An optimization algorithm for designing phase I cancer clinical trials.

Numerous phase I dose finding clinical trials are conducted everyday to find the "maximum tolerated dose" (MTD) of a cancer treatment. Although various Bayesian designs for Phase I clinical trials have been proposed in the literature, the traditional 3+3 design is still widely used because of its algorithm-based simplicity in logistics for the clinical investigators to carry out in comparison with model-based Bayesian methods. In this paper, we propose an optimization algorithm for designing phase I cancer clinical trials. This algorithm does not need assumptions on the true dose-response relationship but can readily incorporate available prior information about the true response probabilities. It searches for an approximately optimal design within a design space, in which the 3+3 design is included as a special case. Simulation studies show that the design recommended by this algorithm significantly outperforms the 3+3 design.