Clinical Report: Predicting Prolonged Air Leak Post-Uniportal Video-Assisted Thoracic Surgery Segmentectomy Using Machine Learning Techniques
Overview
This study developed and validated machine learning models to predict prolonged air leak (PAL) after uniportal video-assisted thoracic surgery (uVATS) segmentectomy. The XGBoost model demonstrated the highest predictive performance, identifying key risk factors associated with PAL.
Background
Prolonged air leak (PAL) is a common complication following uVATS segmentectomy, leading to increased hospitalization and healthcare costs. Identifying patients at risk for PAL is important. Traditional statistical models have limitations in predicting PAL due to complex interactions among clinical variables.
Data Highlights
Model
AUC
Patients with PAL
XGBoost
0.874
76 (12.46%)
Key Findings
The XGBoost model outperformed other machine learning algorithms in predicting PAL.
Key predictors of PAL included low body mass index (BMI), prolonged operative time, reduced DLCO%, diabetes, and complex segmentectomy.
SHAP analysis provided insights into the contributions of various clinical factors to the prediction model.
Internal validation of the model was conducted with a cohort of 610 patients.
Clinical Implications
The identified predictors can assist in risk stratification for patients undergoing uVATS segmentectomy.
Conclusion
Machine learning models, particularly XGBoost, show potential in predicting PAL after uVATS segmentectomy.
