Investigating type I error rate of clinical trials after using stratified permuted block randomization and the permutation test
In this project, we compared different analysis methods following stratified permuted block design, a commonly used method of covariate-adaptive randomization in clinical trials, the permutation test, and the traditional t-test, or the t-value for treatment from the full model in linear regression. Type I error rates were studied under three different scenarios. The t-test yielded very conservative type I error rates for the three scenarios when covariate information, which was used in covariate-adaptive design, was not included in the data analysis. We proposed to use the permutation test to fix this problem, and found that the type I error rates were preserved at the nominal level with our proposed methods.^ Preserving the nominal type I error rates is of special importance to validate the clinical trial results. Having valid results is crucial in determining whether or not a treatment is worth the potential side effects and cost to patients. If the results do not reflect the true effect, then an intervention that may be more harmful or less effective than anticipated could reach the target populations without having a true benefit. Ensuring clinical trial results are valid is very important from ethical, financial, and physical standpoints. Treatments that may cause too much harm or are too expensive to patients should be properly assessed, since patients would ultimately want to invest in proven treatments that will help improve their health and quality of life. In addition, when evaluating treatments, using covariate information in the study design and analysis is beneficial since certain covariates might affect the treatment in different ways. However, excluding randomization covariates from the data analysis will lead to conservative type I error. This project successfully investigated the data analysis methods to preserve the nominal type I error rate.^
Olefsky, Maxine Mei Zheng, "Investigating type I error rate of clinical trials after using stratified permuted block randomization and the permutation test" (2016). Texas Medical Center Dissertations (via ProQuest). AAI10248803.