FDA AI/ML proposed Reg Framework
Executive Summary:
The Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) outlines a regulatory framework for modifications to Artificial Intelligence/Machine Learning (AI/ML)-based Software as a Medical Device (SaMD). AI/ML technologies have transformative potential in healthcare, offering benefits such as earlier disease detection, personalized diagnostics, and continuous learning from real-world data. However, their adaptive nature challenges traditional regulatory paradigms, necessitating a Total Product Lifecycle (TPLC) approach to ensure safety and effectiveness while enabling iterative improvements.
Key highlights of the framework include:
Risk-Based Categorization: AI/ML-based SaMD is categorized based on the significance of information provided and the healthcare situation or condition, ranging from low-risk (Category I) to high-risk (Category IV).
Types of Modifications: Modifications are grouped into three categories:
Performance improvements.
Changes to input data types.
Changes to intended use.
Total Product Lifecycle (TPLC) Approach: The framework emphasizes continuous oversight, including premarket review, postmarket monitoring, and real-world performance reporting. It incorporates Good Machine Learning Practices (GMLP) to ensure quality and safety throughout the lifecycle.
Predetermined Change Control Plan: Manufacturers can submit SaMD Pre-Specifications (SPS) and Algorithm Change Protocols (ACP) during premarket review to outline anticipated modifications and risk management strategies.
Transparency and Real-World Monitoring: Manufacturers are expected to provide updates on algorithm changes, performance improvements, and labeling modifications, supported by real-world evidence.
The framework aims to balance innovation with patient safety, enabling adaptive AI/ML technologies to evolve while maintaining regulatory safeguards. Feedback is sought on various aspects, including modification categories, GMLP considerations, SPS/ACP elements, and mechanisms for transparency and real-world monitoring.
