Retrospective on U.S. Health Risk Assessment:
How Others Can Benefit*

Dennis J. Paustenbach**

Contents

• Introduction
  Notes
• Health Risk Assessment: A Brief History
  Notes
• The Debate on the Appropriateness of Risk Assessment
  Notes
• Improving the Practice of Risk Assessment
  Notes
  • Hazard Identification
    Notes
  • Dose-Response Assessments
    Notes
  • Exposure Assessment
    Notes
  • Risk Characterization
    Notes
• Alternative Approaches to Health Risk Assessment
  Notes
• Conclusions
  Notes
• Appendix: Health Risk Assessment in Other Countries
  
  • Australia and New Zealand
  • Canada
  • The European Community
  • Germany
  • The Netherlands
  • Scandinavia
  • United Kingdom
  Notes

  ---

• Notes on this file.
  • From 6 Risk: Health, Safety & Environment 283-332 (Fall 1995).
  • Tables and figures are not included here.
  • With Netscape, readers can open this file in two windows and separately scroll text and notes for simultaneous viewing.

Introduction

Health risk assessment, however is a separate and distinct discipline which uses toxicology data collected from animal studies and human epidemiology, combined with information about the degree of exposure, to quantitatively predict the likelihood that a particular adverse response will be seen in a specific human population.3 The assessment of toxicology data to predict health risks is not entirely new;4 the risk-assessment process has been used by agencies for almost 40 years, most notably within the U.S. Food and Drug Administration (FDA).5 However, the difference between assessments performed in the 1950's and 1960's and those performed in the 1980's and 1990's is the incorporation of a complex and quantitative exposure assessment. With the emergence of quantitative methods, risk-assessment models can better estimate the probability that a specific adverse effect will occur over a wide range of doses.6

Since 1980, many environmental regulations and some occupational health standards have, at least in part, been based on the results of low-dose extrapolation models and exposure assessments.7 For example, risk-assessment methodologies have been used to set standards for pesticide residues, food additives, pharmaceutical agents, drinking water, soil and ambient air, as well as exposure limits for contaminants found in indoor air, consumer products and other media.8 Risk managers increasingly rely on risk assessments to decide whether a broad array of environmental risks are significant or trivial, an important task since over 300 of about 5,000 chemicals routinely used in industry have been labelled carcinogens as a result of animal studies.9 Risk assessment has become prominent in both the U.S. and Canada because it was seen as better than prior attempts to make decisions from volumes of toxicology and epidemiology data collected over the past 40 years.10

In 1980, for example, the U.S. Supreme Court considered the results of risk assessment to conclude that more stringent regulation of benzene was unwarranted unless it lowered significant risks,11 but the challenge of deciding which risks are significant faces legislators and regulators worldwide. The practice of risk assessment has helped illustrate that decisions about industrial and agricultural practices in nearly any country in some circumstances can adversely affect the health of people and the environment elsewhere. For example, the destruction of the rain forests in South America or changes in the ozone layer in Antarctica are believed to influence the climate in North America and Europe. To help insure that the most important environmental hazards are tackled ahead of lesser ones, it would be useful if all countries had a consistent and objective approach to assessing environmental risks so that political or social pressures were not given much weight.12

Contents
Skip notes

Contents

Health Risk Assessment: A Brief History

A thoughtful review of the history of risk analysis has been published.13 In it, the authors noted that perhaps the world's first professional risk assessors were the Asipu people who lived in the Tigris-Euphrates valley about 3,200 B.C. According to Covello and Mumpower,14 their primary function was to serve as consultants for risky, uncertain or difficult decisions. For risky ventures such as a proposed marriage arrangement or a suitable building site, the Asipu would identify the important dimensions of the problem, identify alternative actions and collect data on the likely outcomes of each alternative. The best available data from their perspective were signs from the gods, which the priest-like Asipu were especially qualified to interpret. After an analysis of benefits and costs of each alternative was completed, the Asipu would recommend to their client the most favorable alternative, etched upon a clay tablet.

The similarities between the practices of modern risk assessors and those of ancient Babylon underscore the historical concerns of society regarding the problems of risk and appreciation of cause and effect relationships in everyday life. By the 16th to 18th Centuries, the basis for the current approach to health risk assessment was established, including a sensitivity to the importance of exposure and response. During the early decades (1900-1940) of the 20th Century, qualitative understanding of health risk assessment improved as health scientists and factory managers learned of the hazards of occupational exposure to the more than 300 chemicals then routinely used in the workplace.15

The emergence of modern health risk assessment can be traced to about 1975. Since then, the considerable knowledge gained from scientists in many disciplines has been slowly transferred to environmental regulators, who often take a lead role in formulating improvements in various risk assessment methods.16 For example, nearly two dozen guidance documents (about 5,000 pages) on how risk assessments should be conducted have been written by the U.S. Environmental Protection Agency (EPA) since 1986. Unfortunately, as EPA standardized the process, assessments often became too inflexible to properly characterize the more likely risks. Specifically, most regulatory attempts to standardize assessment methods have introduced several levels of conservatism because such assessments have, for the sake of public safety, been structured to over-estimate true risks.17

Contents
Skip notes

Contents

The Debate on the Appropriateness of Risk Assessment

Although many in society are concerned about the uncertainty and variability of assessment processes used in today's environmental decision making, the public can take solace in the fact that new and better information concerning the uptake and metabolism of toxins is continually being developed, the degree of exposure and the likelihood that an adverse effect will occur at a specific dose. Nonetheless, it is true that chemical hazards are not, and may never be, completely eliminated when decisions are based on a risk assessment. For this reason, some nations, including Germany, Sweden and others, have traditionally regulated chemicals to reduce emissions and exposure to the lowest level achievable using engineering controls or to impose a ban, raising yet another set of risk and technology issues. Also, an approach can be problematic in that it diverts financial resources that might be used to address other risks such as AIDS or to provide immunization programs.

Reducing health risks to levels that are "as low as reasonably achievable" (ALARA) or requiring the use of the "best available technology" (BAT) can produce significant reductions in the degree of exposure, but either policy has two possible shortcomings. First, adopting an ALARA or BAT approach can be costly and may not result in an appreciable overall benefit (reduction of risk) to society.20 Second, a reliance on banning may not ensure that a significant or even measurable level of risk reduction will occur.21 History demonstrates that banning chemicals may well eliminate one risk, but often this hazard is replaced by another and the financial sacrifice is often not to the overall benefit of society. For example, chemical substitutions, as in the case of saccharin and Nutrasweetreg., may raise new uncertainties about other risks to health. In an ideal world, the costs of decisions concerning risk reduction should be weighed against the benefits of applying the same resources to reduce other important risks such as adequate medical care for all citizens or immunization.22 However, no one has yet determined how to distribute limited financial resources optimally between very different portions of the U.S. budget dealing with, e.g., education, space engineering or the military.

Contents
Skip Notes

Contents

Improving the Practice of Risk Assessment

Nearly twenty years of U.S. experience have taught scientists a great deal about how to perform each step in risk assessment more efficiently and accurately.25 With the ongoing and accelerating application of risk analyses worldwide, the time is ideal to adopt many improvements to risk assessment that have been identified which could significantly advance the usefulness of risk assessment in the twenty-first century.

Contents
Skip notes

Contents

This helps explain why more than 400 chemicals have been found in animal studies to produce tumors, yet fewer than twenty are known human carcinogens.31 Even after accounting for the typical epidemiology shortcomings, it is clear that some, if not many, rodent carcinogens do not pose an equivalent cancer hazard in humans.32 Although it is plausible that some carcinogens may pose a greater human hazard than that suggested by rodent studies, there are few examples where appropriate animal tests were conducted.

As a result of knowledge gained during 1975-95, the first twenty years of health risk assessment, it has become clear that most animal carcinogens (at some dose) will probably pose a human cancer hazard. However, actual human cancer risk at very low doses remains unclear and will continue to be debated for many years.35 The criteria by which a risk assessor determines that a chemical could pose a significant carcinogenic or developmental threat to humans is based on animal studies and consideration of at least six factors. For carcinogens, these include the number of animal species affected, the number and types of tumors occurring in them, the dose (relative to the acute toxic dose) at which they are affected, the dose/response relationship, and the genotoxicity of the chemical.36 For developmental toxicants, the key issues in hazard identification are similar to those for carcinogens and include the number of species affected, the severity of effects, the slope of the dose-response curve and the relationship between the dose which affects the mother compared to that affecting the offspring.37

Over the past two decadess, scientists have also learned not to place equal weight on all data. Regulators, for example, need to resist the temptation to emphasize any data that suggests that a chemical poses a carcinogenic or developmental hazard but put little weight on any that suggests a failure to cause such effects at environmentally relevant exposures. This approach has heretofore been considered prudently health-protective. Yet, it is not scientific, nor is it necessarily in the public's best interest.38 For example, extraordinary confidence has sometimes been placed on studies suggesting that a chemical may pose a particular hazard with modest consideration of the study's quality.39

During the 1990's, the scientific community and most regulators have come to accept that not all data are equal, and that only data of similar quality should be judged equally.40 In the U.S., this is known as the "weight-of-evidence" approach. It represents an important refinement and is applicable not only to hazard identification, but also to exposure and dose-response evaluations. One benefit of using a "weight-of-evidence" approach is that it minimizes the possibility that huge sums of money will be spent to conduct several high quality toxicity studies, only to have the results refuted by one or two poor ones.41 This approach has been embraced by many environmental agencies outside the U.S. Some of the best examples where an agency attempted to use a weight-of-evidence approach, coupled with understanding of the mechanism of action, include d-limonene,42 formaldehyde,43 NTA,44 methylene chloride,45 and chloroform.46

contents
Skip notes

46. Rory A Corley et al., Development of a Physiologically-Based Pharmacokinetic Model for Chloroform, 103 Toxicol. Appl. Pharm. 512 (1990).

contents

As we enter the mid-1990's, it is clear that the most uncertain aspect of chemical assessments, especially carcinogens, is the low-dose extrapolation. Although perhaps humbling, most toxicologists agree that they are limited in ability to estimate the risks associated with typical levels of environmental exposure based on results of standard rodent bioassays.47 There are many reasons why this is so. First, we do not fully understand all of the possible mechanisms of action for carcinogens.48 Second, the doses at which we conduct the animal tests are so high that they often produce effects that would not occur at the doses to which people are exposed.49 Third, there are usually significant differences between animals and humans with respect to the rate at which chemicals are metabolized, distributed and excreted. Fourth, the delivered dose to specific target tissues in animals will often be much higher than that delivered to human target tissues.50 Thus, scientists must rely on a model or theory to estimate the human response at experimental doses that are often one-thousand-fold below the lowest tested animal dose as illustrated below.51

Figure 1 depicts a dose-response curve from an exceptionally thorough (8 dose groups) study. The solid line is a best fit of the eight data points identified. The three lowest indicate that at these doses, no increased incidence in tumors was observed in the test animals. The error bars on the three lowest doses indicate the statistical uncertainty in the test results since a limited number of animals were tested at each dose (n = 100). To insure that risk estimates are not underestimated, the models most frequently used by regulators are based on the upper bound of the plausible response, rather than the best estimate.

To date, in the U.S. most regulatory approaches for identifying safe or acceptable levels of exposure to air, water and soil contaminants have been based on statistical rather than biologically-based models. As discussed below, the physiologically-based pharmacokinetic (PB-PK) models are the most scientifically rigorous (and likely most valid) models for estimating safe levels of human exposure based on animal data.54

Finally, as shown in Table 2,56 the results of commonly-used models usually vary in a predictable manner because they use different mathematical equations for predicting the chemical's carcinogenic potency.

It is noteworthy that rodent studies now used to predict the magnitude of human risk were never intended for that purpose.58 They were designed to qualitatively identify potential human cancer hazards, not to quantitatively estimate the human risk at low levels of exposure.59 As in hazard identification, whenever adherence to strict regulatory guidance requires that dose-response assessments must use a single mathematical model, the assessment can become so constrained that valuable biological information (which could dramatically alter the results) is not fully accounted for.60

We have now learned a good deal about the biology of cancer, and this information should be considered in the dose-response assessment whenever possible. For example, we now know that there are at least three broad classes of mechanisms by which chemicals may produce a carcinogenic response in rodents: repeated cytotoxicity, promotion and initiation.61 Some have suggested as many as eight different classes of mechanisms through which chemicals produce tumors.62 These distinctions are important since the appropriate model for estimating the cancer risk for humans exposed to low doses of a cytotoxicant or promotor should be different than that for an initiator.63 Because at relatively high doses non-genotoxicants may produce repeated cytotoxicity, the primary reason for excessive cell turnover, many scientists expect them to possess a threshold dose below which no cancer hazard would be present.64 This is in contrast to genotoxicants or mutagens that may pose some risk, albeit small, at very low doses.

In general, the scientific underpinnings of the dose-response models used for assessing carcinogens are based on our understanding of ionizing radiation and genotoxic chemicals.65 Both types of agents may well have a linear, or a nearly linear, response in the low-dose region. However, promoters and cytotoxicants need not have a linear dose-response curve. Scientific data increasingly suggest that promoters and other non-genotoxic agents will have a very non-linear dose-response relationship at low doses and, as importantly, probably have a genuine or practical threshold.66 The increased acceptance of this postulate is evidenced by EPA's position that the linearized multi-stage model is inappropriate for dioxin,67 thyroid type carcinogens,68 nitrilotriacetic acid (NTA),69 d-limonene70 and, presumably, similar non-genotoxic chemicals.

Over the past twenty years, the scientific community has learned at least six ways to improve the way that regulatory agencies in the U.S. have predicted the possible human risk at low levels of exposure to carcinogens (Table 1).

First, we have learned that it is important to identify the best estimates from the cancer models, as well as the upper and lower bounds of the risk.71 The objective of the bounding techniques is to attempt to account for the statistical uncertainty in the results of the animal tests, however, we have rarely presented the degree of potential conservatism within the bounding procedure.72 The problem of not presenting all of the results is that the risk manager will not be fully aware of the breadth of equally plausible risk estimates.73 For example, the cancer risk associated with exposure to chloroform in most U.S. drinking water has been reported to be as high as one in 10,000 using the upper bound risk estimate of the multi-stage model;74 however, using the same model, the best or maximum likelihood estimate (MLE) of the risk is about one in 1M and the lower bound estimate is virtually zero (about one in ten million). Therefore, the plausible range of risk is as high as one in 10,000 and as low as zero. When biological factors are considered, such as its weak genotoxicity and pharmacokinetics, the carcinogenic risk associated with low concentrations of chloroform in chlorinated drinking water is likely to be quite small or negligible.75 Yet, as a matter of policy, the risk of one in 10,000 is most cited by regulators and appears in newspapers.

Second, we have learned not to rely on only one mathematical model. There are at least six different models that may need to be considered when estimating the risks at low doses. Each of them can yield results which are plausible, depending on the mechanism of action and pharmacokinetics of the chemical, as well as the characteristics of the dose-response curve.76 With better understanding of carcinogenesis and the shortcomings of statistical models, regulators have recently become more willing to consider models that can quantitatively account for chemical-specific mechanisms of action.77 However, support for encouraging flexibility in the risk-estimation process has been criticized on the grounds that our knowledge about carcinogenesis is insufficient to regulate in other than a very conservative manner.78 The application of Monte Carlo techniques to dose-response assessment and the polling of experts should help resolve some of these philosophical differences in how to establish regulatory goals.

Third, we need to give greater weight to epidemiology studies. It is usually claimed that these studies are almost never as statistically robust as the animal studies and, therefore, are not very useful.79 Yet, total acceptance of this assertion is inappropriate because epidemiological studies can, at least, establish the degree of confidence that should be placed in the results of low-dose extrapolation models.80 Often, even less-than-perfect epidemiology, coupled with retrospective exposure assessments, can yield much more defensible estimates of likely human health hazards than statistical models based on animal studies. Recent work evaluating benzene furnishes an example.81

Fourth, we should quantitatively scale up data from rodents to predict the human response. For example, when evaluating most toxicologic effects, statisticians and biologists have generally assumed that at a given dose (mg/kg-day) the rodent response to a chemical will be nearly identical to the human response. This approach is usually reasonably accurate for noncarcinogenic effects. In contrast, for carcinogens, several factors need to be considered when trying to predict how humans will respond compared to rodents. First, the biologic half-life between rodents and humans can be expected to be different for virtually all chemicals. Often, for a given chemical, these differences will vary in a predictable manner based simply on the body weight to surface area ratio and/or the life span.82 This may be valid for those chemical carcinogens which require metabolic activation, but may not be very accurate for those carcinogens that do not. Consequently, for regulatory purposes, surface area corrections have been used in an attempt to adjust for the pharmacokinetic differences between rodents and humans. Recent work suggests that body weight to the 2/3 power is the most valid scale-up factor if no compelling information to the contrary is available.83

Figure 385 depicts a physiologically-based pharmacokinetic (PB-PK) model for inhalation exposure to C-14 Carbon tetrachloride.86 Such models allow scientists to predict how humans will respond to a chemical based on data collected in rodents. Basically, the movement and transformation of the test chemical within the rodent is described by mathematical equations. The same is done for the human. If the rodent data tracks a chemical's behavior in humans, one can be confident that the model can quantitatively predict the human response at several different doses.

The fifth, and possibly most important, lesson we have learned is that low-dose models are useful for ranking classes of carcinogens objectively, but they should not be expected to precisely predict the cancer risk because they cannot account for biological information such as the mechanism of action, types of tumor, differences among species, metabolism and genotoxicity. Perhaps the most promising approach to incorporating biologic factors into cancer risk estimates is the use of biologically-based models such as the one by Moolgavkar-Knudson-Venzon (MKV).90 This model accounts for the number of mutations required for malignancy and the role of target cell birth and death processes associated with these mutations. A key element is a quantitative description of how the carcinogen affects the cellular birth, death and mutation rates. Unfortunately, because most of the information needed to perform these analyses is not yet available, and may not be measurable, it holds limited promise in the near future.

Sixth, experience has shown that EPA-recommended cancer models often do not respond to the characteristics of the dose-response curve. As discussed by Sielken,91 it does not seem appropriate to base important regulatory decisions on the results of models which are minimally "responsive" to the very costly information collected in standard lifetime rodent studies. Another way to look at this is to understand the fragility or sturdiness of these models not only w hen interpreting bioassay data, but also when selecting the dosing regimen to be used in future bioassays. If one conducts only one statistical test to select the form of the model, this limits the ability to learn as much as possible from rodent data. A way to avoid this problem is to conduct simulations of the model's responsiveness to alternative, but similar, data sets to ensure that the extrapolation is reasonable.92 As shown by Reith and Starr,93 the potential range of tumor incidences that may be analyzed with the multistage model are restricted by the experimental design (specifically, by numbers of doses and animals) of the bioassay; this in turn limits the possible values of carcinogenic potency that can arise. This needs to be considered when selecting the doses to be used in biassays.

Contents
Skip notes

93. James P. Reith & Thomas B. Starr, Experimental Design Constraints on Carcinogenic Potency Estimates, 27 J. Toxicol. Environ. Health 287 (1989) and James P. Reith & Thomas B. Starr, Chronic Bioassays: Relevance to Quantitative Risk Assessment of Carcinogens, 10 Regul. Toxicol. Pharm. 160 (1989).

Contents

The primary routes of exposure to chemicals in the environment are inhalation of dusts and vapors, dermal contact with contaminated soils or dusts, and ingestion of contaminated foods, water or soil.98 Initial efforts to quantitatively estimate the uptake of environmental contaminants by humans were first conducted by scientists in the field of radiological health99 and their work can be a source of valuable information when conducting assessments of chemical contaminants. Numerous methodologies for estimating the human uptake of contaminants have been proposed and refined in recent years.100

We have learned at least eleven rather significant lessons about conducting exposure assessments that can immediately be adopted elsewhere. These could potentially save other nations hundreds of millions of dollars and thousands of person-years of work (Table 1).

First, U.S. experience has shown that in attempts to be prudent, we have overemphasized the "maximally exposed individual" (MEI).101 Often, the results of these analyses were misinterpreted by the public and/or misrepresented by some scientists or lawyers. For example, some assessments addressed only risks to the MEI. Yet, that the risk may affect only the 99.5th or 99.9th percentile of all exposed populations was rarely stated. Current EPA exposure guidelines and NAS guidance acknowledge this deficiency and note that a worst-case or MEI analysis should be used only to decide if an exposure is insignificant and should not be used to characterize the actual or plausible human risks. In short, most MEI analyses should only be used in screening assessments, although some have noted that so-called "reasonable" person assessments often don't present the full picture.102

Second, as we have learned to characterize accurately the risks of exposure for about 95% of the population, more emphasis has been placed on evaluating various special groups (e.g., Eskimos, subsistence fisherman, dairy farmers), who can be exposed to particularly high doses (the 95%-99.9% group). Although the risk for these populations needs to be understood, the typical levels of exposure for the majority of exposed individuals should be the initial focus of most risk assessments. For example, if a regulatory agency bases its decision on the results of an assessment of persons who eat about 100g of freshwater fish every day (99th percentile), yet the average American eats only 18g of fish per day (lifetime average), the analysis should reflect the fact that 99 of 100 persons are not represented by the corresponding risk estimate. To help minimize the potential for misunderstanding, it is important to describe the number of exposed persons at each anticipated dose level, along with the most likely and upper estimates of exposure and the associated plausible risk.103 With this information, risk managers can decide whether large or small sums of money are warranted to reduce the health hazard.104

Third, don't allow the repeated use of conservative assumptions to dictate assessment results. In recent years, several investigators have discussed this issue and have demonstrated its importance.105 The problem can be illustrated by one assessment of the dioxin hazard posed by municipal waste incinerators. Before deciding to issue a permit, the EPA conducted a screening level assessment, evaluating the theoretical cancer risk for a child who lived within a short distance (0.8 KM) from the hypothetical incinerator. They assumed that a child could eat as much as about one small spoonful of dirt each day, that his house was down-wind of the stack, that he ate fish from a pond near the incinerator, that his fish consumption was at the 95th percentile level, that he drank contaminated water from the pond, that he ate food grown primarily from the family garden, and that he drank milk from a cow which grazed on forage at the farm. Not surprisingly, it was predicted that siting the incinerator could plausibility increase the child's lifetime cancer risk (if he lived there for 70 years) by one in 100; but he could hardly be portrayed as a typical person living near a municipal incinerator. Regrettably, the associated upper-estimate of the risk was the only one reported by the press. Certainly, it would have been more appropriate to have studied and presented the number of persons likely to be exposed to this amount of contaminant, as well as the level of exposure for the typical person living within ten miles of the facility. It may also have been useful to note that few farms are located near most incinerators since operators try to minimize transportation costs from urban areas.

Fourth, in exposure assessment the problems associated with the repeated use of overly conservative assumptions and the need to properly account for small (but highly exposed) populations can now be overcome with Monte Carlo techniques.106 The probabilistic or Monte Carlo technique addresses the main deficiencies of the point estimate approach because it imparts a great deal more information to the risk manager.

Figure 4 illustrates how a Monte Carlo analysis is applied to understand the distribution of time needed to go shopping based on the three activities involved in shopping. Time spent shopping each month (minutes) is estimated by the product of two parameters: the number of trips per month and the total time spent in the store (minutes). Total time spent in the store is the sum of time spent shopping and time spent in line. Using Monte Carlo techniques, a distribution of likely values is associated with each of these parameters. These distributions are dependent upon the detail of information available to characterize each parameter. For example, the distribution compares all of the information such as those days when the line at the check-out counter is short, as well as those when it's long. It is noteworthy that each parameter has a different distribution: log-normal, gaussian, and square. Total time spent shopping is then calculated repeatedly by combining parameter values that are randomly selected from these distributions. The result is a distribution of likely time spent shopping each month. Using this technique, information concerning each parameter is carried along to the final estimate.

Instead of presenting a single point estimate of risk, probabilistic analyses characterize a range of potential risks and their likelihood of occurrence. In addition, those factors which most affect the results can be easily identified. For example, in a probabilistic analysis, one can present the risk manager with the following type of statement:

The plausible increased cancer risks for the 50th, 95th, and 99th percentiles of the exposed population are 1 x 10-8, 5 x 10-7 and 1 x 10-6, respectively. However, if there was to be a fishing ban on catfish and other bottom feeders, the risks for the 50th, 95th, and 99th percentile groups would drop by 100-fold.

Our understanding of the concentration of DDT in the edible portion of smallmouth bass, is based on high quality, reliable data; therefore, our confidence in the risk estimates is high. Our analysis indicates a 95% level of confidence that 90% of the increased cancer risk could be eliminated by a ban on catching carp or catfish in the river and that little additional risk reduction would be achieved if bass and trout fishing were also banned.

This type of characterization provides the kind of information risk managers need to make informed decisions regarding the necessary magnitude of response, the cost/benefit of particular decisions, and whether a range of risk-reduction alternatives should be considered.

We have also refined our approach to conducting statistical analyses of environmental data (lesson six). For example, we now see the importance of accounting for the log-normal and truncated distributions. As has been shown repeatedly, most environmental and occupational data are log-normally distributed rather than conform to a gaussian distribution,108 but few environmental scientists were aware of the significance until recently. As noted in the recently pro posed EPA guidelines for exposure assessment, inappropriate statistical analysis of environmental data is one of the most easily corrected of the common errors in exposure analysis.109

Seventh, we have improved techniques for statistically handling samples having no detectable amount of contaminants.110 Frequently, agencies have used the limit of detection (LOD) in exposure calculations on the premise that the contaminant might be present at that level. Many agencies have suggested that 50% of the LOD should be used to calculate the plausible degree of human exposure when no detectable amount is identified; others have suggested that the LOD divided by the square root of two, and other much more complex approaches have been recommended. In short, when such approaches are used on a site that may only be 2-10% contaminated (based on surface area) the impact of a few samples on the calculated average level of contamination will be much higher than what is likely to be present across the site unless proper statistical methods are applied. Several approaches have been suggested to help avoid this problem.111

Eighth, we need to account for the environmental fate of chemicals. Many factors such as degradation by sunlight, soil and water microbes, and evaporation can dramatically influence the degree of human exposure. Yet, many assessments have frequently assumed concentrations measured today will exist and remain constant for 30-50 years.112 For instance, the public health hazard posed by the potential release of dioxin vapors from incinerators was recently evaluated. After it was alleged that the vapors posed a serious health hazard to surrounding residents, a risk assessment was conducted. Yet, the environmental half-life of dioxin (as a vapor) proved to be critical because it has a half-life of only 90 minutes. In contrast, TCDD in deep soil, fly-ash or sediment may have an environmental half-life of 50-500 years. What had been portrayed as a potentially serious health hazard was shown to be insignificant, given its photolytic half-life.113

Ninth, we have recognized the need to validate exposure assumptions or model estimates by using field measurements. Although past sampling and analytical procedures were inadequate to measure the low concentrations of toxicants found in the environment, better ones are now available. As field measurement techniques are further refined, less reliance should be placed on mathematical models for predicting chemical distributions in the environment, and more confidence should be placed on environmental samples.114

Tenth, and perhaps the most significant advance in exposure assessment of airborne contaminants, is recognition of the need to account for indirect exposure pathways. For example, the uptake of a contaminant in water by direct ingestion is obvious, but uptake of the same contaminant from watering garden vegetables or inhaling volatiles while showering have not always been evaluated.115 When regulating airborne nonvolatiles, the most important indirect route of exposure, not considered before 1986, may be ingestion of particulates deposited on plants subsequently eaten by grazing animals.116 The ingestion of the meat and milk from these animals can produce, depending on the chemical and conditions, risks 50-200 fold greater than that from inhalation.117 Methods for estimating uptake through virtually all indirect routes have been developed and continue to be refined.118

Finally, whenever possible one should use biological monitoring to confirm predicted human exposures. Over the past five to ten years, analytical chemists can increasingly detect very small quantities of dozens of chemicals in the blood, urine, hair, feces, breath and fat of exposed persons. Measuring parts per trillion and part per quadrillion is now possible. For many chemicals, this represents a direct indicator of recent exposure, and in some cases like PCBs and dioxins, chronic exposure. For example, the uptake of dioxin by Vietnam veterans exposed to 2,4,5-T was recently evaluated by analyzing the amount of dioxin in their blood. This study, conducted almost fifteen to twenty years after the last day of service in Vietnam, allowed epidemiologists to conclude that the vast majority of veterans had only a modest degree of exposure to a chemical alleged to produce numerous adverse health effects in field soldiers.119

Contents
Skip notes

119. Larry Needham, Serum 2,3,7,8-Tetrachlorodibenzo-p-dioxin Levels in U.S. Army Vietnam-Era Veterans: CDC Veterans Heath Studies, 260 J. AMA 1249 (1988) and William H. Wolfe et al., Determinants of TCDD Half-Life in Veterans of Operation "Ranch Hand," 41 J. Toxicol. Env. Health 481 (1994).

Contents

A first-rate risk characterization now offers numerous opportunities to describe the "big picture." It has become generally accepted in the U.S. that theoretical increases in cancer risk of one in 1M or even one in 10,000 from exposure to an environmental contaminant should not be depicted as a serious health risk.121 Risks predicted in most assessments are usually the upper, not the best, estimate of risk. It has been recommended that every assessment should state that risk estimates represent an upper bound of plausible risk (when appropriate), are unlikely to underestimate risk, and that actual risk may be much lower, even zero.122 In short, it is important to communicate that conservative procedures and models are likely to overestimate human risk.123 Yet, if risks may be underestimated, this, too, needs to be stated.124

Second, we have learned to account for background contaminant levels.125 In assessments of the 1970's and 1980's, risks from natural and/or background levels likely to have influenced risk managers were often overlooked. For example, in a preliminary decision regarding the necessary level of clean-up at a Superfund site in a rural community, the state agency suggested that soil levels of a cyclodiene be reduced to 10 ppm -- to keep the theoretical cancer risk below one in 1M. However, the administrator decided not to require remediation when it was recognized that the background levels of some cyclodienes in many areas in the U.S. exceed 50 ppm and that no adverse health effects had been observed or suspected.126

Third, and perhaps most interesting, we learned during early years of the environmental movement that the hazard may not always be as it appears. For example, at several sites where large sums of money were used to remove lead- or PCB-contaminated soil, there was no decrease in the blood levels of these contaminants in local children or adults.127 How could this be? As we have since learned, a primary route of uptake is through the dust in the home which has been transferred from the soil via tracking or deposition of airborne particulate. Thus, cleaning homes is often more effective in reducing the health risks than removing the contaminated soil from the community; and certainly one must accompany the other. In many locations, this type of rather simple but practical solution may be the only financially plausible one for dealing with soil contaminated by persistent chemicals. Thus only paving over contaminated soil (like mine waste that covers hundreds of acres) and then conducting a thorough cleaning of the homes may be a reasonable, although temporary, approach to bringing about a rapid decrease in the health risk.

Fourth, we have learned that the significance of model-estimated cancer risks needs to be communicated thoroughly and understandably. For example, the goal of some standards, such as the maximum contaminant levels (MCLs) for drinking water, is to keep the maximum plausible cancer risk between one in 100,000 and one in 1M. However, it might be useful to note that about 30% of all Americans will develop cancer and about 25% will eventually die from it. Accordingly, a 1 x 10-6 risk is equivalent to ensuring that the lifetime cancer risk for persons exposed to this level of contamination will be no greater than 250,001 in 1M (25.0001%), rather than the background rate of 250,000 in 1M (25%). If society demands this standard of care, that is its choice. However, both society and its risk managers should understand the relative magnitude of various risks before deciding to spend money on one hazard over another. As recently noted by Gough,128 even if the EPA were to regulate all carcinogens in air, soil and water at levels generally considered insignificant, the decrease in cancer incidence in the U.S. would probably be only about 0.25% to 1.3% of the annual cancer mortality.

Fifth, since the late 1980's, risk characterizations have been found much more useful if they contain cost/benefit analyses. In general, the public and its elected officials want to know, for each dollar spent, how much risks are apt to be reduced for various portions of the population. Most risk scholars and managers believe that such analyses are the primary justification for performing risk assessments.129 For these applications the use of Monte Carlo techniques has been invaluable.

Sixth, we now recognize that the hallmark of a good risk characterization is a discussion of the uncertainties in the risk estimates. Since 1990, many improvements have been made concerning how to conduct quantitative, rather than qualitative, uncertainty analyses and how to present them.130 The key element of these analyses is a statistical evaluation of each of the various parameters and the implication with respect to the risk estimate for various segments of the population. Although at first blush this may sound like a scientific nuance, the public and the courts now insist on understanding the level of confidence in risk predictions.

A key trait of a high-quality risk characterization is the accurate and unbiased discussion of our confidence in the risk estimates. Often, regulatory agencies and the press have erroneously stated or implied that the results of low-dose models actually predict the increased cancer risk for exposed individuals. As noted previously, because statistical models cannot account for all biological mechanisms (including repair), they cannot accurately predict the actual cancer risk. As discussed by Dr. Frank Young, a former FDA Commissioner, risk estimates should not be portrayed as anything more than relative indicators of risk.131 He noted, for example, that when the FDA uses the risk level of one in 1M, it is confident that the risk to humans is virtually nonexistent, rather than one in 1M exposed persons is likely to develop cancer.132 As noted by Finkel, risk scholars and policy makers have rarely used model-derived estimates to predict the incidence of cancer but rather only as a risk-ranking tool.

Although the U.S. may well have spent a large fraction of its environmental monies over the past twenty years attacking problems which were of modest importance, and many of the decisions were influenced by the poor quality of risk characterizations, this level of regulatory activity clearly raised public awareness. Much of their concern however, focussed on the possibility that exposure to industrial chemicals might increase the cancer risk and the public had been convinced by many scientists that there was "no safe level of exposure" to a carcinogen. To most toxicologists, the public's anxiety about exposure to very low concentrations (e.g., doses) of carcinogens was generally greater than it needed to be, nonetheless these concerns fueled the fire to pass more regulations. Fortunately, a significant amount of work has been invested by researchers in an attempt to improve the way we characterize risk so that decision making is likely to be more reasoned in the future. One issue over which we have argued was how to define the word "safe" when talking about exposure to carcinogens.

A very recent paper appears to do a good job at describing what most people would agree was a "safe" level of exposure or what would constitute "acceptable risk." The authors suggest that if dose is portrayed in units of molecules of exposure, then it would be easier to identify at what dose a cancer risk can be considered "safe." The authors discuss the potent animal carcinogen dioxin (TCDD) in their paper:133

The probability of developing cancer with exposure to one molecule of 2,3,7,8-TCDD a day over a lifetime is calculated by the upper bound EPA approach to be less than 2 x 10-15. Given the astronomical numbers of cells and molecules that exist in the human body and the myriad of factors that could prevent a single interaction of a carcinogen molecule with a strand of DNA from ultimately creating a tumor, it is not surprising that we should calculate an extremely small risk for such an infinitesimally small level of exposure.

To empower individuals to decide if a "safe" level exists it is instructive to explore the dimensions of this calculated risk for the smallest possible indivisible daily dose. Using the calculated risk value and assuming similar lifetime exposure to 2,3,7,8-TCDD for the entire current population of the world indicates that there would be less than 1 in a 100,000 chance of even a single case of cancer arising from that infinitesimal level of exposure. From another perspective, the calculated risk value indicates that all humans who have ever lived on the planet could have been exposed at this infinitesimal level without ever producing a single case of cancer in the history of the human race.

This analysis does not attempt to consider the current levels of exposure to 2,3,7,8-TCDD nor to evaluate what risk levels may be associated with those particular human exposure levels because different mechanisms likely apply. Furthermore, this calculation for a hypothetical infinitesimal exposure neither supports nor refutes arguments on the contentious issues surrounding the human health significance of 2,3,7,8-TCDD as an environmental contaminant. However, it does illustrate that there is a quantifiable low level of exposure, which most reasonable individuals would likely find to be below their own worry threshold.[Note omitted.]

Contents
Skip notes

133. Steve E. Hrudey and Daniel Krewski, Is There a Safe Level of Exposure to a Carcinogen? Envir. Sci & Technol. (in press 1995).

Contents

Alternative Approaches to Health Risk Assessment

One alternative is technology-based approaches. This has, in fact, been adopted in the 1992 Amendments to the U.S. Clean Air Act. Another is to ban substances or legislate prohibitions on industrial chemical emissions. Both approaches rely exclusively on hazard identification and are very dependent on the skills of analytical chemists since any measurable quantity could initiate action. In support of this approach, some have argued that bans are the only successful means to significantly reduce environmental risks.135 Bans on the chemicals DDT, PCB and lead in gasoline are often identified as major success stories.

A third alternative to risk assessment, that appeals to some environmentalist groups, is to adopt simpler rules for conducting health assessments. For example, the European approach involves applying a safety or uncertainty factor to the no-observed-effect level from the best animal study and assume that this will prevent the adverse effect.136 This scheme places an equal weight on both carcinogenic and non-carcinogenic chemicals. The advantage is that it is fast and inexpensive, but a perceived disadvantage is that it concedes that some level of exposure is probably safe for virtually all persons. In California and a few other states in the U.S., risk assessment methods have been standardized so that they can be conducted rapidly and inexpensively. This, too, can promote or introduce many other scientific problems because all the data are often unaccounted for.

A fourth alternative some have advocated relies upon public pressure to minimize hazards. This approach has been used in several states in the 1980's and has been argued to be highly effective. The best documented approaches are California's Proposition 65 and California's Assembly Bill AB-2588, wherein acceptable levels of exposure were established using a single method, and any entity that appears to expose persons above these levels must report both to the agency and to exposed persons.137 Sending letters of notification to potentially affected persons is widely considered effective for encouraging chemical users to minimize or eliminate emissions. This sidesteps the many scientific problems with risk assessments by using public pressure to prioritize hazards. Many within the environmental community believe that this approach is quite workable.

Contents
Skip notes

137. Copeland et al. (1994), supra note 17.

Contents

Conclusions

The scientific underpinnings of about twenty years of health risk assessment practice and the implications for environmental policy have been discussed in more than 600 peer-reviewed and published papers, providing a wealth of information to other countries now in the process of evaluating whether to use risk assessment to resolve difficult environmental issues. Without question, other countries can learn from the U.S. experiences, saving themselves billions of dollars spent as a result of well-intended, but misguided, decisions made during the early years of the environmental revolution.140

In spite of numerous scientific or methodological uncertainties, properly conducted risk assessments can provide reasonably accurate predictions of the exposure of various populations and relatively accurate estimates of the magnitude of health risk (except, perhaps, for carcinogens).141 Risk-assessment procedures have matured a great deal over the past ten years. Use of less rigid approaches to interpreting the significance of animal bioassay data should produce much more defensible hazard identifications in the coming years. Biologically-based disposition and cancer models should provide better estimates of the actual human cancer risk at low levels of exposure, thus improving the dose-response segment. Reliance on worst-case exposure scenarios is no longer necessary in light of better information on specific exposure parameters and more sensitive techniques for measuring concentrations of contaminants in the environment. The use of Monte Carlo techniques has revolutionized the practice of exposure assessment and soon will change the practice of dose-response analysis. New and highly sensitive analytical procedures will permit us to assay hair, blood, urine, adipose and other biologic media to validate the reasonableness of the exposure estimates. Statistical procedures to account for the distribution of various factors within exposed populations will almost certainly be integral to future risk assessments.

Our awareness of numerous scientific uncertainties, as well as knowledge of how to best characterize them, will almost certainly lead to more credible health risk assessments that will be more useful to risk managers worldwide. If refinements are incorporated into future risk assessments and are used to establish acceptable levels of chemical exposure, scarce financial resources can be devoted to problems that, when resolved, will yield the largest improvement in public health.

For these reasons, the EPA's Science Advisory Board (SAB)142 and the NAS143 have promoted the routine use of quantitative risk assessment in regulatory decision making. It appears that Australia, New Zealand, Canada, the European community, countries in the Pacific Rim and others will derive significant benefit from studying the costly learning curve through which the U.S. has passed.

Contents
Skip Notes

138. Center for Risk Analysis, supra note 7 and Carnegie Commission, supra note 8.

139. See, e.g., Linda-Jo Schierow, Comparison of Environmental Risk Provisions in the 103d Congress, 5 Risk 283 (1994).

140. Barnes, supra note 10; Viscusi, supra note 1 and EPA, supra note 12.

141. Paustenbach, supra note 17 and Lavy et al., supra note 36.

142. EPA, supra note 8.

143. NAS, supra note 3.

Contents

One major difference between the U.S. and other countries is how carcinogens are evaluated and regulated. In countries such as the U.K. and Germany, that rely primarily on expert judgments, chemical carcinogens are regulated using a case-by-case approach. Known or suspected chemical carcinogens are subjected to an individual review that considers both the mechanism of action and the epidemiology data. Risk assessment usually involves the formation of expert advisory committees that make the decisions regarding exposure standards or regulations, rather than an agency. The advisory body commonly uses a "weight-of-the-evidence" approach, in which all of the available information and test data are used to formulate a scientific position for consideration as the basis for regulatory decision. This approach has historically been poorly received in the U.S. due to pressure to establish public policy that errs considerably on the side of safety. In the wake of unrelenting financial pressure from competing social needs, and the European experience, the weight-of-evidence approach has gained momentum within EPA in recent years.

Contents

Australia and New Zealand

Second, there is a commitment to use risk assessment to help identify the most appropriate level of clean-up. According to recent guidelines, a contaminated site should be cleaned as necessary to minimize short- and long-term, on- and off-site, risks.147 To accomplish this, there is an emphasis on flexibility to allow the most logical solution to be selected and implemented. For example, regulators are encouraged to consider that it may be appropriate in cases where there is no threat to human health and the environment to accept a strategy of containing contaminants or using planning controls to limit site use. As a consequence, a fourth characteristic, unlike policies adopted elsewhere, is recognition that it may be much less expensive, and therefore wiser, to wait for new technologies to be developed to address particularly difficult environmental problems.

Finally, health risk assessment guidelines in Australia and New Zealand clearly attempt to combine both the human and environmental risks when addressing chemical contamination. The emergence of ecological risk assessment worldwide appears to be gaining greater recognition than in the U.S. due, in part, to increased concern about preserving our natural resources and the rapidly growing eco-tourism industry which may well become a central part of some economies.

Contents

Canada

In regulating chemical substances, the responsible Canadian authorities consider the probable carcinogenic mechanism of action when making decisions about acceptable levels of exposure. For example, for nongenotoxic chemicals, uncertainty factors are added to the no-observed-adverse-effect level (NOAEL) to calculate a tolerable daily intake similar to the U.S. acceptable daily intake (ADI). Conversely, for genotoxic carcinogens, regulators select one of several methods under a policy aimed to reduce health risks by as much as possible. At this time, the approach used to assess genotoxic carcinogens involves estimating an exposure-potency index (EPI). This index compares the expected exposure of a population with an estimate of the carcinogenic potency of the chemical. The potency estimate is derived from epidemiologic or animal data with the objective of identifying the dose that would cause a carcinogenic response in 5% of the test subjects.

Contents

The European Community

European legislation pertaining to risk assessment has been mainly directed at chemical safety, pesticide residues, food additives and occupational exposures. Most countries mandate both qualitative and quantitative risk assessments, depending on the type of chemical and usage. A common characteristic of most directives passed or proposed in those fields is that the member states or individual employers are responsible for performing risk assessments, not the regulatory agencies. Scientific experts are used throughout, both by those performing risk assessments and by various agencies during the review process. Although as recently as 1993 there was significant optimism that roughly similar risk-assessment procedures would be used throughout Europe, this appears unlikely in the near future.

Contents

Germany

Although German authorities do not widely practice risk assessment, known human carcinogens are strictly regulated. To date, known human carcinogens have been subjected to stringent regulations focusing partially on the best available technology (BAT) or, in the case of drinking water regulations, on international E.C. directives. It is also commonplace for decisions concerning the regulation of chemical carcinogens and other hazardous chemicals in Germany to be made by multipartite expert committees on a case-by-case basis. Typically, an expert committee relies upon a NOAEL/ADI approach to identify acceptable levels of exposure to noncarcinogenic chemicals.

Contents

The Netherlands

In the Dutch risk-assessment process, scientists initially evaluate a chemical to determine its genotoxicity. Subsequently, they use information about functional effects and chemical structure, the results of bioassays, and other relevant data to reduce the uncertainties relating to the genotoxic potential in humans. When it is impossible to eliminate completely the risk of exposure to a genotoxic carcinogen, the Dutch approach relies upon a very simple linear extrapolation model to determine a dose-response value for human exposure.153 A simple linear extrapolation of animal data to humans is considered a very conservative approach, as the metabolic rate of humans is lower than that of animals and is also inversely proportional to age and weight. DNA repair processes appear to be proportional to body weight, and the sensitivity of man to known human carcinogens is about equal to that of experimental animals.154

Like most other European countries, virtually all of the safe levels of exposure for a nongenotoxic carcinogen identified in regulations are based on a NOAEL divided by a safety factor of ten to 1,000, depending on the amount of uncertainty in the data. The resulting value represents an acceptable daily intake for the substance.

Dutch regulators have distinguished themselves by having developed a number of innovative methods for assessing the hazards posed by contaminated soils.155 The Dutch standards are probably the most widely cited soil standards in the world.156 They recently added guidelines for air, groundwater, surface water and sediment. In support of these efforts, they have also conducted a number of research efforts to evaluate the uptake of soil by humans and the environmental fate of the contaminants.

Contents

Scandinavia

Regulators in Denmark employ risk-assessment techniques only to a limited extent when determining exposure standards for carcinogens. In cases where new chemicals may be substituted for a suspected carcinogen, an assessment is required to examine whether exposure can be eliminated effectively. Low-dose extrapolation models are used to estimate risk when a nonthreshold genotoxic carcinogen cannot be replaced by another chemical.160

In Sweden, the EPA sets exposure standards for a variety of carcinogenic and noncarcinogenic chemicals and uses classic risk-assessment methods to assess health risks for industrial emissions.161 Like other European countries that use risk assessment, regulators consider carcinogens with a pronounced genotoxic mechanism as prime candidates for careful analysis. As part of that process, the Swedish EPA uses mathematical models to extrapolate from animals exposed to high doses of carcinogens to predict the effects on humans exposed to lower doses. The agency also evaluates the carcinogens using a case-by-case approach in which each chemical is assessed individually, as opposed to the more generic approach common in U.S. regulatory agencies that follow fairly strict guidelines.

Swedish regulators use a weight-of-the evidence approach to evaluate a chemical's carcinogenic potential. Like most other European countries, genotoxic carcinogens are typically regulated to ensure the lowest possible levels of exposure; for nongenotoxic chemicals, safe levels of exposure are identified from a NOAEL or lowest-observed-adverse-effect level (LOAEL) and a safety factor applied based on the level of uncertainty in the available information. These ADIs are then used to calculate maximum residue levels of pesticides in food and occupational exposure limits.

Contents

United Kingdom

Papers list several reasons for U.K. skepticism of commonly-used U.S. dose-extrapolation models. First, a linear low-dose model has yet to be validated. Second, the bioassay data used to derive a low-dose model of chemical carcinogenicity are incomplete. Third, the low-dose models are based more on mathematical assumptions than established biochemical mechanisms; consequently, risk estimates vary widely depending on the model used. Last, models give an unjustified impression of precision, given the approximations and assumptions upon which they are based.163

The regulation of chemical carcinogens in the U.K. is based on mechanistic considerations. For example, if a chemical acts by a genotoxic mechanism, regulators assume, as a matter of prudence, that the compound does not have a threshold. If a nongenotoxic mechanism is known to be involved, regulators consider it possible to identify a safe level of exposure.164

Chemicals displaying genotoxicity are evaluated using expert judgment and a weight-of-the-evidence approach. In evaluating such compounds, experts consider all available evidence (including human data, animal data, mutagenicity data and structure/activity relationships). If they conclude that the compound should be considered a potential human carcinogen that acts by a genotoxic mechanism, they recommend action to reduce levels to as low as is reasonably practical or to eliminate it entirely. For suspected carcinogenic compounds operating through a well-understood nongenotoxic mechanism, researchers evaluate animal studies to determine the NOAEL, which is then divided by a safety factor to derive an ADI. The safety factor reflects the uncertainties of extrapolating from animals to humans and of variation among individuals.165 ADIs are also used to calculate maximum residue levels for pesticides on food.

Contents

144. National Health and Medical Research of Australia and New Environmental Council, Australian Guidelines for the Assessment and Management of Contaminated Sites (1990).

145. Kenneth M. Murchison, Environmental Law in Australia and the U.S.: A Comparative Overview (two parts), 6 Environ. Plan. L. J. 179 and 8 Environ. Plan. L. J. 254 (1994).

146. Ossama El Saadi & Andrew Langley, The Health Assessment and Management of Contaminated Sites. (South Australian Health Commission 1991).

147. Australian Guidelines, supra note 144.

148.OTA, supra note 136.

149. Commission of the European Communities, Chemical Risk Control in the European Community (1987).

150. Organisation for Economic Cooperation and Development, Environmental Directorate, OECD Cooperative Risk Reduction Activites (1992).

151. R. Turck, Nature, Conservation and Nuclear Safety (Fed. Ministry Environment) (1993).

152. OTA, supra note 136.

153. Richard Kroes, Animal Data: Interpretation and Consequences in Environmental Carcinogenesis (P. Emmelot & E. Kriek, eds. 1979).

154. Richard Kroes, Contribution of Toxicology Toward Risk Assessment of Carcinogens, 60 Arch. Toxicol. 224 (1987).

155.Joop H. van Wijnen, P. Clausing & B. B. Brunekreef, Estimated Soil Ingestion by Children, 51 Environ. Res. 147 (1990) and Joop H. van Wijnen, Health Risk Assessment of Soil Contamination (1990).

156. M. E. J. Van der Weiden, Environmental Quality Objectives in the Netherlands (1994).

157. L. Dragsted, Low Dose Extrapolation of Data from Animal Bioassays, in Risk Managements and Risk Assessment in Different Sectors of Denmark 77 (1992).

158. Department of Health (U.K.), Comm. on Carcinogenicity of Chemicals in Food, Guidelines for Evaluation of Chemicals for Carcinogenicity (1991)

159. Dragsted, supra note 155.

160. Id.

161. OTA, supra note 136.

162. van Wijnen, supra note 155.

163. U.K., supra note 158.

164. OTA, supra note 136.

165. U.K., supra note 158.

Risk Articles Index