Enhancing Causal Inference in the Sentinel System: An Evaluation of Targeted Learning and Propensity Scores for Confounding Control in Drug Safety

Improving causal inference from real-world data (RWD) for purposes of evaluating safety and effectiveness of medical products is an important goal of the Sentinel Initiative. Current causal inference approaches implemented in the Sentinel System include methods based on propensity scores for control of confounding. Targeted Learning, which incorporates machine learning (ML) methods for causal inference, has been proposed as an alternative and potentially better approach.

The goal of this project is to evaluate the potential use of Targeted Learning methods within the Sentinel System compared to existing approaches. As Sentinel expands its data resources to include electronic health records (EHR), the project will focus on comparing the performance of Targeted Learning methods with causal inference approaches currently available to Sentinel in data environments that include linked insurance claims and EHR data. The project will further assess practical aspects of the feasibility of how such approaches might be implemented in a distributed environment.