FDA's device names read like a card catalog

Type “continuous flush catheter” into a search box wired straight to FDA’s own device classification data and it returns nothing. Not because the device isn’t there — it is, filed under product code KRA — but because FDA never wrote the name that way. The record reads Catheter, Continuous Flush. Same device, same words, wrong order, and a naive phrase match cannot tell the difference between absent and misfiled.

A convention older than the API

FDA’s Product Classification Database has been assigning three-letter product codes to generic device categories since 1976, and it names each one the way a card catalog names a book: noun first, modifiers trailing, comma-separated. A device that stimulates a peripheral nerve from an implant is filed as Stimulator, Peripheral Nerve, Implanted. One that evokes a response electrically is Stimulator, Electrical, Evoked Response — product code GWF. The convention groups every catheter under “Catheter,” every stimulator under “Stimulator,” the way a library shelves every book about Rome under R regardless of which historian wrote it. It is a filing system built for browsing a list, not for typing a query, and openFDA republishes it verbatim — the API does not translate the name into the order a person would say it out loud.

What breaks when you search it like English

A phrase search treats word order as part of the match. Ask a classification lookup for “peripheral nerve stimulator” as a quoted phrase and the literal substring never occurs in the record, so the query returns zero results — indistinguishable, on the screen, from “FDA has no such device category.” That is the failure mode worth naming: a well-formed, correctly spelled search that fails silently, not because of a typo, but because English word order and FDA’s filing order disagree, and nothing about an empty result tells you which one is wrong.

An empty result from this database answers a different question than the one you asked. It rarely means “FDA has no such device.” It usually means “not in the order you tried.”

The practice: search the words, not the phrase

The fix is not a smarter guess at FDA’s order — it is refusing to depend on order at all. Split a free-text query into individual words, require each one to appear somewhere in the record, and drop the assumption that they appear adjacent or in sequence. “Peripheral,” “nerve,” and “stimulator” each match Stimulator, Peripheral Nerve, Implanted individually; none of them care what position FDA put them in. FDA Radar’s own onboarding lookup — the “what do you make” box that turns a free-text description into a candidate product code — works exactly this way, and it does so because an early version that echoed the phrase straight back to the API returned nothing for ordinary device descriptions during testing, not because a specification said to build it that way.

Why it matters past onboarding

The same trap catches anyone building their own predicate search or product-code lookup by hand: querying the classification database for a device type by typing it the way you would say it aloud. If the search comes back empty, that is not evidence the category is missing from FDA’s system — it may only be evidence that the query assumed an order FDA never used. The reliable move is to search by product code once you have one, and to treat a free-text miss as a prompt to try individual words before concluding FDA has nothing on file.

That is regulatory intelligence and method — how one FDA database is actually organized, and how to query it without mistaking its filing convention for an absence of data — never a determination about what any device’s classification should be. Per what we monitor and how often, that same classification database is the reference layer behind FDA Radar’s own onboarding lookup, confirmed by the user in one click rather than saved on a guess.