Observed vs expected: are serious events reported more often than background rates predict?

Results

1 of 13 vaccine–event pairs have an observed/expected ratio whose 95% interval lies above 1 (highlighted). "adj." adds reports with unknown onset date in proportion. The sensitivity column re-computes O/E with the alternative background rate shown.

Anaphylaxis: how complete is reporting?

The Vaccine Safety Datalink found medically confirmed anaphylaxis after 1.31 per million doses (McNeil et al., J Allergy Clin Immunol 2016). Anaphylaxis is dramatic, immediate and usually reported, so the VAERS reporting rate below, compared with that confirmed rate, shows how VAERS counts relate to real cases for a well-reported event.

VAERS anaphylaxis reports are not all confirmed anaphylaxis (Brighton level 1–3), so the reporting rate can exceed the confirmed rate.

How many reports document a confirmable case? (Brighton criteria)

The Brighton Collaboration defines each adverse event at three levels of diagnostic certainty (Level 1 = most certain, e.g. GBS with nerve-conduction studies and CSF findings; Level 3 = clinical features only). Each VAERS narrative and lab field was screened for the findings each level requires. Bars show the share of reports whose documentation would reach each level; grey = not enough information to classify, which is not the same as "not a case".

Automated keyword proxy, not case adjudication: keywords can be negated, and "no alternative diagnosis" cannot be checked. A hand check of random narratives agreed with the proxy in 14/15 GBS reports and 7/10 myocarditis reports (myocarditis Level 1 is somewhat overstated because mentioning a test is not the same as an abnormal result). Definitions: Guillain-Barré syndrome, Myocarditis, Thrombocytopenia / ITP, Anaphylaxis, Intussusception. Code: analysis/brighton_proxy.py; data: brighton_proxy.json.

Limitations

• Doses are doses distributed (HRSA, from CDC) or administered (COVID-19, CDC); distributed overstates administered, which lowers E slightly too much.
• Background rates are general-population incidences, mostly from claims or registry codes, which run higher than chart-confirmed rates; VAERS reports are unconfirmed. Using code-based backgrounds tends to push O/E down.
• All-age backgrounds are applied to all-age doses because U.S. doses by age are not published for most vaccines; where the event's incidence differs strongly by age the sensitivity column shows the effect.
• Risk windows follow published AESI protocols; a different window changes both O and E.
• Pre-specification: the pairs, windows and background rates were fixed in the protocol before counts were computed.

Common questions

What does an observed-to-expected ratio above 1 mean?

More reports arrived inside the risk window than background incidence predicts for that many doses, even though VAERS misses many events. That is a strong signal that warrants controlled studies; it is how myocarditis after mRNA vaccines and GBS after some vaccines were first flagged.

Does an O/E below 1 mean a vaccine is safe for that outcome?

No. VAERS under-reports most events, often by a large and unknown factor, so O/E below 1 is expected even if the vaccine has no effect at all. Reassurance comes from controlled studies, which are listed on each vaccine page.

Sources

• Black SB et al. 2021 Vaccine (The critical role of background rates of possible adverse events in the assessment of COVID-19 vaccine safety) 10.1016/j.vaccine.2021.03.016; PMID 33846042; PMC7936550
• Christensen KJ et al. 2021 Acta Neurol Scand 10.1111/ane.13420; PMC8178233
• Gubernot D et al. 2021 Vaccine (FDA/CDC literature review of US rates) 10.1016/j.vaccine.2021.05.016; PMC8118666
• Payne AB et al. 2022 Res Pract Thromb Haemost (CDC) 10.1002/rth2.12682; PMC8901465
• Tate JE et al. 2008 Pediatrics 10.1542/peds.2007-1590; PMC2680116
• Willame C et al. 2023 Vaccine (ACCESS project, EMA-funded, VAC4EU) 10.1016/j.vaccine.2022.11.031; PMC9678835
• Doses: HRSA VICP statistics report (doses distributed 2006–2024, CDC data), edition 09/01/2026; CDC COVID-19 Vaccinations in the United States (data.cdc.gov unsk-b7fc).
• Reports: VAERS public data files (domestic reports), downloaded June 2026 release. Code: analysis/vaers_analyses.py, analysis/observed_vs_expected.py; data: observed_vs_expected.json.