Designing around hardware.
A supplement study and a device study look similar on a timeline and diverge nearly everywhere else. Hardware has a regulatory pathway of its own, a blinding problem pills do not have, and one advantage pills cannot match.
- ▸The first question is not about design. It is whether your device is the subject of the study or just the instrument, which decides whether the device regulations reach you at all.
- ▸If they do, one fork sets your timeline. Significant risk means an application to FDA. Non-significant risk means a written argument to your board, and you can begin.
- ▸You cannot make a light panel look like a placebo the way you can make a capsule look like a capsule. There are answers, and pretending otherwise is not one of them.
- ▸The dose is not fixed the way a capsule is. On a light device it varies with the participant's own biology, which changes screening, instructions and analysis.
This describes published FDA regulations and FTC guidance. It is not legal or regulatory advice.
Is your device the subject, or the instrument?
The device regulations apply when the objective of the investigation is to assess the safety or effectiveness of a device. Same hardware, opposite posture, decided entirely by what the study was designed to find out.
The device as instrument
A wearable recording sleep architecture in a supplement trial is a tool. It is collecting data in a study of something else, and the study is not a device study on that basis.
The device as subject
The same wearable, in a study designed to establish that the wearable improves sleep, is the investigational device. If you are a hardware brand studying your own product to support a claim about it, you are almost certainly here.
The fork that sets your timeline.
In scope, investigations sort into three categories, and which one you land in changes everything downstream.
Significant risk
Four prongs: an implant presenting potential for serious risk; a device represented as supporting or sustaining human life presenting serious risk; one for a use of substantial importance in diagnosing, curing, mitigating or treating disease presenting serious risk; or one that otherwise presents potential for serious risk.
Non-significant risk
Everything else in scope. Treated as having an approved application, unless FDA notifies otherwise, provided you label the device correctly, obtain board approval after presenting a brief explanation of why it is not a significant risk device, document consent, comply with monitoring, and keep the records.
Exempt
Some investigations fall outside the regulations entirely. Those categories are specific and worth reading in the regulation itself rather than in a summary, this page included.
The risk determination is an argument you write and submit, not a box you tick. A board that disagrees with your reasoning sends you to FDA.
What that argument actually looks like
In a twelve-week study of an at-home UVB light therapy device, the case for non-significant risk was not an assertion. It rested on things a board could check.
- ▸A non-invasive device operating within established wavelength and power ranges
- ▸Proximity sensors that cut power if a user gets too close
- ▸Automatic shut-off timers capping session length
- ▸A cool-touch exterior with fan-based temperature regulation
Those are engineering features, and they became the regulatory argument. If you are building hardware you intend to study, the safety design and the risk determination are the same conversation.
The blinding problem.
A capsule can be matched by an identical capsule. A light panel cannot be matched by a panel that emits nothing. Participants generally know whether the thing in front of them is switched on.
Participants and researchers blinded where feasible. And where a double-blind design is not feasible, blinding to the fullest extent possible with steps taken to minimize bias.
Improvement in a treatment group can come from the placebo effect, from spontaneous change, or from the practice effect on whatever you are measuring. The guidance names all three.
Sham device
Matched on everything except the active output.
Dose arms
Intensity levels instead of on versus off, so everyone receives something.
Blinded assessor
Whoever reads the outcome does not know the assignment.
Hard endpoints
Outcomes that expectation cannot easily move.
When none of those is available, the honest move is to scope the claim to the design. The UVB study is single-arm, with no randomization and no placebo, and the protocol says so directly. It commits in writing to analyzing outcomes descriptively with attention to the limits of a single-arm design, and frames the work as signal detection to inform later randomized studies. Not a disclaimer bolted on at the end. The study's stated purpose, written before enrollment.
Two things a capsule study never has to solve.
The dose is not fixed
A capsule delivers the same dose to everyone. On the UVB device, prescribed exposure ran from 30 seconds to 5 minutes depending on the participant's Fitzpatrick skin phototype, two to three times weekly across twelve weeks.
The dose is a function of the participant's own biology, not a constant in the protocol. Instructions have to be personalized. Screening has to capture the variable that sets the dose. And any analysis of exposure has to account for two fully compliant participants having received materially different amounts.
Screening is about the physics
A supplement study screens for things that interact with the ingredient. For the light device, exclusions covered a long list of photosensitivity conditions, medications known to increase photosensitivity across several drug classes, tanning bed or phototherapy use within the previous thirty days, and open wounds in areas the device would expose.
Every eligible participant then completed a phone screening call on top of the survey screening. That is a real recruitment cost, and it belongs in the feasibility maths before anyone commits to an enrollment target.
The device records its own use.
A capsule study asks people whether they took it. In the UVB study, participants logged each session in the app, the app issued automated reminders, and adherence was tracked in real time rather than reconstructed at the end.
Adherence stops being a self-report
It lets you separate a null result from a compliance failure, which decides whether the study was worth running. A product that does nothing and a product nobody used produce the same disappointing chart, and only one of them is a reason to stop.
Compliance defined in advance, in numbers
Not "good adherence" but use on at least 70% of prescribed days, plus completion of every scheduled survey and lab. Worth noting that this is a compliance threshold and not a completion rate. They are different measurements and should never be reported as though they were the same one.
The firmware problem
Consumer devices update. Manufacturers can revise firmware and the algorithms sitting on top of the raw signal at any point, and they are under no obligation to tell a research sponsor. If that happens mid-protocol, the measuring instrument changed partway through the measurement.
For a study using someone else's wearable as an instrument, that threatens the outcome data. For a study of your own device it is worse: you may have finished a study of a version you no longer sell.
What to settle before you commission one.
- ▸Is my device the subject of this study, or an instrument in it?
- ▸If it is the subject, what is my non-significant-risk argument, and would a board accept it?
- ▸What is my control condition, and if participants cannot be blinded, what am I doing instead?
- ▸Is the dose the same for everyone, or does it vary with the participant?
- ▸What does screening actually have to exclude, and what does that do to recruitment?
- ▸Can I capture usage directly, and will I be able to tell a null result from a compliance failure?
- ▸Will the firmware change during the study, and would I know if it did?
Where the answer is unclear and the stakes are commercial, that is a conversation for regulatory counsel rather than a page like this one.
Where we sit in this.
We run decentralized device studies in participants' homes, on their own schedules, with adherence logged rather than recalled.
If you have not settled what you are trying to prove, that is a different question. If you are choosing between a pilot and a full study, that is another one.
Written by the Reputable Health team. This article describes publicly available FDA regulations and FTC guidance and is not legal or regulatory advice.
- 21 CFR Part 812, Investigational Device Exemptions. U.S. Food and Drug Administration.
- Health Products Compliance Guidance. Federal Trade Commission, December 2022.