Summary
Medical care in the emergency department is one of the most important component in the health care system. The clinical burden of emergency service keeps increasing all over the world. Overcrowding in the emergency department not only increases the workload of health care professionals but also increases the chance of unexpected events including the in-hospital mortality and readmission. In the study, we used the large scale administrative data from National Health Insurance. The dataset was composed of the information of hospital levels, diagnosis, management and claims in a longitudinal fashion from 2003-2010. We applied the deep learning model to predict the in-hospital mortality and readmission within 30 days. All patient’s events were lined up to have the full timeline, and patient vectors were generated. For each event, a series of outcomes are calculated. The area under curves (AUC) were 0.96 for in-hospital mortality prediction, 0.69 for 30-day readmission prediction.
 
Applications
The risk prediction can provide the critical information not only for the health care providers but also for the patients and family in emergency department. The information can guide the risk stratification and intensive care for the patients needed and prevent the unexpected event for the emergent and critical patients in the daily clinical practice.
 
Advantages
The more accurate prediction by deep learning from the large scale data provides a breakthrough in the risk management in emergency department. The promising result can be introduced into real clinical practice.
 
Keywords
Emergency department, Deep learning, In-hospital mortality, Readmission, National health insurance