Research tools · Beta
VAERS signal browser
For every vaccine and every reported event in the U.S. Vaccine Adverse Event Reporting System (VAERS), this table shows whether the event is reported more often with that vaccine than with other vaccines — using the five standard statistics that safety agencies use — and then checks whether that excess survives the usual distortions, and what controlled studies have found.
| Event (MedDRA PT) | Reports | PRR 95% CI | ROR | EB05 | IC025 | Methods | Age-adj. | Single-vaccine | Non-COVID comp. | Serious-only | Same era | Category |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | ||||||||||||
How to read the columns
For a vaccine V and an event E, a 2×2 table counts reports: a = V with E, b = V without E, c = other vaccines with E, d = other vaccines without E. Only U.S. reports are used; each report counts once per vaccine and once per event.
- PRR (proportional reporting ratio) = [a/(a+b)] / [c/(c+d)]. Flag (Evans 2001, used by CDC for VAERS): PRR ≥ 2, χ² ≥ 4, at least 3 reports.
- ROR (reporting odds ratio) = ad/bc. Flag: lower 95% limit > 1.
- IC025: lower limit of the Bayesian information component (BCPNN, used by WHO's Uppsala Monitoring Centre). Flag: > 0.
- EB05: lower 5% limit of the empirical-Bayes geometric mean (MGPS, DuMouchel 1999; used by FDA and CDC for VAERS). It shrinks ratios built on few reports toward 1. Flag: ≥ 2.
- Methods: how many of these four flag the pair.
Then four checks recompute the PRR to remove a known distortion. ✓ = the lower 95% limit stays above 1; ✗ = it does not.
- Age-adjusted: Mantel–Haenszel PRR within age bands. Without it, every event common in babies looks like a signal for infant vaccines, because most other reports are from adults.
- Single-vaccine: only reports naming this vaccine alone. Without it, an effect of one vaccine shows up for everything given at the same visit (intussusception, caused by rotavirus vaccine, also appears for DTaP).
- Non-COVID comparator: COVID-19 vaccine reports (about half of VAERS, mostly adults, with mandatory reporting of some outcomes) removed from the comparison group.
- Serious-only: the same comparison among reports of death, life-threatening illness, hospitalisation, disability or birth defect.
- Same era: only reports received from 2021 on. Newer MedDRA terms and coding habits arrived with the COVID-19 surge, so a term used almost only in COVID-19 reports can look like a huge signal; this check compares like with like.
Categories
- Established an association accepted as causal by controlled studies or expert review (GRADE ratings on this site, NASEM 2012, ACIP, FDA labels). The row names the source. Most of these are rare.
- Studied, not found controlled studies looked for this and did not find an increased risk (for example autism after MMR). The VAERS signal reflects reporting, not risk.
- Expected reaction sore arm, fever, tiredness and similar short-lived reactions listed on vaccine labels.
- Target disease the disease the vaccine prevents, vaccine failure or vaccine-strain infection.
- Administrative / test and Work-up terms that are not events (dosing errors, test results, procedures).
- Candidate a clinical event flagged by at least 3 of the 4 methods that none of the above explains. These form the causality review queue.
- Weak flagged by fewer than 3 methods.
Categories are assigned by rules and a curated list of topics (code: analysis/vaccine_signal_triage.py), not by reading reports. A candidate is not a suspected harm; most will turn out to be explained by who gets the vaccine, publicity, or chance across tens of thousands of comparisons.
Causality review queue
Candidates with at least 10 reports that survive the age-adjusted and single-vaccine checks (and, for COVID-19, the same-era check), ranked by EB05. Each will get a written assessment (Bradford Hill viewpoints, controlled-study search, Brighton case documentation) in the style of the flagged signal pairs. Status is shown on each row as assessments are added.
| Vaccine | Event | Reports | EB05 | Age-adj. PRR | Single-vaccine PRR | Status |
|---|---|---|---|---|---|---|
| Loading… | ||||||
Limits
- Disproportionality compares reports with reports. A vaccine that is safe overall can show many signals, and a real harm can be missed when every vaccine is reported for it.
- Reports are not verified. Duplicates exist (about 2% by our estimate); some events are reported because of publicity, litigation or mandatory reporting rules.
- About 46,000 pairs are tested, so hundreds will pass any threshold by chance alone.
- Vaccine classes group products (e.g. all influenza vaccines, mRNA and adenovirus-vector COVID-19 vaccines together); product-level signals can be diluted or mixed.
Beta. Source: VAERS public data files (U.S. reports, 1990–2025). Code: analysis/vaers_pair_signals.py, analysis/vaccine_signal_triage.py. Data: data/signals/. Updated 2026-10-14. Report a correction.