Earlier this week, the United States Food & Drug Administration (“FDA” or “Agency”) released its long-awaited report. final tips (the “Board”) on predetermined change control plans (“PCCPs”) for devices that use artificial intelligence/machine learning (“AI/ML”) software. FDA’s stated goal in this guidance is to “provide a forward-thinking approach to promoting the development of safe and effective AI-based devices,” and it represents notable progress in the Agency’s struggle to keep up – or at least avoid being too far away. overtaken by the rapid pace of AI/ML innovation, as used in digital health technology.
A. The AI challenge
The FDA’s primary challenge in regulating the software functions of AI-enabled devices (“AI-DSF”) has been that the self-modifying nature of these functions simply does not fit within the long-standing framework of the ‘Agency for medical devices, including, more specifically, medical devices. device modifications. Since the Medical Device Amendments of 1976, the FDA has kept a relatively firm grip on post-market device modifications by requiring manufacturers of legally marketed devices that originally required 510(k) clearance, a classification of Novo or pre-market approval (“PMA”). to submit a supplemental 510(k) notification, De Novo request and/or request for approval for post-market modifications that significantly change or modify the design, components, method of manufacture or intended use of the device (i.e., modifications “that could materially affect the safety or effectiveness of the device, or major changes or modifications in the intended use of the device”).(1) However, the regulation of AI-DSF, which are designed to constantly self-modify, clearly requires a different approach. The FDA has been publicly considering such an approach since at least 2019, when it issued its first discussion paper on the topic.(2) Now, nearly six years later, the FDA’s first final guidance is available .
B. New PCCP framework
The guidelines begin by acknowledging that the development of the AI-DSF is an “iterative process” and that its new PCCP framework is designed to provide “reasonable assurance” of safety and effectiveness. Rather than requiring AI-DSF manufacturers to receive authorization and/or approval for significant modifications After a device is legally marketed, the FDA will require AI-DSF manufacturers to include a PCCP with the initial marketing submission (i.e., 510(k) notification, De Novo application, or PMA application). As part of the premarket review process, the FDA will review the PCCP “to ensure the continued safety and effectiveness of the device without requiring additional marketing submissions to implement each change described in the PCCP.” . Rather than reviewing each change as it occurs – which would be impossible to monitor and evaluate – the FDA will review human-defined goals (i.e. algorithms) intended to guide the continuous self-modifications of technology.
The guide outlines the FDA’s recommendations on content a marketing submission for a PCCP AI-DSF, which generally includes: (i) a detailed description of the specific and planned modifications to the device (i.e., the “Description of Modifications”); (ii) the associated methodology for developing, validating and implementing these modifications in a manner that ensures the continued safety and effectiveness of the device in the intended use populations (i.e. the “Modification Protocol” ); and (iii) assessing the benefits and risks of planned changes and risk mitigation (i.e. “impact assessment”). Over 49 pages, the guide gives extensive details on each of the three elements of a PCCP (i.e. description of modifications, modification protocol and impact assessment), but warns that manufacturers should not rely on the guide for a “full analysis”. description of what may be necessary to include in a marketing submission for an AI-DSF” – that is, FDA is careful to clarify that the PCCP is only one element of a submission marketing plan for an AI-DSF.(3)
The PCCP requirement outlined in the guidelines is broad and applies to both automatic AI-DSF changes (i.e., changes implemented automatically by software, also called “continuous learning”), and manual AI-DSF Changes (i.e., changes involving steps that require human input, action, review and/or decision-making and are therefore not implemented automatically). Although the guidelines do not explicitly say so, they are presumed to apply to everyone. future marketing submissions because they do not include any indication that the requirement would apply retroactively.
C. Looking to the future
There is no doubt that the fate of FDA policy, especially one that is implemented solely through informal guidance (rather than notice-and-comment rules), is a bit uncertain in the face of the new administration , who promised big changes. and is centered on a platform of deregulation. In what may have been an attempt to protect its new policy from being reversed during the next administration, the guidelines strive to detail all of the stakeholder input that FDA received and considered during construction. of the proposed framework between 2019 and 2024 – a move we typically see when the Agency issues final regulations, but is less common in informal guidance.
However, given that the FDA’s first discussion paper (addressing the need to regulate AI-DSF in digital health technologies) was issued under the previous Trump administration, coupled with the increased push to regulate AI across industries, this is a rare subset of the FDA’s regulatory framework. project that could enjoy broadly bipartisan support. It remains to be seen whether the next administration will commemorate FDA’s PCCP policy through formal regulation, or whether regulators and other key stakeholders will instead call on Congress to set AI-DSF policy. Regardless, the policy outlined in the new guidelines is valid, at least for now, and medical device manufacturers that use AI/ML should begin to incorporate the preparation of modification descriptions, protocols modification and impact assessments in their overall product development plans.
FOOTNOTES
(1) See 21 CFR 807.81(a)(3).
(3) See, for example., 21 CFR 807.87, 21 CFR 860.220, 21 CFR 814.20.