Recall event IDs by initiation label

openFDA records with report dates in 2025 · deduplicated by event_id

Firm initiated948
FDA mandated23
Missing1
Event IDs are a closer denominator than report rows, but do not represent affected units or the full recall lifecycle.

After grouping FDA device enforcement report records by event ID, 948 of 972 event IDs with report dates in 2025 were labeled “Voluntary: Firm initiated.” Twenty-three were labeled FDA mandated, and one lacked an initiation label.

The firm-initiated share was 97.5%. It reflects the FDA recall process, in which firms usually initiate recalls and the agency oversees their adequacy. “Voluntary” does not mean the action lacked regulatory involvement or that the underlying issue was minor.

MedDataUSA's published record-level article established the denominator problem: 2,908 enforcement rows mapped to 972 event IDs. This follow-up preserves that result but asks a different question—how those deduplicated event IDs were labeled by initiation type.

Quick read

  • The preserved openFDA extract contained 2,908 report records but 972 unique event IDs.
  • Initiation and classification labels were internally consistent across all records sharing an event ID.
  • The median event ID had one report record; the largest had 116, showing why raw rows are a poor event denominator.
  • At the event-ID level, 820 were Class II, 135 Class I, and 17 Class III.

Most event IDs were firm initiated

FDA says companies initiate most recalls voluntarily, sometimes after discovering a problem and sometimes after the agency raises concerns. FDA can request a recall and has mandatory-recall authority in defined circumstances.

The openFDA field captures the initiation label attached to an enforcement record. Grouping by event ID reduces repeated counting when one action covers multiple products, models, lots, or recall numbers. In this snapshot, 948 event IDs were firm initiated and 23 FDA mandated.

That difference should not be framed as company self-policing versus government action. FDA may inspect, investigate, request, classify, monitor, and publicize a firm-initiated recall. The field identifies how the recall was initiated, not every step that led to it.

Event-level classes

Class II accounted for 820 event IDs, or 84.4%. Class I accounted for 135 and Class III for 17. Each event ID had one consistent class across its associated records in the preserved extract.

Class I means there is a reasonable probability that use of or exposure to the violative product will cause serious adverse health consequences or death. Class II covers situations in which consequences may be temporary or medically reversible, or the probability of serious consequences is remote. Class III applies when adverse health consequences are not likely.

Class is a health-hazard category, not a count of affected units or a measure of a company's overall product quality.

One event can fan out across records

Six hundred three event IDs had a single report record. Three hundred sixty-nine had two or more, and 54 had at least ten. The maximum was 116 records for one event ID.

That spread explains why the record-level firm-initiated count, 2,869, should not be compared directly with the 23 event-level FDA-mandated count. A single event's reporting structure can multiply rows.

What the analysis cannot establish

Event IDs are a better denominator for this question than report rows, but they are not a perfect count of every real-world corrective action. Events can evolve, records can be updated, and the report date is not necessarily the initiation date.

The extract does not measure affected units, patient harm, time to detection, remediation speed, installed base, market share, or current recall status. It should not be used as a live alert feed or company ranking.

Methodology and limitations

MedDataUSA reused the checksummed openFDA device enforcement snapshot containing every returned record with a report date from January 1 through December 31, 2025. We grouped records by nonmissing event_id, verified that classification and initiation labels were consistent within each group, and counted each event ID once.

The analysis is anchored to report date and the preserved endpoint vintage, not recall initiation date. openFDA can revise data and warns that its endpoint should not replace current FDA recall communications.

Sources