Prediction of postoperative pulmonary infection after video-assisted thoracoscopic anatomical pulmonary resection using an interpretable machine learning model.

BACKGROUND: Postoperative pulmonary infection (PPI) remains a significant complication after video-assisted thoracoscopic anatomical pulmonary resection (VAT-APR). This study aimed to develop and validate an interpretable machine learning (ML) model for perioperative prediction of PPI using Light Gradient Boosting Machine (LightGBM) integrated with SHapley Additive exPlanations (SHAP) for patients undergoing VAT-APR. METHODS: This retrospective predictive modeling study included 512 patients who underwent VAT-APR between January 2020 and December 2024 at the Department of Thoracic Surgery, The First People's Hospital of Yunnan Province, Kunming, China. Preoperative and intraoperative features were extracted, and the dataset was randomly partitioned into training (70%) and validation (30%) sets. Five ML algorithms were evaluated, with the LightGBM model selected as optimal. RESULTS: The LightGBM model demonstrated strong discriminative performance, achieving an area under the curve (AUC) of 0.832 [95% confidence interval (CI): 0.795-0.869], with an accuracy of 0.781 and an F1-score of 0.710, outperforming five comparator algorithms. Feature importance analysis identified operative time, percentage of predicted forced expiratory volume in 1 second (FEV1%), age, and monocyte count as the most influential predictors. The model showed good calibration (Hosmer-Lemeshow test, P=0.32) and yielded a positive net clinical benefit across a range of threshold probabilities in decision curve analysis. CONCLUSIONS: The interpretable LightGBM-SHAP model provides a robust and transparent approach for perioperative risk assessment of PPI, with potential to inform individualized perioperative management and targeted preventive strategies.