London: Scientists have developed a groundbreaking artificial intelligence model named Delphi-2M that can forecast people’s health problems more than a decade before they occur.
The technology has been trained to analyse patterns in anonymous medical records and calculate the risk of developing over 1,231 diseases, functioning much like a weather forecast that gives probabilities rather than exact outcomes.
Delphi-2M, which is based on the same generative AI principles as language models like ChatGPT, does not provide exact dates for events such as a heart attack, but instead estimates probabilities, for example, a 70 percent chance of developing a specific disease within a given timeframe.
What if you could get a glimpse of the future of your health, today?
Our scientists have developed a new generative AI model, trained using large-scale health records, that can estimate how human health may change over time.
Watch to find out more. pic.twitter.com/QCYNSS1DVo
— EMBL-EBI (@emblebi) September 17, 2025
According to Prof. Ewan Birney, interim executive director of the European Molecular Biology Laboratory (EMBL), this ability to predict multiple diseases simultaneously represents an unprecedented leap in healthcare forecasting: “And we can do that not just for one disease, but all diseases at the same time; we’ve never been able to do that before. I’m excited.”
Delphi-2M was trained using anonymised health data from over 400,000 participants in the UK Biobank, which included hospital admissions, GP records, and lifestyle information such as smoking habits.
It was then validated against additional Biobank data and tested using the records of 1.9 million people in Denmark, where its accuracy proved strong. Prof. Birney explained that when the model predicts a one-in-10 risk of a disease within a year, the real-world outcomes align closely with that probability.
The Delphi-2M is particularly effective at predicting diseases with clear progression patterns, such as type 2 diabetes, heart attacks, and sepsis, but less effective with random events like infections.
Its potential applications are wide-ranging, from offering targeted preventive treatments and lifestyle guidance to supporting national health services in anticipating medical demand, such as projecting the number of heart attacks expected in a specific city by 2030.
Although still in the research stage and not yet ready for clinical use, the AI tool could eventually guide preventive medicine, enhance screening programmes, and personalise healthcare strategies. However, Prof. Birney cautioned that its implementation must be carefully tested and regulated: “Just to stress, this is research, everything needs to be tested and well-regulated and thought about before it’s used, but the technology is here to make these kinds of predictions.”
The Professor compared its trajectory to genomics, which took around a decade to move from scientific validation to routine clinical application. The study, published in the scientific journal Nature, is the result of collaboration between EMBL, the German Cancer Research Centre (DKFZ), and the University of Copenhagen.
Prof. Moritz Gerstung, head of AI in oncology at DKFZ, noted that generative models like Delphi-2M could one day enable personalised care at scale, while Prof. Gustavo Sudre, an AI and neuroimaging researcher at King’s College London, praised the research as a major step toward scalable, interpretable, and ethically responsible predictive modelling in medicine.
ALSO READ | Microsoft backs UK AI with major investment in tech deal
