Random walk and parallel crossing bayesian optimal interval design for dose finding with combined drugs

Abstract

Interval designs have recently attracted enormous attention due to their simplicity, desirable properties, and superior performance. We study random-walk and parallel-crossing Bayesian optimal interval designs for dose finding in drug-combination trials. The entire dose-finding procedures of these two designs are nonparametric (or model-free), which are thus robust and also do not require the typical "nonparametric" prephase used in model-based designs for drug-combination trials. Simulation studies demonstrate the finite-sample performance of the proposed methods under various scenarios. Both designs are illustrated with a phase I two-agent dose-finding trial in prostate cancer.