Clinical Event Knowledge Graphs: Enriching Healthcare Event Data with Entities and Clinical Concepts - Research Paper

Clinical processes include admission, discharge, medication administration, diagnostic testing, and others. Process mining promises to provide insights for improving such processes. An issue in analyzing clinical processes is that recorded events concerning the treatment of patients are often not only related to the patients themselves, but also the event and activities terminology need to be interpretable globally in different health organizations. Specifically, in the case of multimorbidity, it is expected that clinical events recorded for a patient relate to multiple disorders and are linked to many different clinical concepts from various medical specialties. This hampers the application of process mining as extracting a single-entity event log gives an incomplete view of the patient trajectory, and relating events and activities to clinical terminology understood by medical professionals is complex. We propose to address these issues by building a clinical event knowledge graph that combines multi-entity event data from clinical systems with clearly defined clinical terms from coding systems such as ICD10-cm and a systematically organized collection of medical terms such as SNOMED CT. Our contribution is a framework that facilitates building such a clinical event knowledge graph. We evaluate the proposed framework by showing the feasibility of applying it to the MIMIC-IV dataset. We validated a set of process-related questions on multi-morbid patients with clinical experts. By leveraging the graph, these questions can be answered at the abstraction level of clinical terms. This may facilitate the involvement of medical professionals in the analysis, leading to the enhanced management of healthcare processes.