In this paper the regression discontinuity design is adapted to the survival analysis setting with right-censored data, studied in an intensity based counting process framework. In particular, a local polynomial regression version of the Aalen additive hazards estimator is introduced as an estimator of the difference between two covariate dependent cumulative hazard rate functions. Large-sample theory for this estimator is developed, including confidence intervals that take into account the uncertainty associated with bias correction. As is standard in the causality literature, the models and the theory are embedded in the potential outcomes framework. Two general results concerning potential outcomes and the multiplicative hazards model for survival data are presented.
翻译:在本文中,回归不连续性设计适应了以基于强度计数过程框架研究的右审查数据的生存分析环境,特别是Aalen添加危险估计值的局部多数值回归版本,作为两个共变量的累积危险率函数之间的差值的估测符。为这一估计值开发了大样本理论,包括考虑到与纠正偏差有关的不确定性的置信间隔。与因果关系文献的标准一样,模型和理论嵌入了潜在结果框架。介绍了两个关于潜在结果的一般性结果和关于生存数据的多复制危害模型的一般结果。