Methods for refining patient classification for comprehensive payment based on diagnostic group classification are eagerly desired. In this paper, we proposed a patient classification method focused on the distribution of hospital stay. This method measures differences among the hospital stay distribution of patient groups divided by treatment and side effects by using Kullback-Leibler Divergence (KLD), which is an indicator of distribution similarity, and cluster the groups by cluster analysis using these values.
We tried classification by applying this method to the anonymization data of 832 inpatient patients who do not undergo surgery at a hospital in Beijing city. KLD was obtained for a group of patients divided by rehabilitation in treatment, combination of speech impairment, peripheral neuropathy, lung infection, and type 2 diabetes in side effects, and hierarchical clustering was performed. As a result, groups with common distribution characteristics were classified into the same cluster, suggesting the usefulness of this method.
View full abstract