A class of designs for Phase I cancer clinical trials combining Bayesian and likelihood approaches

The Bayesian continual reassessment method (CRM) and its likelihood version (CRML) provide important tools for the design of Phase I cancer clinical trials. However, a poorly chosen prior distribution in CRM may lead to inferior performance of the method in the early stage of a trial, whereas the maximum-likelihood estimate used in CRML may result in initial high variability. These features of CRM and CRML served as the motivations for the development of this new class of designs, which combines the Bayesian and the likelihood approaches and has CRM and CRML as special cases. Simulation studies on a leukaemia trial show that the proposed class of designs significantly outperforms the traditional up-and-down design.