Covariate-Adjusted Response Adaptive Designs for Competing Risk Survival Models

Often in medical research, response to a particular treatment can be classified in terms of failure from multiple causes. In such cases, a competing event precludes the observation of the main event of interest. Such scenarios in survival analysis are termed as competing risks. Covariate-adjusted response-adaptive (CARA) designs skew patient allocation toward the better-performing treatment arm, so far in a clinical trial, for a given patient's covariate profile. When there are competing events occurring in a clinical trial, ignoring such information during the design stage may bias the results for disease-specific treatment comparison. Optimal CARA designs are developed, assuming proportional sub-distribution hazards for two-arm survival trials, where the primary endpoint encounters competing risks. The derived allocation proportions are targeted using biased coin procedure. These allocation proportions that are sequentially estimated, converge empirically to the expected target values, which are functions of the Fine and Gray model coefficients. The proposed methods are shown to be suitable alternatives to the traditional balanced designs through extensive simulation studies and have also been implemented to redesign a real-life clinical trial. Simulation results reveal the need of a theoretical procedure for more complicated semi-parametric survival response models.