Matched design for marginal causal effect on restricted mean survival time in observational studies

Abstract Investigating the causal relationship between exposure and time-to-event outcome is an important topic in biomedical research. Previous literature has discussed potential issues of using hazard ratio (HR) as marginal effect measure due to noncollapsibility. In this article, we advocate restricted mean survival time (RMST) difference a measure, which collapsible simple interpretation area under curves over certain horizon. To address both measured unmeasured confounding, matched design with sensitivity analysis proposed. Matching used pair similar treated untreated subjects together, generally more robust than modeling misspecifications. Our propensity score RMST estimator shown be asymptotically unbiased, corresponding variance calculated by accounting for correlation matching. Simulation studies also demonstrate that our method adequate empirical performance outperforms several competing methods practice. assess impact develop strategy adapting E -value approach data. We apply proposed Atherosclerosis Risk Communities Study (ARIC) examine smoking on stroke-free survival.