Process control is the most common named cause of a device recall

FDA's device recall dataset carries one field the rest of the record does not explain. Root cause is the recalling firm's own answer, picked from a fixed list and recorded when the event is created. Read across the 944 recall events initiated in 2025, the answer firms gave most often was not the design of the device. It was the process that built it.

Process control was named on 166 of those events. Device design was named on 101. Software design was named on 91. The gap is not a feature of 2025: process control has been ahead of device design in every year since 2021, by 147 to 109 in 2021, 110 to 87 in 2022, 130 to 104 in 2023, and 156 to 122 in 2024.

The largest single answer was not a cause at all. On 227 events, close to a quarter of the year, the firm had told FDA that its investigation was still open.

What the field is, and what it is not

Root cause is the root_cause_description field in openFDA's device recall endpoint. It is the firm's answer and not FDA's finding. It is not the recall classification, which grades the health risk. It is not an inspection observation, and it is not an enforcement action. A firm can update the record later, so a share read today is a reading of one export and not a closed account. These figures come from the bulk export dated 2026-09-04, the same export every data page on this site carries a “Data as of” line for.

One code, three different failures

Tyber Medical began a recall on 2025-12-19 of bone fixation plates cleared under K232652, in product code HRS. The stated reason is one sentence: “The supplier manufactured anatomical left plates with an incorrect thread orientation.” The event covers two records, Z-1131-2026 and Z-1132-2026, for 100 units and 64 units. Nothing about the plate's design changed. A supplier cut a thread the wrong way round.

Beaver-Visitec International began a recall on 2025-12-05, recall number Z-1008-2026, of five laser and endoscopy systems cleared under K240615. Two ferrites that the design called for on the USB cable between the nurse monitor and the system controller were not fitted. The firm's own record calls the result a non-conformance to the design specifications. The design specified the parts. The build left them out.

Bard Peripheral Vascular began a recall on 2025-12-12, recall number Z-1367-2026, of 14,935 catheter kits, “due to damage to outer tray that can potentially compromise the sterile barrier”. The device inside was not at issue. The tray around it was.

A part made wrong by a supplier, a component omitted at assembly, a sterile barrier damaged in packaging: one code covers all three, and the three sit in product codes HRS, HQF and LJS. That is the field's use and its limit at once. It tells a reader where in the chain to look and nothing about what to do when they get there.

What these numbers do not say

Four things this count cannot support:

• It is a count of events, not of devices or of units. One event can carry a single record or a hundred and sixteen. The 2,837 recall records initiated in 2025 resolve to 944 events, and counting records instead of events moves the totals, to 514 for process control against 433 for device design, without moving the answer.
• It is not a rate. FDA publishes no denominator of devices shipped, so nothing here says whether a recall is more or less likely than it was.
• It says nothing about which firms are worse. A firm that finds its own defects and reports them completely produces more records than one that does not.
• It is not a warning letter count, an inspection finding or a supplier performance figure. MDM holds none of those and does not print them.

Method

The figures are counts of distinct recall event numbers in FDA's device recall dataset, grouped on the date the recall was initiated rather than the date FDA posted it. The two differ by weeks: Z-1132-2026 was initiated on 2025-12-19 and posted on 2026-01-20. Every event in the window carries a single root cause, so counting by event is well defined. None of the 944 carries two.

The window stops at 2025 because 2026 is still filling. The 2026-09-04 export holds 126 records initiated in July 2026 and 18 in August, against a monthly run rate above 200 for the two years before, so a 2026 share would be read off an incomplete file. The dataset is documented at https://open.fda.gov/apis/device/recall/ and every record named above links to its own page here.

MDM's continuing coverage of the recall record is collected under Recalls.