| AEMS / FAERS signals dataset — 2025Q4 | |
| Metric | Value |
|---|---|
| Distinct drug–event pairs | 881,751 |
| Total reports (sum of observed counts) | 4,802,844 |
| Pairs flagged by ≥ 2 of 4 methods (our signal rule) | 265,108 |
| Pairs flagged by all 4 methods | 96,735 |
Inside the AEMS Data: What 4.8 Million Reports Do and Don’t Tell You
The FDA’s new Adverse Event Monitoring System, the quarterly data behind our signal analyses, and why the strongest ‘signal’ is usually an artifact
What AEMS is
On March 11, 2026 the FDA launched the Adverse Event Monitoring System (AEMS), a single real-time public platform that consolidates the agency’s previously separate adverse-event databases — FAERS (drugs and biologics), VAERS (vaccines), and the animal AERS — with devices, food, supplements, and tobacco folding in through 2026. The stated reasons were cost (the legacy systems ran ~$37M/year) and fragmentation; the new system publishes in real time rather than quarterly and adds an AI-assisted dashboard.
For an independent analyst, the important fact is quieter and reassuring: the bulk quarterly data-extract files continue. They are still published, still free, still in the same ASCII/XML format that begins in 2004 — now under the AEMS name, with the January–March 2026 quarter posted on April 28, 2026. Those extracts are the raw material behind every analysis on this site. This page is a look at the data itself: how much of it there is, and how to read it without fooling yourself.
The current data at a glance
The most recent quarter in our processed dataset is 2025Q4. Within its four-quarter rolling window:
Every one of those 881,751 pairs is scored by four disproportionality methods — the Gamma-Poisson Shrinker (GPS/EBGM), the Proportional Reporting Ratio (PRR), the Reporting Odds Ratio (ROR), and the BCPNN Information Component (IC). A pair becomes a “signal,” in our fixed convention, when its EB05 is at least 2.0 and at least two of the four methods agree.
Here is how the whole quarter distributes across method agreement:
| How many methods agree, across every pair this quarter | ||
| Methods flagged | Pairs | Share |
|---|---|---|
| 0 | 337,516 | 38.3% |
| 1 | 279,127 | 31.7% |
| 2 | 139,892 | 15.9% |
| 3 | 28,481 | 3.2% |
| 4 | 96,735 | 11.0% |
| Most pairs are flagged by no method at all. Agreement by all four is the minority — and, as the next section shows, even that is not enough on its own. | ||
Why we don’t publish a “top signals” leaderboard
It is tempting to sort those 96,735 four-method pairs by EB05 and call the top of the list “the biggest safety signals.” Do not trust that list. The strongest disproportionality scores are dominated not by discoveries but by artifacts — and looking at the actual top of the current quarter makes the point better than any warning:
| Four of the strongest ‘signals’ this quarter — and why each is an artifact | ||||
| Drug | Event | Observed | EB05 | Why it's not a discovery |
|---|---|---|---|---|
| endari | Sickle cell anaemia with crisis | 219 | 2893 | Indication. Endari (L-glutamine) is prescribed for sickle cell disease — so its reports are full of sickle cell crises. The drug treats the condition it appears to ‘signal’ for. |
| gold bond original strength | Mesothelioma | 322 | 2056 | Litigation cluster. A wave of talc-related legal reporting, not a pharmacovigilance discovery. Mass-tort campaigns generate synchronized report bursts unrelated to new biology. |
| nirsevimab | Respiratory syncytial virus bronchiolitis | 77 | 2825 | Indication / route. Nirsevimab is given to prevent RSV disease in infants; RSV terms saturate its reports by construction. |
| heparin | Heparin-induced thrombocytopenia | 72 | 2460 | Definitional. A real, well-known reaction — but an eponymous one already in every label for decades. Disproportionality ‘rediscovering’ it is confirmation, not signal. |
| EB05 and observed counts are pulled live from the current data; the annotations are fixed. | ||||
Each of these clears every statistical bar — high EB05, all four methods, hundreds of reports — and each is meaningless as a discovery. They illustrate the four ways a raw leaderboard lies:
- Confounding by indication — a drug’s reports are full of the disease it treats.
- Litigation and media clusters — synchronized reporting campaigns, not biology.
- Route and administration effects — the reports describe how the drug is given.
- Definitional / known reactions — real, but already on the label for years.
A disproportionality score is a question generator, never an answer. The number tells you where to look; it cannot tell you what you will find.
Analysis done right
Useful signal detection starts from a specific, pre-registered question and spends most of its effort on confounding, not on the score. Two pieces on this site show that pattern on published work (not a raw leaderboard):
- Signal & Noise — the house methods field guide: one real signal, one trap, one fix at a time; fixed rule of EB05 ≥ 2 with ≥ 2 of 4 methods agreeing.
- Tested Against VAERS — Christine Cotton’s trial-era safety categories checked against the post-market VAERS record with the same four-method consensus filter.
Analyses on this site present event names as they appear in the public FAERS/VAERS releases (Preferred Term labels in those extracts). We do not redistribute MedDRA terminology files or hierarchy products.
Explore it yourself
To search the full dataset interactively — any drug, any event, signal trajectories over quarters — use faers.mobi, which runs the same four-method engine on the same AEMS/FAERS extract with server-side search across every drug–event pair.
NoteHow to read anything on this page
These are counts of reports, not counts of patients harmed. Adverse-event reporting is voluntary, unverified, and shaped by attention and litigation. Disproportionality compares a pair’s reporting rate to the rest of the database; it cannot establish causation, and a high score is a reason to investigate, not a conclusion. Aggregate counts like those above carry no patient-identifying information.
Methods & data
Data: FDA Adverse Event Monitoring System (AEMS) quarterly extract files, continuing the FAERS series (https://fis.fda.gov/extensions/FPD-QDE-FAERS/FPD-QDE-FAERS.html), processed into four-quarter rolling windows and scored by GPS/EBGM (EB05 posterior lower bound), PRR, ROR, and BCPNN/IC. Signal rule: EB05 ≥ 2.0 with ≥ 2 of 4 methods, applied uniformly. Figures on this page are computed live from the current processed dataset (latest quarter: 2025Q4).
Sources: FDA, “FDA Launches New Adverse Event Look-Up Tool” (https://www.fda.gov/news-events/press-announcements/fda-launches-new-adverse-event-look-tool); CIDRAP, “FDA announces AEMS, new adverse-event database to replace VAERS” (https://www.cidrap.umn.edu/public-health/fda-announces-aems-new-adverse-event-database-replace-vaers).