Table 1.

Association between arsenic and diabetes in areas of relatively high exposure (≥ 150 µg/L drinking water).

Reference (study design)	Location, subjects	Diabetes diagnosis	Main findinga,b	Exposurec	Factors considered in analysis
Chen et al. 2010 (cross-sectional)	Bangladesh (Araihazar) HEALS, n = 11,319 ♂♀	Self-report prior to baseline	1.11 (95% CI: 0.73, 1.69) adjOR	176.2–864 (Q5) vs. 0.1–8 (Q1) µg As/L drinking water, CEI Cohort: 0.1–864 µg As/L	Age, sex, BMI, smoking status, educational attainment
Lai et al. 1994 (cross-sectional)	Taiwan (southern) As-endemic region, n = 891 ♂♀	Self-report, OGTT, treatment history	10.05 (95% CI: 1.3, 77.9) adjOR	≥ 15 vs. 0 ppm-year drinking water, CEI Cohort: 780 (700–930) µg As/L; median (range) concentrations in artesian wellsd	Age, sex, BMI, physical activity
Nabi et al. 2005e (case–control)	Bangladesh (Chapainowabganj) arsenicosis cases, n = 235 ♂♀	Glucose, blood	2.95 (95% CI: 0.954, 9.279) OR	218.1 vs. 11.3 (mean) µg As/L drinking water Cohort: 218.1 (3–875) µg As/L; mean (range)	Unadjusted
Rahman et al. 1998f (cross-sectional)	Bangladesh (Dhaka) keratosis cases, n = 1,107 ♂♀	Self-report, OGTT, glucosuria	5.2 (95% CI: 2.5, 10.5) adjPR	Keratosis vs. non-keratosis Cohort: < 10–2,100 µg As/L	Age
Rahman et al. 1999f (cross-sectional)	Bangladesh (multisite) with skin lesions, n = 430 ♂♀	Glucosuria	2.9 (95% CI: 1.6, 5.2) adjPR	> 10 vs. < 1 mg-year As/L drinking water, CEI Cohort: < 500 to > 1,000 µg As/L drinking water	Age, sex, BMI
Tsai et al. 1999e (retrospective)	Taiwan (Chiayi County) Blackfoot region, n = 19,536 deaths ♂♀	Death certificate	1.46 (95% CI: 1.28, 1.67) SMR	Blackfoot endemic region vs. national reference Cohort: 780 (250–1,140) µg As/L; median (range)	Age, sex
Tseng et al. 2000a, 2000b (prospective)	Taiwan (southwestern) agricultural and aquacultural regions, n = 446 ♂♀	Fasting blood glucose, OGTT	2.1 (95% CI: 1.1, 4.2) RR	≥ 17 vs. < 17 mg/L-year As (drinking water, CEI) Cohort: 700–930 µg As/L; range of median concentration in artesian wells	Age, sex, BMI
Wang SL et al. 2003g (cross-sectional)	Taiwan (southwestern) As-endemic region, n = 706,314 ♂♀	Insurance claims	2.69 (95% CI: 2.65, 2.73) adjOR	Endemic vs. non-endemic region Cohort: 780 (350–1,140) µg As/L; median (range)d	Age, sex
Abbreviations: adjOR, adjusted odds ratio; adjPR, adjusted prevalence ratio; As, arsenic; BMI, body mass index; CEI, cumulative exposure index; HEALS, Health Effects of Arsenic Longitudinal Study; mg-year, milligram year; OGTT, oral glucose tolerance test; OR, odds ratio; Q, quintile; RR, relative risk; SMR, standardized mortality ratio. aIdentification of main findings was based on the following strategy: for studies that did not report a significant association between arsenic exposure and a health outcome at any exposure level, the main summary finding was based on the highest exposure group compared to the referent group (e.g., 4th quartile vs. 1st quartile). When a study reported a significant association between arsenic exposure and a health outcome, the main finding was based on lowest exposure group where a statistically significant association was observed (e.g., 3rd quartile vs. 1st quartile). bUnless specified, relative risk estimates are crude estimates. cMedian or mean and range of As concentration in drinking water for the cohort is included when reported. dArsenic drinking-water concentrations were taken from other publications based on same populations. eCalculated by entering data presented in publication into OpenEpi software (Dean et al. 2011). fAlthough the arsenic water concentrations are expressed in units of mg/L, the value is supposed to represent the “approximate time-weighted mean arsenic exposure levels that were calculated over the lifetime of each subject as ∑j(ajcj/∑jaj, where aj is the number of years a well with arsenic concentration cj was used, assuming that the current levels of arsenic in the well water were also representative of the past source.” gThere appears to be an error in the number of persons included in the “non-endemic” area category based on the ns provided in Table 1 of Wang et al. 2003.