United States: FDA Issues Draft Guidance On Predetermined Change Control Plans For Artificial Intelligence/Machine Learning-Enabled Device Software Functions

The U.S. Food and Drug Administration (FDA) has issued a draft guidance intended to promote the development of safe and effective medical devices that use a type of artificial intelligence (AI) known as machine learning (ML). The draft guidance further develops FDA's least burdensome regulatory approach for AI/ML-enabled device software functions (ML-DSFs), which aims to increase the pace of innovation while maintaining safety and effectiveness.

The draft guidance follows the FDA's 2019 AI/ML Discussion Paper, 2021 AI/ML Action Plan and the December 2022 enactment of the Food and Drug Omnibus Reform Act of 2022¹, which added Section 515C, authorizing the FDA to approve predetermined change control plans (PCCPs) for devices in premarket submissions. Specifically, the draft guidance provides recommendations on the components of PCCPs for ML-DSFs designed to prospectively avoid the need for new marketing submissions for post-market changes. The comment period for this draft guidance is open until July 3, 2023.

In short, once on the market, ML-DSFs are capable of learning through real-world data to perform a task—sometimes without being explicitly programmed to perform that task.² These modifications to ML-DSFs may have the potential to affect significantly the ML-DSFs' safety or effectiveness. Traditionally, manufacturers seeking to update their already-approved software functions in a manner that would affect the safety and effectiveness would need to consider whether a new marketing submission must be submitted to FDA, which can be an expensive and lengthy process. If a manufacturer has a PCCP authorized by FDA through its initial premarket submission, however, it can proactively preauthorize certain specified, future modifications to an ML-DSF such that those changes can be implemented without new marketing submissions. The draft guidance does not apply to plans that contain only minor modifications that would not require new marketing submissions to FDA.

FDA is clear that PCCPs should contain the following components:

• A description of the planned ML-DSF modifications, i.e., specifics on what a manufacturer expects the software to become as it learns from data;
• The associated methodology to develop, implement and validate those modifications, i.e., how the software is going to learn and change while remaining safe and effective for use; and
• An assessment of the impact of those modifications, i.e., a description of the modifications' benefits and risks, as well as methods of mitigation of those risks.

Modifications appropriate for a PCCP include those that maintain or improve the safety or effectiveness of the device. In contrast, modifications that deduct from the safety of the device will not be acceptable. The types of modifications that may be acceptable are detailed in the draft guidance and include modifications related to quantitative measures of ML-DSF performance specifications and modifications related to device inputs. The existing quality system must be able to validate the modifications. Further, modifications must be within the approved indication for use for the device—meaning that manufacturers should take into account their future intended changes when crafting an indication for use statement.

Transparency to users is of utmost importance in order to ensure end users understand what changes may be made to ML-DSFs that are on the market. FDA would consider continued marketing of a modification outside the scope of a PCCP without approval of a new submission to render the ML-DSF adulterated and misbranded. Thus, manufacturers should begin early, invest significant time and effort into delineating their pre-determined ML-DSF modifications in PCCPs, and pay close attention to their ML-DSF's behavior once on the market. In fact, FDA strongly encourages that manufacturers engage with FDA early regarding a proposed PCCP for ML-DSFs through the Q-submission process.

For More Information

If you have any questions about this Alert, the draft guidance, its potential implication or how to implement its recommendations, please contact Frederick R. Ball, Sean K. Burke, Coleen W. Hill, any of the attorneys in our Life Sciences and Medical Technologies Industry Group, any of the attorneys in our Products Liability and Toxic Torts Group or the attorney in the firm with whom you are regularly in contact.

Footnotes

1. Title III of 131 Division FF of the Consolidated Appropriations Act, 2023, Pub. L. No. 117-328.

2. The draft guidance applies to ML-DSFs that the manufacturer intends to modify over time, regardless of whether those changes to the ML model are implemented automatically by the software or will be implemented manually through human input, action, review or decision-making.

Disclaimer: This Alert has been prepared and published for informational purposes only and is not offered, nor should be construed, as legal advice. For more information, please see the firm's full disclaimer.