Medical device ML — what evidence does FDA expect in a submission?
Informational only, not professional or legal advice. Facts are linked to the source we checked; statuses can change after the date above.
This is the one medical-device ML question with a public, section-by-section answer. FDA’s Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations is marked “Draft – Not for Implementation” and “Document issued on January 7, 2025.” It contains nonbinding recommendations.
What the draft asks for. Its table of contents lists these submission sections: Device Description; User Interface and Labeling; Risk Assessment; Data Management; Model Description and Development; Validation (including Performance Validation); Device Performance Monitoring; Cybersecurity; and Public Submission Summary. Appendix E is an “Example Model Card”.
The draft explains why each section matters. On data: “For an AI-enabled device, the model is part of the mechanism of action.” On validation: “The performance validation for an AI-enabled device provides objective evidence that the device performs predictably and reliably in the target population according to its intended use.” On monitoring: manufacturers “should have a postmarket performance monitoring plan”.
Planned model changes. The PCCP guidance says a plan generally includes a “Description of Modifications”, a “Modification Protocol” (“the associated methodology to develop, validate, and implement those modifications”) and an “Impact Assessment” of “the benefits and risks of the planned modifications”.
EU cross-reference. The AI Act Service Desk Annex I lists “Regulation (EU) 2017/745 of the European Parliament and of the Council of 5 April 2017 on medical devices” in Section A. The notified-body evidence for that route is governed by Regulation (EU) 2017/745; we could not retrieve the EUR-Lex text in this review and quote nothing from it.
Editorial judgment: For a US submission, build the evidence pack to the draft’s section list now, and treat it as draft: the final text may differ. For the EU, do not assume the FDA structure satisfies a notified body; we have not found an official sentence that equates the two.
Sources
- Artificial Intelligence-Enabled Device Software Functions: Lifecycle Management and Marketing Submission Recommendations - FDA
- Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions - FDA
- Regulation (EU) 2017/745 on medical devices - EUR-Lex
- AI Act Service Desk: Annex I - European Commission