Development of Human Exposure and ToxicityFactor Distributions Using Maximum-Entropy Inference.* Robert C. Lee and William E Wright, Golder Associates, 4104 148th Ave. NE, Redmond, WA 98052
The need for a subjective probability assessment procedure exists in the field of stochastic health risk assessment to allow development of input variable distributions. A method based on maximum-entropy inference is proposed and demonstrated for development of exposure and toxicity-factor distributions. This method has been used in the fields of information theory, decision analysis, and engineering; but its application to the field of health risk assessment is novel. Distributions developed using this method maximize uncertainty given limited informational constraints, such as upper and lower bounds, means, and/or quantiles. Thus, risk estimates resulting from stochastic models incorporating such distributions are likely to incorporate the full range of possibilities, and decisions based on upper percentiles are likely to be conservative. Maximum-entropy inference avoids or reduces the potential problems of uncertainty underestimation, bias, undocumented assumptions, credibility, and expense that are associated with other subjective probability assessment techniques. Distributional solutions which maximize uncertainty are derived for exposure variables, as well as for slope factor and reference dose distributions based on information in EPA's Integrated Risk Information System.
*Work funded by Monsanto Corporation and Golder Associates, Inc.