New U.S. FDA Draft Guidance Outlines Path To Faster Modification of AI/ML-Enabled Devices

• Description of Modifications — should describe and provide rationale for the planned modifications and describe device characteristics and performance changes that would result from the implementation of such modifications.¹¹  FDA recommends including a limited number of specific changes that are able to be verified and validated, and each proposed modification should state whether it would be implemented automatically (i.e., by software) or manually (i.e., require human involvement).¹² 

• Modification Protocol — should describe “the verification and validation activities (including pre-defined acceptance criteria) that will support those modifications” and must be compliant with the quality system (QS) regulation under 21 CFR Part 820.¹³ This includes device design controls and production and process controls requirements (see 21 CFR 820.30 and 820.70) and the need for documentation of the changes and relevant approvals (see 21 CFR 820.181).¹⁴

• Impact Assessment — using an existing manufacturer’s quality system as a framework, should document the benefits and risks that the modifications would introduce and any mitigations.¹⁵ Documentation should 1) include a comparison of the device version with each modification with the device version without any modifications; 2) discuss benefits and risk for each modification, including risk of social harm; 3) “discuss how the activities proposed within the Modification Protocol continue to reasonably ensure the safety and effectiveness of the device”; and also discuss “4) how the implementation of one modification impacts the implementation of another, and 5) the collective impact of implementing all modifications.”¹⁶

Additional Considerations

 FDA encourages early engagement using the Q-submission process to obtain Agency feedback regarding a proposed PCCP, particularly when the ML-DSF is combination product or a “high-risk, life-sustaining, life-supporting, or implantable device.”¹⁷

 Another consideration relates specifically to developers of products with a predicate device. The statute and guidance essentially state that when making a substantial equivalence determination using a predicate device authorized with a PCCP, the comparison must use the predicate device version cleared or approved prior to changes made under the PCCP.¹⁸

Unsurprisingly, the draft guidance states that the FDA review division will determine “what evidence and information are required to support proposed modifications in a marketing submission.”¹⁹

Appendix A to the Guidance²⁰ includes example elements of Modification Protocol Components for ML-DSFs, and Appendix B to the Guidance²¹ includes examples that “illustrate different ML-DSF scenarios where a PCCP could be employed.”

Key Takeaway

Overall, this approach has the potential to save companies time and reduce uncertainty around the need for FDA authorization and the support required before making changes to devices with AI/ML. Nevertheless, there are extensive requirements described in the current draft that developers would have to fulfill. This draft guidance will need to be finalized and may change based on comments received.