Thoracic muscle loss increases the use of mechanical ventilation in elderly patients with pulmonary embolism: constructing and validating a machine learning model on a two-center cohort.
Authors
Affiliations (8)
Affiliations (8)
- Department of Radiology, The Second Affiliated Hospital, Jiangxi Medical College, Nanchang University, No. 1 Minde Road, Nanchang, 330006, China.
- Jiangxi Provincial Key Laboratory of Intelligent Medical Imaging, Nanchang, 330006, China.
- Department of Radiology, The Second Affiliated Hospital, Jiangxi Medical College, Nanchang University, No. 1 Minde Road, Nanchang, 330006, China. [email protected].
- Jiangxi Provincial Key Laboratory of Intelligent Medical Imaging, Nanchang, 330006, China. [email protected].
- Department of Radiology, The First Affiliated Hospital of Gannan Medical University, Ganzhou Economic and Technological Development Zone, No. 128, Jinling West Road, Ganzhou, 341001, China.
- Clinical and Technical Support, Philips Healthcare, No.718 Lingshi Road, Jing'an District, Shanghai, China.
- Department of Radiology, The Second Affiliated Hospital, Jiangxi Medical College, Nanchang University, No. 1 Minde Road, Nanchang, 330006, China. [email protected].
- Jiangxi Provincial Key Laboratory of Intelligent Medical Imaging, Nanchang, 330006, China. [email protected].
Abstract
This study aimed to develop and validate a machine learning (ML) model to predict the need for mechanical ventilation (MV) in elderly patients diagnosed with acute pulmonary embolism (APE). The study included a cohort of 321 patients from two centers. Center A contributed 261 patients for the development of four ML models: Random Forest, XGBoost, Logistic Regression (LR), and Support Vector Classifier (SVC). The remaining 60 patients from Center B were used for external testing. Feature selection incorporated CT histogram features related to muscular density and area of the pectoralis muscles on CT images at the level of the fourth thoracic vertebra, common geriatric comorbidities, and routine laboratory tests for APE. The area under the curve (AUC) was used to evaluate the predictive performance of the models; calibration curves were employed to assess calibration performance, and the sPESI score served as a baseline comparator. Shapley Additive exPlanations (SHAP) plots were utilized to visualize the importance of each feature. The final set of features included low oxygen saturation, smoke status, CT_PMA_10th, CT_PMA_90th, CT_PMI_75th, CT_PMA_Fat_ratio, chest pain, syncope, diabetes, gender, chronic heart failure, NT-proBNP/BNP positive, and D-dimer. In the internal validation set, the four models performed well and exhibited similar performance, with AUC values exceeding 0.80. Among the models evaluated, the LR model demonstrated the best performance on the external test set, with an AUC of 0.837, an accuracy of 0.817, a recall of 0.750, a specificity of 0.833, a precision of 0.529, and an F1-score of 0.621. The SHAP plot revealed that low oxygen saturation, CT_PMA_10th, and CT_PMI_75th were highly important features. Loss of pectoral muscle may be associated with the need for MV in elderly patients with APE. The prediction model developed in this study, which includes this factor, could aid in identifying high-risk individuals and may inform future efforts to improve early risk stratification and personalized management in this population.