How the data works

Sources

• GUDID (Global Unique Device Identification Database) — the FDA's registry of devices that carry a Unique Device Identifier (UDI). It gives each device's brand, model, company, product code and other label details. FDA file date: Oct 13, 2026.
• MAUDE (Manufacturer and User Facility Device Experience) — reports of deaths, injuries and malfunctions involving medical devices, submitted by manufacturers, importers, hospitals, health professionals and patients. FDA file date: Oct 13, 2026.
• Medical Device Recalls — the FDA's records of device recalls and corrections, including the reason, root cause and current status. FDA file date: Oct 13, 2026.

All three are downloaded from openFDA. We check for new FDA files regularly and rebuild the site when they change.

How reports are matched to a device

Many MAUDE reports include the device identifier (UDI-DI) of the product involved. A report appears on a device page only when that identifier matches one of the device's own identifiers in GUDID. Identifiers that GUDID assigns to more than one device record are ignored, so a report is never attached to several unrelated devices. Reports that do not include an identifier — common in older reports — are not shown on any device page, which means the counts on this site are lower than the total number of reports in MAUDE.

How recalls are matched

A recall is listed under “Recalls naming this device” only when the recall record itself contains the device's identifier (we check the number's GS1 check digit to avoid matching lot or serial numbers by accident). Recalls that cover the same FDA premarket submission (510(k) or PMA number) but do not name the device are shown separately as “Recalls under the same FDA submission”, because one submission can cover many different models.

What the numbers can and cannot tell you

• MAUDE reports are not verified by the FDA. A report does not prove that a device caused or contributed to an event.
• Report counts depend on how many devices are in use, how long they have been sold, and reporting practices. They cannot be used to calculate failure rates or to rank devices against each other.
• Some manufacturers submit large numbers of reports at once (for example, summary reports), which can create sudden spikes in the monthly chart.
• The event type (death, injury, malfunction) is the category chosen by the reporter.
• openFDA notes that its data is not validated for clinical or production use. Always check the linked FDA record before relying on a detail.

Automatically generated text

The summary sentences on device pages are produced by fixed templates from the counts above. Report narratives are shown exactly as published by the FDA.

AI quick summaries

Some device pages show a labelled "Quick summary" written by an AI model (Claude, by Anthropic). The model receives only the FDA data shown on that page — counts, most reported problems, recent report narratives and recall reasons — and is instructed not to add outside information or judge whether a device is safe. Before a summary is published, a script checks that every number in it matches the data and that it contains no safety verdicts; summaries that fail are not shown. AI summaries can still simplify or miss nuance, so the detailed records on the page remain the source of truth.