Related Experiment Videos
Machine Learning-Based Prediction of Prolonged Mechanical Ventilation After Stanford Type A Aortic Dissection
Zhanhua Wei1, Jiali Zhou2, Xing Li1
1Department of Nursing, Renmin Hospital of Wuhan University, Wuhan, People's Republic of China.
Journal of Multidisciplinary Healthcare
|July 6, 2026
Summary
This study compared machine learning models for predicting prolonged mechanical ventilation (PMV) after Stanford type A aortic dissection (TAAD) surgery. Support vector machine (SVM) showed promising results for risk stratification.
Area of Science:
- Cardiovascular Surgery
- Medical Informatics
- Critical Care Medicine
Background:
- Stanford type A aortic dissection (TAAD) is a critical condition with high mortality.
- Prolonged mechanical ventilation (PMV) is a frequent postoperative complication impacting recovery and resource utilization.
- Existing prediction models for PMV after TAAD surgery primarily use logistic regression, with limited comparison to machine learning methods.
Purpose of the Study:
- To develop and internally validate prediction models for PMV risk following TAAD surgery.
- To compare the performance of logistic regression, support vector machine (SVM), and extreme gradient boosting (XGBoost) in predicting PMV.
- To identify key predictors for PMV in TAAD patients.
Main Methods:
- Retrospective analysis of 511 TAAD patients from two institutions.
- Patients were divided into training (75%) and testing (25%) sets.
- PMV defined as mechanical ventilation >48 hours; risk factors identified using LASSO regression; models built with logistic regression, SVM, and XGBoost; performance evaluated using AUC, calibration curves, and decision curve analysis.
Main Results:
- PMV occurred in 31.1% of patients.
- Six predictors for PMV were identified: preoperative lactate, cardiopulmonary bypass time, peak postoperative creatinine, postoperative plasma transfusion, end-of-surgery lactate, and pulmonary complications.
- In the test set, SVM achieved the highest AUC (0.820), followed by logistic regression (0.805) and XGBoost (0.784), with no statistically significant differences.
Conclusions:
- The developed models, particularly SVM, demonstrate acceptable internal performance for predicting PMV after TAAD surgery.
- Identified predictors are clinically relevant for perioperative risk stratification and postoperative respiratory management.
- Findings are preliminary; prospective multicenter validation is required before clinical implementation.
