Doubly robust estimation of causal survival effects via inverse probability of treatment and censoring weighting and semi-parametric AFT models with variable selection.

Clinical trials and observational studies frequently encounter survival data with right censoring and high-dimensional confounders. This article proposes several estimators for assessing the average causal effect of a binary treatment on survival outcomes in the presence of confounders. These include doubly weighted estimators, regression model-based estimators, and doubly robust estimators. The proposed doubly robust estimators are shown to be consistent when either the treatment model or the semiparametric regression model for survival outcomes is correctly specified. A variable selection technique based on an adaptive sparse group lasso regularization approach is proposed to identify important confounders among high-dimensional covariates for both treatment and outcome models. This method enables simultaneous variable selection and regularization for both models. The performance of the proposed methodologies is thoroughly evaluated through simulation studies. Finally, the practical utility of the proposed methods is illustrated through applications to two real-world datasets.