A Bayesian model for the optimal ordering of a collection of screens

Abstract

SUMMARY Bergman & Gittins (1985) propose a screen design problem in which a collection of screens for therapeutic activity and toxicity of pharmaceutical compounds is to be sequenced. They describe conditions under which a natural ordering heuristic based on failure probability per unit cost for each screen is optimal. We propose a Bayesian model in which measures of activity and toxicity are given a joint prior density. The above conditions for optimality are satisfied for this model under simply stated conditions. A numerical study suggests that the heuristic performs well more generally.