The FDA's Solution to Adaptive AI: Understanding the Predetermined Change Control Plan (PCCP)
The rapid integration of Artificial Intelligence and Machine Learning (AI/ML) into Software as a Medical Device (SaMD) has presented a profound challenge to traditional regulatory frameworks [1]. Unlike static medical devices, AI algorithms are designed to be adaptive, learning and improving from real-world data post-deployment. This continuous evolution, while beneficial for patient care, clashes with a regulatory system historically built on reviewing a fixed product. The core question for regulators has become: How can the safety and effectiveness of an AI-enabled device be assured when its performance characteristics are constantly changing?
The U.S. Food and Drug Administration (FDA) has addressed this challenge with a novel, risk-based approach centered on the Predetermined Change Control Plan (PCCP). This framework, detailed in the FDA's recent guidance, is designed to allow manufacturers to make specific, pre-authorized modifications to their AI/ML-enabled devices without requiring a new pre-market submission for every single update [2]. The PCCP shifts the regulatory focus from reviewing every change after it occurs to reviewing the manufacturer's plan for managing future changes before they occur.
The Two Pillars of the PCCP
A successful PCCP submission is built upon two essential components that together define the scope and methodology of future algorithm updates: the Description of Modifications and the Modification Protocol.
1. Description of Modifications (What Will Change)
This component clearly outlines the types of changes the manufacturer intends to implement in the future. It is a commitment to the FDA regarding the scope of the device's evolution. These modifications fall into three primary categories:
- Performance Changes: Updates intended to improve the clinical performance of the device, such as increasing sensitivity or specificity.
- Retraining Changes: Updates involving the use of new data to retrain the model, often to address issues like dataset drift or to expand the model's generalizability.
- Input Changes: Changes to the type of data the algorithm receives, such as incorporating a new sensor or a different image modality.
The Description of Modifications ensures that the FDA and the public have a clear understanding of the boundaries within which the AI will be allowed to evolve.
2. Modification Protocol (How Changes Will Be Controlled)
The Modification Protocol is the manufacturer's detailed plan for how they will develop, test, validate, and implement the described modifications while maintaining the device's safety and effectiveness. This is the control mechanism of the PCCP. It must include:
- Data Management Practices: A clear plan for how new data will be collected, curated, and managed to ensure quality and relevance for retraining.
- Re-training and Validation Procedures: Specific, pre-defined protocols for how the algorithm will be retrained, including the acceptance criteria (e.g., performance metrics, statistical thresholds) that the updated algorithm must meet before deployment.
- Performance Evaluation: A strategy for real-world monitoring and evaluation of the updated device post-deployment to detect any unforeseen performance degradation or bias.
By establishing these rigorous, pre-specified protocols, the FDA ensures that the manufacturer has a robust Quality Management System (QMS) in place to handle the dynamic nature of the AI/ML algorithm [3].
Streamlining Innovation While Ensuring Safety
The PCCP represents a significant regulatory evolution, moving away from a static, one-time review to a Total Product Lifecycle (TPLC) approach. This TPLC framework recognizes that for AI/ML SaMD, the development process is continuous.
The key benefit of the PCCP is the ability for manufacturers to implement pre-authorized changes quickly, accelerating the pace of innovation and ensuring that patients benefit from the latest algorithmic improvements without unnecessary regulatory delays. This is particularly crucial in fields like digital pathology and radiology, where small improvements in diagnostic accuracy can have a massive impact on patient outcomes.
However, the PCCP is not a blank check. Any modification that falls outside the scope of the approved Description of Modifications or fails to meet the criteria defined in the Modification Protocol still requires a new pre-market submission to the FDA. This maintains the agency's ultimate oversight on significant changes that could impact safety or effectiveness.
The Future of AI Regulation in Digital Health
The FDA's PCCP framework is a leading example of how regulatory bodies are adapting to the unique challenges of AI in healthcare. It strikes a critical balance: fostering the rapid, iterative development inherent to machine learning while upholding the agency's mandate to protect public health. The success of this model will likely influence regulatory approaches globally, setting a precedent for how to manage the lifecycle of adaptive technologies.
For more in-depth analysis on this topic, the resources at www.rasitdinc.com provide expert commentary on the intersection of digital health, AI, and regulatory science, offering valuable insights for professionals navigating this complex landscape.
References
[1] FDA. Artificial Intelligence and Machine Learning (AI/ML) in Software as a Medical Device (SaMD) Action Plan. U.S. Food and Drug Administration. 2021. [2] FDA. Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence-Enabled Device Software Functions. U.S. Food and Drug Administration. 2025. [3] FDA. Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD). U.S. Food and Drug Administration. 2019. [4] Singh, V. United States Food and Drug Administration Regulation of AI/ML SaMD. ScienceDirect. 2025. [https://www.sciencedirect.com/science/article/pii/S2949761225000380] [5] Gilbert, S. Algorithm Change Protocols in the Regulation of Adaptive AI/ML-Based Software as a Medical Device. PMC. 2021. [https://pmc.ncbi.nlm.nih.gov/articles/PMC8579211/]