NurseVetted

NurseVetted submits this comment on Docket No. FDA-2026-N-7874 for the organization. The full comment is attached.
The discussion paper addresses clinicians and uses physician training as the model for premarket competency assessment. Nurses are the clinicians who will see many of these outputs, notice when they are wrong, explain them to patients, and decide what happens next. A framework that does not name that user will understate the risk.
We ask three things. Treat the nurse as a distinct user, not a subset of "healthcare professional." Require evidence, before clearance, that a nurse can use the output, override it, and see what the system recorded. After clearance, accept floor reports that do not have to pass through the manufacturer.
NurseVetted is not asking FDA to adopt a private badge. We are asking that the framework name the nurse as a user, test the device against that use before it is cleared, and watch that use after it is live.

Respectfully submitted,
NurseVetted
https://www.nursevetted.com

NurseVetted
Comment on Docket No. FDA-2026-N-7874
Considerations for the Regulation of Generative AI-Enabled Medical Devices: Discussion Paper
and Request for Feedback

Submitted electronically via Regulations.gov
Date: October 6, 2026

NurseVetted is a nurse-led organization that reviews healthcare AI tools against a fixed set of
nursing domains, before a tool is deployed and again after it is in use. This comment is filed for
the organization. It does not claim a completed national survey. It answers the questions on
which bedside nursing practice gives us direct knowledge. We do not answer every question in
the paper.

Summary
The discussion paper addresses clinicians and uses physician training as the model for premarket
competency assessment. Nurses are the clinicians who will see many of these outputs, notice
when they are wrong, explain them to patients, and decide what happens next. A framework that
does not name that user will understate risk and overstate the value of a benchmark.

Our request is narrow. Treat the nurse as a distinct user, not a subset of "healthcare professional."
Require evidence, before clearance, that a nurse can use the output, override it, and see what the
system recorded. After clearance, accept floor reports that do not have to pass through the
manufacturer.

1. Risk framework
The two-axis idea, how independently the function acts and the consequence of relying
on a wrong output, is a workable start. It is not sufficient for nursing use.

Add three dimensions.

First, who must act on the output. A suggestion read by an attending physician, a suggestion that
lands in a nurse's worklist at 2 a.m., and a message sent to a patient are not the same risk. Nurses
are already the user for drafted documentation, drafted patient messages, discharge instructions,
and shift handoff summaries. The paper addresses patients and clinicians but does not name the
nurse as a distinct user.

Second, whether the user can override the output without penalty or extra labor that makes the
override impractical. An output that is easy to distrust and hard to refuse will be followed. That
is a risk property of the device, not a training failure.

Third, whether the user can inspect what the system recorded about the patient and about the
clinician. A nurse who cannot see the basis for an output, or what was written into the record in
her name, cannot exercise the judgment the paper assumes a licensed professional will retain.
We agree with the paper that a disclaimer telling the user to consult a doctor does not make an
output less directive. The same holds for boilerplate that says "clinician retains responsibility." It
should not lower the risk score. Directiveness should be judged by what the nurse is expected to
do next.

2. Premarket evaluation
A competency assessment modeled on physician training is the wrong sole template.
Physicians and nurses do not receive the output in the same way, and they are not
accountable for the same next action.

Before a generative AI device function reaches patients, FDA should expect evidence on the
intended nursing user, not only on diagnostic accuracy:

• Can a nurse tell when the output is wrong, in the time available on a shift?
• Can she override it, and is that override recorded as a clinical decision rather than
noncompliance?
• Can she see the inputs and the text the system stored?
• Does the function add steps without removing the work it claims to replace?
• Was a nurse involved in defining the intended use, or only in a late usability session?

Non-clinical benchmarking can show that a model gives a dangerous answer. It cannot show that
the device fits a nursing workflow. Clinical confirmation should include the nurse who will
actually use the output, not only a physician reviewer. Where the paper proposes a panel of
qualified clinicians as the comparator for open-ended outputs, that panel should include nurses
for any function a nurse uses. A partial or abbreviated confirmation is reasonable. Skipping the
nursing user is not.

3. Postmarket monitoring
A model meets a premarket benchmark and still fails on a unit. Postmarket monitoring
that depends on the manufacturer, or on adverse events large enough to file, will miss the
ordinary failure: extra work, an override that is discouraged, an output the nurse no
longer trusts, a note the nurse did not write.

FDA should provide a path for a frontline clinician to report this kind of failure without going
through the vendor, and should expect manufacturers to review those reports. Monitoring should
include override rates, inspection of what the system recorded, and whether the claimed time
savings occurred. A change in the underlying model should trigger a look at those same
measures, not only a technical change-control file.

4. What this comment is not
NurseVetted is not asking FDA to adopt a private badge, and we are not asking the
agency to treat nursing judgment as a substitute for device evidence. We are asking that
the framework name the nurse as a user, test the device against that use before it is
cleared, and watch that use after it is live.
Respectfully submitted,
NurseVetted
https://www.nursevetted.com
Docket No. FDA-2026-N-7874