Delphi-2M LLM uses medical records, lifestyle to provide risks for 1k+ diseases (www.nature.com)

Researchers published Delphi-2M, a modified large language model (a GPT-style generative pre-trained transformer) that predicts an individual’s risk of developing 1,258 diseases—including cancers, dermatological and immune conditions—up to 20 years ahead. Trained on longitudinal records from 400,000 UK Biobank participants and conditioned on past medical history plus demographic and lifestyle inputs (age, sex, BMI, smoking and alcohol use), the model produces probabilistic future health trajectories. For most conditions it matched or outperformed existing single-disease risk models and beat a biomarker-based machine-learning algorithm for multi-disease prediction. The advance is significant because it demonstrates that transformer architectures can be repurposed for large-scale, multi-label longitudinal disease forecasting, potentially enabling earlier identification of high-risk patients and prioritization of preventive interventions within clinical workflows. Important caveats remain: Delphi-2M was trained on a single UK dataset, so its generalizability, fairness across populations, and clinical utility need external validation and regulatory scrutiny. Still, the work points to a new technical direction—using LLM sequence modeling to integrate routine health and lifestyle data into long-horizon, multi-disease risk estimates—which could reshape population health management if validated and deployed responsibly.