Hanxu Xi,1,2,* Yudi Zhang,3,* Rui Zuo,2 Wei Li,2 Chen Zhang,2 Yongchang Sun,1 Hong Ji,2 Zhiqiang He,3 Chun Chang1 

1Department of Respiratory and Critical Care Medicine, Peking University Third Hospital, Beijing, 100191, People’s Republic of China; 2Information Management and Big Data Centre, Peking University Third Hospital, Beijing, 100191, People’s Republic of China; 3Key Laboratory of Universal Wireless Communications, Ministry of Education, Beijing University of Posts and Telecommunications, Beijing, 100876, People’s Republic of China

*These authors contributed equally to this work

Correspondence: Chun Chang, Department of Respiratory and Critical Care Medicine, Peking University Third Hospital, No. 49 North Garden Road, Haidian District, Beijing, 100191, People’s Republic of China, Email doudou_1977@163.com Hong Ji, Information Management and Big Data Center, Peking University Third Hospital, No. 49 North Garden Road, Haidian District, Beijing, 100191, People’s Republic of China, Email puh3_imc@bjmu.edu.cn

Background: Asthma is the world’s second most prevalent chronic respiratory disease. Current clinical decisions regarding hospitalization for adult asthma patients in emergency departments (EDs) primarily rely on presenting clinical status, acute exacerbation severity, therapeutic response and high-risk factors. Assessing the need for hospitalization of patients with complex comorbidities remains a significant challenge.
Research Question: This study aims to develop models that integrate various environmental and clinical factors to predict the hospitalization of adult asthma patients in EDs and to interpret these models.
Study Design and Methods: A retrospective analysis was conducted utilizing data from asthma patients at a single ED from 2016 to 2023; the data included demographics, vital signs, illness severity, laboratory test results, and comorbidities, along with environmental variables. Predictive models were constructed using the extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), support vector machine (SVM), logistic regression (LR), and random forest (RF). Area under the receiver operating characteristic curve (AUC), accuracy, and F1 score were the primary metrics used to assess model performance.
Results: The analysis included 1140 ED visits. The median age was 51.0 years (interquartile range: 31.0 to 67.0 years), and 56.5% of the patients (644) were female. Overall, 21.8% of patients (249) required hospitalization after their ED visits. The AUC results for predicting hospitalization without external environmental factors were 0.8075 for XGBoost, 0.8233 for LightGBM, 0.7935 for SVM, 0.8033 for LR, and 0.8272 for RF. After integrating ambient air pollutant and meteorological features, the RF model consistently outperformed the other models, achieving an AUC of 0.8555. The most critical parameters for predicting hospitalization were found to be illness severity, oxygen saturation, age, and heart rate.
Interpretation: Machine learning (ML) models based on clinical, meteorological, and air pollution data can rapidly and accurately predict hospitalization of adult asthma patients in EDs.

Keywords: asthma exacerbation, machine learning, emergency department