Novel Statistical Methods for Complex Time-to-Event Data in Cardiovascular Clinical Trials

项目摘要： 现代心血管（CV）试验经常收集大量致死性和非致死性事件的数据（例如，心脏 衰竭、心脏病发作、中风、胸痛等）对病人健康有不同的影响。近年来， 新的方法已经开始出现，它们试图捕获比每个事件的传统端点更多的事件。 患者的第一个事件。然而，为了说明复合终点的整体性，同时区分 其组件的重要性（例如，死亡vs CV住院）并不容易。目前，调查人员仍然缺乏 足够的工具来衡量治疗效果，设计未来的试验，评估风险因素和建立预测模型。 在这个项目中，我们通过四个具体目标来解决这些差距。在目标1中，我们考虑了一个一般类， 通过成对比较（总体和亚组）定义的非参数效应量被估量， 其中一个组件可以使用分级比较规则容易地优先于另一个组件。逆 概率截尾加权（IPCW）和增强逆概率加权（AIPW）技术， 适用于U统计量估计量，以纠正删失偏倚并提高效率（从而减少试验 成本）使用随机化前和随机化后的患者数据。在目标2中，我们开发了计算功率的例程， 新提出的复合终点方法的样本量，例如有利于 在固定和成组序贯设计下，治疗和存活期间丢失率。在目标3中，我们提出了新的 复合终点的半参数回归模型遵循早期关于比例胜利的工作， 分数（PW）模型。广义半参数比例优势（GSPO）模型 通过将传统的PO模型扩展到多个事件， 严重性在目标4中，我们将生存树作为预测工具从单变量扩展到复合终点。绘图 在有序响应的分类树上，我们开发了加权基尼指数的时间整合版本， 两个方法的节点分裂，和一个广义的一致性指数的交叉验证修剪，从而 考虑结果事件的时间和严重性。开发的方法将用于 最近得出结论的流感疫苗有效阻止心胸事件的二次分析， 失代偿性心力衰竭（INVESTED）试验（ClinicalTrials.gov：NCT 02787044）。与此同时，他们将 在Comprehensive R Archive Network（CRAN.R-project.org）上，将其整合到新的和现有的R软件包中 供从业人员公开使用。

Project Summary: Modern cardiovascular (CV) trials often collect data on a wide array of fatal and nonfatal events (e.g., heart failure, heart attack, stroke, chest pain, and etc.) with different implications for patient health. In recent years, new methods have started to emerge which seek to capture more events than the traditional endpoint of each patient’s first event. However, to account for the totality of a composite endpoint while differentiating the importance of its components (e.g., death vs CV hospitalization) is not easy. As it stands, investigators still lack adequate tools to measure treatment effects, design future trials, assess risk factors, and build prediction models. In this project, we address these gaps via four specific aims. In Aim 1, we consider a general class of nonparametric effect-size estimands defined though pairwise comparison (both overall and subgroup-wise), in which one component can be readily prioritized over another using a hierarchical rule of comparison. The inverse probability censoring weighting (IPCW) and augmented inverse probability weighting (AIPW) techniques are adapted to U-statistic estimators to correct for censoring bias and to improve efficiency (and thus reduce trial cost) using patient data both pre- and post-randomization. In Aim 2, we develop routines to calculate power and sample size for newly proposed methods for composite endpoints, such as the restricted mean time in favor of treatment and while-alive loss rate, under both fixed and group sequential designs. In Aim 3, we propose novel semiparametric regression models for composite endpoints following earlier work on the proportional win- fractions (PW) model. In particular, the generalized semiparametric proportional odds (GSPO) model accommodates nonproportional win fractions by extending traditional PO models to multiple events with ordered severities. In Aim 4, we extend survival trees as a predictive tool from univariate to composite endpoints. Drawing on classification trees for ordinal response, we develop time-integrated versions of the weighted Gini index and twoing approach for node-splitting, and of a generalized concordance index for cross-validative pruning, thereby accounting for both the timing and severity of the outcome events. The methods developed will be used for secondary analyses of the recently concluded INfluenza Vaccine to Effectively Stop cardio-Thoracic Events and Decompensated heart failure (INVESTED) trial (ClinicalTrials.gov: NCT02787044). Meanwhile, they will be incorporated into new and existing R-packages on the Comprehensive R Archive Network (CRAN.R-project.org) for public use by practitioners.