Predicting postoperative delirium after hip arthroplasty for elderly patients using machine learning.

Area of Science:

• Geriatric Medicine
• Artificial Intelligence in Healthcare
• Surgical Complications

Background:

• Postoperative delirium (POD) is a significant complication in elderly patients undergoing hip arthroplasty.
• Early identification and prediction of POD are crucial for patient outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting POD in elderly hip-arthroplasty patients.
• To identify essential features associated with POD in this patient cohort.

Main Methods:

• Utilized electronic health record data from 476 elderly hip-arthroplasty patients (Jan 2017-Apr 2021).
• Employed the Confusion Assessment Method (CAM) for delirium assessment.
• Applied feature selection (mutual information) and logistic regression (LR) for model development and validation.
• Evaluated model performance using AUC, accuracy, sensitivity, specificity, and F1-score.

Main Results:

• The final ML model, combining mutual information and logistic regression, demonstrated high predictive performance.
• Achieved an Area Under the Curve (AUC) of 0.94, accuracy (ACC) of 0.88, sensitivity of 0.85, specificity of 0.90, and F1-score of 0.87.
• Identified key predictors of POD including age, Cystatin C, GFR, CHE, CRP, LDH, monocyte count, mental illness history, psychotropic drug use, and intraoperative blood loss.

Conclusions:

• The developed ML model accurately predicts POD in elderly hip-arthroplasty patients.
• Identifying and managing risk factors such as age, specific biomarkers, and pre-existing conditions can mitigate POD incidence.
• Preoperative interventions targeting identified factors can significantly reduce the occurrence of POD.