MHRA adopts AI framework proposed by UK docs
The Medicines and Healthcare products Regulatory Agency (MHRA) says the UK government has adopted a list of recommendations on the use of artificial intelligence (AI) in medical devices developed by an independent group of doctors in the National Health Service (NHS). The agency also said it is initiating the third phase of its AI Airlock, a regulatory sandbox for evaluating new AI products.
In September, the National Commission into the Regulation of AI in Healthcare, established by the MHRA, published a report with 44 specific recommendations for adapting to the new technology. It concluded that the UK's current regulatory system needs to evolve to keep pace with developments in AI technologies by adopting a more proportionate regulatory framework and a lifecycle approach to the products it allows on the market. It includes recommendations such as improving the device qualification and classification pathways, allowing flexibility in device iterations and updates, harmonizing regulations, and creating more educational resources and tools.
"The independent National AI Commission, led by NHS doctors, has given us a clear and ambitious blueprint for how we make the UK the best place to get AI products to patients safely, quickly and in a way that maintains public trust," said MHRA CEO Lawrence Tallon. "I am pleased that today the Government is accepting its recommendations in full, whilst setting out a clear plan and oversight structure to deliver them."
The report also made the case for greater use of regulatory sandboxes. Heeding the report’s call, MHRA also said it is seeking volunteers to join the third phase of its AI airlock, which was initially launched in 2024. The latest phase will focus on postmarket surveillance of AI-enabled medical devices and track their evolution throughout their lifecycle.
The recommendations proposed by the National AI Commission have some overlap with the machine learning guiding principles developed by MHRA, the US Food and Drug Administration (FDA), and Health Canada. However, Alastair Denniston, a professor in the Department of Applied Health Sciences at the University of Birmingham and Chair of the National AI Commission, told Focus that the commission's recommendations go well beyond the guiding principles in both breadth and depth. (RELATED: Regulators release 10 principles for good machine learning practice, Regulatory Focus 27 October 2021)
Denniston emphasized that the commission's recommendations were developed with input from a broad range of experts and stakeholders who provided important feedback on how the UK's AI-enabled health care should look in the future to ensure it is safe, fast, and trusted. More specifically, he noted that it involved input from 12,000 people from across the UK, including about 10,000 members of the public, and 2,000 health care practitioners and experts.
"We tried to focus on the [recommendations] that would be most critical to achieving that overarching goal of a regulatory framework that was safe, fast, trusted, and person-centered," said Denniston. "Interestingly, whilst there is a clear UK context to some of these, overall I think all international regulators are facing the same challenges – and that’s where it was great to have senior US, Singaporean, and other international experts on our Commission.
"I think that both the overarching principles and the 44 recommendations that sit beneath it are worth digging into, wherever you’re based," he added.
Denniston said the commission tried to consider what it would take in practice to create the type of AI regulatory framework they envisioned. For example, he said that they examined what it means for various kinds of AI technology if their focus is on developing a faster, more risk-proportionate regulatory framework.
"Given that these technologies can be highly operator- and context-dependent, how do we ensure that each actor in the delivery and use of these technologies know their role and responsibilities?" he asked. "How do we ensure that staff are competent and confident to use these technologies? How do we share information more effectively across the system? And, perhaps most important of all, what matters most to patients when AI is being used in their healthcare?
"One of the consistent messages from across all sectors, including the public, is that this wasn’t a question of whether we should be using AI, but rather about how we should be using AI," he added.
Denniston said that, while there's been broad acknowledgment that AI has the potential to improve the quality and experience of healthcare in the UK, as well as efficiency and cost, the existing regulatory system was not designed for the technology, leading to the creation of the National AI Commission to bring stakeholders together to discuss what a modern regulatory system for AI should look like. He highlighted that it took only four weeks for the MHRA and the UK Government’s Department of Health and Social Care to adopt the commission's recommendation.
Denniston also discussed MHRA's AI Airlock, stating that it has been a valuable mechanism for exposing regulatory gaps, exploring potential solutions, and testing them. He noted that as AI technology becomes increasingly complex, especially when considering GenAI and agentic systems, other countries are also exploring their own regulatory sandboxes.
"It is quite interesting to see different ways in which it is being applied, ranging from primarily exploring the regulatory issues to using it as an actual technology evaluation mechanism," said Denniston. "I think it is important to recognize that not every airlock or sandbox is the same, but I think they can all be valuable.
"The UK’s AI Airlock is certainly helping inform the MHRA’s regulatory approaches to these newer technologies," he added.
National AI Commission report, MHRA announcement, AI Airlock
