The code changed. The failure didn't.

Every Medical Device Report carries two coded fields alongside the narrative a reviewer actually reads: a device problem code and a patient problem code, chosen from an FDA-controlled list and required on every report, manufacturer or user-facility, under Form FDA 3500A. They look like the most stable, machine-readable fact in the record — a controlled vocabulary, trending-ready by design. In February 2021, FDA changed the vocabulary. The record does not announce the change at the point a trend line crosses it.

What the code fields are for

A Medical Device Report has to code at least one device problem and one patient problem per event, at the most specific level the list offers, under 21 CFR 803.52 for manufacturers and 21 CFR 803.32 for user facilities. The narrative is free text, read one report at a time. The codes exist precisely so a reader does not have to read one report at a time — tally the same device problem code across a window and the tally is supposed to be the trend.

The vocabulary itself changed

FDA’s own adverse-event-coding program retired its legacy device problem, patient problem, and device component code lists and replaced them with terms harmonized to the International Medical Device Regulators Forum’s adverse event terminology. The cutover was not quiet: FDA delayed the original production date to give filers time to adjust, and set February 28, 2021 as the date the legacy codes were retired and the IMDRF-harmonized codes became the only ones a report could use. FDA describes the result as full harmonization — every legacy code maps to one corresponding IMDRF code — which is a statement about a crosswalk existing, not a statement that the raw code string in a report filed in January 2021 matches the raw code string in a report filed in March describing the same failure.

A code that disappears at the cutover did not stop occurring. It stopped being spelled that way.

What a raw tally does with that boundary

Pull a product code’s device problem codes from openFDA’s device event endpoint across a window that spans early 2021, and the two code sets sit in the same field, unlabeled as to which vocabulary generated which row. A failure mode that was common before the cutover and common after it, under its mapped IMDRF term, reads in a raw tally as one code falling to zero and an unrelated code appearing from nothing, right on schedule. Nothing about the device changed. The list it was described against did.

The practice

A trend built on device or patient problem codes is worth running on one side of February 28, 2021 at a time, or through FDA’s published crosswalk between the legacy and IMDRF terms before treating a code’s disappearance or arrival as a fact about the device. The same caution applies to anyone else’s trend handed to you without that boundary marked — a quarter-over-quarter comparison of a competitor’s or a predicate’s problem-code mix is only a real signal once both sides of it are speaking the same coded language. Whether a specific shift in a specific device’s coded history reflects an actual change in failure pattern is a determination for whoever owns that risk file, reading the narratives behind the codes — this describes a documented change in how FDA codes an event, regulatory intelligence and method, never a reading of what any device’s adverse event history means.