By: Elle Winn (Mittal Consulting)

Exploring the Future of AI Medical Devices: FDA’s Predetermined Change Control Plan Offers Potential, but Challenges Remain

Artificial Intelligence (AI) is continuing to grow at an awe-striking rate. As the artificial intelligence ecosystem continues to develop, we should expect to see AI continue with strong penetration into the med-tech sector.

AI and machine learning (ML) functions have the potential to drastically increase the value and capabilities of modern medical technologies. Especially in the age of big-data, AI- and ML-equipped medical devices have the potential to not just revolutionize diagnostics and treatment, but to also make sense of and find patterns in the massive amount of healthcare data we’ve already amassed. While the potential benefits of AI- and ML-enabled medical devices can be clearly seen at this point, another consideration remains paramount: user and patient safety. This is where regulatory bodies like the FDA comes in.

Regulatory Landscape for AI/ML Up to Now

AI software functions are continually developed in iterative processes throughout the product lifetime. Before this guidance, any new changes to AI-enabled medical devices in these iterations would require a new marketing submission to gain FDA clearance or approval for that device version. This is a major regulatory burden and acts as a major barrier to entry for any AI-enabled device in the med-tech field. To maintain the regulatory integrity of AI- and ML-enabled medical devices, the FDA has created a new program that fosters the iterative development of such devices without requiring a new marketing submission each time. The proposed solution is the Predetermined Change Control Plan (referred to in short as PCCP). PCCPs allow manufacturers to prospectively specify and earn authorization for intended device modifications. This article will explore PCCPs as they pertain to AI- and ML-enabled medical devices based on the newly released final FDA Guidance Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions(December, 2024). This final guidance document comes after the draft guidance which was issued in April 2024 as well as another draft guidance expanding the PCCP program to all medical devices. The final guidance discussed in this article recommends aspects of PCCPs that must be submitted in the formal marketing submission and provides general recommendations for PCCPs intended to be submitted for AI-enabled devices.

PCCP for AI- and ML-Enabled Devices

There are two main points discussed in this guidance:

1. How to submit a PCCP to be authorized
2. What should be included in a proposed PCCP.

Starting with the how, FDA highlights that proposed PCCPs can be submitted in either of the three major marketing applications: Pre-Market Approval (PMA) applications, De Novo Requests, or 510(k) Submissions. Furthermore, FDA clarifies that proposing a PCCP will not require a separate submission – proposed PCCPs should be included as a standalone section within the marketing submission. In addition, FDA notes that the PCCP does not have an impact on what marketing submission a sponsor should choose to complete; the submission pathway should be selected according to regulatory strategy and device classification and risk.

Moving into the “what”, there are three major sections to be included in a proposed PCCP. The overview of these sections is shown in the figure below:

As with any formal submission, but especially for PCCP proposals, as much detail and scientific justification for each change, change implementation method, and change impact should be provided for the FDA to determine if there is reasonable assurance of safety with each iteration.

FDA notes that early and frequent engagement with the FDA regarding PCCPs is highly encouraged due to the complexity of AI-enabled tools and the novelty of the PCCP program.

The Caveats

While PCCPs have an allure of decreased regulatory burden through the modification pre-authorization process, there are still some concerns about how feasible and functional this program will be for device developers.

Primarily, the guidelines to determine what changes will be acceptable in a proposed PCCP are very obscure. In the guidance, FDA notes the following types of modifications may be suitable for a PCCP:

1. “Modifications related to quantitative measures of AI-DSF [AI-enabled Device Software Functions] performance specifications”
2. “Modifications related to device inputs to, and compatibility with, the AI-DSF”
3. “Certain modifications related to the device’s use and performance (e.g. for use within a specific subpopulation”

To add to the ambiguity, FDA notes later in the guidance that a modification in one device may not be appropriate for a modification in another device, and that all decisions are further based on risk. All in all, the primary message we get from this is that Pre-Submissions will be key in developing a PCCP for at this point, pre-submissions are really the only way to get any valuable specific input from FDA on acceptable changes in AI-enabled medical devices for PCCPs.

With the novelty of this program, there are not enough authorized PCCP plans to provide much insight about how functional the program is and how much benefit it will provide device developers. With the amount of ambiguity in the guidance it is clear FDA has ensured they have enough leeway in their decision-making processes. At this point, it seems reasonable to expect that AI-enabled devices will come with the same, amplified regulatory challenges that other cutting edge medical technologies come with. Cutting edge medical technologies face higher regulatory hurdles, higher burdens of proof, and more scrutiny by the FDA – while more common technologies can “copycat” the regulatory strategies of their predecessors. With all the information about PCCPs for AI devices we’ve been provided (and not provided), it continues to beg the question: will innovators of cutting-edge medical technologies always get penalized?