Use of probabilistic expert judgment in uncertainty analysis of carcinogenic potency

A new approach to characterizing the state of knowledge about carcinogenic potency is described. In this approach, the carcinogenic risk posed by a specific dose is characterized by a probability distribution, indicating the relative likelihood of different risk estimates. The approach utilizes expert judgment and a probability tree and is illustrated in a case study of chloroform exposure. Experts in cancer biology/toxicology, pharmacokinetics, and dose-response modeling were identified by a panel of science-policy specialists. In a workshop, experts reviewed the chloroform data, received training in probability elicitation, and constructed a consensual probability tree based on biological theories of cancer causation. Distributions of carcinogenic risk were developed based on the probability tree, chloroform data, judgmental probabilities provided by the experts, and classical statistical techniques. Risk distributions varied considerably between experts, with some predicting essentially no risk from 100 ppb chloroform in drinking water while other have at least some probability on risks generally considered of regulatory significance. Estimated human risk was much lower when extrapolating from liver tumors in animals than from kidney tumors. Issues of scientific disagreement leading to different risk distributions between experts are discussed. The resulting risk distributions are compared to standard EPA risk calculations for the same exposure scenario as well as to the expert judgement of epidemiologists about cancer risks of chlorinated drinking water. Issues in combining expert judgments are discussed, and several alternative methods are presented. Strengths and weaknesses of the distributional approach are discussed.