Clustering of patient trajectories with an auto-stopped bisecting K-Medoids algorithm

Abstract

Abstract Nowadays, more and more hospitals seek to employ well-developed process management and simulation tools in the healthcare organizations to develop the overall patient pathways. Since it is extremely important to define suitable patient groups for constructing process or simulation models, we proposed a sequence mining method, an auto-stopped Bisecting K-Medoids clustering algorithm, to classify patients into groups with homogeneous trajectories within two stages. At the first stage, patients are classified according to the complexity of the care process. Afterwards, groups obtained at the first stage are further classified with the similarity of the trajectories. The proposed approach was executed with a real data set from a medium-size hospital. According to the experimental results, this method can be used to classify patients into manageable groups, where the most frequent trajectories might be extracted to validate a process modelling technique. In addition, data extracted from those groups could be used to feed our simulation models.