Treatment Comparison and Estimation of Relative Risks with Failure Time Data

描述(申请人提供)：失效时间数据的统计分析一直是设计和分析各种医学研究(如随机临床试验)的基石，并发挥着重要作用。故障时间数据区别于所有其他类型数据的一个关键特征是审查，这使得故障时间数据的分析既独特又困难，并且可能以不同的形式出现。这项建议将研究医学研究中的两个基本问题，即治疗比较和相对风险的估计，涉及几种类型的失效时间数据，涉及区间截尾或区间截尾失效时间数据。对这些数据的分析越来越受到包括政府机构和制药公司在内的医学调查人员和统计学家的关注。具体地说，拟议的研究包括三个目标，它们是开发适当和/或有效的统计程序，用于1)治疗比较和样本量计算，2)相对风险估计和回归分析1，以及3)相对风险估计和回归分析2。众所周知，治疗比较可能是医学研究中最基本和最常见的要求任务，对于此类研究的设计，一个关键因素是确定所需的样本量。在区间截尾数据的情况下，提出了一些比较步骤。然而，这些方法大多在应用上受到限制，更重要的是，它们都不能用于样本量的计算。目标1将开发新的比较程序，允许并因此给出计算样本量的公式。估计相对风险或更普遍的协变量效应是医学研究中的另一项常见任务，例如无进展生存肿瘤学研究。当人们面对各种类型的区间删失数据，包括聚集数据时，所提出的研究将开发出适当或有效的方法。目标2将侧重于审查可被认为与所研究的故障时间变量无关的情况，而目标3将处理独立性不真实的情况。这项拟议研究中将开发的统计程序或工具将使有关研究的设计成为可能，并有助于对其进行正确和/或有效的分析。 公共卫生相关性：这项建议的主要目标是解决和调查医学研究中的几个重要问题，如随机临床试验中，当面临各种类型的间隔审查失败时间数据时，治疗比较和相对风险估计，并为这些数据开发适当的和/或有效的统计程序。它将由三个目标组成：1)处理比较和样本量计算，2)相对风险估计和回归分析I，3)相对风险估计和回归分析II。目标2和目标3的主要区别是前者将侧重于审查机制独立于响应变量的情况，而后者将处理审查机制可能依赖于响应变量的情况。

DESCRIPTION (provided by applicant): The statistical analysis of failure time data has been the cornerstone and plays an essential role in the design and analysis of various medical studies such as randomized clinical trials. A key feature of failure time data that separates them from all other types of data is censoring, which makes the analysis of failure time data unique and difficult and can occur in different forms. This proposal will investigate two fundamental problems in medical studies, treatment comparison and estimation of relative risks, with respect to several types of failure time data that involve interval-censoring or interval-censored failure time data. The analysis of such data has been attracting more and more attention among medical investigators and statistician including these in government agencies and pharmaceutical companies. Specifically, the proposed research consists of three aims and they are the development of appropriate and/or efficient statistical procedures for 1) treatment comparison and sample size calculation, 2) estimation of relative risks and regression analysis I, and 3) estimation of relative risks and regression analysis II. It is well-known that treatment comparison is perhaps the most basic and commonly required task in medical studies and for the design of such studies, a key element is the determination of the required sample size. In the case of interval-censored data, some comparison procedures have been proposed. However, most of them are limited in applications and more importantly, none of them can be used for sample size calculation. Aim 1 will develop new comparison procedures that allow and thus give formulas for the sample size calculation. Estimation of relative risks or more generally covariate effects is another common task in medical studies such as progression-free survival oncology studies. The proposed research will develop appropriate or efficient approaches to it when one faces various types of interval-censored data including clustered data. Aim 2 will focus on situations where the censoring can be regarded to be independent of the failure time variable under study, while aim 3 will deal with situations where the independence is not true. The statistical procedures or tools that will be developed in this proposed research will make the design of the concerned studies possible and help one to conduct correct and/or efficient analysis of them. PUBLIC HEALTH RELEVANCE: The main goal of this proposal is to address and investigate several important issues on treatment comparison and relative risk estimation in medical studies such as randomized clinical trials when one faces various types of interval-censored failure time data, and develop appropriate and/or efficient statistical procedures for them. It will consist of three aims and they are; 1) treatment comparison and sample size calculation, 2) estimation of relative risks and regression analysis I, and 3) estimation of relative risks and regression analysis II. A major difference between aims 2 and 3 is that the former will focus on situations where censoring mechanism is independent of response variables, while the latter will deal with situations when the censoring mechanism may depend on response variables.