European researchers say a new AI model can forecast an individual’s risk for more than 1,000 conditions years in advance, offering probability-style health predictions akin to weather reports. Trained on anonymized records from over 400,000 UK Biobank participants and validated on 1.9 million Danish records, the system—Delphi-2M—performed best on progressive illnesses such as Type 2 diabetes, cardiac events and sepsis. The work, published in Nature, underscores how large language model techniques can be adapted to longitudinal medical data to identify who may benefit from early intervention and to help hospitals plan capacity. Scientists caution the tool isn’t ready for clinical deployment, citing demographic biases in the training set and the need for regulatory oversight; future versions aim to integrate imaging, genetics and lab results. Backers say adoption could follow the path of genomics, taking years to move from research confidence to routine care.
Related articles:
International evaluation of an AI system for breast cancer screening
Cardiovascular risk factors from retinal fundus photographs via deep learning
A targeted real-time early warning score (TREWS) for sepsis identification
Ethics and governance of artificial intelligence for health
Highly accurate protein structure prediction with AlphaFold
