Response-adaptive rerandomization

Rerandomization has recently attracted more attention in the literature randomized experiments. It leverages covariate information of participants to achieve a well-balanced allocation, and thus improves the efficiency of inference. However, by only considering covariate information, it may lead to potential ethical issues in clinical trials as a large number of patients might be assigned to the inferior treatment arm. To mitigate this issue, we propose a response-adaptive rerandomization scheme by incorporating response information for two-arm comparative clinical trials. Not only is our method applicable to both continuous and binary outcomes, but it also demonstrates desirable statistical and ethical properties. Extensive simulation studies are performed to illustrate the practicality and superiority of our approach.