Choosing the Right Predicate Device: The Most Important Step in Your 510(k) Strategy

Once you decide your product is a good candidate for the 510(k) pathway, the next critical decision is which predicate device you will compare yourself to. For many SaMD and AI teams, this choice feels abstract at first. In reality, it is one of the most strategic calls you will make.

Your predicate device anchors your claim of substantial equivalence. It influences your risk classification, what testing you need, how much data you must generate, and how comfortable FDA feels with your overall approach. A smart predicate choice can keep you on a relatively fast track. A poor one can push you toward a Not Substantially Equivalent (NSE) decision and into De Novo or PMA territory.

This guide walks through the key questions you should ask when selecting a predicate for a SaMD or AI-enabled device.

1. What is a predicate device and what does “substantial equivalence” really mean?

In a 510(k), you are not trying to prove your device is perfect in a vacuum. You are trying to show that it is as safe and effective as an appropriate device already on the US market, known as the predicate.

FDA considers a device substantially equivalent if:

  • It has the same intended use as the predicate, and

  • Either the technological characteristics are the same, or any differences do not raise new questions of safety or effectiveness, and

  • The information you submit shows your device is as safe and effective as the predicate.

For SaMD and AI products, “technological characteristics” include your algorithms, software architecture, data inputs, user interface, connectivity, and any embedded AI models. Small differences are allowed. Large shifts require careful justification and strong performance data.

2. Is the predicate truly and clearly “legally marketed”?

Your first filter is simple: the predicate has to be legally marketed in the US.

FDA allows several types of devices to serve as predicates, including:

  • Devices cleared through 510(k)

  • Preamendment devices that were on the market before May 28, 1976

  • Certain devices originally approved as Class III that were later down-classified

  • Some 510(k)-exempt devices, if used correctly for comparison

Using a device that is not legally marketed as your predicate will almost certainly lead to an NSE decision. That can force you into a De Novo request or PMA pathway, with more time, cost, and complexity.

For software products, it is also worth checking:

  • Whether the predicate is still actively marketed and supported

  • Whether it has major safety issues, recalls, or cybersecurity incidents in its history

FDA has draft guidance on best practices for predicate selection that explicitly encourages choosing predicates that do not carry known safety problems or outdated technology.

3. Does the predicate’s intended use align with your real product vision?

Predicate selection is not just a technical exercise. It is a claims strategy.

FDA requires that your device have the same intended use as the predicate. If your intended use diverges, it becomes difficult or impossible to show substantial equivalence.

Questions to ask:

  • Does the predicate target the same condition or clinical scenario?

  • Is the user population similar? (for example, adults vs pediatrics)

  • Does it occupy the same position in the workflow? (screening, triage, diagnosis, treatment guidance)

  • Are the claims about performance and clinical impact similar to what you are aiming for?

If your software uses AI, be careful not to write an intended use that sounds more autonomous or high-risk than the predicate you want to use. A claim like “assists clinicians in evaluating X” is very different from “automatically diagnoses X.” Those differences can shift you into a different risk category and away from your desired predicate.

In practice, many startups adjust their Version 1 claims to better match a solid predicate, then plan enhancements and broader claims for later updates.

4. How similar are the technological characteristics, especially for SaMD and AI?

Even when intended use matches, FDA will examine whether your device achieves that use with similar technological characteristics.

For SaMD and AI systems, this means comparing:

  • Input data types and sources (EHR data, imaging, sensor streams, wearables, etc.)

  • Core algorithms and processing steps

  • AI model structure and outputs

  • User interface and decision support workflow

  • Connectivity, cloud components, and interoperability

  • Cybersecurity controls and data handling

Differences are allowed, but they matter. FDA will ask whether these changes introduce new questions of safety or effectiveness compared with the predicate.

Examples that can trigger new questions:

  • Moving from rule-based logic to deep learning classification for the same task

  • Shifting from retrospective analysis to real-time monitoring

  • Adding automated treatment recommendations where the predicate only flags results for review

  • Introducing fully remote or home use where the predicate is used only in supervised clinical settings

With recent AI guidance, FDA is also looking more closely at training data, generalizability across populations, and how you manage model updates and drift over time.

When technological differences become too large to comfortably bridge, that is a signal that you may be heading toward De Novo instead of 510(k).

5. What performance data will you need to bridge the gaps?

Once you know how your device differs from the predicate, the next question is what evidence is needed to show that those differences do not make the device less safe or effective.

FDA will look at:

  • Whether your test methods are appropriate and robust

  • Whether you have adequately challenged the device under realistic and worst-case conditions

  • Whether results demonstrate performance that is at least as good as the predicate

For SaMD and AI, this often includes:

  • Software verification and validation

  • Bench testing using representative datasets

  • Human factors and usability testing

  • Cybersecurity testing and threat modeling

  • For AI models, validation across different demographic and clinical subgroups

If your data show clear, comparable performance and no new safety concerns, FDA can still find substantial equivalence even when the tech stack is not identical. If the data expose unresolved safety or effectiveness issues, an NSE outcome becomes more likely, and you may be pushed toward De Novo or PMA.

This is where predicate choice becomes critical. A predicate that is closer to your technology often requires less complex testing to bridge the gap.

6. Is this predicate strategically aligned with your long-term roadmap?

Finally, even if a predicate is legally marketed, has aligned intended use, and similar technology, you should ask whether it is strategically the right choice for your product and company.

Consider:

  • Is the predicate in a device family you want to be associated with long term?

  • Does it represent a level of performance you can match or exceed comfortably?

  • Does it support your plans to expand indications or move into other regions later?

  • Is it recent enough that FDA is comfortable using it as a touchstone for current practice?

For AI and SaMD, there is an extra dimension: you want a predicate that is conceptually similar to how regulators already think about your type of algorithm. Picking something too far removed can make your story harder to tell.

Putting it all together

Predicate selection is not a checkbox exercise. It is one of the most strategic regulatory choices you will make:

  • It locks in your intended use and risk envelope

  • It shapes your testing plan and evidence burden

  • It influences how FDA views your AI or software architecture

  • It can be the difference between a focused 510(k) review and an unexpected detour into De Novo

For first-time founders, this is not something you should improvise in a weekend. The most efficient teams treat predicate selection as a structured decision, grounded in FDA guidance and aligned with their product roadmap.

If you are planning a 510(k) for a SaMD or AI-enabled device and want help identifying and evaluating predicates, aligning your claims, and mapping out the testing needed to support substantial equivalence, Unigen can work with you to design a predicate strategy that supports both fast clearance and long-term growth. Contact us to schedule an introductory call.

Previous
Previous

The lean SaMD regulatory approval starter kit: like an MVP but for documentation

Next
Next

How Intended Use Shapes Your SaMD Classification and FDA Pathway