Prediction models for postoperative delirium in elderly patients with machine-learning algorithms and SHapley...

Area of Science:

• Geriatric Medicine
• Orthopedic Surgery
• Data Science in Healthcare

Background:

• Postoperative delirium (POD) is a frequent and serious complication in elderly patients undergoing hip fracture surgery.
• Early identification of patients at high risk for POD is crucial for optimizing treatment and improving patient outcomes.

Purpose of the Study:

• To develop and evaluate predictive models for postoperative delirium in elderly hip fracture patients.
• To compare the performance of machine learning algorithms against conventional logistic regression for POD prediction.

Main Methods:

• A retrospective study involving 797 elderly patients (≥65 years) who underwent hip fracture surgery.
• Construction of prediction models using logistic regression and five machine learning algorithms (Random Forest, GBM, AdaBoost, XGBoost, SVM).
• Model performance evaluated using Area Under the Receiver Operating Characteristic Curve (AUC-ROC), accuracy, sensitivity, and precision; feature importance assessed with SHAP.

Main Results:

• The incidence of POD was 9.28% (74/797 patients).
• A logistic regression nomogram achieved an AUC of 0.71.
• Machine learning models demonstrated superior performance, with Random Forest (AUC 0.81), GBM (AUC 0.80), and XGBoost (AUC 0.77) showing the highest predictive power.
• Random Forest achieved the highest sensitivity (91.9%), while SVM had the highest precision (67.8%).

Conclusions:

• Machine learning algorithms offer a more effective approach to predicting POD in elderly hip fracture patients compared to traditional logistic regression.
• These models can facilitate convenient risk stratification, potentially leading to better management and outcomes for this vulnerable patient population.