New AI healthcare tools should be given provisional “L-plates” style approval until manufacturers can demonstrate that they are safe and effective in real-world use, according to recommendations from the UK’s biggest consultation on health technology regulation.
The proposals, announced on Thursday, also call for continuous monitoring of AI tools after approval and easier public access to information about safety incidents, as regulators grapple with technology that can continue to change after it enters clinical use.
The National Commission into the Regulation of AI in Healthcare was launched last year by the UK medicines regulator MHRA, as it sought to develop a regulatory framework for the new era of generative AI tools.
Led by two NHS doctors, its conclusions were based on evidence from 12,000 people including patients, clinicians, healthcare leaders and technology developers. The MHRA will consider the recommendations and issue a formal response as soon as possible.
Lawrence Tallon, MHRA chief executive, said the UK’s ambition was “to become the best place in the world for innovators to build and test AI, for healthcare professionals to use it, and for patients to benefit from it in their care”.
But Jennifer Dixon, chief executive of the Health Foundation, which conducted research into public attitudes towards the use and regulation of AI in healthcare that helped shape the Commission’s recommendations, said “the real test will be whether the NHS has the capacity, skills and systems in place to implement and monitor AI applications safely and effectively at the scale now needed”.
Tallon said AI differed from previous generations of medical devices because it continued to adapt and evolve after it had been licensed as more data was put into it.
“Now there are positive aspects to that, let’s say learning and improvement, [but] potentially there are negative aspects, let’s say drift, where an AI starts to not behave in the way you want it to, and in either event, you need to be able to continuously monitor what’s happening with that AI after the product approval,” he added.
The Commission examined approaches being developed by regulators in the US, Canada and Japan. It recommended a searchable source of safety information on AI-enabled medical devices, including reports of adverse incidents.
A majority of participants believed that patients should be informed when AI had been used in their care, according to research cited by the Commission.
Under the current framework, there was “no universal requirement or expectation” for healthcare professionals to routinely give patients that information. However, “while routine disclosure of every AI use may not always be practical, clear and proportionate communication is essential to maintaining patient and public trust”, it added.
The Commission also recommended stronger MHRA enforcement powers so it could act “decisively and quickly” against AI systems that fell short of required standards.
With the World Health Organization estimating a global shortage of 11mn healthcare workers by the end of the decade, Tallon argued that the UK could not rely on expanding its workforce. “We need to use technology and AI to be able to produce faster, more accurate healthcare with the people we’ve already got,” he added.
The MHRA said in a statement that it had “heard the messages of the Commission, and plans to maintain the momentum generated across the health system, responding in full as soon as possible to give industry, patients and professionals clarity on the direction of travel”.
