FDA Drug and Vaccine Adverse Event Knowledge Graph
FDA Adverse Event Reporting System (FAERS) and Vaccine Adverse Event Reporting System (VAERS) reports as RDF, with patient, drug, vaccine, reaction and outcome fields, linked to FDA applications, package NDCs and substances.
The FDA Drug and Vaccine Adverse Event Knowledge Graph publishes the two largest U.S. post-market safety reporting systems for human medical products as linked data: FAERS (openFDA drug/event, 20,692,690 reports, 19.26 billion triples) and VAERS (vaers.hhs.gov, 3,622,202 reports from 1990 on, domestic and non-domestic, 201 million triples), 19.46 billion triples in all. Every source field is kept except licensed terminology text (below), and every report carries its source dataset, report ID, source URL and download date. It is the seventh graph of the FDA regulated-products family and by far the largest: the FAERS Turtle is 134 GB gzipped in two files.
FAERS. fdaevent:AdverseEventReport, IRI https://bmedesign.org/fda-drug-event-kg/id/report/<safetyreportid>, with receive, receipt and transmission dates (xsd:date), seriousness flags, report source, sender and primary source qualification. Child nodes per report: the patient (age, sex, weight, death date), one fdaevent:FaersDrug per drug (76,979,798 in all: characterization, product name, dosage, route, indication, start and end dates, action taken, openFDA harmonized fields) and one fdaevent:FaersReaction per reaction (62,097,430: MedDRA version and outcome code).
VAERS. fdaevent:VaersReport, IRI .../vaers-report/<VAERS_ID>, with all columns of the VAERS data file (receive date, state, age, sex, outcome flags, recovery, onset interval, reporter narrative text, lab data, medical history, allergies). fdaevent:VaersVaccination (4,357,556, one per vaccine: type, manufacturer, lot, dose series, route, site, name) and fdaevent:VaersSymptomSet (4,994,174, one per symptom row: term count and MedDRA versions) link back to their report.
Links to other graphs. fdaevent:hasApplication, fdaevent:hasPackageNdc and fdaevent:hasSubstance on FAERS reports point at the application, package NDC and UNII substance IRIs of fda-drug-biologic-kg and fda-substance-kg, lifted from each drug entry's openFDA block.
Licensed terminology. Reaction and symptom terms in both sources are MedDRA preferred terms. MedDRA term text is withheld from the published graph under the MedDRA license (MedDRA versions, outcomes and every non-MedDRA field are published in full). Links from every FAERS reaction and VAERS symptom to MedDRA PT and LLT codes, and the SOC to LLT hierarchy, are built under the project's MedDRA subscription and will be added once the MSSO confirms in writing what may be published.
Validation. Every file passes riot --validate. VAERS passes the graph's SHACL shapes in full; FAERS, too large for a single SHACL run, passes a full-coverage streaming check of the same rules plus SHACL on sampled chunks.
Responsible use. These are passive, voluntary reports: counts are not incidence rates, and a report does not establish that a product caused the event. VAERS reports are accepted without verification.
Source and license. openFDA drug/event and the VAERS data downloads (HHS, CDC and FDA): public domain U.S. government works. Funded under NSF award 2535091 (Proto-OKN Theme 2: OKN-Fabric) through subaward 5136002. The graph is released under CC0 1.0.
provenance:
- derivedFrom:
label: openFDA drug/event (FAERS)
uri: https://open.fda.gov/apis/drug/event/
kind: Database
license:
data:
label: Public Domain
spdx: CC0-1.0
uri: https://open.fda.gov/license/
- derivedFrom:
label: VAERS data downloads
uri: https://vaers.hhs.gov/data/datasets.html
kind: Database
license:
data:
label: Public Domain
spdx: CC0-1.0