The 510(k) Predicate Problem: Why FDA Refuses Clearance and How AI Research Changes the Outcome

The most expensive mistake in a 510(k) submission costs nothing to make — and everything to fix. It happens at the very beginning of the process, before you’ve written a single submission section, before you’ve run a single bench test, and well before FDA has touched your file. It’s the choice of your predicate device.

A flawed predicate selection doesn’t just invite Additional Information (AI) requests from FDA. It can produce a Not Substantially Equivalent (NSE) determination that sends your device back to square one — or into the De Novo process — after months of preparation and hundreds of thousands of dollars in clinical and engineering work.

And yet, predicate research at most medical device companies still means a regulatory affairs specialist querying the FDA 510(k) database by keyword and selecting the most recent cleared device that sounds similar to theirs.

We can do considerably better.

Why Predicate Selection Breaks 510(k) Submissions

FDA receives approximately 3,500 510(k) submissions annually. Of those, roughly 15–20% receive a Refuse to Accept (RTA) determination before substantive review begins — often because the submission’s core framework is structurally flawed from the start. Predicate selection is frequently the root cause.

The substantial equivalence standard, established under 21 CFR 807.87 and amplified through FDA’s 510(k) Program guidance documents, requires demonstrating that your device has the same intended use as the predicate and either (a) the same technological characteristics, or (b) different technological characteristics that don’t raise new questions of safety and effectiveness and that perform at least as well.

That second path — different technological characteristics — is where submissions most often collapse. Companies underestimate how carefully FDA reviewers scrutinize comparative performance data. They choose a predicate that’s technically cleared but whose product code carries regulatory history that cuts against their device. They pick predicates that are several technology generations old, creating a side-by-side comparison that highlights how different their device is rather than how similar.

Three predicate selection errors account for the large majority of problems we see during regulatory compliance consulting engagements:

1. Choosing a recalled predicate. A predicate doesn’t disappear from the 510(k) database because it was later recalled. FDA won’t automatically reject a submission built around a recalled predicate, but reviewers will scrutinize every technological equivalence claim far more aggressively. If the reason for recall relates to the same performance characteristics you’re claiming equivalence on, you may be constructing an unwinnable argument.

2. Selecting a predicate from a mismatched product code. The FDA product classification database (organized under 21 CFR Parts 862–892) groups devices by product code, each carrying its own special controls, performance standards, and review history. Selecting a predicate from an adjacent but different product code — even if the devices look functionally similar — can mean your submission gets evaluated against the wrong regulatory framework entirely.

3. Using multiple predicates without a clear primary. Split predicates (one device for intended use, another for technological equivalence) can work, but FDA’s guidance on the 510(k) Program makes clear the agency views this approach with heightened skepticism when the predicates were cleared under different regulatory frameworks. If you use this strategy, your rationale needs to be airtight.

What “Substantial Equivalence” Actually Requires — and Where Teams Miss It

FDA’s standard isn’t purely technical. It’s interpretive — and it has evolved substantially over the past decade.

The agency has repeatedly updated its thinking through device-specific guidance documents, the “Deciding When to Submit a 510(k) for a Change to an Existing Device” guidance, and an accumulating body of NSE decisions that effectively constitute FDA’s evolving interpretation of equivalence across different technology categories. The result is that the substantial equivalence analysis in 2026 is meaningfully more demanding than it was in 2016. Reviewers have a richer body of NSE precedents, more sophisticated performance benchmarks for common device types, and increasingly AI-assisted review tools on their own side of the submission portal.

Performance data requirements vary significantly by device type. For an in vitro diagnostic (IVD), substantial equivalence means analytical validation against the predicate’s performance specifications — sensitivity, specificity, precision, interferences. For a cardiovascular implant, it may require bench mechanical testing, finite element analysis, and fatigue data demonstrating durability over the device’s expected service life. Understanding what FDA has historically required for your specific product code before you design your testing program — not after — is often the difference between a 90-day clearance and an 18-month correction cycle.

The FDA’s 510(k) Summary database and the MAUDE (Manufacturer and User Facility Device Experience) adverse event database together contain a combined total approaching 1.3 million records. Mining them manually for a single submission is impractical. Missing a pattern buried in that data — for example, that devices in your product code with your specific intended use consistently receive additional information requests about biocompatibility — is an avoidable setback that structured analysis catches before FDA does.

How AI-Augmented Predicate Research Reduces NSE Risk

The practical opportunity here is significant. The FDA’s entire 510(k) clearance database is publicly accessible. So is MAUDE, the product classification database, the De Novo database, and the Premarket Approval (PMA) database. Fifty-plus years of regulatory decisions are sitting in structured, queryable data — and the majority of device companies are still using keyword search to navigate it.

AI-augmented regulatory tools engage that corpus in meaningfully different ways:

Semantic predicate matching goes beyond keyword search to identify cleared devices by functional and technical similarity rather than product name. A software-assisted surgical planning tool and a surgical navigation system may share no matching keywords but have directly analogous regulatory histories and performance requirements. Semantic analysis surfaces those connections in seconds.

Recall and enforcement cross-referencing automatically flags whether a candidate predicate device or its manufacturer carries adverse regulatory history — recalls, warning letters, import alerts, Class I or II enforcement actions — and surfaces that context before a team commits to a predicate strategy. A two-minute automated check prevents a months-long problem.

NSE decision pattern analysis examines which technological differences have historically triggered NSE determinations in a given product code, then flags whether your device’s characteristics fall into those categories. This is particularly valuable for Class II devices where the technological characteristics analysis is carrying the most argumentative weight in the submission.

Submission timeline benchmarking draws on historical clearance data to estimate realistic review timelines for a specific device type and product code. This matters practically: project teams routinely set milestone schedules based on aspirational timelines rather than actual FDA review history for comparable devices.

At Aurora TIC, we’ve built predicate research into our AI-augmented regulatory consulting workflow — not as a standalone database query, but as a structured analytical process that produces a decision-grade predicate rationale document defensible in front of an FDA reviewer or during a pre-submission (Q-Sub) meeting.

Building the Predicate Rationale Document FDA Reviewers Want to See

FDA reviewers aren’t simply looking for you to name a predicate. They’re evaluating the rigor of your predicate rationale — and that rationale needs to be documented in your 510(k) submission clearly enough that a reviewer unfamiliar with your device type can follow your logic without asking for clarification.

An airtight predicate rationale document addresses five things explicitly:

1. Intended use comparison. A sentence-by-sentence comparison of your device’s intended use with the predicate’s cleared intended use, directly quoting from the predicate’s 510(k) Summary where possible. Paraphrase invites disagreement; direct quotation does not.

2. Technological characteristics comparison. A structured table identifying each relevant technological characteristic, whether it matches the predicate or differs, and for any difference, the specific performance data that demonstrates the difference doesn’t raise new safety or effectiveness questions.

3. Product code justification. A statement confirming your device falls within the predicate’s product code, along with an explanation of any borderline classification questions you considered and resolved.

4. Predicate regulatory history. A brief note confirming the predicate is currently cleared (not recalled), its clearance date, and its K-number. This takes 60 seconds to document and removes a predictable reviewer question.

5. Alternative predicate consideration. For complex or novel devices, a brief explanation of why the chosen predicate is more appropriate than the other cleared devices you evaluated. This step is optional — but it signals analytical discipline to reviewers and preempts questions about whether you considered stronger predicates.

A predicate rationale built this way isn’t just a submission artifact. It becomes the roadmap for your performance testing program, your biocompatibility assessment plan, and your labeling. Getting it right at the outset — informed by systematic, AI-augmented analysis of the precedent landscape — compresses the entire development timeline, not just the FDA review phase.

The 510(k) pathway is fundamentally about argument quality. FDA isn’t approving your device the way a PMA involves approval; they’re clearing it based on how persuasively you’ve established substantial equivalence. That argument starts with predicate selection. And that’s one decision your team shouldn’t make based on the first device that surfaces in a keyword search.

Written by Sam Sammane, Founder & CEO, Aurora TIC | Founder, Qalitex Group. Learn more about our team

Reserve early access to our AI audit tools Contact us

Related from our network

• ISO 17025-Accredited Testing for Medical Device Manufacturers — Qalitex Laboratories provides analytical testing and method validation data to support 510(k) performance comparisons and design verification packages.
• Health Canada CMDCAS and MDSAP Device Consulting — Androxa supports medical device manufacturers navigating Canadian regulatory requirements alongside FDA 510(k) clearance strategies.