Explainable Machine Learning-Based Prediction of Postoperative Hypoxemia in Elderly Patients Undergoing General Anesthesia

In order to predict the risk of postoperative hypoxemia in elderly patients undergoing general anesthesia, this study developed, validated, and interpreted explainable machine learning-based models. One thousand six hundred senior individuals (≥60 years) participated in a retrospective, single-center cohort research. Comprehensive perioperative data were gathered, including laboratory indices, intraoperative physiological parameters, comorbidities, and demographics. To improve the interpretability of the model, feature selection was carried out using least absolute shrinkage and selection operator (LASSO) regression after thorough data preparation. A held-out test dataset was used to design and internally validate four machine learning algorithms: logistic regression, support vector machine, random forest, and extreme gradient boosting (XGBoost). The area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, calibration curves, and decision curve analysis were used to evaluate the model's performance. To give clear justifications for model predictions, feature importance evaluations were carried out. About 15.0% (240/1,600) of patients experienced postoperative hypoxemia. Preoperative serum albumin level, lowest intraoperative mean arterial pressure, age, total crystalloid infusion volume, and a history of chronic obstructive pulmonary disease were important prognostic factors that were consistently found across models. The XGBoost-based explainable AI model outperformed the other models in terms of predictive performance, with the highest AUC of 0.994 (95% CI: 0.990-0.998), good calibration, and a notable net therapeutic benefit on decision curve analysis. The suggested explainable machine learning XGBoost model predicts postoperative hypoxemia in elderly patients receiving general anesthesia with accuracy and interpretability. This method facilitates focused interventions to improve surgical respiratory outcomes by supporting early risk stratification and informed perioperative decision-making by clarifying important risk factors influencing individual predictions.