A Comparison of Data Sampling Techniques for Predicting Postoperative Delirium in Neurosurgery

Area of Science:

• Neurosurgery
• Artificial Intelligence
• Medical Informatics

Background:

• Postoperative delirium is a significant concern in neurosurgical patients.
• Early prediction and prevention strategies are crucial for improving patient outcomes.

Purpose of the Study:

• To develop and validate an XGBoost model for predicting postoperative delirium in neurosurgical patients.
• To identify key risk factors associated with delirium development.

Main Methods:

• Utilized a dataset of neurosurgical patients from 2017-2023.
• Employed various data sampling techniques to address class imbalance, including SMOTETomek.
• Developed an XGBoost classification model to predict delirium risk.

Main Results:

• The SMOTETomek under-over sampling technique demonstrated superior performance.
• Achieved an Area Under the Receiver Operating Characteristic Curve (AUROC) of 0.97.
• Identified significant risk factors: preoperative hospital stay, ICU admission, multiple diagnoses, fall risk, and prior neurological procedures.

Conclusions:

• The developed XGBoost model effectively predicts postoperative delirium in neurosurgical patients.
• The model facilitates early identification of high-risk individuals.
• Findings support tailored prevention strategies to mitigate delirium incidence.