Overview

• Built by teams at EMBL, the German Cancer Research Center and the University of Copenhagen, Delphi-2M analyzes medical histories and lifestyle factors to estimate probabilities and timing for more than 1,000 conditions.
• The model was trained on 400,000 UK Biobank records and externally validated on about 1.9 million patients from Denmark’s National Patient Registry, producing meaningful forecasts across two healthcare systems.
• For most conditions, its accuracy matched or exceeded leading single‑disease risk models and outperformed a biomarker‑based multi‑disease algorithm, with stronger results over shorter horizons and for diseases with consistent progression.
• As a generative system, Delphi-2M can sample synthetic future health trajectories, enabling calibrated population‑level risk estimates and long‑range planning up to two decades.
• Researchers highlight data‑set biases and privacy risks, estimate potential clinical use in roughly 5–10 years, and call for governance, randomized trials and safeguards to prevent misuse.