OPTIMAL METHODS FOR HIGH DIMENSIONAL CENSORED DATA

描述(改编自申请者的摘要)：建议的研究将 开发和测试统计半参数方法以估计A 的(多元)时间随机变量的(多元)分布 何时审查观测时间(不准确的数据)以及何时感兴趣 经审查/未经审查，可能是高维的、依赖时间的和/或 与时间无关的协变量是可用的。所开发的方法有两种 对现有方法的重要扩展：(1)它们将允许 感兴趣的时间随机变量与截尾随机变量的相依性 变量；(2)它们能够使用附加的依赖于时间的和 非参数局部高维时间无关协变量 最佳方式。例如，在研究传染病时， 在免疫系统中，效率大幅提高的估计方法可以 通过对免疫反应之间的关系进行建模而获得 系统和感兴趣的时间变量。 所提出的方法得到的估计量在少数情况下是一致的。 建模假设，但它们的效率将取决于一个假设的好坏 能够对感兴趣的时间和 时间相关和/或时间无关的协变量。该方法将是 应用于动物致癌实验和其他癌症的数据 估计肿瘤发病时间分布的研究， 艾滋病数据(传染病数据)，该方法也将用于 人体内植入物(肝分流术)存活功能的评估 尸体。除了大量的协变量外，现有的数据还包含 具有时间变量预测值的时间相关协变量 利息。该方法的性能将通过将其应用于 已知真实模型的模拟数据。模拟也将被 用来与现有方法进行效率比较。

DESCRIPTION (Adapted from applicant's abstract): The proposed research will develop and test statistical semiparametric methods for estimation of a (multivariate) distribution of a (multivariate) time-random variable of interest when the observed time is censored (data which is inexact) and when censored/uncensored, possibly high dimensional, time-dependent and/or time-independent covariates are available. The developed methods have two imporant extensions to the existing methods: (1) they will allow for dependence of the time random variable of interest and the censoring random variables; (2) they are able to use additional time-dependent and high-dimensional time-independent covariates in a nonparametric and locally optimal way. For example, in the study of infectious diseases which change the immune system, a large gain in efficiency of the estimation method can be obtained by modelling the relationship between responses of the immune system and the time-variable of interest. The estimators resulting from the developed methods are consistent under few modelling assumptions, but their efficiency will depend on how well one is able to model the relationships between time of interest and the time-dependent and/or time-independent covariates. The method will be applied to data from animal carcinogenicity experiments and other cancer studies for estimation of the distribution of the time of onset of tumors, to AIDS data (infectious disease data) and the method will also be used for estimation of the survival function of an implant (liver shunt) in the human body. The data available contain, beyond a large number of covariates, time-dependent covariates with predictable value for the time variable of interest. The performance of the method will be studied by applying it to simulated data where the true models are known. Simulations will also be used to do an efficiency comparison of the methods with existing methods.