Objective:

To highlight the discrepancies between FDA clearance of AI medical devices and the actual validation of their safety and effectiveness, emphasizing the misconceptions held by clinicians.

Key Findings:

• Over 1,400 AI medical devices are on the market, with 97% cleared via the 510(k) pathway requiring only substantial equivalence.
• Most physicians mistakenly believe that randomized trials are conducted prior to AI device clinical use, which is often not the case.
• Existing postmarket surveillance systems, like the MAUDE database, capture a very small percentage of adverse events.
• There is a significant gap in understanding what FDA clearance entails versus actual device performance and validation.

Interpretation:

There is a critical need for clinicians to understand the limitations of FDA clearance and the actual performance of AI medical devices to ensure safe and effective integration into clinical practice.

Limitations:

• Postmarket surveillance systems are inadequate for capturing AI/ML performance, leading to potential safety risks.
• Limited industry participation in monitoring programs hampers data collection and understanding of device performance.
• Proposed policy changes may reduce transparency for AI tools outside the FDA device pathway, complicating oversight.

Conclusion:

Healthcare providers should not consider FDA clearance as sufficient for due diligence and must demand comprehensive validation data for AI tools, ensuring they understand the limitations of clearance.