Estimating Average Treatment Effects for Time-to-Event Outcomes with Instrumental Variables: A Doubly Robust Approach

Oregon Convention Center 

Instrumental variable methods have been developed to estimate causal effects in observational studies in the presence of unmeasured confounding. However, existing approaches fall short in estimating average treatment effects (ATE) for time-to-event outcomes, often restricted to specific survival models and lacking desired statistical properties. In this study, we introduce a novel instrumental variable estimator of ATE in time-to-event outcomes, based on the cumulative incidence functions, accommodating scenarios with or without competing risks. Derived from efficient influence function, our estimator possesses double robustness and asymptotic efficiency, as theoretically proved and demonstrated via simulations. Our method enables the incorporation of various models for outcome, treatment and censoring, including machine learning and ensemble learning algorithms.

Causal inference

Competing risks

Conditional average treatment effect

Efficient influence function

Survival Analysis

Targeted maximum likelihood estimation 

Lifetime Data Science Section