Quantitative assessment of lung and bladder cancer risk and oral exposure to inorganic arsenic: Meta-regression analyses of epidemiological data

Inorganic arsenic (iAs) in drinking water varies geographically and is prevalent worldwide. While exposures in the US are generally low, there are some areas with higher levels of naturally occurring iAs (potentially> 100 mu g/L) where residents rely on unregulated drinking water wells. Much of the evidence on the association between iAs and cancer comes from epidemiological studies conducted in South American and Asian populations. These populations have generally been exposed to much higher levels of iAs and have differing underlying characteristics, both of which make comparing them to Western populations difficult. A key question is whether and how one should extrapolate from these high exposure studies to estimate cancer risk at lower exposures. We conducted an independent analysis to determine the most appropriate cancer endpoints, studies, and models to support an oral carcinogenicity assessment of iAs, taking into consideration factors that affect the apparent potency of iAs across geographically and culturally distinct populations. We identified bladder and lung cancer as high-priority endpoints and used meta-regression to pool data across studies from different regions of the world to derive oral cancer slope factors (CSFs) and unit risks (excess risk per mu g/L) for iAs based on the background risks of bladder and lung cancer in the US. We also calculated concentrations of iAs in water that are not likely to result in cancer risk above what is considered acceptable by the United States Environmental Protection Agency (US EPA). While we derived these factors assuming a linear, no-threshold relationship between iAs and cancer risk, we also evaluated the shape of the dose-response curves and assessed the evidence for overall nonlinearity. Overall, we found that the incremental risks of bladder and lung cancer associated with iAs were relatively low. The sensitivity analyses we conducted suggested that populations with relatively high iAs exposures appeared to drive the pooled cancer risk estimates, but many of our other tested assumptions did not substantially alter these estimates. Finally, we found that the mode of action evidence supports there being a threshold, but making a robust quantitative demonstration of a threshold using epidemiological data is difficult. When considered in the context of typical exposure levels in the US, our potency estimates indicate that iAsinduced cancer risk is much lower than observed bladder and lung cancer incidences. This suggests that the low iAs levels to which much of the general US population is exposed likely do not result in substantial additional cancer risk.