Chapter 4. Risk Assessment

• Risk assessment is the systematic scientific characterization of potential adverse health effects resulting from human exposures to hazardous agents or situations.
• Risk is defined as the probability of an adverse outcome under specified conditions.
• Risk management refers to the process by which policy actions are chosen to control hazards.

Toxicologic research and toxicity testing conducted and interpreted by toxicologists constitute the scientific core of an important activity known as risk assessment for chemical exposures. The National Research Council detailed the steps of hazard identification, dose–response assessment, exposure analysis, and characterization of risks in Risk Assessment in the Federal Government: Managing the Process (widely known as The Red Book). The scheme shown in Figure 4–1 provides a consistent framework for risk assessment across agencies with bidirectional arrows showing an ideal situation where mechanistic research feeds directly into risk assessments and critical data uncertainty drives research. Often, public policy objectives require extrapolations that go far beyond the observation of actual effects and reflect different tolerances for risks, generating controversy.

The Presidential/Congressional Commission on Risk Assessment and Risk Management formulated a comprehensive framework that applies two crucial concepts: (1) putting each environmental problem or issue into public health and/or ecological context and (2) proactively engaging the relevant stakeholders, affected or potentially affected community groups, from the very beginning of the six-stage process shown in Figure 4–2. Particular exposures and potential health effects must be evaluated across sources and exposure pathways and in light of multiple endpoints, and not the current general approach of evaluating one chemical in one environmental medium (air, water, soil, food, and products) for one health effect at a time.

Risk assessment is the systematic scientific characterization of potential adverse health effects resulting from human exposures to hazardous agents or situations. Risk is defined as the probability of an adverse outcome based on the exposure and potency of the hazardous agent(s). The term hazard is used in the United States and Canada to refer to intrinsic toxic prop-erties, whereas internationally this term is defined as the probability of an adverse outcome. Risk assessment requires qualitative information about the strength of the evidence and the nature of the outcomes—as well as quantitative assessment of the exposures, host susceptibility factors, and potential magnitude of the risk—and then a description of the uncertainties in the estimates and conclusions. The objectives of risk assessment are outlined in Table 4–1.

The phrase characterization of risk reflects the combination of qualitative and quantitative analyses. Unfortunately, many users tend to equate risk assessment with quantitative risk assessment, generating a number for an overly precise risk estimate, while ignoring crucial information about the mechanism of effect across species, inconsistent findings across studies, multiple variable health effects, and means of avoiding or reversing the effects of exposures.

Risk management refers to the process by which policy actions are chosen to control hazards identified in the risk assessment/risk characterization stage of the six-stage framework (Figure 4–2). Risk managers consider scientific evidence and risk estimates—along with statutory, engineering, economic, social, and political factors—in evaluating alternative options and choosing among those options.

Risk communication is the challenging process of making risk assessment and risk management information comprehensible to community groups, lawyers, local elected officials, judges, business people, labor, and environmentalists. A crucial, too-often neglected requirement for communication is listening to the fears, perceptions, priorities, and proposed remedies of these “stakeholders.”

Risk management decisions are reached under diverse statutes in the United States and many other countries. Some statutes specify reliance on risk alone, whereas others require a balancing of risks and benefits of the product or activity (Table 4–1). Risk assessments provide a valuable framework for priority setting within regulatory and health agencies, in the chemical development process within companies, and in resource allocation by environmental organizations. Currently, there are significant efforts toward the harmonization of testing protocols and the assessment of risks and standards.

A major challenge for risk assessment, risk communication, and risk management is to work across disciplines to demonstrate the biological plausibility and clinical significance of the conclusions from studies of chemicals thought to have potential adverse effects. Biomarkers of exposure, effect, or individual susceptibility can link the presence of a chemical in various environmental compartments to specific sites of action in target organs and to host responses. Individual behavioral and social risk factors may be critically important to both the risk and the reduction of risk. Finally, public and media attitudes toward local polluters, other responsible parties, and relevant government agencies can greatly influence the communication process and the choices for risk management.

Assessing Toxicity of Chemicals—Methods

Structure/Activity Relationships

Given the cost of $2 to $4 million and the 3 to 5 years required for testing a single chemical in a lifetime rodent carcinogenicity bioassay, initial decisions on whether to continue development of a chemical, submit a premanufacturing notice, or require additional testing may be based largely on structure/activity relationships and limited short-term assays. A chemical's structure, solubility, stability, pH sensitivity, electrophilicity, volatility, and chemical reactivity can be important information for hazard identification.

Structure–activity relationships have been used for assessment of complex mixtures. However, it is difficult to predict activity across chemical classes and especially across multiple toxic endpoints using a single biological response. Computerized SAR methods have given limited results, whereas three-dimensional (3D) molecular modeling, which utilizes pharmacophore mapping, 3D searching and molecular design, and establishment of 3D quantitative structure–activity relationships, has been more successful.

In Vitro and Short-Term Tests

The next approach for hazard identification comprises tests ranging from in vitro bacterial mutation assays to more elaborate short-term tests such as skin-painting studies in mice or altered rat liver-foci assays conducted in vivo. Other assays evaluate development, reproduction, neurotoxicity, and immunotoxicity. New assay methods from molecular and developmental biology for developmental toxicity risk assessment should accelerate use of a broader range of model organisms and assay approaches for noncancer risk assessments. Short-term assay validation and application is particularly important to risk assessment because such assays can provide information about mechanisms of effects while being faster and less expensive than lifetime bioassays.

Animal Bioassays

Animal bioassay data are key components of the hazard identification process. A basic premise of risk assessment is that chemicals that cause tumors in animals can cause tumors in humans. All human carcinogens that have been adequately tested in animals produce positive results in at least one animal model. Although this association cannot establish that all agents and mixtures that cause cancer in experimental animals also cause cancer in humans, nevertheless, in the absence of adequate data on humans, it is biologically plausible and prudent to regard agents and mixtures for which there is sufficient evidence of carcinogenicity in experimental animals as if they presented a carcinogenic risk to humans—a reflection of the “precautionary principle.” The USEPA assumes relevance of animal bioassays unless lack of relevance for human assessment is specifically determined. In general, the most appropriate rodent bioassays are those that test exposure pathways of most relevance to predicted or known human exposure pathways. Bioassays for reproductive and developmental toxicity and other noncancer endpoints have a similar rationale.

Consistent features in the design of standard cancer bioassays include testing in two species and both sexes, with 50 animals per dose group and near lifetime exposure. Important choices include the strains of rats and mice, the number of doses, and dose levels (typically 90, 50, and 10 to 25 percent of the maximally tolerated dose [MTD]), and the details of the required histopathology (number of organs to be examined, choice of interim sacrifice pathology, etc.). Positive evidence of chemical carcinogenicity can include increases in number of tumors at a particular organ site, induction of rare tumors, earlier induction (shorter latency) of commonly observed tumors, and/or increases in the total number of observed tumors.

Critical problems exist in using the hazard identification data from rodent bioassays for quantitative risk assessments. This is because of the limited dose–response data available from these rodent bioassays and nonexistent response information for environmentally relevant exposures. Results thus have traditionally been extrapolated from a dose–response curve in the 10 to 100 percent biologically observable tumor response range down to 10−6 risk estimates (upper confidence limit) or to a benchmark or reference dose-related risk.

Addition of mechanistic investigations and assessment of multiple noncancer endpoints into the bioassay design represent important enhancements of lifetime bioassays. It is feasible and desirable to tie these bioassays together with mechanistically oriented short-term tests and biomarker and genetic studies in lower doses than those leading to frank tumor development and help address the issues of extrapolation over multiple orders of magnitude to predict response at environmentally relevant doses.

In an attempt to improve the prediction of cancer risk to humans, transgenic mouse models have been developed as possible alternatives to the standard 2-year cancer bioassay. By using mice that incorporate or eliminate a gene that is linked to human cancer, these transgenic models have the power to improve the characterization of key cellular processes and the mode of action of toxicological responses. It is suggested that these models currently should not replace the 2-year assay, but should be used in conjunction to assist in the interpretation of additional toxicologic and mechanistic evidence.

Use of Epidemiologic Data in Risk Assessment

The most convincing line of evidence for human risk is a well-conducted epidemiologic study in which a positive association between exposure and disease has been observed. Table 4–2 shows examples of epidemiologic study designs and provides clues on types of outcomes and exposures evaluated. There are important limitations. When the study is exploratory, hypotheses are often weak. Exposure estimates are often crude and retrospective, especially for conditions with long latency before clinical manifestations appear. Generally, there are multiple exposures, especially when a lifetime is considered. There is always a trade-off between detailed information on relatively few persons and very limited information on large numbers of persons. Contributions from lifestyle factors, such as smoking and diet, are a challenge to sort out. Humans are highly outbred, so the method must consider variation in susceptibility among those who are exposed.

Nevertheless, human epidemiology studies provide very useful information for hazard identification and sometimes quantitative information for data characterization. Three major types of epidemiology study designs are available: cross-sectional studies, cohort studies, and case–control studies (Table 4–2). Cross-sectional studies survey groups of humans to identify risk factors (exposure) and disease but are not useful for establishing cause and effect. Cohort studies evaluate individuals selected on the basis of their exposure to an agent under study. These prospective studies monitor over time individuals who initially are disease-free to determine the rates at which they develop disease. In case–control studies, subjects are selected on the basis of disease status: disease cases and matched cases of disease-free individuals. Exposure histories of the two groups are compared to determine key consistent features in their exposure histories. All case–control studies are retrospective studies.

Epidemiologic findings are judged by the following criteria: strength of association, consistency of observations (reproducibility in time and space), specificity (uniqueness in quality or quantity of response), appropriateness of temporal relationship (did the exposure precede responses?), dose–responsiveness, biological plausibility and coherence, verification, and analogy (biological extrapolation). In addition, epidemiologic study designs should be evaluated for their power of detection, appropriateness of outcomes, verification of exposure assessments, completeness of assessing confounding factors, and general applicability of the outcomes to other populations at risk. Power of detection is calculated using study size, variability, accepted detection limits for endpoints under study, and a specified significance level.

Recent advances from the human genome project, increased sophistication of molecular biomarkers, and improved mechanistic bases for epidemiologic hypotheses have allowed epidemiologists to expand our understanding of biological plausibility and clinical relevance. “Molecular epidemiology” with improved molecular biomarkers of exposure, effect, and susceptibility has allowed investigators to more effectively link molecular events in the causative disease pathway. The range of biomarkers has grown dramatically and includes identification of single nucleotide polymorphisms, genomic profiling, transcriptome analysis, and proteomic analysis.

Integrating Quantitative Aspects of Risk Assessment

Quantitative considerations in risk assessment include dose–response assessment, exposure assessment, variation in susceptibility, and characterization of uncertainty.

The fundamental basis of the quantitative relationships between exposure to an agent and the incidence of an adverse response is the dose–response assessment. Analysis of dose–response relationships must start with the determination of the critical effects to be quantitatively evaluated. It is usual practice to choose the data sets with adverse effects occurring at the lowest levels of exposure; the “critical” adverse effect is defined as the significant adverse biological effect that occurs at the lowest exposure level.

Threshold Approaches

Threshold dose–response relationships characterization includes identification of “no or lowest observed adverse effect levels” (NOAELs or LOAELs). On the dose–response curve illustrated in Figure 4–3, the threshold, indicated as T, represents the dose below which no additional increase in response is observed. The NOAEL is identified as the highest nonstatistically significant dose tested; in this example it is point F, at 2 mg/kg body weight. Point G is the LOAEL (2.5 mg/kg body weight), as it is the lowest dose tested with a statistically significant effect.

In general, animal bioassays are constructed with sufficient numbers of animals to biological responses at the 10 percent response range. Significance usually refers to both biological and statistical criteria and is dependent on the number of dose levels tested, the number of animals tested at each dose, and background incidence of the adverse response in the nonexposed control groups. The NOAEL should not be perceived as risk-free.

As described in Chapter 2, approaches for characterizing dose–response relationships include identification of effect levels such as LD50 (dose producing 50 percent lethality), LC50 (concentration producing 50 percent lethality), ED10 (dose producing 10 percent response), as well as NOAELs.

NOAELs have traditionally served as the basis for risk assessment calculations, such as reference doses or acceptable daily intake (ADI) values. References doses (RfDs) or concentrations (RfCs) are estimates of a daily exposure to an agent that is assumed to be without adverse health impact on the human population. ADI values may be defined as the daily intake of chemical during an entire lifetime, which appears to be without appreciable risk on the basis of all known facts at that time. Reference doses and ADI values typically are calculated from NOAEL values by dividing by uncertainty (UF) and/or modifying factors (MF):

Tolerable daily intakes (TDI) can be used to describe intakes for chemicals that are not “acceptable” but are “tolerable” as they are below levels thought to cause adverse health effects. These are calculated in a manner similar to ADI. In principle, dividing by these factors allows for interspecies (animal-to-human) and intraspecies (human-to-human) variability with default values of 10 each. An additional UF can be used to account for experimental inadequacies—for example, to extrapolate from short-exposure-duration studies to a situation more relevant for chronic study or to account for inadequate numbers of animals or other experimental limitations. If only a LOAEL value is available, then an additional 10-fold factor commonly is used to arrive at a value more comparable to a NOAEL. Traditionally, a safety factor of 100 would be used for RfD calculations to extrapolate from a well-conducted animal bioassay (10-fold factor animal-to-human) and to account for human variability in response (10-fold factor human-to-human variability).

MF can be used to adjust the UF if data on mechanisms, pharmacokinetics, or relevance of the animal response to human risk justify such modification.

Recent efforts have focused on using data-derived factors to replace the 10-fold UF traditionally used in calculating RfDs and ADIs. Such efforts have included reviewing the human pharmacologic literature from published clinical trials and developing human variability databases for a large range of exposures and clinical conditions. Intra- and interspecies UF have two components: toxicokinetic and toxicodynamic aspects. Figure 4–4 shows these distinctions. This approach provides a structure for incorporating scientific information on specific aspects of the overall toxicologic process into the reference dose calculations; thus, relevant data can replace a portion of the overall “uncertainty” surrounding these extrapolations.

The NOAEL approach has been criticized on several points, including that (1) the NOAEL must, by definition, be one of the experimental doses tested; and (2) once this is identified, the rest of the dose–response curve is ignored. Because of these limitations, an alternative to the NOAEL approach, the benchmark dose (BMD) method, was proposed. In this approach, the dose–response is modeled and the lower confidence bound for a dose at a specified response level (benchmark response [BMR]) is calculated. The BMR is usually specified at 1, 5, or 10 percent. The BMDx (with x representing the percent BMR) is used as an alternative to the NOAEL value for reference dose calculations. Thus, the RfD would be:

The proposed values to be used for the UF and MF for BMDs can range from the same factors as for the NOAEL to lower values due to increased confidence in the response level and increased recognition of experimental variability owing to use of a lower confidence bound on dose.

Advantages of the BMD approach can include (1) the ability to take into account the full dose–response curve; (2) the inclusion of a measure of variability (confidence limit); and (3) the use of a consistent BMR level for RfD calculations across studies. Obviously, limitations in the animal bioassays in regard to minimal test doses for evaluation, shallow dose–responses, and use of study designs with widely spaced test doses will limit the utility of these assays for any type of quantitative assessments, whether NOAEL- or BMD-based approaches.

Nonthreshold Approaches

As Figure 4–3 shows, numerous dose–response curves can be proposed in the low-dose region of the dose–response curve if a threshold assumption is not made. Because the risk assessor generally needs to extrapolate beyond the region of the dose–response curve for which experimentally observed data are available, the choice of models to generate curves in this region has received lots of attention. For nonthreshold responses, methods for dose–response assessments have also utilized models for extrapolation to de minimus (10−4 to 10−6) risk levels at very low doses, far below the biologically observed response range and far below the effect levels evaluated for threshold responses. Two general types of dose–response models exist: statistical (or probability distribution models) and mechanistic models.

The distribution models are based on the assumption that each individual has a tolerance level for a test agent and that this response level is a variable following a specific probability distribution function. These responses can be modeled using a cumulative dose–response function. However, extrapolation of the experimental data from 50 percent response levels to a “safe,” “acceptable,” or “de minimus” level of exposure—for example, one in a million risk above background—illustrates the huge gap between scientific observations and highly protective risk limits (sometimes called virtually safe doses, or those corresponding to a 95 percent upper confidence limit on adverse response rates).

Models Derived from Mechanistic Assumptions

This modeling approach designs a mathematical equation to describe dose–response relationships that are consistent with postulated biological mechanisms of response. These models are based on the idea that a response (toxic effect) in a particular biological unit (animal or human) is the result of the random occurrence of one or more biological events (stochastic events).

A series of “hit models” exist for cancer modeling, where a hit is defined as a critical cellular event that must occur before a toxic effect is produced. The simplest mechanistic model is the one-hit (one-stage) linear model in which only one hit or critical cellular interaction is required for a cell to be altered. As theories of cancer have grown in complexity, multihit models have been developed that can describe hypothesized single-target multihit events, as well as multitarget, multihit events in carcinogenesis.

Toxicologic Enhancements of the Models

Three exemplary areas of research that have improved the models used in risk extrapolation are time to tumor information, physiologically based toxicokinetic modeling (described in Chapter 7), and biologically based dose–response (BBDR) modeling. The BBDR model aims to make the generalized mechanistic models discussed in the previous section more clearly reflect specific biological processes. Measured rates are incorporated into the mechanistic equations to replace default or computer-generated values.

Development of BBDR models for endpoints other than cancer is limited; however, several approaches have been explored in developmental toxicity utilizing cell cycle kinetics, enzyme activity, litter effects, and cytotoxicity as critical endpoints. Approaches have been proposed that link pregnancy-specific toxicokinetic models with temporally sensitive toxicodynamic models for developmental impacts. Unfortunately, the lack of specific, quantitative biological information for most toxicants and for most endpoints limits study and utilization of these models.

Variation in Susceptibility

Toxicology has been slow to recognize the marked variation among humans. Generally, assay results and toxicokinetic modeling utilize means and standard deviations to measure variation, or even standard errors of the mean, thereby ignoring variability in response due to differences in age, sex, health status, and genetics.

One key challenge for risk assessment will be interpretation and linking of observations from highly sensitive molecular and genome-based methods with the overall process of toxicity. Biomarkers of early effects, like frank clinical pathology, arise as a function of exposure, response, and time. Early, subtle, and possibly reversible effects can generally be distinguished from irreversible disease states.

The challenge for interpretation of early and highly sensitive response biomarkers is made clear in the analysis of data from gene expression arrays. Because our relatively routine ability to monitor gene responses has grown exponentially in the last decade, the need for toxicologists to interpret such observations for risk assessment and the overall process of toxicity has increased with equal or greater intensity.

Microarray analysis for risk assessment requires sophisticated analyses to arrive at a functional interpretation and linkage to a conventional toxicologic endpoint. Because of the vast number of measured responses with gene expression arrays, pattern analysis techniques are being used. The extensive databases across chemical classes, pathological conditions, and stages of disease progression that are essential for these analyses are being developed.

The primary objectives of exposure assessment are to determine source, type, magnitude, and duration of contact with the agent of interest. Obviously a key element of the risk assessment process, hazard does not occur in the absence of exposure. However, exposure data are frequently identified as the key area of uncertainty in overall risk determination. The primary goal of using exposure information in quantitative risk assessment is not only to determine the type and amount of total exposure, but also to find out specifically how much may be reaching target tissues. A key step in making an exposure assessment is determining what exposure pathways are relevant for the risk scenario under development. The subsequent steps entail quantitation of each pathway identified as a potentially relevant exposure and then summarizing these pathway-specific exposures for calculation of overall exposure.

Additional considerations for exposure assessments include how time and duration of exposures are evaluated in risk assessments. In general, estimates for cancer risk use averages over a lifetime. In a few cases, short-term exposure limits (STELs) are required and characterization of brief but high levels of exposure is required. In these cases exposures are not averaged over the lifetime and the effects of high, short-term doses are estimated. With developmental toxicity, a single exposure can be sufficient to produce an adverse developmental effect; thus, daily doses are used, rather than lifetime weighted averages.

The Toxicology Data Network (TOXNET) from the National Library of Medicine (http://sis.nlm.nih.gov/) provides access to databases on toxicology, hazardous chemicals, and related areas. These information sources vary in the included level of assessment, ranging from just listings of scientific references without comment to extensive peer-reviewed risk assessment information.

The World Health Organization (http://www.who.int/) provides chemical-specific information through the International Programme on Chemical Safety (http://www.who.int/pcs/IPCS/index.htm) criteria documents and health and safety documents. The International Agency for Research on Cancer (IARC) provides data on specific classes of carcinogens as well as individual agents. The National Institutes for Environmental Health Sciences (NIEHS) National Toxicology Program provides technical reports on the compounds tested as a part of this national program (http://ehis.niehs.nih.gov/roc/toc9.html).

Individuals respond very differently to information about hazardous situations and products, as do communities and whole societies. Understanding these behavioral responses is critical in stimulating constructive risk communication and evaluating potential risk management options. In a classic study, students, League of Women Voters members, active club members, and scientific experts were asked to rank 30 activities or agents in order of their annual contribution to deaths. Club members ranked pesticides, spray cans, and nuclear power as safer than did other lay persons. Students ranked contraceptives and food preservatives as riskier and mountain climbing as safer than did others. Experts ranked electric power, surgery, swimming, and X-rays as more risky and nuclear power and police work as less risky than did lay persons. There are also group differences in perceptions of risk from chemicals among toxicologists, correlated with their employment in industry, academia, or government.

Psychological factors such as dread, perceived uncontrollability, and involuntary exposure interact with factors that represent the extent to which a hazard is familiar, observable, and “essential” for daily living. Figure 4–5 presents a grid on the parameters controllable/uncontrollable and observable/not observable for a large number of risky activities; for each of the two paired main factors, highly correlated factors are described in the boxes.

Public demand for government regulations often focuses on involuntary exposures (especially in the food supply, drinking water, and air) and unfamiliar hazards, such as radioactive waste, electromagnetic fields, asbestos insulation, and genetically modified crops and foods. Many people respond very negatively when they perceive that information about hazards or even about new technologies without reported hazards has been withheld by the manufacturers (genetically modified foods) or by government agencies (HIV-contaminated blood transfusions in the 1980s; extent of hazardous chemical or radioactive wastes).

Most people regularly compare risks of alternative activities—on the job, in recreational pursuits, in interpersonal interactions, and in investments. Determining how best to conduct comparative risk analyses has proved difficult due to the great variety of health and environmental benefits, the gross uncertainties of dollar estimates of benefits and costs, and the different distributions of benefits and costs across the population.

The objectives of risk assessments vary with the issues, risk management needs, and statutory requirements. However, the National Research Council and Risk Commission frameworks are sufficiently flexible to address these various objectives, accommodate new knowledge, and provide guidance for priority setting in industry, environmental organizations, and government regulatory and public health agencies. Toxicology, epidemiology, exposure assessment, and clinical observations can be linked with biomarkers, cross-species investigations of mechanisms of effects, and systematic approaches to risk assessment, risk communication, and risk management. Advances in toxicology are certain to improve the quality of risk assessments for a broad array of health endpoints as scientific findings substitute data for assumptions and help to describe and model uncertainty more credibly.