Rebecca Mathias

Built-in user feedback and postmarket monitoring

Built-in user feedback and postmarket monitoring
Comments on FDA Discussion Paper | Docket FDA-2026-N-7874

Submitted by: Rebecca Mathias1, Stephen Gilbert1,2

Affiliations:
Else Kröner Fresenius Center for Digital Health, Faculty of Medicine, TUD Dresden University of
Technology, Dresden, Germany
Faculty of Business and Economics, TUD Dresden University of Technology, Dresden, Germany

We agree with CDRH's observation that the variable, open-ended outputs of GenAI-enabled devices,
together with their potential to change after deployment, make it difficult for premarket testing to fully
capture performance. Postmarket monitoring needs ways to identify problems as they arise during realworld use. The FDA discussion paper describes periodic re-benchmarking, sample-based clinician
review and performance degradation monitoring as possible approaches. These can provide important
information about device performance, but may not capture all problems arising in day-to-day use.
Patients and clinicians may identify misleading outputs, difficulties recognising errors or verifying
information, disruptions to workflows, unexpected changes in device behaviour, or near misses that are
not readily apparent from periodic technical assessments. They can also report positive experiences
and benefits that conventional incident reporting does not capture. However, much of this goes missed
because of the time and effort it takes to file a report using a separate channel.

In our previously published work, we propose transparent, mandatory feedback mechanisms built
directly into AI-enabled digital health technology interfaces (Mathias et al., 2025). Rather than requiring
users to find separate complaint or incident-reporting channels, devices could allow feedback at the
point of use through a simple mechanism combining ratings, structured responses and, where
appropriate, free-text reports. We also proposed linking these mechanisms through standardised
interfaces to a national or regulator-supervised platform. Ratings and reviews could be publicly
accessible, while sensitive underlying data could be restricted or aggregated. Measures such as
verified anonymity could encourage open reporting while protecting users.

The strength in numbers from collecting feedback across large numbers of users and devices is an
important part of this proposal. Individual reports may not show how widespread a problem is, but
aggregated feedback could reveal recurring problems or changes in user experiences that would
otherwise go unnoticed. Combining this feedback with structured signals, such as device-use metrics
or telemetry, could help identify these patterns. Defined thresholds could flag increases in reported
problems or other concerning changes and trigger investigation or corrective action. This could help
identify problems before they result in serious harm, although individual serious reports should not
have to reach a threshold before action is taken.

Collecting feedback is only part of the process. It also needs to be systematically analysed and acted
upon, with clear responsibilities, thresholds for action, and documented follow-up and mitigation.
Integrated feedback should complement existing adverse-event reporting, premarket assessment,
formal postmarket surveillance and other monitoring approaches, not replace them. Importantly,
feedback should cover the broader interaction with the device, including human review, workflow fit and
how outputs influence subsequent decisions. It cannot replace the evidence needed before market
entry, but could identify problems that were not apparent during premarket evaluation.
We encourage CDRH to consider built-in, accessible and transparent feedback as an additional
component of postmarket monitoring for GenAI-enabled medical devices, particularly in relation to
Sections VI.A and VI.B and discussion Questions 19 and 21. We propose working towards routine
integration and, ultimately, mandatory collection and appropriate sharing of feedback where feasible
and proportionate to the device and its context of use. The proposal concerns a wider system for
collecting, aggregating, analysing and sharing feedback from real-world use at the point and time of
use to identify problems early and support appropriate action.

Reference: U.S. Food and Drug Administration. Considerations for the Regulation of Generative AIEnabled Medical Devices: Discussion Paper and Request for Feedback. August 2026.
https://www.fda.gov/media/194242/download
Mathias R, Vasey B, Chalkidou A, et al. Safe AI-enabled digital health technologies need built-in open
feedback. Nature Medicine. 2025;31:370-375. https://doi.org/10.1038/s41591-024-03397-6

AI Assistance and Ongoing Research Disclosure
ChatGPT 5.6 (OpenAI) was used to help summarise our research work for the purposes of this
comment. The content was reviewed and edited by the authors of this comment, who remain
responsible for the final submission. The views expressed in this comment are those of the authors and
do not necessarily reflect those of their affiliated institutions.