A 510(k) usually does not need clinical data, because bench, software, biocompatibility and animal testing are typically enough to show substantial equivalence. FDA asks for clinical data when it cannot judge equivalence without it, mainly when your indications or technology differ from the predicate, or when no bench or animal model can answer the question.
Key takeaways
- We reviewed the 3,070 public 510(k) summaries for devices FDA cleared in 2025. At least 17% described clinical data of some kind, and about 9% described a new clinical study in people or with patient specimens.
- FDA's 2023 draft guidance names four cases where clinical data may be needed: new indications, new technology, no good lab or animal model, and new risks with the predicate.
- Device type matters a lot. We found clinical data in 2.5% of orthopedic summaries, but a new clinical study in 21% of cardiovascular summaries, including 20 of 28 blood pressure monitors.
- Special 510(k)s for changes to your own device rarely included clinical data: 7%, against 19% for Traditional 510(k)s.
- Clinical data does not have to be a new trial. It can come from published studies, real-world data, or studies run outside the US, if they meet FDA's conditions.
Do most 510(k)s need clinical data?
No. FDA's 2023 draft guidance says that in many cases, strong non-clinical test data alone can show substantial equivalence. In those cases, FDA says, asking for clinical data would go against the law's least burdensome principle [1].
FDA has said this for years. In its 2014 guidance on the 510(k) Program, FDA wrote that it requested clinical data for less than 10 percent of the 510(k)s reviewed in its Office of Device Evaluation [2]. The same guidance noted one exception: 510(k)s reviewed in CBER for products that keep blood and blood products safe typically include clinical data [2].
When clinical data is part of your case, FDA's rules require your 510(k) summary to say so. It must briefly discuss the clinical tests you submitted, referenced or relied on, including, where applicable, the subjects and the safety and effectiveness results [3]. That rule is what made our 2025 analysis possible.
When does FDA ask for clinical data in a 510(k)?
When non-clinical evidence cannot settle the question. FDA may ask for it when bench, animal or other lab data are not enough. It may also ask when the test methods are not clinically validated or lack a sound scientific basis [1, 2].
FDA's 2023 draft guidance describes four scenarios. In each, clinical data may be needed, not must [1]:
| Scenario | When clinical data may be needed | When it is usually not needed |
|---|---|---|
| 1. Different indications for use | A higher-risk population, such as a different disease stage, or a new anatomical site with more risk | A new site that poses similar or lower risk than the predicate's |
| 2. Different technology | A significant change in materials, design, energy source or other features. FDA's example: a resorbable implant where similar devices are non-resorbable | New implant sizes inside the range of cleared sizes. A new maximum or minimum size may need clinical data |
| 3. Non-clinical testing cannot show equivalence | No bench or animal model exists, or it does not predict clinical outcomes. FDA's examples include hemostasis devices, aesthetic devices, and IVDs moving to point-of-care use | When a validated bench, analytical or animal model can answer the question |
| 4. New or increased risk with the predicate | Recalls, adverse events or literature show a new risk, and testing alone cannot show your device avoids it | When detailed non-clinical testing plus labeling can show the risk is controlled |
FDA also says the need for clinical data can change over time, as experience with a device type grows [1]. A requirement that applied to early devices of a type may fall away later, and the reverse can happen after safety problems emerge.
Clinical data cannot fix every difference. In scenario 2, clinical data helps only when the technological differences do not raise different questions of safety and effectiveness [1]. If they do, FDA finds the device not substantially equivalent. You may then need a PMA or, if appropriate, a De Novo request [2]. Our De Novo guide explains that route. Complizen's Regulatory Strategy service settles this before you build: it sets your pathway and predicates, and assesses which claims your current evidence supports and which need more.
Device-specific rules come first. The draft guidance does not override special controls or device-specific guidance [1]. Some special controls require clinical performance testing outright. For example, software that uses machine learning to flag possible heart conditions must be clinically validated on real-world data from a representative group of patients [6]. FDA calls this device type cardiovascular machine learning-based notification software.
How often did 2025 510(k)s include clinical data?
About one in five, and a new clinical study in about one in ten. We analyzed the 510(k) summaries of devices FDA cleared in 2025 [9, 10]:
| Measure | Count | Share |
|---|---|---|
| 510(k)s cleared in 2025 | 3,198 | |
| Public 510(k) summaries we screened | 3,070 | 100% |
| Described clinical data of any kind | at least 530 | 17% |
| Described a new clinical study in people or with patient specimens | at least 286 | 9% |
What we counted. We counted these as clinical data of any kind:
- a new clinical study
- tests on existing patient data, such as images or recordings
- sample clinical images taken with the device
- earlier clinical studies or published literature that the 510(k) relied on
Human factors testing did not count, because FDA does not treat it as clinical data [1]. Neither did skin patch tests run for biocompatibility.
How we did it. We downloaded every 2025 clearance from openFDA, extracted the text of each public 510(k) summary, and screened for clinical data with keyword rules. Every summary the screen flagged was then read and classified one by one against fixed rules, with spot checks. We also reviewed a random sample of 151 summaries the screen did not flag. Six of the 143 we could classify described clinical data, so the true share of all 2025 summaries is probably about one in five, between roughly 18% and 23%. The 128 clearances without a public summary, mostly 510(k) statements, are not included.
Two patterns stand out:
- Changes to your own device rarely need it. Special 510(k)s described clinical data in 31 of 462 summaries (7%), and only 3 described a new clinical study. Traditional 510(k)s did so in 477 of 2,547 (19%).
- A CLIA waiver changes the picture. All 20 Dual 510(k) and CLIA Waiver by Application submissions described clinical data.
These figures come from what applicants wrote in their summaries. Summaries vary in detail, so treat the counts as a floor, not an exact measure.
Which device types usually need clinical data?
Diagnostics and devices that measure the body. FDA groups devices by review panel, and the differences between panels are large [9, 10]:
| Review panel | 2025 summaries | Any clinical data | New clinical study |
|---|---|---|---|
| Radiology | 475 | 164 (35%) | 11 (2%) |
| Orthopedic | 401 | 10 (2.5%) | 4 (1%) |
| General and plastic surgery | 364 | 28 (8%) | 21 (6%) |
| Cardiovascular | 347 | 93 (27%) | 73 (21%) |
| Gastroenterology and urology | 256 | 20 (8%) | 11 (4%) |
| Dental | 215 | 13 (6%) | 2 (1%) |
| Neurology | 201 | 26 (13%) | 12 (6%) |
| General hospital | 178 | 8 (4.5%) | 7 (4%) |
| Anesthesiology | 99 | 20 (20%) | 15 (15%) |
| Microbiology | 97 | 73 (75%) | 65 (67%) |
| Physical medicine | 92 | 1 (1%) | 1 (1%) |
Counts are what we found, so they are minimums. What sits behind the numbers:
- Radiology is mostly software tested on existing images. Of the 164 radiology summaries with clinical data, 118 described testing software on existing patient data, and 33 described sample clinical images. Only 11 described a new clinical study.
- Measuring devices usually test on people. Twenty of the 28 cleared non-invasive blood pressure monitors described a clinical validation in people, and so did 14 of the 15 oximeters [9, 10]. FDA's 2013 guidance for pulse oximeters recommends clinical studies to determine SpO2 accuracy [7]. A January 2025 draft would update its testing recommendations, partly because accuracy can be affected by skin pigmentation [8].
- IVDs rely on patient specimens. Two thirds of microbiology summaries (65 of 97) described a clinical performance study with patient specimens.
- Implants and hardware rarely do. None of the 54 bone fixation plate summaries and none of the 39 dental implant abutment summaries described clinical data [9, 10].
The same pattern shows at the level of individual device types [9, 10]:
| Device type (product code) | 2025 summaries | Any clinical data | New clinical study |
|---|---|---|---|
| Non-invasive blood pressure monitors (DXN) | 28 | 23 | 20 |
| Oximeters (DQA) | 15 | 14 | 14 |
| Multi-analyte respiratory virus antigen tests (SCA) | 8 | 8 | 8 |
| Automated radiological image processing software (QIH) | 93 | 34 | 1 |
| Lumbar intervertebral fusion devices with bone graft (MAX) | 28 | 1 | 0 |
| Bone fixation plates (HRS) | 54 | 0 | 0 |
| Dental implant abutments (NHA) | 39 | 0 | 0 |
If your device is in a low-rate group, your 510(k) will most likely rest on bench and other non-clinical testing. Our performance testing guide covers that side. If it is in a high-rate group, plan for clinical data from the start.
What counts as clinical data?
More than a new trial. FDA's draft guidance lists sources that may count, drawn from an international regulators' definition [1]:
- clinical investigations of your device, before or after it reaches the market
- clinical studies of a comparable device, published in the scientific literature
- published or unpublished reports of clinical experience with your device or a comparable one
- registries, adverse event databases and medical records, which are real-world data
Whatever the source, it should be valid scientific evidence under 21 CFR 860.7(c)(2) [1, 11]. If you use data from a comparable device, such as an earlier version, you need a justification for why it represents your new device. You may also need non-clinical data to show the devices are comparable [1].
Two things do not count. FDA says human factors testing is not clinical data [1]. Animal studies are non-clinical testing [1].
Can you use clinical data from outside the US?
Yes, under conditions. FDA accepts data from a clinical investigation conducted outside the US to support a 510(k) if the investigation is well designed and well conducted, and [4]:
- you provide a statement that it followed good clinical practice, which includes ethics committee review and informed consent
- you describe what you did to ensure it met good clinical practice, such as the investigators, their sites and qualifications, and a summary of the protocol and results
- you show the device studied is identical to your device, or explain how it differs
- FDA can validate the data through an onsite inspection or other means, if it decides to
This matters for manufacturers who already hold CE marking or another approval. Our guide to using CE-mark clinical data in a 510(k) explains what FDA checks.
If you run a new study in the US, the investigational device exemption (IDE) rules apply. For a study that is not significant risk, you can follow lighter rules, including approval from an ethics board (IRB). FDA then treats the study as having an approved IDE [5]. A significant risk study needs FDA to approve an IDE application first. Our IDE guide covers both, and our guide to clinical trial costs helps you budget one.
How do you find out early whether your 510(k) needs clinical data?
Check four sources, in this order:
- Your device's special controls and device-specific guidance. These take priority over the general scenarios [1]. If they require clinical performance testing, the question is settled. Complizen's Device Classification tool returns your product code with every applicable special control, its 21 CFR citation, and the FDA guidance for your device type.
- Your predicate's 510(k) summary. It must discuss any clinical tests the predicate relied on [3]. Search FDA's 510(k) database for the predicate's K number, open the summary, and find the section on clinical tests, often titled "Non-Clinical and/or Clinical Tests Summary" or "Clinical Testing" [10]. If your predicate relied on clinical data, plan for FDA to ask whether you need it too. Our guide to choosing a predicate explains how to find and compare them.
- The four scenarios. Compare your indications and technology with the predicate's honestly. A higher-risk population, a riskier anatomical site and a significant change in materials are among the triggers FDA lists [1].
- A Pre-Submission. FDA's draft guidance recommends a Pre-Sub to discuss how its recommendations apply to your device [1]. FDA's feedback is most effective before you run your planned testing [12]. Our Q-Submission hub explains how to request one.
If FDA asks for clinical data during review, the request usually comes in an Additional Information letter. Our guide to answering FDA's additional information requests covers your options.
The order matters. Finding out late can mean designing a clinical study after your bench testing is done. Finding out at the planning stage lets you design testing around it, or narrow your claims so non-clinical data can support them.
Frequently asked questions
Does a 510(k) need clinical data?
Usually not. Most 510(k)s show substantial equivalence with bench, software, biocompatibility and animal testing. In our review of 3,070 public summaries for 2025 clearances, at least 17% described clinical data of some kind, and 9% described a new clinical study in people or with patient specimens.
What percentage of 510(k)s include clinical data?
In its 2014 guidance, FDA said it requested clinical data for less than 10 percent of 510(k)s in its Office of Device Evaluation. Our 2025 review found new clinical studies in 9% of public summaries, and clinical data of any kind, including retrospective data and literature, in at least 17%.
When does FDA require clinical data for a 510(k)?
FDA's 2023 draft guidance names four cases: new indications for use, new technology, no good bench or animal model, and a new or increased risk with the predicate. Special controls or device-specific guidance can also require clinical performance testing for a device type.
Do Class II devices need clinical trials?
Most do not. Class II devices usually reach the market through a 510(k), and most 510(k)s rest on non-clinical testing. Some types usually include it: in 2025, most blood pressure monitors and oximeters described clinical validation in people, and FDA's pulse oximeter guidance recommends clinical studies.
Does a Special 510(k) need clinical data?
Rarely. Special 510(k)s cover changes to your own cleared device. In our review of 2025 clearances, 31 of 462 Special 510(k) summaries described clinical data, about 7%. Only 3 of them described a new clinical study. Traditional 510(k)s described clinical data in about 19% of summaries.
Can I use published literature as clinical data in a 510(k)?
Yes. FDA's draft guidance lists clinical studies of a comparable device reported in the scientific literature as a source of clinical data. The data should be valid scientific evidence, and you need to justify why results from another device apply to yours.
Is human factors testing clinical data?
No. FDA's 2023 draft guidance states that data from human factors testing is not considered clinical data for its purposes. Human factors validation may still be required for your device, but it does not answer the question of whether FDA needs clinical performance data.
Can I use clinical data from a study outside the US?
Yes, if the study was well designed, well run and followed good clinical practice. Under 21 CFR 812.28, you describe how you ensured that. You also name the investigators and sites and summarize the protocol and results. FDA must be able to check the data.
Do I need an IDE for a 510(k) clinical study?
It depends on risk. For a study that is not significant risk, you can follow lighter rules, including approval from an ethics board (IRB), and FDA treats it as having an approved IDE. A significant risk study needs an FDA-approved IDE before it starts. Data from studies outside the US are accepted under 21 CFR 812.28.
Do software and AI devices need clinical data?
Often, but usually not a new trial. In our review, radiology software mostly described testing on existing patient images. Some special controls require it. For example, machine learning software that flags possible heart conditions must be clinically validated on real-world patient data.
How do I know if my device needs clinical data?
Check your device's special controls and FDA guidance, then your predicate's 510(k) summary. Compare your indications and technology against FDA's four scenarios. If the answer is still unclear, FDA's draft guidance recommends asking the review division through a Pre-Submission before you plan testing.
Can FDA ask for clinical data after I submit my 510(k)?
Yes. FDA can ask for clinical data during review if non-clinical data cannot show substantial equivalence. If it asks because of a new or increased risk with the predicate, FDA's draft guidance says it intends to explain why it needs the data, in line with its least burdensome principles.
References
- FDA. Recommendations for the Use of Clinical Data in Premarket Notification (510(k)) Submissions. Draft guidance, September 7, 2023.
- FDA. The 510(k) Program: Evaluating Substantial Equivalence in Premarket Notifications (510(k)). Final guidance, July 28, 2014.
- eCFR. 21 CFR 807.92, Content and format of a 510(k) summary. Up to date as of October 7, 2026.
- eCFR. 21 CFR 812.28, Acceptance of data from clinical investigations conducted outside the United States. Up to date as of October 7, 2026.
- eCFR. 21 CFR 812.2, Applicability. Up to date as of October 7, 2026.
- eCFR. 21 CFR 870.2380, Cardiovascular machine learning-based notification software. Up to date as of October 7, 2026.
- FDA. Pulse Oximeters: Premarket Notification Submissions (510(k)s). Final guidance, March 4, 2013.
- FDA. Pulse Oximeters for Medical Purposes: Non-Clinical and Clinical Performance Testing, Labeling, and Premarket Submission Recommendations. Draft guidance, January 2025.
- openFDA. 510(k) API, decisions dated January 1 to December 31, 2025. Accessed October 9, 2026.
- FDA. 510(k) Premarket Notification database, 510(k) summaries for 2025 clearances. Accessed October 9, 2026.
- eCFR. 21 CFR 860.7, Determination of safety and effectiveness. Up to date as of October 7, 2026.
- FDA. Requests for Feedback and Meetings for Medical Device Submissions: The Q-Submission Program. Final guidance, May 29, 2025.