VigiBase, EudraVigilance, and FAERS — Comparing the Big Databases

The three databases covered in this lesson each answer the same underlying question this module has been building toward — is a drug-event pair statistically overrepresented — but they do it with meaningfully different scope, governance, and preferred metrics, and understanding those differences is part of correctly interpreting any specific signal that gets discussed by name in the literature or in regulatory communications.

VigiBase, run by the WHO-UMC (the same Uppsala Monitoring Centre behind the WHO-UMC causality scale from Module 6), is the broadest in population terms: it aggregates individual case safety reports from more than 150 national pharmacovigilance centres worldwide, giving it genuine global reach that neither of the other two databases can match on its own. Its primary disproportionality metric is IC, the Bayesian log2-scale measure introduced in Lesson 7.3, and its breadth makes it particularly valuable for surfacing signals that might be too rare to detect reliably within any single country’s data alone.

EudraVigilance, run by the EMA and covered from the regulatory-process side in the previous lesson, is scoped to the EU/EEA and is deeply integrated with the PRAC signal management pipeline — a signal surfaced in EudraVigilance flows fairly directly into the same regulatory assessment process that can lead to a label change or referral. FAERS, run by the FDA, covers US-marketed drugs and biologics, uses EBGM (with EB05 as its practical screening threshold, per Lesson 7.5) as its primary metric, and is distinctive for something the other two aren’t: full case-level data, with identifying information removed, is publicly downloadable via openFDA — which is a large part of why so much independent academic signal-detection research, even on globally marketed drugs, specifically uses FAERS data.

The practical consequence worth internalising is that a signal appearing strongly in one of these three databases and appearing weak or absent in another isn’t necessarily a contradiction — it can genuinely reflect differences in population (a drug prescribed more heavily in one region than another), reporting culture (some countries and healthcare systems report more completely than others), and the sensitivity of each database’s preferred metric. A PV professional reading "a signal was identified in FAERS" needs to hold that claim distinctly from "a signal was identified in VigiBase" — they’re not interchangeable statements, even when they’re describing what might ultimately turn out to be the same underlying safety issue.