Prediction Model of Length of Stay in Intensive Care Unit for
Patients Undergoing Craniotomy Based on Interpretable Machine Learning
WANG Shaobo1, WANG Qiqi2, JIAO Zengtao2, LIU Youjun1, YU Rongguo3
1. Faculty of Environment and Life, Beijing University of Technology, Beijing 100124, China; 2. Yidu Cloud (Beijing) Technology
Co., Ltd., Beijing 100191, China; 3. Surgical Intensive Care Unit, Fujian Provincial Hospital, Fuzhou Fujian 350001, China
Abstract：Objective To predict whether a patient with craniotomy will stay in intensive care unit (ICU) for a long period (≥8 d) by
establishing a prediction model of length of stay based on three explainable machine learning algorithms, and to mine the risk factors
affecting the length of stay in ICU. Methods A total of 677 patients who had undergone craniotomy in Fujian Provincial Hospital
from 2005 to 2018 were selected. The prediction model was established for 67 features of patients based on logistic regression of
machine learning, random forest, and gradient boosting decision tree (GBDT) algorithms. According to the model evaluation method,
the optimal model was selected and analyzed. Results The effect of GBDT model was the best, the accuracy was 85%, and the area
under the curve of receiver operating characteristic was 0.90. Conclusion The above results proved the effectiveness of the prediction
model established in this study, which can provide clinicians with auxiliary decision-making suggestions, make corresponding
intervention and decision in advance, and reduce the burden of patients and medical institutions.
王绍博1，王琪琪2，焦增涛2，刘有军1，于荣国3. 基于可解释性机器学习算法的开颅手术患者重症监护室住院时间预测模型[J]. 中国医疗设备, 2022, 37(5): 23-28.
WANG Shaobo1, WANG Qiqi2, JIAO Zengtao2, LIU Youjun1, YU Rongguo3. Prediction Model of Length of Stay in Intensive Care Unit for
Patients Undergoing Craniotomy Based on Interpretable Machine Learning. China Medical Devices, 2022, 37(5): 23-28.
