Machine Learning Model Predicts Postoperative Delirium in Elderly CHD Patients

Overview

A retrospective cohort study developed a gradient boosting machine (GBM) model to predict postoperative delirium (POD) in elderly patients with coronary heart disease (CHD) undergoing non-cardiac surgery. The model demonstrated strong predictive performance with an AUC of 0.856 and identified key predictors including Clinical Frailty Scale (CFS), Mini-mental State Examination (MMSE), and Athens Insomnia Scale (AIS) scores.

Background

Postoperative delirium is a common and serious complication in elderly surgical patients, associated with increased morbidity and mortality. Elderly patients with CHD are at particularly high risk due to combined effects of aging and cardiac disease. Early identification of patients at risk for POD is critical for implementing preventive strategies. Machine learning offers advanced methods to analyze complex clinical data and improve risk prediction accuracy beyond traditional approaches.

Data Highlights

Key Findings

• Seven key features were selected for model development using Boruta, LASSO, and logistic regression methods.
• The GBM model outperformed nine other machine learning models in predicting POD.
• Higher Clinical Frailty Scale (CFS) grades significantly increased POD risk prediction.
• Lower Mini-mental State Examination (MMSE) scores were strongly associated with increased POD risk.
• Higher Athens Insomnia Scale (AIS) scores contributed to better prediction of POD occurrence.
• An easy-to-use online calculator based on the GBM model was developed to facilitate clinical application.

Clinical Implications

The GBM model provides clinicians with a reliable tool to identify elderly CHD patients at high risk for postoperative delirium, enabling targeted preventive interventions. Incorporating assessments of frailty, cognitive function, and sleep quality into preoperative evaluations may improve risk stratification. External validation is necessary before widespread clinical implementation.

Conclusion

This study successfully developed and internally validated a machine learning model with strong predictive performance for POD in elderly patients with CHD. Key clinical features such as frailty, cognition, and insomnia are critical predictors, supporting their inclusion in preoperative risk assessments.

References

1. Study Authors/Chinese Clinical Trial Registry/2025 -- Predictive Model Utilizing Machine Learning for Postoperative Delirium in Elderly Coronary Heart Disease Patients Undergoing Non-Cardiac Surgery