Clinical Process Analysis for Grasping Inpatient Condition with Machine Learning and Directed Graph

Abstract

　Standardization and structuring of data are necessary for analyzing medical data between medical institutions. An interface and repository are developed for data linkage of the electronic medical record, have simplified the work of integrating data. In this paper, we propose an analysis method that can determine the priority of clinical intervention by prioritizing and visualizing time series of patient condition during the hospitalization period. Time series analysis of long-term hospitalizations and discharge destination was performed using DPC and clinical pathway data stored in the institutions using machine learning and the directed graphs. Long-term hospitalization was AUC=0.913, rolling over, changing clothes, respiratory status, and circulatory status were extracted. The discharge destination was AUC=0.773, oral care, and meal were extracted. The directed graph showed the relationship between the time-series variables, and was confirmed to be useful for the analysis of clinical process.