Do device recall patterns differ by manufacturer size?

Every device founder has heard some version of the same claim. Small manufacturers are riskier, the argument goes, because they lack the quality systems, the regulatory staff and the supplier buying power of a large one. Buyers repeat it. Distributors repeat it. Acquirers price it in.

The public recall record lets that claim be tested, at least partly. FDA publishes every device recall with the recalling firm, a root-cause category and a hazard class. The 510(k) file publishes every clearance with the applicant’s name. Joined on firm name, the two files show how recall patterns change with portfolio size, and where they do not change at all.

Does a bigger manufacturer recall more often?

Yes, and steeply. Among firms holding one 510(k) clearance in 2016-2025, 2.8% had a recall event initiated in 2020-2025. Among firms with 26 to 100 clearances it was 67.5%. Among the 13 firms with more than 100, it was 92.3%, a 33x difference in the chance of any recall at all.

Part of that is simple exposure. A firm with a hundred cleared products has a hundred chances to recall one. So the better comparison scales recalls by portfolio size. Recall events per 100 clearances rise from 8.2 in the single-clearance tier to 33.4 in the largest, a 4.1x difference. Even per product cleared, the largest filers recall more often.

The concentration is visible from the other direction too. The two largest portfolio tiers hold 133 of the 8984 applicant names in the 510(k) file, 1.5%. They account for 1847 of the 5487 recall events in the window, 33.7%. Each recalling firm in the 101+ tier averaged 52.2 events over six years. Each recalling firm in the single-clearance tier averaged 2.9.

Are larger manufacturers’ recalls more serious?

No. The share of recall events classified Class 1 (the most serious of FDA’s three hazard categories) ranges from 5.1% to 12.5% across the six size groups, with no direction to it. The single-clearance tier has the highest share. The 26-to-100 tier has the lowest. The largest filers sit in between at 8.9%.

Class 2 dominates every tier, between 82.4% and 93.3% of events. Across the whole window, 4839 of 5487 events were Class 2, 88.2%. Class 1 made up 503 events, 9.2%, and Class 3 145 events, 2.6%. That mix barely moves when the recalling firm changes size.

One small shift runs the other way from the folk wisdom. Class 3, the least serious category, is 5.1% of events in the single-clearance tier and 1.1% in the largest. Small filers’ recall mix leans slightly toward the mildest class. The counts behind the smallest tiers are modest, so treat this as a tendency in the data and nothing firmer.

FDA writes the three recall classes in Roman numerals; this piece uses digits, with 1 the most serious and 3 the least. They are not the same thing as a device’s own risk class. The recall class is assigned per product by FDA, reflecting the relative degree of hazard. It says nothing about the firm as a whole. It is still the only standardised severity measure in the public file, and on that measure size does not separate firms.

Why would recall frequency rise with portfolio size?

The data shows the pattern without the reason. Three explanations fit it, and they are not mutually exclusive. Larger firms sell more units into more settings, so defects surface faster. They carry older products cleared before the window. And mature quality systems find and report more problems themselves.

The installed-base effect is the most mechanical. A device sold in the hundreds of thousands meets more edge cases than one sold in the hundreds. The file contains no unit volumes, so the per-clearance rate above still cannot adjust for it.

The legacy-portfolio effect is a denominator problem. A large firm recalls products cleared in 2008 as readily as products cleared in 2023. The size measure here counts only the last ten years of clearances, and only 510(k) clearances. Premarket approvals and exempt devices recalled by the same firm sit in the numerator without a matching entry in the denominator.

The detection effect is the least comfortable for the folk wisdom. A recall is, in FDA’s words, an action taken to address a problem with a device. A firm with a complaint-handling team, field service engineers and a post-market surveillance function will find more problems to act on. A recall count partly measures how hard a company looks.

The public file cannot separate those three. Any article that claims it can is asserting more than the data holds.

Do the root causes differ by size?

Somewhat, at the edges. Manufacturing and change-control causes lead every tier, at 36.1% to 44.2% of events. Device design, as FDA codes it, falls steadily with size: 14.0% of single-clearance firms’ events, 8.5% of the largest firms’. Process control peaks in the largest tier at 20.9%.

Read together, that is a plausible story about where each kind of company is exposed. A small firm with one or two products carries its design risk in a handful of devices, so a design issue is a larger share of its recall history. A large firm runs many production lines and supplier changes, so its recall mix leans toward process control and change control.

Software design sits at 5.6% of single-clearance events and between 8.6% and 11.9% in every larger tier. Software-heavy products tend to come from firms with more than one clearance, which likely explains most of that gap.

The largest tier also bundles more products into each event: 4.04 recalled product numbers per event, against 2.57 to 2.87 in the smaller tiers. A large firm’s recall tends to span a product family rather than a single model.

Root cause needs its own caution. FDA records the category and can revise it until the recall is closed. Across all events, 25.1% carry a cause that is not yet determined, mostly “under investigation by firm”. That share peaks at 33.1% in the 26-to-100 tier, which may reflect more recent, still-open recalls there.

What about recalling firms that never appear in the 510(k) file?

They are a large group, and a mixed one. After normalisation there are 1332 recalling firm names; 761 matched a 510(k) applicant name and 571 did not. Those unmatched firms produced 1163 events. Their Class 1 share is 11.1%, inside the same band as everyone else. Their mix differs in one place: regulatory-status causes.

Regulatory-status causes, chiefly FDA’s “No Marketing Application” category, account for 7.9% of unmatched firms’ events. That category means a product was marketed without the clearance it needed. In the largest tier the share is 1.1%. That gap is partly built into the method: a firm selling without a clearance is, almost by definition, a firm without clearances to match.

The unmatched group also holds firms that are large by any other measure. Some sell mostly premarket-approved or exempt products. Some recall under a sales or distribution entity whose name differs from the one on their 510(k) filings. Some are importers and relabelers. The name join cannot tell them apart, so this group is reported separately rather than folded into a tier.

Does the answer change if size is measured differently?

No. Grouping firms by their own recall count instead (one event, 2-5, 6-20, 21 or more) gives a Class 1 share between 7.7% and 10.7% in every group. The firms with the most recalls have the same severity mix as firms that recalled once. Two independent size proxies agree that frequency varies and severity does not.

This second cut uses no name join at all, so it is immune to the matching problems above. It is also circular in one respect: firms are grouped by the very thing being counted. That is why it serves only as a check on the primary measure.

How was this calculated?

Three openFDA pulls, retrieved 2026-09-30. Device recalls initiated 2020-2025, one row per recalled product: 15865 rows. Device enforcement reports for the same years, joined on recall number to attach FDA’s class; every recalled product matched. And every 510(k) applicant’s clearance count for 2016-2025.

The unit is the recall event, FDA’s event number. The window holds 5487 events. Each event takes the most severe class among its products, and the most common root cause among them. Counting products instead would let one event covering dozens of catalogue numbers outweigh dozens of single-product events.

Size is the recalling firm’s number of 510(k) clearances over the prior decade, joined on a normalised firm name. Normalisation removes case, punctuation, corporate-form words and trailing divisions or cities, so spellings of one legal entity merge. It never merges two different entities. 4324 of the events, 78.8%, came from a firm whose name matched.

The tiers (one clearance, 2-5, 6-25, 26-100, more than 100) were fixed before the class mix was examined. The share of firms with any recall uses every 510(k) applicant name in the tier as its denominator, including the many that recalled nothing. The full normalisation rule is in the dataset module, and all three result sets are downloadable above.

What can this data not tell you?

It cannot measure company size. A 510(k) count measures regulatory activity only. A large firm selling a few high-volume products can hold fewer clearances than a small firm filing many variants. The measure also splits corporate groups that file through several subsidiaries into several smaller entities.

It cannot measure harm. The recall class describes FDA’s assessment of the hazard a problem could present. It does not record whether anyone was hurt, how many units were affected in use, or how quickly the firm acted.

It cannot adjust for exposure. There are no unit volumes, installed-base figures or years-on-market in either file. A per-clearance rate is the best available scaling, and it is a weak one.

It sees only recalls. Market withdrawals, stock recoveries and safety communications that fall short of a recall are outside the file. So are problems a firm never detected or reported.

It describes the market as a whole. A tier average says nothing reliable about a particular company. A single firm’s recall history needs to be read record by record, with the reason for recall in front of you.

How do you run this for your own category?

The same endpoints work at product-code level. Count recall events and class mix for your code, then count the applicants who hold clearances in it. That gives the recall base rate a buyer or distributor is implicitly comparing you against, before anyone raises it in a meeting.

# recalled products in one product code, 2020-2025, by root cause
https://api.fda.gov/device/recall.json?search=product_code:IRP+AND+event_date_initiated:[2020-01-01+TO+2025-12-31]&count=root_cause_description.exact

# class mix for recalls initiated in the same window
https://api.fda.gov/device/enforcement.json?search=recall_initiation_date:[20200101+TO+20251231]&count=classification.exact

# 510(k) holders in the same product code
https://api.fda.gov/device/510k.json?search=product_code:IRP+AND+decision_date:[2016-01-01+TO+2025-12-31]&count=applicant.exact&limit=1000

The enforcement records carry no product-code field, so a code-level class mix means pulling the recall records first and joining on recall number, exactly as this analysis does. For a single category that is a few hundred rows at most, and a spreadsheet handles it.

What should a founder take from this?

The public record does not support “smaller means riskier.” On FDA’s own severity scale, small and large filers’ recalls look alike. What differs is frequency, and frequency rises with portfolio size. A founder facing the small-company objection has a checkable answer to it.

That answer has to be used carefully. The regulated-market rule still applies: do not market against a named competitor’s recall history, and do not imply a product is safer because its maker is small. The aggregate finding supports a narrower point: size alone is not evidence of risk. That is the claim to make.

The more useful move is structural. Buyers, distributors and acquirers judge quality systems through documents: complaint handling, supplier controls, change control, corrective-action records. The root-cause data says change control and process control are where recalls concentrate at every size. A founder who can show those systems working has something more persuasive than any size argument.

There is a planning point as well. A single-clearance firm with a recall carries it in almost every conversation, because it is most of the firm’s record. A large firm with dozens of recalls carries none of them individually. Communication planning for a recall (who is told, in what order, with what documentation) matters more to a small firm. It belongs in the launch plan rather than the crisis plan.

Carmine’s view is that regulatory data belongs in positioning work as well as regulatory affairs. The panel-level clearance analysis shows who is entering a category, and the clinical evidence ledger shows who is producing the evidence about it. Recall patterns complete the picture of how a category is judged. More on building that authority sits in the category authority hub, and on the operating systems behind it in the marketing operations hub.