How Are Medical Devices Approved, and Why Most Are Not

A hospital buys a new infusion pump, an advert says the device is FDA approved, and the phrase does a great deal of work it has not earned. Understanding how are medical devices approved means learning that most of them are not approved at all, in the sense the word normally carries.

They are cleared. Clearance means the manufacturer showed the regulator that the device is substantially equivalent to something already on the market, not that anyone tested whether it works. The distinction is technical, it is stated plainly in the regulator’s own material, and almost nobody outside the industry knows it exists.

Three classes, three routes

Devices are sorted into three classes by risk, and the class decides how much scrutiny applies before anything can be sold. The classification is set out in regulation for each device type rather than argued case by case, so a manufacturer usually knows its route before it starts.

Only the third route is approval in the ordinary sense. Premarket approval requires the manufacturer to show, with its own evidence, that the device is safe and effective for its intended use. The second route asks for something quite different, and it is the route the vast majority of devices take.

What substantial equivalence actually requires

A 510(k) submission is named after the section of the law that created it, and what it asks for is comparison rather than performance. There is no form to fill in; the contents are specified in the regulation itself. The submitter picks a legally marketed device, sets its own product beside it, and argues the two are alike.

The regulator’s own description is that a 510(k) demonstrates the device is as safe and effective, that is, substantially equivalent, to a legally marketed device. The comparison device is called the predicate, and choosing a good one is the central skill in the whole exercise.

Substantial equivalence has two limbs. The new device must have the same intended use as the predicate, and the same technological characteristics — or, where those characteristics differ, the submitter has to show that the differences raise no new questions of safety and effectiveness, which is a considerably lower bar than showing that the device does what the label on the box says it does.

The predicate has to be legally marketed, which means it cannot be a device that is itself in violation of the statute.

Notice what is absent from that test. There is no general requirement to demonstrate that the device benefits a patient, because the benefit is inherited from the predicate rather than established afresh. The agency can ask for clinical data where the differences warrant it, but asking is the exception rather than the default.

Cleared is not approved

The vocabulary here is precise, and the precision is routinely lost in translation. When a 510(k) succeeds, the regulator issues an order finding the device substantially equivalent and stating that it can be marketed, and that order clears the device for commercial distribution.

Approval is the word reserved for the premarket approval pathway, where the agency has reviewed evidence that the device itself works. That pathway is slow, expensive and comparatively rare, which is exactly why manufacturers avoid it when a predicate is available.

So “FDA cleared” and “FDA approved” describe two genuinely different levels of scrutiny, and marketing copy tends to blur them into one reassuring phrase that carries the weight of the stronger term.

The structure is familiar from other consumer markings. A symbol or phrase that reads as an official endorsement, while the actual meaning is narrower and technical, is the same problem as the recycling triangle described in what happens to recycling.

The review is lighter than most people assume

Three details from the regulator’s published process are worth stating together, because their combination surprises people.

• The equivalence determination is usually made within 90 days
• The agency does not typically perform facility inspections before clearance
• The manufacturer may market the device immediately once clearance is granted

A quality system inspection can follow at any point afterwards, and the manufacturer is expected to be ready for one. It is a post-market check rather than a gate.

None of this is hidden. It is published on the agency’s own guidance pages, in plain language, and it simply does not travel into public understanding — partly because nobody reads regulatory guidance, and partly because every commercial incentive runs the other way.

Predicate chains and where they lead

Here is the structural consequence, and it deserves more attention than it gets. A device cleared today is equivalent to a predicate; that predicate was itself cleared by equivalence to an earlier device, which was cleared against something earlier still.

Each step is small and defensible on its own terms. Across a long chain, the device on the market can end up substantially different from the one at the start, without any single link in the chain having required clinical evidence at the point it was granted.

Researchers call this predicate creep, and it has been documented in peer-reviewed case studies tracing individual product lineages back through successive clearances.

The regime also contains an oddity of history. Devices already on sale when the modern framework began were grandfathered in, which means some chains trace back to products that were never evaluated by anyone.

Whether that matters depends entirely on how far the chain has drifted, and drift is not something the equivalence test was ever designed to measure. Each comparison looks at two devices, not at the sequence they belong to.

There is a third route for the awkward cases. A genuinely novel device with no suitable predicate, but low or moderate risk, can go through a pathway called De Novo, which classifies it and — this is the part that compounds — then becomes a predicate other manufacturers can cite.

Software has made all of this harder. A program that analyses a scan is regulated as a device, and it can be cleared by equivalence to earlier software, but a framework built around unchanging physical objects sits awkwardly with a product that ships an update every month.

Where the comparison with drugs breaks down

The contrast with medicines is the fastest way to see what is unusual here.

A new drug cannot reach a pharmacy without passing through sequential human trials, and the attrition is severe — combining the regulator’s own published phase-transition rates gives roughly six survivors in every hundred candidates, as the mechanics of how clinical trials work set out.

A moderate-risk device can reach a hospital without a single clinical trial, provided a suitable predicate exists.

There are defensible reasons for the asymmetry. Devices are often mechanical, their failure modes are frequently visible rather than biochemical, and requiring full trials for every incremental redesign of a catheter would stop useful iteration dead while doing very little for patients.

The argument is weaker for devices that are essentially software, or that sit inside the body for decades, where failure is neither immediate nor obvious and the patient has no way to inspect the thing that is failing.

What happens after it is on the market

Because so much scrutiny sits after clearance rather than before it, post-market surveillance carries more weight in this system than in the drug system.

Manufacturers, importers and device user facilities are required to report deaths, serious injuries and certain malfunctions to the regulator, and those reports feed a public database.

The weakness is well known. Adverse event reporting is passive: it depends on somebody noticing a problem, connecting it to the device rather than to the underlying illness, and then filing. Under-reporting is assumed rather than debated, and the database is treated as a signal detector rather than a count.

Recalls are the other mechanism, and most are initiated by the manufacturer rather than ordered by the agency. A recall does not always mean removal from hospitals either — in regulatory usage it can mean a correction, a software patch or a change to the instructions, with the device staying exactly where it is.

A recalled predicate does not automatically invalidate the devices cleared against it either. The chain does not unwind on its own.

What to check when something is described as FDA approved

Four questions separate a meaningful claim from a marketing one, and none requires specialist knowledge.

Which word is actually used. Cleared and approved are not synonyms, and the more impressive one is the rarer one.

What the stated intended use is, because equivalence is judged against that specific use and a device cleared for one purpose is frequently promoted for adjacent ones.

Whether any clinical data exists at all, which for a 510(k) device is a genuine question rather than a formality — the pathway permits clearance without it.

And how new the underlying design really is, since a product marketed as an innovation may be riding a predicate chain that has been in service for decades.

The regulatory records are public. Clearance letters, the predicate cited in each submission and the adverse event database are all searchable, which makes this one of the rare areas where a determined reader can check the claim rather than trust it.

The system is not a scandal, and it was designed deliberately to let low-risk iteration proceed without forcing manufacturers to repeat work that has already been done. What it is not is the thing the phrase on the box implies, and the gap between those two is where the useful scepticism belongs.

For a related mechanism, see our explainer on how plane crashes are investigated.