The FDA De Novo Pathway, Explained

A new medical device can lack a legally marketed predicate and still be suitable for class I or class II regulation, and the FDA De Novo pathway is designed for that situation. It lets FDA evaluate a novel device type, identify its risks, decide whether regulatory controls can manage those risks, and create a new classification when the evidence supports doing so.

“Novel” is not enough. A sponsor needs a defined intended use, a coherent risk analysis, and valid evidence that the device performs safely and effectively for that use. If general controls, or general and special controls, cannot provide reasonable assurance of safety and effectiveness, De Novo is not the proper class I or II destination.

A grant has consequences beyond the first product. It establishes a device type, classification regulation, and controls. The granted device may then serve as a predicate for later 510(k) submissions when appropriate.

Why a new device defaults toward class III#

The Federal Food, Drug, and Cosmetic Act classifies devices according to the controls needed to provide reasonable assurance of safety and effectiveness. Class I generally relies on general controls. Class II uses general controls plus special controls. Class III devices generally require premarket approval unless another statutory route applies.

Under section 513(f)(1), a device type first introduced after the 1976 Medical Device Amendments is automatically class III if it is not substantially equivalent to a preamendments device, unless FDA reclassifies it or grants De Novo classification.

That default prevents a novel technology from bypassing review simply because no comparison exists. It can also overclassify a low-to-moderate-risk device if no pathway allows a risk-based review. Section 513(f)(2) supplies that pathway. De Novo does not weaken the evidence question. It changes the classification question from “is this substantially equivalent to a predicate?” to “what class and controls are appropriate for this new type?”

The central eligibility test#

A De Novo request is appropriate when no legally marketed device of the same type provides a suitable predicate and the device can be classified into class I or II under the statutory criteria.

Class I is appropriate when general controls are sufficient to provide reasonable assurance of safety and effectiveness, or the device is not life-supporting or life-sustaining and does not present a potential unreasonable risk of illness or injury. Class II is appropriate when general controls alone are insufficient but special controls can provide the needed assurance.

If the device is life-supporting, life-sustaining, of substantial importance in preventing impairment of human health, or presents a potential unreasonable risk, and insufficient information exists to determine that controls can manage risk, class III may be required. The analysis is device-specific. Software, an in vitro diagnostic, an implant, and a home-use monitor can all use De Novo in principle, but their evidence and controls will differ.

Two routes to the same classification request#

The original route followed a 510(k). FDA would find the device not substantially equivalent, often because no predicate existed or because a different intended use or technology raised different safety and effectiveness questions. The sponsor could then request De Novo classification.

Congress later created a direct route. A requester that determines there is no legally marketed predicate may submit a De Novo request without first filing a 510(k). This avoids a knowingly unproductive predicate comparison.

Both routes lead FDA to a risk-based classification evaluation. A prior 510(k) NSE decision does not guarantee a grant. A direct submission does not lower the standard.

If a predicate actually exists and the device can be evaluated for substantial equivalence, the 510(k) route may be the correct pathway. FDA's Q-Submission program can support early discussion when classification or evidence strategy is uncertain.

Intended use defines the regulatory claim#

The intended use and indications for use establish what the device is for, who uses it, target population, setting, input, output, and clinical role; they anchor the relevant risks and performance claims.

A tool that helps a clinician prioritize images is different from one that provides an autonomous diagnosis, and a home test is different from a laboratory test even if they measure the same analyte. A wellness output is different from a claim to detect disease.

Seemingly small wording changes can alter risk. “Provides information for consideration” does not automatically make a function low risk if users are expected to rely on it. FDA evaluates the actual role, not only cautious verbs.

The evidence must match the proposed use. Data from adults may not support children. A study in a specialty center may not support untrained home users. Performance on curated files may not support real-time operation with missing or poor-quality inputs.

What goes into a De Novo request#

The exact content depends on device type, but 21 CFR part 860, Subpart D and FDA guidance establish core elements.

The request identifies the requester and device, provides regulatory history, and explains why no legally marketed predicate exists; it describes intended use, indications, technological characteristics, components, principles of operation, and relevant accessories.

The requester recommends class I or II and explains why. For class II, it proposes special controls. The request identifies known and reasonably foreseeable risks to health and maps each risk to mitigation measures.

Performance information may include bench, software, cybersecurity, electrical safety, electromagnetic compatibility, biocompatibility, sterility, shelf life, animal, human factors, and clinical evidence. Not every category applies to every device, but omissions need a scientific basis.

Proposed labeling explains intended users, instructions, warnings, limitations, interpretation, maintenance, and training. Manufacturing information supports confidence that production units will match those evaluated. The submission also includes required administrative forms and certifications. A complete request tells one consistent story across use, risk, evidence, controls, and label.

Risk analysis is the organizing spine#

A risk table should not be a generic list of hazards: it links how the device can fail or be misused to the resulting clinical harm, then identifies evidence and controls that reduce probability or severity.

For a diagnostic, risks may include false positive, false negative, invalid result, cross-reactivity, specimen error, delayed result, and misinterpretation. For software, risks may include corrupted input, wrong output, automation bias, cybersecurity compromise, unavailable function, and performance differences across sites or populations.

For a physical device, mechanical failure, electrical injury, thermal injury, infection, material toxicity, breakage, migration, and use error may matter. Home use adds lighting, storage, connectivity, language, dexterity, and untrained-user conditions.

Mitigation can come from design, verification, alarms, lockouts, user interface, training, labeling, quality controls, performance testing, postmarket monitoring, and restrictions on use. Labeling alone is weak when a design control can prevent the hazard. Residual risk remains after controls. FDA considers whether probable benefits outweigh probable risks and whether the total evidence supports reasonable assurance of safety and effectiveness.

Special controls turn evidence into a device-type rule#

For a class II De Novo, special controls are requirements beyond general controls that address the risks of the device type, and they may require specific performance testing, design characteristics, labeling, clinical data, postmarket surveillance, patient registries, or guidelines.

The controls should be measurable. “Device must be accurate” is too vague. A useful control identifies the kind of analytical or clinical performance, population, interfering conditions, or validation needed.

FDA may adopt controls different from those proposed. Once established in the classification regulation and associated documentation, they govern future devices of the type. A later 510(k) submitter must demonstrate substantial equivalence to a predicate and comply with applicable special controls. The existence of a predicate does not excuse new performance testing when technology, indications, or risk questions differ.

Evidence is tailored to uncertainty and consequence#

De Novo has no single clinical-trial template. FDA considers the device's risk, novelty, intended use, failure consequences, available nonclinical evidence, and uncertainty.

Bench evidence may be sufficient for some devices. Others need prospective clinical performance, diagnostic accuracy, usability studies, or comparative outcomes. An algorithm can require representative independent test data, locked-model evaluation, subgroup analysis, missing-input handling, and evaluation within the intended workflow.

Diagnostic evidence should define the target condition and reference standard, spectrum of samples, sites, readers, invalid results, and uncertainty. A high summary metric does not repair a biased sample or weak reference.

Human-factors validation tests whether intended users can perform critical tasks under realistic conditions. Training used in the study must match proposed labeling. Rescue by study staff can conceal use errors that would occur after launch. Cybersecurity and software documentation scale with risk and architecture. Connected devices need threat modeling, update processes, authentication, data integrity, and a plan for supported components.

The acceptance review is not the substantive decision#

FDA first examines whether the request contains the information needed for substantive review. The 2021 acceptance-review guidance describes an acceptance checklist and process aligned with the De Novo final rule.

An accepted request has crossed an administrative completeness gate. It has not been granted. During substantive review, FDA may ask for additional information, discuss revisions, inspect facilities, or place the request on hold under applicable procedures.

An incomplete request can receive a refuse-to-accept decision. A sponsor can address deficiencies and resubmit. Good pre-submission planning reduces preventable cycles, especially when the novel risk-control strategy needs early agreement. Watch for a company that announces “accepted for review” as though FDA had reached a conclusion about safety and effectiveness. It has not.

eSTAR is now the default submission format#

FDA's electronic Submission Template and Resource, eSTAR, guides content and validation for several device submission types. According to FDA's current De Novo request page, De Novo requests submitted on or after October 1, 2025 generally must use eSTAR unless an exemption described by FDA applies.

The template is a submission container and interactive guide. It does not choose the correct intended use, prove performance, or design special controls. A technically validated file can still contain an inadequate scientific argument.

Electronic requirements can change. Use the current FDA template, portal instructions, fee information, and guidance at the time of filing. An archived checklist is not good enough.

FDA's possible final actions#

FDA may grant or decline a De Novo request. It may also consider a request withdrawn in situations described by regulation, including failure to provide requested information within the applicable period or refusal to permit an authorized inspection.

FDA grants when the evidence supports class I or II and the relevant controls can provide reasonable assurance of safety and effectiveness, with probable benefits outweighing probable risks. The grant letter identifies the device and classification.

A decline can result when the device does not meet class I or II criteria, information is insufficient, proposed controls are inadequate, data are unreliable, benefit-risk is unfavorable, or other statutory or regulatory grounds apply.

A decline does not mean every future version is impossible. It means the submitted request did not support the requested classification. Redesign, new evidence, a narrower use, or a different pathway may be needed.

What a grant creates#

A grant authorizes the specific device to be marketed subject to applicable requirements. FDA establishes a new classification regulation for the device type, assigns a class and product code, and identifies special controls when class II.

The granted device can serve as a predicate for future 510(k) submissions when appropriate. This is one reason the first classification has system-wide importance: it defines the comparison and control framework for later products.

FDA publishes a decision summary and lists the request in its De Novo database. These records can reveal the evidence, risks, and controls behind the decision.

The authorization is bounded by the label. It does not establish safety or effectiveness for an unreviewed population, setting, output, or clinical claim. It also does not guarantee superior outcomes compared with existing care.

De Novo is not PMA “approval”#

Regulatory verbs communicate pathway. FDA generally describes a successful De Novo as “granted” and the device as “authorized.” A 510(k) receives “clearance.” A premarket approval application receives “approval.”

The pathways differ in statutory questions. A 510(k) asks substantial equivalence to a predicate. De Novo classifies a novel type into class I or II and creates controls. PMA is the class III premarket approval route and requires valid scientific evidence supporting safety and effectiveness.

Saying “FDA approved” for every device obscures these differences and may mislead readers about the review. The most accurate statement names the pathway, decision, indications, and date. Authorization is meaningful. Precision about the verb does not diminish the review; it describes what FDA actually decided.

De Novo versus other device pathways#

510(k)#

A 510(k) is appropriate when a device can be shown substantially equivalent to a legally marketed predicate. The comparison addresses intended use and technological characteristics, including whether differences raise different questions of safety and effectiveness.

Premarket approval#

PMA applies principally to class III devices. It evaluates whether valid scientific evidence provides reasonable assurance of safety and effectiveness for the proposed use. A device that needs class III controls is not made suitable for De Novo merely by being novel.

Humanitarian device exemption#

An HDE is designed for devices intended to benefit people with diseases or conditions affecting not more than the statutory population limit per year in the United States. It has distinct requirements, including probable benefit relative to risk and institutional oversight provisions.

Emergency use authorization#

An EUA is a temporary statutory mechanism available during specified emergency declarations when criteria are met. It is not a device classification and does not create an ordinary predicate.

Investigational device exemption#

An IDE permits a device to be studied in a clinical investigation under applicable requirements. It is not marketing authorization. A study may support a later De Novo, 510(k), or PMA.

Breakthrough designation is not a marketing pathway#

The Breakthrough Devices Program can provide interactive and prioritized FDA review for qualifying technologies that may offer more effective treatment or diagnosis of life-threatening or irreversibly debilitating conditions and meet another statutory criterion.

Designation does not authorize marketing, predetermine the classification, or guarantee a positive decision. A Breakthrough device still needs the appropriate marketing submission, which may be De Novo, 510(k), or PMA. Likewise, Safer Technologies Program participation and Q-Submission feedback are not authorizations. Public descriptions should distinguish process milestones from final decisions.

General controls continue after grant#

Class I and II devices remain subject to applicable general controls. These can include establishment registration, device listing, labeling requirements, medical device reporting, corrections and removals, records, and restrictions against adulteration and misbranding.

Manufacturers must comply with applicable quality-system requirements. FDA's Quality Management System Regulation became effective in 2026, aligning the device current good manufacturing practice framework more closely with ISO 13485 while retaining FDA authorities and requirements.

Unique device identification, tracking, postmarket surveillance, and other obligations may apply depending on device and regulation. Special controls add to, rather than replace, the general framework. A successful premarket review is therefore the start of a product life cycle, not the end of oversight.

Changes after authorization need regulatory analysis#

Devices change through software updates, new materials, manufacturing modifications, cybersecurity patches, model retraining, new users, and expanded claims. The manufacturer must assess whether a change requires a new marketing submission and maintain appropriate design and risk records.

For AI-enabled device software, a predetermined change control plan may prospectively describe certain planned modifications and their validation under current FDA guidance. Such a plan is reviewed as part of the marketing submission; it is not permission for unrestricted self-modification.

A new intended use or change that raises different safety and effectiveness questions can move beyond the original authorization. Postmarket data can also identify risks requiring labeling changes, correction, removal, or other action. The device being sold and used should still be the device FDA evaluated, apart from changes managed under the applicable framework.

Reading a De Novo decision summary#

Start with indications for use. Identify the user, patient population, specimen or input, setting, output, and intended clinical action. Then examine device description and technological characteristics.

Read the risks-to-health table beside the special controls. That pairing is where FDA shows you its regulatory logic. Review performance sections for sample size, sites, comparator, reference, endpoints, usability, invalid results, and limitations.

Check whether evidence was prospective or retrospective and whether it measured accuracy, workflow, or patient outcomes. Look for subgroup uncertainty and excluded situations. A statistically strong result can coexist with a narrow label.

Finally, read labeling and decision date. Later safety communications, recalls, database updates, or supplements may change the current picture. The original summary is necessary, but it is not always enough to support a claim you want to make in the present tense.

Common claims that go too far#

“FDA registered” usually means an establishment or device listing requirement was completed, not that FDA reviewed safety and effectiveness. “FDA compliant” is too vague without naming the requirement.

“De Novo approved” uses the wrong pathway verb. “Clinically validated” can hide the study population, endpoint, and comparator. “FDA authorized AI” can imply that every future model version was reviewed when only one version or change plan was.

“Creates a new category” can be accurate in a regulatory sense, but it does not prove first invention in science or commercial superiority. “Can be used as a predicate” does not mean every later device will be equivalent.

If you are writing the claim, give the grant number, the date, the indications, the device version where that matters, the classification, and the limitations that came with it.

A sponsor-facing readiness checklist#

Before you file, your team should be able to answer:

1. Why is there no suitable legally marketed predicate?
2. What exact device type and intended use are proposed?
3. Why can general or special controls manage every material risk?
4. Does each risk map to design, testing, labeling, or another control?
5. Are performance data representative of intended users, inputs, and settings?
6. Are software, cybersecurity, human factors, manufacturing, and shelf-life evidence complete where relevant?
7. Does the label match the tested conditions and known limitations?
8. Are proposed special controls clear enough to govern later devices?
9. Does the eSTAR package pass technical validation and scientific consistency review?
10. Is there a postmarket monitoring and change-management plan?

The hardest questions are usually classification logic and evidence fit, not file assembly.

The classification-centered conclusion#

The De Novo pathway solves a specific regulatory problem. A device is new enough that no suitable predicate exists, yet its risks may be manageable under class I or II controls. FDA reviews the evidence, benefit-risk balance, and proposed controls to decide whether to create that new device type.

A grant is both product-specific and structural. It authorizes the first device for its labeled use, establishes classification and controls, and may create a predicate for later 510(k)s.

The reading rule is simple. Name the pathway, the intended use, the evidence, the controls, and the decision accurately. “Novel,” “submitted,” “accepted,” “designated,” “granted,” “cleared,” and “approved” all describe different facts, and only one of them is the one you are being told about.

References#

• FDA. De Novo Classification Request.
• Electronic Code of Federal Regulations. 21 CFR part 860, Medical Device Classification Procedures.
• United States Code. 21 U.S.C. section 360c, device classification.
• FDA. Acceptance review for De Novo classification requests. 2021.
• FDA. Electronic submission template for medical device De Novo requests.
• FDA. De Novo database.
• FDA. Benefit-risk factors for medical device premarket review.