Bayesian Interpretation of Animal Bioassay Results. K. P. Brand, J. K. Hammitt, and J. S. Evans, Dept. of Environmental Health and Harvard Center for Risk Analysis, Harvard School of Public Health, Boston, MA, 02115
Weight of evidence summaries of animal bioassays results (both cancer and non-cancer) are generally undermined by large uncertainties. Reasonable interpretations of the evidence can be disparate despite sharing (roughly) equal consistency with the evidence. In such cases the inadequacy of just one of these interpretations (i.e., point-estimates) is obvious and has spurred the use/development of uncertainty analysis techniques. Bayesian methods provide some important advantages over traditional uncertainty analysis techniques (e.g., Monte-Carlo simulation) including: a convenient basis for integrating newly gathered evidence into the overall assessment; facilitates the assessment of value-of-information indices; and can identify the limited/constrained resolving power of typical bioassays. We focus on the latter advantage. Inattention to the precision and potential bias inherent in a data-generating process (DGP) can lead to miss-interpretations of bioassay results. We give examples of such miss-interpretations and demonstrate how the application of Bayesian methods would have led to more appropriate interpretations. Finally the limited resolving-power of bioassay DGP’s are considered in a decision-analytic context wherein other factors affect the value of bioassay data (e.g., relevance, promise, and a priori uncertainty).
Work supported by EPA Cooperative Agreement 822917-01-0.