RP27 - Hospital Processes Mining from real-world FHIR Graphs

Efficient patient care hinges on navigating hospital workflows, including clinical pathways, administrative procedures, and resource interactions. Analysing these processes helps identify bottlenecks, optimise resource use, and improve care delivery [1]. The FHIR standard uses a resource-based model to transform healthcare data exchange by structuring data as interconnected “resources” (e.g., patient demographics, diagnoses, medications), creating graphs that reflect real-world hospital processes. Building on RP14 and RP17, this project aims to develop techniques to translate FHIR graph data into structured representations of hospital procedures, benchmark these against best practices, and apply findings to optimise healthcare operations. Methodologically, FHIR graphs from the local digital health platform (SHIP) will be analysed using Graph Neural Networks (GNNs) to uncover complex relationships and translate them into formal process descriptions (e.g., BPMN process models, FHIR PlanDefinition/ ActivityDefinition, workflow diagrams, decision trees). These will be compared with published healthcare guidelines and hospital SOPs to identify areas for improvement. Expected outcomes include a repository of process models, detailed comparative analysis, and actionable recommendations for enhancing efficiency, reducing redundancies, and optimising resource allocation. Bringing these insights to the PoC via integration via RP36 into the Melanoma Dashboard application can provide clinicians with a comprehensive view of a patient’s healthcare journey, ensuring personalised care by highlighting potential delays and optimising resource allocation. This project advances the understanding of hospital workflows through FHIR data, contributing to scientific knowledge and practical improvements in healthcare operations.

[1] E. Rojas, J. Munoz-Gama, M. Sepúlveda, and D. Capurro, “Process mining in healthcare: A literature review,” J. Biomed. Inform., vol. 61, pp. 224–236, Jun. 2016, doi: 10.1016/j.jbi.2016.04.007.