Signal Detection for Small National Pharmacovigilance Centres: A Reference-Standard Evaluation of Local-Only Screening Versus External Screening with Local Triage, with an Arab-Region Case Study

Preprint. Disproportionality algorithms were validated in databases holding millions of reports, yet many national pharmacovigilance centres, including most in the Arab region, hold only thousands. Using all 2024 FAERS reports (1,480,544 deduplicated cases) and 475 evaluable OMOP/EU-ADR reference pairs (196 positive, 279 negative controls), we evaluated PRR, ROR, the BCPNN information component (IC) and MGPS EBGM in random subsamples of 1,000–1,000,000 reports and in the 5,839 FAERS reports that occurred in Arab League states.

Local-only IC sensitivity was 3.3% at 5,000 reports, 23.2% at 50,000 and 31.9% at 100,000 (full database 54.6%). In the Arab-occurrence database, local-only screening detected 3.1% of positive controls; screening the rest of FAERS and keeping only signals for locally reported drugs detected 45.4% (specificity 94.6%; +42.3 percentage points, 95% CI 35.7–49.2; McNemar p<0.001), and 82.4% among locally used drugs. Small centres should screen large external databases and triage locally, report database size with any "no signal" finding, and build reporting volume.

Files: manuscript PDF; code and derived data (raw FAERS not redistributed; downloaded by the code), including an open list of 1,854 screening signals for clinical review. Reporting follows READUS-PV.