AI is poised to revolutionise every aspect of healthcare. It will transform every stage of the therapeutics value chain, including the discovery, manufacturing, and repurposing of drugs. It will enable better diagnostics, and faster, more precise clinical decision-making and AI-guided precision treatments. And it will play a valuable role in preventing ill health.
Right now, we have only just begun to unlock the incredible power of AI models trained on digital health data to transform healthcare. But patients and medical professionals are already voicing justified concerns about whether AI is being delivered effectively and ethically. In healthcare, we need to work even harder than in other domains to ensure that AI tools we use are effective and trustworthy – lives are at stake. On the one hand this requires us to develop AI methods that are not “off-the shelf” but from the ground up designed for patient-ready healthcare use in mind.
The comprehensive datasets available here in the UK under public stewardship provide an important contribution to this. It makes the country an indispensable partner to the US and other countries in helping driving forward healthcare AI.
US and UK synergies
The U.S. leads the world in healthcare innovation due to its market size and access to extensive funding opportunities such as venture capital, not to mention its renowned universities and its technology and healthcare giants. While the UK compares well to the US in some of these respects, such as its world-leading science, there is one respect in which the UK is far ahead of every other country – the national scale of its healthcare system and the data it generates every day.
As a professor working to advance AI for healthcare technologies, I collaborate daily with the UK’s National Health Service (NHS), which offers free healthcare at the point of use for over 65 million people and operates under a relatively centralised management system. This creates the opportunity to generate, regulate and work on dataset that is unprecedented in scale and therefore far less biased than comparable data from the US, where access to healthcare is often limited by economic competition between healthcare providers and economic barriers to patients. In the UK, regions such as London (e.g. via the HDR Discover Now Hub), Wales and other metropolitan areas have already established secure data environments containing medical and healthcare data for millions of patients. These secure data environments enable researchers and developers to work on healthcare data within the NHS but without being able to take the data “away”.
What makes the NHS data especially powerful is its comprehensive linkage between primary care (GP) records and secondary care (hospital) records. This interconnected dataset allows for an unprecedented ability to track patients throughout their healthcare journey, providing not just short-term insights into diagnostic efficacy or short-term treatment outcomes but also the ability to evaluate long-term impacts and trends, include follow-ups and long-term disease management e.g. for chronic conditions.
The UK has a longstanding tradition of creating and maintaining secure large-scale datasets beyond healthcare records that are invaluable for research. Among these is the UK Biobank, the largest resource of its kind in the world. This vast repository contains in-depth genetic, imaging, lifestyle, and health information from over 500,000 UK participants. The biobank’s ability to link this data to genetic and imaging information, as well as lifestyle factors, offers an unparalleled opportunity for AI-driven healthcare innovations.
AI factories for large-scale datasets
One reason that AI systems often perform better than humans is the sheer volume of data they are trained on. Our AI Clinician, a system we have implemented in several UK hospitals to offer semi-autonomous treatment of sepsis in intensive care units, uses records from 10,000s of patients to learn optimal treatment strategies and hour-by-hour precision dosing recommendations – in comparison to the few 1,000 patients a human clinician may see in their lifetime. This allows our AI Clinician to chart more effective courses of care than clinicians would without it.
While it was possible to build the AI Clinician using US healthcare data, it’s only thanks to UK data that we can embark on even more ambitious project, known as Nightingale AI – an AI foundation model that can leverage the treasures of electronic patient records, biomedical data, and published medical literature to develop an advanced health-focused AI model. We have partnered with Europe’s first true AI factory, the Isambard-AI supercomputer to enable us to develop methods that allow us to train generative AI at scale on secure medical data.
With the ability to read and understand X-rays, imaging, electrocardiograms, genetic data, electronic health records, doctors’ letters, data from wearables, clinical trial results and a range of other multimodal life sciences data, Nightingale AI can be used flexibly to support a wide range of healthcare applications that include medical research, clinical decision-making, and drug discovery – in some respects like ChatGPT but trained from high quality, specialist biomedical data. Unlike ChatGPT it is not focussed on creating text, but on patient data most of which is not text based.
The NHS partners we are working with towards building Nightingale AI do more than just provide the sheer scale of data we need to train this model – it will help ensure the model is comprehensive, high-quality, and representative of our diverse UK population. And because UK’s health datasets benefit are under public stewardship, users can be confident we are using them in a way that aligns with public values. This is why we designed Nightingale AI as a platform technology that learns from healthcare data and allows others who want to build onto this platform to query the medical knowledge the system has distilled from the data – providing a novel, knowledge access point instead of direct access to a far more sensitive data access point.
The UK as a development testbed for digital health
While it’s important that countries develop their sovereign AI capacities, allowing them to retain control over models, data, compute and governance – we do not believe this means to have to do it alone. For Nightingale AI we have forged partnerships within the UK (such as the AI Hub in Generative Models), with the Children’s Hospital of Orange County and Grady Health in California, and Europe through the DVPS Horizon Europe program. The pragmatism, reputation and agility of the UK healthcare regulators and the easy integration of clinical trials into NHS operations as supported by the UK’s NIHR – enables us to co-create, develop, validate and deploy medical technology by bringing together patients, clinicians and developers together from Day One. .A transatlantic alliance will make it possible to develop healthcare AI applications that are not only incredibly powerful but also ethical and robust. By combining the U.S.’s innovation capabilities for scaling up healthcare technology with the UK’s comprehensive healthcare data for rapid prototyping and initial validation, we can create AI-driven solutions that will transform healthcare on a global scale.
