Pervasive uncertainty is a dominant analytical difficulty that continues to hinder the EPA's risk assessment process for setting standards for environmental contaminants, particularly within the dose-response step.
Currently, the EPA handles this by applying deterministic factors referred to as safety or uncertainty factors. This approach has long been criticized as arbitrary, obscuring the true uncertainty, and limiting the ability of policy-makers to make adequately informed risk management decisions. We propose a hierarchical Bayesian model approach to synthesize evidence from toxicological and epidemiological studies, allowing for explicit statement of uncertainty assumptions in the prior distributions, and pre-processing data using Bayesian Model Averaging (BMA) to account for model uncertainty. We apply this model to a case study of chloroform, a disinfection byproduct, in drinking water. We use the same data set considered by the EPA when setting their regulatory standards for chloroform, exploring four different health outcomes that were either cancer or considered pre-cursors to cancer. Final model estimates demonstrated that incorporating more scientific information into the priors had minimal impacts on mean estimates, but reduced the uncertainty surrounding the final estimates.
Benchmark dose (BMD) and lower-bound benchmark (BMDL) dose estimates from the model were mostly lower than those estimated by the EPA, indicating that not considering the full body of scientific evidence fails to capture the true uncertainty surrounding the final estimate. As a result, Maximum Contaminant Level Goal (MCLG) estimates using the Bayesian model were consistently lower than EPA estimates, and in particular were lower than the MCLG standard for chloroform currently in place. This Bayesian model provide an alternative approach to incorporating and quantifying various sources of uncertainty in the dose-response step, and may be applicable in other aspects of risk assessment.
No comments:
Post a Comment