How should we regulate medical AI in the UK? With Professor Stephen Gilbert- Medical Device Regulatory Science expert

Key Points

• AI in medical devices presents unique regulatory challenges due to its adaptable nature, high-frequency update potential, and broad societal risks and opportunities, necessitating specialized regulatory approaches beyond traditional medical device frameworks.
• Effective regulation must balance preventing patient harm from irresponsible innovation with avoiding harm from lack of access due to over-regulation, emphasizing the need for "better" rather than simply more or less regulation.
• A significant divergence exists between EU and US regulatory strategies: the EU's MDR and AI Act generally classify AI-based decision support as medical devices requiring notified body assessment, while the US FDA's guidance creates a "non-device" category for certain explainable decision support tools, exempting them from pre-market review.
• Regulatory science and adaptive frameworks, like predetermined change control plans, are crucial for keeping pace with AI innovation while ensuring safety, highlighting a global shift towards more dynamic oversight models.

Summary

This podcast episode features a discussion between Dr. Annabelle Painter and Professor Steven Gilbert on the critical issue of AI regulation in healthcare, particularly within the UK context. Professor Gilbert, drawing from his background in veterinary medicine, academic research, and industry roles in medical device regulation, argues that AI medical devices are fundamentally different from traditional devices like implants. Their adaptable, software-based nature allows for rapid, beneficial updates but also introduces unique risks, justifying specialized regulatory consideration as seen in the EU's horizontal AI Act approach.

A central theme is the regulatory balance between fostering innovation and ensuring patient safety. Gilbert illustrates this pendulum swing with historical examples: scandals like faulty metal hip implants led to calls for stricter regulation (MDR), while subsequent access shortages for orphan devices sparked campaigns for less red tape. He advocates for "better regulation," achieved through regulatory science—where authorities collaborate with academia to preemptively address emerging tech challenges—and adaptive frameworks that accommodate iterative AI improvements.

The conversation highlights a major transatlantic regulatory divergence. The EU system, which the UK currently aligns with, typically classifies AI-driven diagnostic decision support as a medical device requiring conformity assessment. In contrast, a pivotal US FDA guidance carves out a "non-device" category for certain clinical decision support software, exempting it from pre-market review if it meets criteria like not performing direct image analysis and being explainable. Gilbert views this as a transformative, order-of-magnitude difference that grants the US sector greater flexibility and growth potential, though it raises questions about oversight and risk. The discussion underscores that as the UK updates its regulations, it must navigate these complex, globally inconsistent landscapes to both protect patients and position itself as a competitive AI hub.