FDA issues draft guidance on pre-determined change control plans
FDA has published long-awaited draft guidance on predetermined change control plans (PCCPs) for artificial intelligence/machine learning (AI/ML)-enabled medical devices. This draft guidance is the latest development in the process to establish an updated regulatory framework that accounts for the iterative nature of device software.1
The draft guidance builds on feedback received in relation to FDA’s 2019 discussion paper on a proposed framework for AI/ML software as a medical device (SaMD) and fulfils FDA’s 2021 AI/ML-based SaMD action plan commitment to issue draft guidance for comment. Notably, the PCCP concept set out in the new draft guidance applies to all AI/ML-enabled device software functions (not just SaMD) and introduces new nomenclature and concepts.
Over the last few years, a number of manufacturers have been submitting PCCPs based on the limited information set out in FDA’s 2019 discussion paper and 2021 action plan. This new 43-page draft guidance is a significant step forward, providing much anticipated detail into FDA’s approach to PCCPs.
According to the draft guidance, the PCCP framework is designed to be used by manufacturers who intend to modify their ML-enabled device over time. The PCCP should be included as a standalone section within a marketing submission, and it should feature prominently in the submission cover letter. FDA will review a PCCP “as part of a marketing submission to ensure the continued safety and effectiveness of the device without necessitating additional marketing submissions for implementing each modification described in the PCCP.”
PCCPs are intended to reduce the regulatory burden on iterative ML-enabled device products by providing a means to prespecify and authorize certain modifications (that would otherwise require additional marketing submissions prior to implementation) and methods of implementation, allowing authorized changes to be made without further FDA involvement.
The draft guidance lists three general areas of modification that could be appropriate for a PCCP: (i) modifications related to quantitative measures of device performance specifications; (ii) modifications related to inputs to the device; and (iii) limited modifications related to the device’s use and performance (eg, for use within a specific subpopulation). However, the PCCP framework cannot be used, and a new submission is required, if the envisioned modification exceeds the device’s existing indications for use.
FDA can authorize some components of a proposed PCCP and not others. Once authorized, the PCCP is considered part of the device. Therefore, any deviation from the PCCP (ie, implementing modifications or using methods of implementation not included in the authorized PCCP) would cause a device to be adulterated or misbranded.
When modifications are undertaken in line with an authorized PCCP, the manufacturer should document the modification and associated analysis in accordance with the manufacturer’s quality system.
The PCCP framework can be applied to ML-enabled devices of any class as well as ML-enabled device components of combination products.
The majority of the draft guidance is given over to describing the components of a PCCP. FDA’s 2021 action plan SaMD Pre-Specifications (SPS) and Algorithm Change Protocol (ACP) terminology does not feature. According to the draft guidance, to provide FDA with the information needed to review the proposed modifications a PCCP should consist of the following:
- Description of Modifications, providing detailed information on all proposed software changes along with rationale for each change. This includes describing changes to the device characteristics and performance resulting from implementation of the modifications and how the modifications would be implemented (eg, automatic or manual modification, labeling changes).
- Modification Protocol, describing the methods that will be followed when developing, validating and implementing the described modifications. The draft guidance identifies four primary components of a Modification Protocol, that “outline a manufacturer’s 1) data management practices; 2) re-training practices; 3) performance evaluation protocols; and 4) update procedures, including communication and transparency to users and real-world monitoring plans, for each modification in a PCCP,” but also acknowledges that in certain circumstances additional components may be required, and
- Impact Assessment, identifying the benefits and risks (interestingly “including risks of social harm”) introduced by the described modifications and explaining how the activities of the Modification Protocol will assure the safety and effectiveness of the device. Impact Assessment documentation in a marketing submission should discuss how the individual modifications included in the PCCP impact not only the ML-device software functions, but also how they impact the overall functionality of the device, including how they impact device hardware and other device software functions.
The draft guidance draws attention to the need for traceability between the Description of Modifications section and the Modifications Protocol section, and notes that this may be accomplished through a traceability table2 for a PCCP with multiple modifications.
Limitations and open questions
There are several areas where the draft guidance does not go as far as some digital health manufacturers may have hoped, including the following:
- The draft guidance “is not intended to provide a complete description of what may be necessary to include in a marketing submission” and “is also not intended to delineate the types of modifications [FDA] would consider acceptable in a PCCP.” While the draft guidance does illustrate various examples, it does not provide a complete set of instructions for PCCP submissions. The draft guidance states that the foundational assessment of whether a particular modification is appropriate for a PCCP, and what supporting information will be required, will ultimately be determined by the relevant FDA review division. We expect the incomplete definition around when a PCCP might be appropriate to be the subject of significant feedback from stakeholders.
- Limitations on FDA’s current ability to apply PCCP framework to continuous learning. FDA is proposing to consider PCCPs for ML-enabled devices where modifications are implemented automatically “to the extent [FDA] can properly review them.” FDA notes that modifications implemented automatically by software, without human intervention, is an evolving area posing an additional degree of complexity where FDA is still building up its experience.
- A PCCP should include “only a limited number of modifications that are specific, and that can be verified and validated.” Therefore, presumably more complex devices with numerous envisioned modifications may require multiple PCCPs.
- Any modification to an authorized PCCP requires resubmission of the full marketing submission along with the modified PCCP. The draft guidance states that if a “device has been relatively unchanged since FDA’s prior authorization and a new PCCP is proposed, FDA would focus its review on the proposed PCCP.” However, this is not intended to imply a review of the PCCP only and it is not clear how this type of review will be performed.
On April 13, 2023, the FDA will host a webinar for those interested in discussing, and posing questions about, the draft guidance. For those interested in submitting written comments to FDA on this draft guidance, the deadline for doing so is July 3, 2023. To discuss the implications of the draft guidance for your business, please contact any of the authors.
1 In December 2022, as part of the Food and Drug Omnibus Reform Act (FDORA), the FDA gained new authority related to PCCPs. Namely, FDORA amended the Federal Food, Drug and Cosmetic Act to clarify that a new regulatory submission shall not be required for the types of changes authorized under a PCCP, if they are within the scope of an established PCCP.
2 The draft guidance includes a sample traceability table to provide an example of how to depict traceability with clear references.