Electronic health records (EHRs) contain longitudinal data that can be modeled as graphs to reveal disease trajectories. A common approach is to identify branching motifs—patterns where a disease course splits into different outcomes. But as the authors of a new arXiv preprint note, a trajectory that is well-supported by the data is not necessarily clinically novel or meaningful.

The paper presents a two-level prioritization framework designed to rank such branching motifs. The framework was applied to 99,527 motifs extracted from longitudinal EHRs. While the abstract does not detail the exact mechanics of the two levels, the core aim is to separate patterns that merely have structural support from those that carry genuine semantic novelty.

The authors argue that frequency or graph-based support alone is insufficient to identify trajectories worth clinical attention. By adding a prioritization step, the framework offers a way to focus on disease trajectories that are more likely to be useful for clinical insight and hypothesis generation.