AI can forecast your future health, just like the weather (www.bbc.com)

Researchers unveiled Delphi-2M, an AI model that can forecast an individual’s risk of 1,231 diseases years into the future by learning patterns in anonymized medical records. Built with language-model-style training (similar technology to ChatGPT) it predicts “what comes next” in a patient’s health timeline rather than exact dates—outputting calibrated probabilities like a weather forecast (e.g., a 70% chance of an event). The model was trained on UK Biobank data from over 400,000 people (GP records, hospital admissions, lifestyle) and validated against other biobank participants and 1.9 million Danish records, where predicted risks matched observed outcomes well for progressive conditions such as type 2 diabetes, heart attacks and sepsis. The work, published in Nature and developed by EMBL, DKFZ and University of Copenhagen, could help flag high‑risk patients for early intervention, personalize screening, and forecast local healthcare demand. Key technical implications include scalable, multi‑disease risk estimation and good calibration across cohorts, but important limitations remain: the model is research‑grade (not yet clinical), was trained on an age‑skewed UK cohort, and may carry biases. Developers are expanding inputs to include imaging, genetics and lab data and emphasize rigorous testing, regulation and ethical safeguards before deployment—positioning this as a potential paradigm shift in population and personalized health planning rather than an immediate clinical tool.