纵向数据因果推断中的双稳健半参数效应模型研究

项目名称： 纵向数据因果推断中的双稳健半参数效应模型研究

项目编号： No.81473071

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 医药、卫生

项目作者： 王素珍

作者单位： 潍坊医学院

项目金额： 70万元

中文摘要： 纵向数据研究中，估计处理效应、做出因果推断是非常复杂的，因为暴露于不同处理组的个体特征不同，因而组间协变量的分布并不均衡，处理效应的估计可能会出现偏倚。倾向评分（Propensity Score,PS）综合运用建模、回归、分层、加权等方法，能够较好地消除协变量不均衡引起的偏倚，对处理效应评价的准确性大大提高。然而PS本身存在许多尚未解决的问题，尤其是PS模型的指定以及与其相关的统计学假设检验方法的选择问题，限制了其在纵向数据偏倚控制中的应用。本研究拟从肿瘤患者纵向随访数据入手，研究组间协变量不均衡引起的偏倚及其对处理因素和结局变量的影响,探讨利用PS方法调整组间混杂因素的重要性，阐明纵向数据因果推断中处理效应无偏估计的理论依据，确定双稳健半参数无偏效应模型的构建方法。

中文关键词： 双稳健性；倾向评分；纵向数据；因果推断；半参数模型

英文摘要： Estimation of treatment effects with causal interpretation from longitudinal data is complicated because exposure to treatment may be confounded with subject characteristics. Propensity Score, which utilizes modeling construction, regression and covariate adjustment synthetically, can effectively eliminate the biases caused by the imbalance of covariates and offer improved precision for the evaluation of treatment effect. However, there are many unresolved issues for PS methods. The applications of PS in the bias control of longitudinal data have been limited due to uncorrect specifications of PS model or statistical hypothesis.In this study we will dig the biases resulted from unbalance of covariates and explore the effect of biases on treatment and outcome variables for longitudinal data from tumor patients so as to make it clear that it is very important to adjust the cofoundings among groups by PS methods. The theories of unbised treatment effect estimation will be clarified and the construction of double robust semiparametric unbiased effect model for causal effect inference in longitudinal data will be determined.

英文关键词： Double robust;propensity score;longitudinal data;causal effect;semiparametric model