GraD-IBD: Graph Model Detects Inflammatory Bowel Disease from Diagnosis Trajectories

Researchers propose GraD-IBD, a graph-based model that reformulates longitudinal ICD code sequences into visit-bucketized, temporally directed graphs for early detection of inflammatory bowel disease (IBD). The model introduces a context-aware, time-decay message passing mechanism to capture temporal dependencies while reducing computational complexity. Experiments on a real-world clinical dataset show consistent and robust improvements over state-of-the-art methods, with significant reductions in computational demands. The approach addresses challenges posed by the irregular and hierarchical nature of ICD code sequences, which complicate traditional N-D lattice-based sequential modeling.