Researchers have developed HealthFormer, a generative multimodal model designed to simulate human physiology and predict health trajectories. Trained on data from over 15,000 individuals across various health domains, the model tokenizes physiological measurements to forecast individual health changes. HealthFormer has demonstrated strong transferability to independent cohorts, improving disease and mortality endpoint predictions and outperforming existing clinical risk scores. Additionally, it can simulate interventions in silico, accurately predicting biomarker changes and intervention effects from published trials, positioning it as a foundational health world model. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Potential to accelerate personalized medicine and clinical trial simulations by providing a 'health world model'.
RANK_REASON Academic paper introducing a new generative model for simulating human physiology and predicting health outcomes.