Introduction
On December 4, 2024 [3] [5] [6], the FDA released final guidance on the content required for a Predetermined Change Control Plan (PCCP) in marketing submissions for modifications of AI-enabled medical devices [1] [5]. This guidance underscores the FDA’s commitment to fostering innovative regulatory approaches for AI-enabled devices [3], recognizing their iterative development process and the necessity for manufacturers to implement machine learning updates in a controlled and pre-approved manner.
Description
The guidance outlines essential concepts of PCCPs [6], allowing manufacturers to make specified post-market changes without needing additional FDA authorizations after obtaining 510(k) clearance, de novo authorization [6], or premarket approval [3] [6]. The FDA acknowledges the dynamic nature of AI devices [5], which continuously evolve through building [5], testing [2] [5], and learning [3] [4] [5]. Approved plans enable manufacturers to document significant changes to their devices that incorporate AI software functions, provided these modifications are well-documented in compliance with FDA requirements [5].
The final guidance broadens the scope to include a wider range of AI-enabled devices compared to the previous draft [5]. Key updates include a new section on version control and maintenance [5], enhanced content requirements for labeling [1], and detailed documentation requirements for modifications. Manufacturers must include a description of changes, a modification protocol detailing development and validation methods [1], and an impact assessment that identifies the benefits and risks associated with the modifications. This impact assessment is crucial for linking the description of modifications to the modification protocol [2], ensuring that verification and validation activities continue to guarantee the safety and effectiveness of the device [2].
To ensure that proposed modifications align with the expectations for a PCCP [5], manufacturers are encouraged to engage with the relevant FDA review division early in the pre-submission process [5]. Additionally, a webinar will be held on January 14, 2025 [1], to address questions regarding the final guidance [1].
The guidance outlines core types of modifications suitable for inclusion in the PCCP [2], including quantitative measures of AI-Driven Software Function (AI-DSF) performance specifications, device inputs and compatibility [2], and aspects of device use and performance [2]. Modifications that enhance AI-DSF compatible software/hardware and promote interoperability are deemed appropriate for the PCCP [2].
Manufacturers are also encouraged to consider additional questions regarding data management practices [2], retraining strategies [2], performance evaluation protocols [2], and update procedures for proposed modifications [2]. The new section on data management emphasizes the importance of segregating development datasets from test datasets [2], while the retraining practices section highlights the need to mitigate risks such as overfitting and bias [2]. The performance evaluation protocols section has been expanded to include specific triggers for evaluations and additional testing requirements [2], necessitating a more detailed performance evaluation plan for modification protocols [2].
The update procedures section includes expanded questions on validation in integrated environments with interoperable devices and addresses cybersecurity risk management [2], referencing FDA guidance on Cybersecurity in Medical Devices [2]. Appendix B has been updated with more detailed scenarios to clarify implementation processes [2], increasing the number of examples from four to six [2]. This suggests the FDA’s growing experience in approving AI-DSFs with PCCPs and a wider array of modification types for manufacturers to consider [2]. The scenarios also illustrate the consequences of omitting specific modifications from the PCCP [2], which may lead to the need for a new submission [2].
Distributing devices that do not comply with the PCCP is considered adulteration and misbranding [2], violating Section 301 of the FD&C Act [2] [3]. Establishing a new PCCP for an existing device typically requires a new marketing submission [2], as does creating a PCCP for a device that lacks one [2]. Significant modifications outside the PCCP also require a new marketing submission [2], and the original PCCP must be reviewed and established as part of this submission to remain valid [2]. This comprehensive approach ensures that any changes made to AI-enabled medical devices uphold their safety and effectiveness, facilitating the evolution of adaptive AI technologies while ensuring patient safety.
Conclusion
The FDA’s final guidance on PCCPs for AI-enabled medical devices represents a significant step towards accommodating the unique challenges posed by the iterative nature of AI technology. By allowing specified post-market changes without additional authorizations [1] [5], the guidance facilitates innovation while maintaining rigorous safety and effectiveness standards. This approach not only supports the advancement of adaptive AI technologies but also ensures that patient safety remains a top priority.
References
[1] https://www.lexology.com/library/detail.aspx?g=baa8af29-b07e-41f8-91f2-88c21e264419
[2] https://www.linkedin.com/pulse/from-draft-final-key-insights-fdas-guidance-pccps-madison-phd-mph-ubvhe
[3] https://clarkstonconsulting.com/insights/pccp-for-ai/
[4] https://clarimed.com/resources/blog/fda-s-new-guidance-is-a-game-changer-for-ai-medical-devices
[5] https://www.jdsupra.com/legalnews/fda-s-final-guidance-provides-practical-8003333/
[6] https://www.jdsupra.com/legalnews/fda-issues-final-guidance-on-4708484/
