News
FDA finalizes Predetermined Change Control Plan guidance for AI/ML-enabled medical devices
The FDA released the final version of its Predetermined Change Control Plan (PCCP) guidance for AI/ML-enabled medical device functions on September 15, formalizing the framework that lets manufacturers pre-authorize model updates without a new 510(k) submission. The final guidance clarifies documentation expectations for the Modification Protocol and tightens the boundary between changes that fit within an approved PCCP and those that require a new submission.
- FDA
- PCCP
- regulatory
- AI-ML-device
- 510k
- SaMD
The final PCCP guidance closes a long-open regulatory question for AI/ML-enabled devices: how does a device that is designed to keep learning stay within its cleared indication as the underlying model evolves? The answer the FDA has landed on is the same conceptual answer the draft floated — manufacturers pre-specify the change space, the validation protocol, and the impact assessment, and the agency clears that plan alongside the device. What the final version does is make the documentation expectations concrete enough that manufacturers can build the internal quality processes to match.
The most operationally significant tightening is around the “impact assessment” section of the Modification Protocol. The final guidance is more specific about the evidence a manufacturer must generate to demonstrate that an in-scope modification does not degrade performance across the intended use population and subpopulations. That subpopulation performance monitoring — including demographic slices where model drift has historically been a source of harm — is now an explicit expected element of any PCCP-authorized retraining cycle.
For AI-enabled device manufacturers, the practical implication is that the PCCP is not a shortcut for avoiding regulatory work. It is a way to schedule that work predictably. Companies that already have MLOps discipline — versioned training data, reproducible validation pipelines, monitored production performance — will be able to file substantive PCCPs and iterate their devices on the schedule their engineering teams can sustain. Companies without that discipline will find that filing a credible PCCP is more work than filing periodic 510(k)s.
The final guidance also nudges the field toward standardization of the language used to describe the change space. Expect to see PCCP filings converge on a common vocabulary over the next 12 months as the first wave of post-final-guidance submissions works through the Center for Devices and Radiological Health.
Related coverage: FDA AI-enabled devices topic.
Primary source: Read the full original on U.S. Food and Drug Administration ↗