相依删失下基于倾向得分协变量修正的限制均值寿命推断与应用
Restricted Mean Survival Time Inference and Application with Covariate Adjustment Based on Propensity Scores under Dependent Censoring

在临床医学和流行病学研究中，由于研究时间和随访的限制，生存数据通常存在右删失，因此限制均值寿命常用于评估实际研究中的治疗效应。本文结合协变量修正和删失数据处理方法，分析非随机化分组在随机删失和相依删失条件下的治疗效应差异问题。首先，利用倾向得分协变量修正法消除混杂影响；然后，采用逆概率删失加权方法调整相依删失引起的偏差，并考虑删失时间与协变量及基线风险函数的关系。最后，将本文提出的方法应用于70岁以上乳腺癌浸润性导管癌女性患者的两种治疗策略的效果差异分析。
In clinical medicine and epidemiology research, due to limitations in study duration and follow up, survival data is often subject to right censoring, making restricted mean survival time (RMST) a common measure for assessing treatment effects in practical studies. This paper combines covariate adjustment and methods for handling censored data to analyze treatment effect differences in non-randomized groups under conditions of random and dependent censoring. First, propensity score covariate adjustment is used to eliminate confounding effects. Then, inverse probability of censoring weighting (IPCW) is applied to adjust for biases caused by dependent censoring, taking into account the relationship between censoring time, covariates, and baseline hazard function. Finally, the proposed method is applied to analyze the effect differences between two treatment strategies for women over 70 years old with invasive ductal carcinoma of the breast.