Development of a machine learning algorithm model to predict intraoperative hypotension in elderly patients undergoing thoracic and abdominal surgeries.

Objectives: To develop and validate machine learning (ML) models for identifying key predictors and estimating the risk of intraoperative hypotension (IOH) in elderly patients undergoing general anesthesia. Methods: This secondary analysis included 1,720 elderly surgical patients from a randomized controlled trial. Data were split chronologically into training sets. Feature selection was performed using univariate analysis and the Boruta algorithm. Eight ML models - logistic regression, Bayesian model, K-nearest neighbor, support vector machine, neural network, classification and regression tree, extreme gradient boosting, and random forest - were developed with cross-validation, hyperparameter tuning, and random oversampling. Model performance was evaluated using ROC, PRC, calibration, and decision curve analyses, and interpretability was enhanced using SHapley Additive exPlanations (SHAP). Results: Key predictors included anesthesia protocol, Charlson comorbidity index, preoperative sodium, creatinine, BUN/creatinine ratio, intraoperative drug use (e.g., sevoflurane, lidocaine, morphine), preoperative MAP and MHR, surgical and anesthesia duration, and surgical site. The random forest model achieved the best performance (accuracy=0.9917; MCC=0.9832; AUC-ROC=0.9998; AUC-PRC=0.9998). Conclusions: A robust ML-based model was established to accurately predict IOH in elderly patients. These findings may support individualized anesthesia management and targeted preventive strategies to reduce IOH incidence.