A key focus for the MHRA as we develop our forthcoming strategy is enabling safe and timely access to innovative technologies, while maintaining public trust. This includes strengthening our approach to regulating adaptive AI, enhancing both pre-market evaluation and robust post-market surveillance, and ensuring that safety, performance and equity remain central as technologies evolve in real-world settings.In this blog, Dame Jennifer Dixon, CEO of the Health Foundation, highlights both the opportunities and challenges ahead. Through collaboration and the ongoing work of the National Commission for the Regulation of AI in Healthcare, she underscores the importance of strengthening evaluation, oversight and system-wide approaches to support safe, effective and equitable adoption of innovation.
Foreword by Jennifer Dixon
It’s too easy to luxuriate in a doom funk in the UK. Just look at the latest OBR projections on sluggish economic and productivity growth, the ONS projections of a shrinking and ageing population, the size of the NHS waiting list, the rising burden of morbidity and the marked inequalities in health across the population. And that’s just for starters.
But flip all this on its head. Look at our assets and the opportunities ahead instead. We are a rich country (overall), with stable, solid institutions and with an enviable track record in science, innovation and public service, amongst many other things.
The key question is how to seize growing opportunities now and ahead of us, in particular from technology and artificial intelligence. And at a speed to keep health care affordable, accessible and good quality for all into the future.
I am often struck by how complex and fragmented health care is in other countries. We have a national system, where coordinated strategic direction is more possible – such as for investment, regulation, payment incentives, price control, standards of clinical care and access, and of course across data infrastructure and cybersecurity. Where 30 years ago the NHS might have seemed to reformers like a large elephant that was difficult to move, now it might just be that a national system with good coordination could come into its own as we try to respond and get ahead of very rapidly developing technology to benefit the UK population.
Take AI and regulation. I’m fortunate to be on the National Commission for the Regulation of AI in Healthcare where we’ve been discussing, among other things, how to respond to artificial intelligence applications that are adapting in the real world, not static. In the purview of the MHRA are questions such as: is the AI model safe and accurate? Does it continue to be as it adapts in a real-world setting? And how does it compare to other similar AI applications? Our goal is to ensure AI is safe and accurate for all, and that no bias creeps in that could lead to unfair and suboptimal care. Because of all this the Commission is focusing a lot on how to develop better pre-market testing and good post-market surveillance – all very important.
But as we know, AI and tech when used in the real world are not just a ‘product’ – a bounded technical thing - but more a ‘socio-technical’ activity used by and for humans. ‘Socio’ questions on implementation include: is the AI application useable (as intended), acceptable and fair (to clinicians and patients) and safe (or were there unsafe or biassed workarounds in practice etc)? And there are wider questions such as: did it shift care into other settings? Was there a return on investment? Did it save clinical or patient time? Or even, is there likely to be a large disruption on jobs to be anticipated and planned for? These are questions which may well be beyond the scope of MHRA. But they are crucially important if we are all to benefit.
These immediate implementation and wider questions are more the purview of research evaluations, many of which are funded by NIHR and supported by its many research centres. In my experience, while formal (summative) research in the NHS is generally conducted well (if not always rapidly or cheaply), the more formative local service evaluations can be of very mixed quality. Given the increasing need for cheap, fast and at scale formative evaluations of AI applications, plus the linked need for MHRA real world ongoing post-market surveillance, this whole area of evaluation needs to be strengthened, standardised and formalised to pass muster with clinicians and patients. Because many local implementation evaluations in the NHS are weak there are often strong incentives to overclaim the results.
Clearly these assessments of new technologies must include scrutiny of equity and bias, so that all patients can benefit. But as we develop this new system of AI governance, including evaluation and post-market surveillance, it’s also true that current access to health care is suboptimal – neither perfectly equitable nor perfectly safe. So, the big task ahead is to get better at demand signalling what are the highest priority technologies to be developed and tested, test them better, faster and more cheaply, and work out effective ways to spread them safely. All while keeping public and clinical support and trust high. This agenda is the number one priority to reform care in the UK.
seen at 11:50, 14 April in MHRA Inspectorate.