Development of an emergency department length-of-stay prediction model based on machine learning

Weiming Wu; Min Li; Huilin Jiang; Min Sun; Yongcheng Zhu; Gongxu Zhu; Yanling Li; Yunmei Li; Junrong Mo; Xiaohui Chen; Haifeng Mao

doi:10.5847/wjem.j.1920-8642.2025.048

Development of an emergency department length-of-stay prediction model based on machine learning

World J Emerg Med. 2025 May 1;16(3):220-224. doi: 10.5847/wjem.j.1920-8642.2025.048.

Authors

Weiming Wu¹, Min Li¹, Huilin Jiang¹, Min Sun², Yongcheng Zhu¹, Gongxu Zhu¹, Yanling Li¹, Yunmei Li¹, Junrong Mo¹, Xiaohui Chen¹, Haifeng Mao¹

Affiliations

¹ 1Emergency Department, the Second Affiliated Hospital of Guangzhou Medical University, Guangzhou 510260, China.
² 2Goodwill Hessian Health Technology Co., Ltd., Beijing 100007, China.

Abstract

Background: The problem of prolonged emergency department length of stay (EDLOS) is becoming increasingly crucial. This study aims to develop a machine learning (ML) model to predict EDLOS, with EDLOS as the outcome variable and demographic characteristics, triage level, and medical resource utilization as predictive factors.

Methods: A retrospective analysis was performed on the patients who visited the emergency department of the Second Affiliated Hospital of Guangzhou Medical University from March 2019 to September 2021, and a total of 321,012 cases were identified. According to the inclusion and exclusion criteria, 187,028 cases were finally included in the analysis. ML analysis was performed using R-squared (R²), and the predictive factors and the EDLOS were used as independent variables and dependent variables, respectively, to establish models. The performance evaluation of the ML models was conducted through the utilization of the mean absolute error (MAE), root mean square error (RMSE), and R², enabling an objective comparative analysis.

Results: In the comparative analysis of the six ML models, light gradient boosting machine (LightGBM) model demonstrated the lowest MAE (443.519) and RMSE (826.783), and the highest R² value (0.48), indicating better model fit and predictive performance. Among the top 10 predictive factors associated with EDLOS according to the LightGBM model, the emergency waiting time, age, and emergency arrival time had the most significant impact on the EDLOS.

Conclusion: The LightGBM model suggests that the emergency waiting time, age, and emergency arrival time may be used to predict the EDLOS.

Keywords: Emergency department; Length of stay; Light gradient boosting machine; Machine learning.