Predicate Devices and Substantial Equivalence, Explained for Medical Software

A predicate device is a product already legally on the market that a new device points to and says, in effect, that it is close enough to that cleared thing to be allowed too. Substantial equivalence is the regulatory finding that the new device shares the same intended use and the same basic technology as its predicate, or that whatever differs does not raise new questions of safety or effectiveness. In the United States this is the logic behind the 510(k) pathway, and it lets many devices reach patients without a fresh clinical trial. That is efficient for well understood hardware. It also means clearance demonstrates likeness to a prior product rather than direct proof of benefit, and clinical software pulls hard on that distinction.

Key points#

• A predicate is an already cleared device that a new one is compared against.
• Substantial equivalence asks whether two products are alike, not whether either helps patients.
• Clearances can chain: device C matched to B, B matched to A, drifting from the original evidence.
• Software breaks three assumptions the framework was built on: sameness, fixedness, and population fit.
• A clearance is a floor to read carefully, not a grade for how well a tool works.

The bargain the pathway makes#

It helps to start with why this route exists, because the rationale is sound within its range. Most medical devices are iterations rather than inventions. A new infusion pump, a redesigned catheter, another blood pressure cuff arrives in a mature category where the ways things fail are already catalogued and the engineering standards are settled. Requiring a full clinical trial for every incremental version would slow useful improvements, raise costs, and rarely reveal anything new about a well characterized technology.

Substantial equivalence lets regulators spend their scrutiny where novelty and risk are highest, and wave through the routine. Applied to physical devices in understood categories, it is a defensible piece of engineering governance. The trouble starts when the same comparison is asked to carry weight it was never designed to bear.

What the comparison actually tests#

Two questions sit at the center of a substantial equivalence review. The first is intended use: does the new device serve the same clinical purpose, in the same setting, for the same kind of patient as its predicate. The second is technological characteristics: does it work in the same basic way, or, if the technology differs, is there evidence that the differences do not raise new safety or effectiveness questions. Clear both, and the device can be marketed on the strength of the comparison.

Notice the shape of that question. It asks whether the new device is enough like something already trusted. On its own it does not ask whether either device improves a health outcome. The original trust was often established years earlier, sometimes through a series of products each cleared against the one before it. The reasoning is comparative, and a comparison is only ever as strong as the thing on the other side of it.

What a clearance proves, and what it leaves unsaid#

For anyone trying to judge a cleared device, one distinction matters more than any other. Clearance through equivalence tells you the device resembles a lawful predicate closely enough that regulators did not demand new safety questions be answered. It does not tell you the device was shown, in a study, to help patients. Those are separate claims, and marketing language blurs them easily.

The chain problem#

Consider what happens across generations. If device C is cleared against device B, and B was cleared against A, the newest product can sit several steps removed from any original body of clinical evidence. Each hop introduced small differences that were individually reasonable and collectively meaningful, and no single equivalence decision ever steps back to examine the full chain. Researchers who study the 510(k) process have described this drift, sometimes called predicate creep, as a structural feature rather than an occasional lapse.

The inherited-flaw problem#

Equivalence also inherits whatever its predicate carried, strengths and weaknesses alike. If a predicate was cleared on thin evidence, or was later found to have problems, the devices that leaned on it are not automatically re-examined. The common mistake in reading a clearance is to treat it as a verdict on the product when it is really a statement about a relationship between two products.

Where software breaks the analogy#

The framework was built for objects you can hold, and software as a medical device strains several of its assumptions at once.

Take sameness first. Two catheters can be compared on materials, dimensions, and mechanical behavior. Two clinical algorithms that share an intended use can operate in completely different ways inside: one a transparent rule set, the other a model trained on data whose contents shape every output it produces. Declaring them equivalent because they take similar inputs and return similar outputs skips over how each one arrives at its answer, and for software the internals are much of the safety story.

Then there is the moving target. A physical predicate is fixed the moment it ships. Software is not. It gets retrained, retuned, wired to new data sources, and pushed updates on a cadence hardware never had. A tool judged equivalent at one moment can behave differently after an update, which means the original comparison describes a version that may no longer be the one running in the clinic.

Population fit is the third strain. A model can be genuinely equivalent to its predicate on the data both were tested against and still perform poorly on a group underrepresented in that data. Equivalence checks likeness to another product, not adequacy for the full range of people the software will actually meet. Two systems can match each other and both miss the same patients.

Reading a clearance well#

None of this makes the pathway illegitimate. It makes a clearance a floor rather than a ceiling, and a narrow floor for software. A useful habit when weighing a digital health tool is to read the intended use statement first, because that single sentence bounds everything the clearance covers, and real world use tends to wander past it.

From there the questions are about evidence beyond the equivalence finding: is there prospective testing, external validation on a population that resembles the intended users, and a plan for monitoring performance once the tool is in use. Cleared status answers a regulatory question. Whether a tool earns clinical trust is a separate inquiry, and it is the one that should govern how much anyone relies on it. Substantial equivalence is a sensible instrument for stable hardware and a stretched one for adaptive software; knowing which situation you are in changes how much a clearance should reassure you.