Strengthening inferential studies in the FDA Sentinel initiative: results from a methodological demonstration project

This is the authors' abstract. We don't add key points for this paper.

The FDA Sentinel Real-World Evidence Data Enterprise (RWE-DE) contains linked electronic health records with claims data for over 25 million patients. To demonstrate the applicability of the RWE-DE to a study previously considered infeasible in claims-based Sentinel network, we emulated a target trial using a case study comparing acute pancreatitis among new users of sodium-glucose cotransporter-2 inhibitors (SGLT-2i) with new users of dipeptidyl peptidase-4 inhibitors (DPP-4i) among 97,119 patients with type 2 diabetes mellitus from HealthVerity [2018-2020] and TriNetX [2013-2024] databases. We applied a previously validated computable phenotyping algorithm using EHRs to identify acute pancreatitis as the primary outcome. After confounding adjustment for >135 variables using propensity score fine stratification weighting, the hazard ratio (95% confidence interval) for acute pancreatitis following SGLT-2i initiation compared to DPP-4i initiation was 0.85 (0.67-1.07) for intent-to-treat and 0.84 (0.58-1.22) for per-protocol analysis. This study serves as a proof-of-concept for future safety assessments in Sentinel.