Development and Validation of a Machine Learning-Based Risk Prediction Model for Postoperative Delirium in Older Patients with Hip Fracture

Abstract

Background Postoperative delirium (POD) is associated with impaired cognitive function, increased morbidity, and mortality. Early identification of high-risk patients is critical for effective intervention. Methods Data from 2,516 older patients with hip fractures treated at the First Medical Center of the Chinese PLA General Hospital were retrospectively collected. Logistic Regression (LR), Random Forest (RF), Classification and Regression Tree (CART), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost) were used to construct the prediction models. SHapley Additive exPlanation (SHAP) analysis was performed to visualize the optimal model. External validation was conducted on 176 patients from March 2022 to November 2023 to assess the model's clinical applicability. Results The training dataset included 2,516 older patients, of which 367 (14.59%) developed POD. XGBoost demonstrated the best predictive performance (AUC = 0.92; accuracy = 86.4%; sensitivity = 87.7%; specificity = 85.1%; Brier score = 0.15). SHAP analysis ranked PNI (Prognostic Nutritional Index), ASA (American Society of Anesthesiologists classification), and age as the top three predictors. External validation on 176 patients showed the XGBoost model maintained strong performance (AUC = 0.89; accuracy = 83.0%; sensitivity = 95.8%; specificity = 80.9%; Brier score = 0.15). Conclusions An ML-based model was developed and validated to predict postoperative delirium risk in older patients with hip fracture. These findings may help to develop personalized interventions to provide better treatment plans and optimal resource allocation. The interpretable framework can increase the transparency of the model and facilitate understanding the reliability of the predictive model for the physicians.