The 510(k) record doesn't name the predicate

A 510(k) exists to answer one question: is this device substantially equivalent to something already on the market. The device that answer points back to — the predicate— is the single fact a competitor-tracking read of the clearance record wants most. It is also the one fact that record does not carry.

The requirement is real, and it sits one document away

21 CFR 807.92(a)(3)requires a 510(k) summary to include “an identification of the legally marketed device to which the submitter claims equivalence” — the predicate, named. 807.92(a)(6)goes further: where the subject device shares the predicate’s technological characteristics, the summary must compare them directly. FDA’s own guidance on choosing a predicate confirms the same premise from the applicant’s side — a 510(k) can name more than one, and often does. None of this is buried. It is a named, numbered requirement.

What the structured feed actually keeps

None of it survives into the queryable data. openFDA’s 510(k) endpoint — the same one FDA Radar ingests daily — returns a K-number, applicant, device name, product code, decision code and date, review panel, and the third-party-review flag. FDA Radar’s own ingestion stores exactly that set as structured facts on every clearance. Neither carries a field for the predicate’s K-number or name. The 807.92 requirement lives inside free text — a PDF, if the applicant chose the summary route rather than the 807.93statement — not inside a column either system exposes to a query.

Every 510(k) has to name its predicate. No public database lets you search by one.

The practice

Answering “what did this device cite as its predicate” means opening that device’s own 510(k) summary and reading it, K-number by K-number — not filtering a table. The summary route is itself a choice the applicant made before submitting: a device cleared under the alternative 807.93 statement has no public summary to open at all, predicate included, only FDA’s own thirty-day-on-request commitment standing in its place. A predicate-landscape read built only from what a database returns will quietly skip every clearance that took that second path, with no field anywhere flagging that it did.

Which predicate a specific device should have cited, or whether one it did cite still holds up, is a judgment for a regulatory professional working the file directly — this is a reading method for where the fact lives in the public record, never a substitute for that judgment. Per what FDA Radar monitors and how often, the clearance corpus is ingested for exactly this kind of cross-reference; the predicate name is the field it still cannot hand you.