违背分组情形下生存数据的半参数因果推断

项目名称： 违背分组情形下生存数据的半参数因果推断

项目编号： No.11271081

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 数理科学和化学

项目作者： 郑明

作者单位： 复旦大学

项目金额： 58万元

中文摘要： 在医学、心理学、经济学、社会学的大量研究中，对干预措施与感兴趣的结果变量进行因果推断是人们最关心的。因果推断比相关分析更为复杂与困难。进行因果推断的重要基础之一是随机化设计，但在现实中，样本个体或试验者常常不能按照随机化设计的要求到试验组或控制组完成试验，即出现违背分组的情况，一种典型的违背分组类型为全部-从不-遵循（all-or-none compliance）。另一方面，试验中获得的结果变量往往存在各种形式的截断或删失或缺失。上述两种情况的共存，给因果推断带来了更大的难度。迄今为止，关于生存数据因果推断的研究成果不多。本项目将针对全部-从不-遵循的情形，在前人研究工作及申请人前一项目获得的研究成果的基础上，探讨生存数据的因果推断问题。具体地，在若干种半参数模型假设下，基于半参数似然思想首次为各种形式的截断与删失数据提供合理与高效的因果推断方法，从而丰富随机试验数据的统计分析手段。

中文关键词： 全部-从不-遵循；因果推断；Cox模型；线性变换模型；生存数据

英文摘要： In many studies of medical science, psychology, economics and sociology, people care most about the causal inference between the intervention and the outcome variable. Causal inference is usually more complicated and difficult than correlation analysis, and randomization is one of the most important bases to make casual inference. In many real applications, some subjects involved in the study may not obey the randomization assignments and thus non-compliance phenomena appear. One typical form is the so-called all-or-none compliance. On the other hand, some outcome variables, such as survival time, rising from the randomization studies often subject to various kinds of censoring, truncation and missing. Both the non-compliance and the incomplete observation bring great challenge to the causal inference. So far, little literature has discussed the causal inference for survival data with non-compliance. In this project, we will study the causal inference for survival data based on the existing literature and the achievement of the previous NSFC project of the applier. Specially, we will propose new causal inference procedures for survival data with various kinds of censoring, truncation and missing under several commonly used semi-parametric models. Our basic idea is to use the semi-parametric likelihood method and

英文关键词： all-or-none compliance；causal inference；Cox model；transformation model；survival data