FDA Draft Guidance on Predetermined Change Control Plans for Artificial Intelligence/Machine Learning-Enabled Medical Devices

By Tiffany Yang-Tran

In recent years, the FDA has embarked on an ongoing journey to develop a premarket review approach for artificial intelligence (AI)/machine learning (ML) software modifications. This journey included the Agency’s 2019 discussion paper and request for feedback on the proposed regulatory framework, several workshops to gather inputs from various stakeholders, as well as the Agency’s action plan released in 2021 that described the FDA’s strategy for addressing AI/ML-enabled medical devices in a holistic, collaborative, and multidisciplinary manner. In April 2023, the US Food and Drug Administration (FDA) released the guidance, Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence/Machine Learning (AI/ML)-Enabled Device Software Functions. Furthermore, in March of 2024, the Center for Biologics Evaluation and Research (CBER), Center for Drug Evaluation and Research (CDER), Center for Devices and Radiological Health (CDRH), and Office of Combination Products (OCP) published a brief paper “which outlines the agency's commitment and cross-center collaboration to protect public health while fostering responsible and ethical medical product innovation through Artificial Intelligence.”

This article will look more closely at the April 2023 guidance and some of its direct impacts on AI/ML product development, and particularly ML-enabled device software functions (ML-DSFs) which the guidance focuses on.

Controlling the iterative nature of AI/ML

ML is a powerful tool that allows software to learn through real-world data to augment a device’s functionality. ML-enabled technologies represent a transformative shift in healthcare providing solutions with the ability to assist healthcare providers and allow for better patient outcomes. ML is greatly characterized by its adaptability and its capability to improve its performance through iterative input from the manufacturer, or from the ML itself (i.e., the software’s functionality). As such, ML-device software functions (ML-DSFs) present challenges for traditional regulatory pathways (i.e., 510(k) premarket clearance, De Novo classification, premarket approval) intended for devices that are developed and released as a final, “static” form. As AI and ML continue to transform medical devices and healthcare, FDA’s new draft guidance attempts to accommodate the “adaptive” nature of these AI/ML technologies and “proposes a least burdensome approach to support iterative improvement through modifications to an ML-enabled device software functions (ML-DSFs) while continuing to provide a reasonable assurance of device safety and effectiveness”.

Utilizing a Predetermined Change Control Plan (PCCP)

With this April 2023 guidance, manufacturers can proactively specify and seek premarket authorization for intended modifications to an ML-DSF. This proactive approach requires the submission of a predetermined change control plan (PCCP) as a standalone section within the marketing submission. The PCCP should describe the modifications made to the ML-DSF, how they will be implemented, and how those modifications will be assessed. Modifications covered in the authorized PCCP can then be implemented without the need for a new marketing submission.

The PCCP’s main components are:

1. Description of modifications. This section should include a list of specific, and verifiable proposed device modifications as well as rationale for each change. Notably, each modification should state whether the change will be implemented automatically (i.e., via software) versus manually (i.e., via human input/review).
2. Modification protocol. This section should describe the methods that will be followed to develop, validate, and implement the modifications. Specifically, data management practices, ML re-training methods, performance evaluation metrics, ML update procedures, and user and stakeholder communication strategies should all be described in a clear and traceable manner. All verification and validation activities, including pre-defined acceptance criteria, should be included in this section.
3. Impact assessment. This section should discuss the benefits and risks of implementing the modifications, as well as any risk mitigation plans. The scope of this section should capture how each modification impacts the ML-DSF, how each individual modification impacts each other, as well has how they impact the other software and hardware functions of the device. It should also address any risks of social harm and how the device continues to ensure the safety and effectiveness of the device.

Once a device has submitted a marketing submission and has an authorized PCCP, manufacturers must ensure the modification is specified in the PCCP and is implemented in accordance with steps delineated in the modifications protocol.  The manufacturer can proceed to implement that modification and document the change in their quality system.

If a modification is not covered in the PCCP, the manufacturer must take the appropriate steps to submit a new marketing submission.  A manufacturer could attempt to modify an existing authorized PCCP, and submit it for review. However, the FDA currently believes that such modifications could likely reveal changes that could significantly affect the safety or effectiveness of the device and thus require a new marketing submission for the device. The FDA intends to publish guidance on PCCPs in general later this year.

A human factors perspective

As part of the FDA’s overall vision to “address AI/ML-enabled medical devices in a holistic, collaborative, and multidisciplinary manner,” this guidance helps manufacturers developing ML-DSF-enabled devices to rapidly iterate, develop, and improve under the highly adaptive nature of the AI and ML technology. 

From Emergo’s standpoint, it is imperative that human factors remain a core consideration when modifying AI/ML-DSFs. The PCCP guidance states that “modifications must be made in a manner that ensures the continued safety and effectiveness of the device” and, as such, manufacturers should consider how the modifications impact the user experience. If the modification results in changes to the device labelling, appropriate human factors measures should be taken to assess any potential impacts to the user’s performance, their understanding of the change, and any shifts that might occur in the user’s workflow and/or responsibilities. AI/ML-DSFs should provide transparency and visibility into key data inputs, and especially continue to do so when modifications are made.

Navigate the FDA’s expectations with Emergo by UL

Please reach out if we can help you consider how this new guidance, and/or other existing HFE guidance documents, impacts the HFE work needed for your SaMD product. We remain eager to help our customers plan HFE in a compliant and efficient manner, as well as to help our customers respond to Additional Information and other regulator requests for HFE-related information.

Tiffany Yang-Tran is a Senior Human Factors Specialist at Emergo by UL.