GSK Vaccines, Rue de l'Institut 89, Rixesart, Belgium.
Randstad India Pvt Ltd (Employee Contracted for GSK Asia Pvt Ltd), Bangalore, India.
Stat Methods Med Res. 2021 Apr;30(4):1072-1080. doi: 10.1177/0962280220985564. Epub 2021 Jan 27.
In clinical trials, several covariate-adaptive designs have been proposed to balance treatment arms with respect to key covariates. Although some argue that conventional asymptotic tests are still appropriate when covariate-adaptive randomization is used, others think that re-randomization tests should be used. In this manuscript, we compare by simulation the performance of asymptotic and re-randomization tests under covariate-adaptive randomization. Our simulation study confirms results expected by the existing theory (e.g. asymptotic tests do not control type I error when the model is miss-specified). Furthermore, it shows that (i) re-randomization tests are as powerful as the asymptotic tests if the model is correct; (ii) re-randomization tests are more powerful when adjusting for covariates; (iii) minimization and permuted blocks provide similar results.
在临床试验中，已经提出了几种协变量自适应设计，以平衡治疗组与关键协变量的关系。尽管有人认为，当使用协变量自适应随机化时，传统的渐近检验仍然适用，但也有人认为应该使用再随机化检验。在本文中，我们通过模拟比较了协变量自适应随机化下渐近检验和再随机化检验的性能。我们的模拟研究证实了现有理论所预期的结果（例如，当模型指定错误时，渐近检验不能控制Ⅰ类错误）。此外，它还表明：（i）如果模型正确，再随机化检验与渐近检验同样有效；（ii）在调整协变量时，再随机化检验更有效；（iii）最小化和置换块提供了相似的结果。
