Evaluating transparency in AI/ML model characteristics for FDA-reviewed medical devices

Area of Science:

• Medical Technology
• Regulatory Science
• Artificial Intelligence

Background:

• The integration of artificial intelligence (AI) and machine learning (ML) in medical devices necessitates transparent regulatory reporting.
• The U.S. Food and Drug Administration (FDA) introduced Good Machine Learning Practice (GMLP) guidelines in 2021 to address this need.
• Uncertainty remains regarding the adherence to GMLP principles in FDA-reviewed AI/ML medical devices.

Purpose of the Study:

• To assess the transparency of regulatory reporting for AI/ML-enabled medical devices approved by the FDA.
• To evaluate the impact of the 2021 GMLP guidelines on reporting transparency.
• To identify gaps in transparency concerning model development and performance metrics.

Main Methods:

• A review of 1,012 FDA Summaries of Safety and Effectiveness (SSEDs) for AI/ML devices approved between 1970 and December 2024.
• Utilized a novel AI Characteristics Transparency Reporting (ACTR) score across 17 categories to quantify transparency.
• Analyzed changes in ACTR scores before and after the 2021 GMLP guideline issuance.

Main Results:

• The average ACTR score was low at 3.3 out of 17, indicating significant transparency deficits.
• A modest improvement of 0.88 points in ACTR scores was observed post-2021 guidelines.
• Nearly half of the devices lacked clinical study reporting, and over half omitted performance metrics.

Conclusions:

• Significant transparency gaps exist in the regulatory reporting of AI/ML-enabled medical devices.
• Current adherence to GMLP principles is insufficient to ensure adequate transparency.
• Enforceable standards are crucial to enhance trust and ensure the safety of AI/ML medical technologies.