During a session at the 2024 American College of Clinical Pharmacy (ACCP) Annual Meeting, Omar Badawi, PharmD, MPH, FCCM, shed light on the complexities and challenges faced by the U.S. Food and Drug Administration (FDA) in regulating medical devices, particularly those incorporating artificial intelligence (AI) and machine learning (ML). Badawi heads the Division of Data Sciences at the US Telemedicine and Advanced Technology Research Center.

The FDA categorizes medical devices into three classes based on risk. Class I devices are considered low risk and typically exempt from the pre-market approval process. However, they must still adhere to general controls and labelling requirements. Class II devices present moderate risk and usually require FDA clearance, which is a procedure distinct from full approval. Notably, most Class I and Class II devices are exempt from the 510(k) premarket notification, implying they have demonstrated equivalence to similar pre-existing devices. In contrast, Class III devices, regarded as high-risk, are subject to rigorous pre-market approval processes.

Despite FDA efforts to track AI and ML-enabled devices, the categorization remains inconsistent, Badawi noted. As of August 2024, the FDA had listed 950 AI/ML medical devices. However, not all devices with potential AI/ML functionalities, such as continuous glucose monitors, are recognised under this listing. This is because these devices are approved following the same protocols as lab assays for glucose measurement, bypassing the detailed scrutiny intended for AI/ML devices.

The FDA insists on post-market safety surveillance for AI/ML devices and assessments to identify bias, but these standards aren’t uniformly applied across all devices employing similar technologies. Wearable technology like smart clothing, fitness trackers, and certain smart glasses escape stringent FDA review, primarily because they are positioned as consumer electronics rather than medical devices. Meanwhile, serious ethical and privacy concerns have surfaced, such as a recent incident involving Harvard students using smart glasses for unauthorised facial recognition.

The discussion also brought up the pressing issue of transparency in AI tool development and validation. A recent investigation revealed that only a small percentage of FDA-cleared AI/ML devices underwent prospective validation or randomised controlled trials. This lack of transparency stokes concerns about the evidence quality underlying these medical tools. Badawi pointed to studies highlighting biases in AI-driven medical assessments, citing racial disparities in pulse oximeter readings and differential diagnosis biases within AI models like ChatGPT-4.

Such findings underscore the critical need for robust validation methodologies and greater transparency in AI applications within healthcare. Researchers and clinicians are urged to consider these biases, which can lead to inequities in care delivery and outcomes across diverse patient demographics.

In conclusion, while AI and ML hold transformative potential for medical device efficacy and innovation, these technologies introduce considerable regulatory, ethical, and safety challenges that demand careful oversight and continuous examination.

Source: Noah Wire Services