A review of low-dose arsenic risks and human cancers.

The linear no-threshold (LNT) model has historically been the default assumption in assessing carcinogenic risk from arsenic ingestion based on epidemiological studies. This contrasts with the threshold model used in assessing carcinogenic risk from arsenic ingestion derived from toxicological investigations of experimental animals. We present here a review of our epidemiological work that has examined models that may better explain the human cancer risk from the ingestion of arsenic, particularly from low level exposures, than does the LNT model. While previous epidemiology studies have demonstrated increased risks of bladder, lung, and skin cancers at arsenic exposures of 200 ug/L or greater, we seek here to examine the dose-response patterns at lower exposure levels. These include ecological, case/control, and cohort designs. Methodologic issues include choice of continuous or stratified analysis of exposure data, search for sources of non-conformity or variability, and distinctions in water sources and geography. Multiple studies have yielded useful data-based models, including threshold models, hockey-stick models, and "J-shaped" linear-quadratic models. These models have found that increased cancer risk may only begin at specific arsenic exposure levels greater than zero. These results provide guidance in seeking toxicological explanations and public health reference levels.