Researchers have developed a framework for creating and refining clinical knowledge graphs. These graphs are used to make informed decisions about patient care, but current methods can introduce errors and inconsistencies. The new framework uses multiple agents to propose relations between patients' data, normalizes the data using ontologies, scores evidence, and refines the graph using a technique called JEPA. The system learns to predict missing relationships based on observed patterns in the graph. In tests, the framework improved leave-one-out edge recovery by 31% when augmented with discharge-note representations. --- Why it matters: This matters because accurate clinical knowledge graphs can improve patient outcomes and reduce healthcare costs. By predicting patient-state representation more effectively, clinicians can make better-informed decisions about treatment and care. Source: https://arxiv.org/abs/2608.22583