Permutation Test Following Covariate-Adaptive Randomization in Randomized Controlled Trials

Abstract

In randomized controlled trials, patients are recruited and randomly allocated to treatments. Patients are never randomly sampled from large population of patients on treatments under study. Therefore, it is important to consider the design and statistical analysis based on the randomization model.

In this article, we show theoretically that a permutation test based on the difference in means is identical to analysis of covariance if marginal covariate balance is completely attained. Our theoretical results and Monte Carlo simulation study suggest that the permutation test following Pocock–Simon's covariate-adaptive randomization can be a useful alternative to traditional population-based tests in a confirmatory randomized controlled trial with important prognostic factors. The proposed procedure is illustrated with modified data from the randomized placebo-controlled trial of pirfenidone.

Key Words:

• Covariate-adaptive randomization
• Invoked population model
• Linear models
• Monte Carlo method
• Permutation tests
• Randomization model

ACKNOWLEDGMENT

We are much obliged to Professor William F. Rosenberger at the George Mason University for his comments which improved this paper.

Notes

10,000 simulations were conducted. P test: permutation test.

10,000 simulations were conducted. P test: permutation test.