Evaluating treatment effect modifiers using data from randomized two-sequence, two-period crossover clinical trials: application to a diabetes study
本文针对2型糖尿病交叉临床试验，提出分析治疗效果修饰因子的统计框架，通过患者自身对照消除混杂并提高效率，发现基因表达标记物可预测不同治疗推荐。
Abstract Type 2 diabetes is known as a heterogeneous disease with diverse pathophysiology. The identification of treatment effect modifiers for personalized medicine, however, has been difficult in basic research due to complex biological mechanisms related to various genetic and nongenetic factors, and in clinical research due to confounding bias and limited sample sizes. In this paper, we focus on a two-sequence, two-period crossover (CO) clinical trial of type 2 diabetes with baseline markers as a new strategy for analyzing treatment effect modification that allows confounding elimination and efficiency enhancement by within-patient treatment comparison. We provide a framework for statistical analysis of treatment effect modification and develop methods for testing for treatment effect modification and estimating individualized predictors of treatment differences in the CO trial with limited sample size. Numerical assessments showed that the efficiency of the proposed CO analysis was substantially higher than the standard analysis based on parallel-group comparison in these analyses. Application to the diabetes trial showed that markers based on gene expression in peripheral blood cells before pharmacological treatment significantly modified the effect of the treatment on the response and that marker-based predictors produced two subgroups of patients for whom different treatments should be recommended.