竞争风险数据下关于时变暴露的逆概率处理加权研究

Inverse Probability of Treatment Weighting for Time-Varying Exposure in Competing Risks Data

摘要: 在生存分析中，当研究个体的暴露因素随时间动态变化，且结局事件存在竞争风险时，传统Cox比例风险模型的估计会产生偏倚。本文旨在将逆概率处理加权方法与Fine-Gray竞争风险模型相结合，提出一种双重加权模型，以精准估计竞争风险数据下时变暴露对感兴趣事件的因果效应。首先，构建逆概率权重以调整时间依赖性混杂。然后，将逆概率权重与Fine-Gray模型的删失权重结合，形成双重权重并嵌入加权得分方程。最后，利用恶性黑色素瘤患者数据，分别建立时变分段暴露变量和静态二元暴露变量模型进行对比分析。实例分析表明，在双重加权竞争风险模型中，相较于静态暴露变量，时变暴露变量能更精确地捕捉暴露的动态效应，估计结果更为稳健。

Abstract: In survival analysis, when the exposure factors of study subjects change dynamically over time and the outcome events involve competing risks, traditional Cox proportional hazards models may yield biased estimates. This paper aims to integrate the inverse probability of treatment weighting method with the Fine-Gray competing risks model to propose a doubly-weighted model, in order to accurately estimate the causal effect of time-varying exposure on the event of interest in the presence of competing risks. First, inverse probability weights are constructed to adjust for time-dependent confounding. Then, these weights are combined with the censoring weights from the Fine-Gray competing risks model to form double weights, which are embedded into a weighted score equation. Finally, using data from malignant melanoma patients, models with time-varying segmented exposure variables and static binary exposure variables are established for comparative analysis. The empirical analysis demonstrates that, in the doubly-weighted competing risks model, time-varying exposure variables more accurately capture the dynamic effects of exposure and yield more robust estimates compared to static exposure variables.