Using Machine Learning Algorithms to Predict High-Risk Factors for Postoperative Delirium in Elderly Patients

Area of Science:

• Geriatric Medicine
• Medical Informatics
• Surgical Complications

Background:

• Postoperative delirium (POD) is a frequent complication in elderly patients undergoing non-brain surgery.
• POD significantly impacts both short-term and long-term patient prognosis.
• Accurate prediction of POD is crucial for timely clinical intervention.

Purpose of the Study:

• To develop and validate a machine learning model for predicting POD in elderly patients.
• To identify key preoperative, intraoperative, and postoperative risk factors associated with POD.
• To enhance clinical decision-making for preventing and managing POD.

Main Methods:

• Utilized data from 950 elderly patients, including 132 with POD.
• Collected 30 characteristic variables encompassing demographics, medical history, and surgical details.
• Applied and compared three machine learning algorithms: MLP, XGBoost, and KNN, using k-fold cross-validation, ROC curves, calibration curves, and DCA for evaluation.

Main Results:

• XGBoost exhibited superior performance with an AUC of 0.982 (training set) and 0.924 (validation set).
• External validation showed an AUC of 0.88, indicating strong generalizability.
• The XGBoost model demonstrated high predictive accuracy, stability, and clinical utility.

Conclusions:

• The developed machine learning model, particularly XGBoost, offers high prediction accuracy for POD in elderly patients.
• The model aids clinicians in timely diagnosis and treatment planning.
• This predictive tool enhances the management of postoperative complications in geriatric surgical patients.