A1331
Title: Robust trial augmentation for survival outcomes
Authors: Guanbo Wang - Dartmouth College (United States) [presenting]
Abstract: Randomized trials often lack sufficient sample sizes to provide precise estimates of average treatment effects, particularly when outcomes are time-to-event and subject to censoring. Incorporating external data is a promising strategy to improve statistical efficiency, but incompatibility between external and trial data can introduce bias in estimating effects for the trial population. A method has been proposed that leverages external data to improve efficiency while preserving consistency, even when incompatibility is present. However, this approach does not accommodate survival outcomes. An extension of the framework is developed to create a robust augmentation procedure for survival endpoints. The method guarantees that, whenever trial-only analyses yield consistent estimates of the ATE, the augmented estimator remains consistent and never performs worse in terms of asymptotic efficiency than the most efficient estimator based solely on trial data.
