701 – Dose Finding, Dose Response in Clinical Trials
A Response-Adaptive Covariate-Balanced Randomization for Multi-Arm Clinical Trials
Cassandra Ballou
Oregon Health & Science University
Yiyi Chen
Oregon Health & Science University
Randomization is a key characteristic of clinical trials which makes them the gold standard for determining treatment effectiveness. Response-adaptive randomization is often desired because it allows more patients to receive the winning treatment. Compared to traditional randomization response-adaptive randomization is more likely to allow imbalance in baseline covariates. We propose a simple yet flexible response-adaptive covariate-balanced randomization for multi-arm trials. For the response-adaptive component, we compared two approaches, the generalized drop-the-loser urn model and Ridit scoring. Likewise, for the covariate-balancing component, two approaches were considered, simple stratification and prognostic scoring. The operating characteristics of the proposed design were assessed via simulation for a variety of scenarios in which number of treatment arms, values of treatment success probability, and patient response delay time were varied. The performance of the newly proposed method was compared to non-response-adaptive and to Bayesian response-adaptive randomization methods to delineate for clinicians the circumstances where the purposed method should be adopted.