Evaluation of Patient Grouping by Community Detection of the Networks Derived from Clinical Data

Abstract

 Our study aims to develop patient grouping by community detection of networks derived from clinical data. We constructed the networks with nodes consisting of patient IDs and the other field’s data of the database in the hospital information system. We found that the networks were divides into several communities and as a result, a set of patients was divided into several groups according to the communities (PGN: Patient Grouping by Network community detection). Using the same data, we also tried to group the set of patients by cluster analysis and MDC (Major Diagnostic Categories). And we get the two other patient groupings (PGC: Patient Grouping by Cluster analysis, PGM: Patient Grouping from MDC). We constructed a new index to measure the similarity degree between two patient groupings (SIG: Similarity Index between two Groupings), and we calculated the SIGs for all combinations of groupings. As a result, the similarity degree between PGN and PGC was comparatively low. The similarity degree between PGN and PGM was higher than that between PGC and PGM. This result indicates that the patient grouping by network community detection is clinically more meaningful than that by cluster analysis. We can produce another grouping by a choice of different field as the nodes of the networks. Therefore, our grouping method has wide applications. For example, using nursing data can gives us a new patient grouping for nursing care.