基于观察性研究进行处理/暴露因素效应估计已成为目前研究热点，倾向性评分法作为最常用分析方法之一，在实际数据分析中可能存在倾向性评分值估计模型误设、出现极端权重等问题，导致效应量估计不准确。为解决上述问题，申请者分别研究了广义增强模型和重叠权重加权法，并在前期模拟研究中初步将两种方法联合，发现在复杂数据情况下其较现有方法的结果估计更准确，提示新的分析思路可同时解决模型误设和极端权重的问题。在此基础上，首先本课题深入探索适合重叠权重的广义增强模型迭代停止规则并优化参数设定，拟联合构建广义增强模型重叠权重加权方法。其次采用多场景数据模拟研究对方法的有效性进行验证，并依据模拟结果对方法进一步优化。最后将构建的方法编写为分析程序包，实现分析方法的自动化处理，并在前列腺癌专病队列数据中进行应用，验证其实际应用效果。研究成果可为观察性研究中处理/暴露因素效应估计提供新的分析手段与软件支持。

The treatment/exposure factor effect estimation based on observational research has become a research hotspot. Propensity score method is one of popular methods to estimate the effect of treatment/exposure factor in observational study. However, when the propensity score method is applied in the actual data analysis, there are potential risks of propensity score estimation model misspecification and extreme weight. These issues will lead to inaccurate estimation of effect size. In order to solve these two problems, we had performed in-depth study on the generalized boosted model (GBM) and the overlap weighting method. In the early simulation study, we found combination of these two methods could obtain more accurate result estimation than present methods in the data with complex relationship among covariates. Therefore, it is suggested that the new method can solve the problems of model misspecification and extreme weight at the same time. Based on the early research, first of all, we will make in-depth research on the stopping rules and parameter setting of GBM based on the overlap weight. And we intend to combine two methods to construct the generalized boosted model overlap weighting method (GBM-OW). Secondly, we will conduct simulation studies with simulation data of multi scenes to verify the validity of the GBM-OW method. The GBM-OW method will be improved and optimized according to the results of simulation studies. Finally, the analysis package of the GBM-OW method will be constructed to realize the automatic processing of the analysis method. And we will apply the GBM-OW method to the prostate cancer cohort data (National Key R&D Program of China) to verify the practical application effect of the method. Through in-depth research, we can provide a new analysis method and software support for estimating the effect of treatment/exposure factor in observational research.