Cardiovascular Effects

Earlier epidemiological studies have indicated that the cardiovascular system might be sensitive to chronic ingestion of arsenic. Effects seen following chronic exposure to arsenic in drinking water include hypertension and increased cardiovascular-disease mortality (NRC 1999).

Rahman et al. (1999) conducted a cross-sectional evaluation of blood pressure in 1,595 adults (above 30 years of age) who resided their entire lives in one of four villages in a rural area of Bangladesh. Many villages in Bangladesh have high arsenic exposures resulting from the use of groundwater for drinking water; wells were drilled because of microbial contamination of surface waters. Well-water arsenic concentrations were determined by reference to a database compiled from recent surveys, and most contained arsenic concentrations in excess of 0.05 mg/L. Examiners were not completely blinded to a subject's general exposure status but had no knowledge of precise levels of exposure. No subjects were taking antihypertensive medication, and the diet, lifestyle, and socioeconomic status of all subjects were similar. Prevalence ratios for hypertension were assessed after adjustment for age, sex, and body-mass index. The Mantel-Haenszel-adjusted prevalence ratios for hypertension increased with increasing arsenic in drinking water. The ratios were 1.2 (95% confidence interval (CI)=0.6–2.3; 50 cases, 573 controls), 2.2 (95% CI=1.1–4.3; 93 cases, 483 controls), and 2.5 (95% CI=1.2–4.9; 55 cases, 227 controls) for exposure category I (arsenic at <0.5 mg/L), II (0.5–1.0 mg/L), and III (>1.0 mg/L), respectively. The chi-square test for trend was highly significant (p ≪ 0.001). Cumulative arsenic consumption (arsenic concentration of the well water multiplied by the years of consumption) was also analyzed. The adjusted prevalence ratios for hypertension were 0.8 (95% CI=0.3–1.7; 13 cases, 225 controls); 1.5 (95% CI=0.7–2.9; 83 cases, 610 controls), 2.2 (95% CI=1.1–4.4; 40 cases, 239 controls), and 3.0 (95% CI= 1.5–5.8; 62 cases, 209 controls) for cumulative exposures of less than 1.0 mg/L-years, 1.0–5.0 mg/L-years, greater than 5.0–10.0 mg/L-years, and greater than 10.0 mg/L-years, respectively. The chi-square test for trend was highly significant (p ≪ 0.001). In a linear regression model that took into account age, sex, and body-mass index, mean blood pressure increased with both exposure measures. This study is consistent with early reports in Taiwan associating average and cumulative arsenic exposure in drinking water with a risk of hypertension.

Lewis et al. (1999a) conducted a retrospective cohort mortality study of residents of Millard County, Utah, an area where some drinking-water wells contained concentrations of arsenic up to several hundred micrograms per liter. This study is critiqued in detail in the Cancer Effects section of this chapter. Relative to statewide rates, the cohort had an increased risk of death from “hypertensive heart disease” in males (standardized mortality ratio (SMR)=2.20, 95% CI=1.36–3.36) and in females (SMR=1.73, 95% CI= 1.11–2.58). Although statistical power was limited, there was no evidence of a positive dose-response relationship. It is uncertain whether this finding reflects a chronic hypertensive effect of arsenic, because other causes of mortality more commonly associated with hypertension, particularly ischemic heart disease and cerebrovascular disease, were significantly decreased in the cohort. “Hypertensive heart disease” is infrequently encountered as a coded cause of death, and because the increased SMRs were based on small numbers (21 deaths in males and 24 deaths in females), only a slight degree of misclassification bias might have influenced the results. Deaths in the broad category of “nephritis and nephrosis” were increased in males (SMR=1.72, 95% CI=1.13–2.50), and the possibility exists that the grouping might have subsumed some cases of nonspecific nephrosclerosis associated with hypertension.

Hertz-Picciotto et al. (2000) conducted a reanalysis of circulatory disease mortality among a cohort of smelter workers to assess whether the healthy worker survivor effect (HWSE) might have obscured a potential contributory role of cumulative airborne arsenic exposure. Although the route of exposure might have been predominantly by inhalation, the study addresses an impact on a nonpulmonary systemic end point that might also have relevance to arsenic ingestion. Using the least-exposed cohort members as internal controls, analytical approaches that applied a time-lagging method for exposure assessment and adjusted for employment status as a time-dependent variable in each year of follow-up revealed a stronger association between arsenic exposure and cardiovascular disease mortality. A G-null approach that used time-period-dependent exposure rates rather than cumulative exposure to control for the HWSE found no relationship between arsenic and cardiovascular disease mortality; however, because of data-set limitations, the power of that analysis was low. The authors concluded that the HWSE might have contributed to the apparent lack of arsenic-associated cardiovascular disease mortality in prior occupational cohort studies, in contrast to positive effects observed in several drinking-water studies.

Dermal Effects

Chronic arsenic exposure causes a characteristic pattern of noncancer dermal effects that begins with spotted hyperpigmentation and might later include palmar and plantar hyperkeratosis. Many studies about those skin lesions in humans have been published (NRC 1999). Several recent studies have investigated adverse health effects associated with ingestion of arsenic present in groundwater in the Gangetic plane of West Bengal, India, and neighboring Bangladesh, where over the past decade more than 30 million people might have been consuming water with arsenic concentrations in excess of 50 µg/L (Chowdbury et al. 2000).

Mazumder et al. (1998) conducted a cross-sectional survey of the prevalence of truncal hyperpigmentation and palmar-plantar keratosis in a region of West Bengal, India, with groundwater arsenic concentrations ranging from nondetectable to 3,400 µg/L. Medical examinations, current volume of water consumption, and well-water arsenic measurements were obtained on 7,683 children and adults (4,093 females and 3,590 males) recruited from rural villages in a region known to have high groundwater arsenic concentrations and in a referent population from a region thought to have lower exposures (total population at risk was 150,457). Heterogeneity of exposure existed within each region, and examiners were blinded to the exact exposures of the subjects. For greater than 80% of the participants, well-water arsenic concentration was less than 500 µg/L. Age-adjusted prevalence of hyperpigmentation was 0.3/100 in females in the lowest exposure level (<50 µg/L) and increased to 11.5/100 in the highest exposure level (≥800 µg/L). Corresponding prevalences for males were 0.4/100 and 22.7/100. For keratosis, the prevalence for females ranged from zero in the lowest exposure level (<50 µg/L) to 8.5/100 in the highest exposure level (≥800 µg/L), and for males, it ranged from 0.2/100 to 10.7/100. Of note, 29 subjects with hyperpigmentation and 12 subjects with keratosis were consuming water from domestic wells containing arsenic concentrations less than 100 µg/L. However, as noted by the investigators, because of the cross-sectional nature of the investigation, the possibility existed that these individuals might have consumed water containing higher concentrations of arsenic at their worksites or at past residences.

A separate analysis (Mazumder et al. 1998) of 4,443 subjects examined the prevalence of hyperpigmentation and keratosis according to tertiles of arsenic dose calculated on a body-weight basis. The prevalence (not age adjusted) of hyperpigmentation was 0.0/100 for females and 0.4/100 for males in the lowest textile (0 to 3.2 µg/kg/day); 3.0/100 for females and 6.5/100 for males in the middle textile (3.2 to 14.9 µg/kg/day); and 5.9/100 for females and 16.6/100 for males in the highest tertile (14.9 to 73.9 µg/kg/day). One-tailed chi-square tests of trend were significant (p < 0.001) for males and females. The corresponding prevalences for keratosis in females were 0.7/ 100, 2.3/100, and 3.5/100 (test for trend p=0.028) and for keratosis in males were 0.8/100, 4.2/100, and 12.7/100 (test for trend p < 0.001).

Tondel et al. (1999) conducted a cross-sectional survey for characteristic arsenic-related skin lesions in 1,481 subjects (903 males, and 578 females) at least 30 years of age who resided in four rural villages of Bangladesh. The villages were selected because they exhibited a range of arsenic concentration in the drinking water (nondetectable to 2,040 µg/L). Although specific information was collected on the water source of each individual, the daily volume of water consumed was not ascertained, and examiners were apparently not blinded to subjects' exposure levels. The crude prevalence of any arsenic-associated skin lesion (pigmentary changes or keratosis) was 29/100. Data were analyzed in five exposure categories. The age-adjusted prevalence of any arsenic-related skin lesion increased in relation to the arsenic concentration of the drinking water. In males in the lowest exposure category (drinking-water arsenic, ≤150 µg/L), the age-adjusted prevalence of any arsenic-associated skin lesion was 18.6/100 and increased to 37.0/100 in the highest exposure category (>1,000 µg/L) (chi-square dose for trend, p < 0.001). The corresponding rates for females were 17.9/100 in the lowest exposure category and 24.9 in the highest exposure category (test for trend, p < 0.02). At similar levels of arsenic in drinking water, males had a higher prevalence of skin lesions than females. A dose-index, calculated as the arsenic concentration of a subject's drinking water divided by his or her body weight, was used to examine the trend in age-adjusted prevalence of skin lesions across three categories (≤5 µg/L/kg, >5–10 µg/L/kg, and >10 µg/L/kg). The corresponding age-adjusted prevalences for males were 19.6/100, 30.2/100, and 34.8/100 (test for trend, p < 0.001); for females, the respective rates were 19.7/100, 22.1/100, and 30.8/100 (test for trend, p < 0.001).

Ahsan et al. (2000) conducted a cross-sectional survey of “melanosis” (hyperpigmentation) and keratosis in three contiguous rural villages in Bangladesh within a region suspected of having elevated arsenic in groundwater. In the study, 87 males and 80 females were drawn from a total population of 300 residents; the investigators indicated that there might have been a greater number of affected individuals volunteering than unaffected individuals, resulting in an overestimate of actual prevalence rates. Exposure variables included the concentration of total arsenic in a spot urine sample, the arsenic concentration of drinking water contained in storage pitchers found in each subject's residence, and a cumulative exposure index calculated by multiplying the arsenic concentration in the pitcher water by the estimated amount of water consumed per year and by the estimated number of years that the current tube-well had been used for drinking. Nine of the 167 participants had melanosis, 2 had keratosis, and 25 had both melanosis and keratosis. A surprising finding in this study was the relatively high prevalence of skin lesions in subjects whose current drinking-water samples contained very low concentrations of arsenic; 13.9% of all subjects with skin lesions were currently drinking water with an arsenic concentration of <10 µg/L. In the analyses that evaluated skin lesions as a function of cumulative arsenic index in milligrams, 7 of 38 subjects (18.4%) in the lowest quartile (<116.4 mg) had skin lesions. In the highest quartile (1,279.9–22,147.1 mg), 14 of 39 subjects (35.9%) had skin lesions. Compared with subjects in the lowest quartile of total urinary arsenic excretion, subjects in the highest quartile had an odds ratio of 3.6 for the presence of any skin lesion (95% CI=1.2, 12.1; urinary arsenic at 471– 1,840 µg/L). However, the correlation between creatinine-adjusted total urinary arsenic and arsenic concentration in drinking water was only 0.5.

Tucker et al. (2001) examined the relationship between skin lesions and arsenic ingestion in a cross-sectional study of 3,228 children and adults exposed to arsenic in drinking water in Inner Mongolia. Design aspects of this study are critiqued in the Cancer Effects section (see below). In an analytical approach that assessed exposure by using the “peak arsenic concentration” of the well water that was used during the subject's lifetime, there was a dose-dependent increase in the age-adjusted prevalence of both keratoses and “dyspigmentation” (truncal hyperpigmentation). Thirty-seven of the 172 subjects with keratoses and 25 of the 121 subjects with dyspigmentation were assigned peak arsenic concentration exposures of less than 100 µg/L.

Reproductive and Developmental Effects

Arsenic exposures have been associated with a number of adverse health outcomes. Relatively little attention, however, has been directed toward assessing the potential impact of arsenic on human reproductive health effects, despite studies in both humans and experimental animals demonstrating that arsenic and its methylated metabolites easily pass the placenta (Concha et al. 1998). Evidence from human studies suggests the potential for adverse effects on several reproductive end points. A small number of epidemiological studies investigating the relationship between arsenic exposure in humans and adverse reproductive effects have been published since the 1999 report, but they suffer from some limitations and are not necessarily applicable in populations exposed to arsenic in drinking water.

A hospital case-control study found an increase of stillbirths in relation to the proximity of maternal residence to an arsenical pesticide production plant in Texas (Ihrig et al. 1998). The study was small, however, and the findings seem to be restricted to specific subgroups. In addition, exposures to other agents from the chemical plants were not assessed.

An increase in infant mortality (divided into three subcategories: stillbirths, neonatal, and postneonatal) was observed in a county in northern Chile during a time when there was a substantial increase in the arsenic concentration of the public drinking-water supply (from 90 to 800 µg/L) (Hopenhayn-Rich et al. 2000). A subsequent decrease in arsenic was accompanied by a return to the normal Chilean trend in infant mortality over time.

A retrospective survey in Bangladesh (Ahmad et al. 2001) compared several outcomes in women exposed to high (mean 240 µg/L) and low (below 20 µg/L) arsenic concentrations in drinking water and found increases in spontaneous abortions, stillbirths, and preterm births. This study was based on recall of previous pregnancies, however, and ascertainment of the outcomes was not clearly defined.

In summary, although several studies have addressed the potential reproductive effects of arsenic exposures in humans, the evidence is not conclusive. Many of the studies lacked information on lifestyle or personal factors that affect birth weight, congenital malformation, and other outcomes and information on other potential exposures. Several studies are ecological in nature and therefore subject to additional potential biases.

Neurological Effects

Previous studies of arsenic's neurological effects have generally focused on central-nervous-system effects seen following acute, high-dose intoxication and on the peripheral neuropathy that occurs following subacute or chronic exposure (NRC 1999). Two recent studies discussed below have focused on subtle cognitive effects in children following chronic exposure to arsenic.

Siripitayakunkit et al. (1999) investigated the association between environmental arsenic exposure and the intelligence of children in the Ronpiboon district of Thailand, where shallow artesian wells are contaminated with arsenic at concentrations as high as several milligrams per liter. Head-hair arsenic concentration and performance on the Weschler Intelligence Scale Test for Children (WISTC) in 529 schoolchildren (6 to 9 years of age) were analyzed in a cross-sectional evaluation. Median head-hair arsenic concentration was 2.42 µg/g (from 0.48 to 26.94 µg/g). Only 8.3% of the subjects had head-hair arsenic concentrations of 1 µg/g or less. Head-hair arsenic concentration was inversely associated with full-scale intelligence quotient (IQ) in an analysis adjusted for age, father's occupation, maternal intelligence, and family income. There were no data on exposure to lead, a major potential confounder, or on nutritional factors, such as iron. Although subjects lived in areas that apparently had different environmental arsenic concentrations, it was not stated whether the examiners were blinded to the location of residence. In addition, no data were presented regarding the concentration of arsenic in the children's drinking water, the concentration in the subjects' urine, or the presence of arsenic-related skin lesions in any subject. These factors limit the interpretation of the association between IQ and head-hair arsenic concentration.

A recent cross-sectional study in San Luis Potosi, Mexico (Calderon et al. 2001) examined the impact of arsenic and lead on the neuropsychological performance of schoolchildren aged 6 to 9 years. Subjects included 41 children living within 1.5 kilometers (km) of a smelter complex (Morales zone) with increased arsenic and lead concentrations and 39 children living 7 km upwind from the smelter (Martinez zone). The geometric mean total arsenic concentration in urine was higher in the Morales children (62.91 µg/g of creatinine; from 27.54 to 186.21) than in the Martinez children (40.28 µg/g of creatinine; from 18.20 to 70.79) (p < 0.05). Blood lead concentrations were similar in the two groups. Maternal and paternal educational attainment, socioeconomic status, and iron status were lower in the Martinez group. Neuropsychological performance was assessed using the Weschler Intelligence Scale for Children, Revised Version, for Mexico (WISC-RM). In a comparison unadjusted for indices of metal exposure or other covariates, the Morales children scored significantly higher than the Martinez children on the full-scale IQ test and other neuropsychological subscores. However, investigators reported an inverse correlation between log- transformed total urinary arsenic concentration (micrograms per gram of creatinine) and verbal IQ after adjustment by age, sex, socioeconomic status, maternal and paternal education, nutritional factors (transferrin saturation and height by age index), and blood lead. The relationship between log-transformed total urinary arsenic concentration and performance IQ was not significant. Speciation of arsenic in the urine (inorganic arsenic, monomethylarsonic acid (MMA), and dimethylarsinic acid (DMA)) was not performed, and the impact of past exposure to high levels of arsenic and lead emissions near the smelter was not assessed.

Respiratory Effects

Noncancer respiratory effects have been reported in populations exposed to arsenic in drinking water (NRC 1999), but the database on this topic has been sparse. Mazumder et al. (2000) recently reported an association between arsenic ingestion in drinking water and the prevalence of respiratory effects in a large cross-sectional survey of subjects residing in one of the arsenic-affected districts of West Bengal, India. Dermal effects have also been studied in this population (Mazumder et al. 1998). The analysis of the respiratory effects excluded 819 of the 7,683 individuals because of current or past history of smoking. Prevalence odds ratio (POR) estimates for abnormal chest sounds on physical examination were increased for subjects who had arsenic skin lesions and who consumed water with arsenic at greater than 500 µg/L, compared with subjects who had no skin lesions and who consumed water with arsenic at less than 50 µg/L. For females, the age-adjusted POR for cough was 7.8 (95% CI=3.1–19.5) and for chest sounds, 9.6 (95% CI=4.0– 22.9). For males, the age-adjusted POR for cough was 5.0 (95% CI=2.6–9.9) and for chest sounds, 6.9 (95% CI=3.1–15.0). Examiners were not blinded to the presence of arsenic-related skin lesions. However, the large size of the study, the high odds ratios observed, and the positive trend with current arsenic exposure were such that the study provides support for reports of arsenic-associated pulmonary effects previously noted in Chile in the 1960s and 1970s (see NRC 1999).

Milton et al. (2001) reported an association between chronic arsenic ingestion and “chronic bronchitis” in a small cross-sectional study in Bangladesh of 94 individuals with arsenic-associated skin lesions. These individuals were attending “health awareness” meetings in three villages. The mean concentration of arsenic in drinking water was 614 µg/L, and the range was 136 to 1,000 µg/L. The study included 124 nonexposed individuals recruited from three villages “known to be not contaminated with arsenic.” All participants never smoked, and all denied a history of asthma or tuberculosis. Chronic bronchitis was defined as a history of cough productive of sputum on most days for at least 3 consecutive months for more than 2 successive years combined with the presence of chest rhonchi and/or crepitations on physical examination. Chronic bronchitis was present in 14 of 40 exposed males, 11 of 50 nonexposed males, 15 of 54 exposed females, and 2 of 74 nonexposed females. The crude prevalence ratios for chronic bronchitis were 1.6 (95% CI =0.8–3.1) and 10.3 (95% CI=2.4–43.1) for males and females, respectively. After Mantel-Haenszel adjustment for sex, the prevalence ratio was 3.0 (95% CI=1.6–5.3). Although this study could have considerable recruitment bias and observer bias, it contributes limited evidence to the studies that suggest an adverse respiratory effect of chronic high-dose arsenic ingestion.

In an ecological study of mortality in an area of southwestern Taiwan where blackfoot disease is endemic (Tsai et al. 1999; see Diabetes section for description), the SMR for “bronchitis” increased significantly relative to a nearby reference population (SMR=1.53; 95% CI=1.30–1.80), and to all of Taiwan (SMR=1.95; 95% CI=1.65–2.29). The SMR for emphysema was not significantly different from either reference population. The authors noted that it is unlikely that differences in the rate of smoking account for the increased bronchitis mortality.

Hepatotoxic Effects

Hernandez-Zavala et al. (1998) studied liver function in individuals from three towns in the Region Lagunera of Mexico. The mean arsenic concentration in drinking water in each village was 14.0±3.1 µg/L, 116.0±37 µg/L, and 239.0±88 µg/L, and the corresponding mean total arsenic concentration in urine for each group (n=17 per village) was 88.0±27, 398.0±258, and 2,058.0±833 µg/L. The duration of exposure was not stated, but it was known that the middle-exposure town reduced the arsenic concentration (from 400 µg/L) in its water supply 3 years prior to the study. No subjects had recently consumed alcohol or had a history of chronic alcoholism. The mean concentrations of serum alkaline phosphatase and total bilirubin were significantly increased in individuals from the highest exposure town compared with the lowest- exposure town. Total urinary arsenic concentration was also correlated with those end points. Although data were not shown, the authors stated that those correlations were not significantly changed by adjusting for age, pesticide exposure, or history of alcohol or tobacco use or by examining separate correlations with urinary inorganic arsenic, MMA, or DMA. Serum transaminases (ALT, AST, and GGT) and albumin were normal and did not differ significantly between the groups. Twenty-six percent of the subjects from the middle- and high-exposure towns were reported to have at least one skin lesion associated with arsenic exposure. The potential impact of skin lesions as a covariate in the analysis was not reported. The impact of potential outliers on the results was not discussed. Although the correlations between urinary arsenic and both alkaline phosphatase and bilirubin concentrations observed in this study are interesting and might suggest an element of cholestasis, the biochemical findings noted are not specific for this hepatic end point. In the recent study by Santra et al. (1999), bilirubin or alkaline phosphatase increases were not a characteristic finding in 93 patients with firm hepatomegaly attributed to chronic arsenicosis in West Bengal, India. In that study, liver biopsy results from 69 patients with a clinical diagnosis of chronic arsenic poisoning revealed portal fibrosis in 63 (91.3%) cases, cirrhosis in 2 cases (2.9%), and normal histology in 4 (5.8%) cases. The degree of fibrosis was considered mild (grade 1) in 34 (53.9%) of the cases. Clinical evidence of portal hypertension (e.g., esophageal varices) was uncommon.

Hematological Effects

Hernandez-Zavala et al. (1999) also examined urinary porphyrin excretion and erythrocyte heme synthesis pathway enzymes in the same population described previously in the section Hepatotoxic Effects (Hernandez-Zavala et al. 1998). A dose- dependent increase in the ratio of urinary coproporphyrin III to coproporphyrn I and in the ratio of total coproporphyrin to total uroporphyrin was observed among the groups (17 individuals per group; mean urinary arsenic excretion, 88.0±27, 398±258, and 2,058±833 µg/L). Those ratios are in contrast to earlier findings by those authors in a population with exposures that were somewhat higher and of longer duration (Garcia Vargas et al. 1994). In that study, chronic high-dose arsenic ingestion was associated with inversions of the ratio of coproporphyrin to uroporphyrin (i.e., ratio <1) and of coproporphyrin III to coproporphyrin I. It is not clear whether differences in the intensity and temporal pattern of arsenic dose contributed to the disparate findings in those studies.

Diabetes

The main focus of research looking at the effects of arsenic on the endocrine system is the association between arsenic exposure and diabetes mellitus. The NRC (1999) report reviewed studies in Taiwan and Bangladesh that associated chronic ingestion of arsenic in drinking water with an increased risk of diabetes mellitus.

More recently, as part of an ecological study examining multiple causes of mortality in a section of the area of southwestern Taiwan where blackfoot-disease is endemic, Tsai et al. (1999) examined mortality from diabetes mellitus in four townships where artesian well water containing arsenic (median concentration of 0.78 mg/L (780 µg/L); from 0.25 to 1.14 mg/L) had been consumed from the early 1900s until the mid- to late-1970s. Observed mortality between 1971 and 1994 was compared with age and sex-specific expected mortality based on data from (1) a local reference group derived from two nearby counties, and (2) all of Taiwan. The local reference group was considered to be similar to the study group with respect to lifestyle factors; however, the drinking-water arsenic concentration of the local reference area was not stated. The extended time of follow-up resulted in a relatively large number of deaths being available for analysis. Within the high arsenic area, there were 11,193 recorded deaths during 1,508,623 person-years of observation for males and 8,874 recorded deaths during 1,404,759 person-years of observation for females. Diabetes mellitus (World Health Organization's International Classification of Disease (ICD) (CDC 2001) 8 and 9, code 250) was listed as the underlying cause of death for 188 males and 343 females. For males, the SMRs were 1.35 (95% CI=1.16–1.55) and 1.14 (95% CI=0.98–1.31), using the local and the national reference group, respectively. For females, the corresponding SMRs were 1.55 (95% CI=1.39–1.72) and 1.23 (95% CI= 1.11–1.37). Because only mortality attributed primarily to diabetes mellitus was examined, studies such as this one might underestimate an association between arsenic exposure and the population risk of the disease.

Tseng et al. (2000) recently reported the results of a prospective cohort study examining the incidence of diabetes mellitus in three villiages from the arsenic endemic area of southwestern Taiwan. The study population consisted of three villages where artesian well water (median arsenic concentration from 0.70 to 0.93 mg/L) was used for drinking and cooking until the mid-to-late 1970s. The study enrolled 632 subjects over 30 years of age who were determined not to have diabetes mellitus at the time the cohort was assembled (January and February, 1989). Details of the cohort are described by Lai et al. (1994). In 1991 and 1993, 446 of the cohort agreed to participate in a follow-up examination that included a fasting blood glucose and an oral glucose tolerance test. Diabetes mellitus was defined in accordance with the World Health Organization criteria. The incidence of diabetes mellitus in the study population was calculated as the total number of incident cases divided by the sum of follow-up person-time in all subjects. Data on each subject included age, sex, body-mass index, and an index of lifetime cumulative arsenic exposure (CAE). CAE (in units of milligrams per liter-years) was calculated as the product of the median arsenic concentration of the well water in every village that a subject inhabited at some point in his or her life multiplied by the length of time they consumed well water in that village. Individuals were excluded if complete arsenic exposure data were not available. Incidences for diabetes mellitus in the study population were compared with those reported for a demographically similar control population that was studied contemporaneously (Wang et al. 1997). Tseng et al. (2000) reported that during a follow-up period that included 1,499.5 person-years, 41 of 446 subjects developed diabetes mellitus (all noninsulin-dependent diabetes mellitus or type II diabetes mellitus). The incidence for new cases was particularly increased in subjects 55 years of age or older (50.8 per 1,000 person-years). The age-specific incidence density ratios were 3.6 (95% CI=3.5–3.6), 2.3 (95% CI=1.1–4.9), 4.3 (95% CI=2.4–7.7), and 5.5 (95% CI=2.2–13.5) for individuals 35–44, 45–54, 55–64, and 65–74 years of age, respectively. The relative risk for developing diabetes mellitus among those with more than 17 mg/L-years CAE compared with those with less than 17 mg/L-years CAE was 2.1 (95% CI=1.1–4.2), adjusted for age, sex, and body-mass index in a multivariate Cox proportional hazards model. When considered as a continuous variable, the CAE was associated with an adjusted relative risk of developing diabetes mellitus of 1.03 for every 1 mg/L-years of exposure (p < 0.05).

General Issues with Exposure Measurement

In reviewing relevant data developed on measuring arsenic exposure since the 1999 NRC report Arsenic and Drinking Water, it is worthwhile to begin by considering concepts of exposure and dose, because numerous definitions and methods of estimating exposure have been developed. Exposure and dose are related but separate concepts. Exposure to chemicals occurs when there is a chemical source, transportation of the chemical from the source to the subject, and contact between the chemical and the subject (i.e., ingestion of water or food, dermal contact, or inhalation). Dose refers to the amount of chemical transferred to the exposed subject. Dose should be described in terms of its relationship to the exposed subject, whether as the available, administered, absorbed, or active or biologically effective dose.

Measurement of the dose (as opposed to simply the fact of exposure) does not just provide the opportunity for a more detailed and informative description of the relationship between exposure and a disease; it is also important in inferring the absence or presence of a cause-and-effect relationship. One limitation of most epidemiological studies of arsenic is that dose is estimated from measuring the concentration of arsenic in drinking water and assuming an exposure based on drinking-water rates. Details of the exposure assessment conducted in each study are discussed below. The variability and uncertainty associated with the exposure assessment and their impact on a human-health risk assessment for arsenic are discussed in Chapters 4 and 5.

Chiou et al. 2001 Study

Chiou et al. (2001) conducted a prospective cohort study of 8,102 persons in the Lanyang basin of northeastern Taiwan, a region where arsenic-contaminated shallow wells were used for drinking water from the late-1940s through the mid-1990s. All cohort members had used shallow private wells for their primary drinking-water supply. Subjects were interviewed at home between October 1991 and September 1994. There were 4,586 homes represented in the study and 3,901 well-water samples taken and analyzed for arsenic with the use of hydride generation with flame atomic absorption spectrometry. Samples were not taken from 1,136 homes because their wells no longer existed. No biomarkers of arsenic exposure were used. The interview included information on history of well-water consumption, residential history, sociodemographic characteristics, cigarette smoking, personal and family medical history, occupation, and other potential risk factors. Subjects were followed for cancer incidence from the time of enrollment (1991 to 1994) through December 31, 1996, with the use of multiple resources. Detailed analyses of total urinary cancer (includes kidney, bladder, and urethral cancer) and specifically of the most common cell type of urinary cancer, transitional cell carcinoma (TCC), are presented. Nine subjects were diagnosed with bladder cancer, eight with kidney cancer, and one with both. Among those 18 subjects with urinary-tract cancer, 17 had pathological confirmation, and 11 were diagnosed with TCC. The most useful analyses were internal comparisons. Age- and sex-adjusted rates of urinary-tract cancer and TCC were first calculated within each exposure stratum. Incidences of urinary-tract cancer for subjects who drank well water with arsenic concentrations of 10.0 or less, 10.1–50.0, 50.1–100.0, and greater than 100.0 µg/L were 37.6 (three subjects), 44.8 (three subjects), 66.4 (two subjects), and 134.1 (seven subjects) per 100,000, respectively. The corresponding incidences for TCC were 12.5 (one subject), 14.9 (one subject), 66.4 (two subjects), and 114.9 (six subjects) per 100,000, respectively. When evaluated by duration of exposure to well water (<20.0, 20.1–39.9, and 40.0 years), rates for urinary cancer were 61.1 (one subject), 46.7 (four subjects), and 77.3 (10 subjects) per 100,000 persons, and rates for TCC were 0, 46.7 (four subjects), and 46.4 (six subjects) per 100,000, respectively. Cox's proportional hazards regression analysis (adjusted for sex, age, and cigarette smoking) was conducted to estimate relative risks and 95% confidence intervals by concentration and duration of arsenic exposure.

The multivariate relative risks for urinary cancer by concentration of arsenic in drinking water, using as a nonexposed referent group subjects with exposures at 10 µg/L or less, were 1.6 (95% CI=0.3–8.0) for arsenic concentrations at 10.1–50.0 µg/L, 2.3 (95% CI=0.4–14.4) at 50.1–100.0 µg/L, and 4.9 (CI=1.2–15.3) at greater than 100.0 µg/L. Corresponding relative risks for TCC were 1.9 (0.1–32.2), 8.1 (0.7–98.2), and 15.1 (1.7–138.5). In a time-window analysis, the increase in arsenic-induced TCC was more prominent for subjects who had drunk well water for more than 40 years, implying the possibility of a long latency between arsenic exposure and the occurrence of TCC and/or the importance of cumulative dose.

A strength of this study is that multiple follow-up resources were used to detect newly diagnosed (incident) cancer cases, ensuring that most newly diagnosed cancers were correctly identified. Other strengths are that residential and water-use histories were obtained from most cohort members and that individual information on other risk factors, such as cigarette smoking, was available. The authors used appropriate analytical techniques to estimate relative risks and 95% confidence intervals. The approach to exposure estimation has both strengths and weaknesses. The study was conducted in an area of Taiwan with a residentially stable population. Under the assumption that arsenic concentrations from household wells had been relatively constant over time, past individual exposures could be estimated by measuring arsenic in the current household well water of most cohort members. It should be noted that few data directly address the issue of arsenic-concentration consistency over time, and the question of historical consistency of arsenic concentrations in various types of groundwater deserves exploration. Consistency is of particular concern with shallow groundwaters, which might be subject to greater fluctuation than water from deeper wells. There were 4,074 study subjects (50.3%) who had drunk well water for more than 40 years, and over 70% of subjects greater than 50 years of age had drunk well water for more than 30 years. The average duration of drinking well water was 40.7 years. For 2,119 subjects (26.2%), more than one well had been used in their houses, and their past arsenic exposure data were derived from their current well data (Cantor 2001). Those 2,119 subjects had used their current well for approximately 60% of their life spans (standard deviation=21 years), and most had used their current well for at least 10 years (Chen and Chiou 2001). Although exposure assessment was individualized, for all subjects only a single well-water measurement obtained at one point in time was used to characterize long-term arsenic exposure.

Foremost among the limitations of this study is the short duration of follow-up, which limited the number of person-years of observation; hence, only a few cases of incident urinary-tract cancer were recorded: a total of 18 cases (15 with exposure information), of which 11 were TCC (10 with exposure information). In calculating relative risks, the small number of cases resulted in very wide CIs. For example, the adjusted relative risk of TCC in the highest exposure category (>100.0 µg/L) was 15.1, with a 95% CI of 1.7– 138.5. Thus, the study by Chiou et al. (2001) is limited by its relatively small numbers, and risk estimates based on these data might be too imprecise for use in a quantitative risk assessment. However, the data can serve as supplementary information, along with data from other selected studies. Since the cohort continues to be studied by the investigators, it is likely to yield more stable risk estimates in the future as more person-years of follow-up are accrued.

Ferreccio et al. 2000 Study

In a region of northern Chile with a history of increased concentrations of arsenic in drinking water, Ferreccio et al. (2000) conducted a case-control study of incident lung cancer. Eligible cases included all lung cancer cases diagnosed in the eight public hospitals in regions I, II, and III from November 1994 through July 1996. There were 217 eligible subjects identified, and 151 (70%) participated. Nonparticipation was largely due to inability to locate the subject. Controls were selected from the hospitals where the lung cancer cases were diagnosed. Two control series were selected: cancers other than lung cancer and noncancer controls. Another set of similarly identified controls was selected for a parallel study of incident bladder cancer. Cases and controls were interviewed regarding residential history, socioeconomic status, occupational history (to ascertain employment in copper smelting), and smoking.

An unusual characteristic of this study is the method used for control selection. Because control selection is a central element of the study design and might influence interpretation, it is worthwhile to describe the selection procedure. If it is assumed that most cases of lung cancer that occurred in the population were included in the study, an appropriate control group would be healthy persons selected randomly from the population at large, matched on age and sex. Instead, the authors selected hospitalized non-lung-cancer patients as controls. These patients could be more readily identified and interviewed. This technique has frequently been used in cancer epidemiology and can be an acceptable approach. In an effort to avoid overmatching by geographic area, and thus relative concentrations of arsenic in drinking water, the authors sought to select controls from the study hospitals in a manner that would reflect the distribution of arsenic exposure in the overall population from which the cases arose. To do that, the probability of selecting a control from each of the eight hospitals was based on the relative frequency of admission to that institution in 1991. Controls were also matched on sex and age within 4 years of the index case. Two hospital-based controls were selected for each case using this method: a patient diagnosed with a cancer that has not been related to arsenic exposure and a noncancer patient.

Historical exposure to arsenic in drinking water for each respondent was estimated by linking residential history information with a database of information on arsenic concentrations in public-water supplies collected for the years 1950 through 1994. Arsenic concentrations in the years prior to 1950 were based on concentrations in the 1950s. Average arsenic concentration in the place of residence was assigned to each subject on a year-by-year basis for the period of 1930–1994. Population coverage of public-water systems in the main cities in regions I and II was over 90% and was between 80% and 90% in the major cities of region III. The coverage in smaller cities varied between 64% and 91%.

The authors conducted a series of validation checks for the controls. Control- group distribution among hospitals was compared with the target distribution based on admissions in 1991. Major discrepancies were found, the main differences being from the main hospitals of Arica and Antofagasta, the two largest cities in the study area. There was a deficit in the proportion of observed-to-expected controls from the Arica hospital (0.6) and an excess from the Antofagasta hospital (1.2). In another comparison, the authors used the distribution of arsenic concentration in the period of 1958–1970 based on population figures from the 1992 census. Population numbers in 1992 were used to estimate an expected water-arsenic distribution for 419 randomly selected controls. This distribution was then compared with the actual distribution of the selected controls, and the ratio of the selected numbers of controls to that expected was calculated. The baseline exposure stratum (0–49 µg/L) showed a selected-to-expected ratio of 0.8. However, the high-exposure category of 400 µg/L and above (selected-to-expected=1.4) was overrepresented, owing to overselection of controls in Antofagasta, which had unusually high concentrations of arsenic in its drinking water during 1958–1970. Selected-to-expected ratios for intermediate groups were 1.3 for 50–99 µg/L and 0.5 for 100–399 µg/L. The expected impact of the overselection in the highest-exposure category is to diminish estimates of risk, and underselection in the lower stratum is to enhance risk estimates. As the authors noted, this assessment is indirect, because in the data analysis the actual historical residential location of cases and controls rather than their location at the time of the study was used to determine arsenic exposure levels. However, the direction of bias is probably as indicated.

Odds ratios were used to estimate relative risk of exposure to various concentrations of arsenic in drinking water relative to a referent concentration of 0–10 µg/L. Odds ratios were calculated using unconditional logistic regression, adjusted for age, socioeconomic status, smoking, and working in a copper smelter. In separate analyses, arsenic in drinking water was expressed as average yearly concentration during the peak years of exposure 1958–1970 or as average yearly concentration in the period of 1930–1994. In Antofagasta, there was a peak in exposure during 1958–1970, when arsenic concentrations averaged 860 µg/L. When 1958–1970 average arsenic concentrations were used as the estimate of exposure, the odds ratios were 5.7 (95% CI=1.9–16.9) for the 400–699 µg/L exposure stratum and 7.1 (95% CI=3.1–12.8) for 700– 999 µg/L, relative to 0–10 µg/L. Results from the analysis based on average exposures during 1930–1994 and using all controls (419) showed an increase in the odds ratio with arsenic concentration, reaching an odds ratio of 8.9 (95% CI=4.0–19.6) for the highest exposure group (average exposure 200– 400 (µg/L). Results from calculations using cancer controls (n=167), noncancer controls (n=252), or controls frequency-matched to lung cancer cases (n =237) were of similar magnitude. Risk estimates based on long-term average exposure were higher per unit exposure than those that used the peak period of exposure (the mid-point of which is about 30 years before diagnosis of study cases). It is not clear at this time whether it is appropriate to give more weight to the risks based on 40+ years of exposure or to risks based on the 12-year peak exposure period about 30 years prior to diagnosis. When the study population was stratified by smoking status, there was suggestive evidence of a synergistic interaction between smoking and exposure to arsenic in drinking water. Relative to nonsmokers with average arsenic exposures of 49 µg/L or less (1930–1994), nonsmokers with average exposures of 200 µg/L or more had an odds ratio of 8.0 (95% CI=1.7–52.3), whereas smokers with average exposures of 200 µg/L or more had an odds ratio of 32.0 (95% CI=7.22– 198.0).

Strengths of this study include an acceptable response rate, unbiased ascertainment of exposure, individual estimates of exposure, exposure coverage of most of the life span for most study subjects, incorporation of individual data on other potentially confounding risk factors for lung cancer, appropriate analyses of study data, and an adequate study size. This is the only study available for risk assessment that has individual estimates of exposure on all subjects for more than 40 years, well beyond the minimum latency for lung cancer. The major limitation of the study, as discussed, is related to methods used for control selection. The data from this study can be used in a quantitative assessment of risk of arsenic in drinking water, along with data from other selected studies.

Lewis et al. 1999 Study

Lewis et al. (1999a) conducted a retrospective cohort mortality study of 4,058 residents of Millard County, Utah, a region where drinking water is derived from wells and where arsenic concentrations in well water range from undetectable to up to a few hundred micrograms per liter. In a letter to the editor (Lewis et al. 1999b), the purpose of the study is described as follows: “We were interested in determining whether studies of health effects related to arsenic in drinking water could be conducted in U.S. populations exposed to relatively low concentrations of arsenic.” The study cohort included two groups: 2,073 persons from an earlier study (Southwick et al. 1982) and 1,985 additional individuals identified from special census records and other records maintained by the Church of Latter Day Saints (Mormons). The earliest entry in the cohort was in 1900, and the most recent was in 1945. Cohort members had lived in the towns of Delta (1,191 persons; 29.4%), Hinckley (1,192 persons; 29.4%), or smaller towns, such as Deseret, Abraham, or Oasis (1,675 persons; 41.2%). Vital status follow-up was through November 27, 1996. At the closing date, 38.2% (1,551) were alive, 54.3% had died (2,203), and 7.4% (300) were lost to follow-up. All death certificates were coded according to rubrics of the ICD-9, and SMRs were calculated by cause of death, separately for males and females in each of three exposure groups, as described below, and for all exposure groups combined. The SMR is the cause-specific ratio of observed-to-expected numbers of deaths, where the expected number is calculated for the age and sex distribution of the study population by using age- and sex-specific rates of a comparison population. In calculating SMRs, the comparison was sex-specific rates for the state of Utah. That metric was calculated by multiplying the concentration of arsenic in the drinking water in parts per billion (micrograms per liter) by the number of years of exposure to yield an exposure with units of parts per billion-years, arranged into three groups: <1,000, 1,000–4,999, and 5,000 ppb-yr. This metric represents an estimate of cumulative arsenic exposure for years of residence in study towns. Information on water intake was not available. For several causes of death, statistically significant increases in SMRs occurred in a generally uneven pattern based on exposure level. Among men, the causes of death included hypertensive heart disease, nephritis, nephrosis, and prostate cancer. Among women, they were hypertensive heart disease and all other heart disease (pulmonary heart disease, pericarditis, and other diseases of the pericardium).

This study has several strengths. Vital status was determined for all but a small number of cohort members (7.4%), selection into the cohort was unbiased with respect to outcome, and numbers of cohort members and deaths within the cohort were adequate to evaluate risk for major causes of death.

Several of the limitations concern the estimation of exposure and the consequences for study interpretation. Lewis et al. (1999a) present results of 151 arsenic analyses of drinking water in public and private water supplies for the study towns from a survey of drinking-water wells conducted by the Utah State Health Laboratory since 1976. These data “were used in assessing the potential exposure of cohort members to arsenic in drinking water” (Lewis et al. 1999a). Lewis et al. (1999a) present the number of samples taken in each town and the average, median, minimum, and maximum amounts of measured arsenic. The number of wells that contribute to the municipal water supply of each town is not presented, nor is it stated whether multiple measurements in each town represent repeat samples taken from one well, multiple samples from different wells, or a combination of both. For each year a cohort member lived in one of the study communities, he or she was assigned the median concentration of measured arsenic for the wells of that community. Presumably, exposure that occurred during periods when a person did not live in one of the study communities was not included in the estimate of exposure used in the analysis, creating the possibility that cumulative lifetime arsenic exposure might have been incompletely ascertained for some of the cohort.

The range of arsenic concentrations reported in Lewis et al. (1999a) for each town was quite broad. For example, arsenic concentrations measured since 1976 in Hinckley ranged from 80 to 285 µg/L; in Delta, 3.5 to 125 µg/L; and in Deseret, 30 to 620 µg/L. However, the median concentration was assigned to individual cohort members during their years of residence. Use of a median level to estimate exposure when the concentrations vary so widely necessarily resulted in misclassification of exposure for many subjects. A similar approach to estimating exposure was taken in the ecological studies from southwestern Taiwan of Chen et al. (1985) and Wu et al. (1989) that were used in the EPA risk assessment. The estimates of exposure from the Taiwanese studies might be more representative of actual exposures than the estimates from Utah. In Utah, arsenic samples came from different types of water supplies, both public and private.

The exposure metric used for the analysis (parts per billion-years) was the product of water arsenic concentration (in parts per billion) and duration of exposure (in years). One consequence of using this exposure metric, which inextricably combines intensity and temporal aspects of exposure, is that comparisons of results with findings of many other studies that used measurement of average arsenic over extended periods are not possible. Another consequence of using this metric is that persons with very different exposure histories are grouped in the same exposure stratum during data analysis. For example, a value of 1,000 ppb-yr could result from exposure to 20 µg/L for 50 years or 200 µg/L for 5 years. Such strikingly different exposure scenarios could have very different health consequences. Other studies (e.g., Chiou et al. 2001) suggest that average arsenic concentration over a long period is more strongly associated with risk than is the duration of exposure to increased concentrations, suggesting the need to consider concentration and duration of exposure independently. The duration of exposure for cohort members is not explicitly presented and might have been relatively brief for some.

In addition to limitations regarding exposure ascertainment, the subcommittee noted a few epidemiological issues. As the authors recognized in the discussion, “SMRs cannot be directly compared in an analysis that uses indirect adjustment,” especially where the distributions of age among groups are not comparable. There was a significant difference in the age distribution of the three exposure groups, indicating that comparison of SMRs across exposure groups is not appropriate. The difference in age distribution is not surprising among exposure groups in which the exposure metric itself is age-dependent. Older people would be likely to have higher cumulative exposure. “Based on this, any conclusions on whether arsenic is an etiologic factor in consideration of increased or decreased SMRs among the groups is uncertain” (Lewis et al. 1999a).

Comparison rates used in the analysis of risk were for the state of Utah, and this comparison likely resulted in underestimates of risk for some causes of death and overestimates for others. The study cohort was composed of Mormons, with strict religious prohibitions against smoking and consumption of beverages containing alcohol or caffeine. In addition, the study population was largely rural. As the study authors noted, smoking rates for the state of Utah are low, around 12–13%. However, even this relatively low smoking prevalence is expected to result in rates of several cancers (lung, bladder, kidney, and pancreas) and cardiovascular diseases that are increased relative to a nonsmoking population, such as the study cohort. That factor and the rural setting of the cohort were possible contributors to the deficits observed in SMRs for urinary and pulmonary cancers.

A reanalysis of the study data for bladder and lung cancers, without a comparison population, was conducted by the EPA Office of Water (EPA 2000) using Cox proportional hazards regression. The Lewis et al. (1999a) study identified a total of 34 respiratory cancers, of which 30 were lung cancers, and a total of five bladder cancers. The risk detected in the EPA reanalysis was not statistically distinguishable either from zero or from the levels predicted by model 1 of Morales et al. (2000) and used by EPA in its arsenic risk analysis (EPA 2000). In addition to limitations in exposure assessment, it is apparent that the power of the Utah study is too low to allow a precise enough estimate for use in quantitative risk assessment. It is possible that further exploration of the exposure assessment used for this study would be fruitful. If so, then additional follow-up and reanalysis using an internal referent is to be encouraged. In summary, although this study in its current state has several strengths, several limitations preclude its use for quantitative risk assessment.

Tsai et al. 1999 Study

Sex-specific mortality due to several cancer and noncancer causes in the area of southwestern Taiwan where blackfoot-disease is endemic was evaluated by Tsai et al. (1999) for the years 1971 through 1994. Formerly, this area had high concentrations of arsenic in drinking water. Standard mortality ratios were calculated twice using two referent groups: the first referent was the mortality experience of the whole of Taiwan and the second referent was the mortality experience of the two counties of southwestern Taiwan where blackfoot disease is endemic. This study has special significance because it speaks to the issue of possible factors, such as differences in diet, cultural background, smoking, occupational exposures, access to medical care, or other differences (other than arsenic exposures) between the population of southwestern Taiwan and the remaining population of the country that might have influenced cancer mortality. In particular, it has been argued that differences in general nutritional status or selenium intake between the relatively poor farming and fishing communities of southwestern Taiwan (the locale of high arsenic exposures from water and the studies used for the EPA risk assessment) and the remainder of the country were great enough to result in increases of risk that might erroneously be attributed to drinking-water arsenic exposures. Those concerns apparently led EPA, in conducting its risk assessment, to reject the notion that the mortality experience of the whole Taiwanese population was appropriate as a low-exposure referent. Although there remains some possibility that population differences in nutrition, socioeconomic status, or other factors between southwestern Taiwan and the remainder of the country have some influence on their respective cancer rates, results from the sex- and cancer-specific calculations of Tsai et al. (1999) provide evidence that such differences are relatively unimportant. As examples, the subcommittee used the SMRs from Tsai et al. (1999) for three cancer sites of interest—lung, bladder, and kidney cancers among males and females—that were calculated using regional and national cancer rates, respectively, as the referent. Among males in the area where blackfoot-disease is endemic, there were 699 deaths due to lung cancer. Using regional rates as the referent, the lung cancer SMR for males was 3.10 (95% CI=2.88–3.34), and using national rates, the SMR was 2.64 (95% CI=2.45–2.84). There were 312 male deaths due to bladder cancer. Using regional rates as the referent, the SMR for bladder cancer was 8.92 (95% CI=7.96–9.96), and using national rates, the SMR was 10.50 (95% CI=9.37–11.73). There were 94 male deaths due to kidney cancer, the corresponding SMRs being 6.76 (95% CI=5.46–8.27) using regional rates, and 6.80 (95% CI=5.49–8.32) using national rates. Among females, there were 471 deaths due to lung cancer. The SMR for women was 4.13 (95% CI=3.77–4.52) using regional rates as the referent, and the SMR was 3.50 (95% CI=3.19–3.84) using national rates. There were 295 female deaths due to bladder cancer. The SMR for bladder cancer was 14.07 (95% CI=12.51–15.78) using regional rates, and the SMR was 17.65 (95% CI= 5.70–19.79) using national rates. There were 128 kidney cancer deaths among women. SMRs for kidney cancer mortality were 8.89 (95% CI=7.42–10.57) using regional rates and 10.49 (95% CI=8.75–12.47) using national rates. As noted by Tsai et al. (1999), the regional referent population was very similar to the area where blackfoot-disease is endemic in lifestyle factors, and thus the similarity of SMRs analyzed with either the regional control population or the national population indicates that significant confounding was unlikely.

Finally, it is interesting to note that the study by Tsai et al. (1999) determined that the area of southwestern Taiwan where arsenic is endemic experienced significant increases in mortality from many other cancers besides those commonly linked to arsenic. It is possible that the relatively high arsenic exposure of the study population and the large size of the study in terms of person-years of observation might have enabled these statistically significant increases to emerge.

Kurttio et al. 1999 Study

Water samples from drilled wells in some regions of Finland have been reported to have high concentrations of inorganic arsenic. To test whether drinking water from such sources was associated with risk of bladder or kidney cancer, Kurttio et al. (1999) conducted a population-based case-cohort study in Finland of 61 bladder cancer and 49 kidney cancer cases. Cases, diagnosed in the period of 1981–1995, were selected from among Finnish residents of places not served by municipal water supplies who had used drinking water from their own drilled wells between 1967 and 1980 and for whom a water sample was analyzed for arsenic concentrations. A reference group of 275 persons, matched by age and sex to the combined case series, was selected from the same population. Contrary to expectation, arsenic concentrations in the cancer group and the control group were relatively low. Among the 61 bladder cancer cases, 42 drank water from wells with arsenic below 0.5 µg/L. The 95th percentile of the distribution of arsenic in well water among bladder cancer cases was 3.0 µg/L, and among the referent cohort, 4.5 µg/L. In one analysis, the study population was stratified by period of exposure—one group with exposure in the third to ninth calendar year before cancer diagnosis (shorter latency) and another group with exposure in the tenth calendar year and earlier before diagnosis (longer latency). Among persons in the shorter latency group, a relative risk of 2.44 (95% CI=1.11– 5.37, 19 cases) was observed for an average arsenic exposure of 0.5 µg/L or more, relative to persons with average arsenic exposure of less than 0.1 µg/L. Relative risks for the comparable exposure groups among persons with longer latency were not significantly above one. Increased risk with arsenic exposure was confined to persons who smoked during the 1970s. Among the shorter latency group, smokers with average arsenic concentrations of 0.5 µg/L or above had a relative risk of 10.3 (95% CI=1.16–92.6); the relative risk among nonsmokers in this arsenic exposure group was 0.87 (95% CI=0.25–3.02). Both calculations were relative to people (smokers or nonsmokers, respectively) with arsenic concentrations of less than 0.1 µg/L.

The finding of increased bladder cancer risk in this study is striking, as the risk is far above that expected at relatively low concentrations of arsenic exposure. Given the small number of bladder cancer cases in the study (61) and the likelihood that much more arsenic would come from food than water in this low-exposure population, that finding might be the result of a chance observation or an unmeasured bias.

Karagas et al. 2001 Study

Using levels of arsenic in toenails as a biomarker of internal exposure, Karagas et al. (2001) conducted a case-control study of 587 basal-cell carcinoma cases (BCC) and 284 squamous-cell carcinoma cases (SCC) in New Hampshire, where private wells are used by about 40% of the population as the primary source of drinking water. Although the arsenic concentration in a majority of New Hampshire wells is below 1.0 µg/L, the concentration in many underground water sources is higher, with concentrations above 50 µg/L in some instances. Earlier, Karagas et al. (2000) had demonstrated a correlation (r) between arsenic in toenails and arsenic in water in persons exposed to arsenic at 1.0 µg/L or more of water (r=0.65, p < 0.001). In the case-control study, controls were randomly selected from the general population of New Hampshire, matching on age and sex to the overall distribution of the two case series. Cases and controls were interviewed at home starting in January 1994. The questionnaire included questions on type of water supply and the number of years of use. Toenail clippings were collected at the time of the interview and analyzed using instrumental neutron activation analysis for arsenic concentration. Standard logistic regression methods were used to calculate odds ratios for BCC and SCC relative to toenail arsenic concentration, as stratified into six strata according to percentile of the overall distribution among controls. The odds ratios for SCC and BCC were close to unity in all but the highest category. Among subjects with toenail arsenic concentrations above the 97th percentile, the adjusted odds ratios were 2.07 (95% CI=0.92–4.66) for SCC and 1.44 (95% CI=0.74–2.81) for BCC, compared with those with concentrations at or below the median.

The concentration of drinking-water arsenic that corresponds to the 97th percentile lower cutoff level for the highest stratum of toenail arsenic in this study (0.345 µg/g), where increased odds ratios were observed, is approximately 30–50 µg/L (estimated by reference to the earlier publication of Karagas et al. 2000), showing a correlation between water and toenail arsenic concentrations in controls in that study. The highest level of toenail arsenic among controls was 0.81 µg/g, and given the overall log-normal distribution of toenail arsenic in this population, the median level of toenail arsenic in this stratum was likely to be about 0.450 µg/g, corresponding to approximately 100–150 µg/L of drinking water.

There are several uncertainties regarding the exposure assessment in this study. Arsenic bound to the toenail matrix can result from external exposure as well as intake of inorganic arsenic in food and water. These and other factors result in significant variability in toenail arsenic levels among individuals with similar exposure to arsenic in drinking water. Little is known about the toxicokinetics of arsenic uptake into toenails. In the setting of this study, that lack of knowledge leads to uncertainty about the concentration of arsenic in drinking-water exposures corresponding to toenail arsenic concentrations at which increased risk was observed. Arsenic in toenail clippings typically integrates exposures over a few week period about a year prior to sample collection. The latency for skin cancer is not well defined, but appears to be more than a decade. Thus, the time period of exposures reflected in toenail arsenic does not correspond to the critical exposure period for skin cancer. Over 50% of the study population had used the same water supply for over 15 years. Relative-risk estimates did not vary significantly according to length of use by subjects, further complicating interpretation of study results. An additional limitation of this study is the relatively low control participation rate, 50%, which might have resulted in a biased sample from the general population.

Tucker et al. 2001 Study

Tucker et al. (2001) analyzed data describing prevalent skin cancer and other skin disorders (hyperkeratoses and dyspigmentation) from a cross-sectional study in a region of Inner Mongolia with increased concentrations of arsenic in drinking water. A total of 3,179 persons in three villages were examined in 1992, and well-water-use histories for these individuals were gathered. Water samples were collected from 184 of the 187 local wells in these villages and analyzed for arsenic content. The median age of participants was 29 years, and the average well-use history was 25 years. In the study population, 79.6% were less than 50 years of age. Arsenic in drinking water ranged from below detection (10 µg/L) to 2,000 µg/L. Skin cancer was observed in eight subjects. In addition, 172 subjects had keratosis, 121 had dyspigmentation, and 94 subjects were diagnosed with both types of lesions. Among the skin cancer cases, all eight had both keratosis and dyspigmentation. Two exposure metrics were used in the data analysis: peak arsenic concentration (PAC), defined as the highest well-water arsenic concentration reported by an individual, and cumulative arsenic dosage (CAD), estimated by multiplying the arsenic concentration by the duration of well use for each well reported to have been used by each subject and then adding the products together to get cumulative exposure. Thirty-five percent of the study population had a PAC exposure of less than 50 µg/L, and 86% had a PAC exposure of less than 150 µg/L. Several statistical models (frequency weighted, simple linear regression, hockey stick, and maximum likely estimate) were used to analyze data for each of the effects noted. Two measures of exposure were used with each model.

Dose-response relationships were found for skin cancer and the other health end points, using both metrics of exposure to arsenic in drinking water. Several of the statistical models used in the analysis of these data used an a priori assumption of a threshold exposure concentration of arsenic in drinking water below which skin cancer (and other dermatological effects) would not occur. The data appear to be adequately described by such statistical models; however, they are also well-described by a nonthreshold linear model. That is not surprising in view of the small number (eight) of skin-cancer cases observed. The authors do not discuss whether differences in results among the various models used to describe the data were statistically significant.

There are limitations in the two exposure metrics (CAD and PAC) that were used in this study. Limitations in using CAD as a measure of exposure are described above in the discussion of the Lewis et al. (1999a) study. The other exposure metric, PAC, was used without reference to the time of peak exposures. That lack of information is problematic, because many of the study subjects might have used more than one well during relevant exposure periods, and the peak exposures might have occurred many years before this cross-sectional study and for as little as 1 year.