epidemiological findings, experimental data on mode-of-action-related end points, and available information regarding the anticipated variability in human susceptibility.
At present, studies from the arsenic endemic area of Taiwan continue to provide the best available empirical human data for use in assessing the dose-response relationship for arsenic-induced cancer. The current state of knowledge is insufficient to reliably apply a biologically based model to those data. In accordance with EPA's "Proposed Guidelines for Carcinogen Risk Assessment" (EPA 1996), the subcommittee reviewed modes of action based on markers of tumor response and on available data that can determine the shape of the dose-response curve in the range of extrapolation. As discussed in Chapter 7, the several modes of action that are considered most plausible would lead to a dose-response curve that exhibits sublinear characteristics at some undetermined region in the low-dose range. Nonetheless, in the context of its task, the subcommittee considered the magnitude of the likely cancer risks within the range of human exposure at approximately the current MCL.
In vitro studies of the genotoxic effect of submicromolar concentrations of arsenite on human and animal cells and one study of bladder-cell micronuclei in humans with arsenic concentrations of 57 to 137 µg/L in urine indicate that perturbations in cellular function related to plausible modes of carcinogenesis might be operating at arsenic exposure concentrations associated with the current MCL. The subcommittee believes that those data and the confidence with which they can be linked to arsenic-induced neoplasia are insufficient to determine the shape of the dose-response curve between the point of departure and the current MCL. The subcommittee also finds that existing scientific knowledge regarding the pattern of arsenic metabolism and disposition across this dose range does not establish mechanisms that mitigate neoplastic effects. In light of all the uncertainties on mode of action, the current evidence does not meet EPA's stated criteria (EPA 1996) for departure from the default assumption of linearity in this range of extrapolation.
In Chapters 2 and 10, the subcommittee reviewed the strengths and limitations of the Taiwanese data. Chapter 10 also discussed the implications of applying different statistical models to the Taiwanese internal-cancer data for the purpose of characterizing cancer risk at the current MCL in the United States. With respect to EPA's 1988 risk assessment for arsenic-induced skin cancer in which the multistage Weibull model was used, a sensitivity analysis, within the limits of the available data, suggests that misclassification arising from the ecological-study design and the grouping of exposures would likely have only a modest impact on EPA's risk estimates. Sensitivity analyses applied to male bladder-cancer risk estimated by the multistage Weibull model had a greater impact on results. However, a more stable and reliable fit was provided by Poisson regression models that characterized the log relative risk