PDF(Pubmed)
Abstract：
In precision medicine, there is much interest in estimating the expected-to-benefit (EB) subset, i.e. the subset of patients who are expected to benefit from a new treatment based on a collection of baseline characteristics. There are many statistical methods for estimating the EB subset, most of which produce a \'point estimate\' without a confidence statement to address uncertainty. Confidence intervals for the EB subset have been defined only recently, and their construction is a new area for methodological research. This article proposes a pseudo-response approach to EB subset estimation and confidence interval construction. Compared to existing methods, the pseudo-response approach allows us to focus on modelling a conditional treatment effect function (as opposed to the conditional mean outcome given treatment and baseline covariates) and is able to incorporate information from baseline covariates that are not involved in defining the EB subset. Simulation results show that incorporating such covariates can improve estimation efficiency and reduce the size of the confidence interval for the EB subset. The methodology is applied to a randomized clinical trial comparing two drugs for treating HIV infection.
摘要：
在精准医学中，估计预期收益(EB)子集有很大的兴趣，即,基于基线特征的集合,预期受益于新的治疗的患者的子集。有许多统计方法来估计EB子集，其中大多数都会产生“点估计”，而没有解决不确定性的信心声明。EB子集的置信区间最近才被定义，它们的构建是方法论研究的新领域。本文提出了一种用于EB子集估计和置信区间构造的伪响应方法。与现有方法相比，伪反应方法使我们能够专注于对条件治疗效应函数进行建模(与给定治疗和基线协变量的条件均值结果相反),并且能够整合来自基线协变量的信息,这些信息不参与定义EB子集.仿真结果表明，合并此类协变量可以提高估计效率并减少EB子集的置信区间大小。该方法适用于比较两种治疗HIV感染的药物的随机临床试验。
