GLP covers the safety study, not the device

A device sponsor reading 21 CFR Part 58 for the first time usually reads it as a lab-conduct standard — the rule that decides whether a bench test was run properly, full stop. The regulation’s own scope section draws a narrower line than that, and FDA’s attempt to walk device companies through exactly where the line falls has been sitting in draft for more than a decade.

A safety-study rule, not a testing rule

21 CFR 58.3 defines the two terms that do the actual work. A nonclinical laboratory study is an in vivo or in vitro experiment in which a test article is studied prospectively, under laboratory conditions, to determine its safety — and the same definition excludes any study using human subjects, any clinical study, and any field trial in animals. A test article expressly includes a medical device for human use. Put the two together and GLP attaches to exactly one kind of work: a safety study, on a device, that is neither a bench characterization nor a human trial.

What the definition carves out on its own

The same definition excludes “basic exploratory studies” — work done to find out whether a device has any potential utility at all, or to pin down its physical or chemical characteristics, before anyone has designed a safety study around it. That carve-out exists in the regulation’s own text, not in a guidance document, and it means a sponsor can run real bench work, on a real device, and never touch Part 58 — not because the work was sloppy, but because it was never the kind of study the regulation was written to reach.

The guidance that explains the rest, still in draft

In August 2013, CDRH published a draft Q&A, The Applicability of Good Laboratory Practice in Premarket Device Submissions, meant to walk sponsors through exactly which studies behind a 510(k), PMA, De Novo, or HDE actually need a GLP compliance statement — and it told sponsors plainly that most in vitro diagnostic studies, because they run on human subjects or human specimens, fall outside Part 58 for the same reason a clinical study does. More than twelve years later, no Federal Register notice has finalized it. The agency’s own walk-through of where the regulation’s boundary actually falls remains exactly what it was issued as: a draft, not a rule a sponsor can cite as settled.

The regulation decided what counts as a safety study in 1978. FDA is still drafting the sentence that tells you whether yours is one.

The practice

Deciding whether a specific bench or animal study behind a submission needed to run under Part 58 is not a question the draft guidance can answer for you — it never became final, and a sponsor citing it as binding is citing a document FDA itself has not finished. The regulation’s own text is what a compliance statement actually has to answer to: whether the study was prospective, whether it was run to establish safety rather than just feasibility, and whether it ever touched a human subject or specimen. Getting that determination wrong on a specific study, in either direction, is a call for the regulatory and nonclinical professionals who designed the protocol — this is regulatory intelligence and method, about where the regulation’s own line sits, never advice on a specific study’s own compliance.