How to Use a De Novo as a Predicate for Your 510(k) AI Device Submission

When FDA grants a De Novo request, it does three things simultaneously: it clears the specific device, it creates a new product code, and it establishes the special controls that will govern all future devices in that category. For the companies that submit 510(k)s after the De Novo, this is enormously valuable — you have a cleared device, a defined regulatory framework, and a product code to submit under.

But using a De Novo as a predicate requires understanding what the De Novo actually established and how your device compares to it.

What Makes a De Novo a Good Predicate

Not all De Novos make equally good predicates. The best De Novo predicates have:

• A clearly defined intended use that overlaps with yours
• Special controls that your device can demonstrably meet
• A product code that fits your device's indication
• Downstream 510(k) clearances that show FDA has accepted subsequent submissions

That last point matters more than most people realize. A De Novo that has been cited as a predicate multiple times in subsequent 510(k)s gives you evidence that FDA reviewers have repeatedly accepted the predicate logic. A De Novo that has never been used as a predicate may be technically valid but introduces more uncertainty.

The Most-Used AI De Novo Predicates

DEN220043 — Eko Health (QMF, 2022)

The Eko CORE Analysis De Novo created product code QMF and has become the most-cited predicate in radiology and cardiac AI. With 47+ downstream 510(k)s, it's the gold standard for AI analysis software predicates. If your device analyzes any kind of biomedical signal or image with an AI algorithm, there's a good chance QMF is relevant.

DEN200080 — Caption Health (QFP, 2021)

The Caption Guidance echocardiography De Novo created QFP and governs cardiac image acquisition AI. 12 downstream clearances. Strong predicate for any AI that guides clinicians through an imaging acquisition workflow.

DEN180005 — iCAD (QNC, 2019)

Established the mammography AI product code QNC. Predicate for density assessment, lesion detection, and other mammography AI applications.

DEN160026 — Digital Diagnostics (OWJ, 2018)

The IDx-DR autonomous diagnostic AI. Rarely used as a predicate because its special controls are strict — it covers fully autonomous AI that provides diagnoses without clinician review. Only use this predicate if your device truly operates autonomously.

How to Structure Your Predicate Comparison

When you use a De Novo as your predicate in a 510(k), FDA expects you to demonstrate substantial equivalence to the De Novo device. Your submission should include:

First, a comparison of intended use. Your device's intended use must be the same as or more narrow than the De Novo predicate. Expanding beyond the De Novo's intended use is a common reason for non-substantial equivalence (NSE) findings.

Second, a feature comparison. For AI devices, this means comparing your algorithm's design, inputs, outputs, and intended clinical workflow to the De Novo device. Technological differences are acceptable — you just need to show they don't raise new safety or effectiveness questions.

Third, special controls compliance. The De Novo order established specific special controls. Your submission must demonstrate that your device meets each of them. This is where AI devices most often stumble — the special controls for AI typically include performance requirements, labeling requirements, and post-market monitoring obligations.

When to Use a Downstream 510(k) Instead

Sometimes a 510(k)-cleared device that itself used a De Novo as a predicate is actually a better predicate for your submission than the De Novo itself. This is true when:

• The downstream device is more similar to yours in technology or intended use
• The downstream device demonstrates FDA has accepted a specific approach you're planning to take
• The De Novo's special controls were written broadly and a downstream clearance shows how FDA applied them in practice

In practice, many strong submissions use both: the De Novo as the primary predicate that establishes the product code, and a downstream 510(k) as a secondary predicate that demonstrates precedent for specific technical approaches.

Predicate Strategy for Novel AI Indications

If your AI device addresses an indication without a relevant De Novo, you have two options: find a predicate from a different indication where the technology is similar (a cross-indication predicate strategy), or pursue De Novo yourself.

Cross-indication predicates can work — FDA cares more about technological similarity than indication similarity in some contexts — but they require careful justification and are more likely to result in Additional Information requests.

De Novo is appropriate when your device is genuinely novel and no predicate exists. The benefit is that you establish the product code and special controls for your category. The cost is time (typically 18-24 months vs 9 months for 510k) and resources.