openFDA FAERS Adverse Event API

Every product on an American drugstore shelf carries a second, quieter paper trail: the reports of what went wrong afterward. The openFDA FAERS Adverse Event API is the FDA Adverse Event Reporting System made addressable - 20,692,690 reports at the July 2026 count, each a structured case of a suspected side effect, a medication error, a product quality problem or a therapeutic failure involving a prescription or over-the-counter medicine. Cases follow the international ICH E2b/M2 safety-reporting standard, so a report filed in Canada parses identically to one filed in Ohio.

For Drug Retail work this is the risk ledger behind the shelf: who got hurt, how old they were, what they were taking, which product the reporter named, and how the case ended. Get a sample.

One genuine case out of the twenty million, flattened for reading - a fatal transdermal case from Canada riding on an eight-digit safety report identifier whose final digit is a checksum:

safetyreportid                5801206-7
receivedate                   20080707
serious                       1        seriousnessdeath: 1
primarysource.reportercountry CANADA
patient.patientonsetage       26       unit code 801 = years
patient.patientsex            1        1 = male
patient.drug[0].medicinalproduct         DURAGESIC-100
patient.drug[0].drugcharacterization     1        1 = suspect
patient.drug[0].drugadministrationroute  041      041 = transdermal
patient.drug[0].drugindication           DRUG ABUSE
patient.reaction[0].reactionmeddrapt     DRUG ADMINISTRATION ERROR
patient.reaction[1].reactionmeddrapt     OVERDOSE

Cases are append-only: a new report arrives under a fresh identifier rather than as an edit to an old one. Retained consecutive deliveries therefore build an arrival history - you can watch reaction vocabulary accumulate on a product quarter by quarter instead of seeing only today's cumulative total.

Seventeen documented core fields per case, arranged in three tiers. Report metadata dates the submission through receipt, transmission and FDA-receipt timestamps and sets the seriousness flags - death, hospitalization, life-threatening, disability and congenital anomaly among them. The patient block holds age at onset with its unit code, sex, and two repeating arrays: reactions coded as MedDRA preferred terms, and drugs carrying the product name exactly as the reporter typed it, the drug's role in the event, the indication it was taken for and its administration route. A provenance tier adds the primary reporter's country and the manufacturer's own case number. Every core field, with type, definition and a genuine example:

Additional fields on request. Nested structures outrun the flat view - the drug array repeats once per product named in the case and the reaction array once per reported term, so a single multi-drug case flattens into many drug-reaction pairs under one identifier.

Geo: global submissions to FDA - American cases sit alongside foreign ones, and the primary reporter's country rides on every case, so domestic signal can be separated from worldwide noise. Temporal: 2004 Q1 to the present, more than twenty years of post-market reporting; the newest window keeps collecting late-arriving cases after it closes, so treat recent months as provisional rather than final. Granularity: one record per safety report with drug and reaction arrays nested inside.

One structural caveat outweighs every coverage edge: when several drugs and several reactions share a single report, nothing in the record ties a specific drug to a specific reaction. These counts measure reporting behavior at least as much as incidence, and any analysis that forgets that will find patterns the data never claimed.

API, files, or your warehouse. Daily, weekly, or hourly.

Pick the slice, pick the cadence, pick the landing zone - the same twenty-million-case corpus arrives whichever way you take it, arrays intact. Because cases are only ever added, standing deliveries compound: yesterday's snapshot stays correct forever and the delta between deliveries is pure new intake, so longitudinal watch costs no historical re-pulls.

• Pharmacovigilance and safety teams run disproportionality and signal-detection screens over the full twenty-year span, using suspect-drug roles and outcome flags as the priors their methods expect; the modeling notes sit on the data scientists use cases page.
• Competitive intelligence and product teams watch reaction vocabulary accumulate on rival products - a new preferred term clustering on a competitor's molecule is observable weeks before it becomes a headline; playbooks live on the competitive intel use cases page.
• Compliance and regulatory affairs test whether observed reaction profiles still match what the label warns about, turning label language into something falsifiable rather than assumed.
• Retail pharmacy and category managers read the age, sex and outcome mix around the categories they stock, putting a risk texture behind planogram decisions that usually run on margin alone.
• Investors and quants treat clusters of serious-outcome reports around a molecule as early litigation, recall and withdrawal risk - a screen input, not a verdict; screens are on the investors quants use cases page.

• Data scientists and ML engineers get the cleanest nesting-to-table story in the drug catalog: flatten arrays into drug-reaction pairs keyed by the safety report identifier, deduplicate on identifier plus combination, and the corpus is model-ready.
• Developers and builders get flat JSON records that load without parsing gymnastics, with coded values that resolve through the code key below rather than bespoke dictionaries per client.
• Market researchers get reporter-entered product names and indications as evidence of how medicines are actually used, off-label uses included - usage reality no marketing deck supplies.
• Journalists and academics get citable case patterns - a specific outcome, a specific country, a specific product spelling - instead of anecdotes about drug safety.

• openFDA NDC Directory API - the identity spine; resolving reporter-entered product names to harmonized drug codes is what turns reaction profiles into panels you can roll up by labeler or ingredient.
• openFDA Drug Label API (Structured Product Labeling) - what the label warns about, so observed reaction profiles can be tested against claimed warnings.
• openFDA Drug Enforcement / Recall API - enforcement actions matched on product identity, the natural next question after a signal hardens.
• openFDA Drugs@FDA API - approval histories and regulatory actions, adding regulatory age next to report counts.
• GoodRx Drug Prices & Pharmacy Finder - the commercial layer; pairing safety signal with price movement separates perception effects from access effects.

Three reference points frame the corpus before you commit: the definition of a FAERS adverse event report, how a MedDRA-coded reaction gets its vocabulary, and the National Drug Code (NDC) that anchors the join to product identity. The full ranking sits on best drug-retail datasets, the wider catalog on the drug retail data hub, and provenance notes across the agency's catalog on the U.S. FDA openFDA source profile.