Linguistic analysis of large-scale medical incident reports for patient safety

The analysis of medical incident reports is indispensable for patient safety. The cycles between analysis of incident reports and proposals to medical staffs are a key point for improving the patient safety in the hospital. Most incident reports are composed from freely written descriptions, but an analysis of such free descriptions is not sufficient in the medical field. In this study, we aim to accumulate and reinterpret findings using structured incident information, to clarify improvements that should be made to solve the root cause of the accident, and to ensure safe medical treatment through such improvements. We employ natural language processing (NLP) and network analysis to identify effective categories of medical incident reports. Network analysis can find various relationships that are not only direct but also indirect. In addition, we compare bottom-up results obtained by NLP with existing categories based on experts' judgment. By the bottom-up analysis, the class of patient managements regarding patients' fallings and medicines in top-down analysis is created clearly. Finally, we present new perspectives on ways of improving patient safety.