Predicting postoperative pulmonary infection risk in patients with diabetes using machine learning

Abstract

Background: Patients with diabetes face an increased risk of postoperative pulmonary infection (PPI). However, precise predictive models specific to this patient group are lacking. Objective: To develop and validate a machine learning model for predicting PPI risk in patients with diabetes. Methods: This retrospective study enrolled 1,269 patients with diabetes who underwent elective non-cardiac, non-neurological surgeries at our institution from January 2020 to December 2023. Predictive models were constructed using nine different machine learning algorithms. Feature selection was conducted using Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression. Model performance was assessed via the Area Under the Curve (AUC), precision, accuracy, specificity and F1-score. Results: The Ada Boost classifier (ADA) model exhibited the best performance with an AUC of 0.901, Accuracy of 0.91, Precision of 0.82, specificity of 0.98, PPV of 0.82, and NPV of 0.82. LASSO feature selection identified six optimal predictive factors: postoperative transfer to the ICU, Age, American Society of Anesthesiologists (ASA) physical status score, chronic obstructive pulmonary disease (COPD) status, surgical department, and duration of surgery. Conclusion: Our study developed a robust predictive model using six clinical features, offering a valuable tool for clinical decision-making and personalized prevention strategies for PPI in patients with diabetes.