A world model simulates the latent dynamics of human health

Human health is a single underlying state that no measurement observes directly. Diagnoses, blood tests, molecular profiles and images capture different facets at different times. Inferring health from such evidence requires a representation that integrates every modality, persists between observations and is revised as new evidence arrives. Here we introduce HealthFlux, a pan-modal world model that learns the latent dynamics of health from 5,647 features across eleven data domains in 502,166 UK Biobank participants. Its hybrid state-space architecture combines ODE-based evolution between observations with continuous-time recurrent updates when new measurements arrive. In held-out participants, recursive simulations without further observations predicted 1,010 diseases and mortality over twenty years, outperforming previously published state-of-the-art models. HealthFlux also achieved the highest median disease-level AUROC in three independent US cohorts and exceeded specialized clinical risk scores for disease and mortality. HealthFlux predicts diseases excluded entirely from training, with a mean AUROC of 0.783, evidence that it has learned health itself rather than the diseases it was trained on. Medication-conditioned virtual clinical trials simulated drug-associated physiological changes and downstream disease risk. Predicted effects were evaluated against published results from 36 clinical trials. Each modality contributes information the others lack, and integrating them identifies individuals at risk whom single-modality models miss. HealthFlux treats health itself as the object of prediction through one continuously updated state, informed by any measurement, that supports disease-risk estimates years before diagnosis.