Machine learning-based prediction model for postoperative delirium in non-cardiac surgery

Area of Science:

• Medical Informatics
• Surgical Oncology
• Geriatric Medicine

Background:

• Postoperative delirium is a frequent and distressing complication following surgery.
• Accurate prediction of delirium risk is crucial for timely intervention and improved patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning-based prediction model for postoperative delirium.
• To identify key predictors of delirium in patients undergoing non-cardiac surgery.

Main Methods:

• Utilized extreme gradient boosting algorithm on a large dataset (203,374 patients) from Samsung Medical Center.
• Selected top five predictive variables: age, operation duration, physical status, sex, and surgical risk.
• Validated the model's performance using area under the receiver operating characteristic (AUROC) curve and survival analysis.

Main Results:

• The prediction model achieved an AUROC of 0.870 (internal) and 0.867 (external validation).
• Key predictors identified were age, operation duration, physical status, male sex, and surgical risk.
• Patients predicted to be at high risk for delirium showed significantly higher incidence (p < 0.001).

Conclusions:

• Machine learning effectively predicts delirium risk in non-cardiac surgery patients.
• Implementing this prediction model can aid in targeted screening and improved postoperative care.
• The developed model is available online for further research and validation in diverse populations.