Minimizing the Costs of Screening Designs

Keywords: design of experiments, adaptive, sequential

Abstract: Screening designs are used to rapidly decide if drugs or other agents are worthy of further study. We provide a program for optimizing such designs, assuming that tests have Bernoulli outcomes. Our program provides an integrated framework for accurately modeling the situation, giving the user a great deal of flexibility. The program incorporates testing costs and constraints and the costs of incorrect decisions. It allows for grouped allocation of tests, ranging from a single round of tests to fully sequential designs. Previous researchers had fixed the form of the design, such as two-stage, and optimized the parameters of that form. By instead providing a comprehensive framework, more design variations can be considered and the costs can be reduced.