STATISTICAL METHODS FOR INFORMATIVELY CENSORED DATA

在许多临床和流行病学研究中，目标之一是 以评估响应变量随时间的变化。在此类研究中 受试者被跟踪，并在几个时间测量他们的反应 在研究期间。在这些研究中，失败是很常见的 受试者死亡、退出和失访。 进一步 损耗可能与响应有关，导致数据被 信息审查。例如，提款可能发生在更高的 对治疗无反应者的频率高于 谁知道信息权审查对统计人员构成挑战 由于许多常用的统计方法假设， 损耗概率与响应变量无关。 拟议项目将侧重于纵向数据分析 当审查过程被认为是信息丰富的， 连续和离散纵向数据的统计方法。 将强调应用于癌症研究，并开发 方法将使用来自临床的实际数据进行说明， 癌症的流行病学研究。

In many clinical and epidemiological studies one of the objectives is to evaluate changes in the response variable over time. In such studies subjects are followed, and their responses are measured at several time points during the study period. It is common in these studies to lose subjects due to death, withdrawal and loss to follow-up. Further attrition may be related to the response, causing the data to be informatively censored. For example, withdrawals may occur at greater frequency among those who do not respond to the treatment than those who do. Informative right censoring poses a challenge to statisticians since many of the popularly used statistical methods assume that the attrition probability is independent of the response variable. The proposed project will focus on the analysis of longitudinal data when the censoring process is considered informative and will develop statistical methods for both continuous and discrete longitudinal data. Application to cancer studies will be emphasized, and the developed methods will be illustrated using actual data from clinical and epidemiological studies of cancer.