An adaptive dose-finding approach for correlated bivariate binary and continuous outcomes in phase I oncology trials.

In this study, we developed a novel adaptive dose-finding approach for inclusion of correlated bivariate binary and continuous outcomes in designing phase I oncology trials. For this approach, binary toxicity and continuous efficacy outcomes are modeled jointly with a factorization model. The basic strategy of the proposed approach is based primarily on the Bayesian method. We based the dose escalation/de-escalation decision rules on the posterior distributions of both toxicity and efficacy outcomes. We compared the operating characteristics of the proposed and existing methods through simulation studies under various scenarios. We found that the recommendation rate of the true recommended dose (RD) in the proposed method was more favorable than that in the existing method when the true RD was relatively at the tail end among the tested doses. It was similar to that of the existing method when the true RD was relatively at the top end.