Chapter 4: Risk Assessment

In the 1970s Congress established a basic plan for environmental laws that authorized regulatory actions to protect public health and the environment. These science-based actions provided the foundation for environmental and human health risk assessment. Toxicological research and toxicity testing constitute the scientific core of risk assessment, which is used for evaluating potential adverse health impacts from chemical exposures. However, such considerations for risk evaluation were not new, since for decades the American Conference of Governmental Industrial Hygienists (ACGIH) set threshold limit values for occupational exposures and the US Food and Drug Administration (FDA) established acceptable daily intakes (ADIs) for pesticide residues and food additives. In 1958, the US Congress instructed the FDA in the Delaney Clause to prohibit the addition of all substances found to cause cancer in animals or humans to the food supply. Pragmatically, this policy allowed food sources that had nondetectable levels of these additives to be declared “safe.” As advances in analytical chemistry revealed that “nondetects” were not equivalent to “not present,” regulatory agencies were forced to develop “tolerance levels” and “acceptable risk levels.” Risk assessment methodologies blossomed in the 1970s with the rising need to address these issues and statues and to provide a common framework for considering human and ecological health effects (Albert, 1994).

Risk assessment was defined as a specific activity in 1983 in the National Research Council (NRC) publication Risk Assessment in the Federal Government: Managing the Process (Red Book). This book detailed steps for hazard identification, dose–response assessment, exposure analysis, and characterization of risks (NRC, 1983). These 4 basic framework steps now form the foundation framework for risk assessment and risk management approaches for many contexts such as children’s health and ecological health and they provide a consistent context for risk assessment frameworks across agencies nationally and worldwide. Risk assessment frameworks have been used by the Health and Safety Executive Risk Assessment Policy Unit of the United Kingdom (Great Britain and Health and Safety Executive, 1999) and by the World Health Organization (NRC, 2009; WHO, 2010) and all of these build from these common components. Fig. 4-1 illustrates a framework with bidirectional arrows demonstrating an ideal situation where mechanistic research feeds directly into risk assessments and critical data uncertainty drives research. The common elements of the basic risk assessment framework are shown in red highlight. Initially, attention in risk assessment was focused on cancer risks; in subsequent years, noncancer end points were examined with similar methods and risk assessments were performed across life stages and for ecosystems. Continuing advances in toxicology, epidemiology, exposure assessment, biologically based modeling of adverse responses, and modeling of variability and uncertainty, as well as the rapidly growing area of genomics, have contributed to improvements in risk assessment. Risk assessment links with the public policy objectives from risk management often require extrapolations that go far beyond the observation of actual effects. Extrapolating data to address risks continues to generate controversy.

The National Academy of Sciences (NAS) 1994 report entitled Science and Judgment in Risk Assessment captured in its title the combination of qualitative and quantitative approaches essential for effective assessment of risks (NRC, 1994). The report discussed in detail the challenges and provided approaches for incorporating new scientific findings into the risk assessment process. It also highlighted ways to deal with uncertainty when insufficient scientific information was available. To address these challenges, the Presidential/Congressional Commission on Risk Assessment and Risk Management (Risk Commission, 1997) formulated a comprehensive framework that (1) placed each environmental problem or issue into public health and/or ecological context and (2) proactively called for engaging relevant stakeholders, often affected or potentially affected community groups, from the very beginning of the 6-stage process shown in Fig. 4-2. This report emphasized that particular exposures and potential health effects must be evaluated across sources and exposure pathways and in light of multiple end points, not just one chemical, in one environmental medium (air, water, soil, food, products), for one health effect at a time. The single-chemical, single-exposure approach had been the general approach up to this time. The importance of defining the risk problem based on this initial input is critical to risk assessment. Problem formulation and scoping has now been defined as Phase 1 for risk-based decision making and has been outlined by the US Environmental Protection Agency (EPA) (US EPA, 2011b) and reinforced by Science and Decisions, the Silver Book (NRC, 2009).

Risk assessment is the systematic scientific evaluation of potential adverse health effects resulting from human exposures to hazardous agents or situations (NRC, 1983, 1994; Omenn, 2000). Risk is defined as the probability of an adverse outcome based on the exposure and potency of the hazardous agent(s). Although historically there have been differences in how the term hazard has been used, international risk harmonization activities have allowed for current consensus on hazard as reference to intrinsic toxic properties, whereas exposure becomes an essential consideration along with hazard for risk. Risk assessment requires an integration of both qualitative and quantitative scientific information. For example, qualitative information about the overall evidence and nature of the end points and hazards is integrated with quantitative assessment of the exposures, host susceptibility factors, and the magnitude of the hazard. A description of the uncertainties and variability in the estimates is a significant part of risk characterization and an essential component of risk assessment. Analogous approaches are applied to ecological risks, as demonstrated by EPA’s ecological risk assessment guidelines (US EPA, 2011b). The objectives of risk assessment are outlined in Table 4-1.

The phrase “characterization of risk” reflects the combination of qualitative and quantitative analysis. Increasingly, the use of the term “hazard characterization” versus “hazard identification” has been seen in both the US EPA (2012a) and WHO (1999) documents. This shift recognizes the difficulties in evaluating only hazard information without dose–response information. Unfortunately, many users equate risk assessment with quantitative risk assessment, generating a number for an overly precise risk estimate, while ignoring crucial information about the uncertainties of the risk assessment, mode of action (MOA), and type of effect across species or context.

Risk management refers to the process by which policy actions are chosen to control hazards identified in the risk assessment stage of the framework (Fig. 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 Commission, 1997).

Risk communication is the challenging process of making risk assessment and risk management information comprehensible to the public, community groups, lawyers, local elected officials, judges, business people, labor, environmentalists, etc (Morgan, 1993; Sandman, 1993; Fischhoff et al., 1996; NRC, 1996). Such communication requires an understanding of level of risk response needed, how and which individuals and organizations should be informed, and what the audience hears versus what was intended as the risk communication message. Involving stakeholders in the initial problem formulation, as well as in the translation of the bottom line has become an essential part of risk management. Part of this analytical–deliberative process is the crucial, too-often neglected or delayed need for considering the fears, perceptions, priorities, and proposed remedies of these stakeholders (Risk Commission, 1997; Drew et al., 2003, 2006; Judd et al., 2005; Kramer et al., 2006; Cullen et al., 2008). Underestimating potential public and media attitudes toward local polluters, other responsible parties, and relevant government agencies may lead to what has been labeled “the outrage factor” (Sandman, 1993), greatly influencing the communication process and the choices for risk management. Sometimes the decision makers and stakeholders simply want to know the “bottom line”: whether a substance or a situation is “safe” or not. Others will be keenly interested in knowing how and why the risk estimates are uncertain and may be well prepared to challenge underlying assumptions about context and methodology. Stakeholders can also be part of a risk management solution (Judd et al., 2005; Drew et al., 2006). Perception of risk is further discussed at the end of this chapter.

Risk management decisions are reached under diverse statutes in the United States. Table 4-2 lists examples of major environmental statutes and the year of initial enactment. Some statutes specify reliance on risk alone, whereas others require a balancing of risks and benefits of the product or activity. 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. Similar approaches for risk assessment have been developed in many other countries and through such international organizations as the International Programme on Chemical Safety (IPCS) within the WHO. Currently within the IPCS, there are significant efforts toward a global harmonization of risk assessment methodology (WHO, 2006). The WHO Human Health Risk Assessment Toolkit: Chemical Hazards is an example (WHO, 2010). Their goals include use of a common risk assessment framework and development of confidence and acceptance of different risk assessment approaches through a common understanding and agreement on basic principles of testing and evaluation.

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 epidemiologic studies, animal bioassays, short-term in vitro and in vivo tests, and structure–activity 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 potential action in target organs and to host health impacts (NRC, 1989a,b, 1992a,b; Adami et al., 2011). Mechanistic investigations of the actions of specific chemicals can help us penetrate the “black box” approach of simply counting tumors, for example, in exposed animals in routine bioassays. Greater appreciation of the mechanisms and extent of individual variation in susceptibility to adverse health impacts among humans can improve protection of susceptible populations and better relate findings in animals to the characterization of risk for humans. Identifying individual behavioral and social risk factors can be critically important both to the characterization of risk and to the reduction of risk.

Assessing Toxicity of Chemicals—Introduction

In order to assess the toxicity of chemicals, information from four types of studies is used: structure–activity relationships (SAR), in vitro or short-term studies, in vivo animal bioassays, and information from human epidemiologic studies. In many cases, toxicity information for chemicals is limited. For example, in 1998, the EPA evaluated high production volume (HPV) chemicals (those produced in excess of one million lb per year) to ascertain the availability of chemical hazard data. Their study found that for 43% of these HPV chemicals, there were no publicly available studies for any of the basic toxicity end points (US EPA, 1998a). In response, EPA established a voluntary program called the HPV Challenge Program (US EPA, 2011c). Industry participants in this program committed to filling the data gaps for HPV chemicals. Since its inception, there have been over 2200 chemicals sponsored for additional testing. International efforts such as the Organization for Economic Cooperation and Development’s screening information data set (OECD/SIDS) program are also addressing data needs related to HPV, highlighted by the publication of their Manual for Investigation of HPV Chemicals (OECD, 2011).

Data requirements for specific chemicals can vary greatly by compound type and applicable regulatory statutes. Introduced in 2003 and approved in 2006, the European Union released a regulatory framework for Registration, Evaluation and Authorisation of Chemicals (REACH) (REACH, 2011). Under the REACH framework, since 2007, all stakeholders submit dossiers that include physical, chemical, and toxicological data as well as risk assessment studies for all chemicals in use in Europe. Such chemical safety assessments are submitted prior to approval in approaches similar to premanufacturing notices (PMNs) in the United States for the US EPA (US EPA, 2011d). Table 4-3 shows requirements and costs for one example class of agents, pesticides (Stevens, 1997; US EPA, 1998b; EPA, 2000), in the United States (40 CFR 158.340). It also illustrates current international efforts to align these testing guidelines by listing examples of the harmonized 870 test guidelines (US EPA, 2010a). The emphasis in REACH is on non–in vivo animal tests. Also, REACH has brought about new international labeling laws for hazard identification.

Assessing Toxicity of Chemicals—Approaches

Structure–Activity Relationships

Given the cost of two to four million dollars and the three to five years required for testing a single chemical in a lifetime rodent carcinogenicity bioassay, initial decisions on whether to continue development of a chemical, to submit PMN, or to require additional testing may be based largely on results from SARs 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. Historically, certain key molecular structures have provided regulators with some of the most readily available information on the basis of which to assess hazard potential. For example, 8 of the first 14 occupational carcinogens were regulated together by the Occupational Safety and Health Administration (OSHA) as belonging to the aromatic amine chemical class. The EPA Office of Toxic Substances relies on SARs to meet deadlines to respond to PMN for new chemical manufacture under the Toxic Substances Control Act (TSCA). Structural alerts such as N-nitroso or aromatic amine groups, amino azo dye structures, or phenanthrene nuclei are clues to prioritize chemicals for additional evaluation as potential carcinogens. SAR information for specific noncancer health end points can be challenging. The database of known developmental toxicants limits SARs to a few chemical classes, including chemicals with structures related to those of valproic acid, retinoic acid, phthalate esters, and glycol ethers (NRC, 2000). More recently, omic technologies have been used to supplement SAR relationship databases, as seen with the creation of the National Cancer Institute’s (NCI) gene expression database (http://dtp.nci.nih.gov/) and the US EPA Computational Toxicity Program Screening Database (ToxCast^TM) (US EPA ACToR, 2011).

SARs have been used for assessment of complex mixtures of structurally related compounds. A prominent application has been the assessment of risks associated with 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD), related chlorinated and brominated dibenzo-p-dioxins, dibenzofurans, and planar biphenyls, and chemicals generally present as mixtures in the environment. Toxicity equivalence factors (TEFs) are used to evaluate health risks associated with closely related chemicals. For the TCDD class, this is based on a common mechanism of aryl hydrocarbon (Ah) receptor induction (US EPA, 1994a). The estimated toxicity of environmental mixtures containing these chemicals is calculated as the sum of the product of the concentration of each chemical multiplied by its TEF value. The World Health Organization has organized efforts to reach international consensus on the TEFs used for polychlorinated biphenyls (PCBs), polychlorinated dibenzo-p-dioxins (PCDDs), and polychlorinated dibenzofurans (PCDFs) for both humans and wildlife, and has updated its values and published the supporting database (Van den Berg et al., 1998, 2006; Haws et al., 2006). Under the auspices of WHO, the dioxin-like PCB congeners have been assigned TEFs reflecting their toxicity relative to TCDD, which itself has been assigned a TEF of 1.0.

Computerized SAR methods have, in general, given disappointing results in the National Toxicology Program (NTP) rodent carcinogenicity prediction challenges (Ashby and Tennant, 1994; Omenn et al., 1995; Benigni and Zito, 2004). More successes are those of pharmaceutical companies using combinatorial chemistry and 3-dimensional (3D) molecular modeling approaches to design ligands (new drugs) that can sterically fit into the “receptors of interest.” However, for environmental pollutants where selective binding to specific receptors is rare, these applications of SAR have had limited success within risk assessment. A renewed interest in quantitative SAR (QSAR) approaches has resulted from the need to evaluate nano-engineered materials where the tremendous number of unique new products has highlighted the necessity of using QSAR approaches to handle the avalanche of novel untested materials (Maynard et al., 2006; Liu et al., 2011).

Efforts within REACH have also emphasized the potential for use of SAR as similar chemicals are collectively evaluated using a concept of “read-across.” Substances whose physicochemical, toxicological, and ecotoxicological properties are similar can be grouped as a “category” of substances when they have a common functional group, common precursor or breakdown product, or a common pattern of potency.

In Vitro and Short-Term Tests

The next level of biological information obtained within the hazard identification process includes assessment of the test chemical in in vitro or short-term tests, ranging from bacterial mutation assays performed entirely in vitro to more elaborate short-term tests, such as skin painting studies in mice or altered rat liver foci assays conducted in vivo. For example, EPA mutagenicity guidelines call for assessment of reverse mutations using the Ames Salmonella typhimurium assay; forward mutations using mammalian cells, mouse lymphoma L5178Y, Chinese hamster ovary, or Chinese hamster lung fibroblasts; and in vivo cytogenetics assessment (bone marrow metaphase analysis or micronucleus tests) (US EPA, 2005). Chap. 8 discusses uses of these assays for identifying chemical carcinogens and Chap. 9 describes in detail various assays of genetic and mutagenic end points. Other assays evaluate specific health end points such as developmental toxicity (Faustman, 1988; Whittaker and Faustman, 1994; Brown et al., 1995; Lewandowski et al., 2000; NRC, 2000; Spielmann et al., 2006), reproductive toxicity (Gray, 1988; Harris et al., 1992; Shelby et al., 1993; Yu et al., 2009), neurotoxicity (Atterwill et al., 1992; Costa, 2000), and immunotoxicity (ICCVAM, 1999, 2011a) (Chap. 12). Less information is available on the extrapolation of these test results for noncancer risk assessment than for the mutagenicity or carcinogenicity end points; however, mechanistic information obtained in these systems has been applied to risk assessment (Abbott et al., 1992; US EPA, 1994b; Leroux et al., 1996; NRC, 2000).

Overall, progress in developing and validating new in vitro assays has been slow and frustrating. The Interagency Coordinating Committee on the Validation of Alternative Methods (ICCVAM) of NTP reinvigorated the validation process in the United States as a result of Public Law 103-43 and coordinates cross-agency issues relating to development, validation, acceptance, and national/international harmonization of toxicological test methods for use in risk assessments. The committee has put forth recommendations for over 40 alternative safety testing methods using various short-term/in vitro assays, such as the cell-free corrosivity test, and for the mouse local lymph node assay for assessing chemical potential to elicit allergic reactions (ICCVAM, 1999, 2011a; NIEHS, 1999a,b). In 2006 the committee released a document that extensively reviews in vitro acute toxicity methods (National Toxicology Program Center for the Evaluation of Alternative Toxicological Methods (US) and Interagency Coordinating Committee on the Validation of Alternative Methods (US), 2006). The European Centre on the Validation of Alternative Methods (ECVAM) has been very active given the visibility of animal rights issues in the European Union, and this Center was originally formed to “support the development, validation, and acceptance of methods that could reduce, refine, or replace [3 Rs] the use of laboratory animals” (http://ihcp.jrc.ec.europa.eu/our_labs/eurl-ecvam). Early successes of ECVAM are described in Hartung et al. (2003). However, a recent report for the Center for Alternatives for Animal Testing highlights the slow progress on full replacement of animal testing, especially for repeated dose toxicity testing, carcinogenicity, or reproductive toxicity testing (Adler et al., 2011; Hartung et al., 2011). It does highlight numerous successes for evaluating sensitization and toxicokinetics (TK). International efforts to reduce the number of animals required for chemical safety testing have been established between ICCVAM, Korea, Japan, Canada, and ECVAM (http://ntp.niehs.nih.gov/pubhealth/evalatm/iccvam/international-partnerships/index.html).

The validation and application of short-term assays is particularly important to risk assessment because such assays can be designed to provide information about mechanisms of effects, and, moreover, they are fast and inexpensive compared with lifetime bioassays (McGregor et al., 1999). Validation of in vitro assays, like other kinds of tests, requires determination of their sensitivity (eg, ability to identify true carcinogens), specificity (eg, ability to recognize noncarcinogens as noncarcinogens), and predictive value for the toxic end point under evaluation. The societal costs of relying on such tests, with false positives (noncarcinogens classified as carcinogens) and false negatives (true carcinogens not detected), are the subject of a value-of-information model for testing in risk assessment and risk management (Lave and Omenn, 1986; Omenn and Lampen, 1988).

Efforts to improve our ability to utilize short-term tests for carcinogenicity prediction include increased attention to improving the mechanistic basis of short-term testing. Examples of this approach include the development and application of several knockout transgenic mouse models as shorter-term in vivo assays to identify carcinogens (Nebert and Duffy, 1997; Tennant et al., 1999). Specific assays for evaluating mechanisms such as the function of the estrogen receptor are an example where in vitro test methods are used to predict endocrine disruptor actions (ICCVAM, 2011b, 2012). The primary use of short-term tests in the United States continues to be for mechanistic evaluations with the hope of informing chemical-specific information or overall MOA. In that context, results from short-term assays have impacted risk assessments. For example, evidence of nonmutagenicity in both in vitro and in vivo short-term assays plays an essential role, allowing regulators to consider nonlinear cancer risk assessment paradigms for nongenotoxic carcinogens (US EPA, 1999a). Mechanistic information from short-term in vitro assays can also be used to extend the range of biological observations available for dose–response assessment. In addition, for developmental toxicity assessment, assay methods that acknowledge the highly conserved nature of developmental pathways across species have accelerated the use of a broader range of model organisms and assay approaches for noncancer risk assessments (NRC, 2000). Toxicity Testing in the 21st Century and Toxicity Pathway-Based Risk Assessment emphasized such approaches for using model organisms and tiered toxicity strategies for Risk Assessment (NRC, 2007a; NRC, 2010).

Animal Bioassays

Animal bioassays are a key component 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. Thus, “although this association cannot establish that all chemicals 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 chemicals and mixtures for which there is sufficient evidence of carcinogenicity in experimental animals as if they presented a carcinogenic risk to humans” (IARC, 2000)—a reflection of the “precautionary principle.” The US EPA cancer guidelines (US EPA, 2005) also assume 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 and biological pathways of most relevance to predicted or known human exposure pathways. Bioassays for reproductive and developmental toxicity and other noncancer end points have a similar rationale. The National Toxicology Program (NTP) serves as a resource for conducting, designing, and evaluating bioassays for cancer as well as noncancer evaluation. The NTP Office of Human Health Assessment and Translation serves as a resource to the public and regulatory agencies regarding the interpretation and assessment of adverse effects of chemicals (National Toxicology Program, 2011b). The WHO International Agency for Research on Cancer (IARC) has evaluated over 900 agents “of which more than 400 have been identified as carcinogenic, probably carcinogenic or possibly carcinogenic to humans” (WHO, 2011b).

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%–25% of the maximally tolerated dose [MTD]), and details of the required histopathology (number of organs to be examined, choice of interim sacrifice pathology, etc). The NTP Web site lists details on study designs and protocols (National Toxicology Program, 2011a). 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. Recently, NTP has added an in utero exposure period to the start of their cancer bioassays to more directly evaluate the significance of early life exposures for cancer incidence.

The cancer bioassay, originally designed for hazard identification, is frequently used to evaluate dose–response. The relatively limited number of evaluated doses and the use of high doses have caused issues for low-dose extrapolations and have limited the use of cancer bioassays as a “gold standard” for prediction of human carcinogenicity risk (McClain, 1994; Cohen, 1995; Risk Commission, 1997; Rodericks et al., 1997; Capen et al., 1999; Rice et al., 1999). Tumors may be increased only at the highest dose tested, which is usually at or near a dose that causes systemic toxicity (Ames and Gold, 1990). Second, even without toxicity, the high dose may trigger different events than do low-dose exposures and high doses can saturate important metabolism and elimination pathways.

Rats and mice give concordant positive or negative results in approximately 70% of bioassays, so it is unlikely that rodent/human concordance would be higher (Lave et al., 1988). Haseman and Lockhart (1993) concluded that most target sites in cancer bioassays showed a strong correlation (65%) between males and females—especially for forestomach, liver, and thyroid tumors—so they suggested, for efficiency, that bioassays could rely on a combination of male rats and female mice. Even when concordant, positive results are observed, there can still be large differences in potency, as observed in aflatoxin-induced tumors in rats and mice. In this example, an almost 100,000-fold difference in susceptibility to aflatoxin B₁ (AFB₁)–induced liver tumors is seen between the sensitive rat and trout species and the more resistant mouse strains. Genetic differences in the expression of cytochrome P450 and glutathione S-transferases explain most of these species differences and suggest that humans may be as sensitive to AFB₁-induced liver tumors as rats (Eaton and Gallagher, 1994; Eaton et al., 1995, 2001). These species differences have been supported by research results (Groopman and Kensler, 1999; Kensler et al., 2011) and have been extended within epidemiologic studies to demonstrate the interaction of hepatitis C infection with AFB₁ exposure to fully explain elevated human liver cancer risks.

Lifetime bioassays have been enhanced with the collection of additional mechanistic data and with the assessment of multiple noncancer end points. It is feasible and desirable to integrate such information together with data from mechanistically oriented short-term tests and biomarker and genetic studies in epidemiology (Perera and Weinstein, 2000). In the example of AFB₁-induced liver tumors, AFB₁–DNA adducts have proved to be an extremely useful biomarker. A highly linear relationship was observed between liver tumor incidence (in rats, mice, and trout) and AFB₁–DNA adduct formation over a dose range of 5 orders of magnitude (Eaton and Gallagher, 1994). Such approaches may allow for an extension of biologically observable phenomena to doses lower than those leading to frank tumor development and help to address the issues of extrapolation over multiple orders of magnitude to predict response at environmentally relevant doses.

Table 4-4 presents some mechanistic details about rodent tumor responses that are no longer thought to be predictive of cancer risk for humans. This table lists examples of both qualitative and quantitative considerations useful for determining relevance of rodent tumor responses for human risk evaluations. An example of qualitative considerations is the male rat kidney tumors observed following exposure to chemicals that bind to a_2u-globulin (eg, unleaded gasoline, 1,4-dichlorobenzene, d-limonene). The a_2u-globulin is a male-rat-specific low-molecular-weight protein not found in female rats, humans, or other species, including mice and monkeys (McClain, 1994; Neumann and Olin, 1995; Oberdorster, 1995; Omenn et al., 1995; Risk Commission, 1997; Rodericks et al., 1997).

Table 4-4 also illustrates quantitative considerations important for determining human relevance of animal bioassay information. For example, doses of compounds so high as to exceed solubility in the urinary tract outflow lead to tumors of the urinary bladder in male rats following crystal precipitation and local irritation leading to hyperplasia. Such precipitates are known to occur following saccharin or nitriloacetic acid exposure (Cohen et al., 2000). The decision to exclude saccharin from the NTP list of suspected human carcinogens reaffirms the nonrelevance of such high-dose responses for likely human exposure considerations (Neumann and Olin, 1995; National Toxicology Program, 2005). A gross overloading of the particle clearance mechanism of rat lungs via directly administered particles, as was seen in titanium dioxide (TDO) exposures, resulted in EPA’s delisting TDO as a reportable toxicant for the Clean Air Act Toxic Release Inventory (US EPA, 1988; Oberdorster, 1995).

Other rodent responses not likely to be predictive for humans include localized forestomach tumors after gavage. Ethyl acrylate, which produces such tumors, was delisted on the basis of extensive mechanistic studies (National Toxicology Program, 2005). In general, for risk assessment, it is desirable to use the same route of administration as the likely exposure pathway in humans to avoid such extrapolation issues. Despite the example of forestomach tumors, tumors in unusual sites—such as the pituitary gland, the eighth cranial nerve, or the Zymbal gland—should not be immediately dismissed as irrelevant, since organ-to-organ correlation is often lacking (NRC, 1994). The EPA cancer guidelines provide a good list of considerations for evaluating relevance in the sections on evaluating weight of evidence (US EPA, 2005).

In an attempt to improve the prediction of cancer risk to humans, transgenic mouse models have been developed as possible alternative to the standard 2-year cancer bioassay. Transgenic models use knockout or transgenic mice that incorporate or eliminate a gene that has been linked to human cancer. NTP evaluated some of these models and found the p53-deficient (p53+/− heterozygous) and Tg.AC (v-Ha-ras transgene) models to be particularly useful in identifying carcinogens and mechanisms of action (Bucher, 1998; Chhabra et al., 2003). The use of transgenic models has the power to improve the characterization of key cellular and MOA of toxicological responses (Mendoza et al., 2002; Gribble et al., 2005). However, these studies have been used primarily for mechanistic characterization than for hazard identification. Transgenic models have been shown to reduce cost and time as compared with the standard 2-year assay but they have also been shown to be somewhat limited in their sensitivity (Cohen, 2001). As stated in the current EPA cancer guidelines, transgenic models should not be used to replace the standard 2-year assay, but can be used in conjunction with other types of data to assist in the interpretation of additional toxicological and mechanistic evidence (US EPA, 2005). A series of genetically defined (fully sequenced genomes, 20 × coverage) mice referred to as the Collaborative Cross have been established to investigate genetic and environmental influences on toxicological response in mice (Chesler et al., 2008) and these strains should improve our understanding of mammalian genes that predispose to cancer (Koturbash et al., 2011).

Use of Epidemiologic Data in Risk Assessment

The most convincing lines of evidence for human risk are well-conducted epidemiologic studies in which a positive association between exposure and disease has been observed (NRC, 1983). Environmental and occupational epidemiologic studies are frequently opportunistic. Studies begin with known or presumed exposures, comparing exposed with nonexposed individuals, or with known cases, compared with persons lacking the particular diagnosis.

Table 4-5 shows examples of epidemiologic study designs and provides clues on types of outcomes and exposures evaluated. Although convincing, there are important limitations inherent in epidemiologic studies. Robust exposure estimates are often difficult to obtain as they are frequently done retrospectively (eg, through retrospective job history records). Also, since many important health effects have long latency before clinical manifestations appear, reconsideration of relevant populations can be challenging. Another challenge for interpretation is that there are often exposures to multiple chemicals, especially when a lifetime exposure period is considered. There is frequently 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 important to assess as they can have a significant impact on cancer development. Human epidemiologic studies can provide both very useful information for hazard assessment and quantitative information for data characterization. Good illustrations of epidemiologic studies and their interpretation for toxicological evaluation are available (Gill et al., 2011 and Regalado et al., 2006).

Three types of epidemiologic study designs—cross-sectional studies, cohort studies, and case–control studies—are detailed in Table 4-5. 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 can evaluate individuals selected on the basis of their exposure to a chemical under study. Thus, based on exposure status, these individuals are monitored for development of disease. 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 2 groups are compared to determine key consistent features in their exposure histories. All case–control studies are retrospective studies.

In risk assessment, 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) (Hill, 1965; Faustman et al., 1997; World Health Organization and International Programme on Chemical Safety, 1999; Adami et al., 2011). These same criteria have been used for evaluating MOAs where integrated considerations of both human and animal studies are done.

Epidemiologic study designs should also 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 end points under study, and a specified significance level (Healey, 1987; EGRET, 1994; Dean et al., 1995). Meta-analysis is used with epidemiologic studies to combine results from different studies using weighting of results to account for sample size across studies. The importance of human studies for risk assessment is shown in evaluations with arsenic and dioxin (US EPA, 2010b).

Advance from the human genome project have increase sophistication of molecular biomarkers and have improved the mechanistic bases for epidemiologic hypotheses. This has allowed epidemiologists to get within the “black box” of statistical associations and forward our understanding of biological plausibility and clinical relevance. “Molecular epidemiology,” the integration of molecular biology into traditional epidemiologic research, is an important focus of human studies where improved molecular biomarkers of exposure, effect, and susceptibility have allowed investigators to more effectively link molecular events in the causal disease pathway. Epidemiologists can now include the contribution of potential genetic factors with environmental risk factors for the determination of the etiology, distribution, and prevention of disease. Highlighting the potential power of genetic information to epidemiologic studies, the Human Genome Epidemiology (HuGE) Network was launched in 1998, providing a literature database of published, population-based epidemiologic studies of human genes (Khoury, 1999).

With the advance of genomics, the range of biomarkers has grown dramatically and includes identification of single-nucleotide polymorphisms (SNPs), genomic profiling, transcriptome analysis, and proteomic analysis (Simon and Wang, 2006). Implications of these improvements for risk assessment are tremendous, as they provide an improved biological basis for extrapolation across the diversity of human populations and allow for improved cross-species comparisons with rodent bioassay information because of evolutionarily conserved response pathways (NRC, 2000). In addition, genomics allows for “systems-based” understanding of disease and response, moving risk assessment away from a linear, single-event-based concept and improving the biological plausibility of epidemiologic associations (Toscano and Oehlke, 2005; NRC, 2010).

Integrating Qualitative Aspects of Risk Assessment

Qualitative assessment of hazard information should include a consideration of the consistency and concordance of findings, including a determination of the consistency of the toxicological findings across species and target organs, an evaluation of consistency across duplicate experimental conditions, and a determination of the adequacy of the experiments to consistently detect the adverse end points of interest.

Qualitative assessment of animal or human evidence is done by many agencies, including the EPA and IARC. Similar evidence classifications have been used for both the animal and human evidence categories by both agencies. These evidence classifications have included levels of “sufficient, limited, inadequate, and no evidence” (US EPA, 1994b, 2005) or “evidence suggesting lack of carcinogenicity” (IARC, 2000). For both agencies, these classifications are used for an overall weight-of-evidence approach for carcinogenicity classification.

Weight of evidence is an integrative step used by the EPA to “characterize the extent to which the available data support the hypothesis that an agent causes cancer in humans” (US EPA, 2005). It is the process of “weighing” all of the evidence to reach a conclusion about carcinogenicity. With this method, the likelihood of human carcinogenic effect is evaluated within an evaluation of the conditions under which such effects may be expressed. Weight of evidence can consider both the quality and quantity of data as well as any underlying assumptions. The evidence includes data from all of the hazard assessment and characterization studies such as SAR data, in vivo and/or in vitro studies, and epidemiologic data. Using this type of information and weight-of-evidence approach, the EPA includes hazard descriptors to define carcinogenic potential and to provide a measure of clarity and consistency in the characterization narrative: “carcinogenic to humans,” “likely to be carcinogenic to humans,” “suggestive evidence of carcinogenic potential,” “inadequate information to assess carcinogenic potential,” and “not likely to be carcinogenic to humans.” In this section, approaches for evaluating cancer end points are discussed for carcinogens. Similar weight-of-evidence approaches have been proposed for reproductive risk assessment (refer to sufficient and insufficient evidence categories in EPA’s guidelines for reproductive risk [US EPA, 1996a] and considerations by NTP).

The Institute for Evaluating Health Risks defined an “evaluation process” by which reproductive and developmental toxicity data can be consistently evaluated and integrated to ascertain their relevance for human health risk assessment (Moore et al., 1995; Faustman et al., 2011). This evaluation process has served as the basis for US EPA’s guidelines for developmental toxicity risk assessment (US EPA, 1991) and for NTP’s Office of Health Assessment and Translation (Formerly CERHR) (National Toxicology Program, 2011b). Application of such carefully deliberated approaches for assessing noncancer end points has helped avoid the tendency to list chemicals as yes or no (positive or negative) without human relevancy information.

Mode of Action

The EPA has emphasized in their revised cancer guidelines the importance of using “weight of evidence” to arrive at insights to possible “MOA” (US EPA, 2005). MOA information describes key events and processes leading to molecular and functional effects that would in general explain the overall process of cancer development. In many cases these could be plausible hypothesized MOAs for specific toxicity end points, but the detailed mechanistic nuances of the pathway might not yet be fully known. EPA is using such MOA information to suggest non-default approaches for cancer risk assessments and for evaluating toxicity of compounds with common MOAs in cumulative risk assessments (US EPA, 1996a, 1998a).

Within the EPA’s carcinogenic risk assessment guidelines, the MOA framework considers evidence from animal studies, relevance to humans, and life stage or population susceptibility (US EPA, 2005). Chemical-specific adjustment factors for interspecies differences and human variability have been proposed and build upon guidance developed by the WHO’s International Programme on Chemical Safety Harmonization Project (WHO, 2000). Critical to the MOA development is the use of “criteria of causality” considerations, which build upon Hill criteria used in epidemiology (Hill, 1965; Faustman et al., 1996; US EPA, 1999a; Klaunig et al., 2003), and consider dose–response relationships and temporal associations, as well as the biological plausibility, coherence, strength, consistency, and specificity of the postulated MOA.

Integrating Quantitative Aspects of Risk Assessment

Quantitative considerations in risk assessment include dose–response assessment, exposure assessment, variation in susceptibility, and characterization of uncertainty. For dose–response assessment, varying approaches have been proposed for threshold versus nonthreshold end points. Traditionally, in the United States, threshold approaches have been applied for assessment of noncancer end points, and nonthreshold approaches have been used for cancer end points. As we have learned more about nongenotoxic mechanisms of carcinogenicity, these processes have been evaluated using threshold approaches. Each approach and its inherent assumptions are discussed below, as are efforts to include more detailed mechanistic considerations to harmonize these approaches (Bogdanffy et al., 2001).

In general, human exposure data for prediction of human response to environmental chemicals are quite limited; thus, animal bioassay data have primarily served as the basis for most quantitative risk assessments and have required extrapolation for human health risk prediction. The risk assessor, however, is normally interested in low environmental exposures when considering human risk, exposures that are well below the experimentally observable range of responses in most animal assays. Thus, methods for extrapolating from high dose to low dose as well as extrapolating from animal risk to human risk are required and comprise a major emphasis of dose–response assessment.

The fundamental basis of the quantitative relationships between exposure to a chemical 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 most robust data sets with adverse effects occurring at the lowest levels of exposure from studies using the most relevant exposure routes. The “critical” adverse effect is defined as the significant adverse biological effect that occurs at the lowest exposure level (Barnes and Dourson, 1988). EPA has issued toxicity-specific guidelines that are useful in identifying such critical effects (for developmental toxicity [US EPA, 1991], reproductive toxicity [US EPA, 1996a], neurotoxicity [US EPA, 1998c], and cancer [US EPA, 1994b, 1996a, 1999a, 2005]). IPCS has compiled a document cited as the Principles for Modeling Dose–Responses for the Risk Assessment of Chemicals (International Programme on Chemical Safety, 2006). It outlines key concepts and considerations for dose–response evaluations within the context of a risk assessment.

Threshold Approaches

Approaches for characterizing threshold dose–response relationships include identification of “no observed adverse effect level” (NOAEL) or “lowest observed adverse effect levels” (LOAELs). On the dose–response curve illustrated in Fig. 4-3, the doses tested in the bioassay are given as F, G, H, and I. The statistical significance of points G, H, and I is indicated using an asterisk (*). The NOAEL (F) is identified as the highest nonstatistically significant dose tested; in this example the NOAEL occurs at approximately 2 mg/kg body weight. Point G is the LOAEL (∼2.3 mg/kg body weight), as it is the lowest dose tested with a statistically significant effect. Lines A to D represent possible extrapolations below the point of departure (POD), which is represented on this figure as a square (▪) and is labeled as point E. The POD is used to specify the estimated dose near the lower end of the observed dose range, below which extrapolation to lower exposures is necessary (US EPA, 2005). In Fig. 4-3, the POD occurs at 10% effective dose or ED₁₀. The type of extrapolation below the POD is depending on the type of data available. The importance of choosing an appropriate POD and extrapolation type is discussed further in the following sections of this chapter. From these modeled approaches, model D shows a model where the threshold (T) represents the dose below which no additional increase in response is observed.

In general, most animal bioassays are constructed with sufficient numbers of test animals to detect biological responses at the 10% response range; however, this is also dependent on end point and the background rate of the end point in control animals. The risk assessor should always understand the biological significance of the responses being evaluated in order to put statistical observations in context. Significance thus usually refers to both biological and statistical criteria (Faustman et al., 1994) 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 several reports have shown that the response of NOAELs for continuous end points averages 5% risk, and NOAELs based on quantal end points can be associated with risk of greater than 10% (Allen et al., 1994a; Faustman et al., 1994).

As described in Chap. 2, approaches for characterizing dose–response relationships include identification of effect levels such as LD₅₀ (dose producing 50% lethality), LC₅₀ (concentration producing 50% lethality), ED₁₀ (dose producing 10% response), as well as NOAELs. NOAELs have traditionally served as the basis for risk assessment calculations, such as reference doses (RfDs) or ADI values. RfDs or concentrations (RfCs) are estimates of daily exposure (oral or inhalation, respectively) to an chemical that is assumed to be without adverse health impact in humans. The ADIs are used by WHO for pesticides and food additives to define “the daily intake of chemical, which during an entire lifetime appears to be without appreciable risk on the basis of all known facts at that time” (WHO, 1962; Dourson et al., 1985). RfDs (first introduced in Chap. 2) and ADI values typically are calculated from NOAEL values by dividing by uncertainty factor (UF) and/or modifying factor (MF) (Dourson and Stara, 1983; Dourson and DeRosa, 1991; US EPA, 1991):

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. Historically, dividing by the UFs allows for interspecies (animal-to-human) and intraspecies (human-to-human) variability with default values of 10 each. An additional UF is 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 is commonly used to arrive at a value more comparable to a NOAEL. For developmental toxicity end points, it has been demonstrated that the application of the 10-fold factor for LOAEL-to-NOAEL conversion is too large (Allen et al., 1994a). Traditionally, a safety factor of 100 would be used for RfD calculations to extrapolate from a well-conducted animal bioassay (10-fold factor for animal-to-human variability) and to account for human variability in response (10-fold factor for human-to-human variability).

MFs are values that can be used to adjust the UFs if data on mechanisms, pharmacokinetics, or relevance of the animal response to human risk are available. For example, if there is kinetic information suggesting there is very similar metabolism for a particular compound in rats and humans, producing the same active target metabolite, then—rather than using a 10-fold UF to divide the NOAEL from the animal toxicity study to obtain a human relevant RfD—a factor of 3 for that UF might be used. Of particular interest is the addition of an extra 10-fold factor through the Food Quality Protection Act (FQPA) to ensure protection of infants and children (US EPA, 1996b). Under this law an additional UF is added to ensure protection of children’s health and it is currently being used for determining allowable pesticide chemical residues. This factor is designed to take into account potential prenatal and postnatal toxicity and to overcome the incompleteness of toxicity and exposure data (FQPA, PL 104-170). Illustrative discussions on how such a legislatively mandated UF might be applied are available for chlorpyrifos (Schardein and Scialli, 1999; US EPA, 2000a). In this case, this uncertainty factor was reduced due to the availability of specific animal experiments assessing developmental toxicity at sensitive life stages (Zhao et al., 2006; US EPA, 2011e).

To reduce uncertainty in calculating RfDs and ADIs, there has been a transition from the use of traditional 10-fold UFs to the use of data-derived and chemical-specific adjustment factors. Such efforts have included reviewing the human pharmacologic literature from published clinical trials (Silverman et al., 1999) and developing human variability databases for a large range of exposures and clinical conditions (Renwick, 1991, 1999; Johnson et al., 1997). Toward this goal, Renwick has separated the intraspecies and interspecies UFs into 2 components: toxicokinetics (TK) and toxicodynamic (TD) aspects (Renwick, 1991, 1999; Johnson et al., 1997). Fig. 4-4 shows these distinctions. A key advantage of this approach is that it provides a structure for incorporating scientific information on specific aspects of the overall toxicological process into the RfD calculations; thus, relevant data can replace a portion of the overall “uncertainty” surrounding these extrapolations. Current WHO guidance uses a 4.0- and 2.5-fold factor for the TK and TD interspecies components, respectively (WHO, 2005), and interindividual TK and TD factors of 3.16 (Renwick and Lazarus, 1998). The recent NAS publications (Silver Book) have emphasized the need to consider population variability and context in risk assessment. In examples described in this NRC report, high background incidence of cancers or other adverse health outcomes in the population should be considered before setting “acceptable” levels of additional risks (NRC, 2009).

NOAEL values have also been utilized for risk assessment by evaluating a “margin of exposure” (MOE), where the ratio of the NOAEL determined in animals and expressed as mg/kg per day is compared with the level to which a human may be exposed. For example, consider the case where human exposures to a specific chemical are calculated to be solely via drinking water, and the total daily intake of the compound is 0.04 mg/kg per day. If the NOAEL for neurotoxicity is 100 mg/kg per day, then the MOE would be 2500 for the oral exposure route for neurotoxicity. Such a large value is reassuring to public health officials. Low values of MOE indicate that the human levels of exposure are close to levels for the NOAEL in animals. There is usually no factor included in this calculation for differences in human or animal susceptibility or animal-to-human extrapolation; thus, MOE values of less than 100 have been used by regulatory agencies as flags for requiring further evaluation.

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 (Crump, 1980). 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%. Fig. 4-5 shows the BMD using a 10% BMR (BMD₁₀) and a 95% lower confidence bound on dose (BMDL₁₀). The BMD_x (with x representing the percent BMR) is used as an alternative to the NOAEL value for RfD calculations. Thus, the RfD would be:

EPA has developed software for the application of BMD methods and a technical guidance document to provide guidelines for application of BMDs for both cancer and noncancer end points (EPA, 2000). It recently released an updated version (2.2) of its BMD software (http://www.epa.gov/ncea/bmds/index.html). The 2009 WHO guidelines for dose–response modeling for risk assessment also discuss BMD approaches (International Programme on Chemical Safety, 2006). Both the EPA and WHO guidelines distinguish NOAEL versus BMD-based approaches. Harmonization of approaches available for assessments is discussed in the WHO documents.

The BMD approach has been applied to study noncancer end points, including developmental (Allen et al., 1994a) and reproductive toxicity (Auton, 1994). The most extensive studies with developmental toxicity have shown that BMD₀₅ values were similar to a statistically derived NOAEL for a wide range of developmental toxicity end points and that results from using generalized dose–response models were similar to biologically based dose–response (BBDR) models designed specifically to represent unique features of developmental toxicity testing (Faustman et al., 1999).

Advantages of the BMD approach include (1) the ability to take into account the 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. 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 Fig. 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 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.

EPA cancer guidelines define the dose–response methods as requiring two steps: (1) defining the “POD” or the lowest dose associated with adverse effects within the range of the experimental data and (2) the extrapolation from the POD to low environmentally relevant exposure levels based on experimental data. The extrapolation can be done with a linear model or a nonlinear model with this choice dependent on the amount and type of experimental data available. Risk estimates using the linear model (biological response increases proportionally with level of exposure) are generally higher than those using nonlinear models. For example, in 2003, the EPA released a Dioxin Reassessment and used the BMD dose–response method. The reassessment used a POD based on a 1% response with a linear extrapolation model based on the position that the scientific data were inadequate to rule out the EPA’s standard default linear assumption. The EPA Dioxin Reassessment was evaluated in a National Academies Report in 2006 (National Academy of Science, 2006) and this review discussed remaining extrapolation issues. In EPA’s recent draft of dioxin risk assessment (US EPA, 2010b) revised TCDD cancer risk assessment is presented using the upper bound on the regression slope for estimating cancer mortality risk from the epidemiology evaluation by Cheng et al. (2006). The top 5% of exposure estimates from this epidemiologic study were excluded as the dose-related changes in cancer mortality plateaued at high doses. EPA lagged exposures by 15 years and adjusted cancer risk based on cumulative TCDD fat concentrations. Their analysis included an estimate of TCDD cancer risks above the TCDD background levels found in the NIOSH worker cohort analyzed by Chang et al. Below the POD, the EPA assumed a default linear slope value. (For more information, please go to US EPA as this draft document is undergoing final comments and discussion.) This document represents one of the most recent interpretations of cancer risk assessment by EPA for data-rich compounds with multiple cancer risk estimates for animal and human studies as well as how extensive mechanistic and kinetic information is considered. This draft report provides insight into recent interpretations of US EPA cancer guidelines (US EPA, 2005).

Statistical or Probability Distribution Models

Two general types of dose–response models exist for extrapolation: statistical (or probability distribution models) and mechanistic models (Krewski and Van Ryzin, 1981). The distribution models are based on the assumption that each individual has a tolerance level for a test chemical 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. Chap. 2 discusses the common normal distribution pattern (see Fig. 2-3). A log probit model estimates the probability of response at a specified dose (d); thus, P(d) = φ [a + β log d], where φ is the cumulative function for a standard normal distribution of the log tolerances with standard deviations σ and mean μ, a equals μ/σ, and β equals the slope of the probit line (−1/σ). The probit curve at low doses usually assumes an S shape. Chap. 2 discusses determination of the LD₅₀ value from such a curve. However, extrapolation of the experimental data from 50% 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 [VSDs] or those corresponding to a 95% upper confidence limit on adverse response rates).

The log-logistic model was derived from chemical kinetic theory. The probability of response at dose d is defined as P(d) = [1 exp(a + β log d)]⁻¹. Like the probit model, this model defines sigmoidal curves that are symmetrical around the 50% response level; however, the log-logistic curves approach the 0% and 100% response levels with a shallow curve shape. The logit and probit curves are indistinguishable in fitting the data in the region of the response curve where experimentally derived data are present (Brown, 1984; Hartung, 1987; Allen et al., 1994b).

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, human, pup, etc) is the result of the random occurrence of one or more biological events (stochastic events).

Radiation research has spawned a series of such “hit models” 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. For example, based on somatic mutation theory, a single mutational change would be sufficient for a cell to become cancerous through a transformational event and dose-independent clonal expansion. The probability statement for these models is P(d) = 1 − exp^{(− λd)}, where λ d equals the number of hits occurring during a time period. Using this approach, a single molecule of a genotoxic carcinogen would have a minute but finite chance of causing a mutational event.

As theories of cancer have grown in complexity, so too have these hit-based mechanistic models. Multihit models have been developed that can describe hypothesized single-target multihit events, as well as multi-target, multi-hit events in carcinogenesis. The probability statements for these models are P(d) = ∫^λd x^k−1 exp(−x)/Γ(k)dx, where Γ(k) denotes the gamma function with k being the critical number of hits for the adverse response. The Weibull model has a dose–response function with characteristics similar to those of the multihit models, where the response equation is P(d) = 1 − exp[−λd^k]. Here again, k is the critical number of hits for the toxic cellular response.

Armitage and Doll (1957) developed a multistage model for carcinogenesis that was based on these equations and on the hypothesis that a series of ordered stages was required before a cell could undergo mutation, initiation, transformation, and progression to form a tumor. This relationship was generalized by Crump (1984) by maximizing the likelihood function over polynomials, so that the probability statement is:

If the true value of λ₁ is replaced with λ^*₁ (the upper confidence limit of λ₁), then a linearized-multistage model can be derived where the expression is dominated by λ^*₁ at low doses. The slope on this confidence interval, q^*₁, has been used by EPA for quantitative cancer assessment. To obtain an upper 95% confidence interval on risk, the q^*₁ value (risk/Δ dose in mg/kg per day) is multiplied by the amount of exposure (mg/kg per day). Thus, the upper bound estimate on risk (R) is calculated as:

R = q^*₁ [risk (mg/kg per day)⁻¹] × exposure (mg/kg per day).

This relationship has been used to calculate a “VSD,” which represents the lower 95% confidence limit on a dose that gives an “acceptable level” of risk (eg, upper confidence limit for 10⁻⁶ excess risk). The Integrated Risk Information System (IRIS) (US EPA, 2011e) developed by EPA gives values for many environmental carcinogens (US EPA, 2000b). Because both the q^*₁ and VSD values are calculated using 95% confidence intervals, the values are believed to represent conservative, protective estimates.

The EPA has utilized the LMS model to calculate “unit risk estimates” in which the upper confidence limit on increased individual lifetime risk of cancer for a 70-kg human breathing 1 μg/m³ of contaminated air or drinking 2 L per day of water containing 1 ppm (1 mg/L) is estimated over a 70-year life span. The example given in Fig. 4-6 shows the calculation of incremental lifetime cancer risk (ILCR) of skin cancer using soil exposure and q^* values for inorganic arsenic. If a POD-based approach is used for low-dose extrapolation, the model would be used to extrapolate from the POD or upper confidence limit and these slopes would be considered for the q^* values.

Toxicological 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 TK modeling, and BBDR modeling (Albert, 1994). Chap. 7 discusses in detail improvements in our estimation of exposure and offers approaches on how to model “target internal effective dose” in risk assessment rather than just using single-value “external exposure doses.” In this chapter we discuss the BBDR modeling.

BBDR modeling 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. For example, the Moolgavkar–Venson–Knudson (MVK) model is based on a two-stage model for carcinogenesis, where two mutations are required for carcinogenesis and birth and death rates of cells are modeled through clonal expansion and tumor formation. This model has been applied effectively to human epidemiologic data on retinoblastoma. In animal studies, kidney and liver tumors in the 2-acetylaminofluorene (2-AAF) “mega mouse” study, rat lung tumors following radiation exposure, rat liver tumors following N-nitrosomorpholine exposure, respiratory tract tumors following benzo[a]pyrene exposure, and mouse liver tumors following chlordane exposure have been modeled (Cohen and Ellwein, 1990; Moolgavkar and Luebeck, 1990).

There have been concerted research efforts to improve our mechanistic understanding of response at lower and lower exposures. Two excellent examples of the use of biomarkers to supplement our exploration of these dose–response relationships are seen in Bailey et al. (2009) and Vlaanderen et al. (2011). Hormetic responses are U-shaped dose–response relationships where small doses of a chemical can be beneficial but high doses are toxic. These are discussed in Chap. 2. These responses pose challenges to risk assessors. In general, science-based mechanistic conditions of low-dose response are always critical. See approaches used by the Institute of Medicine for vitamins and essential elements as these compounds consistently exhibit hormesis (Institute of Medicine (US), 2008).

Development of BBDR models for end points other than cancer is limited; however, several approaches have been explored in developmental toxicity, utilizing MOA information on cell cycle kinetics, enzyme activity, and cytotoxicity as critical end points (Faustman et al., 1989, 2005a; Shuey et al., 1994; Leroux et al., 1996; Gohlke et al., 2002, 2004, 2005, 2007; Daston et al., 2004). Approaches have been proposed that link pregnancy-specific TK models with temporally sensitive TD models for developmental impacts (Faustman et al., 1999; Faustman and Omenn, 2006). Unfortunately, there is a lack of specific, quantitative biological information on both kinetics and dynamics for most toxicants and end points (NRC, 2000; Faustman et al., 2005a).

The primary objectives of exposure assessment are to determine source, type, magnitude, and duration of contact with the chemical(s) of interest. This is a critical element of the risk assessment process, as hazard does not occur in the absence of exposure. However, it is also frequently identified as the key area of uncertainty in the overall risk determination. Here, the primary focus is on uses of exposure information in quantitative risk assessment.

The primary goal of exposure calculations is not only to determine the type and amount of total exposure but also to find out specifically who may be exposed and how large a dose 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. Fig. 4-7 shows an example exposure diagram used to illustrate possible exposure pathways from a hazardous chemical release. The subsequent steps entail quantitation of each pathway identified as a relevant exposure and then summarizing these pathway-specific exposures for calculation of overall exposure. Such calculations can include an estimation of total exposures for a specified population as well as calculation of exposure for highly exposed individuals. The EPA has published numerous documents that provide detailed definitions and guidelines for determining such exposures (US EPA, 1992, 2008, 2011a), with the recognition that special considerations need to be adopted when assessing childhood exposures (US EPA, 2008).

Conceptually, calculations are designed to represent “a plausible estimate” of exposure of individuals in the upper 90-95th percentile of the exposure distribution. Upper bound estimations would be “bounding calculations” designed to represent exposures at levels that exceed the exposures experienced by all individuals in the exposure distribution and are calculated by assuming upper limits for all exposure variables. A calculation for individuals exposed at concentrations near the middle of the exposure distribution is a central estimate. Fig. 4-7 gives example risk calculations using two types of exposure estimation procedures (US EPA, 1989, 1992, 2011a). Part A shows an example point estimation method for the calculation of arsenic (As) exposure via a soil ingestion route. In this hypothetical scenario, As exposure is calculated using point estimates for the upper 95th percent confidence limit of the arithmetic mean for each value in the lifetime average daily dose (LADD) equation. The LADD is calculated as follows:

In this scenario, the average time (AT) is lifetime (25,550 days) and the contact or ingestion rate (CR) is equal to 100 mg soil per day. The exposure frequency and duration (EDF) is equal to the exposure frequency (daily exposure = 365 days per year) times the exposure duration (30 years). To obtain the ILCR, the LADD must be multiplied by the bioavailability to obtain an absorbed dose and q^* where the upper 95% CLM slope (q^*) is 1.5 × change in risk per mg/kg body weight per day in this example.

Many exposures are now estimated using probability distributions for exposures rather than single-point estimates for the factors within the LADD equation (Finley et al., 1994; Cullen and Frey, 1999). Such approaches can provide a reality check and can be useful for generating more realistic exposure profiles reflective of population exposures. Good sources for US exposures can be obtained from the EPA Exposure Factors Handbook (US EPA, 2011a). Fig. 4-7B shows how this is done using an example arsenic risk scenario with soil As concentration, ingestion rate, exposure duration, frequency, body weight, and bioavailability modeled as distributed variables. Using Monte Carlo simulation techniques, an overall ILCR distribution can be generated and a 95th percentile for population risk obtained.

The FQPA of 1996 has highlighted the need for considerations of potential special risks for children and the need to evaluate child-specific exposures as well as cumulative exposures and risks (US EPA, 1996b). In 2006, the EPA released a supplemental document with child-specific exposure factors to address the special considerations necessary when evaluating childhood exposures (US EPA, 2008). Table 4-6 shows example exposure factors for drinking water intake for children and adults. The EPA Child-Specific Exposure Factors Handbook, which is online, provides useful information about such exposure distributions. The FQPA also identified the need to evaluate total exposures by determining aggregate exposure measures for all exposures to a single substance. Cross-media exposure analyses, such as those conducted for lead and mercury, are good examples of the value of looking at such total exposures in evaluating human risks, and complex models are available for some compounds (see EPA’s integrated exposure uptake biokinetic [IEUBK] model for lead in children; US EPA, 2010c). Cumulative exposures and cumulative risk refer to the total exposure and risks from a group of compounds. In FQPA chemicals with similar modes of toxicity are evaluated using cumulative exposure and risk estimates. For example, EPA is identifying and categorizing pesticides that act by a common MOA, and cumulative exposures to classes of organophosphates with similar MOAs (cholinesterase inhibition) have been used as examples of classes of pesticides for which cumulative exposure and cumulative risk estimates are needed (US EPA, 1996c; ILSI, 1999; Ryan, 2010). WHO has recently published a framework for combined exposures (Meek et al., 2011).

Additional considerations for exposure assessments include how time and duration of exposure are evaluated in risk assessments. In general, estimates for cancer risk use average exposure over a lifetime (see LADD example above). In a few cases, short-term exposure limits (STELs) are required (eg, ethylene oxide) and characterization of brief but high levels of exposure is significant. 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 if exposures occur during a window of developmental susceptibility; thus, daily doses are used, rather than lifetime weighted averages (US EPA, 1991; Weller et al., 1999).

Risk characterization is a summary of the risk assessment components and serves to outline the key findings and inform the risk manager in public health decisions. It is an analysis and integration of the conclusions from the hazard assessment, the dose–response, and the exposure assessment. EPA outlines the science policy and elements for completing a risk characterization in their Risk Characterization Handbook (US EPA, 2000c). As shown in Fig. 4-1, risk characterization considers the nature, estimated incidence, and reversibility of adverse effects in a given population; how robust the evidence is; how certain the evaluation is; if susceptible populations are characterized; and if there is a relevant mode of action.

For many years there has been an information sharing process aimed at harmonization of chemical testing regimes and clinical trials methodologies, so that data might be accepted in multiple countries. Efforts by WHO and IPCS have included the Harmonization of Approaches to the Assessment of Risk from Exposure to Chemicals Project (WHO, 2011a). This project has the goal of globally harmonizing but not standardizing the approaches to risk assessment by increasing understanding and developing basic principles and guidance on specific chemical risk assessment issues. The WHO has worked to harmonize dose–response methodologies (DRM) and released a report providing descriptive guidance for risk assessors in using DRM in hazard characterization (International Programme on Chemical Safety, 2006). Examples of other reports include those on terminology (International Programme on Chemical Safety, 2004) and exposure assessment (World Health Organization, 2006) and Uncertainty and Data Quality in Exposure Assessment (International Programme on Chemical Safety, 2008).

Uncertainty analysis is an essential component in our final risk characterization and includes factors such as variability and lack of knowledge (NRC, 1994). Variability refers to true differences such as is reflected in temporal, spatial, or interindividual differences. It usually cannot be reduced with further study. Lack of knowledge can also be discussed as a source of uncertainty and these factors may be reduced with further study. For example, if only a few environmental samples are taken from a contaminated site for analysis, then uncertainty can be reduced by taking further samples. Formal methods for dealing with uncertainty analysis are described in several references including IPCS (2008) and Bogen et al. (2009). Using probabilistic-based approaches (PRA) for exposure assessment and assuming distributional versus dichotomous data for exposure factors is one way of addressing uncertainty that is commonly applied.

Variation in Susceptibility

Risk assessment methodologies incorporating human variability have been slow to develop. Generally, assay results utilize means and standard deviations to measure variation, or even standard errors of the mean. This ignores variability in response due to specific differences in age, sex, health status, and genetics. Default factors of 10× are overutilized to describe cross- and between-species differences and TK and dynamic data are rare. Nevertheless, EPA and OSHA are expected under the Clean Air Act and the Occupational Safety and Health Act to promulgate standards that protect the most susceptible subgroups or individuals in the population (Omenn et al., 1990; Cullen and Frey, 1999; Faustman and Omenn, 2006; Kramer et al., 2006; Cullen et al., 2008). Sensitive or susceptible populations have been defined and interpreted for the Clean Air Act and in this context “particularly sensitive citizens, such as bronchial asthmatics and emphysematics”, are to be protected by the standards. This illustrates not only the importance but also the challenge to risk assessors to ensure protection for human populations.

Ecogenetics has been defined as the study of critical genetic determinants that define susceptibility to environmentally influenced adverse health effects (Costa and Eaton, 2006). Ecogenetic variation can affect biotransformation systems that activate and detoxify chemicals or alter the response in target tissues. With the completion of the Human Genome Project in April 2003, the identification of human polymorphisms has greatly expanded our potential for understanding how genetic variability can impact biological response and susceptibility. There have been numerous activities initiated with the goal of understanding the linkage between genes and the environment. The following databases and initiatives are designed to identify polymorphisms, including single-nucleotide polymorphisms (SNP), insertions, deletions, amplifications, and alternative splicing. These include the International HapMap Project (HapMap, 2009), the DNA Polymorphism Discovery Resource (National Human Genome Research Institute, 2010), the National Center for Biotechnology Information (NCBI) SNP databases (NCBI, 2011a,b), and the NIH GAIN program (National Human Genome Research Institute, 2010). Using the information from such databases, researchers hope to identify genetic similarities and differences in human beings with the goal of identifying genes that affect health, disease, and individual responses to medications and environmental factors (Omen, 2012). Recent observations now demonstrate that not only genetics is significant, but also epigenetics (Sartor et al., 2012). SNP arrays are now available that can identify and characterize approximately 1 × 10⁶ SNPs in the human population. Arrays also exist that evaluate SNPs in pathways significant for drug metabolism enzymes and transporters. These are in FDA-approved pathways and have been used for evaluation of appropriate drug therapies. However, they are just starting to be used for environmental risk assessment (Guerrette et al., 2011). Next-generation sequencing, where coverage of the genome can be extended to 40×, has revealed rare copy number variants that can be linked to disease and provide another level of understanding concerning human variation (Sudmant et al., 2010). It should be noted that this genetic information on individuals can raise ethical, social, and legal concerns regarding protections of individuals and their rights (Hsieh, 2004). Recent legislation that protects individuals from discrimination on the basis of genetic variation helps to ensure protection and not exclusion for all workers (Human Genome Program, 2008; United States Act, 2008).

One of the key challenges for toxicologists doing risk assessments will be the interpretation and linking of observations from highly sensitive molecular and genome-based methods with the overall process of toxicity (Eisen et al., 1998; Limbird and Taylor, 1998; Andersen and Barton, 1999; NRC, 2000). The basic need for linkage of observations was highlighted in early biomarker work. NRC reports on biomarkers (NRC, 1989a,b, 1992a,b) drew distinctions for biomarkers of effect, exposure, and susceptibility across a continuum of exposure, effect, and disease/toxicity. Biomarkers of early effects, such as 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. Chemical-specific biomarkers have the potential to provide critical information, but there is an inherent complexity in pulling the information together. If biomarkers are considered a reflection of exposure or disease state, then considerations have to be made regarding the interactions between genes and the environment and the difference between a population-based assessment and individual assessment (Groopman and Kensler, 1999).

Nowhere is the challenge for interpretation of early and highly sensitive response biomarkers clearer than in the complicated data from gene expression arrays (toxicogenomics). Our continued ability to monitor changes in response in thousands of genes has proven challenging for toxicologists. Emphasis on toxicity-based risk assessment methods and analysis has been strongly encouraged by several significant NAS reports (NRC, 2000, 2007a,b; NRC, 2010) as well as development and application of pathway-based methods (Yu et al., 2006). Toxicogenomics projects have confirmed the repeatability and cross-platform concordance of microarray data (Bammler et al., 2005; Guo et al., 2006; Nature Biotechnology Editorial, 2006; Patterson et al., 2006) and standards for data interpretation (minimum information about a microarray experiment [MIAME]) accepted (Brazma et al., 2001).

Microarray analysis for risk assessment requires sophisticated analyses beyond basic cluster analysis (Eisen et al., 1998) to arrive at a functional interpretation and linkage to conventional toxicological (apical) end points. Because of the vast number of measured responses with gene expression arrays, pattern analysis techniques are being used through pathway mapping platforms such as MAPPFinder and GenMAPP (Moggs et al., 2004; Currie et al., 2005). The Gene Ontology Consortium has developed a controlled vocabulary (ontology) for sharing biological information across species. Database users can then annotate their gene products with gene ontology (GO) terms, establishing a consistent description and definition across databases and studies, and this facilitates risk assessment. Through the power of GO terms, analysis methods are being developed to allow for a quantitative time- and dose-dependent interpretation of genomic response using systems-based approaches to integrate responses at the biological and cellular level (Yu et al., 2006). A very good approach to integrated analysis of multiple omics datasets is Conceptgen (Sartor et al., 2010).

Both the EPA and FDA have formally recognized the power of genomic information as well as its potential limitations. Both agencies have issued interim policies guiding how to include genomic information into risk assessments and they have agreed to use genomic information in conjunction with (but not in isolation of) standard risk assessment data (FDA, 2011; US EPA, 2011f). Extensive research efforts by the US EPA Computational Toxicology program will facilitate future risk assessment uses (US EPA, 2012b).

There are numerous information resources available for risk assessment and a few are listed below to provide the reader with examples but not a comprehensive list of risk assessment resources and databases. Such resources include the Toxicology Data Network (http://toxnet.nlm.nih.gov/) from the US National Library of Medicine, the US NTP (National Toxicology Program, 2011a) including the Report on Carcinogens and the Office of Health Assessment and Translation, the WHO IPCS (WHO, 2011a), and the IARC (WHO, 2011b). The WHO Toolkit for Risk Assessment is especially user friendly for new risk assessors (WHO, 2010). WHO, through its IPCS, produces useful risk documents entitled Concise International Chemical Assessment Documents that cover a variety of chemical substances (WHO, 2011c). The Codex Alimentarius Commission created by FAO and WHO produces guidelines and reports relevant for food safety and standards (FAO and WHO, 2011). The IRIS is US EPA’s human health assessment program that evaluates health risks for over 500 chemical substances (US EPA, 2011e). IRIS lists specific risk values used by the US EPA. It is anticipated that additional databases will be available from REACH.

There is a need for a note of caution for users of these risk-relevant databases as with most databases there is a need to consider multiple contributions to differences between recommended values. These differences can be due to variations in how and when new scientific information is incorporated in the evaluations. Also, different regulatory drivers for national values or issues related to legal priorities or delays in implementing draft to final recommended values can occur. Such factors have contributed to differences in arsenic and dioxin risk estimates in the United States.

Recently, new toxicogenomic databases that identify and, in some cases, provide characterization of chemicals have become available. The NCBI (2011a,b) provides access to an enormous set of biomedical and genomic information. Its portfolio of comprehensive data and tools can be valuable for risk assessment and the aim to integrate with toxicologically relevant end points and disease is laudable. It has Chemicals and Bioassays, Data and Software, DNA and RNA, Protein Domains and Structures, Genes and Expression, genetic information databases linked with decision outcomes, Genomes and Gene Maps, and Cross-Species Homology. Resources covering databases and tools help in the study of SARs. A publicly available functional genomics data repository (Gene Expression Omnibus [GEO] database) supports MIAME-compliant data submission, and allows for easily accessible microarray- and sequence-based data, which are searchable for specific profiles of interest based on gene annotation or precomputed profile characteristics. Also variations such as data on genomic variation, interaction of genotypes and phenotypes, and the single-nucleotide polymorphisms (dbSNP) can be obtained through the NCBI. ACToR (http://actor.epa.gov/actor/faces/ACToRHome.jsp), EPA’s online database on chemical toxicity data and potential chemical risks to human health and the environment, is another useful resource for risk assessments. It allows users to search and query data including Toxicity Reference Database (ToxRefDB) (30 years and $2 billion worth of animal toxicity studies), ToxCastDB (data from screening 1000 chemicals in over 500 high-throughput assays), ExpoCastDB, and DSSTox (provides high-quality chemical structures and annotations). The ToxRefDB captures thousands of in vivo animal toxicity studies on hundreds of chemicals. Another curated Comparative Toxicogenomics Database (CTD, http://ctd.mdibl.org/) includes data describing cross-species chemical–gene/protein interactions and chemical– and gene–disease relationships to illuminate molecular mechanisms underlying variable susceptibility and environmentally influenced diseases. These data provide insight into complex chemical–gene and protein interaction networks. These databases can be especially useful for hazard identification and mechanistic information; few emphasize exposure information.

It is well known that individuals respond differently to information about hazardous situations and products (Fischhoff, 1981; Fischhoff et al., 1993, 1996; Sandman, 1993; NRC, 1996; Risk Commission, 1997; Institute of Medicine, 1999; Slovic, 2010). Understanding these behavioral responses at the individual, community, and population levels is critical in stimulating constructive risk communication and evaluating potential risk management options for risk assessment issues. 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 (Slovic et al., 1979, 2005). Club members ranked pesticides, spray cans, and nuclear power as safer than did other laypersons. 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, but nuclear power and police work as less risky than did laypersons. From studies like these, we now know that there are cultural and gender differences in perception of risks. There are also group differences in perceptions of risk from chemicals among toxicologists, correlated with their employment in industry, academia, or government (Neal et al., 1994). Recent risk perception research has emphasized the importance of knowing the balance between analytical thinking and “affect.” Our immediate emotional reaction can shape our initial risk perceptions (“affect”), while our thoughtful, deliberate “analytical” thinking also impacts our risk understanding but the latter is a slower response. Understanding this balance can help us explain why there is a complex relationship between perceived risk and benefits (Slovic, 2010).

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 (Lowrance, 1976; Morgan, 1993). Fig. 4-8 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. The public can respond negatively when they perceive that information about hazards or new technologies has been withheld or under-rated. This can explain some of the very strong responses to genetically modified foods, HIV-contaminated blood transfusions in the 1980s, or hazardous chemical and radioactive wastes. Loss of trust is exemplified by Japanese reactions following the Fukushima Daiichi nuclear reactor meltdown (March 2011), where initial perceived benefits of nuclear power were transformed to distrust during follow-up actions and responses to the earthquake and tsunami.

Perceptions of risk led to the addition of an extra safety factor (default value 10) for children in the FQPA of 1996. Engineering-based “as low as reasonably achievable” (ALARA) approaches also reflect the general “precautionary principle,” which is strongly favored by those who, justifiably, believe we are far from knowing all risks given frequently limited toxicity testing data (Roe et al., 1997).

In order for risk assessment to inform environmental risk management, there is a need to ensure that the initial “problem formulation” of the risk question(s) is succinctly framed to answer questions in the real world. Environmental health is very dynamic and many divergent emerging environmental challenges such as climate change, energy shortages, and engineered nanoparticles will require an expansion of our context well beyond single-chemical, single-exposure scenarios. In order to accomplish this goal, several factors need to be considered. These factors include defining not only health but also well-being and sustainability and will require a context of global and international scale.

Well-being is increasingly being used to describe human health and the goal of sustainable environmental risk management. The WHO has championed the use of well-being rather than health to describe the Millennium Development Goals (WHO, 2011d). Well-being goes beyond “disease-free” existence to freedom from want (including food and water security) and fear (personal safety) and sustainable futures. Concepts such as food security (abundance and quality of foods), water security (plentiful supplies and high quality of water), and sustainability form internationally recognized environmental and developmental goals. Sustainability embraces the risk management concept that “development that meets the needs of the present, without compromising the ability of future generations” to thrive and hence well-being is one of the goals of environmental actions and decisions. Recognition that environmental problems are global is essential to our understanding of how we manage risks and how we address sustainability (Leiserowitz et al., 2006; Gohlke et al., 2008; Kite-Powell et al., 2008). Ocean health and air pollution are excellent examples of the need for understanding the global context where pollutants do not honor country and national borders. World Trade Organization (WTO) agreements have also placed risk assessment approaches in the forefront as tools for managers working to meet the global goals for sustainable trade and economic health. At the most basic level new risk assessments are being done with a new global emphasis (Carruth and Goldstein, 2004; US EPA Office of Research and Development, 2011). Recent reorganization of EPA along management lines of sustainability now provides an integrative framework for working toward sustainability goals. As a response for this broader framing of health impacts, the EPA Office of Research and Development realigned their research within a new structure of Chemical Safety for Sustainability; Sustainable and Healthy Communities; Safe and Sustainable Water Resources; and Air, Climate and Energy. Risk assessment is viewed as one of the critical approaches for integrating these goals and toxicology is one of the tools used within that approach (Teichman and US EPA, 2011).

Associated with concepts of well-being and sustainability is a public health orientation to risk assessment and risk management. Public health as a discipline is very compatible with toxicology, where toxicological tests are performed to identify and characterize potential health risks and to prevent the unsafe use of such agents. Public health also has an emphasis on approaches for identifying, characterizing, and preventing risks.

Within public health risk management, there are three stages of prevention: primary, whose goal is prevention and risk or hazard avoidance; secondary, whose goal is mitigation or preparedness including risk or vulnerability reduction and risk transfer; and tertiary, where prompt response or recovery is an approach for decreasing residual risk or risk reduction (Frumkin, 2010). Fig. 4-9 shows an overview of risk assessment and management for public health where concepts of capacity assessment, vulnerability, and impact assessment are included. In this context, vulnerability assessment would include consideration of exposure and susceptibility as part of the vulnerability assessment. Hazard analysis refers to both hazard identification and probability-based frequency of anticipated events. Capacity assessment has been used for identifying strengths and resiliency of a system to impact. Recent disasters such as Hurricane Katrina, the Gulf Oil Spill, and the Fukushima Daiichi nuclear reactor meltdown all point to the need for environmental risk assessment to be considered as a part of an even larger context in order to understand critical infrastructure for determining human and environmental risks. Life-cycle analysis, green chemistry, and built environment all require such a broad context for evaluation (Keim, 2002; NRC, 2009). Fig. 4-4 also suggests that understanding population variability and vulnerability in a context of “built environment” is important. The “built environment” encompasses all man-made resources and infrastructure, including buildings, spaces, and transportation systems, which support human activity and have a strong influence on public health (Perdue et al., 2003). These types of frameworks can allow for easier consideration of public health concepts for sustainability, environmental disaster response, and life-cycle systems analysis than many of our traditional frameworks for environmental chemical risk assessment, which can be done in relative isolation.

The NRC and Risk Commission frameworks for risk assessment and risk management provide a consistent framework-based approach for evaluating risks and taking action to reduce risks. The objectives of risk assessments vary with the issues, risk management needs, and statutory requirements. Hence, setting the context and problem formation for risk evaluation is essential (NRC, 2009). The frameworks are sufficiently flexible to address various objectives and to accommodate new knowledge while also providing guidance for priority setting in industry, environmental organizations, and government regulatory and public health agencies. Risk assessment analyzes the science and, if incomplete, identifies uncertainty and provides approaches for moving forward with decisions (NRC, 2009). 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 end points as scientific findings substitute data for assumptions and help to describe and model uncertainty more credibly.