510(k), De Novo, and PMA: How FDA Device Pathway Applies and What Pushes a Device onto the Harder Route
Selecting the wrong premarket pathway — or failing to anticipate what will disqualify a device from the easier one — can add years and tens of millions of dollars to a development program. For regulatory and clinical teams, understanding the structural logic FDA uses to sort devices into 510(k), De Novo, or PMA is therefore a foundational planning requirement, not a late-stage administrative detail.
The analysis below works through the statutory and regulatory criteria that govern pathway assignment, explains how predicate availability and device class interact, and identifies the specific factors — novel technology, new intended uses, absent predicates, and unresolved safety questions — that have historically moved devices out of the 510(k) lane and into De Novo or PMA.
Want to ask Rhizome your own regulatory questions? Try it for free.
510(k), De Novo, or PMA: which pathway applies, and what pushes a device onto the harder route
For a US medical device, the premarket route is not something a sponsor picks for convenience. It falls out of two questions the agency asks in sequence: does a suitable predicate exist, and how much residual risk remains after the applicable controls are applied. The answer sorts almost every device into one of three lanes: 510(k) premarket notification, De Novo classification, or premarket approval (PMA). The interesting regulatory work is understanding what nudges a device out of the easy lane and into a harder one.
The pathway is set by class and predicate, not by preference
Device class (I, II, or III) describes the regulatory risk tier, while the submission type tells you the premarket mechanism 1525. In the Product Classification database, submission type 4 maps explicitly to "premarket notification submission (510(k))" 1234. Class does not translate one-to-one to a submission, though. Many Class I and a large share of Class II devices are exempt from premarket notification entirely: if the device stays within the limits of its exemption it needs no 510(k), and only when it exceeds those limits does a 510(k) become mandatory 1234678. So "Class II" does not automatically mean "file a 510(k)" 1234.
The three active pathways differ mainly in the evidence and the comparator they rest on:
- 510(k) rests on comparison to a legally marketed predicate.
- De Novo creates a new classification when no predicate exists but the risk is low to moderate.
- PMA rests on independent clinical evidence for high-risk (Class III) devices.
510(k): the substantial equivalence route
A 510(k) clearance turns on a finding of substantial equivalence (SE) to a predicate. A device is substantially equivalent if it has the same intended use and the same technological characteristics as the predicate, or the same intended use with different technological characteristics where those differences do not raise new questions of safety and effectiveness and the device is shown to be as safe and effective as the predicate 135136137138140.
Two conditions do the heavy lifting. Intended use must match the predicate; that is the threshold requirement in every SE comparison 135136137140. Technological differences are tolerated, but only when performance data show they do not create new or different safety and effectiveness questions 136137138139. Reviewers make this explicit in decision summaries, framing it as a direct question ("Do performance data demonstrate equivalence? Yes") 93.
De Novo: novel, but low-to-moderate risk
De Novo exists for the device that has no predicate to compare against yet does not carry enough risk to justify Class III. The statutory logic is that a device not in commercial distribution before the 1976 amendments is automatically Class III (and would need PMA) unless FDA classifies or reclassifies it into Class I or II, and to do that the class must carry enough regulatory controls to give a reasonable assurance of safety and effectiveness 128. De Novo is the request that asks FDA to make that risk-based classification without first filing a doomed 510(k) 128.
In practice a granted De Novo lands the device in Class II with general controls (including design controls) plus device-specific special controls, such as labeling limitations, verification and validation requirements, and other mitigations that FDA judges sufficient to keep the residual risk acceptable 127. The recurring theme in these decisions is that the risk is low to moderate and can be adequately mitigated, so PMA-level Class III regulation is not warranted 127.
PMA: the high-risk route
PMA is the approval standard for Class III devices. FDA grants a PMA only on valid scientific evidence that provides a reasonable assurance of safety and effectiveness, with safety defined as a reasonable assurance, based on valid scientific evidence, that the probable benefits outweigh the probable risks 5152535562677173. The devices in this lane are the ones you would expect: implanted or life-sustaining systems such as replacement heart valves, cochlear implants, spinal cord stimulators, carotid stents, coronary stents, and artificial discs 51525455606265666770717273. Recent original PMAs in the database follow the same pattern, including the Mosaic replacement heart valve 199, the TMJ Concepts patient-fitted temporomandibular joint prosthesis 220, an implantable cardioverter defibrillator platform 222, and an implanted programmable infusion pump 208.
What pushes a device onto the harder pathway
This is the crux of the question. Five triggers move a device off 510(k) toward De Novo or PMA.
1. No suitable predicate exists. Substantial equivalence is a comparison; with no legally marketed device to compare against, the 510(k) argument has no foundation 128. Genuinely new technology is the usual cause. De Novo grants routinely note that no predicate applied, for example the Illumina MiSeqDx next-generation sequencing platform ("Predicate Device Name(s) and 510(k) numbers: Not applicable") 274 and the CDC DENV-1-4 dengue RT-PCR assay, a first-of-kind multiplex test with predicate comparison marked "not applicable" 279. AI/algorithm-driven devices land here frequently: DermaSensor pairs optical spectroscopy with a machine-learning algorithm to assess skin lesions 273, and the Sepsis ImmunoScore is a cloud-based machine-learning model that generates a sepsis risk score from real-time parameters 275.
2. A "not substantially equivalent" (NSE) determination. Even with a candidate predicate, the 510(k) fails if the device cannot fit the SE framework: a different intended use breaks the threshold requirement 91; technological differences that raise new safety or effectiveness questions cannot be reconciled 91; or the performance data simply do not demonstrate equivalence 9193. Any one of those forces the device off the pathway, at which point the sponsor must pursue De Novo (if risk is low-to-moderate) or PMA.
3. The automatic Class III default. The two triggers above interact with a statutory backstop: a novel post-1976 device, or one found NSE, defaults into Class III and its PMA requirement unless it is reclassified into Class I or II 128. FDA treats affected devices as Class III subject to PMA, and a sponsor who wants out must succeed on a reclassification request rather than assume the lighter route 223235245. De Novo is the mechanism that interrupts that default for lower-risk devices 128.
4. High, implant-level, or life-sustaining risk. When the residual risk cannot be controlled by general and special controls, the device stays in Class III and PMA applies. The implanted and life-sustaining systems above illustrate the profile that keeps a device in the PMA lane rather than allowing a De Novo down-classification 5162.
5. New questions of safety and effectiveness. This is the hinge across all three pathways. Technological differences are acceptable in a 510(k) only when they raise no new questions 135136137138140; the moment they do, the device needs either the special controls of a De Novo or the clinical evidence of a PMA to resolve those questions.
Worked contrast: De Novo versus PMA in recent decisions
| Device (submission) | Pathway | Why this lane |
|---|---|---|
| DermaSensor (DEN230008) | De Novo | Novel optical-spectroscopy + AI algorithm, no predicate; adjunctive only, not a standalone diagnostic and not a biopsy replacement, so risk stayed low-to-moderate 273 |
| Sepsis ImmunoScore (DEN230036) | De Novo | Novel cloud ML model; intended for use alongside other findings, outputs a risk score rather than a treatment decision 275 |
| Illumina MiSeqDx (DEN130011) | De Novo | First-of-kind NGS platform, predicate "not applicable"; characterized through analytical studies 274 |
| Teal Wand (DEN240045) | De Novo | New device type granted a novel product code (SEP) 339 |
| ArteraAI Prostate (DEN240068) | De Novo | Novel AI prognostic device, new product code (SFH) 332 |
| FARAPULSE Pulsed Field Ablation (PMA) | PMA | Prospective, Bayesian adaptive, multicenter 1:1 randomized non-inferiority pivotal trial with 12-month follow-up 292284 |
| TriClip G4 (PMA) | PMA | Randomized pivotal trial, hierarchical composite endpoint at 12 months, advisory panel review and post-approval study 287289 |
| Guardant Shield (PMA) | PMA | Diagnostic PMA taken to an advisory panel with formal safety/effectiveness/benefit-risk votes 283 |
| Mosaic Bioprosthesis (P990064) | PMA | Implanted replacement heart valve, Class III 199 |
The PMA evidentiary and procedural burden
The reason pathway assignment matters so much commercially is the gap in burden. A PMA is typically built on a pivotal clinical study, often prospective and sometimes randomized, with prespecified endpoints and long follow-up: FARAPULSE (randomized non-inferiority, 12-month follow-up) 292, TriClip (randomized, composite endpoint at 12 months) 287289, IN.PACT Admiral (randomized, follow-up planned to five years) 290, and ReActiv8 (Level 1 evidence from a pivotal randomized trial) 296. Many PMAs go to an FDA advisory panel that votes separately on safety, effectiveness, and benefit-risk, as with AtriCure Synergy, Guardant Shield, TriClip, and OCS Liver 282283289298; others are approved without a panel when the evidence package is already well vetted 284286291295. The PMA record can also carry a Quality System (21 CFR 820) inspection finding, as documented for SurVeil 303, and frequently commits the sponsor to a post-approval study covering longer-term durability, training, or labeling questions, as with AtriCure, TriClip, and the OCS organ-perfusion systems 282289297298. None of that machinery attaches to a routine 510(k), and a De Novo sits in between: no independent clinical-effectiveness burden of a PMA, but a set of special controls the sponsor must meet and maintain 127.
Practical read for regulatory teams
Start every pathway assessment with the predicate search and the intended-use statement, because those two determine whether 510(k) is even available 13591. If no predicate exists, the real question is not "510(k) or not" but "De Novo or PMA," and that turns on whether general and special controls can hold the residual risk to a reasonable assurance of safety and effectiveness 127128. Watch the intended-use wording and any technological change most closely: a shift in intended use or a difference that raises a new safety or effectiveness question is the most common way a device that looked like a clean 510(k) ends up needing a De Novo or a full PMA dossier 91135136. When the device is implanted, life-sustaining, or otherwise high-risk, plan for the PMA burden from the outset, including a pivotal trial, likely panel review, and post-approval commitments 292289303.