Sequential matched randomization and a case for covariate-adaptive randomization
Background: Sequential Matched Randomization (SMR) is one of multiple recent covariate-adaptive randomization (CAR) procedures that utilize a distance matrix to improve covariate-balance and estimation efficiency. Randomization occurs within mates whose distance meet an a-priori, fixed similarity quantile of random distances. Methods: We extend SMR to allow multiple participants to be randomized simultaneously, to allow matches to break and rematch if a better match later enrolls (Sequential Rematched Randomization; SRR), and to use a dynamic threshold. In simplified settings which vary covariate distribution and association upon outcome, we compare end-study covariate-balance and estimator efficiency in SMR before and after extensions. In a real-world application, we compare covariate-balance, power, and estimator efficiency of SMR before and after extensions when adjusting for priority covariates and all covariates of interest. We compare with Complete Randomization (CR) and CR followed by a flexible, covariate-adjusted regression model. As side-by-side comparisons, we include stratified randomization, D_A optimality biased coin design (D_A-BCD), and Pairwise Sequential Randomization (PSR). Results: In both the simplified and real-world application, we observe benefits of each extension upon covariate balance and estimator efficiency. In the real-world application, SRR with a dynamic threshold, D_A-BCD, and PSR provide greater power than CR followed by a covariate-adjusted regression model. Matching methods achieved greater covariate-balance when adjusting for all covariates yet greater power and efficiency when adjusting for priority covariates. Conclusion: We improve upon SMR and show the potential for CAR methods – that adjusting for covariates in randomization can outperform covariate adjustment in a flexible regression model.
READ FULL TEXT