Algorithm Change Protocols in the Regulation of Adaptive Machine Learning-Based Medical Devices.

Area of Science:

• Healthcare innovation
• Artificial intelligence
• Machine learning

Background:

• Current healthcare regulations, designed for static medical devices, hinder the continuous improvement of AI and ML models.
• Real-time or near real-time data updates for ML models present significant regulatory challenges under existing frameworks.

Purpose of the Study:

• To examine current regulatory frameworks for AI/ML in healthcare.
• To advocate for innovative regulatory approaches that match the dynamic nature of AI/ML.
• To explore international perspectives and propose a balanced regulatory paradigm.

Main Methods:

• Review of the status quo and recent developments in healthcare AI/ML regulation.
• Analysis of international regulatory proposals, including those from the WHO and FDA.
• Examination of the draft EU regulatory framework.

Main Results:

• Existing regulations pose significant hurdles for continuously learning healthcare ML models.
• The FDA's proposed approach focuses on quality management systems and algorithm change protocols.
• The EU framework suggests similar approaches but lacks implementation details for algorithm change protocols.

Conclusions:

• Innovative healthcare AI/ML necessitates adaptive regulatory strategies.
• A paradigm shift towards oversight of developer quality systems and defined algorithm change protocols is crucial.
• Detailed implementation of algorithm change protocols is essential for realizing AI/ML benefits in the EU while ensuring patient safety.