On December 4, 2024, the U.S. Food and Drug Administration ("FDA") released its final guidance for industry on predetermined change control plans ("PCCPs") for devices that utilize artificial intelligence-enabled software functions (the "Final Guidance"). PCCPs enable device manufacturers to implement device modifications that have been pre-authorized by FDA as part of an initial marketing application instead of submitting a new marketing application, which may be particularly useful for AI-enabled devices designed to evolve over time. The Final Guidance details the information to include in a PCCP for AI-enabled device software functions ("AI-DSF") and the types of modifications to AI-DSF that are appropriate for inclusion in a PCCP.
The Final Guidance is largely consistent with and maintains the same structure and general policy approach as the April 2023 draft guidance on the subject (the "Draft Guidance," discussed in a previous Ropes & Gray Alert). Notable revisions in the Final Guidance include: expansion of the scope to all AI-enabled devices instead of just machine learning ("ML")-enabled devices; additional detail regarding the interplay between PCCPs and modifications that require new marketing submissions, content to include in the Description of Modifications section of a PCCP, and labeling considerations; and clarification regarding diversity considerations, intended use, and the applicability of PCCPs to drug-device combination products.
The Final Guidance builds upon FDA's efforts in recent years to develop and implement a regulatory framework tailored to AI-based software whose iterative, self-modifying nature does not fit neatly within FDA's existing framework for modifications to medical devices. The Final Guidance also continues FDA's development of PCCP policies, following the Agency's August 2024 draft guidance on PCCPs for medical devices generally (as described in our prior Ropes & Gray Alert).
This Alert summarizes the key updates in the Final Guidance as well as important takeaways for medical device manufacturers, digital health product developers, and other industry stakeholders.
Key Updates in the Final Guidance
Scope. Whereas the Draft Guidance focused specifically on ML, a subset of AI, the Final Guidance applies to all AI-enabled devices, theoretically sweeping in a broader array of digital health products. Despite this expanded scope, FDA notes in the Final Guidance that its recommendations are largely tailored to ML devices given that most submissions containing PCCPs reviewed by FDA have been for devices incorporating ML. In addition to expanding the scope of the Final Guidance to include all AI-enabled devices, the Final Guidance also adopts definitions of AI and ML that align with the Biden administration's 2023 Executive Order on the Safety, Secure, and Trustworthy Development and Use of Artificial Intelligence, which directed a coordinated federal approach to AI.1
Applicability to Combination Products. Comments from industry stakeholders on the Draft Guidance requested that FDA provide clarity regarding how PCCPs may be used for combination products. Stakeholders were particularly interested in guidance from FDA as to whether PCCPs could be part of a New Drug Application or Biologic License Application, and how PCCPs would be considered where the Center for Devices and Radiological Health ("CDRH") is not the lead review center. In response to these comments, FDA underscores that its statutory authority on PCCPs specifically cites the device review pathway, ostensibly implying that PCCPs would not be accepted for drug-led combination products. Additionally, FDA includes a new example PCCP scenario for a device-led combination product in the appendix to the Final Guidance: an imaging system that is co-packaged with an approved optical imaging drug.
No PCCPs Via Special 510(k)s. Device manufacturers advocated in comments to the Draft Guidance for FDA to permit inclusion of PCCPs in special 510(k)s. However, FDA makes clear in the Final Guidance that for devices for which a 510(k) is required, a PCCP may be authorized only through the traditional and abbreviated 510(k) pathways.
Modifications to Indications through PCCPs. In response to comments from industry requesting greater flexibility regarding inclusion in a PCCP of modifications impacting the intended use of a device, FDA has softened its approach somewhat in the Final Guidance. While FDA upholds its position in the Final Guidance that modifications under a PCCP should maintain the device's intended use (or allow it to remain substantially equivalent for a 510(k) cleared product), FDA acknowledges that some modifications to the indications for use, such as specifying the use of the device with an additional device or component, may be appropriate for inclusion in a PCCP.
Describing Proposed Modifications. The Final Guidance provides additional information regarding what to include in the Description of Modifications section of a PCCP. Specifically, the Final Guidance recommends that manufacturers clearly establish guardrails that define the range of automatic updates as well as information regarding the expected frequency of updates – ranging from periodic updates of primarily locked devices to continuously updated devices.
Addressing PCCP Modifications in New Marketing Submissions. The Final Guidance addresses how modifications made pursuant to an authorized PCCP will interplay with modifications not included in the PCCP that require a new marketing authorization. When submitting a new marketing submission for device modifications not in scope of a previously authorized PCCP, the manufacturer should summarize the modifications that have been implemented in accordance with the previously authorized PCCP so that FDA can understand the current device characteristics and performance.
Labeling Considerations. Requests for FDA to provide additional clarity on labeling implications were featured prominently in the comments submitted on the Draft Guidance. In response, FDA reiterates in the Final Guidance that labeling of an AI/ML-enabled device with a PCCP should clearly state that the device incorporates machine learning and has an authorized PCCP so that users understand that the device may require software updates that may modify its performance, inputs, or use. Additionally, the Final Guidance clarifies that labeling should be updated as modifications are implemented to include "relevant information" such as a description of which modifications were implemented; how the modifications were implemented; and how users will be informed of implemented modifications, such as through updated instructions for use. The Final Guidance does not address industry requests for FDA to clarify that labeling updates can be accomplished through electronic labeling and not only physical labeling; FDA is silent on this issue.
Public Notice Regarding PCCPs. FDA expands upon its recommendation in the Draft Guidance that publicly available device summaries, such as a 510(k) summary, De Novo decision summary, or PMA summary of safety and effectiveness document and approval order, contain a description of a device's authorized PCCP to support transparency. FDA further explains in the Final Guidance that publicly available device summaries should include, as appropriate, a description of planned modifications, testing methods, validation activities and performance requirements required for implementation of modifications, and how users will be informed of implementation of modifications specified in the PCCP.
Diversity Considerations. The Final Guidance highlights FDA's commitment to ensuring that AI-DSFs are developed with consideration of the needs of diverse populations. It "affirms that digital health technologies should be designed and targeted to meet the needs of diverse populations," as FDA explained in an announcement accompanying the release of the new guidance. The Final Guidance advises that when developing a PCCP, manufacturers "should consider the intended use populations (with respect to race, ethnicity, disease severity, gender, age, or others, as appropriate) and intended environments of use, so that the device continues to reflect these populations and environments as the device is modified."
Takeaways and Implications for Industry
Many of the considerations related to FDA's draft guidance on PCCPs for all medical devices, as discussed in our prior Alert, apply in the AI-DSF context as well. For example, there is ambiguity regarding the number and types of modifications appropriate for inclusion in a PCCP, but this is likely to become clearer as industry and FDA gain more experience with PCCPs. Additionally, obtaining authorization of a PCCP requires substantial upfront work, and it remains to be seen whether the upfront work required will result in modifications in fact being implemented more expeditiously. A few additional points are outlined below.
Implications for Manufacturers' Quality Management Systems. The Final Guidance does not address the interplay between PCCPs and device manufacturers' quality management systems. Because the Final Guidance is focused on marketing submission recommendations, it does not explain how design control requirements apply to modifications implemented pursuant to a PCCP. FDA's revised Quality Management System Regulation, as described in a previous Ropes & Gray Alert, is slated to take effect in February 2026. While manufacturers of AI/ML-enabled devices consider possible PCCP strategies, they should also ensure they have established systems and processes to ensure device modifications are implemented in accordance with the applicable quality system requirements.
Emphasis on the Q-Submission Process. FDA recommends that manufacturers utilize the Q-Submission process to seek feedback from FDA prior to a formal marketing submission on numerous topics, including inclusion of modifications to the indications of use in a PCCP as well as plans for automatic updates under a PCCP. FDA's insistence on using the Q-Submission process to provide individualized feedback to manufacturers on such key issues has the potential to result in inconsistency in FDA feedback for different manufacturers, and the emphasis on obtaining feedback via Q-Submissions might discourage some companies from pursuing PCCPs altogether.
Looking to 2025. Although it remains to be seen precisely what impact the Trump administration will have on the development of AI, the new administration is expected to rescind the Biden administration's Executive Order on AI and replace it with a policy that imposes fewer controls. Additionally, the new Trump administration may, in an effort to foster AI development, direct federal agencies to adopt a more hands-off approach to the regulation of AI. It is unclear whether such general shifts in AI policy will affect FDA's oversight of AI-enabled medical devices, given that FDA has already been pursuing policies that provide regulatory flexibility. For example, the PCCP concept, which was originally introduced by FDA in a 2019 discussion paper during the first Trump administration and then codified in the Food and Drug Administration Omnibus Reform Act of 2022, is intended to reduce regulatory burdens on industry.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.