A Bayesian framework for longitudinal EHR and genetic discovery

Researchers have developed a new model called ALADYNOULLI to better understand how health risks change over a person's life. By combining large-scale electronic health records with genetic data, this tool identifies latent disease signatures that show how different medical conditions co-evolve and cluster together over time.

Traditional medical analysis often treats diseases as separate, isolated events. This new framework takes a different path by modeling these conditions as part of a connected, dynamic system. Because it integrates genetic information directly into its architecture, the model provides personalized risk profiles that update as new clinical information enters the patient file.

Testing across three large independent biobanks including the UK Biobank, Mass General Brigham, and All of Us, the model proved consistent at identifying shared disease signatures. These patterns link cardiovascular, metabolic, and pulmonary conditions in ways that align with established clinical phenotypes. The model also offers a principled way to correct for selection bias, which is a common challenge in large-scale health data research.

Beyond just predicting risk, this approach helps uncover new genetic architecture. By linking rare and common genetic variants to these signatures rather than just individual diseases, researchers found new insights into the biology of conditions like coronary artery disease and diabetes. This work provides a new way to group patients based on their specific disease trajectories rather than just their current diagnostic label, which could lead to more targeted preventive care.