Treatment Effect Estimation Under Additive Hazards Models With High-Dimensional Confounding

Estimating treatment effects for survival outcomes in the high-dimensional setting is critical many biomedical applications and any application with censored observations. This article establishes an “orthogonal” score learning effects, using observational data a potentially large number of confounders. The estimator allows root-n, asymptotically valid confidence intervals, despite bias induced by regularization. Moreover, we develop novel hazard difference (HDi), estimator. We establish rate double robustness through cross-fitting formulation. Numerical experiments illustrate finite sample performance, where observe that cross-fitted HDi has best performance. study radical prostatectomy’s effect on conservative prostate cancer management SEER-Medicare linked data. Last, provide extension to machine both approaches heterogeneous effects. Supplementary materials this are available online.