Abstract

A response adaptive design uses the information collected during a trial to modify the treatment allocation to allocate more patients to the potential better treatment thus having ethical advantages over the traditional clinical trials. This talk will introduce formatting the adaptation of treatment allocation as a sequential decision problem and the approximation of the optimal value functions under different optimality criterion. The property of a span-contractor operator for the sequential decision process will be discussed and the convergence in approximation to the optimal value functions will be established. When the information of the responses and treatment allocation of previously treated patients are summarized by sufficient statistics for the unknown parameters, the decision process becomes a Markov process. An algorithm will be introduced to identify the optimal allocation for response adaptive designs and numeric studies will be presented.