Pharmacoepidemiology
Safety signal identification in the Sentinel System: Consistent findings for medical products over various study designs Ashley I. Michnick* Ashley I. Michnick Leah B. Herity Geetha S. Iyer Suraj Rajasimhan Jummai Apata Jillian Burk Derek Campbell José J. Hernández-Muñoz Michelle Hines Nathan Kim Joy Kolonoski Nora McElroy Sampada Nandyala Amelia Thyen Megan Wiley Judith C. Maro Monica A. Muñoz
Introduction. The U.S. FDA performs medical product (MP) safety signal identification (SI) in its Sentinel System using tree-based scan statistics that detect unexpected health outcomes of interest (HOIs) after MP initiation. HOIs occurring more frequently (p<0.05) are considered alerts, which are investigated and classified as a clinically meaningful safety signal or not.
Methods. In the Sentinel Distributed Database, we performed SI for new users of risankizumab-rzaa and guselkumab using propensity score-matched active comparator (PSMAC) and self-controlled risk interval (SCRI) study designs where each agent served both as exposure and comparator. Tree-based methods scanned across inpatient and emergency department diagnosis codes. The SCRI was inherently adjusted, and the PSMAC design used 1:1 high-dimensional PSM. We assessed whether HOIs occurred more frequently in the risk vs. control period (SCRI) or group (PSMAC) and evaluated whether clusters of HOIs occurred more frequently in variable risk windows after treatment compared to all other periods in the same window.
Results. The SCRI had approximately 30,000 new users of each drug, and the PSMAC had 14,819 matched new users. For both MPs, no HOIs occurred more frequently in the risk vs. control period (SCRI) or group (PSMAC). No clusters of HOIs occurred more frequently in variable risk windows after guselkumab treatment, but we identified an increase (p<0.05) in cholelithiasis without cholecystitis on days 9-11 after risankizumab-rzaa initiation compared to all other days in the six months after. Further investigation of line-level healthcare claims suggested this was likely incidental, therefore it wasn’t considered a safety signal.
Conclusions. This study demonstrates the feasibility of using two study designs with varying strengths to perform signal identification among large numbers of MP users. Tree-based scan statistics are valuable complementary methods to augment MP safety surveillance.
Disclosures. This project was supported by Task Order 75F40123F19009 under Master Agreement 75F40119D10037 from the US Food and Drug Administration (FDA). The contents are those of the authors and do not necessarily represent the official views of, nor an endorsement, by FDA/HHS, or the U.S. Government.