Development and Validation of Machine Learning-Based Models for Prediction of Intensive Care Unit Admission and In-Hospital Mortality in Patients with Acute Exacerbations of Chronic Obstructive Pulmonary Disease

• Published in: Chronic obstructive pulmonary diseases Vol. 11; no. 5; pp. 460 - 471
• Main Authors: Jia, Qinyao, Chen, Yao, Zen, Qiang, Chen, Shaoping, Liu, Shengming, Wang, Tao, Yuan, XinQi
• Format: Journal Article
• Language: English
• Published: United States 27.09.2024
• Subjects: in-hospital mortality; risk assessment; machine learning; ICU admission; acute exacerbations of chronic obstructive pulmonary disease
• ISSN: 2372-952X; 2372-952X
• DOI: 10.15326/jcopdf.2023.0446

This present work focused on predicting prognostic outcomes of inpatients developing acute exacerbation of chronic obstructive pulmonary disease (AECOPD), and enhancing patient monitoring and treatment by using objective clinical indicators. The present retrospective study enrolled 322 AECOPD patients. Registry data downloaded based on the chronic obstructive pulmonary disease (COPD) Pay-for-Performance Program database from January 2012 to December 2018 were used to check whether the enrolled patients were eligible. Our primary and secondary outcomes were intensive care unit (ICU) admission and in-hospital mortality, respectively. The best feature subset was chosen by recursive feature elimination. Moreover, 7 machine learning (ML) models were trained for forecasting ICU admission among AECOPD patients, and the model with the most excellent performance was used. According to our findings, a random forest (RF) model showed superb discrimination performance, and the values of area under the receiver operating characteristic curve were 0.973 and 0.828 in training and test cohorts, separately. Additionally, according to decision curve analysis, the net benefit of the RF model was higher when differentiating patients with a high risk of ICU admission at a <0.55 threshold probability. Moreover, the ML-based prediction model was also constructed to predict in-hospital mortality, and it showed excellent calibration and discrimination capacities. The ML model was highly accurate in assessing the ICU admission and in-hospital mortality risk for AECOPD cases. Maintenance of model interpretability helped effectively provide accurate and lucid risk prediction of different individuals.