(209) A Systematic Process Using Robust Signal Detection and Contextualization Methods in Treescan® and E360® to Reduce False Positives in Routine Pharmacovigilance

Publication Number: 1254

Background: Traditional pharmacovigilance (PV) approaches relying primarily on disproportionality analysis of adverse event (AE) reporting data often result in numerous false positives, few new signals, and consume significant resources to differentiate true signals. A more efficient process leveraging electronic healthcare data and robust signal detection and refinement analytic capabilities may complement traditional approaches.

Objectives: To describe and compare possible AE signals obtained via disproportionality analysis in FDA Adverse Event Reporting System (FAERS) data to potential signals detected in PharMetrics® claims using both TreeScan® software and E360® analytic platform.

Methods: We performed a retrospective analysis using FAERS and PharMetrics® using baricitinib, a recently approved JAK inhibitor indicated for rheumatoid arthritis (RA), as an example. In FAERS through May 2021, the reporting odds ratio (ROR) with a signaling cut-point of ≥2.0 was applied to detect potential signals. A cohort of baricitinib-exposed patients with RA and propensity-score matched controls consisting of RA patients on other JAK inhibitors or TNF blockers was built in E360®. Tree temporal scan statistics using self-controlled and propensity-score matched designs were generated on 3-level ICD-10 codes. Potential signals were identified based on log likelihood ratios (LLR) and p-values and were further evaluated using descriptive and inferential analysis in E360®.

Results: In FAERS, 25 possible signals with ROR ≥2.0 were detected. Most were either AEs included in the U.S. product label (n=14, 56%) or reflected confounding by indication (n=9, 36%). Using propensity score matched analysis in TreeScan®, 4 potential signals were detected at p< 0.05 (LLRs: 7.0–11.1); 3 of which (75%) were related to the indication and another of unknown relevance (“Fracture of foot/toe”). Using a self-controlled analysis, 2 potential signals were detected: “Eyelid disorder” (LLR=7.6,p=0.02) and “Fracture of foot/toe” (LLR=7.2,p=0.02). The latter was further evaluated in E360®. None of these FAERS AEs were detected in TreeScan when compared to time before treatment and to other advanced RA therapies. Baricitinib patients with a claim for “fracture of foot/toe” were significantly more likely to have osteoporosis (p < 0.001) and to have experienced a fall (p < 0.01) in the year preceding treatment initiation compared to baricitinib patients who did not have a foot/toe fracture.

Conclusions: A systematic approach to PV combining robust signal detection and evaluation with focus on reducing false positives from type-I error and confounding may effectively and efficiently complement more traditional analytic approaches to inform PV monitoring.