Decoding The FDA's Final Guidance On PCCPs For AI-Enabled Device Software Functions

By Yu Zhao, Bridging Consulting

The FDA’s final guidance on Predetermined Change Control Plans (PCCPs) for AI-enabled device software functions (AI-DSFs) has arrived, replacing the draft guidance issued in April 2023. While the final guidance largely builds on the earlier draft, it incorporates several critical refinements aimed at addressing practical challenges and advancing the regulatory framework for current and future AI-driven innovation. Here’s an in-depth look at the key updates, from terminology shifts to expanded transparency measures and clarified regulatory expectations. This article does not cover many other important PCCP concepts and requirements that remain unchanged from the draft to the final guidance.

The order of the 13 takeaways below is based on the order of content appearance in the final guidance.

1. Terminology Update: ML-DSF To AI-DSF

One of the most visible changes in the final guidance is the terminology update, from machine learning-enabled device software function (ML-DSF) to AI-DSF. This terminology shift reflects the FDA’s intention to future-proof the guidance, ensuring its applicability to both current, foreseeable, and even unforeseeable advancements in AI technologies beyond machine learning. However, most recommendations and examples remain tailored to the unique characteristics of ML-based algorithms. This terminology update signals the FDA’s recognition of the evolving nature of AI technologies while maintaining the focus on technologies already in active use in today’s healthcare.

2. Narrowed Scope For Combination Products

The draft guidance indicated that the guidance and recommendations “apply to the device constituent part of a combination product,” which includes the device constituent part of any combination product regardless of whether their primary mode of action (PMOA) is device, drug, or biologic. However, the final guidance restricts its scope to the device constituent part of "device-led combination products." It stays silent on the device constituent part of drug-led or biologic-led combination products. To explore whether modifications to the device constituent part of drug- or biologic-led combination products are appropriate for PCCPs, manufacturers should seek case-specific regulatory feedback, potentially through the Q-Submission process, from the FDA.

3. Submission Type Clarifications

The draft guidance indicated that the guidance is for devices subject to 510(k), PMA, or De Novo premarket submissions, without specifying submission types appropriate for the PCCP submission. The final guidance provides more granular details on appropriate submission types for PCCPs.

Submission Types Appropriate To Establish A PCCP

For PMA devices:

• Original PMA
• Modular PMA (where a PCCP comprises a module of review)
• 180-Day PMA Supplement
• Panel Track PMA Supplement
• Real-Time Review (RTR) Supplement

For 510(k) devices:

• Traditional 510(k)
• Abbreviated 510(k)

For Class II or Class I devices subject to De Novo pathway:

• Original De Novo request

Submission Types NOT Appropriate To Establish A PCCP (though not specifically listed in the final guidance)

• For PMA devices: 30-Day Notice, Change Being Effected (CBE), and PMA Annual Report
• For 510(k) devices: Special 510(k) and CBE 510(k)

These excluded submission types are either ones that do not require prior approval from the FDA (CBEs and PMA annual report) or notifications that have shorter review cycles (30-Day Notice and Special 510(k)) and therefore are not appropriate for comprehensive review required for a PCCP.

Among the allowable PMA supplement types, RTR has the shortest review cycle (i.e., 90 days). The final guidance indicates that RTR is only appropriate “where a PCCP comprises minor changes and the manufacturer and FDA agree that the review can be achieved in a real-time setting.”

Notably, Manufacturing Site Change Supplement for PMA devices is not specifically listed. However, it is also a type of 180-Day Supplement, so it could be interpreted as covered under 180-Day Supplement.

Similarly, if manufacturers plan to leverage PCCPs for making modifications to the device constituent part of drug- or biologic-led combination products, and to determine the appropriate NDA or BLA submission types, they should seek case-specific regulatory feedback from the FDA.

4. Regulatory History And QSR Compliance

The final guidance highlights that the FDA “may under certain case-by-case circumstances withhold clearance of a PCCP submitted in a 510(k) based on findings in the regulatory history of the manufacturer that demonstrate failure to comply with QSR.”

Although QSR compliance or manufacturing information is not required in 510(k) submissions, the FDA reserves the right to withhold PCCP clearance if a manufacturer’s regulatory history indicates noncompliance. This underscores the need for device manufacturers to have robust quality systems in place and to maintain a strong compliance record to support PCCP and device safety and efficacy.

5. Emphasis On Transparency

The final guidance mandates detailed documentation of PCCP-related information in device labeling and public summaries, including PMA Summary of Safety and Effectiveness Documents (SSEDs), 510(k) summaries, and De Novo decision summaries. These requirements ensure that stakeholders, particularly device users such as healthcare providers and patients, have access to critical information about software functions, implemented updates, and the type of regulatory oversight applied. Such transparency is especially critical when software modifications are made automatically by an adaptive ML algorithm (also referred to as a continuous learning algorithm), which, once deployed, can learn from new data and update itself without human intervention.

6. Guardrails For Continuous Learning Algorithms

To date, the FDA has not cleared, granted, or approved any device incorporating continuous learning algorithms. A PCCP provides an ideal pathway for obtaining authorization for such algorithms, as it allows the FDA to ensure safety and effectiveness by requiring manufacturers to include detailed information in the modification protocol and the impact assessment sections. These details, which often exceed the requirements of a typical 510(k), De Novo, or even PMA submission, help the agency build confidence in the adaptive algorithm’s ability to operate within defined boundaries while adhering to regulatory compliance standards.

In fact, the final guidance underscores the importance of setting boundaries by emphasizing that manufacturers must “clearly establish boundaries or guardrails that define the range of automatic updates” for continuous learning algorithms in their PCCPs. This highlights the critical need for clear process distinctions between fully automated software updates and those that involve a human-in-the-loop approach for testing and deployment.

7. Softer Stance On Indications For Use Modifications

A significant shift in tone appears in the guidance’s treatment of indications for use modifications:

• Draft Guidance: “At this time, FDA expects that modifications included in a PCCP should also maintain the device within the device’s indications for use. As with modifications to the intended use, FDA believes modifications to the indications for use would not allow the device to remain safe and effective.”
• Final Guidance: The FDA still maintains that any change to the intended use of the device is not appropriate for the PCCP. However, it softens the tone regarding changes to indications for use: “In general, FDA believes that modifications included in a PCCP should also maintain the device within the device’s indications for use. As with modifications to the intended use, FDA believes that most modifications to the indications for use included in a PCCP would be difficult for FDA to assess prospectively to determine whether the device would remain safe and effective. However, there may be certain modifications to the indications for use (e.g., certain changes in the indications for use to specify use of the device with an additional device or component) that may be appropriate for inclusion in a PCCP.”

This nuanced approach provides greater flexibility for manufacturers seeking to expand the utility of their devices within defined parameters. Manufacturers should engage with the FDA through the Q-Submission Program when planning to include indications for use modifications in their PCCPs. This proactive step can help clarify agency expectations and streamline the submission process.

8. Verification And Validation Protocols

The final guidance requires manufacturers to indicate whether verification and validation protocols differ from previously cleared or approved versions. For instance, if the modification protocol involves clinical validation for certain modifications, the PCCP should specify whether there are any differences in the study type, protocol (including the truthing process), and performance acceptance criteria compared to the clinical validation study previously approved by the FDA. Furthermore, manufacturers must provide justifications for any deviations. This step ensures consistency while promoting continuous improvement in the evaluation of device performance, ensuring that devices with these modifications maintain or enhance safety and effectiveness.

9. Performance Testing Guidance

In general, repeatedly testing an ML algorithm until success is not appropriate due to concerns about overfitting and confirmation bias. A helpful note in the final guidance clarifies that a failure is not “considered to be unresolvable if a root cause analysis of the failure reveals that it is not related to specific aspects of the PCCP,” and that in such cases, “performance testing may be conducted again.” This addition supports a practical approach to device testing and validation.

However, whether the same test data set can be reused fully or partially in retesting must be carefully assessed on a case-by-case basis.

10. Impact Assessment for Combination Products

The Impact Assessment section in the final guidance explicitly requires manufacturers to evaluate how each individual modification to the device constituent affects:

• drug or biologic constituents and
• the combination product as a whole.

This holistic approach ensures that any changes made to the device constituent do not compromise the safety and effectiveness of the combination products. By considering the interplay between all constituents, manufacturers can address potential risks and maintain the overall integrity and intended functionality of the product.

11. Strengthened Postmarket Surveillance

The final guidance formalizes the draft guidance’s recommendation that the update procedures, as part of the modification protocol, should include “post-market surveillance plans and procedures,” such as real-world monitoring and user notification requirements. Additionally, the final guidance adds the following question for consideration in Appendix A: “Will there be criteria and/or plan to roll-back an update to reset devices to a previous version, if applicable?”

These measures align with FDA’s focus on ensuring ongoing safety and effectiveness in the real-world environment.

12. UDI Requirements

In Appendix A, under Update Procedures, the final guidance references the unique device identifier (UDI) requirements, stating, “New unique device identifiers (UDIs) are required for devices that are required to bear a UDI on its label when there is a new version and/or model, and for new device packages.”

However, determining whether every single software update requires a new device identifier (DI) is less black and white. This decision depends on factors such as the nature and scope of the update and its potential impact on the device’s intended use and/or functionality.

Manufacturers should follow their quality system procedures to assess UDI requirements. If uncertainties exist, manufacturers may consider consulting an external UDI expert to ensure compliance.

13. New Example in Appendix B

Appendix B of the final guidance now includes Example #6, an “Optical Imaging System Co-packaged with Imaging Drug,” specifically tailored to device-led combination products. This new example and the post-authorization modification scenarios help manufacturers navigate through postmarket change management for their combination products.

Final Thoughts

The FDA’s final PCCP guidance for AI-DSFs marks an important step forward in aligning regulatory practices with the rapid evolution of AI in healthcare. By addressing stakeholder feedback and refining its recommendations, the guidance provides a clearer framework for navigating the complex and evolving regulatory landscape.

While the final guidance closely mirrors the 2023 draft version, it aligns well with real-world PCCP review experiences with CDRH in the last two years. Designed to encompass all device types across varying risk classifications, the recommendations and requirements are comprehensive enough to cover PCCPs for high-risk PMA device modifications.

Manufacturers should always follow the least burdensome approach, using this guidance as a helpful tool rather than rigidly adhering to every single requirement. For those who are new to PCCPs, or experienced manufacturers tackling a new device type or modification type, engaging with the FDA through the Q-Submission process is highly recommended.

Finally, although this final guidance is for AI-enabled device software functions, its principles can extend to other device types. Manufacturers can reference this final guidance and the August 2024 draft guidance, PCCPs for Medical Devices, to guide their PCCP preparation for non-AI-DSF modifications.

As the AI landscape continues to evolve, this guidance serves as both a milestone and a foundation for future regulatory innovation.

About The Author:

Yu Zhao founded Bridging Consulting LLC, a consultancy dedicated to assisting AI startups and medical device companies in achieving innovation and regulatory compliance, in 2020. He has more than two decades of experience in the life sciences and technology space, and specializes in medical device regulatory, quality, and clinical affairs. His clients range from startups to large industry companies. During his tenure at Medtronic, he led regulatory affairs departments for multi-billion-dollar business units.