CMStatistics 2020: Start Registration
View Submission - CMStatistics
Title: Robust estimation of the average treatment effects in presence of right-censoring and competing risks Authors:  Brice Ozenne - University of Copenhagen (Denmark) [presenting]
Thomas Scheike - University of Copenhagen (Denmark)
Thomas Alexander Gerds - University of Copenhagen (Denmark)
Abstract: Average treatment effects (ATE) are important parameters in pharmacoepidemiology where the aim is to evaluate differences between treatments based on health care databases. Estimation of the ATE is complicated by the occurrence of competing events (e.g. death), patient drop-out, and confounders present in non-randomized data. Several types of estimators for the ATE can be derived based on working regression models for the outcome, censoring, and treatment distributions. However, the traditional G-formula or inverse probability weighting (IPW) estimators require well-specified working models to be unbiased. We will present how results from the semi-parametric theory can be used to derive a doubly robust estimator for the ATE. We show, both theoretically and using simulation studies, that the proposed estimator is robust to misspecification of some of the working models and compare it to G-formula and IPW estimators. We will also discuss the use of the functional delta method to obtain the asymptotic distribution of the robust ATE estimator. The proposed robust ATE estimator is implemented in the ate function of the riskRegression package (available on CRAN).