AI Model 'Foresight' Trained on 57 Million NHS Health Records to Predict Future Illnesses

Medical AI trained on whopping 57 million health records

A generative artificial-intelligence (AI) model has been let loose on an entire nation's health records for the first time. The model, called Foresight, predicts hospitalizations, heart attacks and hundreds of other conditions, and researchers trained it on deidentified data from 57 million patients in England's National Health Service (NHS). Foresight can currently be used only for research related to COVID-19, which must be done in a secure 'data environment' operated by the NHS. But if the AI's predictions prove valuable in diverse populations, Foresight could eventually guide individual patient care and help apportion NHS resources, Angela Wood, a health-data scientist at the University of Cambridge, UK, said at a press briefing on 6 May. "This is the first time an AI model has been used within health research on 57 million people. This is a real step forward," she added. The development comes as AI is increasingly being integrated into medical research through models that diagnose diseases and perform other medical tasks -- in some cases better than humans. A previous iteration of the Foresight was trained on health records from around 1.5 million people in London and tested on its ability to determine future diagnoses when presented with a patients' medical history and made correct judgements most of the time. The model is based on several data sources, including hospital and vaccination records, general practitioners' (GP) visits and a UK national death registry. The latest version was trained on data from 2018 to 2023, totalling around 10 billion medical 'events'. These data were stripped of identifying information, such as names, addresses and birthdates. As added precautions, the model can be run only on NHS computer systems, and any research predictions it generates will be screened before release, Michael Chapman, the director of data access at NHS England, said at the briefing. The NHS allows individuals to request that GP records not be used for research or planning, but people cannot withdraw other types of health data from the model, Chapman added. But it might be difficult to make it impossible to extract private patient data from the model, Luc Rocher, a data-privacy researcher at the University of Oxford, UK, said in a statement to the UK Science Media Centre, which organized the press briefing. "The very richness of data that makes it valuable for AI also makes it incredibly hard to anonymize. These models should remain under strict NHS control where they can be safely used." Foresight is currently limited to use in around 100 existing projects related to COVID-19 that are part of a collaboration between the British Heart Foundation and NHS England to study the effects of the pandemic, such as those caused by cancelled medical appointments and other disruptions, using patient data. Any research use beyond this would require extra approvals, Chapman said. Researchers will also test how well Foresight can predict which of around 1,000 conditions a patient will develop in 2023, given their medical history in 2018-22. "That allows us to actually get as close to a ground truth as is possible," Chris Tomlinson, a health-data scientist at University College London who is leading the evaluation, said at the briefing. Moritz Gerstung, a health data scientist at the German Cancer Research Centre in Heidelberg, says the sheer scale of NHS data will make for more accurate healthcare AIs, which are better able to predict uncommon conditions or interpret atypical medical histories. He has applied for permission to train a model his team developed, called Delphi, on NHS data. "It's very exciting, the potential NHS data offers in training such generative health models."

Concerns raised over AI trained on 57 million NHS medical records

An artificial intelligence model trained on the medical data of 57 million people who have used the National Health Service in England could one day assist doctors in predicting disease or forecast hospitalisation rates, its creators have claimed. However, other researchers say there are still significant privacy and data protection concerns around such large-scale use of health data, while even the AI's architects say they can't guarantee that it won't inadvertently reveal sensitive patient data. The model, called Foresight, was first developed in 2023. That initial version used OpenAI's GPT-3, the large language model (LLM) behind the first version of ChatGPT, and trained on 1.5 million real patient records from two London hospitals. Now, Chris Tomlinson at University College London and his colleagues have scaled up Foresight to create what they say is the world's first "national-scale generative AI model of health data" and the largest of its kind. Foresight uses eight different datasets of medical information routinely collected by the NHS in England between November 2018 to December 2023 and is based on Meta's open-source LLM Llama 2. These datasets include outpatient appointments, hospital visits, vaccination data and records, comprising a total of 10 billion different health events for 57 million people - essentially everyone in England. Tomlinson says his team isn't releasing information about how well Foresight performs because the model is still being tested, but he claims it could one day be used to do everything from making individual diagnoses to predicting broad future health trends, such as hospitalisations or heart attacks. "The real potential of Foresight is to predict disease complications before they happen, giving us a valuable window to intervene early, and enabling a shift towards more preventative healthcare at scale," he told a press conference on 6 May. While the potential benefits are yet to be supported, there are already concerns about people's medical data being fed to an AI at such a large scale. The researchers insist all records were "de-identified" before being used to train the AI, but the risks of someone being able to use patterns in the data to re-identify the records are well-recorded, particularly when it comes to large datasets. "Building powerful generative AI models that protect patient privacy is an open, unsolved scientific problem," says Luc Rocher at the University of Oxford. "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." "The data that goes into the model is de-identified, so the direct identifiers are removed," said Michael Chapman at NHS Digital, speaking at the press conference. But Chapman, who oversees the data used to train Foresight, admitted that there is always a risk of re-identification: "It's then very hard with rich health data to give 100 per cent certainty that somebody couldn't be spotted in that dataset." To mitigate this risk, Chapman said the AI is operating within a custom-built "secure" NHS data environment to ensure that information isn't leaked out of the model and is accessible only to approved researchers. Amazon Web Services and data company Databricks have also supplied "computational infrastructure", but can't access the data, said Tomlinson. Yves-Alexandre de Montjoye at Imperial College London says one way to check whether models can reveal sensitive information is to verify whether they can memorise data seen during training. When asked by New Scientist whether the Foresight team had conducted these tests, Tomlinson said it hadn't, but that it was looking at doing so in the future. Using such a vast dataset without communicating to people how the data has been used can also weaken public trust, says Caroline Green at the University of Oxford. "Even if it is being anonymised, it's something that people feel very strongly about from an ethical point of view, because people usually want to keep control over their data and they want to know where it's going." But existing controls give people little chance to opt out of their data being used by Foresight. All of the data used to train the model comes from nationally collected NHS datasets, and because it has been "de-identified", existing opt-out mechanisms don't apply, says David Harris at NHS England, part of the Foresight team, though people who have chosen not to share data from their family doctor won't have this fed into the model. Under the General Data Protection Regulation (GDPR), people must have the option to withdraw consent for the use of their personal data, but because of the way LLMs like Foresight are trained, it isn't possible to remove a single record from an AI tool. Harris says that "as the data used to train the model is anonymised, it is not using personal data and GDPR would therefore not apply". Exactly how the GDPR should address the impossibility of removing data from an LLM is an untested legal question, but the UK Information Commissioner's Office's website states that "de-identified" data should not be used as a synonym for anonymous data. "This is because UK data protection law doesn't define the term, so using it can lead to confusion," it states. The legal position is further complicated because Foresight is currently being used only for research related to covid-19, says Tomlinson. That means exceptions to data protection laws enacted during the pandemic still apply, says Sam Smith at medConfidential, a UK data privacy organisation. "This covid-only AI almost certainly has patient data embedded in it, which cannot be let out of the lab," he says. "Patients should have control over how their data is used." Ultimately, the competing rights and responsibilities around using medical data for AI leave Foresight in an uncertain position. "There is a bit of a problem when it comes to AI development, where the ethics and people are a second thought, rather than the starting point," says Green. "But what we need is the humans and the ethics need to be the starting point, and then comes the technology."

AI trained on de-identified patient data to predict health care needs in pilot study

Foresight, a generative AI model, learns to predict what happens next based on previous medical events. It's similar to models like ChatGPT, which predicts the next word in a sentence based on what it's seen previously from data across the internet. Foresight is being trained on routinely collected, de-identified NHS data, like hospital admissions and rates of COVID-19 vaccination, to predict potential health outcomes for patient groups across England. This could be events such as hospitalization, heart attacks or a new diagnosis. Predicting these events early could enable targeted intervention, shifting towards more preventative health care at scale. The pilot study operates entirely within the NHS England Secure Data Environment (SDE), a secure data and research analysis platform that uniquely enables this groundbreaking work by providing controlled access to de-identified health data from the 57 million people living in England. Access to data at this scale is only made possible through the NHS England SDE, where both the AI model and all patient data remain under strict NHS control. By including data covering England's entire population, the model can make predictions about health outcomes across all demographics and for rare conditions. Dr. Chris Tomlinson, lead researcher from UCL, said, "AI models are only as good as the data on which they're trained. So if we want a model that can benefit all patients, with all conditions, then the AI needs to have seen that during training. Using national-scale data allows us to represent the kaleidoscopic diversity of England's population, particularly for minority groups and rare diseases, which are often excluded from research." Through rigorous approval processes, the British Heart Foundation Data Science Center at Health Data Research UK made it possible for the researchers to access and work in the SDE. The Center also involved members of the public, who continue to contribute to approving and shaping the research. The researchers believe the model's predictive power could pinpoint high-risk patient groups, opening up a window of opportunity to intervene to improve and save lives. Due to the diversity and completeness of the training data, the model could also help to highlight and address health care inequalities. And the ability to analyze health care risks and outcomes on a population level could offer critical support to the NHS when it comes to planning. Simon Ellershaw, an AI Ph.D. researcher from UCL, said, "Combining the computing resources needed for AI with NHS data has always been challenging, but thanks to the support of our partners we've been able to safely and securely apply state-of-the-art AI methods to NHS data at unprecedented scale." The pilot study is an opportunity to test the model in a secure and safe environment, protecting privacy, and all predictions are rigorously tested for accuracy against real-world outcomes. Currently the model is using recent data, from November 2018 to the end of 2023, for a limited number of datasets and made available for COVID-19 research. Professor Richard Dobson, Deputy Director of the NIHR Maudsley Biomedical Research Center and another lead researcher at King's and UCL, said, "This pilot is building on previous research that demonstrated Foresight's ability to predict health trajectories from data from two NHS trusts. To be able to use it in a national setting is very exciting as it will potentially demonstrate more powerful predictions that can inform services nationally and locally. "Currently, the data in this pilot is broad but shallow, and ultimately we'd like to harness the expertise and AI platforms behind Foresight by including richer sources of information like clinicians' notes, or results of investigations such as blood tests and scans if they become available." In the future, the researchers would like to train the model further on deeper data sources, going back further in time. They're also exploring how to responsibly expand the scope of the model, which is currently restricted to COVID-related research. However, the researchers are clear: patients and the public must be at the heart of any guidance developed around the model's use and predictions, to make sure this research and its applications are in their best interests. A BHF Data Science Center public contributor involved in reviewing and approving this project, said, "As a patient, I'm interested in how this research could help identify linked health conditions, reduce the risk of developing new ones, and support those who face challenges accessing health care. "It's important that people know how their health data is being used, so it's encouraging to see a focus on transparency and making sure AI is used in the NHS in a safe, ethical way with public benefit at its heart. Dr. Vin Diwakar, National Director of Transformation at NHS England, said, "AI has the potential to transform the way we prevent and treat disease, if trained on large datasets and safely tested. The NHS Secure Data Environment has been fundamental to this pioneering research, shaping a future where earlier treatments and interventions are targeted at those who will benefit, preventing future ill health. This will boost our ability to move quickly towards personalized, preventative care." Professor Angela Wood, Associate Director at the BHF Data Science Center at Health Data Research UK, said, "Harnessing the power of AI with NHS data at this scale represents a huge step forward for health data studies. "The Foresight model has the potential to support health care professionals to make timely and effective medical decisions for high-risk patients, and to help address health inequalities by ensuring everyone is represented in predictive health care. This pioneering work, enabled by the BHF Data Science Center, demonstrates the UK's leadership in the trustworthy use of AI to transform lives." Health data, AI, and the NHS The Government recently announced the development of a Health Data Research Service, designed to support secure access to data for health researchers. The service is designed to ensure that projects like Foresight, which securely and safely access data to drive innovation and improve patient lives, will be much easier to support. Health and Social Care Secretary Wes Streeting said, "Our Plan for Change is harnessing trailblazing AI to radically transform our NHS - while also protecting patient data with strict security procedures. I'm determined that we use this kind of groundbreaking technology to cut down on unnecessary hospital trips, speed up diagnosis times, and free up staff time. "AI will be central as we bring our analog NHS into the digital age to deliver faster and smarter care across the country." Chief Scientific Advisor for the Department of Health and Social Care Lucy Chappell said, "Using AI to drive innovation across the NHS is key in making strides for better care to patients. This study could allow earlier diagnosis and treatment, enabling people to better manage their health and care, while ensuring patient data is safeguarded. The NIHR is proud to be supporting this work through our world-leading infrastructure." Science and Technology Secretary Peter Kyle said, "This ambitious research shows how AI, paired with the NHS's wealth of secure and anonymized data, is set to unlock a health care revolution. This technology is transforming what's possible in tackling a host of debilitating diseases, from diagnosis, to treatment, to prevention. "This is work that will be instrumental to this Government's mission to overhaul health care and grow the economy, which sits at the heart of our Plan for Change. And an unrelenting focus on privacy and security means people can rest assured that their data is in safe hands."

AI to predict future illnesses using NHS patient data

Artificial intelligence will be used to predict future illnesses by analysing all NHS patients' data in a world-first study. The pilot will see a generative AI model, called Foresight, trained to predict disease by using the anonymised health data of 57 million people in England. The model will learn about the patterns that have led to ill-health and hospitalisation among people in England with the aim of being able to spot trends and predict people at risk of health issues such as heart attacks. It will work in a similar way to other AI models such as ChatGPT, which predicts the next word in a sentence based on the data it has accumulated from across the internet. By better understanding the predictors of disease, heart attacks and the causes of hospitalisations, it will be easier to intervene in an individual's life and take measures to prevent such an event from happening. The research is being led by experts from University College London (UCL) and King's College London (King's) and has the power to "save lives", they said. Dr Chris Tomlinson, lead researcher from UCL, said the AI model could play a key role in the Government's mission to shift to preventing illness rather than just treating it. Predicting diseases before they happen "AI models are only as good as the data on which they're trained. So if we want a model that can benefit all patients, with all conditions, then the AI needs to have seen that during training," he said. "Foresight is a really exciting step towards being able to predict disease and complications before they happen, giving us a window to intervene and enabling a shift towards more preventative health care at scale," he added. It could predict the risk of someone being hospitalised and "use Foresight to understand the drivers for that deterioration, and potentially suggest personalising opportunities for intervention so that might include, for example, optimising medications to improve blood pressure control and reduce the risk of, say, stroke", Dr Tomlinson said. Wes Streeting, the Health Secretary, said the Government was "harnessing trailblazing AI to radically transform our NHS - while also protecting patient data with strict security procedures". "I'm determined that we use this kind of groundbreaking technology to cut down on unnecessary hospital trips, speed up diagnosis times, and free up staff time," he added. "AI will be central as we bring our analogue NHS into the digital age to deliver faster and smarter care across the country." It has previously had access to a limited number of datasets for Covid-19 research from between November 2018 to the end of 2023, but is now being rolled out at "unprecedented scale". It comes after an investigation found last month that Chinese researchers were to be granted access to half a million GP records through UK Biobank, a research hub, despite MI5 fears that Beijing's regime could acquire sensitive information. However, those involved in the Foresight study are adamant the research follows strict security protocols with personal information stripped away. They said it was operating within the NHS England Secure Data Environment, which is a research analysis platform where the "de-identified" data from 57 million people exists. Peter Kyle, the Science Secretary, said "an unrelenting focus on privacy and security means people can rest assured that their data is in safe hands". The new pilot comes after a study, published in the Lancet Digital Health in March 2024, found Foresight was able to predict the type of health conditions a patient is likely to develop in the future. Professor Richard Dobson, deputy director of the NIHR Maudsley Biomedical Research Centre and another lead researcher at KCL and UCL, said: "This pilot is building on previous research that demonstrated Foresight's ability to predict health trajectories from data from two NHS trusts. "To be able to use it in a national setting is very exciting as it will potentially demonstrate more powerful predictions that can inform services nationally and locally."