Statistical Methods for HIV/AIDS Research

项目描述（由申请人提供）：本项目旨在发展和应用艾滋病研究中的统计方法，以改善采用纵向随访设计时的科学推断。本研究的目的如下：(1)建立具有信息性终末事件的标记数据的统计方法。在艾滋病毒/艾滋病研究的纵向研究中，经常收集或观察到对复发事件发生的影响。目标1将考虑两种类型的标记物测量：复发标记物过程数据和治疗后标记物数据在失败事件时评估。标记数据的收集通常会因行政审查或发生终端事件（如死亡）而终止，其中终端事件可能与感兴趣的标记测量相关。Aim 1下的工作将包括用于分析复发标记过程数据的推理、建模和估计方法的开发，以及用于从处理后标记数据估计因果治疗效果的分析程序。(2)开发新的组序方法，用于监测HIV/AIDS长期临床试验中审查的时间到事件终点。数据和安全监测委员会（DSMB）通常需要进行中期分析，以监测艾滋病毒/艾滋病长期临床试验中预防团或治疗治疗的有效性和安全性。对于这样的长期艾滋病毒/艾滋病试验，收集了经过审查的时间到事件的终点，传统统计方法的幼稚应用，如比例风险模型（Cox, 1972），往往不足以描述长期随访期间不同治疗方案之间治疗效果的时变性质。在这一目标下，我们将开发新的基于具有变化点的危害函数的群体序列方法来监测时变治疗效果。(3)发展回归方法以适应不断变化的协变量效应。艾滋病毒/艾滋病干预措施很少有持续的效果。无论是艾滋病毒预防试验还是艾滋病治疗研究，期望干预在随机化后立即完全生效是不现实的。此外，随着时间的推移，耐药性可能会产生，从而削弱干预效果。对时变干预效果进行表征和量化，将为研究干预机制提供有价值的科学认识。然而，大多数可用的方法只能适应常数效应。我们计划开发统计模型和推理程序来解决这个问题，重点是推广加速故障时间模型和循环事件数据。(4)开发增强的敏感性分析程序，用于分析过早失去随访的HIV随机研究，特别是由于终止治疗。在许多申请FDA批准的HIV药物注册试验中，患者在治疗结束后没有被跟踪。在FDA要求患者在治疗结束后继续随访之前，本研究的目的是为FDA临床审查员和更广泛的科学界提供有关治疗效果意图的信息，否则这些信息将无法获得。我们将开发的方法将普遍适用于随机试验和观察性研究，这些研究可能缺乏随访信息。公共

DESCRIPTION (provided by applicant): The objective of this project is to develop and apply statistical methods in AIDS research to improve scientific inferences when longitudinal follow-up designs are employed. The proposed research includes the following aims: (1) To develop statistical methods for marker data with informative terminal events. In longitudinal studies of HIV/AIDS research, marker measurements are frequently collected or observed conditioning on the occurrence of recurrent events. Two types of marker measurements will be considered under Aim 1: recurrent- marker process data, and post-treatment marker data evaluated at time of failure event. The collection of marker data is typically terminated by administrative censoring or occurrence of a terminal event such as death, where the terminal event is possibly correlated with the marker measurements of interest. The work under Aim 1 will include the development of inference, modeling and estimation methods for analyzing recurrent marker process data, and analytical procedures for estimation of causal treatment effects from post-treatment marker data. (2) To develop new group sequential methods for monitoring censored time-to-event endpoints in long-term HIV/AIDS clinical trials. Interim analyses are usually required by the Data and Safety Monitoring Board (DSMB) to monitor the efficacy and safety of a prevention regiment or therapeutic treatment in long-term HIV/AIDS clinical trials. For such long-term HIV/AIDS trials that collect censored time-to-event endpoints, naive applications of the conventional statistical methods, such as the proportional hazards model (Cox, 1972), are often insufficient to characterize the time-varying nature of treatment effect between different treatment regiments during long-term follow-up. Under this aim, we will develop new group sequential methods based on the hazard functions with change points in monitoring time-varying treatment effect for censored time-to-event outcomes. (3) To develop regression methods to accommodate evolving covariate effects for recurrent events data. HIV/AIDS interventions rarely have constant effects. Whether it is an HIV prevention trial or AIDS therapeutic study, it is unrealistic to expect the intervention to take full effect instantaneously after randomization. Furthermore, drug resistance might develop over time, which erodes the intervention effect. Characterizing and quantifying time- varying intervention effect would provide valuable scientific insight to the mechanism of the intervention. However, most available methods only accommodate constant effects. We plan to develop statistical models and inference procedures to address this issue, with a focus on generalizing the accelerated failure time model and on recurrent events data. (4) To develop enhanced sensitivity analysis procedures for analyzing HIV randomized studies with premature loss of follow-up, especially due to termination of treatment. In many registration trials for FDA approval of HIV medicines, patients are not followed after treatment termination. Until the FDA mandates continued follow-up of patients after treatment termination, the aim of this research is to provide FDA clinical reviewers and the broader scientific community information about intention to treat effects that would otherwise be unavailable. The methods we will develop will be generally applicable to randomized trials and observational studies with potentially informative loss of follow-up. PUBLIC HEALTH RELEVANCE: New statistical models and methods are proposed to study survival, recurrent events and marker process data in HIV/AIDS clinical trials and cohort studies. Statistical tools and techniques are developed to deal with some of the sophisticated and important problems arising in AIDS studies with longitudinal nature.