preventability of effects) and quantitatively (the probability of effects of various kinds and severities).

Performing full-scale risk assessments is a formidable task, requiring data, technical expertise, and peer review. Deciding to go forward with a risk assessment is a risk-management decision, and scaling the effort to the importance of the problem, with respect to scientific issues and regulatory impact, is crucial.

This section examines some of the risk-assessment issues that are under debate, such as assessing toxicity and relevance to humans, accounting for variations in population exposures and susceptibility, describing uncertainties, evaluating risks of chemical mixtures, conducting ecologic risk assessments, and assessing risks associated with microorganisms and radiation.

Science and Judgment in Risk Assessment (NRC 1994) evaluated EPA's cancer risk-assessment practices and concluded that, although they could use fine tuning, they were fundamentally sound. EPA responded promptly to many specific recommendations of that report and has just issued Proposed Guidelines for Carcinogen Risk Assessment (EPA 1996). The Commission evaluated cancer risk-assessment practices at EPA and other federal regulatory agencies from a risk-management perspective--a broader context. We identified several aspects that we believe can be improved further. We address here the need for a common metric to compare reduction of risk of cancer and noncancer effects, how to evaluate chemical mixtures, and how to clarify factors that affect susceptibility to toxicity resulting from chemical exposures.

FINDING 3.1.1: Scientific information on a chemical's mode of action is used to make weight-of-evidence decisions about the relevance of toxicity-test results to humans and to extrapolate doses from laboratory animals to humans. Quantitative assessment of the relationship between dose and response is a mathematical procedure that has suffered from a regulatory policy dichotomy: chemicals that are suspected of causing cancer are regulated by assuming that every exposure has some risk, but chemicals suspected to cause other effects, such as developmental or reproductive toxicity, are regulated by assuming that there is a safe level of exposure. That simple dichotomy is not fully supportable by current scientific evidence for either carcinogens or noncarcinogens. It has resulted in health risk assessments for carcinogenic and noncarcinogenic chemicals that cannot be compared and in striking discrepancies among maximal exposures considered to have negligible risk.

RECOMMENDATION: To assist in comparative risk assessment and risk characterization, a margin-of-exposure approach should be evaluated as an addition to current methods for expressing risks for carcinogens. The margin-of-exposure approach that is currently used by EPA for noncarcinogens and is proposed for carcinogens with nonlinear dose-response characteristics should be extended to carcinogens with linear dose-response characteristics, for comparison. For all types of cancer and noncancer health effects, risk assessments should use all relevant scientific information about a chemical's mode of action and disposition in the body.

RATIONALE

The distinction between "nonthreshold" carcinogens and "threshold" noncarcinogens is increasingly blurred. The standard assumption that all carcinogens are mutagens and that their dose-response relationships can be modeled by assuming low-dose linearity is inconsistent with a variety of "secondary" mechanisms of carcinogenesis now identified. Meanwhile, knowledge of mechanisms of toxicity and of variation in susceptibility undercuts the standard assumption that all noncarcinogens have a definable threshold dose below which there is no effect. Disputes over the existence of a threshold for the activity of a particular chemical or over its dose-response characteristics below what can be observed in rodent bioassays or in biologic-marker studies are unlikely to be resolved experimentally. The result of this regulatory dichotomy is that carcinogens tend to be regulated more stringently and noncarcinogens less stringently than would be the case if there were a consistent method to compare them.

A large part of the debate about cancer risk assessment has focused on identifying and arguing over the best mathematical dose-response models to apply to rodent bioassay or epidemiologic data to extrapolate, for protection of the general population, often far below the range of effects that can be observed only at high doses. Because effects of such low exposures cannot be directly observed, the accuracy of those models beyond the observable range cannot be validated experimentally. Consequently, the accuracy or validity of the potency estimates derived from the models will remain in dispute. Public-health protection is not well served by unresolved debates about mathematical dose-response models, which delay or paralyze a regulatory agency's ability to evaluate toxicity or set standards. No similar extrapolation is performed for chemicals that can cause other effects, such as lung damage or reproductive toxicity.

A margin of exposure is a ratio defined by EPA as a dose derived from a tumor bioassay, epidemiologic study, or biologic marker study, such as the dose associated with a 10% response rate, divided by an actual or projected human exposure (EPA 1996). The risk manager evaluates a particular margin of exposure and decides whether it provides an appropriate level of protection given the relevant risk-management criteria. Stakeholders can make their own judgments. The margin-of-exposure approach is similar to, but more flexible than, the method EPA uses to derive estimates of reference doses or concentrations (RfD, RfC) for noncancer effects. Criteria for evaluating the acceptability of a margin of exposure include the slope of the dose-response relationship in the observable range, the nature and extent of the uncertainties, human variability to the response of concern, and human sensitivity as compared with laboratory animals.

Crump et al. (1996) advocate harmonization of cancer and noncancer risk-assessment practices through the use of techniques that do not rely on predicting low-dose risks. They state that the hope that dose-response models consistent with knowledge of the mechanisms of carcinogenesis would provide better estimates of low-dose cancer risk is unlikely to be realized in the near future. Because there is still so much uncertainty about low-dose mechanisms, even biologically based dose-response models must rely mostly on assumptions and generally predict risks similar to those predicted using completely empirical models such as the linearized multistage modeling procedure. The questionable biologic basis of current methods for estimating carcinogenic potency is suggested by the correlation that has been described between maximum tolerated doses (MTDs) and potency (NRC 1993). The observation that MTDs of carcinogens are about 400,000 times the concentrations of carcinogens estimated to be associated with a 10^-6 upper-bound risk level is a direct arithmetic result of the linear extrapolation, not a confirmation of a biologic mechanism.

Using a margin-of-exposure approach to evaluate risks from diverse toxicants might have some advantages. First, the distinct but complementary roles of risk assessment and risk management would be transparent: identifying an appropriate effect and dose to use as the basis of risk assessment would be a science-based activity (as it is now), and identifying appropriate margins of exposure would clearly be a risk-management responsibility, requiring consensus as to the level of protection that is desired and feasible for different effects or situations. For example, FDA uses a larger margin of exposure for a substance in food that is consumed by most of the U.S. population compared to what OSHA might use for protection of workers exposed to a solvent used in only one process in one industry.

Second, a margin-of-exposure approach for all carcinogens could improve risk communication. The information base on carcinogens is often restricted to observable dose-response data from bioassays. In only a limited number of cases do additional mechanistic data aid in extrapolating between species and from high to low exposures. It therefore seems misleading to express cancer risk in terms of predicted incidence or numbers of deaths per unit of the population when cancer risk often is based on no more information than is available on noncancer effects but is expressed in a manner that implies an unwarranted degree of precision. Third, harmonizing risk assessment methods for carcinogens and noncarcinogens might permit noncarcinogens greater emphasis than they now tend to receive. And finally, it would be easier to compare cancer risks to noncancer risks for making risk-management decisions. It is difficult to know whether cleaning up a hazardous-waste site classified as posing an upper-bound incremental lifetime cancer risk of 1 in 10,000 should receive a higher or lower priority than cleaning up a site classified as having a noncancer hazard index of 10. The same problem will emerge when residual risks are characterized and compared (see section 6.1.1).

EPA's Proposed Guidelines for Carcinogen Risk Assessment (EPA 1996) use that very approach for carcinogens with nonlinear dose-response characteristics, with a margin of exposure generated from the lowest effective dose (LED) associated with a 10% response rate (LED₁₀) or another point of departure, such as a no-observed-adverse-effect level (NOAEL). The LED₁₀ concept is similar to the BMD ₁₀, or benchmark dose for a 10% response rate, widely recommended for developmental effects but not yet rigorously evaluated for other types of effects.⁽¹⁾ The toxicology and risk-assessment communities are currently engaged in a major debate about whether the benchmark-dose approach should be used for evaluating the dose-response characteristics of other types of toxicity, especially neurotoxicity; about the comparability of standards based on benchmark doses and reference doses; and about the appropriateness and consequences of using lower confidence intervals on benchmark doses derived from bioassays conducted with different testing protocols.

European countries use a margin-of-exposure approach for nongenotoxic carcinogens, and the Commission was told by Daniel Krewski, acting director of Health and Welfare Canada's Bureau of Chemical Hazards, that Canada is expected to adopt that approach for all carcinogens in the near future. Canada uses a risk measure similar to the margin of exposure called the exposure potency index, defined as the margin between estimated exposure concentrations and the dose that produces a 5% response rate (TD₀₅).

A disadvantage of a margin-of-exposure approach is that it produces results that are considered inconsistent with the needs of current methods used to perform economic analysis. We address this matter in section 4.3.

Instead of relying on estimates of cancer potency, FDA has used a "threshold of regulation" method for many years to regulate chemical additives that can migrate into foods from packaging material. FDA compiled a computerized database on hundreds of carcinogens, found that their potencies were lognormally distributed, and chose a concentration that generally would be associated with a cancer risk of 10^-6 or less no matter how potent a carcinogen might be (Rulis 1989). According to FDA's testimony before the Commission, that concentration, 0.5 part per billion, is a concentration that generally represents a "reasonable certainty of no harm"; if a compound were present in food at that concentration, even if it were found to be a carcinogen, its risk to humans would be well below the highest risk that is considered negligible. To protect at a level sufficient to ensure less than an upper-bound lifetime risk of 10^-6 associated with carcinogens as potent as the dioxin 2,3,7,8-TCDD or the fungal toxin aflatoxin B₁, however, the threshold of regulation would have to be so low (lower than 1 part per trillion in the diet) that virtually nothing would be able to pass the threshold-of-regulation criterion. (If an additive is known to be a carcinogen, it is regulated under the "Delaney clause" or another authority and is not permitted to be added to food at any concentration.) The advantage of the threshold-of-regulation approach is that regulatory activity for additives present below the threshold concentration is avoided; resources are focused instead on substances that might be of greater concern (see section 5.3 on bright lines).

Consideration of the relationships between dose and response is the fundamental principle of toxicology. Regulatory priority should be given to incorporating information about mechanisms of action and disposition to override default assumptions used to estimate small risks. We are hopeful that biologic markers of early effects can be validated as essential intermediate points in the pathophysiologic pathways of carcinogenesis and thereby provide more information about dose-response relationships and a better basis for relating animal and human responses.

FINDING 3.1.2: Chemicals that cause cancer in rodents are appropriately considered potentially carcinogenic in humans. However, some chemicals elicit tumors in rodents through mechanisms that are unlikely to have any corresponding effect in humans. Others elicit tumors only at very high doses that are unlikely to be relevant to human exposures. Lingering controversies about those responses undermine the general reliance on rodent-bioassay results. Regulatory agencies have been cautious in recognizing the distinctions and in issuing guidance on when such rodent responses should be discounted or disregarded. As this report was going to press, EPA released its Proposed Guidelines for Carcinogen Risk Assessment, which include a category "not likely" to be carcinogenic in humans that includes chemicals with irrelevant modes of action.

RECOMMENDATION: Although the results of rodent bioassays provide valuable information about chemicals' potential risks to humans, some rodent cancer responses should be classified as irrelevant to human cancer risk assessment. If, after adequate testing, a chemical is found to produce only tumors that occur as a result of mechanisms or doses that have been deemed not relevant to humans, it should be unnecessary to generate cancer-potency estimates. Regulatory agencies should develop consistent criteria for making those distinctions, and the criteria should be updated as scientific knowledge about the mechanisms responsible evolves.

RATIONALE

The policy of presuming that a chemical that causes cancer when tested in laboratory rodents is potentially carcinogenic in humans is justified by considerable evidence and by the precautionary principle of being protective when uncertain. That policy is undercut, however, when rodent tumor responses that can be shown to be irrelevant to humans are not excluded from consideration. Furthermore, from a risk-management perspective, it is wasteful to expend risk-assessment resources, risk-management time, and public and legal involvement nonproductively revisiting such policy issues chemical by chemical.

Table 3.2 lists examples of rodent mechanisms and tumor responses that are leading candidates for classification as "not likely" to be carcinogenic in humans according to EPA's Proposed Guidelines for Carcinogen Risk Assessment. That classification includes a subcategory of agents that elicit only rodent tumors that are irrelevant to human risk and another of agents that produce tumors at doses and via routes of exposure that need to be compared with known human occupational and general-population exposures to determine relevance. Chemicals that produce tumors only in rodents because of striking pharmacokinetic differences can also be addressed. In general, the chemicals listed in table 3.2 are not genotoxic; that is, they do not react directly with DNA. Instead, they cause local injury or otherwise stimulate local hyperplasia and cell division, which might be associated with a low incidence of tumor formation because of chronic overstimulation.

Tumor Site	Tumor Mechanism	Rodent Carcinogens
Male rat kidney	-2u globulin-induced nephropathy	D-limonene, unleaded gasoline, isophorons
Forestomach	Local hyperplasia	BHA, propionic acid, ethyl acrylate (administered by gavage)
Male rat bladder	Reactive hyperplasia from cytotoxic precipitated chemicals	Saccharin, cyromazine, melamine, nitriloacetic acid, fosetyl-Al
Lung	Overwhelming of clearance mechanism	Various particles, including titanium dioxide and carbon black
Thyroid	Sustained excessive hormonal stimulation	EBDC fungicides, goitrogens, amitrol, sulfamethazine

Some chemicals are recognized to induce the accumulation of large amounts of a protein containing -2u globulin in the male rat kidney. Most scientists agree that this accumulation leads to damage to the kidney tubules, cell death, sustained cell proliferation, and tumor formation (see Melnick et al. 1996 for alternative viewpoint). It does not occur in female rats or in other species, including humans. It was extensively studied and reviewed by EPA's Risk Assessment Forum and Science Advisory Board from 1988 to 1991, and it was decided to disregard that particular rodent response for certain chemicals (EPA 1991). If that response is disregarded, risk assessment and regulation can be directed, as appropriate, at other adverse effects, including kidney tumors not due to this protein-mediated mechanism. Two problems remain, however: the very long time that it took to reach a decision, and the reluctance to apply the decision to the tumor response instead of to individual chemicals producing the response.

Another tumor response that is irrelevant to humans is that which occurs in the rodent forestomach after administration of a chemical by gavage (that is, via a tube placed in the stomach). Gavage is convenient for determining whether a chemical can cause tumors in organs distant from the stomach after absorption into the bloodstream, but it can result in local cytotoxicity and hyperplasia. At least three commercially important chemicals (table 3.2) have been found to produce tumors only in the forestomach. For example, butylated hydroxyanisole (BHA) was reviewed for FDA by a Federation of American Societies for Experimental Biology panel, which concluded in 1994 that there is a threshold for its tumor-producing effect, cell proliferation. There is no evidence of a similar effect in humans (who lack forestomachs) and no scenario in which similar high-dose local exposure would occur.

Rodent bladder tumors became famous during the saccharin debate of 1978-1979. Regulatory agencies later supported an International Life Sciences Institute panel on rodent bladder carcinogenesis, which concluded that chemicals that precipitate in urine, or that elicit effects leading to precipitation of other chemicals should be considered carcinogens only at high doses (Neumann and Olin 1995). If human exposures to such chemicals are much lower than the doses tested, the rodent response can be disregarded. Bladder tumors can arise by other mechanisms (Cohen et al. 1995).

Grossly overloading the lung's clearance mechanisms by administering particles directly to the lung was considered irrelevant to humans by EPA in the case of titanium dioxide, which was delisted from the Toxic Release Inventory in 1988 for this reason (Fed Reg 53:23107-23202, 1988). The phenomenon is applicable to particles in general, not only to titanium dioxide, but it has been declared irrelevant to humans only in the case of titanium dioxide. Criteria are needed to determine what are "gross" particle overloads. Ultrafine preparations of carbon black or other materials might present a risk at lower concentrations.

Thyroid tumorigenesis in the rat has been under review by EPA at least since 1988. Various pesticides and fungicides induce liver enzymes or thyroid enzymes that affect thyroid hormone levels and lead to hyperplasia and ultimately tumor formation in rodents. The response might be evoked by high doses only. The feedback and transport systems for rodent thyroid hormones are very different from those in humans (McClain 1994). Although there is no doubt that humans are far less sensitive to this response, agency consensus appears to be that the animal model cannot be disregarded.

Finally, there have been many challenges to the interpretation of mouse liver-tumor formation (not listed in table 3.2). At least six potential mechanisms have been described, some of which occur in humans and some of which do not. Mouse liver tumors are among the most common seen in bioassays and pose particularly vexing problems for interpreting effects of chlorinated organic solvents.

Bringing a risk-management perspective to the scientific-review process might galvanize action in a way that normal agency procedures have not. At least 10 years passed before EPA acted on the male rat kidney-tumor response. Over 15 years have passed since the human relevance of saccharin's carcinogenicity was doubted, but packages of sugar substitutes including saccharin still carry warning labels required by Congress. Banning the detergent nitriloacetic acid, a rodent bladder carcinogen, led to increased use of phosphate detergents, with serious ecological effects. The Commission recognizes that time is required to investigate chemicals' modes of action and endorses EPA's current plans to identify tumor responses in rodents that are not likely to be relevant to humans. We encourage EPA to apply those distinctions as early as possible in the risk-assessment process, before time and resources are wasted. Other agencies should follow similar practices.

FINDING 3.1.3: Current regulatory approaches for reducing risks associated with chemical exposures generally do not reflect differences in individual susceptibility or encourage getting evidence to identify them. Genetic, nutritional, metabolic, and other differences make some segments of a population more susceptible than others to the effects of a given exposure to a given chemical.

RECOMMENDATION: Risk assessments should include consideration of genetic and other host differences in susceptibility and identify especially susceptible human subpopulations for specific chemical exposures. Available information on the range of a population's susceptibility should be considered. Where appropriate, knowledge of differences in susceptibility should be used to support additional "bright lines" or standards for chemical exposure concentrations, to protect especially susceptible subpopulations (see section 5.3).

RATIONALE

Susceptibility to the effects of chemical exposures depends on the sensitivity of a person's response to different doses. Susceptibility is influenced by many factors, including age, sex, genetic variation in metabolism of chemicals, genetic variation in response to agents or stressors at their sites of action, ethnic origin and ethnic practices, socioeconomic status, geographic location, and lifestyle factors, such as smoking, alcoholic-beverage consumption, diet, physical activity, and recreational habits. Dose-response relationships are chemical-specific and depend on modes of action; people are not hypersusceptible to all kinds of exposures (Omenn et al. 1982). The influence of concurrent exposures on risk is discussed in section 3.2. The following are examples of subpopulations potentially at higher risk.

Population	Factor Affecting Response to Exposure
Asthmatics	Increased airway responsiveness to allergens, respiratory irritants, and infectious agents
Fetuses	Sensitivity of developing organs to toxicants that cause birth defects
Infants and young children	Sensitivity of developing brain to neurotoxic agents such as lead
1-Antitrypsin-deficient persons	Inherited deficiency of a protein that protects against chemical damage
Glutathione-S-transferase-deficient	Diminished detoxification of some carcinogens and medicines
Elderly	Diminished detoxification and elimination mechanisms

There are opportunities to identify, evaluate, and reduce risks to sensitive people. Asthmatics, for example, make up 5-10% of the general population in the United States. Some air pollutants, especially sulfur oxides, particles, and ozone, are respiratory irritants that pose a greater risk to this subpopulation than to the general public. Both the number of cases of asthma and the number of deaths from asthma are increasing in the United States. Blacks have a 15% higher prevalence of asthma than whites. Likewise, susceptibility to lung cancer appears to vary among ethnic groups; in the United States, the incidence of lung cancer in black men is 1.5 times that in white men, 2.5 times that in Hispanic men, 2-4 times that in Asian men, and 8 times that in American Indian men (NCI 1984). One source of individual and ethnic differences in susceptibility is differences in the activity of enzymes that affect chemical toxicity. Increased risks of cancers of the bladder, skin, colon, lung, and stomach have been associated with differences in the activity of specific enzymes that can activate or deactivate carcinogens. Susceptibility to organophosphate pesticide toxicity is also markedly influenced by the activity of a specific enzyme.

Congressional amendments to the Safe Drinking Water Act, amendments to the Federal Insecticide, Fungicide and Rodenticide Act, regulatory-reform legislation, and other bills would require such recognizable subpopulations as the elderly, children, and women of child-bearing age to be identified and considered in risk characterization and in standard-setting. Recognition of subgroup susceptibility does not necessarily result in more stringent regulation. For example, people allergic to particular chemicals or pet-animal proteins might modify their exposures or modify their responses (with medication). Identifying the size of the population at risk and describing the risk peculiar to that population during risk characterization, perhaps relying on biologic markers of susceptibility, will make it possible to characterize risks more realistically than is possible using only estimates for the general population. Risk-communication messages can then be targeted more effectively.

1For a detailed discussion of the benchmark dose, see the report prepared for the Commission by Elaine Faustman (abstract found in appendix A.5).