A new artificial intelligence system, Delphi-2M, has been developed to predict over 1,000 diseases by analyzing the medical histories of patients. Utilizing data from the UK Biobank of approximately 403,000 participants and validated on 1.93 million Danish patients, the model displayed a high predictive accuracy (average AUC ~0.76) for various diseases. Performance varied across conditions, with mental health clustering over time, while pregnancy-related conditions dissipated faster. Beyond diagnosis, Delphi-2M can simulate health trajectories up to 20 years. The study highlights both the effectiveness and limitations of the model, such as possible bias and reduced accuracy for certain diseases.

Source: Nature