Properties of Log-Rank Test Following Covariate-Adaptive Randomization in Multi-Center Oncology Trials

• Activity Number: 318 - Adaptive (and Other) Clinical Trial Designs
• Type: Contributed
• Date/Time: Wednesday, August 11, 2021 : 3:30 PM to 5:20 PM
• Sponsor: Biopharmaceutical Section
• Author(s): Jerry J Li* and Olga M Kuznetsova and Gang Han
• Companies: Daiichi Sankyo Inc. and Merck & Co., Inc. and Department of Statistics, Iowa State University
• Keywords: minimization; covariate-adaptive randomization; stratified log-rank test; re-randomization test; multi-center trial

Abstract

Population in oncology trials is often highly heterogeneous, and balancing subjects allocation in strong prognostic factors is required. A dynamic allocation procedure, most commonly a Pocock and Simon (1975) covariate-adaptive randomization, is employed to balance the covariates.

We studied the performance of log-rank test following minimization that balances on several prognostic factors in an oncology study where center is a strong prognostic factor, the number of centers is large, and the stratified log-rank test (ST-LR) does not stratify by center. Through simulations we observed the followings: 1) The ST-LR preserves the Type I error when center is not included in minimization, but is conservative when center is included in minimization; 2) Power of the ST-LR is higher when center is included in minimization vs. excluded from minimization; 3) In presence of treatment effect, when center is included in minimization, P-value from the re-randomization test is generally lower than that from ST-LR , but the two P-values are very close when the treatment effect is absent.