All Times EDT
Keywords: Noncollapsibility, matched design, sensitivity analysis
Time to event outcome is commonly encountered in clinical research and drug discovery. Multiple papers have discussed the limitation of using hazard ratio as a marginal causal effect measure due to its noncollapsibility and the time-varying nature. We adopt the restricted mean survival time (RMST) difference as a marginal measure, since it essentially captures the mean discrepancy and has simple interpretation as the difference of areas under survival curves. To remove observed confounding, both propensity score matching and weighting based methods are developed. The matching method provides an asymptotically unbiased estimator and does not rely on correct outcome model specification. The weighting method has the doubly robust property and may lead to more efficient estimation. Simulation studies demonstrate that our estimators have favorable performance as compared to other competing methods. Sensitivity analysis based on E-value is also performed in a real world setup, to assess the impact due to potential unmeasured confounding.