Katina Dorsey

These comments reflect my own views as a registered nurse and do not represent the views of my employer or any organization I am affiliated with.

I am a registered nurse with 27 years of experience in healthcare, including clinical nursing, case management, and payer-side clinical review. I am responding to FDA’s discussion paper, "Considerations for the Regulation of Generative AI-Enabled Medical Devices" (Docket No. FDA-2026-N-7874), and specifically to Questions 19 and 21 in Section VI. I write in support of a clear, defined role for practicing clinicians in post-market monitoring of generative AI-enabled devices.

My daily work is reviewing clinical documentation and judging whether it supports the care described. That work has shown me something a testing framework may not capture on its own. An AI tool can produce documentation that reads perfectly and still miss the one detail that changes the clinical picture. I have seen summaries that flattened the record, leaving out a single sentence that carried the nuance behind the actual decision. The same risk applies to any device function that summarizes a patient record for a clinician. Benchmark testing may not catch this, because the output can be fluent and plausible and still wrong about what mattered. Section V.C of the paper acknowledges that benchmarking may not fully establish how a device performs in real clinical use.

Post-market monitoring should not rest on manufacturer self-reporting and automated metrics alone. Section VI lists periodic sample-based clinician review as one possible approach, and I support it. I ask FDA to pair it with a structured channel for clinicians who use these tools in daily practice to report when a tool’s output drifted from sound clinical judgment, with protection for those who report. Manufacturers would remain responsible for monitoring their devices, as the paper states, and clinician reports would give them and FDA a signal that periodic sampling may not surface. Nurses are well placed to notice this drift early. We read what the tool produced and we see what the patient needed, so we are often the first to see the difference. FDA’s approach to postmarket monitoring should treat that position as a monitoring asset.

On competency, Section V of the paper draws its examples from physician training and credentialing. Nursing has its own long tradition of competency validation, from precepted practice to ongoing skills verification. If FDA models device competency on human clinician competency, nurses belong in that conversation, because we are among those who sit between the tool’s output and the patient. Section VI.C proposes using the premarket assessment as the baseline for later re-benchmarking, so the people who help define competency at the start shape what gets monitored afterward. I ask that qualified adjudicators include nurses.

I ask the FDA to name two elements of postmarket monitoring for generative AI-enabled devices: clinician review of real cases, and a protected channel for clinicians to report when a tool’s output drifts from sound clinical judgment. Nurses should be explicitly included in both. Thank you for the opportunity to comment.

Katina A. Dorsey, RN, CCM-R