Methodology

FDA now refers to FAERS as the FDA Adverse Event Monitoring System (AEMS); openFDA still publishes the data under the FAERS name, and this site uses that name. Every figure on this site is a count of reports in the openFDA bulk exports: the FAERS release of 28 Sep 2026 (20,646,523 reports) and the MAUDE release of 28 Sep 2026 (26,136,888 reports). Nothing is modelled or estimated. This page explains how a report becomes a count.

What one report is

A FAERS report is one submission to FDA about one patient: the drugs involved and their role, the reactions coded with MedDRA preferred terms, seriousness and outcome flags, the patient's age and sex, and the reporter's qualification. A MAUDE report is one submission about one event involving a device: the brand, generic name, manufacturer and product code, the event type (death, injury, malfunction, other) and coded device and patient problems. Most reports in both systems are filed by manufacturers, who are required to forward what they receive.

How a drug is named

Reporters write the product name themselves, as a brand ("Lipitor"), a generic ("atorvastatin calcium") or something in between. openFDA attaches to each drug entry the brand, generic and pharmacologic class names from the FDA label when its own harmonisation matches the reported name exactly. This site keys each entry on the reported name, resolved as follows: if the name matches a brand on the label, the drug page is that brand; if it matches a generic or substance name, the page is that generic; if openFDA attached exactly one generic name to an unmatched entry, that generic is used; otherwise the name is looked up in the National Drug Code Directory after common salt abbreviations and dosage-form words are removed; and if nothing matches, the name is kept as reported. A brand and its generic are therefore separate pages, because reporters name them separately; a brand page shows its generic name and, where a class is known, the same-class table lists both.

Suspect, concomitant and interacting

FDA records each drug in a report as suspect, concomitant or interacting. A drug page counts reports in which the drug is a suspect or interacting product. Reports that list it only as a concomitant medication are counted separately and shown in the page's opening paragraph, but they are not in any figure, share or trend. Listing a drug as suspect is the reporter's judgement; FDA states that "a causal relationship cannot be established between product and reactions listed in a report."

Classes

A pharmacologic class is FDA's Established Pharmacologic Class (EPC) from the drug label. A drug is placed in the class its reports carry most often, or in the class the NDC Directory gives its name where reports carry none. A class page is built when at least 2 drugs with pages fall in it. The class median for a reaction is the middle value of that reaction's share across the member drugs; it is the reference bar on each member's page.

Counting rules

• A report that lists three suspect drugs counts once for each of them, and once for the class of each.
• A report that records five reaction terms counts once for each term, so shares of terms add to more than 100%. Terms are pooled across capitalisation and spacing variants.
• "Marked serious" is the report's own seriousness flag. Outcome flags (death, life-threatening, hospitalisation, disability, congenital anomaly, other serious) are as the reporter set them; a report can carry several; none is verified by FDA.
• Age bands convert the reported age unit (years, months, weeks, days, hours, decades) to years. Sex and reporter qualification are as coded. "United States" means the country of the event or of the reporter is recorded as US.
• Months are the month FDA first received the report. 0 records with no usable receive date are left out.
• Co-reported drugs are other products, in any role, listed in reports where the page's drug is a suspect product. Only the 2,500 most-reported names are tracked, and pairs with fewer than 5 reports are not shown.
• Device brand names are pooled after punctuation and capitalisation are normalised; one product can still appear under several spellings. Manufacturer names are pooled after company suffixes are removed. Product codes come from the report and are named from FDA's device classification.
• Device problem and patient problem codes are FDA's own code lists; placeholder codes ("no apparent adverse event", "insufficient information") are excluded.

Duplicates and follow-up reports

FDA notes that when a manufacturer learns more about an event, "subsequent updates are submitted in separate reports", and that "there are many instances of duplicative reports and some reports do not contain all the necessary information." One event can therefore appear as an initial report, one or more follow-ups, and a voluntary report filed by someone else. This site counts records as openFDA publishes them, without de-duplication, because the public files do not carry enough to link them reliably; every count on the site is a count of records, not of events or patients.

No people

No narrative, dosage text, lot number, address, date of birth or contact name is read from any record. Product, brand and manufacturer names are checked against patterns that read as a person's name (a surname-comma-forename form, professional titles, a plain two-to-four-word name with no organisational marker for manufacturers) and such names are never published. Reporter identities are not in the public files.

Thresholds

A page is built and indexed only when there is enough to say: drug names with 500 or more suspect-product reports (names with 200 to 499 exist for links but are marked noindex), classes with 2 or more drug pages, reaction terms with 5,000 or more reports, device brands with 300 or more (150 to 299 noindex), product codes with 300 or more, manufacturers with 1,000 or more, months with 300 or more drug reports.

Comparisons, shares and what they mean

Each page sets its entity beside a wider pool: a drug beside its class and the whole database; a device beside its product code; a class, reaction or manufacturer beside the whole database. Shares are shares of reports. They are not rates of occurrence, because the number of people taking a drug or using a device is not in the data, and no population or sales figures are used. FDA states that the reports "cannot be used to estimate the incidence of these events" and that "common products may have a higher number of adverse events due to the higher total number of people using the product." Tables that list several products are ordered by report count for navigation and carry that statement.

Trends

The trend compares the latest 12 months of receipt with the 12 months before. FDA releases FAERS quarterly and MAUDE monthly, and reports arrive months after events, so the newest months are incomplete and grow in later releases; a fall in the latest period is expected. FDA attributes much of the long-run rise in report volume to improved reporting.

Plain-language glosses

MedDRA terms and FDA problem codes are shown exactly as coded, with a short gloss from a fixed dictionary maintained by this site ("Dyspnoea: shortness of breath"). The gloss explains the word and adds nothing to the report. Terms without an entry are shown without one.

Text on the pages

Summaries and answers are assembled from fixed sentence templates filled with the counts. Nothing is written by a language model, no sentence attributes an event to a product, and a fixed list of phrasings is checked against every built page.

Build

The FAERS export is about 114 GB compressed across some 1,770 files and the MAUDE export about 18 GB across 370; together they hold roughly 500 GB of JSON. Each file is streamed once through a reader that reduces it to counts, and each year's counts are saved as a compact snapshot. Long-tail cells (a reaction recorded once for a drug in a year) are dropped from the snapshots to keep them small, so the least frequent terms on a page can be slightly undercounted. The site is rebuilt daily from the snapshots; a year is re-read only when FDA's file listing changes for it, and the newest files not yet in a snapshot are read live within a time budget.