Mass General Brigham Develops New Model to Predict Health Outcomes Using EHR Data
A study conducted by Mass General Brigham and collaborators has introduced a Bayesian model to transform electronic health record (EHR) data into meaningful disease trajectories. The research, led by Sarah Urbut, MD, PhD, and Pradeep Natarajan, MD, MMSc, addresses the limitations of current medical practices that treat diagnoses as static labels. The model, named ALADYNOULLI, integrates longitudinal diagnosis patterns with genetic risk information to uncover latent disease signatures. This approach allows for the identification of shared mechanisms across different diseases, which are often missed when analyzed separately. The model has been tested across three biobanks, involving over 683,000 individuals, and has successfully compressed 348 diseases into 21 latent signatures. These findings suggest that the model can generalize across different health systems, potentially improving dynamic risk assessments. 
