select search filters
briefings
roundups & rapid reactions
Fiona fox's blog
Scientists comment on Foresight AI, an AI model trained on de-identified NHS data to predict healthcare needs.
Dr Junade Ali CEng FIET, Consultant Software Engineer and Fellow at the Institution for Engineering and Technology, said:
“It’s exciting to see generative AI technology combined with healthcare data to improve healthcare quality. This opens the door to large-scale advances in improving human health.
“This is a bespoke model created for healthcare use rather than off-the-shelf solutions like ChatGPT. Therefore, it is important to provide a ‘health warning’ about using more general-purpose AI assistants to attempt to predict the future, especially when human behaviour is involved.
“As with every advance of AI technology, we need to be mindful that AI is not yet a substitute for human decision-making and connection.”
Dr Prasanth Kamma, Lead Architect at the Center of Excellence and Fellow at the Institution of Engineering and Technology, said:
“This is the first time a predictive system has been trained on the entire health footprint of a nation, with about 57 million patient records. That level of scale and inclusion allows the model to reflect how people actually move through care, across different regions, age groups, and health conditions. Most predictive systems are built using limited or siloed data. This one is based on real population-level journeys, which makes its predictions far more practical and relevant.
“Foresight does not just predict who might get sick, but reveals where care is delayed, inconsistent, or missing entirely. That changes how the NHS can respond. The model helps identify which communities are being overlooked, allowing support to reach people earlier. It represents a shift from reactive healthcare policy to a more proactive and preventative model.
“The model currently uses broad data such as diagnoses and admissions, but does not yet include clinical notes, test results, or social factors, all of which influence outcomes. Expanding these data inputs will improve accuracy. Even though the data is de-identified, trust must be actively maintained. Transparency, clear governance, and accountability are just as important as technical performance.
“This is more than a research initiative. It is a national example of how healthcare data can be used with care, security, and public value at the forefront. The NHS is showing that meaningful innovation can happen without compromising trust. As someone who leads similar data driven AI platforms in the United States, I see this as a responsible model that many countries can learn from.”
Prof Peter Bannister, Fellow and Healthcare expert at the Institution of Engineering and Technology, said:
“Training a generative AI model on a national, diverse dataset representing patient journeys across the whole of the NHS is an important step preventative and sustainable models of healthcare delivery.
“However, at this early stage, it is necessary to define how commercial organisations – essential partners in developing and deploying new healthcare technology as well as those who provide treatments – will be able to interact with the model, both during initial development and ongoing improvement of arising innovations.
“More fundamentally, improved prediction alone will not drive better health: cutting-edge engineering alongside real-world testing of these complex solutions will be required to ensure these early indicators can in fact enable better treatment decisions, particularly for the most disadvantaged in society.”
Rimesh Patel CEng, Former Chair of the Central London Network for the Institution of Engineering and Technology and Independent Cyber Security Specialist, said:
“The brief’s main goal is to use de-identified NHS data to perform ‘predictive’ next steps for medical triage for the patient across different health conditions. This is a novel endeavour for humanity on using emerging state of the art technology to fast-track research outcomes. With government approval, this could be a watershed moment in public healthcare, hence the significance of this approach.
“This is an industry defining moment, however what is not clear are the ancillary outputs beyond how the data will be used to optimise ‘patient medical needs’, where the brief mentions “the model could also help to highlight and address healthcare inequalities. And the ability to analyse healthcare risks and outcomes on a population level could offer critical support to the NHS when it comes to planning”, therefore, this is a pilot study where the stakeholders need to have a conservative approach so that grey areas of abuse are not in the equation of the pilot. The ‘study’ should clarify it’s hypothesis (i.e. to serve underrepresented communities, identify opportunities to make the medicines more accessible, identify commercial opportunities, identify next steps in patient treatment, etc.) so that the scope of data required be limited to that goal only to alleviate any bias or abuse.
“In regards to the potential limitations of Foresight AI, from a technology perspective, there is mention of secure data environments (SDE) which will be key to ensure the pilot is not open to abuse of data, but this will still depend on how the data is assessed and under what control mechanisms.
“The product itself (Foresight) should make information available on how it will consume data and if it can identify stages of analysis where researchers can intervene and sanitise any data records when required. It should also clarify how any areas of bias factors have been identified and remediated.”
“To add, it is worth mentioning that the use of the word ‘de-identify’ is a different category of reference to current GDPR laws on ‘sensitive data’ and the pilot should identify the need for it and how it differs to achieve the goal of “This will boost our ability to move quickly towards personalised, preventative care.”
Dr Luc Rocher, Senior Research Fellow, Oxford Internet Institute (OII), University of Oxford, said:
“The scale and richness of NHS data required to train generative AI models makes ‘de-identifying’ such complex patient information notoriously challenging. De-identification carries a significant risk that patterns remain which could, inadvertently, lead back to individuals.
“Building powerful generative AI models for healthcare that protect patient privacy is an open, unsolved scientific problem. The very richness of data that makes it valuable for AI also makes it incredibly hard to anonymise. These models should remain under strict NHS control where they can be safely used.”
Dr Omid Rohanian, Senior Research Associate, Department of Engineering Science, University of Oxford, said:
“The Foresight AI initiative is a promising step forward in using AI to enhance healthcare. By training on de-identified NHS data at such a large scale, it has the potential to identify health issues early and strengthen preventative care across England and beyond. However, its success depends on addressing key challenges: ensuring the model is privacy-aware, fair, and free from biases that could impact certain groups; making its predictions transparent and understandable for clinicians; and continually validating its accuracy in real-world settings. If these challenges are addressed, projects like Foresight could transform how we approach healthcare and make a real difference in people’s lives.”
Dr Wahbi El-Bouri, Senior Lecturer in Digital Twins and In Silico Trials, University of Liverpool, said:
“The NHS is uniquely positioned in the world as having centralised healthcare data across a diverse, large population. Exploitation of this data is vital to help improve health outcomes in the UK. There are numerous projects that are trying to use local NHS Trust data to develop similar AI models, but Foresight goes beyond that by attempting to develop these models over all Trust data in England.
“Developing these AI models requires good quality data. Researchers who have worked with NHS data will know that data quality can often be poor, with large amounts of missing data, or incorrect reporting. This will be a fundamental problem the project will come up against, as the AI model can only be as good as the data input.
“I should also note that prevention of disease is a key goal of the NHS to reduce pressure on the health service. While such a project may go some way towards this goal by predicting what may happen next for a patient, for example if they are at higher risk of a heart attack, it does not tackle real prevention of illness. NHS data is the wrong type of data to tackle prevention as when someone has visited the NHS it is because something is already wrong. As a result, we miss out on learning from healthy individuals whose data is rarely collected.”
The announcement of the development of Foresight AI is embargoed until 00:01 UK time Wednesday 7th May, further information is detailed in the HDRUK/ BHF Data Science Centre press release. There will also be a recorded SMC briefing on the announcement at 10:30 UK time Tuesday 6th of May.
Declared interests:
Dr Prasanth Kamma: “I am currently working as Principal Architect at CVS Health, where I lead the design of predictive AI healthcare technology platforms at scale. I am also a Fellow of the Institution of Engineering and Technology (FIET). I have no financial, advisory, research, or employment relationship with the Foresight AI project, NHS England, University College London, King’s College London, or Health Data Research UK.
“I have not received any funding or compensation related to this initiative and have no commercial interests in its outcome. My comments are given independently, based on my experience working with predictive systems in healthcare.”
Prof Peter Bannister: “Digital Health, Diagnostics and Imaging Product Strategy Expert and HealthTech, Artificial Intelligence, and Sustainability Policy and Delivery”
Dr Junade Ali CEng FIET: “No conflicts of interest.”
Rimesh Patel CEng: “I confirm I have no Industry funding or links particularly important to declare. I confirm I was the former Chair of the IET Central London Network where I organised monthly lectures that discussed ALL trending technology topics in the spirit of innovation. I confirm I am the director of my own company, SAIBER LTD, that has held multiple panel discussions, including on the topic of ‘A.i and Cyber Security in the medical field’ 3 years ago as stated here https://saiber.uk/ which had no funding or commercial outputs, it gave the opportunity to facilitate a professional discussion for the subject area. I confirm that SAIBER LTD has not received any fundings to research any topics. I confirm I have written 3 technical papers available as part of the IET Digital Library on the links below: https://digital-library.theiet.org/doi/10.1049/etr.2015.0054 (Cryptography) https://digital-library.theiet.org/doi/10.1049/etr.2016.0132 (Security Transformation; SOC Programme Management) https://digital-library.theiet.org/doi/10.1049/etr.2016.0189 (Application of defence in depth and diversity)”
Dr Luc Rocher: “I don’t have any COI” “In the past five years, my work has been financially supported by UK and Belgian taxpayers, by UK Research and Innovation (UKRI), Engineering and Physical Sciences Research Council (EPSRC), Innovate UK, the Information Commissioner’s Office (ICO), the John Fell Fund from Oxford University Press, and the Belgian National Fund for Scientific Research (F.R.S.-FNRS).”
Dr Omid Rohanian: “I am a Senior Research Associate at the University of Oxford, working on clinical machine learning projects including large-scale annotation tools for biomedical research. I have no conflicts of interest or industry ties to declare.”
Dr Wahbi El-Bouri: “Received research funding from UKRI, Horizon Europe, and the Royal Society; work as an advisor for a startup developing biosensing technology (Performr).”