National Academies Press: OpenBook

Arsenic in Drinking Water: 2001 Update (2001)

Chapter: 5 Quantitative Assessment of Risks Using Modeling Approaches

« Previous: 4 Variability and Uncertainty
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 169
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 170
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 171
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 172
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 173
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 174
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 175
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 176
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 177
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 178
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 179
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 180
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 181
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 182
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 183
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 184
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 185
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 186
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 187
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 188
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 189
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 190
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 191
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 192
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 193
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 194
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 195
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 196
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 197
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 198
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 199
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 200
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 201
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 202
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 203
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 204
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 205
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 206
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 207
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 208
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 209
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 210
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 211
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 212
Suggested Citation:"5 Quantitative Assessment of Risks Using Modeling Approaches." National Research Council. 2001. Arsenic in Drinking Water: 2001 Update. Washington, DC: The National Academies Press. doi: 10.17226/10194.
×
Page 213

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

Quantitative Assessment of Risks Using Docketing Approaches OVERVIEW OF TlIE SCIENCE UNDERLYING EPA'S 2001 PROPOSED REGULATION On January 22, 2001, following the publication of a proposed rule for arsenic in drinking water (EPA 2000a) and a period of public comment, EPA published a final rule for arsenic in drinking water in the Federal Register, setting a maximum contaminant level goal (MECG) of zero for arsenic in drinking water and a maximum contaminant level (MCL) for arsenic of 10 fig in drinking water (EPA 2001~. Typically, when developing an MUNG and an MCL, a risk assessment is conducted. Two important components of a risk assessment are hazard identification and dose-response assessment ARC 1983~. Exposure assessment end risk characterization are also important steps in a risk assessment, but they are beyond the scope of this subcommittee's charge and, therefore, will not be discussed here. The purpose of hazard identification is to determine whether the agent in question causes adverse effects. Deciding which end point is the most sensitive and which studies or data sets are most appropriate for assessing the risks from a chemical are major conclusions from a hazard identification. In the case of EPA's assess- ment of arsenic, the risk being assessed is the risk to the U.S. population from consumption of arsenic in drinking water. The purpose of dose-response assessment is to determine the relationship between the dose and the incidence of an adverse effect in humans. Major conclusions from the dose-response 169

~ 70 ARSENIC IN DRINKING WA TER: 2001 UPDA TE assessment include the model or models that can be used to best determine the risks to the U.S. population from arsenic in drinking water and understanding ofthe impacts of different model choices on the risk estimates from that analy- sis. The details of EPA's hazard identification (choice of endpoint and choice of study) and dose-response assessment (choice of model, selection of a com- parison group, and adjustments for water intake, diet, and mortality versus incidence) are discussed below. Hazard Identification Choice of End Point EPA' s hazard analysis is included in Section m ofthe proposed rule (EPA 2000a) and in Section m.D. ~ of the final rule (EPA 2001~. EPA "relied upon the NRC ~ ~ 999] report as presenting the best available, peer reviewed science as of its completion and has augmented it with more recently published, peer reviewed information" in its proposed rule. EPA (2000a) concludes that acute or short-term effects are not seen at 50,ug/L (the MCL at the time of the pro- posal) and, therefore, addresses the "Iong-term, chronic effects of exposure to Tow concentrations of inorganic arsenic in drinking water." With respect to Tong-term effects, EPA concludes that "arsenic is a multi- site human carcinogen by the drinking water route," and on the basis of epide- miological studies of Asian, Mexican, and South American populations, those "with exposures to arsenic in drinking water generally at or above several hundred micrograms per liter are reported to have increased risks of skin, bladder, and lung cancer." EPA also notes that increased risk of liver and kidney cancer have been associated with arsenic exposure and that skin cancer has been associated with inorganic arsenic contamination in Argentina (re- viewed by Neubauer 1947, as cited in EPA 2000a), in Poland (EPA 2000a), and in a dose-dependent manner following exposure to arsenic in drinking water in Taiwan (Tseng et al. 196S, Tseng 19774. Other epidemiological studies also support an association between arsenic exposure and skin cancer (Roth 1956; Albores et al. 1979; Cuzick et al. 1982; Cebrian et al. 1983~. EPA discussed data from two studies in Taiwan demonstrating a statisti- cally significant increase in mortality risks for bladder, kidney, lung, liver, and colon cancer (Chen et al.1985), and a significant dose-response relationship for death from bladder, kidney, skin, and lung cancer in both sexes and from

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES ~ 71 liver and prostate cancer in males (Wu et al. ~ 989~. An increase in internal cancers was also seen in Argentina (bladder, lung, and kidney cancer) (Hopenhayn-Rich et al. 1996, 1998) and in Chile (bladder, kidney, and lung cancer) (Smith et al. 1998~. Tsai et al. (1999) reported that lung, bladder, intestinal, rectal, and laryngeal cancer were associated with chronic exposure to arsenic in drinking water in Taiwan. EPA also reviewed a study by Lewis et al. ~ 1999) that reported mortality of a population in Utah exposed to lower concentrations (average, 18-191 ~g/~) of arsenic in drinking water in which there was a statistically significant increase in prostate-cancer mortality, but no increase in bladder or lung cancer mortality. EPA also discussed a study by Kurttio et al. (1999) that found a significant association in a case-control study in Finland between bladder cancer and exposure to very Tow concentra- tions of arsenic in drinking water (odds ratio of ~ .53,95% confidence interval (C~ = 0.75-3.09 at 0.1-0.5,ug/~; 2.44,95% C! = ~ . ~ l-5.37 at greater than 0.5 ,ug/~. No association was seen for kidney cancer. EPA reviewed noncancer effects that are observed following chronic exposure to arsenic including dermal effects (Yeh 1973; Tseng 1977; Cuzick et al. 1982), gastrointestinal effects (Morris et al. 1974; Nevens et al. 1990; Mazumder et al. 1997), peripheral vascular disease (Tseng 1977; Zaldivar ~ 974; Cebrian ~ 987; Lewis et al. ~ 999), and diabetes (Lad et al. ~ 994; Rahman and Axelson 1995; Rahman et al. 1998~. in the final rule, EPA again summarized the acute and chronic effects of arsenic, and added a discussion of a study from Japan (Tsuda et al. 1995~. The study found an association between exposure to arsenic in drinking water and lung and bladder cancer. in addition, EPA (2001) added a short discus- sion on the potential susceptibility of children to arsenic. EPA agreed with the conclusion of the majority of the EPA Science Advisory Board (SAB) mem- bers (EPA 2000b) that children are generally at greater risk for a toxic re- sponse to any agent in water because of their greater drinking-water consump- tion (on a unit-body-weight basis), but that the available data, including a study of infant mortality in Chile (Hopenhayn-Rich et al. 2000), do not dem- onstrate a heightened susceptibility to arsenic. After discussing all the toxic effects of arsenic, the water concentrations at which they occur, and the NRC (1999) report, EPA chose cancer as the most sensitive end point, stating that it "focused its risk assessment on the carcinogenic effects of inorganic arsenic" (EPA 2000a). EPA (2001) states that lung and bladder cancer are the internal cancers most consistently seen and best characterized in epidemiological studies, and its quantitative risk assessment is based on data for those two cancers.

72 ARSENIC IN DRINKING WA TER: 2001 UPDA TE Choice of Study An important decision in a quantitative risk assessment is the choice of critical study (or studies) to be used in the dose-response assessment. At the time of EPA's proposed rule, few animal carcinogenicity bioassays had been conducted for arsenic, and there were no positive animal models for dose- response modeling. There was, however, a "large data base on the effects of arsenic on humans" (EPA 2000a, p. 38902~. EPA concluded that questions remain about the shape of the dose-response relationship at Tow concentra- tions. The advantages of using the studies from southwestern Taiwan (Chen et al. 1985; Wu et al. 1989) for quantitative risk assessment, according to EPA, are the duration of exposure and follow-up, the size of the population (more than 40,000 individuals), the extensive pathology data, and the homoge- nous lifestyles ofthe population. Those studies are limited, however, by their design (i.e., they are ecological epidemiology studies), which makes quantita- tive evaluation of dose-response relationships more difficult. EPA also stated that the studies from Chile (Smith et al. 1998) and Argentina (Hopenhayn- Rich et al.1996; 1998) are more limited than the Taiwanese studies (Chen et al. 1985; Wu et al. 1989) and not suitable for quantitative dose-response as- sessment, but that they provide supportive evidence for the effects seen in southwestern Taiwan. EPA concluded that "tt~hese epidemiological studies provide the basis for assessing potential risk from Tower concentrations of inorganic arsenic in drinking water" (EPA 2000a, p.38902~. In its final rule, EPA also concluded that the Utah study by Lewis et al. (1999) "is not power- ful enough to estimate excess risks with enough precision to be useful for the Agency's arsenic risk analysis." Therefore, in its final rule, EPA (2001) still considered the southwestern Taiwan data to be the critical data set for conducting a quantitative risk assess- ment for exposure to arsenic in drinking water. Dose-Response Modeling Model Choice and Selection of a Comparison Group In its proposed arsenic rule, EPA concluded, on the basis of the NEC (1999) report, that there is "no basis for determining the shape of a sublinear dose-response curve for inorganic arsenic" (EPA 2000a). Therefore, EPA estimated the risks of cancer from exposure to arsenic in drinking water using

QUANTITA TI BE ASSESSMENT OF RISKS USING MODELING APPROA CHES ~ 73 a linear extrapolation from the data observed in the southwestern Taiwanese epidemiological studies down to the origin. EPA's default to a linear extrapo- lation in the absence of adequate mode-of-action data (EPA ~ 996) is in part a policy decision. For the proposed rule, EPA used the bladder cancer risk estimates presented in the NRC (1999) report (see Table 5-1 for examples). EPA cited a lifetime risk estimate with a 95% upper confidence limit of ~ to ~ .347 per 1,000, calculated by a Poisson regression mode] not using any base- line data (i.e., no comparison group) (NRC 1999), and EPA used four distribu- tions of risk estimates (mean and 95% CT) from NRC (1999) as representative risks in a Monte Cario analysis to estimate the potential health benefits from the proposed rule. Those four distributions all come from analyses of the southwestern Taiwanese data (Chen et al. 1985; Wu et al. 1989) using a Pois- son regression model with age entered as a quadratic function and dose en- tered as a linear function, either with or without baseline data, or a Poisson regression mode] with a point-of-departure approach, with or without baseline data. On October 20, 2000, EPA published a Notice of Data Availability in the Federal Register (EPA 2000b) in which it discussed statistical modeling published by Morales et al. (2000) and indicated that those analyses would be considered in its final rule for arsenic in drinking water. Morales et al. (2000) estimated bladder, lung, and liver cancer risks for the southwestern Taiwanese population based on the same data set that was analyzed by NRC (1999~. Morales et al. (2000) calculated cancer risk estimates using 10 risk models and also considered how well those models fit the data sets. Of those models, EPA chose a single model that did not use an external comparison population either from all of Taiwan or part of southwestern Taiwan, because most of the models that incorporate a comparison population result in a dose-response curve that is supralinear at low doses. EPA indicated that there is no biologi- cal basis for a supraTinear curve. In addition, differences other than arsenic exposure between the study population and the comparison population could affect the results. The decision to use a dose-response model that does not incorporate a comparison population agreed with the SAB's recommendation that the analysis should be conducted without a comparison group (see discus- sion below). Of the models that did not incorporate a comparison population, mode] ~ from Morales et al. (2000), in which "the relative risk of mortality at any time is assumed to increase exponentially, with a linear function of dose and a quadratic function of age," was used because it best fit the data based on the Akaike information criterion (EPA 2001~. However, EPA did not

~ 74 ARSENIC IN DRINKING WA TER: 2001 UPDA TE TABLE 5-1 NRC's Risk Estimates for Bladder Cancer Mortality from 1999 NRC Reporta Margin of Point of Risk at 50 ~g/L Exposure at Method ofAnalysis Departure, pub (X 1,OOO) 50 ALL Poisson model, linear dose, no background data Poisson model, linear dose, background data included a Estimated points of departure at the 1% excess risk level, corresponding margin of exposure at 50 ~g/L arid corresponding excess lifetime risk estimates at 50 ,ug/L for bladder cancer in males. Figures in parentheses are 95% confidence limits (lower for the point-of-departure estimates; upper for estimated risk at 50 vigil). Risk estimates are those predicted in Taiwan using U.S. ingestion rates. b Me point of departure represents an estimate or observed level of exposure or dose associated with an increase in adverse effects in the study population. An example of a point of departure is ail EDo~ c A margin-of-exposure analysis compares the levels of arsenic to which the U.S. population is exposed with the point of departure to characterize the risk to the U.S. population. The larger the ratio, the greater degree of assumed safety for the population. Abbreviation: EDo,, 1% effective dose. Source: Modified from NRC (1999~. 404 (323) 1.237 (1.548) 8.08 (6.46) 443 (407) 1.129 (1.229) 8.86 (8.14) publish the theoretical risk estimates on which it based its analyses in the Federal Register (EPA 2000a, 200 ~ ). The risk estimates that it presents (EPA 2001) are adjusted for the occurrence of arsenic in U.S. drinking water; con- sideration of such an adjustment is beyond the charge to this subcommittee. Because EPA did not present theoretical lifetime excess bladder or lung can- cer risk estimates, the subcommittee used linear extrapolation from the EDo~s presented in Morales et al. (2000) to estimate these risks at 3, 5, 10, and 20 ~g/L (Table 5-2~. Adjustments for Water Intake To estimate cancer risks associated with a given arsenic concentration in drinking water, assumptions must be made about water consumption in both

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES 1 75 TABLE 5-2 Theoretical Estimates of Excess Lifetime Risk (Incidence per 10,000 people) of Lung and Bladder Cancer at Various Concentrations of Arsenic in Drinlcing Water Based on EDo, Values from Morales et al. (2000)a Arsenic Bladder Cancer Lung Cancer Concentration (,ug/L) Females Males Females Males 3 1.2 .76 1.2 0.82 5 2.0 1.3 1.9 1.4 10 4.0 2.5 3.9 2.7 20 7.9 5.1 7.8 5.5 a Excess cancer risk estimates were calculated using the EDo,s estimated by Morales et al. (2000) using a model in which the relative risk of mortality at arty time is assumed to increase exponen- tially, with a linear function of dose arid a quadratic function of age (i.e., a multiplicative Pois- son linear regression); no external comparison population was used (see Model 1 of Table 8 from Morales et al. 2000~. Risk estimates are rounded to two significant figures. The Taiwar~- ese exposure per kilogram of body weight is assumed to be 2.2 times the U.S. exposure. the U.S. population and the study population. EPA estimated mean daily average per capita consumption of water by individuals in the United States is ~ L/person/day for "community tap water" and 1.2 L/person/day for "total water" (which includes bottled water) based on data from the 1994-1996 Continuing Survey of Food Intakes by Individuals (CSF~ (EPA 20003~. The 90th percentile is 2. l L/person/day and 2.3 L/person/day for community tap water and total water, respectively. Rather than only using a point estimate for its risk assessment, EPA conducted a Monte Cario analysis using the CFSU data to incorporate water intake. Those distributions take into account age, sex, and weight. EPA assumed that the Taiwanese consumed relatively more water per unit of body weight than Americans, estimating consumption of 3.5 and 2.0 L/day for men and women in Taiwan, respectively. As dis- cussed in the following section, EPA also added water consumption to account for water used in cooking in Taiwan. It should be noted that assumptions that increase the amount of arsenic consumed (~inking water and diet) by the study population reduce the "potency factor" or estimated slope of the linear dose-response function when applied to other populations, thereby decreasing the estimated risk in other populations. Conversely, underestimation of the

~ 76 ARSENIC IN DRINKING WA TER: 2001 UPDA TE actual arsenic intake in the study population increases risk estimates in other populations. Therefore, assumptions about total arsenic exposure in the study population can have a large impact on risk estimates. Adjustments for Dietary Intake of Arsenic The staple foods in the southwestern Taiwan region where the study popu- lation resided were rice and sweet potatoes. Those foods absorb a great deal of water when cooked. As part of its risk assessment of arsenic in drinking water, EPA (200 ~ ~ adjusted its lower-bound estimates to account for exposure to arsenic in food from cooking water. For that adjustment, EPA assumed that people in the study population eat ~ cup of dry rice and 2 pounds of sweet potatoes per day. To adjust for arsenic absorbed during cooking, EPA added ~ ~ of water consumption to the Taiwanese population. Therefore, in its analyses, EPA assumed that Taiwanese men and women consumed the equiva- lent of a total of 4.5 L/day and 3.0 L/day of water, respectively. Although EPA used a Monte Cario analysis to account for variability in U.S. water consumption rates, its analyses did not incorporate analogous variability in the Taiwanese water consumption rates. EPA also discussed the fact that the food in Taiwan contains more arsenic than the food in the U.S., even prior to cooking. NRC (1999) presented data indicating that individuals in Taiwan consume food containing inorganic arsenic at 50 Gay, compared with ~ O Gay for Americans. To account for the intake of arsenic from food, EPA multiplied the Tower-bound risk esti- mates by the fraction of arsenic consumed per kilogram contributed by drink- ing water (calculated by dividing the arsenic ingested from drinking water (pa/kg/day) by the total arsenic consumed from drinking water, cooking water, and food) (J. Bennett, EPA, personal commun. May 22, 2001). Adjustments for Mortality Versus Incidence EPA's dose-response assessment is derived from data on mortality from bladder and lung cancer in the Taiwanese study (Chen et al. 1985, 1992; Wu et al. 1989). Extrapolating the mortality-risk estimates calculated in the Tai- wanese population to the incident risks in the U.S. population requires an adjustment for the survival rate for bladder and lung cancer. EPA (2001)

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES ~ 77 noted that the Taiwanese people in the study population had Tow incomes and poor diets, and that "the availability and quality of medical care is not of high quality, by U.S. standards." Therefore, EPA assumed that the bladder cancer incidence was relatively close to the bladder cancer mortality in the Taiwanese study area. EPA calculated the survival rate for bladder cancer by considering the survival-rate data compiled by the World Health Organization (WHO) for bladder cancer in developing countries from ~ 982 through ~ 992 (lARC ~ 999) and by comparing the annual bladder cancer mortality and incidence for the general population of Taiwan in 1996. From those data, EPA concluded that "bladder cancer incidence could be no more than two-fold bladder cancer mortality; and that an 80% mortality rate would be plausible" (EPA 2001; page 7009~. Therefore, when calculating the bladder cancer cases avoided at a given MCL, EPA adjusted the upper bound by a factor of 1.25 to reflect the mortality for bladder cancer. With respect to lung cancer, EPA concluded that "because lung cancer "mortality] rates are quite high, about 88% in the U.S. [EPA 1998], the assumption was made that all Jung cancers in the Taiwan study area resulted in fatalities." OVERVIEW OF TlIE SAB'S REPORT ON THE 2001 RISK ASSESSMENT EPA charged the SAB to review the proposed arsenic rule (EPA 2000a) and to specifically review (1~ EPA's focus on inorganic arsenic as the princi- ple form of arsenic causing health effects; (2) the implications of exposure to natural arsenic through food; and (3) the appropriateness of EPA' s precaution- ary advice to use low-arsenic water in the preparation of infant formula. They also requested the SAB to address several questions related to treatment op- tions for arsenic in drinking water. On December ~ 2, 2000, the SAB issued a report on the proposed drinking- water regulation (EPA 2000c), responding to those questions on the scientific basis of EPA's health risk assessment and on the economic and engineering aspects of the final rule. The exposure assessment, costs, benefits, control technologies, and policy issues discussed by the SAB are beyond the charge to this NRC subcommittee and will not be discussed. The SAB's responses to EPA's three questions on the health effects of arsenic are discussed in this section. EPA addressed some ofthe SAB's comments in its final arsenic rule (EPA 2001~.

78 ARSENICIN DRINKING WATER: 2001 UPDATE Inorganic Arsenic As Principal Form of Arsenic Causing Health Effects The SAB pointed out that new data released since the ~ 999 NRC report indicated that inorganic forms of arsenic are not solely responsible for the toxic effects of arsenic (see Chapter 3 for discussion of new data). Because exposures to other forms of arsenic can produce health effects, the SAB rec- ommended that future risk assessments provide quantitative information on how the intake of inorganic arsenic is related to the concentration of arsenic metabolites in the urine and to bladder cancer. However, because the princi- pal forms of arsenic in drinking water are inorganic, the SAB believed "that it is appropriate for the Agency tEPA] to make "inorganic arsenic] its regula- tory focus." Implications of Exposure to Natural Arsenic Through Food The SAB concluded that on average, for the general U.S. population, ingestion of inorganic arsenic via foodwas considerably greater then ingestion of inorganic arsenic via drinking water. The SAB agreed that data were not available to determine "a well-defined nonlinear dose-response curve." Fur- thermore, the SAB concluded that insufficient data on the distribution of food intakes existed to adequately consider them in the analysis. Therefore, the SAB concluded that EPA had no choice but to calculate marginal risk reduc- lions based solely on arsenic concentrations in drinking water. The SAB noted that "there is a limit to the benefits that can be realized by reducing arsenic in drinking water" as long as the concentrations in food remain un- changed. The SAB also reiterated the recommendation in the NRC (1999) report to obtain additional studies on the noncancer effects of arsenic and incorporate that information into a risk-assessment and cost-benefit analysis. Health Advisory on Low-Arsenic Water and Infant Formula The SAB also reviewed EPA's plan to issue a health advisory for the use of Tow-arsenic water in the preparation of infant formula. Some of SAB's responses focused on risk communication and the public's ability to follow such an advisory; those comments are beyond the scope of this subcommit- tee's charge. However, the SAB also discussed the issue of children's suscep- tibility to arsenic. Most of the SAB members agreed that special circum-

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES ~ 79 stances make infants unique in regard to their response to drinking-water contaminants, but the SAB did not reach consensus on endorsing EPA' s intent to issue a health advisory. An SAB consultant wrote and one SAB member endorsed a minority report on the issue of infant and children's risk. The minority report stated that differences in respiratory and circulatory flow rates, cell-proliferation rates, enzymatic pathways, developmental process, life expectancies, and the disposition of chemicals in infants and other differences in cells during devel- opment can make children more susceptible to toxic chemicals, including arsenic. It further concluded that data from Hopenhayn-Rich et al. (2000) and Concha et al. (1998) "indicate that young children are a uniquely sensitive population for adverse health effects of arsenic." The minority report "departs from the majority opinion contained in the iSAB] report in strength of its conclusions if not the general reasonableness of the need for increased con- cern for children, which is also held by the tSAB]." The basis for increased concern for children is uncertainty about pulmonary and cardiovascular risks to infants, high exposure of infants on a per kilogram basis, and the longer period of exposure and outcome relative to adults. The latter is particularly relevant if the latency period for cancer development from low arsenic expo- sure is long or if the appropriate dose metric involves a less-than-lifetime exposure as discussed in Chapter 4. Although the SAB recognized that children differ from adults in many ways that could make them more susceptible to toxic chemicals, "the majority of the tSAB] did not feel that data available to them on arsenic had demon- strated an increased sensitivity to arsenic in children." The SAB concluded that available data on U.S. drinking-water consumption indicate that infants who consume food made from drinking water could have a higher dose of arsenic per unit of body weight than adults. The majority ofthe SAB did not believe that the study by Hopenhayn-Rich et al. (2000) demonstrated "a heightened sensitivity or susceptibility to arsenic" but that the study "appears to be a hypothesis generating study that, in light of the limitations [the SAB described], merits and requires further study before drawing final conclu- sions." SAB's Comments on EPA's Interpretation of the NRC Report The SAB also commented on EPA's interpretation of the 1999 NBC report. The SAB, in general, agreed with the ~ 999 NRC report, which formed

~ 80 ARSENIC IN DRINKING WA TER: 2001 UPDA TE part of the basis for EPA's proposed regulation. The SAB reiterated some of the cautions made in the 1999 NRC report surrounding the use ofthe ecologi- cal Taiwanese data to characterize risks to the U.S. populations, including (1) the potential for measurement error and confounding because of uncertainty in exposure assessment; (2) the impact of the choice of model; and (3) other factors "such as poor nutrition and low selenium concentrations in Taiwan, genetic and cultural characteristics, and arsenic intake from food." The SAB stated that EPA "may have taken the modeling activity in the NRC report as prescriptive." It cited a study by Morales et al. (2000), published after EPA published its proposed regulation (EPA 2000a), which could impact its risk assessment. The SAB also concluded "that the comparison populations [used in some analyses] were not appropriate control groups for the study area" in the Taiwanese study and, therefore, recommended dose-response models that did not use the comparison population. Following this recommendation, EPA used the Morales et al. (2000) study and analyses conducted without the use of a comparison population in its final rule (EPA 200I).' The SAB did not believe, however, that "resolution of all these factors can nor must be accomplished before EPA promulgates a final arsenic rule." Furthermore, the SAB agreed that the "available data do not yet meet EPA' s new criteria for departing from linear extrapolation of cancer risk." THE SUBCOMMITTEE'S EVALUATION This section critiques various elements that influenced EPA's decision to propose a new MCL for arsenic of 10 ,ug/L (ppb) of drinking water based on what was known at that time and on new information that has emerged since publication of the ~ 999 NRC report. Issues to be addressed include choice of end point (bladder and lung cancer), and the use of the southwestern Taiwan- ese study as the basis for EPA's risk assessment. This section will also cri- tique and discuss a number of the statistical modeling assumptions used by EPA, including (~) the inclusion of an external comparison population in the analysis, (2) the choice of dose-response model, and (3) the approach used to ~ The SAB report stated that it "focused on evidence provided in the NRC ~ 999 report that indicated that the population in the study area differed substantially from these comparison populations socioeconomically and in diet." (EPA 2000c, p. 26~. However, it did not specifically identify any part of the NRC report that compared nutritional or socioeconomic status between the study population and the southwestern Taiwanese referent population.

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES 181 address the impact of assumptions about drinking-water intake in the study population. Choice of End Point and Study EPA's decision to base its risk assessment on the bladder and lung cancer mortality data from the southwestern Taiwanese studies (Chen et al. 1985, 198S, 1992; Wu et al. 1989, as analyzed by Morales et al. 2000) was largely in response to the recommendations in the 1999 NRC report. Arsenic has been implicated in a variety of adverse health effects, including dermal ef- fects, gastrointestinal effects, peripheral vascular disease, diabetes, and can- cer. Cancer, however, is the end point most consistently seen as a conse- quence of long-term chronic exposure. It was also the end point with the most extensive quantitative information available on dose-response. Although several studies that were available at the time of the 1999 NRC report con- vincingly confirmed the carcinogenic effects of arsenic (e.g., Hopenhayn-Rich et al. 1996, 1998; Smith et al. 1998), the southwestern Taiwanese studies documenting bladder and lung cancer (Chen et al. ~ 985, ~ 988, ~ 992; Wu et al. ~ 989) were the only ones that quantified exposure levels well enough to sup- port a quantitative dose-response analysis. Since the 1999 ARC report, con- siderable new information has emerged related to the health effects of arsenic in drinking water. Cancer, however, remains the end point with the most reliable data available for the purpose of quantifying dose-response effects (see Chapter 2~. Although the southwestern Taiwanese studies still stand out as the strongest sources of dose-response information, two recently published studies (Ferreccio et al. 2000; Chiou et al. 2001) are of sufficient quality to warrant serious consideration as part of the quantitative risk assessment for arsenic in drinking water. Both studies are described and critiqued in detail in Chapter 2. They are summarized briefly here, and their potential for use in a quantitative risk assessment is also discussed. In a prospective cohort study, Chiou et al. (200 I) assessed arsenic expo- sure at the individual level for over 8,000 subjects in northeastern Taiwan. Drinking-water arsenic concentrations ranged from below detection to 3,843 ,ug/L of drinking water, with the majority of study subjects (75%) exposed at concentrations less than 100 A/. More than half of the subjects were ex- posed at drinking-water concentrations less than 50 ,ug/L, and 2,346 consumed drinking water with no detectable arsenic (less than 0.15 ,ug/~. Data are also available at the individual level on potential confounding factors, such as

~ 82 ARSENIC IN DRINKING WA TER. 2001 UPDA TE smoking and socioeconomic status. The incidences of urinary-tract cancer overall and the more specific urinary-tract transitional-cell carcinoma (TCC) are the end points of interest. Although the study involves a relatively large number of subjects, because of the relatively short follow-up period (approxi- mately 4 years), the number of cancers observed is relatively Tow (~8 urinary cancers, of which 11 were classified as TCC). Ferreccio et al. (2000) conducted a case-control study in an area of north- ern Chile, where 15 ~ lung cancer cases and 419 controls were studied. Aver- age exposures to arsenic during the period 1930 to 1994 were grouped as 0-10 ~g/L (referent group), 10-29 ,ug/L, 30-39 ,ug/L, 40-49 ,ug/L, 50-199 ~g/L, and 200400 ,ug/~. There is individual information on residential history, socio- economic status, occupational history, and smoking. Another strength ofthat study is that it has detailed, individual-specific exposure assessment; arsenic exposure was determined for each subject by residence and historical arsenic concentrations on a yearly basis from 1930 to 1994. Drinking-water arsenic concentrations were very high (860 ,ug/~) in the major city of the study area, Antofagasta, from 1958 to 1970. Risk calculations based on exposure during this period were somewhat lower than risk calculations based on average exposure from 1930 to 1994. Quantitative Risk Estimates The usefulness of the Chiou et al. study to quantitative risk analysis are limited by its follow-up period compared with that of the studies from south- western Taiwan (Chen et al. 1985, 1992; Wu et al. 1989~. It is useful, how- ever, to consider a qualitative synthesis of the results from the different stud- ies and end points to assess the concordance of the estimates of the effective doses under different modeling assumptions. In this section, 1°/O effective dose LEDGE calculations for the two recent studies are discussed (Table 5-3~. In the next section, some of the modeling assumptions and choices that can influence ED calculations are described, and finally, a summary ofthe subcom- mittee's findings is presented. EPA based its risk assessment on EDo~s calculated by the NRC (1999), and also by Morales et al. (2000~. The Educes were based on Poisson regres- sion models. Although simpler methods could also be applied, the subcom- 2The 1% effective dose is the concentration of arsenic in drinking water in the study that is associated with a 1% increase in the excess risk of cancer.

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES ~83 mittee thinks that the general Poisson approachis appropriate. Poissonregres- sion does not impose any strong model assumptions. It is simply a probabilis- tic formulation for handling the analysis of rare events. As such, it is appro- priate for the analysis of cancer data collected in a cohort study setting (see Bresiow and Day (1988) for related discussion). Modeling assumptions play a critical role, however, when it comes to the choice of the parametric form relating covariates to the rate parameter of interest (in the present context, cancer mortality). Although the standard formulation of Poisson regression involves modeling the rate parameter of interest as an exponential function of covariates, other formulations are possible with specialized programing. In the following sections, the subcommittee considers models that charac- terize cancer mortality as a baseline function of age (and possibly other covariates) multiplied by a relative function, gall, where a' is dose. The sub- committee considers two models of relative risk. One ofthese, the multiplica- tive model, assumes that the logarithm of good can be expressed as a linear function of exposure. This model can be fit using standard statistical software for Poisson regression. The second formulation, the additive model, assumes that grail is simply a linear function of dose. (This formulation requires spe- cialized programing.) Although a variety of other formulations could be considered, relative risk models have a lengthy precedent in cancer epidemiol- ogy and there are good reasons to believe that relative risks are similar across populations, even when baseline risks vary or synergy is operating (e.g., as might possibly be the case for smoking and arsenic exposure). The formula used by the NRC (1999; also see Morales et al. 2000) for computing an EDIT based on the results ofthe Poisson regression analysis had some limitations; the primary one being that it was not straightforward to incorporate the baseline cancer risk based on the U.S. population. A better approach is to base the ED calculation on the formula presented in Appendix of the BEIR IV analysis of Jung cancer mortality associated with radon exposure (NRC ~ 988~. Given a specified simple functional form (e.g., linear) for the relative mortality risk associated with exposure to arsenic in drinking water, the formula can be used to compute a lifetime risk estimate, and hence an ED, based only on the value of the relative risk at a single specified value of exposure (Gail ~ 975~. This property is particularly advantageous for com- puting EDs in settings where raw data are unavailable but where estimates of relative risks and their associated confidence intervals (CIs) have been pub- lished (e.g., lung cancer odds ratios reported by Ferreccio et al. 2000~. The BE1R IV formula can be applied with either incidence or mortality. In the following section, the application of BEIR IV using mortality is considered; later in the chapter, its use with incidence is considered.

184 cd o · _ U' U. ¢ a' To ,~ ._ n On .O ._ a Ct Cal Ct v CC o a an TO OF - o a of Ct a ^ ~ V ~ onto ~' cq - Ct an a .= - a' ._ a an ~ ~ ~ ~ ~ + + ~ + ~ ~ ~^ ,^ T~^ ~ 0^ ~ ~ ~ ~ ~ ~ ~ ~ ='—~ 0——= _~—= =—==—— _ ~ ~ oo _ ~ _ — ~ t— —~ ~} tS, ~ oo c~] + + + + _~ _,~ _~ _~ _~ + _~ _~ _~ _~ _~ _ - ~ —~ O O O O ~ 0N ~ ~ ~ O ~ ~ ~t o _ _ =\ _ O O O O ~ 0- ~ ~ O O —~ d"—) d"—, ~ oo _ ~ ~ ~ ~r, er~ ~ ~ ~ ~ ~ ~ ~ ~ _ ~ _ ~ _ O O ~ ~ ~ ~ ~ ~ O ~ ~ . . . . . · ~ · ~ OX ~] _ ~4 ~ _ ~ ~ _1 —) —} ~ _ _ _ ~ - , ~ o0 oo oo O — O O ~ oo oo ~ _ oo ~ ~ _. _ o o O O ~ O O _ _ _~ _~ ~ C ) C~ ~t — O — ~t C'J ~ d- . . . . . . . . . . . ~ - - - ~ - ~ - - ~ ~ ~ - 4 ~o a~ o ~ ~ o ~ ~ ~ o · ~ ~ - v cn ~ u, ~ ~ 4_ c-= c-= ~ ~ ~ 3 ' ~ 3 ~ j o ~ ~ , ~ ~ ~ ·U, ~ U) ·U:} ~ ~ ~ ~ ~ ~ ~ .s >~y =f ~ ~ ~ ~ ~ ~ ~ .O .O ~ .O .O ~ .O .O ~ _ _ _ _ ~ _ _ ~ _ _ ~ ~ O5 ~ S ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ = ·= Ct Ct o ._ ~ ~ O .0 0 oo ~o = - V~

185 ~ a' ~ _ _ _ ^ ~ ~ ~ C%] ~ \~ ~ ~ _ Boo ~ ~ o ~ ~ _ ~ ~ ~ oo ~ ~ _— i_ _' _' _' _ `_ _' _' `_ `_ _' t en, ~ ~ _ oo d- A, oo d. ~ ~ — ~ O oo ~ ~ ON ~ ~ ~ ~ _ ~ ~ _ us us cn cat O ~ O ~ O an ° ~ U) ~ Cat ~ U. ~ U. ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ =° (V .= o ~ .= to0= ~ o== ~ == _ _ ~ _ _ , - _ _ ~ _ _ ~ .~ .~.8.~. .8.- .~.8.~.~.8 C~3 Cq! C<~ C;~ C<! C<~ Ct C\t — .O .O a~ .O .o ~ .O .O ~ .O .O ·= ·= :~ ·= ·~: - · - ·~ : - · - · - :~ 5 00 a~ C~ ~ m . ~ · Ct ~ Cq! au ~ a., ~ ~ Ir,^ v ~ v _ ~ ~ _ U~ S~ U~ U) C ~ .O ~ ~ ,0 .~ C CS ~ ~ ~ ~ O_ ~ ._ ,' X ~ .= ~ ~ c., ,,, s E -_ ~ ~ e o o ~ o ~ 3 . C: ~ . ~ ~ ~ ,= D ~— E ~, ° , o e ~ S '_ ~ u, ~ U' ,~ cq ~ (L) E s ~ ~ ~ C ° ~ <,, 2 ° ~ ~ o _ 0 ~ ~ o ~ c "g° ~ ~ 2 ~ ° o 3 ~ ~ ~ .= ~ ._ Cd ~ ~ ~ ,~: ~ ~ ~ o o - U) ~) ~ ~ ~ ~ ~ ~·s ° ', ~,4 ° 3 o ~ ° ~ '= ~ ° ~ °N t ° ''^ B .,,.~: o° ° ~ 2 C, ~ ~ ~ ~ ~ _ ~ _ ca & E .' & 2 & & E s ~ 2 o ~ ' ~ ~ ° ~ 2 s

86 ARSENIC IN DRINKING WA TER: 2001 UPDA TE BEIR IV To describe how the BEIR {V formula can be used, it is necessary to specify some notation. Let hi be defined as the hazard for cancer death in age interval i. Similarly, hi* is defined as the corresponding overall hazard of death. We define qi to be the conditional probability of surviving the ith time interval, given survival to the beginning of the interval, qi = exp(-S x hi) Finally, Si is the probability of surviving to the beginning of the ith interval, Si=~~qi) j=! Then the lifetime cancer mortality risk (Ro) for an unexposed individual is Ro = ~ hi/in, Si (l - qi) ~ ~ age groups (~2) (~3) The lifetime cancer mortality risk for an individual exposed to an arsenic concentration of a, ,ug/L (Rd) can be obtained by replacing hi with highly, where gig) is the relative risk associated with that exposure concentration, and the overall mortality hazard for an individual exposed at a dose, al, is h' + (it) - I)hi (~4) The ED is obtained as the solution to Rd - Ro = 0.01. Data on overall U.S. population deaths and cancer-specific deaths were based on the vital statistics records for 1996 obtained from the National Center for Health Statistics (GMWK I table for deaths). Bladder cancers were those corresponding to ICD9 code 188; Jung cancers were those corresponding to ICD9 code 162. As can be seen from the above discussion, the BEIR IV approach uses relative risks from the epidemiological studies anal estimates EDo~s for the U.S. population using baseline risks for the U.S. population. In contrast, in

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES ~7 NRC (1999) and Morales et al. (2000), where the BEIR IV approach was not used, the EDo~s were estimated using Taiwanese baseline risks. That differ- ence in the two approaches affects the resulting EDo~s when the baseline risk in the two countries differs. If the baseline risk in the U.S. population is higher than that in the Taiwanese population, then the EDo,s determined using the BEIR IV approach would be lower than those determined using the alter- nate approach. The subcommittee's comparison of recent data on lung cancer incidence in the United States (Feriay et al. 2001) with Jung cancer incidence in Taiwan (You et al. 2001) indicates that lung cancer incidence is approxi- mately three times higher in the United States than in Taiwan for females and two times higher for mates. For bladder cancer, the incidence is approxi- mately three times higher in mates (Feriay et al. 2001; You et al. 2001) and two times higher in females (FerIay et al. 2001; You et al. 2001; C.~. Chen, National Taiwan University, personal cornmun., August 2S, 2001~. The dif- ferent baseline incidences result in a Towering of the EDo~s calculated using the BEIR IV approach relative to the EDo~s calculated applying the approach used by NRC (1999) and Morales et al. (2000~; the corresponding risk esti- mates are increased. It is interesting to note that when hi is small relative to hi*, then the ap- proach described above using additive Poisson regression and the BEIR IV approach yields results very similar to those resulting from the approach rec- ommended by Smith et al. (1992~. Although not as accurate as the Poisson modeling approach, the approach used by Smith et al. (1992) has the advan- tage of being simpler and less computationally burdensome. In that approach, given the lifetime cancer mortality in an unexposed population (Ro in the notation used here), along with the relative risk, good, then the EDo, is the solution to Ro x (~`l) - 1) = 0.01. This simple formula, in fact, can be used as an approximate check of the results based on the BEIR IV formula pre- sented in Table 5-3. For its results in Table 5-3, the subcommittee used cancer mortality data reported by the National Center for Health Statistics for 1996 (CDC 2001~. Baseline risks (Ro) for lung cancer were 0.076 and 0.046 for mates and females, respectively. The corresponding risks for bladder cancer were 0.007 and 0.003. Calculations of cancer risk estimates presented later in the chapter use the simple approximate formula (Ro x (gyp - 1) = 0.01) to derive EDo~s with respect to lifetime cancer incidence rather than mortality. Chiou et al. (2001) reported relative risks associated with the incidence of all urinary tract cancers (ICD-9 codes 188 and ~ 89) and specific data for TCC (ICD-9 codes 8120.2, 8120.3, or 8130.3) grouped in several exposure

~ 88 ARSENIC IN DRINKING WA TER: 2001 UPDA TE categories, includingO-IO.O ~rg/L, 10.1-50.0 ,ug/L, 50.~-100.0 ,ug/L, andgreat- er than 100.0 ,ug/L. Treating the O-10-)lg/L group as baseline, the authors report relative risks (based on a multivariate mode] that included a linear effect of age, as well as a gender and smoking effect) ranging from ~ .5 for the lO-SO-,ug/l~ group to 15.3 for the greater than1OO-,ug/L group. The relative risk in the lowest group was not statistically different from ~ and had a very wide CT of 0.3 to 8.0. The authors, however, have made the data available to the subcommittee, allowing exploration of some dose-response models based on the ungrouped exposure variable. Exploratory analysis reveals consider- able model sensitivity, which is not surprising given the relatively small num- ber of cancers seen in the short observation time of study (15 urinary tract cancers, including 10 TCCs, in the 4 years for which exposure information was available). The dose-response pattern is quite similar to that seen in the data from southwestern Taiwan, with a somewhat supralinear pattern at low doses and flattening out at high doses. The best fits were provided by a mode! with a Tog transformation of dose (including smoking, age, and gender, as described by Chiou et al. 2001) and an additive mode] with linear dose. As summarized in Table 5-3, EDGY estimates based on the additive linear model tended to be higher than those based on the multiplicative model; that pattern is consistent with the bladder cancer results obtained from the southwestern region. When the models were rerun restricting the study to subjects exposed to less than 400 ,ug/L and again to subjects exposed to less than 200 ,ug/L, there was a much higher concordance between the various models. The BEIR {V (NRC 1988) formula described above was used to compute Edgers. To apply those EDo~s to the U.S. population, the fact that the Chiou et al. (2001) analysis is based on cancer incidence, not mortality, needs to be addressed. For simplicity, it is assumed here that the relative risks are the same for cancer mortality and incidence. That assumption is reasonably accurate in the case of lung cancer, which has a very high case mortality. It might be less accurate for bladder cancer. Assumptions about possible differences between the United States and Taiwan with respect to drinking-water rates are also neces- sary. It was assumed that the Taiwanese, Chilean, and U.S. populations all drank the same amount of water but that a person in the United States weighs 70 kg and a person in Taiwan or Chile weighs 50 kg. Those assumptions result in a conversion factor of 1.4 between the United States and the other countries (see further discussion below). Ferreccio et al. (2000) report odds ratios for lung cancer associated with arsenic exposure through public drinking-water supplies for Chile. The au- thors report their results under several different modeling approaches and

QUANTITA TI VE ASSESSMENT OF RISKS USING MODELING APPROA CHES I 89 choices of control group. The odds ratio associated with the 30-59-pg/L group (based on a multivariate model for the peak-period dose metric that adjusted for age, gender, smoking, and occupational exposure) was i.8 (95% CI = 0.5-6.9~. To obtain a more precise estimate using all the available data, the subcommittee applied a simple linear regression line to the odds ratios listed in Tables 5 and 6 of Ferreccio et al. (2000), using the midpoint of each exposure interval and forcing the line through unity for the lowest exposure group. The analysis yielded an estimated odds ratio of approximately 2.4 ~sso/O cr = 1.9-2.9) based on the data in Table 5 of Ferreccio et al. (2000) and an estimated odds ratio of approximately 1.4 (95% CT = 1.3-1.5) based on the data in Table 6. No adjustments were made for drinking-water intakes and body-weight differences between the Chilean and U.S. populations. The estimated risk ratio of ~ .4 translates to an EDo, of ~ 7 ~g/L for males and 27 for females. Because the values in Table 6 of Ferreccio et al. (2000) are based on average arsenic exposure levels during the high exposure period (1958- ~ 970) and Table 5 uses exposure levels averaged over the entire study period (1930-1994), the reported odds ratios are higher in Table 5 than in Table 6. Table 5-3 shows the risk ratios associated with exposure to 50 ~g/L for the two studies discussed above (Ferreccio et al. 2000; Chiou et al. 200 I) and corresponding EDo, estimates. Results are also included for the southwestern Taiwanese data previously reported by Chen et al. (1985) and reanalyzed by Morales et al. (2000) but now recalculated using the BETR IV lifetime-risk formula. The ranges reported along with each risk ratio generally correspond to 95% CTs obtained from fitted models except where risk ratios were taken from published papers, in which case, they refer to feasible ranges obtained by smoothing or extrapolating the published results. For example, in the case of the Ferreccio et al. (2000) data, the subcommittee applied linear regression to the odds ratios reported in the paper and computed confidence limits based on the fitted regression line. It should be emphasized that such calculations are somewhat imprecise as a result and should mainly be used for qualitative comparisons and for providing information on the range of possible EDo,s. Statistical Analyses and Dose-Response Modeling In addition to the issue of study selection for an arsenic risk assessment, questions have arisen regarding statistical modeling and sources of uncertainty that might affect the ED calculation in EPA's risk assessment. Those issues are discussed below.

~ 90 ARSENIC IN DRINKING WA TER: 2001 UPDA TE Mode] Choice Use of a Comparison Population The 1999 NRC report recommended that calculation of EDs be based on a Poisson regression model or SMR approach applied to the southwestern Taiwanese data (Chen et al. ~ 985, 1992; Wu et al. 1989~. A variety of feasible mode! formulations were presented (see ~ 999 report for dose curves), and it was recommended that those models be considered and compared in some type of sensitivity analysis. As discussed earlier, the data comprised age- and gender-specific cancer mortality data from 42 villages in the region of south- westem Taiwan where arsenic is endemic. The 1999 NRC report discussed two possible approaches to the dose-response analysis: (~) analyzing only the 42 villages and using the variation in cancer rates from vilIage-to-village to determine the nature of the estimated dose-response relationship (using an intemal comparison group); and (2) incorporating data from an external com- parison population, for example, nationwide data. The latter approach is classically used in the analysis of cohort data (Bresiow and Day 1988) and has the advantage of providing a much more accurate estimate of the baseline cancer rates. Use of an unexposed external comparison population also mini- mizes the impact of exposure misclassification in the Tow-dose range within the study population. A potential disadvantage, however, of using an external comparison group is that the analysis can be biased if the study population differs from the comparison population in important ways. Such issues have been discussed extensively in the context of occupational cohort studies, in which, for example, there is often reason to believe that a population of healthy workers might differ from the general population. In the case of the ~ 999 report, which used a linear dose multiplicative Poisson model, it turned out that ED calculations were very similar, regardless of whether a compari- son population was used. Subsequent to the 1999 N~C report, however, Morales et al. (2000) expanded the modeling options beyond those considered by the previous Subcommittee on Arsenic in Drinking Water (NRC ~ 999) and found considerable model sensitivity. In addition to including exposure as a linear term in the model, Morales et al. (2000) also considered Tog and square- root transformations and fit additive versions of the dose-response model in addition to the multiplicative models discussed in the 1999 report. As in`di- cated in Table ~ of Morales et al. (2000), EDGE estimates from the best-fitting models ranged from 250 to 400 ~g/L for male and female Jung and bladder

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES ~91 cancer, based on models without an external comparison group. The 10- to 150-,ug/L range was based on models that used the entire Taiwanese popula- tion as a comparison group. In general, estimated EDo~s tended to be lower for models that included an external comparison population, primarily because the lung and bladder cancer rates in the comparison populations were much Tower than the rates seen even in the study villages with Tow exposures. Con- sequently, the fitted dose-response models tended to be steeper (Morales et al. 20004. Furthermore, when comparison populations were included, models based on a log or square-root transformation of dose tended to fit better statis- tically, since they allowed the dose-response curve to take on a supralinear shape involving a steep initial slope, followed by a flattening out (see Figures 5-la-c). As discussed earlier, EPA based its ED on the linear multiplicative mode] without a comparison population. The SAB recommended this choice primar- ily because the results based on the no-comparison-group models were more stable (i.e., less sensitive to model choice), and the comparison-group models resulted in extremely low ED estimates, which did not seem heuristically reasonable. Further justification for the choice was provided by the argument that the poor rural areas represented in the study population would not be similar to the comparison populations, which were weighted heavily toward the more prosperous urban areas in Taiwan. This subcommittee carefully considered the issues surrounding the use of a comparison population and ultimately concluded that a comparison popula- tion should be used. Rationale for this conclusion included the following: · It has been argued that the area of southwestern Taiwan where arsenic is endemic is different from the whole of Taiwan in important ways (other than arsenic in the drinking water) that could affect cancer death rates; hence, national rates were inappropriate to use as a standard referent when calculat- ing SMRs. That argument is not supported by a recent paper by Tsai et al. (1999~. As discussed in Chapter 2, those authors demonstrate that SMRs for the area where arsenic is endemic based on regional population rates are simi- lar in magnitude to SMRs based on rates for the national population. (The vest majority of individuals in the southwestern Taiwan referent region are not exposed to increased arsenic concentrations in drinking water.) Therefore, the use of an external comparison population is unlikely to introduce significant confounding. ~ Available dose-response data cannot distinguish between linear and nonlinear models.

92 ARSENIC IN DRINKING WA TER: 2001 UPDA TE Indeed, one ofthe reasons cited for not using models that incorporated a comparison population is that they tend to produce supraTinear dose-re- sponse curves, when it has been argued that the mechanistic data point to a sublinear model. As discussed in Chapter 3, however, the mechanistic evi- dence for a sublinear dose-response relationship do not necessarily apply at the population level. Furthermore, the fact remains that there is some empiri- cal evidence suggesting that a supralinear model might indeed hold. For example, statistical goodness-of-f~t criteria applied to both the southwestern Taiwanese data (Chen et al. 1985, 1988, 1992; Wu et al. 1989) and the recent data from northeastern Taiwan (Chiou et al. 2001) support models based on a log transformation of dose, leading to supraTinear models. The data from other published papers point to similar patterns. The study by Ferreccio et al. (2000) also yielded some evidence of a supralinear dose-response relationship. A supralinear dose-response relationship is plausible; for example, it could result from a subpopulation that is susceptible to low doses of arsenic. Another possible reason for an apparent supraTinear curve is dose misclassification. The exposure to high concentrations of arsenic in food (i.e., on the order of 50 ~g/day) would result in a substantial increase in the dose of arsenic received by people in villages with low concentrations of arsenic in drinking water. The impact of the added arsenic would be relatively much greater for individuals with Tow concentrations of arsenic in drinking water than for individuals with high concentrations of arsenic in drinking water. If the true total dose of arsenic in food and water were plotted against the re- sponges, the curve might no longer be supraTinear. Dose misclassification might also result from the movement of people among the different villages. For example, even though a person might live in one particular village, it is logical to think that he or she might visit neigh- boring villages on occasion and drink the water there or eat food grown in other villages. The 1999 NRC report also discusses the fact that some ofthe villages had multiple wells with high variation in measured arsenic concentra- tions. Tfthe villages assigned low exposure status actually contained a signifi- cant number of individuals with higher exposure, such misclassification bias would result in a substantial underestimation ofthe slope ofthe dose-response relationship in models limited to the study villages. Results based on models that include an external comparison population are likely to be less affected by such measurement error, because the model fit is anchored by a large con- troT group.

QUANTITATIVEASSESSA1ENT OF RISKS USING MODELING APPROACHES 193 A 0.10- 0.08 by In Al 0.06 _ := 0.04 0.02 0~ B 0.10 0.08 - <,, 0.06 .E := ._ 0.04 - 0.02 - 0.00- C 0.10- 0.08 - y In - 0.06 - a) Model 1 - Model 2 MSW . :~ · ~ l l l l l 0 500 1,000 1,500 2,000 U.S. equivalent concentration (pglL) - - Model 3 --- Model 5 . Model 6 Model 9 MSW Taiwanese population . . 0 500 1,000 1,500 2,000 U.S. equivalent concentration ()lg/L) 0.04 - 0.02 - — Model 4 -- - Model ~ --- Model 7 Model 8 —MSW Southwest Taiwanese population 0.00- l l l 0 500 1,000 1,500 2,000 U.S. equivalent concentration (pg/L) Figure 5-1 Estimated lifetime death risk over background rates in Taiwan for male bladder cancer (A) without comparison population, (B) with Taiwan-wide comparison population, and (C) with southwestern Taiwan comparison population. Source: Mo- rales et al. (20001.

~ 94 ARSENIC IN DRINKING WA TER: 2001 UPDA TE Functional Form of Dose-Response Relationship As discussed in the previous section, when an external comparison popu- lation was included, the ED estimates from the analyses of the southwestern Taiwanese data used by EPA in setting its proposed standard varied depending on the functional form of the dose-response model being fitted. According to a statistical measure of model fit, the Akaike information criteria (AIC) (see Hastie and Tibshirani ~990) for discussion of model-fit criteria), statistically superior fits were produced using models that included a log or square-root transformation of dose (Morales st al. 2000), despite the fact that those models are not as biologically plausible as others. Note that for ATC values a Tower number indicates a better fit. Table 5-4 shows ATC values for some of the models shown in Morales et al. (2000) and other models, including one with a quadratic term in dose and one with an interaction between dose and age. Models based on both a multiplicative relative risk and an additive relative risk are presented. The quadratic dose-response model, which was not investi- gated by Morales et al. (2000), also resulted in a relatively good fit. ~ addi- tion to the AIC values, Table 54 includes a column of Bayesian posterior model probabilities (PMPs) (see Carlin and Louis ~ 996 for discussion). PMPs are computed by starting with an initial assumption that all models under consideration are equally likely and then computing the conditional probabil- ity that each one is true, given the observed data. The effect ofthese posterior probabilities is to downweight models that do not fit well and emphasize models that do. As can be seen Table 54, the PMPs tend to correspond (in- versely) reasonably well with the ATC values (i.e., the PMPs are high when AlCs are low, and visa versa) but have the important advantage of being easier to interpret. Looking at the PMP values suggests that the additive models, using either a linear or log dose, or the multiplicative mode] with quadratic dose best fit the data. Only the results for mate lung cancer are shown in Table 54. Similar patterns emerged for female lung cancer, and male and female bladder cancer. in light of considerations of biological plausibility and consistency with other recommended approaches to quantitative dose-re- sponse assessment (Smith et al. ~ 992), the subcommittee recommends the use of the additive Poisson model with dose entered as a linear term. In a recent doctoral dissertation, Morales (2001) used the technique of Bayesian model averaging (Carlin and Louis 1996) to compute an EDDY that combines the results from a set of models, such as those listed in Table 5-3. The technique can be useful in situations, such as the one for the arsenic data, where several different models are feasible. Another advantage of this tech-

QUANTITA TI BE ASSESSMENT OF RISKS USING MODELING APPROA CHES ~ 95 TABLE 5-4 Akaike Information Criteria Values and Bayesian Posterior Model Probabilitiesa AIC PMP Relative Risk Model Model Descnption Value Value 1 + ,B d Additive, linear dose 425.8 0.152 exp(,BO + ,B, d) Multiplicative, linear dose 435.9 0.000 Multiplicative, quadratic 425.1 0.432 exp(pO + ,D, d + p2 d^2 ~ dose exp(pO + ,B~ d + p2 d*t) Multiplicative, interaction 437.6 0.000 between age and dose 1 + ,B logged) Additive, log dose 424.3 0.413 exp(,BO + p~ logged)) Multiplicative, log dose 431.1 0.003 exp(,BO + p~ logged) + p2 Multiplicative, interaction 432.7 0.000 log~d)*t) between age and log dose a Values are based on male lung cancer data from southwestern Taiwan (Chen et al. 1985, 1992; Wu et al. 1989), with southwestern Taiwan serving as the comparison population. All models include a quadratic effect of age. Models are distinguished by their assumed relative risk associated with an exposure level, d. Abbreviations: AIC, Akaike information criteria; BPM, Bayesiar~ posterior model probabilities. nique is that it appropriately quantifies the variability due to model uncer- tainty. Because it is a non-standard statistical method that has not been exten- sively evaluated in the context of environmental risk assessment, the subcom- mittee does not recommend it as the basis of the arsenic risk assessment. Impact of Dietary Intake of Arsenic Some attention has been focused on the issue of differences between the United States and Taiwan with respect to dietary intake of arsenic. As dis- cussed earlier, it has been suggested that effective exposures in Taiwan are higher than represented simply by arsenic concentrations in well water, be- cause (1) the Taiwanese people are exposed to additional arsenic through

~ 96 ARSENIC IN DRINKING WA TER: 2001 UPDA TE water used for cooking owing to the predominance of rice and dried sweet potatoes in the diet; and (2) people in the region of Taiwan where arsenic is endemic are exposed to a high general background level of inorganic arsenic in their food (50 ,ug/day in Taiwan compared with 10 ,ug/day in the United States) (NRC 1999), although there is substantial uncertainty and variability in both of those estimates. EPA adjusted its Tower-bound risk estimates to account for the extra ar- senic, adding ~ ~ of water to drinking-water consumption to account for water used in cooking and multiplying the risk estimates by the fraction of arsenic contributed by drinking water. The subcommittee believes that the method used by EPA to account for the arsenic present in cooking water is valid and easily accomplished. However, the source of the data underlying EPA's decision to add 1 ~ of cooking water is not documented. EPA assumed that Taiwanese men and women ate one cup (0.23 L) of dry rice and 2 pounds (0.9 kg) of sweet potatoes each day, although it is not clear how those intake data were estimated by EPA. EPA estimated that the amount of arsenic present in food (before cooking) is higher in Taiwan than the United States. The sub- committee found that there is little evidence to support the estimates used. The subcommittee questions EPA's approach to adjusting for background levels of arsenic in food. A preferable approach would involve refitting the dose-response models with the appropriate background amounts added to individuals living in the region where arsenic is endemic. The subcommittee addressed the issue of background arsenic in food by adding a constant concentration of arsenic to the exposure rates for all individ- uals in the study villages. The assumed background rate in food was 30 fig/ day (corresponding to 0.6 ~glkg/day, assuming a 50-kg weight for a Taiwan- ese person). The impact on the ED estimates by adding this amount was relatively small (approximately a I% increase in ED estimates). Impact of Variability in Drinking-Water Intake The first three rows of Table 5-5 show the impact of varying the assumed ratio between mean water-consumption rates in Taiwan and the United States. EDo~s and LEDo~s were calculated using a multiplicative model, because it is computationally easier to implement; similar results would be expected using an additive model. As expected, the assumed ratio has a fairly dramatic im- pact on estimated EDo~s. The subcommittee considered the impact of varia- tion in individual drinking-water intake on risk estimates based on the Taiwan-

QUANTITA TIVE ASSESSMENT OF RISKS USING MODELING APPROACHES ~ 97 ese data used by EPA in arriving at their regulatory level of 10 ,ug/~. The discussion of this topic has centered around two main issues: The Taiwanese study used by EPA was ecological in nature, meaning that exposure was measured at the village level rather than the individual level. Differences among individuals in the amount of water they drink results in interindividual variability in exposure concentrations. A typical individual in the United States is likely to have a lower drinking-water intake (per unit of body weight) than a typical individual in Taiwan. It should be noted, however, that the assumption of major differences in water consumption between the Taiwanese study population and the U.S. population has been questioned (Mushak and Crocetti ~ 995~. To address those issues, a hierarchical model that formulated the true relative-risk model in terms of micrograms of arsenic per unit of body weight was fitted to the data. Drinking-water intake rates for the United States were based on data reported by EPA (20006~. Intake rates were also assumed to follow a gamma distribution with a mean 21 mL/kg/day and a standard devia- tion of 15 mL/kg/day, which compares well with the actual data shown in Figure 4-~. Distributions of intake rates for Taiwan were also assumed to follow a gamma distribution, but with the mean multiplied either by ~ (imply- ingno difference), 2.2 (the factor used in EPA ~ 988), or 3 (allowing implicitly for additional arsenic exposure through food and cooking water, as discussed above and used by EPA in its analysis). The models were fit using a Bayesian formulation (Carlin and Louis 1996) in the statistical package BUGS (SpiegeThalter et al. ~ 996) and a specialized program written for the subcom- mittee in Fortran. Table 5-5 shows the impact of drinking-water variability on EDGE estimates and standard errors. As can be seen in Table 5-5, estimates of the EDo~s tend to increase when individual variability in both Taiwan and the United States are taken into account. On the other hand, the lower limits on the EDo~s decrease. Therefore, as expected, the spread between the Tower limits and the EDo~s increases when adjustments are made. Effects of Exposure-Measurement Error Measurement errors in assigning village-specific exposures have been cited as another source of concern for analyses based on the Taiwanese data. As indicated in the ~ 999 NRC report, some ofthe villages had multiple wells with

~ 98 ARSENIC IN DRINKING WA TER: 2001 UPDA TE TABLE 5-5 Effect of Variability in Water Consumption on 1% Effective Dose Estimates Based on Male Lung Cancer in Sou~westem Taiwana Adjusted for Individual Taiwanese/U.S. Mean LEDo, Variability Drinldng-Water Rates EDGE (pg/L) (pg/L) No 1 65 57 2.2 145 129 3 195 173 Yes 1 117 41 2.2 191 66 3 246 85 a Multiplicative model with exposure entered as a linear term and age as a quadratic. A similar finding would be expected if the calculations were conducted using an additive model. The southwestern region of Taiwan is used as comparison population. BIER IV formula is used for EDo, calculations. Abbreviations: EDo,, 1% effective dose; LEDGE, lower 95% confidence limit on the 1% effec- tive dose. wide ranges of arsenic concentrations (see Table AlO-l, NRC ~ 999), yet all individuals living in that village would be assigned the median level as their exposure concentration. The statistical theory of measurement error (see Carroll et al. 1995) would predict a bias in estimated EDs under such circum- stances. To assess the degree of such bias, a small sensitivity analysis, fitting a range offeasible measurement-error models, was conducted. In the analysis, the concentration observed for the particular village is generated from a gamma distribution whose mean is the unknown true median concentration for that village and whose variance is defined as ~ 2. As discussed by Carroll et al. (1995), fitting a measurement-error model also requires specification ofthe distribution of true values; it was assumed that the true village-specific medi- ans were uniformly distributed between O and 1,200 ,ug/~. Models were fit in the package BUGS (SpiegeThalter et al. ~ 996~. Table 5-6 shows the results of this analysis based on a multiplicative model with a linear dose effect, using the southwestern region of Taiwan as an external comparison population. Although the subcommittee has recommended the use of the additive model, it has used the multiplicative model for the purpose of this exercise, because it is easier to implement computationally. As can be seen in the table, the estimated EDIT is attenuated as the magnitude of measurement error increases. The attenuation is modest and very small relative to the variability associated

QUANTITA TI BE ASSESSMENT OF RISKS USING MODELING APPROA CHES ~ 99 with model uncertainty. Furthermore, the subcommittee is recommending the use of an external comparison population, and under those circumstances, the impact of the measurement error would be expected to be much smaller (anal- yses not shown). The analysis reported here is based on strong assumptions and should not be overinterpreted as an actual assessment of the measurement error. The analysis is presented solely to assess the possible impact of mea- surement errors. To address the issue in depth would require more extensive analysis and is beyond the scope of this chapter. Mortality Versus Incidence The EDGE calculations reported by NBC (1999) and Morales et al. (2000) based on the data from southwestern Taiwan referred to lifetime cancer mor- tality (bladder or lung). Some assumptions were needed to compare these results with those based on the bladder cancer data from the study of Chiou et al. (2001) in northeastern Taiwan and the data on lung cancer from the study by Ferreccio et al. (2000) in Chile, since both studies report incidence data. In Table 5-3, the subcommittee reports EDo~s for cancer mortality based on the results from the Ferreccio et al. (2000) study and the Chiou et al. (2001) study. It was assumed for those calculations that the same relative risks would apply for cancer incidence and cancer mortality. For the purpose of discuss- ing risks in the U.S. population more generally, however, it is risk of disease incidence, not mortality, that is the endpoint of interest. As discussed earlier, EPA converted risks calculated from Taiwanese mortality data to cancer incidence by assuming an 80% mortality for bladder cancer and a 100% mor- taTity for lung cancer (see Table 5-2~. Even though that assumption is proba- bly appropriate for lung cancer, the subcommittee is concerned that a calcula- tion based on the 80% mortality for bladder cancer in Taiwan might not be appropriate for the U.S. population. According to the SEER (2001) cancer registry data, the U.S. lifetime bladder cancer incidence is approximately five times that of lifetime bladder mortality, indicating that the U.S. case mortality is close to 20%. Calculated Risk Estimates In this section, the subcommittee reports risk estimates that are based on incidence data taken directly from the SEER (2001) registry. Use ofthe BEIR

200 ARSENIC IN DRINKING WA TER: 2001 UPDA TE TABLE 5-6 1% Effective Dose and Lower 95°/0 Confidence Limit on 1% Effective Dose Computed Under Varying Degrees of Assumed Measurement Errora Measurement Error, S.D. (02), ~g) EDo1 (,ug/L) LEDol (lig/L) 0 (0) 145 129 10 (100) 145 129 32 (1,000) 144 127 45 (2,000) 143 127 55 (3,000) 142 125 _ . a On village-specific median arsenic levels. Calculations presented for male lung cancer data from the southwestern Taiwanese study using the Poisson multiplicative linear model. The ratio of the Taiwar~ese-to-U.S. Pearl drinking-water rate was assumed to be 2.2. Abbreviations: EDo,, 1% effective dose; LEDo,, lower 95% confidence limit on the 1% effec- tive dose: S.D.~ standard deviation; or, vanar~ce. IV formula (or its simple approximation see earlier discussion) easily facili- tates this approach. The lifetime cancer incidence assumed for the calcula- tions reported here are 7.85% and 5.75°/O for male and female lung cancer, respectively, and 3.42% and 1.13% for male and female bladder cancer, re- spectively. It is useful to note again for comparison purposes that the corre- sponding lifetime death rates are 7.62% and 4.85% for lifetime-lung cancer mortality (male and female, respectively), and 0.73% and 0.3% for lifetime bladder cancer mortality (male and female, respectively) (SEER 2001~. Tables 5-7,5-8, and 5-9 present estimated lifetime excess risks for lung and bladder cancer incidence for populations exposed to arsenic at 3,5,10, and 20 ~g/L of drinking water. The risk estimates in Tables 5-7 and 5-S are for lung and bladder cancer, respectively, and were calculated for the U.S. population using the lifetime background incidence for these diseases noted above. Risk estimates in Table 5-9 are adjusted by the difference in background incidence for bladder and lung cancer between the United States and Taiwan. Although lifetime incidences for these cancers in Taiwan are not readily available, age- adjusted incidences for bladder and lung cancer have been reported for the period 1993-1997 (You et al. 2001~. Assuming that life expectancy is not dramatically different between Taiwan and the United States during this pe- riod, the ratio of age-adjusted incidence reported for Taiwan can be compared

QUANTITA TIVE ASSESSMENT OF RISKS USING MODELING APPROACHES 201 with that reported for the United States during a similar period from the SEER (2001) data. For the same period for which data are available for Taiwan (1993-1997), the age-adjusted incidence of lung cancer in the United States was 58.6 (mates) and 34.0 (females) per 100,000. The background incidence of lung cancer in the United States, relative to Taiwan, is about 3 times greater for both mates and females. Likewise, the background incidence for bladder cancer in the United States was 23.3 (mates) and 5.4 (females) per 100,000. Thus, the background incidence for bladder cancer in the United States, rela- tive to Taiwan, was approximately 3.4-fold higher for males and 1.6-fold higher for females. Those factors have been used to calculate estimated risks using the background incidence for bladder and Jung cancer of Taiwan (Table 5-9) rather than that of the United States (Tables 5-7 and 5-~. The net effect of using U.S. background rates results in an increase in the risk estimates for arsenic-related cancers in the U.S. populations relative to what is obtained if the background rate of Taiwan is used, projecting a greater risk per unit dose of arsenic in the United States. Although not common practice for some researchers outside the field of epidemiology, the use of relative risks to infer risks across different populations is a widely accepted practice in epidemiol- ogy and has considerable scientific support (Bresiow and Day 1988~. It should be noted that the previous NRC (1999) report and the EPA's risk as- sessment (EPA 2001 ) <lid not use the higher U.S. background rates for bladder and lung cancer in their final risk estimates, although the previous NRC report discusses an approach using U.S. background rates (NRC 1999), and the previous subcommittee considered it a valid option. The maximum-likelihood estimates (MLE) in Table 5-7 and the lower and upper confidence limits on those estimates are presented. It is important to note that the confidence limits, which for the Taiwanese data are all less than 12% above or below the MLE, reflect only statistical variability, not uncer- tainty resulting from the issues discussed earlier in this chapter and in Chapter 4, including interindividual variability. For that reason, the confidence limits are presented in Table 5-7 as examples, but the individual limits are not pre- sented in other tables. The choice of which point on the distribution to use for regulatory purposes, be it the MLE or the upper or lower confidence limit, is a policy choice. All risk estimates presented in Tables 5-7, AS, and 5-9 are calculated using risk ratios determined by the additive Poisson model with dose entered as a linear term and using the BEIR IV formula. For the Taiwanese data set, the mortality data from all of southwestern Taiwan was used in the model to represent an unexposed external comparison population. It should be noted

202 ARSENIC INDRINK[NG WATER: 2001 UPDATE TABLE 5-7 Theoretical Excess Lifetime Risk Estimates (Incidence per 10,000 people) of Lung Cancer for U.S. Populations Exposed at Various Concentrations of Arsenic In Drinking Water Using Chilean and Taiwanese Data a ~ . . I Maximum-Likelihood Estimate (95°/O Lower L~m~t' 95°/O Upper Lit) . Arsenic Chilean Datab Taiwanese DataC Concentration (,ug/L) Females Males Females Males 14 (11, 18) 20 (14, 25) 5.4 (5.1, 5.7) 4.0 (3.9, 4.2) 5 24 (18, 29) 33 (24, 42) 8.9 (8.5, 9.4) 6.8 (6.5, 7.0) 10 48 (36,.59) 67 (48, 83) 18 (17, 19) 14 (13, 14) 20 95 (71, 120) 130 (95, 170) 36 (34,38) 27 (26, 28) aThese risks are estimated using assumptions considered to be reasonable by the subcommittee; it is possible to get higher and lower estimates using other assumptions. Confidence limits presented reflect statistical variability only, reflecting primarily the sample size. As such, they are not indicative of the true uncertainty associated with the estimates. Risk estimates are rounded to two significant figures. b Based on the data from Chile using the average arsenic concentration from the peak years of exposure (1958 to 1970) (Ferreccio et al. 2000), assuming that the typical U.S. and Chilean resident weighs 70 kg, that drinking rates in both countries are the same, and hence that the Chilean exposure per kilogram of body weight is 1.0 times the U.S. exposure, calculated using an additive Poisson model with linear dose. Risks were also estimated using the same assump- tions but based on the average arsenic concentrations from 1930 to 1994. The risk estimates (per 10,000) and corresponding upper and lower confidence limits in females exposed at 3,5, 10, and 20)1g/L are 50 (33,75),83 (56,130),170 (110,250), and 330 (220,500) respectively. The corresponding estimates for mares are 75 (43,100),130 (71,170),250 (140,330), and 500 (290, 670) respectively. c Estimates were calculated using data from individuals in the arsenic-endemic region of south- western Taiwan (Chen et al. l 985,1992; Wu et al. l 989), and data from an external comparison group from the overall southwestern Taiwan area. It was assumed that the typical U.S. resident weighs 70 kg, compared with 50 kg for the typical Taiwanese, and that the typical Taiwanese drinks just over 2 L of water per day, compared with 1 L per day in the United States; thus, the Taiwanese exposure per kilogram of body weight is 3 times the U.S. exposure, calculated using an additive Poisson model with linear dose. that the vast majority of individuals in all of southwestern Taiwan are not exposed to increased arsenic concentrations in drinking water. To permit comparisons of cancer risks for Taiwan with those for Chile and United States, the subcommittee has calculated risk estimates using different studies and assumptions. Thus, a range of possible risks from arsenic in drink- i

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES 203 TABLE 5-8 Theoretical Maxi Likelihood Estimates of Excess Lifetime Risk (Incidence per 10,000 People) of Bladder Cancer for U.S. Populations Exposed at Vanous Concentrations of Arsenic In Drinking Water Using Different Ratios for Taiwanese-to-U.S. Drinking-Water Ingestion on a Per-Body-Weight Basisa Taiwanese to U.S. Drinking Taiwanese to U.S. Drinking Arsenic Water Ratio = 1.4b Water Ratio = 3c Concentration (pg/L) Females Males Females Males 3 7.7 15 3.6 6.8 5 13 25 6.0 11 10 26 50 12 23 20 51 100 24 45 a These risks are estimated using assumptions considered to be reasonable by the subcommittee; it is possible to get higher and lower estimates using other assumptions. Estimates were calcu- lated using data from individuals in the arsenic-endemic region of southwestern Taiwan (Chen et al. 1985, 1992; Wu et al. 1989), and data from an external comparison group from the overall southwestern Taiwan area. Risk estimates are rounded to two significant figures. All 95% confidence limits are less than +/- 12% of the maximum-likelihood estimate and are not pre- sented. Those confidence limits reflect statistical variability only, reflecting primarily the sample size. As such, they are not indicative of the true uncertainty associated with the estimates. b Risks were estimated assuming that the typical U.S. resident weighs 70 kg, compared with 50 kg for the typical Taiwanese, and that the typical Taiwanese drinks just over 1 L of water per day, the same as the 1 L per day in the United States; thus, the Taiwanese exposure per kilogram of body weight is 1.4 times the U.S. exposure, calculated using an additive Poisson model with linear dose. c Risks were estimated assuming that the typical U.S. resident weighs 70 kg, compared with 50 kg for the typical Taiwanese, and that the typical Taiwanese drinks just over 2 L of water per day, compared with 1 L per day in the United States; thus, the Taiwanese exposure per kilogram of body weight is 3 times the U.S. exposure, calculated using an additive Poisson model with linear dose. ing water are presented. Those values, however, should not be considered bounds on the possible risk estimates, because other assumptions could be made that would result in higher or lower values. When estimating the risks to the U.S. population, assumptions must be made about body weights and water consumption in both the United States and the Taiwanese populations. Then, when comparing cancer risk estimates, it is important to be aware of how those assumptions affect the estimates. For example, the higher the ratio of water ingestion in Taiwan relative to the United States in terms of liters per body weight per day, the smaller the U.S. cancer risk estimate will be. The

204 ARSENIC IN DRINKING WA TER: 2001 UPDA TE assumptions that the subcommittee used to calculate the various estimates are discussed below and presented in the footnotes of the tables. Table 5-7 presents estimates for the excess lifetime risk (incidence) of lung cancer in females and mates in the U.S. population from exposure to arsenic at 3, 5, ~ 0, and 20 ~g/L of drinking water. The estimates based on the Chilean data (Ferreccio et al. 2000) were calculated assuming that the typical U.S. and Chilean resident both weigh 70 kg and that the drinking-water ingestion rates in both countries are the same that is, that exposures in Chile on a per-body- weight basis, given the same concentration of arsenic in the drinking water, are equal to those in the United States (i.e. a factor of 1~. The estimates based on the southwestern Taiwanese data set (Chen et al. 198S, 1992; Wu et al. 1989), however, were calculated assuming that the typical U.S. resident weighs 70 kg and drinks ~ ~ of water per day, and the typical Taiwanese resident weighs 50 kg and drinks just over 2 ~ of water per day. Therefore, given the same concentration of arsenic in drinking water, exposures in Tai- wan on a per-body-weight basis would be three times that of a U.S. resident. The data in Table 5-7 utilize the background rate of lung cancer in the United States. To illustrate the importance of the background rate, Table S-9 shows the same risk projections using the Taiwanese data set and the background TABLE 5-9 Theoretical Maximum-Likelihood Estimates of Excess Lifetime Risk (Incidence per 10,000 people) of Lung and Bladder Cancer for Populations Exposed at Various Concentrations of Arsenic in Droning Water, Using the Background Cancer Incidence Rate for Taiwana Arsenic Bladder Cancer Lung Cancer Concentration (,ug/L) Females Males Females Males 3 2.3 2.0 1.8 1.7 5 3.8 3.2 3.0 3.0 10 7.5 6.8 6.2 6.1 20 15 13 12 12 . a These risks are estimated using the assumptions noted in footnotes (a) and (c) of Table 5-8, and assuming a Taiwanese to U.S. drinking water ratio of 3. The estimated background incidence rates for bladder cancer in Taiwan (derived from the subcommittee's cancer risk estimates presented in Tables 5-7 and 5-8 and adjusted by the ratio of Taiwanese incidence data (You et al.2001) to U.S. incidence data (Ferlay et al. 2001~) are 6.9 (males) and 3.4 (females) per 100,000 and for lung cancer are 25.8 (males) and 11.9 (females) per 100,000.

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES 205 incidence rates for lung cancer in Taiwan, which are approximately 2- and 3- fold lower than those in the United States for males and females, respectively. For lung cancer, the MLEs of risk for mates based on the Chilean data (Fernccio et al. 2000) range from 20 per 10,000 at 3 ~g/L to 130 per 10,000 at 20 ~g/L using the average arsenic concentration during the peak exposure period of 1958 to 1970. The corresponding risk estimates in females are 14 per 10,000 to 95 per 10,000. Using the average arsenic concentration from 1930 to 1994, the risk estimates range from 75 to 500 per 10,000 males and from 15 to 330 per 10,000 females. Using the southwestern Taiwanese data, the risk estimates for arsenic at 3,ug/L of drinking water range from 1.7 to 4.0 per 10,000 for mares and from 1.8 to 5.4 per 10,000 for females, depending on which assumptions are used (Tables 5-8 and 5-9~. Different studies have estimated the risks of lung cancer following expo- sure to arsenic, and it is possible and useful to compare the risk estimates generated in the different analyses at a given arsenic concentration. Because the case mortality for lung cancer is close to 100% (SEER 2001), lung cancer incidence approximates lung cancer mortality, and the risk estimates for lung cancer mortality can be compared with risk estimates for lung cancer inci- dence. As can be seen in Table 5-7, the subcommittee's analysis ofthe south- western Taiwanese data yields an estimate for lifetime lung cancer incidence in the United States (using the U.S. background rate) at an arsenic concentra- tion in drinking water of 10 ~g/L of approximately 14 per 10,000 and 18 per 10,000 in males and females, respectively. It is noteworthy that nearly 10 years ago, Smith et al. (1992) published a risk assessment based on the same ecological southwestern Taiwanese data analyzed by the subcommittee. In that risk assessment, lifetime lung cancer mortality risks at 10 ~g/L were estimated to be 11 per 10,000 and 17 per 10,000 for males and females, re- spectively. Therefore, lung cancer risk estimates generated by this subcom- mittee and those published by Smith et al. (1992) are very consistent. The lung cane. er risk estimates derived from the Taiwanese data in Table 5-7 can be compared with those derived from the data in the recent case-control study in northern Chile by Ferreccio et al. (2000~. When the average arsenic con- centration in northern Chile Tom 1930 to 1994 is used as the dose-metric, the risk estimates for lung cancer incidence in the United States calculated by the subcommittee for an arsenic concentration of 10 ,ug/L are 250 per 10,000 mates and 170 per 10,000 females. Those estimates are approximately an order of magnitude higher than the estimates the subcommittee derived from the Taiwanese data. However, when the dose metric selected for the Chilean data is the peak years of arsenic exposure from 1958 to 1970, the correspond-

206 ARSENIC IN DRINKING WA TER: 2001 UPDA TE ing U.S. lung cancer risk estimates are 67 per 10,000 mates and 48 per 10,000 females. Those estimates are approximately 3 to 4 times higher than the sub- committee's estimates derived from the Taiwanese data. For comparative purposes, the subcommittee also derived cancer risk estimates at 10 ,ug/L using the relative risks (see Table 2-~) for lung cancer associated with the peak period of arsenic exposure in northern Chile in the ecological study by Smith et al. (1998~. If the same formula is applied to those relative risks, as was used to estimate the subcommittee's other cancer estimates, the U.S. lung cancer estimates at 10 ,ug/L are 38 per 10,000 and 21 per 10,000 in mates and females, respectively. Overall the peak period exposure data in northern Chile and the data from southwestern Taiwan yield coherent lifetime excess risk estimates ranging from 1.4 to 6.7 per 1,000 for lung cancer in the United States at a drinking- water arsenic concentration of 10 Vigil. The finding that risk estimates de- rived from studies of individuals exposed to arsenic in Chile are similar to those estimated from Taiwanese data provides confidence in the validity ofthe risk estimates. Tables 5-8 and 5-9 present estimates for the excess lifetime risk (incidence) for bladder cancer in females and mates in the U.S. population from exposure to arsenic at 3, 5, ~ 0, and 20 ~g/L of drinking water based on the southwestern Taiwanese study (Chen et al. 1985, 1992; Wu et al. 1989), using either the U.S. background rate for bladder cancer (Table 5-~) or the Taiwanese back- ground rate (Table 5-9~. In one water-intake scenario, the subcommittee assumed that the typical U.S. resident weighs 70 kg and drinks ~ ~ of water per day, and the typical Taiwanese resident weighs 50 kg and also drinks 1 L of water per day. Therefore, given the same concentration of arsenic in drink- ing water, exposures in Taiwan on a per-body-weight basis are 1.4 times that of a U.S. resident. in a second water-intake scenario, calculated using the same data set, it was assumed that the typical U.S. resident weighs 70 kg and drinks ~ ~ of water per day, and the typical Taiwanese resident weighs 50 kg and drinks just over 2.2 ~ of water per day. Therefore, as indicated in the lung cancer estimates earlier at the same concentration of arsenic in drinking water, exposures in Taiwan on a per-body-weight basis are 3 times that of a U.S. resident. For bladder cancer, the maximum-likelihood estimate of risk using a ratio of 1.4 for Taiwanese-to-U.S. drinking-water rates on a per-body-weight basis for males range from approximately 15 per 10,000 at an arsenic concentration of 3 ,ug/L of drinking water to 100 per 10,000 at 20 ~g/L. The corresponding estimates in females range from ~ to 5 ~ per 10,000. Using a drinking water

QUANTITA TIVE ASSESSMENT OF RISKS USING MODELING APPROA CHES 20 7 ratio of 3.0, the corresponding ranges are approximately 7 to 45 per 10,000 in males and 4 to 24 per 10,000 in females, using the U.S. background rate. Risk estimates are 3-fold Tower for mates and 2-fold lower for females if the Tai- wanese background rate is used (Table 5-9~. Further discussion of the subcommittee's quantitative risk estimates and comparison of them with the results of other analyses are presented in Chapter 6. SUMMARY AND CONCLUSIONS · Since EPA issued a pending standard of 10 ,ug/L, based on lung and bladder cancer data from the southwestern Taiwanese study in 2000, two additional studies have appeared in the literature that are of sufficient size and quality and with adequate quantification of dose to be considered in comput- ing EDs for arsenic in drinking water. One is a study that examined urinary tract cancer, and TCC in particular, in northeastern Taiwan (Chiou et al. 2001~; the second is a case-control study of lung cancer in Chile (Ferreccio et al. 2000). · Although it can be argued that an external comparison group for dose- response analysis of the original Taiwanese data should not be used, the sub- committee believes that such arguments are outweighed by evidence in favor of using a comparison population. A recent paper by Tsai et al. (1999) de- creases concerns about the potential role of confounding in using either the southwestern Taiwanese population or the entire Taiwanese population as an external comparison group. · Poisson models provide a flexible and useful framework for analysis of cohort data of the form seen in both the southwestern and northeastern Taiwanese studies. Although dose can be entered into the model in a variety of ways, the additive model with a linear dose effect is consistent with other analyses that have been applied to cancer cohort data. · The BEIR IV formula provides a useful approach to computing an EDGE for the United States based on relative risks obtained from a different population. The BEIR IV formula allows the incorporation of relative risk estimates obtained from case-control studies. The simple formula presented by Smith et al. (1992) is a close approximation to the BEIR IV approach. · There is insufficient knowledge on the mode of action of arsenic to justify the choice of any specific dose-response mode] the subcommittee explored a variety of models and ultimately used the additive mode] with

208 ARSENICIN DRINKING WATER: 2001 UPDATE linear dose effect. Although different results might be obtained from other reasonable model choices, the estimates do not differ by more than an order of magnitude. · Accounting for individual exposure rate variability causes the uncer- tainty in the estimate to increase. Therefore, as shown in Table 5-S, the cen- tral tendency estimates of the EDIT increase when individual variability in drinking-water rates is considered. The corresponding lower bound estimates for each EDIT decreases, however, because the variance about the mean be- comes larger. · Assumptions regarding the differences between mean drinking-water intakes in Taiwan and the United States can have substantial impacts on the estimated EDo,. There is evidence that drinking-water intakes in Taiwan might be closer to U.S. intakes than previously suggested, and that is an im- portant source of uncertainty. increasing the assumed drinking-water intakes for Taiwan provides a suitable approach to adjusting for arsenic exposure via cooking water, although the addition of ~ it, especially for women, seems excessive and warrants further investigation. · Adjusting for background arsenic exposure through food is likely to have only a modest effect on the estimated EDIT. · Measurement error in assigning the village-specific arsenic exposure concentrations is also likely to have only a modest impact on the estimated EDIT, compared with the variability associated with model uncertainty. Fur- ther exploration of this issue, however, would be useful. · Although there are insufficient data available for a formal combined analysis, it is helpful to compare the results across studies and also across various feasible assumptions to obtain a sense of the magnitude of likely effects. Table 5-3 presents such an analysis. · The northeastern Taiwanese study has several strengths, including exposure assessment and data on potential confounders, which could inform the dose-response assessment. At present, however, the follow-up time is insufficient to provide the precision necessary for quantitative dose-response assessment. ~ Analysis of the data from the period of peak arsenic exposure in northern Chile and the data from southwestern Taiwan results in similar esti- mates of lifetime lung cancer incidence in the United States. The consistency of the results adds to the confidence in the validity of the risk estimates.

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES 209 RECOMMENDATIONS · Quantitative risk assessment for arsenic in drinking water should consider the results from the following two studies: The data set from the southwestern Taiwanese study reported by Chen et al. (1988, 1992;) and Wu et al. (1989), which was considered by the NRC in the 1999 report and by EPA (lung and bladder cancer in males and fe- males). The Chilean case-control study (lung cancer) reported by Ferreccio et al. (2000~. · Dose-response analysis of the southwestern Taiwanese data should incorporate an unexposed comparison group; the southwestern Taiwanese region is the recommended comparison group. · The current mode-of-action data are insufficient to guide the selection of a specific dose-response model. The additive Poisson model with a linear term in dose is a biologically plausible model that provides a satisfactory fit to the epidemiological data and represents a reasonable model choice for use in the arsenic risk assessment. Research should be conducted on techniques for integrating the results from many epidemiological studies into a risk assessment. · Information on remaining uncertainties should be incorporated into future analyses when it is acquired. REFERENCES Albores, A., M.E. Cebrian, I. Tellez, and B. Valdez. 1979. Comparative study of chronic hydroarsenicism in two rural communities in the Laguna region of Mex- ico. tin Spanish]. Boll Of~cina Sanit. Panam. 86~3~:196-205. Breslow, N.E., and N.E. Day. 1988. Statistical Methods in Cancer Research: Vol. 2. The Design and Analysis of Cohort Studies. New York: Oxford University Press. Carlin, B.P., and T.A. Louis. 1996. Bayes and Empirical Bayes Methods for Data Analysis. New York: Chapman & Hall. Carroll, R.J., D. Ruppert, and L.A. Stefanski. 1995. Measurement Error in Nonlinear Models. New York: Chapman & Hall. CDC (Centers for Disease Control and Prevention). 2001. GMWK I Total Deaths for Each Cause by 5-Year Age Groups, United States, 1993, 1994, 1995, 1996, 1997,

2 ~ O ARSENIC IN DRINKING WA TER: 2001 UPDA TE andl998. Mortality Tables. DataWarehouse. NationalCenterforHealthStatis- tics, Centers for Disease Control and Prevention, U.S. Department of Health and Human Services,Hyattsville,MD. Online. Available: http://www.cdc.gov/ nchs/datawh/statab/unpubd/mortabs/gmwki.htm. ~ Sept. 21, 2001 ~ . Cebrian, M. 1987. Some Potential Problems in Assessing the Effects of Chronic Arsenic Exposure in North Mexico [preprint extended abstract]. Preprint of paper presented at the 194th National Meeting of the American Chemical Society, Aug. 30-Sept.4, 1987, New Orleans, LA. Cebrian, M.E., A. Albores, M. Aguilar, and E. Blakely. 1983. Chronic arsenic poi- soning in the north of Mexico. Hum. Toxicol. 2~1~:121-133. Chen, C.J., Y.C. Chuang, T.M. Lin, and H.Y. Wu. 1985. Malignant neoplasms among residents of a blackfoot disease-endemic area in Taiwan: high-arsenic artesian well water and cancers. Cancer Res. 45~11 Pt 2~:5895-5899. Chen, C.J., M. Wu, S.S. Lee, J.D. Wang, S.H. Cheng, and H.Y. Wu. 1988. Atherogenicity and carcinogenicity of high-arsenic artesian well water. Multiple risk factors and related malignant neoplasms of blackfoot disease. Arteriosclero- sis 8~5~:452460. Chen, C.J., C.W. Chen, M.M. Wu, and T.L. Kuo. 1992. Cancer potential in liver, lung, bladder and kidney due to ingested inorganic arsenic in drinldog water. Br. J. Cancer 66~5~:888-892. Chiou, H.Y., S.T. Chiou, Y.H. Hsu, Y.L. Chou, C.H. Tseng, M.L. Wei, and C.J. Chen. 2001. Incidence of transitional cell carcinoma and arsenic in drinldng water: a follow-up study of 8,102 residents in an arseniasis-endemic area in nor~eastem Taiwan. Am J. Epidemiol. 153 (5) :411 -418. Concha, G., B. Nermell, and M. Vahter. 1998. Metabolism of inorganic arsenic in children with chronic high arsenic exposure in northem Argentina. Environ. Health Perspect. 106~6~:355-359. Cuzick, J., S. Evans, M. Gillman, and D.A. Price Evans. 1982. Medicinal arsenic and intemal malignancies. Br. J. Cancer 45~6~:904-911. EPA (U.S. Environmental Protection Agency). 1988. Special report on Ingested Inorganic Arsenic: Skin Cancer; Nutritional Essentiality. EPA/625/3-87/013. Risk Assessment Forum, U.S. Environmental Protection Agency, Washington, DC. July 1988. EPA (U.S. Environmental Protection Agency). 1996. Proposed Guidelines for Carci- nogenic Risk Assessment. Notice. Fed. Regist. 61~79~:17960. EPA (U.S. Environmental Protection Agency). 1998. Cost of Lung Cancer Chapter II.V in Cost of Illness Handbook. Off~ce of Pollution Prevention and Toxics, U.S.EnvironmentalProtection Agency. "Online]. Available: http://www.epa. gov/oppt/ coi/. [August 17, 2001 ~ . EPA (IJ.S. Env~ronmental Protection Agency). 2000a. 40 CFR Parts 141 and 142. National Pr~mary Drinking Water Regulations. Arsenic and Clarif~cations to Compliance and New Source Contaminants Monitoring. Notice of proposed rulemalcing. Fed. Regist. 65~121~:38887-38983.

QUANT7TA TINE ASSESSMENT OF RISKS USING MODELING APPROA CHES 21~ EPA (U.S. Environmental Protection Agency). 2000b. 40 CFR Parts 141 and 142. National Primary Drinking Water Regulations. Arsenic and Clarifications to Compliance end New Source Contaminants Monitoring. Notice of data availabil- ity. Fed. Regist. 65(204): 63027-63035. EPA (U.S. Environmental Protection Agency). 2000c. Arsenic Proposed Drinking Water Regulation: A Science Advisory Board Review of Certain Elements of the Proposal, A Report by the EPA Science Advisory Board. EPA-SAB-DWC-01- 001. S cience Advisory B card, U. S. . Environmental Protection Agency, Washing- ton, DC. December. "Online]. Available: http://www.epa.gov/sab/ f~scalOl.htm. EPA (U.S. Environmental Protection Agency). 2000d. Estimated Per Capita Water Ingestion in the United States: Based on Data Collected by the United States Department of Agriculture's (USDA) 1994-1996 Continuing Survey of Food Intakes by Individuals. EPA-822-00-008. Of lice of Water, Office of Standards and Technology, U.S. Environmental Protection Agency. April 2000. EPA (U.S. Environmental Protection Agency). 2001. 40 CFR Parts 9,141 and 142. National Primary Drinking Water Regulations. Arsenic and Clarifications to Compliance end New Source Contaminants Monitoring. Final Rule. Fed. Regist. 66~14~:6975-7066. Ferlay, J., F. Bray, P. Pisani, and D. M. Parkin. 2001. GLOBOCAN 2000: Cancer Incidence, Mortality and Prevalence Worldwide, Version 1.0. IARC CancerBase, No.5. IARC Press, Lyons, France. "Online]. Available: http://www.dep.iarc.fr/ globocan/ globocan.ht~ tSept. 20, 2001~. Ferreccio, C., C. Gonzalez, V. Milosavljevic, G. Marshall, A.M. Sancha, and A.H. Smith. 2000. Lung cancer and arsenic concentrations in drinking water in Chile. Epidemiology 11~6~:673-679. Gail, M. 1975. Measuring the benefit of reduced exposure to environmental carcino- gens. J. Chronic Dis. 28~3~:135-147. Hastie, T., and R. Tibshirani.1990. Generalized Additive Models. New York: Chap- man and Hall. Hopenhayn-Rich, C., M.L. Biggs, A. Fuchs, R. Bergoglio, E.E. Tello, H. Nicolli, and A.H. Smith. 1996. Bladder cancer mortality associated with arsenic in Linking water in Argentina. Epidemiology 7~2~:117-124. Hopenhayn-Rich, C., M.L. Biggs, and A.H. Smith. 1998. Lung and kidney cancer mortality associated with arsenic in drinking water in Cordoba, Argentina. Int. J. Epidemiol. 27~4~:561-569. Hopenhayn-Rich, C., S.R. Browning, I. Hertz-Picciotto, C. Ferreccio, C. Peralta, and H. Gibb. 2000. Chronic arsenic exposure and risk of infant mortality in two areas of Chile. Environ. Health Perspect. 108~79:667-673. IARC (International Agency for Research on Cancer). 1999. Cancer Survival in Developing Countries, R. Sankaranarayanan, R.J. Black, and D.M. Parkin, eds. Pub. No. 145. Lyon: International Agency for Research on Cancer, World Health Organization.

212 ARSENIC IN DRINKING WA TER: 2001 UPDA TE Kur~tio, P., E. Pukkala, H. Kahelin, A. Auvinen, and J. Pekkanen. 1999. Arsenic concentrations in well water and risk of bladder and kidney cancer in Finland. Environ. Health Perspect. 107~9~:705-710. Lai, M.S., Y.M. Hsueh, C.J. Chen, M.P. Shyu, S.Y. Chen, T.L. Kuo, M.M. Wu, and T.Y. Tail 1994. Ingested inorganic arsenic and prevalence of diabetes mellitus. Am. J. Epidemiol. 139(5):484-492. Lewis, D.R., J.W. Southwick, R. Ouellet-Hellstrom, J. Rench, and R.L. Calderon. 1999. Drinking water arsenic in Utah: a cohort mortality study. Environ. Health Perspect. 107~5~:359-365. Mazumder, D.N., J. Das Gupta, A. Santra, A. Pal, A. Chose, S. Sarkar, N. Chattopadhaya, and D. Chakraborty. 1997. Non-cancer effects of chronic arsenicosis with special reference to liver damage. Pp.112- 123 in Arsenic Expo- sure and Health Effects, C.O. Abernathy, R. L. Calderon, and W. Chappell, eds. London: Chapman & Hall. Morales, K.H. 2001. Statistical Methods for Risk Assessment Based on Epidemio- logical Data. Ph. D. Thesis Submitted to Harvard School of Public Health, De- partment of Biostatistics, Boston, Mass. May 2001. Morales, K.H., L. Ryan, T.L. Kuo, M.M. Wu, and C.J. Chen. 2000. Risk of internal cancers from arsenic in drinking water. Environ. Health Perspect. 108~7~:655- 661. Morris, J.S., M. Schmid, S. Newman, P.J. Scheuer, and S. Sherlock. 1974. Arsenic and noncirrhotic portal hypertension. Gastroenterology 66~1~:86-94. Mushak, P., and A.F. Crocetti. 1995. Risk and revisionism in arsenic cancer risk assessment. Environ. Health Perspect. 103~7-8~:684-689. Nevens, F., J. Fevery, W. Van Steenbergen, R. Sciot, V. Desmet, and J. De Groote. 1990. Arsenic and noncirrhotic portal hypertension: a report of eight cases. J. Hepatol. 11~1~:80-85. Neubauer, O. 1947. Arsenical cancer: a review. Br. J. Cancer l(June):192-251. NRC (National Research Council). 1983. Risk Assessment in the Federal Govern- ment: Managing the Process. Washington, DC: National Academy Press. NRC (National Research Council). 1988. Health Risks of Radon and Other Inter- nallyDepositedAlpha-Emitters: BEIRIV. Washington, DC: National Academy Press. NRC (National Research Council). 1999. Arsenic in Drinlcing Water. Washington, DC: National Academy Press. Rahman, M., and J.O. Axelson. 1995. Diabetes mellitus and arsenic exposure: A second look at case-control data from a Swedish copper smelter. Occup. Environ. Med. 52~119:773-774. Rahman, M., M. Tondel, S.A. Ahmad, and C. Axelson. 1998. Diabetes mellitus associated with arsenic exposure in Bangladesh. Am. J. Epidemiol.148~2~: 198- 203. Roth, F. 1956. Concerning chronic arsenic poisoning of the Moselle wine growers with special emphasis on arsenic carcinomas. Krebsforschung 61:287-319.

QUANTITATIVE ASSESSMENT OF RISKS USING MODELING APPROACHES 213 SEER (Surveillance, Epidemiology, and End Results). 2001. Surveillance, Epidemi- ology, and End Results Program Public-Use Data (1973-1998), National Cancer Institute, DCCPS, Surveillance Research Program, Cancer Statistics Branch. "Online]. Available: http://www-seer.ims.nci.nih.gov/Publications/. [August20, 2001]. Smith, A.H., C. Hopenhayn-Rich, M.N. Bates, H.M. Goeden, I. Hertz-Picciotto, H.M. Duggan, R. Wood, M.J. Kosnett, and M.T. Smith. 1992. Cancer risks from arsenic in drinking water. Environ. Health Perspect. 97:259-267. Smith, A.H., M. Goycolea, R. Haque, and M.L. Biggs. 1998. Marked increase in bladder and lung cancer mortality in a region of northern Chile due to arsenic in dying water. Am. J. Epidemiol. 147~7~:660-669. Spiegelhalter, D.J., A. Thomas, N.G. Best, and W.R. Gilks. 1996. BUGS: Bayesian inference Using Gibbs Sampling, Version 0.5. Cambridge: MRC Biostatistics Unit. "Online]. Available: http://www.mrc-bsu.cam.ac.uk/bugs/. tAugust 20, 2001]. Tsai, S.M., T.N. Wang, and Y.C. Ko. 1999. Mortality for certain diseases in areas with high levels of arsenic in drinking water. Arch. Environ. Health 54~3~: 186- 193. Tseng, W.P. 1977. Effects and dose-response relationships of skin cancer and black- foot disease with arsenic. Environ. Health Perspect. 19: 109-119. Tseng, W.P., H.M. Chu, S.W. How, J.M. Fong, C.S. Lin, and S. Yeh. 1968. Preva- lence of skin cancer in an endemic area of chronic arsenicism in Taiwan. J. Natl. Cancer Inst. 40~3~:453-463. Tsuda, T., A. Babazono, E. Yamarnoto, N. Krumatani, Y. Mino, T. Ogawa, Y. Kishi, and H. Aoyama. 1995. Ingested arsenic and internal cancer: A historical cohort study followed for 33 years. Am. J. Epidemiol. 141~3~:198-209. Wu, M.M., T.L. Kuo, Y.H. Hwang, and C.J. Chen. 1989. Dose-response relation between arsenic concentration in well water and mortality from cancers and vascular diseases. Am. J. Epidemiol. 130~6~: 1123-1132. Yeh, S. 1973. Skin cancer in chronic arsenicism. Hum. Pathol. 4~4~:469-485. You, S.L., T.W. Huang, and C.J. Chen. 2001. Cancer registration in Taiwan. Asian Pac. J. Cancer Prev. 2( IACR suppl.~:75-78. Zaldivar, R. 1974. Arsenic contamination of drinking water and foodstuffs causing endemic chronic poisoning. Beitr. Pathol. 151~4~:384-400.

Next: 6 Hazard Assessment »
Arsenic in Drinking Water: 2001 Update Get This Book
×
Buy Paperback | $53.00 Buy Ebook | $42.99
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

Having safe drinking water is important to all Americans. The Environmental Protection Agency's decision in the summer of 2001 to delay implementing a new, more stringent standard for the maximum allowable level for arsenic in drinking water generated a great deal of criticism and controversy. Ultimately at issue were newer data on arsenic beyond those that had been examined in a 1999 National Research Council report. EPA asked the National Research Council for an evaluation of the new data available.

The committee's analyses and conclusions are presented in Arsenic in Drinking Water: 2001 Update. New epidemiological studies are critically evaluated, as are new experimental data that provide information on how and at what level arsenic in drinking water can lead to cancer. The report's findings are consistent with those of the 1999 report that found high risks of cancer at the previous federal standard of 50 parts per billion. In fact, the new report concludes that men and women who consume water containing 3 parts per billion of arsenic daily have about a 1 in 1,000 increased risk of developing bladder or lung cancer during their lifetime.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!