How Do You Choose a Predicate Device for a 510(k)?

By Kevin Buckley · Published: October 30, 2026

The best predicate is the one that gives you the shortest substantial-equivalence argument without blocking your launch. That makes the choice a regulatory question and a patent question at the same time. The predicate should have published data on its safety and effectiveness that you can use to guide equivalence testing of your own device, and that data should be as easy as possible to meet or beat. Many people overlook whether the predicate is patented: the 35 U.S.C. § 271(e)(1) safe harbor protects your equivalence testing, but your first sale of the cleared device could infringe the predicate owner's patent. That's why, if possible, you should choose a predicate that is off patent or one whose last patent expires soon. A Pre-Submission lets you ask FDA about the regulatory side of the choice before you file.

How does substantial equivalence work?

A 510(k) asks FDA to agree that your device is "substantially equivalent" to a device already legally on the U.S. market, called the predicate. The test has two parts (FDA's 510(k) page). Your device must have the same intended use as the predicate, and it must either share the predicate's technological characteristics or differ in ways that don't raise different questions of safety and effectiveness, with data showing it is as safe and effective. The first part is pass/fail.

Why does intended use come first when choosing a predicate?

Intended use is the general purpose of the device. The indications for use are narrower and describe the specific disease, condition or patient population. Expanding the indications beyond the predicate's can create a new intended use, for example by adding a patient population or anatomical site, or by turning a screening claim into a diagnostic one (FDA's 2014 substantial-equivalence guidance). When that happens, the predicate can't support your submission, and if no other predicate covers the expanded use, you are looking at De Novo classification or premarket approval (PMA).

When we evaluate a candidate predicate, we read its cleared indications word by word against the client's marketing plan. We typically see clients describe what the device does on a website in terms of treating a broad condition, such as neurodegeneration or cardiovascular disease, when the label clears the device only for Parkinson's disease or stroke, for example. FDA treats that copy as evidence of intended use. Under 21 C.F.R. § 801.4, intended use is the "objective intent" of the company, shown by "labeling claims, advertising matter, or oral or written statements."

Some clients also try to go direct to consumer (DTC) with a device before obtaining FDA clearance for that same device. In the DTC channel, they try to avoid medical-use claims and rely on "wellness" or other softer claims, but they often go beyond what FDA will allow, which undermines the clearance application. FDA's General Wellness guidance, reissued January 6, 2026, covers only low-risk products intended solely for general wellness use. We advise clients to keep medical claims out of DTC marketing entirely and to market a cleared device to its label.

How many technological differences can you defend?

Technological characteristics cover design, materials, energy source, software and operating principles. Each difference from the predicate has to be explained, and most have to be tested. Some can't be bridged in a 510(k) at all, such as a new energy type or a fundamentally different mechanism of action, because they raise questions the predicate comparison can't answer. For that reason we would rather work from an unremarkable device built the way our client's is than from a market leader that works on a different principle.

Some clients target the market leader because they sense a big market opportunity. The predicate's safety and effectiveness data then typically turns out to be less available or repeatable than the client thought, and they can't meet or beat it in equivalence testing. And because market leaders usually hold multiple patents on the predicate device, the client has a high risk of infringing one or more of the predicate owner's patents. Designing around those patents doesn't avoid the FDA problem either, since a change made to avoid a patent is a technological difference like any other.

What does FDA's predicate best-practices draft recommend?

FDA's September 2023 draft guidance on choosing among valid predicates is still a draft (FDA draft guidance). It recommends a predicate cleared using well-established methods, meaning current FDA-recognized consensus standards, FDA guidance or accepted published methods. The draft also looks for a predicate that meets or exceeds expected safety and performance, judged partly by adverse event history, with no unmitigated use-related or design-related safety issues and no design-related recall, so search FDA's recall and adverse event databases for each candidate and for the device it was cleared against.

Reviewers read the draft even though it isn't binding. In our experience its recommendations mostly write down what reviewers were asking anyway, so we follow it now.

Does the age of a predicate device matter?

Predicates don't expire. Under 21 C.F.R. § 807.92(a)(3), a pre-1976 device, a reclassified device or a device found substantially equivalent can serve, unless FDA has removed it from the market or a court has found it misbranded or adulterated (FDA's 2014 guidance).

FDA's predicate-device webpage still recommends a recently cleared predicate where possible. Its 2023 draft takes a different approach, proposing that predicates be judged by their characteristics "rather than just their age" (FDA draft guidance). The practical risk with an old predicate is that its testing may not meet current standards, in which case you test to today's expectations regardless. Older predicates are more likely to be off patent, though, and we sometimes prefer one for that reason.

Can you use more than one predicate device?

FDA expects one primary predicate and allows more in some situations, but it won't accept a split predicate, where one device supplies the intended use and another supplies the technology (FDA's predicate-device webpage). A reference device can support your test methods, but it can't supply an intended use or technology your predicate lacks.

How do you find and rank predicate candidates?

The search starts with the device's three-letter product code in FDA's classification database, and searching the 510(k) database by that code produces the candidate list. Then read the 510(k) summaries. Under 21 C.F.R. § 807.92, each summary has to identify the predicate it relied on, compare technological characteristics and discuss the testing behind the clearance.

The mechanism of action is the part of a competitor's summary to read most closely. If you can't test your device apples to apples against the predicate using standard test methods, you likely won't succeed in showing equivalence. If there really is only one candidate predicate and you find this problem, you typically need to consider a De Novo request instead.

Since October 1, 2023, every 510(k) has had to use eSTAR (FDA's 510(k) page), which requires a side-by-side comparison with the predicate (FDA eSTAR guidance). A candidate whose summary is too thin to support that comparison is a weaker one.

Competitors' devices are fair game as predicates, though a client's own cleared device is often better because the client knows everything about it. For some well-understood device types, FDA's Safety and Performance Based Pathway lets you meet FDA-identified performance criteria instead of testing head-to-head against the predicate (FDA).

When we rank candidates, intended-use fit is pass/fail. Among those that pass, we weigh well-established test methods most heavily, because methods FDA already accepts can be followed closely and hold up over time. The number and seriousness of technological differences come next, followed by safety history and how much of the testing is public. Patent status breaks ties, and it can favor an older predicate. We have chosen an older cleared device over a newer one when the newer one was still under patent, though age alone doesn't establish that a device is off patent.

What does your predicate choice mean for patent risk?

The FDA argument and the patent risk pull in opposite directions: the closer your device is to the predicate, the easier the equivalence argument and the more likely your device falls within the predicate owner's patent claims.

The testing and clearance work is protected. 35 U.S.C. § 271(e)(1) provides that it is not infringement to make, use, offer to sell, sell or import a patented invention "solely for uses reasonably related to the development and submission of information" under a federal law regulating drugs. The Supreme Court read that language to cover medical devices in Eli Lilly & Co. v. Medtronic, Inc., 496 U.S. 661 (1990) (Justia), and the Federal Circuit has read it broadly since, as in Edwards Lifesciences Corp. v. Meril Life Sciences Pvt. Ltd., No. 22-1877 (Fed. Cir. Mar. 25, 2024) (Federal Circuit opinion).

Protection stops at commercial sale, when § 271(a) applies to you as it does to anyone. If the predicate is still under patent and your device practices its claims, the first sale may be the first infringing act. Naming a device as your predicate concedes nothing about its patent claims. Your 510(k)'s description of how your device works is public, though, and litigants read it closely. Anyone with a patent on the shared technology can sue, not just the predicate's maker. Age is also a weak guide to expiry. A U.S. patent generally ends 20 years after the filing date of its earliest nonprovisional U.S. application (35 U.S.C. § 154(a)(2)), and improvement patents and continuations can extend coverage after the original lapses.

Predicate patents come up early in our practice because our Invention Disclosure Form (IDF) asks clients to identify competitor patents. When we use the IDF to conduct a patentability study, we typically identify the predicate's patents there. A patentability study only tells us the patent exists, so when one turns up there, we follow up with a freedom-to-operate (FTO) review of its claims. We can also find a predicate patent while conducting an FTO search or drafting an FTO opinion. Device makers' patent marking, on the product or a virtual-marking webpage under 35 U.S.C. § 287(a), is a useful first list. Later, as data comes out of the equivalence studies, we look at the novel and non-obvious aspects of that data, and of any new prototype or proof-of-concept work, and consider filing a new patent application. That typically involves a new patentability study, where we could catch the predicate's patents again.

When we find a live patent, we suggest that the client try to identify another predicate. If the patent is truly unavoidable, the consideration becomes whether to launch anyway, knowing the risk of being sued, including enhanced damages of up to three times the amount found for willful infringement under 35 U.S.C. § 284. In that case, the client needs to plan carefully for a settlement or cross-license before litigation starts. If our client is aggressive, it can try to invalidate the predicate patent or patents, either through inter partes review (IPR) at the Patent Trial and Appeal Board or in litigation if it comes to that. IPR is limited, though: it can challenge claims only as anticipated or obvious, and only over patents and printed publications (35 U.S.C. § 311(b)). A petition is barred if it's filed more than one year after the petitioner is served with an infringement complaint (§ 315(b)). The USPTO can also deny institution on discretionary grounds, including the patent owner's "settled expectations," such as how long the claims have been in force (USPTO). That factor works against challenging the old patents that tend to cover market-leading predicates. USPTO practice here is still being worked out, so expect changes. For a client with minimal resources, an alternate predicate is the smartest choice.

FDA publishes the 510(k) summary within 30 days of clearance (21 C.F.R. § 807.95), so the client's own applications on anything new should be on file before then.

Can an AI-enabled device be a predicate?

Usually, yes. By our count of FDA's list of AI-enabled devices, current as of September 4, 2026, 1,553 of 1,614 authorizations (about 96%) were 510(k) clearances, with 40 De Novo grants and 21 PMA approvals (FDA AI-enabled devices). The list isn't comprehensive, but each entry shows a product code, which makes it a good place to start.

Where AI predicates fall short, in our experience, is the complexity of the underlying algorithm itself. If a competitor or predicate owner has a cleared algorithm, that algorithm is usually somewhat to very different from the one the client is developing. In cancer detection, for example, the client may be using datasets that differ from the predicate owner's, and those datasets are pre-processed, processed and classified in different ways. There may also be a combination of algorithms, or some combination within the model, that looks at the underlying data differently. Because of the AI's underlying workflow, sensitivity and specificity will differ, and so will how predictable the output is, whether that's a report, a prediction or a ranking. That's why the comparison usually comes down to performance testing against the predicate's reported sensitivity and specificity, run on data the client can defend.

Model updates raise a separate question. A locked model doesn't change once it's on the market. An updatable model can be changed under a predetermined change control plan (PCCP) that FDA clears as part of the 510(k) under section 515C of the Federal Food, Drug, and Cosmetic Act (FD&C Act), 21 U.S.C. § 360e-4, and changes made under the plan don't need a new 510(k). For predicate purposes, though, a device authorized with a PCCP is compared as originally cleared, and an updated version becomes an eligible predicate only once it is cleared in its own 510(k) (FDA PCCP guidance).

Does a U.S. predicate work outside the United States?

Plans to sell in Europe and China should shape the predicate choice. The job is much easier when the U.S. predicate is already on the market there, but both places use different standards than the U.S. The EU Medical Device Regulation (MDR) doesn't clear devices the way FDA does. Instead, it asks whether your device is equivalent to another device for purposes of clinical evaluation, and that test is stricter than substantial equivalence (MDCG 2020-5). For implantable and Class III devices, MDR Article 61(5) generally lets you claim equivalence to another company's device only if you have a contract giving you ongoing access to its technical documentation, which often rules out a competitor's device (MDCG 2023-7). China uses a "same-type device" (同品种医疗器械) comparison against devices registered with the National Medical Products Administration (NMPA) (NMPA 2021 Notice No. 73). If the U.S. predicate isn't on the market in Europe and China, expect a challenge in those countries. Find out early whether it can serve as the equivalent device under the MDR and as the same-type device in China.

On the patent side, if you're going outside the United States, you have to determine whether the predicate has patent protection in Europe and China. If it doesn't, easy peasy. If it does, look at claim scope. Each country has its own patent laws, so infringement and validity need a separate analysis in each jurisdiction instead of an assumption that the U.S. analysis carries over. Protection for regulatory testing is uneven, too. Article 75(5) of China's Patent Law expressly covers medical devices (CNIPA translation), while Article 27(d) of the Agreement on a Unified Patent Court doesn't mention devices, and national exemptions in Europe differ.

What if no predicate device fits?

It happens. We've had clients use the De Novo pathway instead of the 510(k) pathway because the technology was so new that there simply wasn't a good predicate. A device with no predicate defaults to Class III, which normally requires a PMA. A De Novo request under section 513(f)(2) of the FD&C Act, 21 U.S.C. § 360c(f)(2), asks FDA to classify it into Class I or II based on its low-to-moderate risk, where general controls, or general and special controls, provide reasonable assurance of safety and effectiveness (FDA's De Novo page). A granted De Novo device can then serve as a predicate for later 510(k)s, including competitors'.

Contacting FDA early, usually through a Pre-Submission, tells you what questions FDA needs answered and what evidence would help show safety and effectiveness. That hopefully avoids the PMA route, which usually means extensive clinical data, often from a full clinical trial. Going without a predicate removes the infringement questions about a predicate owner's patents, though an FTO opinion is still advisable at this stage because new technology can read on someone else's patents.

Should you ask FDA about your predicate before you file?

FDA's Q-Submission guidance lists "Is the proposed predicate device appropriate if we demonstrate substantial equivalence?" as an example of a regulatory strategy question. FDA won't decide until it reviews the 510(k), but its written feedback on the predicate and test plan is the closest thing to a preview. A predicate question works best with a side-by-side comparison table, an explanation of why you chose this predicate over the alternatives and the tests you propose for each difference. That material later goes into the eSTAR comparison. FDA's feedback won't address patents, so we run the patent review alongside the Pre-Sub.

If you're weighing predicate candidates or preparing a Pre-Submission and want a regulatory and patent read before you commit, contact us.

This FAQ provides general information, not legal advice, and does not create an attorney-client relationship. Outcomes depend on specific facts and jurisdictions. See our Legal Notices.