Monitoring and Updating Clinical AI

#Performance needs attention after deployment

IMDRF includes appropriate ongoing monitoring in real-world use, with a risk-based focus on maintained or improved safety and performance. The transparency principles also describe communicating performance monitoring, detected issues and risks across a device's life cycle.

Monitoring is not only a question of whether software is available. Read what is checked about performance, the intended population and the way the device is used. These principles do not establish a universal monitoring interval: the task, risks and evidence determine what a particular program needs to examine.

Evidence: IMDRF: good machine learning practice principles (2025) / FDA, Health Canada and MHRA: device transparency principles

#Investigate changes with clinical context

Finlayson and colleagues describe dataset shift arising from changes in technology, populations, settings and behavior. They argue for combining clinician concerns and technical oversight when investigating whether a model is still functioning reliably.

A concern needs a route to investigation and feedback; a data-change signal is not a complete account of its clinical consequences. Read how the system's users can report concerns and what evidence is used to assess performance. A historical evaluation cannot answer every question after the population or workflow has changed.

Evidence: Finlayson and colleagues: the clinician and dataset shift / IMDRF: good machine learning practice principles (2025)

#Updating is a controlled change, not a promise

IMDRF addresses controls for retraining risks such as overfitting, unintended bias and performance degradation. The transparency principles include communication of modifications and updates, known limitations and ongoing performance. An update should not be described as an automatic improvement simply because it uses newer data.

Ask what changed, why it changed, how the revised system was evaluated and which users were told. Keeping those questions connected helps distinguish evidence about the old system from evidence about the revised one. This page gives reading principles, not a technical or clinical update procedure.

Evidence: IMDRF: good machine learning practice principles (2025) / FDA, Health Canada and MHRA: device transparency principles

#A change plan has a specific regulatory scope

In its United States guidance, FDA describes a predetermined change control plan, or PCCP, for AI-enabled device software. The plan includes specified modifications, the methodology for developing, validating and implementing them, and an assessment of their impact. FDA reviews the plan within the device's marketing submission.

An authorized plan is not blanket permission for arbitrary changes, and this framework does not establish authorization in another jurisdiction. The guidance describes recommendations, not a universal approval checklist. No Mynd device, authorized PCCP or live update service is claimed here.

Evidence: FDA: predetermined change control plan guidance (2025)

Source note

The sections above were checked against the linked sources. No clinical review has been performed. This is general research education, not a clinical guideline.