Most pharmacovigilance teams working on globally marketed products run signal detection on FAERS and EudraVigilance separately. They produce a FAERS signal detection report and a European spontaneous reporting analysis, typically on different schedules, using tools that may not share a common data infrastructure. When the two analyses happen to flag the same association, it's treated as confirming evidence. When they diverge, the divergence is noted but often not deeply investigated.
This parallel-but-separate approach is operationally practical but analytically suboptimal. The two databases are not independent samples of a common patient population — they differ in reporter composition, coding conventions, submission practices, and which drug-event combinations are enriched in each region's patient population. Understanding those structural differences is a prerequisite for interpreting what FAERS-EudraVigilance concordance and discordance actually mean.
What Each Database Contains and Doesn't Contain
FAERS, operated by FDA, receives ICSRs from US-market adverse event reporters: manufacturers with US marketing authorizations (mandatory reporting under 21 CFR Part 314.81/601.12), US healthcare providers and consumers (voluntary), and foreign regulatory authority submissions for cases where the event occurred outside the US but the product has US approval.
EudraVigilance, operated by EMA, receives ICSRs from MAH holders with European Economic Area marketing authorizations, national competent authorities across EU/EEA member states (the "yellow card" systems), and from third-country authority submissions. EudraVigilance data is subject to EudraVigilance access policies; the publicly accessible portion (via the EVWEB portal) provides aggregate-level data and individual case signal detection reports for products with European public assessment reports (EPARs). Full ICSR-level access for signal detection purposes requires MAH licensee status through the EMA's Industry Data service.
The key asymmetry: FAERS full quarterly downloads are publicly accessible with no licensee requirement beyond accepting the terms of use. EudraVigilance ICSR-level data for signal detection requires being a MAH with an EMA agreement. For PV teams that need both datasets, the data governance workflows are different in kind, not just in detail.
The MedDRA Version Problem
Both databases use MedDRA for adverse event coding, but MedDRA is a versioned terminology that releases updated versions twice annually — typically March and September. Each release may introduce new PTs, retire or split existing PTs, and remap HLT/HLGT/SOC relationships. In any given year, FAERS and EudraVigilance may be running on different MedDRA versions, and historical data in each database was coded under the MedDRA version that was current at the time of submission — not the current version.
In practice, this means that a FAERS search and a EudraVigilance search for "the same adverse event" may return different report counts even for the same drug and time period, because the PTs used to code that event have been handled differently across version transitions in the two databases. EMA issues "MedDRA retirement mapping" guidance for handling version transitions in EudraVigilance, and FDA's technical documentation notes MedDRA versioning in the FAERS file structure. But neither database automatically recodes historical reports when a PT changes — historical reports carry the coding that was applied when they were submitted.
For any cross-database analysis, you need an explicit MedDRA version reconciliation step. Queries built against the current MedDRA version hierarchy should be mapped back through the version history to catch reports coded under predecessor PTs. For events where MedDRA has had multiple PT changes over the analysis window, this mapping can require tracing through several version transitions.
Reporter-Type Composition Differences
The reporter-type mix in FAERS and EudraVigilance differs in ways that systematically affect the types of signals that are enriched in each database. FAERS is heavily dominated by MAH-submitted manufacturer reports — roughly 80-85% of recent years' volume. Consumer and patient reports, while growing, remain a minority.
EudraVigilance has historically had a higher proportion of healthcare provider-submitted reports relative to FAERS, partly because EU member state yellow card systems feed into EudraVigilance, and those systems have different incentive structures and reporting cultures than the US voluntary reporting system. The UK Yellow Card scheme, pre-Brexit, was one of the largest contributors of HCP-submitted reports in EudraVigilance. Post-Brexit, UK MHRA data flows to its own system and is no longer included in EudraVigilance, which changed the reporter composition for some product categories.
Why does this matter for cross-database signal interpretation? Because HCP-submitted reports and manufacturer-submitted reports have different sensitivities for different types of signals. HCP-submitted reports tend to capture clinical events that treating physicians find noteworthy; manufacturer-submitted reports reflect the pharmacovigilance team's assessment of which events meet expedited reporting criteria. An adverse event category that is common, non-specific, or expected given the drug's label may be systematically underreported by manufacturers while being captured more fully by HCPs who encounter it clinically.
Indication and Population Differences Between Regions
For many globally marketed drugs, the approved indication in the US and the approved indication in the EU are not identical. A drug approved for a broad indication in the US may have a narrower label in the EU due to different clinical trial evidence submitted in each jurisdiction, different risk management requirements, or timing differences in regulatory review. When this is the case, FAERS and EudraVigilance are not sampling from the same patient population even if the drug is the same molecule.
Dosing conventions can also differ between regions. Some drugs have substantially different approved dose ranges between FDA and EMA labels, which affects the adverse event profile of the reported population. QT prolongation signals, hepatotoxicity signals, and dose-dependent renal adverse events may appear at different apparent rates in FAERS vs. EudraVigilance simply because the typical dose in use differs by region.
Co-medication patterns vary by region as well. A drug approved for second-line use in an oncology indication may be used in combination with different backbone therapies in US vs. European treatment guidelines, which affects what polypharmacy context shows up in the ICSRs for that drug in each database.
Practical Approach to Cross-Database Signal Integration
Given these structural differences, here is the workflow logic we find most defensible for teams that need to integrate FAERS and EudraVigilance signal detection:
First, treat each database's signal detection as producing independent evidence rather than attempting to merge them into a single pooled analysis. The databases are not interoperable at the ICSR level without extensive harmonization work — pooling report counts across them without harmonization produces misleading statistical results.
Second, define a pre-specified concordance framework: what constitutes a concordant signal (flagged in both databases within a defined time window), a FAERS-only signal, and a EudraVigilance-only signal? Each category has a different interpretation. Concordant signals are more likely to reflect a true pharmacological association than regional or methodological artifact. FAERS-only signals may reflect US-specific co-medication patterns, US-specific population characteristics, or differential reporting behavior. EudraVigilance-only signals similarly deserve investigation of the regional specifics before concluding the signal is a FAERS miss.
Third, for MedDRA reconciliation, maintain a running version-mapping table that covers the full analysis window. When a PT was retired or renamed during the analysis period, include predecessor PTs in your query. When comparing report counts across databases, document which MedDRA version was used in each query.
We're not saying that cross-database signal confirmation is the only legitimate evidence standard — a strong FAERS signal with a plausible mechanism and supporting preclinical data is worth a serious signal assessment regardless of what EudraVigilance shows. We're saying that the interpretation of cross-database comparisons depends on understanding why each database shows what it shows, and that the structural differences described here are the primary reasons signals diverge across databases.
The Data Gap That Matters Most
The most practically significant gap in cross-database integration isn't a technical one — it's a timing problem. FAERS quarterly releases are published approximately 60-90 days after the end of the calendar quarter, with a lag that reflects FDA's processing and validation pipeline. EudraVigilance ICSR data for MAH licensees is available more continuously but also with submission lags from member state yellow card systems. The lags are different in magnitude and don't align to the same calendar.
For a PV team running quarterly signal detection cycles, this means that even if you have access to both databases, your FAERS analysis and your EudraVigilance analysis are rarely looking at exactly overlapping time windows. A signal that emerges in European data in Q1 may not appear in FAERS until Q2 or Q3 as US reports accumulate. Temporal misalignment in cross-database comparison is a systematic source of apparent discordance that doesn't reflect genuine regional signal differences.
The practical fix is straightforward but requires discipline: explicitly document the data cutoff date for each database in each analysis cycle and account for the lag differential when interpreting concordance. A signal that appears in EudraVigilance three months before FAERS is not necessarily a European-specific signal — it may simply reflect that EudraVigilance had more complete data for that window at the time of analysis.