MAUDEScope

What MAUDE is, and what a count in it can and cannot support

MAUDE is FDA's database of Manufacturer and User Facility Device Experience reports: individual adverse event reports about medical devices, filed by manufacturers, importers, user facilities and the public. It is the largest public record of medical device field experience in the world, it is used by every regulatory and quality function, and it is structurally misleading in ways that are not written on it.

MAUDE has no denominator. It can tell you how many reports exist. It cannot tell you how often anything happened.

Nothing public says how many units of a device were sold, implanted or used, so a report count cannot be divided by anything. A product code with ten times the reports of another may have a hundred times the installed base, or one manufacturer with a rigorous complaint-handling process and another without one. This tool therefore computes no rate of any kind, and it will not put two product codes in an order.

What this tool measures instead

The database, rather than the devices in it. For one product code, from the public record:

• How many reports, and over what window. The all-time count and the count received in the 365 days ending on the date openFDA says the file was refreshed — a window of fixed length, rather than "the last four quarters", which is three whole quarters plus however much of a fourth happens to have been published.
• Who filed them. The share from manufacturers, user facilities, distributors and voluntary reporters. This is the underreporting signal: a product code whose reports come almost entirely from manufacturers is one where the public record is the manufacturers' account of it.
• How late they arrive, split into the interval from the event to the manufacturer becoming aware and the interval from there to FDA receipt, each reported as p50 / p90 / p95 with the excluded records counted rather than dropped.
• Which fields are actually filled in. Lot number, catalogue number, UDI, 510(k) number, device and patient problem codes, narrative text — each as a share of every matching report, with what its absence costs stated beside it.
• How dirty the identifiers are. How many distinct brand spellings the reports carry, and how many survive folding case and punctuation together.
• Which quarters stand out, against the median of the preceding eight, with any FDA device recall for the same product code initiated in the window shown beside them.

Why the gaps are shown as gaps

Three figures the specification asks for are not on the scorecard, and each one renders as a labelled absence rather than as a zero. An estimated duplicate rate needs a clustering pass over the bulk download, and its exact half needs the event_key field, which openFDA serves empty on every one of the 26,136,889 records in the endpoint. Summary-reporting exposure needs two FDA lists that are not in openFDA and whose source addresses this build has not verified. Revision behaviour — how often an already-published report is silently changed — needs two weekly snapshots to compare, and a query made in a browser has one.

A zero would be read as a finding. "Not measured, and here is the input that is missing" is the true statement, so that is what the page prints. The method page carries every definition in full, including the ones this build does not compute.

About openFDA lag

openFDA mirrors FDA source systems on its own schedule and every response states when it was last refreshed. That date — not today's date — is what this page stamps on every figure, in every source line, and in the disclosure block. The scorecard also reports the gap between the newest report in the file and that refresh date, so the lag is a measured number on the page rather than a caveat in a footnote. Even the thirty-day window on the tail of recent reports is measured back from the file's own date: anchoring it to the reader's clock would silently shorten it by however far behind openFDA happens to be.

Questions we get

What is MAUDE?

MAUDE is the FDA database of Manufacturer and User Facility Device Experience reports — individual adverse event reports about medical devices, submitted by manufacturers, importers, user facilities and members of the public. It holds over 26 million records. FDA does not verify the reports, and publishes them as received.

Why can MAUDE not be used to compare one device against another?

Because there is no denominator. MAUDE records reports, not events, and no public dataset says how many of a device were sold, implanted or used. A product code with ten times the reports may have a hundred times the installed base, or one manufacturer with a diligent complaint-handling process. Every figure on this page is a count or a share of reports, and this tool computes no rate of any kind.

What does this tool measure, then?

The database, not the devices. For one product code it reports how many reports exist, who filed them, how long they took to reach FDA, which fields on them are actually filled in, how many distinct brand spellings the reports use, and which quarters carry more reports than the preceding two years would suggest. Those are properties of the public record, and they decide whether the record can answer a question you want to ask of it.

Does it estimate how many reports are duplicates?

No, and the reason is worth stating. The published method groups reports by event key, and openFDA serves that field empty on every record in the endpoint — all 26,136,889 of them. Estimating duplicate events from the narratives needs the bulk download rather than the API. What this build reports instead is report-number linkage, which is exact and is a smaller claim. The method page states the difference.

What does a marked quarter mean?

That the quarter carried more reports than the median of the preceding eight quarters by more than three median absolute deviations, and held at least twenty reports. It is a statement about a series against its own recent history. Where an FDA device recall for the same product code was initiated in that quarter or the two before it, the recall is shown beside the quarter. Where none was, the quarter is shown as unannotated — which is not the same as having no cause.

How current is the data?

Every openFDA response carries a last_updated date, and that is the date printed on every figure and in the disclosure block. openFDA lags FDA source systems, so that date is normally behind today, and the page also reports the gap between the newest report in the file and that date. Even the thirty-day window on the live tail is measured back from the file's own date rather than from the clock on your laptop.

Do I need an account or an API key?

No. The page queries the public openFDA endpoints straight from your browser with no key and no sign-in, and stores nothing about you. The exact request URLs are printed under every scorecard so you can run them yourself and get the same figures.