B1113
Title: Generalizing treatment effects with incomplete covariates
Authors: Imke Mayer - Charite Universitaetsmedizin Berlin (Germany) [presenting]
Julie Josse - INRIA (France)
Abstract: The focus is on the problem of generalizing a causal effect estimated on a randomized controlled trial (RCT) to a target population described by a set of variables from observational data. Available methods such as inverse propensity sampling weighting are not designed to handle missing values, which are, however, common in both data sources. In addition to coupling the assumptions for causal effect identifiability and the mechanism of the missing value and to defining appropriate estimation strategies, one difficulty to consider is the specific structure of the multi-source data with only partial information on treatment and outcome. We propose multiple imputation strategies to handle missing values when generalizing treatment effects, each handling the multi-source structure of the problem differently. As an alternative, we also propose a machine learning-based estimation approach that treats incomplete covariates as semi-discrete variables. The proposed strategies rely on different sets of assumptions concerning the impact of missing values on identifiability. We discuss these assumptions and assess the methods through an extensive simulation study, as well as on a large major trauma registry and an RCT to study the effect of the drug tranexamic acid on mortality in major trauma patients admitted to ICU. This analysis illustrates how the handling of the missing value can impact the conclusion about the effect generalized from the RCT to the target population.
