Abstract
BACKGROUND: Perioperative hypoalbuminemia is associated with postoperative infection, anastomotic fistula, and a poor prognosis. Compared with the preoperative period, hypoalbuminemia is more prevalent following pulmonary surgery, particularly in the early postoperative phase, which is associated with various postoperative complications. Traditional risk assessment relies on clinical experience and basic laboratory indicators. Currently, no research has been conducted on the application of machine learning (ML) in the prediction of early postoperative hypoalbuminemia (EPH). This study aimed to develop an ML-based predictive model for EPH following pulmonary surgery, offering a novel tool for risk assessment and clinical decision-making in the perioperative management of thoracic surgery. METHODS: The data of patients diagnosed with primary lung cancer who underwent elective pulmonary surgery between January 2022 and December 2024 were retrospectively collected. Based on 1:1 case-control matching, the sample comprised 1,048 cases and 1,048 controls. The outcome variable was binary (the presence or absence of EPH after pulmonary surgery). A logistic regression (LR) model was built with 37 variables; the data were split 8:2 and validated by five-fold stratified cross-validation. Model performance was assessed based on the area under the curve (AUC), accuracy, precision, recall, F1, and Brier score, with SHapley Additive exPlanations (SHAP) used for interpretation. RESULTS: The model performance metrics were as follows: AUC of the receiver operating characteristic (ROC) curve: 0.8543, precision: 0.7947, recall: 0.7309, F1-score: 0.7606, accuracy: 0.771, and Brier score: 0.1551. CONCLUSIONS: The LR-based ML algorithm demonstrated excellent performance and effectively identified patients at high risk of EPH after pulmonary surgery [serum albumin (ALB) <35 g/L within 5 days of pulmonary surgery].