Abstract
We construct networks from a prescription database (PD) in a long-term care facility and detect communities in these networks (PD networks) . The nodes of the networks consist of the field’s data of the PD such as a patient ID, a drug name and a doctor’s ID. A community is a sub networks in which nodes are densely connected internally. Because a community in the PDN involves patient ID nodes, the community fixes the group of patients. We can detect communities in a PD network using two fields: “patient” and “drug” as nodes, but we cannot detect them when using all the fields. By evaluating the drugs in the communities, we can clinically characterize the communities. Next, we define a Network Similarity Index (NSI) to measure similarity between communities. Using this index, we find that there exist many communities in the network which have not changed significantly for a long time. Moreover, we find that the patient groups obtained from community detection are different from those from a cluster analysis. These findings will propose another method to analyze the clinical data.
