Investigating an Ontology-Informed Approach to Event Log Generation in Healthcare

Process mining methods have the potential to reveal useful patterns in hospital data and provide insights into patterns of treatment, but healthcare has proven an exceptionally challenging domain due to the significant heterogeneity of possible events and the diversity of individual patients. When studying a patient’s pathway, their history might contain hundreds of events unrelated to the disease of interest; determining which to include or exclude in an event log is difficult, and highly dependent on the situation. Typically, researchers only include the most common or correlated events (risking missing rare but significant exceptions), or hand-curate a list of events (slow and difficult, especially where there might be thousands of possible events). We discuss an alternative approach that uses logical reasoning over ontologies to deduce plausibly related events, which might have implications in terms of time savings, the facilitation of reproducible research, and the enabling of comparisons across datasets and healthcare systems. We investigate the practical feasibility of such an approach by demonstrating it on real data, finding that the discovered processes approximate those produced through established approaches, with some caveats. The approach is a promising way to support conversations with domain experts and expedite human processes, and highlights the potential of structured domain knowledge to inform and enhance the event log generation process.