Predicting Postoperative Delirium in Orthopedic Surgery Using Explainable Artificial Intelligence.

This study developed a LightGBM model to predict postoperative delirium in orthopedic surgery patients (2017-2023), using SMOTETomek to address data imbalance. LightGBM achieved high performance (AUROC 0.98), identifying older age, multiple diagnoses, and specific medical interventions as key risk factors via SHAP analysis. The model aims to enable targeted preventive interventions, particularly for elderly patients, to reduce delirium incidence and improve recovery. This AI-driven approach supports proactive clinical management and better patient outcomes.