The U.S. Environmental Protection Agency's 1988 Risk Assessment for Arsenic
THIS chapter provides an overview of the U.S. Environmental Protection Agency's (EPA's) 1988 risk assessment for arsenic and a discussion of what was done, how it was done, what issues were considered important, and how those issues were addressed. The risk assessment is then discussed in light of the events and information from the past 10 years that might affect its interpretation and utility.
Overview Of The EPA 1988 Special Report
EPA's 1988 risk assessment for arsenic is a special report of the EPA Risk Assessment Forum. Its development was motivated by major scientific controversies within EPA surrounding the health effects of ingested arsenic. A Technical Panel on Arsenic was charged with preparing a report on the health effects of arsenic for agency-wide concurrence and use. A draft of the report was peer reviewed externally at a workshop of experts, and many reviewer comments were incorporated into the final report. The designation "special report" was used to distinguish this report, which is limited to issues of skin cancer and nutritional essentiality, from the agency's comprehensive hazard-assessment documents. As described in the report, the forum "addresses many of the hazard identification, dose-response assessment, and risk characterization parameters called for in the cancer guidelines, but it does not fully assess or characterize arsenic risks for skin cancer nor does it analyze the other cancers associated with exposure to this element" (EPA 1988, p. 1).
There is general agreement that inhalation of inorganic arsenic is associated with increased lung-cancer risk and that available data are adequate to
estimate the magnitude of the risk. The forum identified three sources of uncertainty and controversy, however, with respect to the assessment of risk from ingestion of arsenic (EPA 1988, p. 3): (1) Evidence of human carcinogenicity is primarily from epidemiological studies conducted in other countries, and its applicability to the U.S. population is uncertain. (2) The primary tumor response in those studies is skin cancer, which is more likely to be detected and successfully treated than many other forms of cancer. (3) Limited animal evidence suggests that arsenic might be an essential nutrient, raising the possibility that reducing the level of arsenic below some critical level (as yet unspecified) might result in a decrement of health in some way (as yet unknown).
The forum concluded that the epidemiological studies conducted in Taiwan on arsenic in drinking water (Tseng et al. 1968) and confirmatory studies (Albores et al. 1979; Cebrian et al. 1983) demonstrate that inorganic arsenic is a human carcinogen by the oral route (group A by EPA's classification scheme) (EPA 1988, p.3). The forum also concluded that the Taiwanese studies provide a reasonable basis for quantitative risk assessment of skin cancer in the United States, despite many uncertainties. Those two conclusions affirm the adequacy of the evidence to address the first two steps (i.e., hazard identification and dose-response assessment) of the four-step risk-assessment paradigm developed by the National Research Council (NRC 1983). EPA's group A classification for hazard identification does not refer to potency but to the strength of the evidence for human carcinogenicity. The result of the dose-response assessment is an estimated increase of U.S. lifetime skin-cancer risk of 3-7 x 10-5 (µg/L)-1 (i.e., 3-7 additional cases of skin cancer per 100,000 persons for each microgram of inorganic arsenic per liter in the drinking water). The "3" and "7" are not statistical bounds, but the outcomes from data on females and males, respectively. A value of 5 x 10-5 (µg/L)-1 has been used to estimate skin-cancer risk for both sexes in the United States. An implicit assumption in those values, which are a consequence of the methodology of EPA's risk-assessment guidelines, is that at low arsenic concentrations the risk of skin cancer increases linearly with the arsenic concentration in drinking water. The forum noted that assumption as a source of uncertainty, stating that the dose-response curve might be less than linear and might not pass through the origin. The forum pointed out, but did not explore in depth, epidemiological evidence of an association between inorganic arsenic and internal cancers (EPA 1988, Appendix C) and concluded that more studies are needed to determine if arsenic is a nutritional requirement in humans (EPA 1988, p. 40).
The remainder of this section examines each chapter of the special report to highlight what was done and how it was done, issues that were or were not resolved, and the remaining sources of uncertainty and possible data gaps.
The objective is to provide a baseline for considering the contributions of research in the past 10 years to current understanding of issues related to exposure and health effects of ingested inorganic arsenic. The discussion to follow is organized according to hazard identification, dose-response assessment, and nutritional essentiality of arsenic.
The EPA forum based its 1988 risk assessment on skin cancer as the health-effects end point. Evidence of an association between arsenic and cancer at internal sites (e.g., liver, kidney, lung, and bladder) had begun to appear in the literature, but the data needed to quantify risk were not yet available to EPA (1988, p. 11). Three studies were found suitable for quantification of skin-cancer risk. The primary study (called the "Tseng study") is a large cross-sectional prevalence survey conducted in a region of southwest Taiwan endemic to blackfoot disease (Tseng et al. 1968). The two supporting studies are a prevalence study conducted in the Region Lagunera of Mexico (Albores et al. 1979; Cebrian et al. 1983) and a retrospective study of clinical patients treated with a 1:1 dilution of Fowler's solution containing 3.8 grams (g) of arsenic per liter (Fierz 1965).
Strengths of the Tseng study noted in the report (EPA 1988, p. 16) are (1) the large size (40,421 exposed and 7,500 controls), (2) a statistically significant increase in skin cancer in the exposed population many years after first exposure, (3) a pronounced skin-cancer response by arsenic exposure level, (4) comparability of the exposed and control populations, aside from arsenic exposure, and (5) a high rate of pathological confirmation of observed skin cancer (over 70%). Important uncertainties identified in the Tseng study are (1) the possibility of confounding by other chemicals in the drinking water, (2) the lack of blinding of the examiners, and (3) the possibility that diet might be a risk modifier.
The forum studied the genotoxicity, metabolism, and pathology of arsenic in relation to the possible mechanism by which arsenic might induce carcinogenic effects but concluded that current knowledge was inadequate to be factored "with confidence" into the risk-assessment process (EPA 1988, p. 7). The interrelationships between various lesions of chronic arsenic poisoning and the progression of lesions were studied for their potential use in dose-response assessment, but the information was insufficient to develop a mechanistic model. Hyperpigmentation, a pathological hallmark of chronic arsenic exposure, was not considered an end point for dose-response assessment because it is not a malignant condition and does not appear to be a premalignant condition (EPA 1988, p. 19).
The weight of evidence from tests for mutagenicity is summarized in the report (p. 22), and possible genotoxicity mechanisms, such as interference with DNA synthesis or repair, are postulated. However, the genotoxicity mechanisms of arsenic were not sufficiently understood to incorporate them into a dose-response model formulation, although their importance was noted. Genotoxicity at low doses is an important indicator of irreversible change in genetic function. Such changes are a critical feature of many postulated mechanisms for chemical carcinogenesis and the basis for ascribing low-dose linearity to carcinogenic processes. Although the lack of genotoxic response does not preclude linearity at low doses, it is potentially important as a consideration in selecting a model for extrapolation of carcinogenic risk (EPA 1988, p. 23).
Metabolism and distribution are described (EPA 1988, p. 24) as important factors in evaluating the carcinogenic properties of arsenic, but the information was inadequate for use in dose-response assessment. The forum noted that methylation appears to be a means of detoxifying inorganic arsenic forms; compounds become less acutely toxic as methyl groups are added. At low intakes of arsenic, intake and excretion seem to be balanced, but at sufficiently high intakes, urinary excretion might be compromised, leading to increased tissue concentrations. There is also evidence that the methylating capacity of the body changes as a function of exposure, and maximal levels of excretion of methylated arsenicals are reached after weeks of exposure to the compound. Similarly, the ability to excrete the methylated arsenicals seems to be lost as a function of time after removal of arsenical exposure (EPA 1988, p. 25). In addition, diet might influence methylating capacity and, if so, might affect the reliability of results extrapolated from the Tseng study to the U.S. population. Thus, metabolism and distribution data are considered important for evaluating the carcinogenic properties of arsenic, but further study is needed (EPA 1988, p. 26).
Lacking definitive data on genotoxicity, pathology, metabolism, and pharmacokinetics to help determine the shape of the dose-response curve, particularly at low doses, EPA used the multistage model to fit the data of the Tseng study and included a factor for duration of exposure. (That model is called the "generalized multistage model" in the report and is also commonly known as the "multistage Weibull model.") Using that model, the forum found that the estimated risk of skin cancer at low doses increased linearly with dose and stated that the estimate should be considered an "upper-bound estimate" because multiple hits or threshold considerations might apply. "The
risk at low doses may be much lower than the current estimates, as low as zero, due to such factors as the metabolism or pharmacokinetics of arsenic" (EPA 1988, p. 28).
To test the model fit, it was used to predict the prevalence of skin cancer in Region Lagunera in Mexico (Cebrian et al. 1983). The prediction was judged consistent with the observed number, which was four. Similarly, model estimates were calculated for the patients given Fowler's solution in Germany, as reported by Fierz (1965). The number of observed skin cancers was judged "not inconsistent" with the model prediction (EPA 1988, p. 30).
The forum attempted to quantify two uncertainties in the dose-response evaluation: use of prevalence data in place of cumulative incidence data in the model, and the intake of arsenic from sources other than drinking water (e.g., food). An implicit assumption in the use of prevalence data is that mortality is the same in persons with skin cancer as in those without. Skin cancer can cause death, but mortality is not high. Of greater concern is the correlation of skin cancer with other diseases, such as blackfoot disease, that affect mortality. The net effect of the association of skin cancer with blackfoot disease and possibly other causes of death is an underestimation of the risk of skin cancer. Persons with blackfoot disease and skin cancer have a shorter life expectancy than do persons with skin cancer alone and thus were less likely to be included in the Tseng study. Under a reasonable set of assumptions, it was calculated that the dose-response relationship would be underestimated by up to 50% (EPA 1988, p. 31).
Information on the concentrations of inorganic arsenic in food was lacking, so the dietary intake from that source could not be calculated. The concern is that food items, principally rice and yams grown in arsenic-contaminated water and fish, might also have been a source of inorganic arsenic. It is noted that arsenic intake from sources other than the drinking water would overestimate the unit arsenic risk calculated from the Taiwan study (EPA 1988, p. 86). Some illustrative calculations are included on how the arsenic intake from food consumption could affect the risk estimate, using rice and sweet potatoes as an example.
In its summary discussion of the uncertainties, the forum repeated a precaution: "absent animal data or reliable human data under conditions of low exposure, the shape of the dose-response, if any, at low doses is uncertain" (EPA 1988, p. 31).
Evidence of nutritional essentiality of trace amounts of inorganic arsenic would affect any interpretation of the health risks at low exposure levels. The
report addressed two questions of uncertainty with respect to nutritional essentiality, one qualitative and the other quantitative. The qualitative question concerns whether inorganic arsenic is nutritionally essential in humans. Experimental studies with rats, chicks, minipigs, and goats have suggested the "plausibility" of essentiality (EPA 1988, p. 38). However, the published information was insufficient to determine the reproducibility of arsenic deficiency syndrome in those species, and a physiological role of arsenic or mechanism of action had not been identified. If arsenic is required in animals, the forum noted, then it is highly probable that it is also required in humans. Although somewhat academic if the essentiality of arsenic has not been established, the quantitative question concerns how essentiality would be incorporated into a risk assessment. Essentiality suggests that below a certain threshold, arsenic is not harmful but needed, and a quantitative value would be needed for implementation in dose-response assessment.
EPA's 1988 Risk Assessment: 10 Years Later
At the time of the 1988 EPA risk assessment and the preceding arsenic risk assessment (EPA 1984), the only source of dose-response data on skin cancer was from the Tseng study. EPA was aware of a mortality study on cancer at multiple internal sites conducted in the same region of Taiwan as the Tseng study (EPA 1988, p. 90), but dose-response data from that study were not published until 1988 (Chen et al. 1988). Currently, more detailed data are available for cancer at multiple internal sites in the "Chen study" than for skin cancer in the Tseng study. Thus, some choices, albeit limited, of data sets and end points are available for use in dose-response assessment, and additional studies might soon provide more choices. Subsequent to the 1988 EPA risk assessment, more detailed information has become available on the distribution of arsenic concentrations in well water used as a measurement of exposure in the Tseng study and the Chen study. Moreover, much more epidemiological and clinical evidence has been reported from other countries, and some advances have been made in understanding arsenic metabolism and the mode of action of arsenic carcinogenicity. These subjects are discussed in more detail in the following chapters. The remainder of this chapter focuses on the 1988 EPA risk assessment and the developments since 1988 that might affect its interpretation and current usefulness.
As stated previously, the EPA forum based its risk assessment on the Tseng study. That study involved a house-to-house survey of 40,421 inhabit-
ants of a southwestern region of Taiwan where artesian wells with a high arsenic concentration had been in use for at least 45 years. The study clearly established a dose-response relationship between arsenic in drinking water and the prevalence of skin cancer and, along with other confirmatory sources, provided ample evidence for EPA's classification of inorganic arsenic as a group A human carcinogen. EPA's dose-response assessment, however, which was also based on the Tseng data, deserves closer scrutiny.
The data limitations of the Tseng study for application to dose-response assessment are due to insufficient detail, at least as reported. The Tseng study, which contains only a summary of the study data, remains the single data source. Only the sex-specific and age-specific (20-year intervals) prevalence rates were reported for skin cancer (Tseng et al. 1968, Table 1). To associate those outcomes with arsenic exposure, other information in the article had to be used to estimate how the total number of persons at risk and the number of skin-cancer cases (by sex and age) were distributed over three broad exposure categories (low, 0-300 µg/L; medium, > 300-600 µg/L; and high, >600 ,g/L) plus an ''undetermined category" as described below. The median value of arsenic well-water tests in each exposure range was then used by Tseng et al. (1968) as the representative arsenic concentration for dose-response assessment. (The effect of grouping data on dose-response assessment is addressed for bladder cancer in Chapter 10.)
Epidemiological data are desirable because they provide observations on humans instead of animals, but human exposures are seldom known precisely. Assumptions about exposure to arsenic are generally indicated in the forum's risk assessment. Briefly, each individual's rate of exposure to arsenic was treated as constant since birth. (Some temporal variability might occur in arsenic concentrations in wells (Tseng et al. 1968), in sources of drinking water or rates of consumption, or in dietary sources of arsenic.) Except for an unspecified number of villages assigned to an "undetermined" category, villages were assigned to a low-, medium-, or high- exposure category on the basis of the outcomes of tests for arsenic in village wells. (As described in Tseng et al. (1968), the undetermined category included those villages where either the artesian wells with arsenic-polluted water were not in use or the difference in the arsenic content in water from various artesian wells in the same village was so great that it was impossible to put them into the low, medium, or high category.) The health-effects data were then summarized across villages in each of the low-, medium-, and high-exposure categories; the undetermined category contained about 40% of the cancer cases. Thus, although over 40,000 people were personally examined in the field, the observations on health effects could only be associated with broad exposure categories.
For the dose-response assessment, the EPA forum assumed that the skincancer rates for the three exposure categories applied to median arsenic concentrations of 170 µg/L, 470 µg/L, and 800 µg/L, concentrations based on a histogram of all arsenic tests of well water for all villages combined. The actual well-water measurements were not reported for each of the villages.
Estimating Risk in the United States
The objective of the dose-response assessment was to estimate the risk of skin cancer at low concentrations of arsenic in drinking water in the United States. That was done in two steps: (1) estimation of risk at low arsenic concentrations in the observed Taiwanese population, and (2) extrapolation of that risk to the U.S. population by rescaling dose to account for differences between Taiwan and the United States in average body weight and average rate of consumption of drinking water. The first step was accomplished by fitting the multistage Weibull model to the Tseng data by maximum likelihood and then evaluating the fitted curve at a specified age and arsenic dose for an estimate of the prevalence of skin cancer at that age and dose. For the second step, the dose was rescaled to an equivalent arsenic intake (per unit of body weight per day) for a U.S. reference person (weighs 70 kilograms (kg) and drinks 2 L of water per day). U.S. Life Tables from the National Center for Health Statistics were used to convert prevalence estimates to estimates of lifetime risk of skin cancer.
The determinants of the outcome of the first step are the data and the choice of model fit to the data. As discussed above, the response data are highly summarized with limited knowledge of the correct representative doses. Aside from concern about data validity, however, data summarization can bias the shape of the estimated dose-response curve (see Chapter 10). The other determinant of the outcome of the first step, the choice of model, provides further reason for caution. The correct formulation of the model to fit to the data is unknown. It is not uncommon for several hypothesized models to fit observed data about equally well but to produce substantially different risk estimates at a low-dose exposure. As noted previously, the EPA dose-response assessment (using the data summarized in the low-, medium-, and high-exposure categories) predicts linearity at low dosei.e., that excess risk of skin cancer increases proportional to rate of arsenic intake for a fixed number of years of exposure. More recently, an expert committee has reviewed the evidence on the mode or modes of action for arsenic and the implications for linearity or nonlinearity (see Chapter 7). Further studies on the metabolism of arsenic have found no evidence of a threshold (see Chapter 5).
With regard to the second step (extrapolation of risk from Taiwan to the United States), after rescaling for difference in body weight and consumption of water in Taiwan and the United States, EPA applied the dose-response assessment of arsenic in drinking water in Taiwan to estimate the risk of arsenic in drinking water in the United States. The contribution to arsenic intake from dietary sources could not be factored into the dose-response assessment for Taiwan or the extrapolation to the United States because of insufficient information on arsenic in food. Limited data on dietary arsenic intake in the blackfoot-disease region now available suggest that arsenic intake from food is higher in Taiwan than in the United States, although more data and improved quantification are needed for confirmation (see discussion of arsenic in food in Chapter 3 and evidence for essentiality and beneficial effects in Chapter 9). Ideally, the health effects from arsenic should be attributable to total intake from both food and drinking water, rather than drinking water alone. Consideration of arsenic in food might affect both the dose-response relationship for arsenic in drinking water in the study population of Taiwan and the implications for risk from arsenic in drinking water in the United States where dietary arsenic might differ from that in the study population in Taiwan (Brown and Abernathy 1997).
EPA's 1988 risk assessment of ingestion of inorganic arsenic is described in this chapter. This special report of the EPA Risk Assessment Forum was limited to the issues of skin cancer and nutritional essentiality. Quantification of skin-cancer risk for hazard identification and dose-response assessment was based on three studies in humans. The primary study was the large prevalence study conducted in southwestern Taiwan (the Tseng study), where people were exposed to high concentrations of arsenic in drinking water. The two supporting studies were a prevalence study of persons exposed to arsenic in groundwater in Mexico and a study of clinical patients treated with arsenicals in Germany. The EPA 1988 report concluded that arsenic is a carcinogen by the oral route and estimated an increase of lifetime skin-cancer risk of 3-7 cases per 100,000 population for each microgram of inorganic arsenic per liter of drinking water. The report cautioned, however, that the risk at low doses could be much lower because of the pharmacokinetics of arsenic.
Since the EPA 1988 report, additional evidence has been found that reinforces the forum's conclusion that ingestion of inorganic arsenic causes skin cancer. Sufficient evidence has also been published to confirm that arsenic ingestion causes cancers more fatal than skin cancer, in particular,
lung cancer and bladder cancer. The risk of those cancers warrant consideration and would add to risk of skin cancer. New information related to EPA's estimate of lifetime risk of skin cancer from arsenic, however, has added additional uncertainty to the sources used in the EPA report. In particular, it has been learned that arsenic exposure among persons and villages grouped together in the data reported in the Tseng study is more variable than previously realized. This variability might be largely accounted for by Tseng's creation of an undetermined category, but the specifics are unclear. Another concern is the need to consider arsenic intake from both food and water. Although recognized as a source of arsenic in the forum's report, dietary intake of arsenic in Taiwan and the United States was not sufficiently documented to be included in the 1988 risk assessment. Only sparse data are currently available, but estimates of lifetime risk of skin cancer from consumption of arsenic in drinking water show that estimates could be sensitive to differences in dietary intake of arsenic.
Albores, A., M.E. Cebrian, I. Tellez, and B. Valdez. 1979. Comparative study of chronic hydroarsenicism in two rural communities in the Region Lagunra of Mexico. Bol. Oficina Sanit. Panam. 86:196-205.
Brown, K.G., and C.O. Abernathy. 1997. The Taiwan skin cancer risk analysis of inorganic arsenic ingestion: Effects of water consumption rates and food arsenic levels. Pp. 260-271 in Arsenic: Exposure and Health Effects, C.O. Abernathy, R.L. Calderon, and W.R. Chappell, eds. London: Chapman & Hall.
Brown, K.G., H.R. Guo, T.L. Kuo, and H.L. Greene. 1997. Skin cancer and inorganic arsenic: Uncertainty-status of risk. Risk Anal. 17:37-42.
Cebrian, M.E., A. Albores, M. Aguilar, and E. Blakely. 1983. Chronic arsenic poisoning in the north of Mexico. Hum. Toxicol. 2:121-133.
Chen, C.J., T.L. Kuo, and M..M. Wu. 1988. Arsenic and cancers [letter]. Lancet i(8582):414-415.
EPA (U.S. Environmental Protection Agency). 1984. Health Assessment Document for Inorganic Arsenic. Final Report. EPA 600/8-83-021F. U.S. Environmental Protection Agency, Office of Health and Environmental Assessment, Environmental Criteria and Assessment Office, Cincinnati, Ohio.
EPA (U.S. Environmental Protection Agency). 1988. Special Report on Ingested Inorganic Arsenic: Skin Cancer; Nutritional Essentiality. EPA 625/3-87/013. U.S. Environmental Protection Agency, Risk Assessment
Forum, Washington, D.C.
Fierz, U. 1965. Catamnestic investigations of the side effects of therapy of skin diseases with inorganic arsenic. Dermatologica 131:41-58.
NRC (National Research Council). 1983. Risk Assessment in the Federal Government: Managing the Process. Washington D.C.: National Academy Press.
Tseng, W.P., H.M. Chu, S.W. How, J.M. Fong, C.S. Lin, and S. Yeh. 1968. Prevalence of skin cancer in an endemic area of chronic arsenicism in Taiwan. J. Natl. Cancer Inst. 40:453-63.