Background: Although studies have reported associations between high concentrations of ingested inorganic arsenic and diabetes mellitus, there is no evidence of this association at low exposures. However, a well-publicized study (JAMA. 2008;300:814–822) recently produced an extraordinary finding of a more than 3-fold increase in diabetes at low concentrations of urinary arsenic. This potentially affects 40 million adults in the United States.
Methods: We used the same cross-sectional data on urinary arsenic and type 2 diabetes mellitus in 795 adults from the 2003–2004 National Health and Nutrition Examination Survey to assess this evidence.
Results: As in the earlier study, we found an odds ratio (OR) near 1.0 for diabetes, comparing the 80th versus 20th percentiles of urinary total arsenic (OR = 0.88 [95% confidence interval = 0.39–1.97]). This OR increased to above 3.0 when urinary arsenobetaine was added to the logistic risk model. However, this high OR was a statistical artifact because arsenobetaine, which is ingested from fish and is essentially nontoxic, is a part of measured total urinary arsenic. These 2 variables are highly correlated (correlation = 0.80). Because arsenobetaine is a part of total arsenic, it should first be subtracted from total arsenic rather than being added to the statistical model. Doing so yields an OR of 1.15 (0.53–2.50).
Conclusion: These findings show no evidence of increased risk of diabetes with arsenic exposure in this dataset. This underscores the importance of valid statistical techniques and careful consideration of scientific plausibility when investigating low-concentration chemical exposures.