Correlation Clustering of Stable Angina Clinical Care Patterns for 506 Thousand Patients

Abstract

Objectives. Our goal was to apply statistical and network science techniques to depict how the clinical pathways of patients can be used to characterize the practices of care providers. Methods. We included the data of 506,087 patients who underwent procedures related to ischemic heart disease. Patients were assigned to one of the 136 primary health-care centers using a voting scheme based on their residence. The clinical pathways were classified, and the spectrum of the pathway types was computed for each center, then a network was built with the centers as nodes and spectrum correlations as edge weights. Then Louvain clustering was used to group centers with similar pathway spectra. Results. We identified 3 clusters with rather distinct characteristics that occupy quite compact spatial areas, though no geographical information was used in clustering. Network analysis and hierarchical clustering show the dominance of medical university clinics in each cluster. Conclusion. Though clinical guidelines provide a uniform regulation for medical decisions, doctors have great freedom in daily clinical practice. This freedom leads to regional preferences of certain clinical pathways, the intercenter professional links, and geographical locality and coupled with quantifiable consequences in terms of care costs and periprocedural risk of patients.