The count is real. The rate is not.

Search MAUDE for a product code and the count sitting at the top of the results looks like a fact you can act on: this many adverse event reports, this device, this window. It invites a question it cannot answer — is that number high. FDA’s own documentation about the database says, plainly, that it cannot answer that question, and the reason is not that the count is wrong. It is that the count has no denominator.

What FDA actually says about its own database

The Manufacturer and User Facility Device Experience database exists because 21 CFR Part 803 requires manufacturers, importers, and device user facilities to report deaths, serious injuries, and certain malfunctions — a reporting duty created under section 519 of the Federal Food, Drug, and Cosmetic Act. FDA’s own page describing the database states the limitation directly: the incidence, prevalence, or cause of an event cannot be determined from the reporting system alone, because of under-reporting, inaccuracies in individual reports, no verification that the device actually caused what was reported, and — the part a raw count hides best — no information about how often the device was used. A report count with no usage figure behind it is a numerator with no denominator. It cannot become a rate no matter how it is sliced.

What a rising count actually tells you

A count that doubles quarter over quarter reads like a device getting more dangerous. It is equally consistent with the device simply being used more, a reporting habit changing at one large user facility, a single filer submitting duplicates, or FDA’s own recordkeeping catching up on a backlog. MAUDE cannot distinguish between these from the count alone, and neither can anyone reading it from outside the reporting chain. The number is real — every entry is a report someone actually filed — but the rate implied by comparing it to a rival device or to last quarter is not a fact sitting in the data. It is an inference the reader supplied.

Why this is a deferred item, not an ignored one

This is the reason FDA Radar’s own sourcing page lists MAUDE as deliberately not covered rather than quietly skipped: per-product adverse-event analytics that are honest about the denominator problem need statistical handling — exposure estimates, duplicate detection, narrative classification — that a raw ingest of the feed does not provide on its own. Shipping the count without that handling would hand a reader exactly the false rate this page is describing. Relevance over completeness cuts both ways. Sometimes it means filtering a feed down to what matters. Sometimes it means not shipping a feed at all until it can be read correctly.

The practice

Read a MAUDE count as a set of individual narratives worth reading one at a time, not as a single number worth comparing to another number. A cluster of reports describing the same failure mode, the same lot, the same use error, is worth a closer look regardless of what the count is doing in aggregate — the pattern in the text is real evidence even when the count around it is not a rate. Whether a given pattern is material enough to act on is a judgment for a regulatory or quality professional looking at the specific device and the usage data the public record does not contain. This is a description of what the count can and cannot support — regulatory intelligence and method, never advice about a specific product’s safety.