Logistic Regression and Machine Learning Algorithms for the Risk Prediction of Perioperative Adverse Cardiovascular...

Area of Science:

• Cardiology
• Geriatric Medicine
• Anesthesiology

Background:

• Elderly patients undergoing noncardiac surgery face increased risks of perioperative adverse cardiovascular events (PACEs), leading to poorer prognosis and higher mortality.
• Early identification of at-risk individuals is crucial for effective prevention and management strategies.

Purpose of the Study:

• To develop and validate predictive models for PACEs in elderly patients undergoing noncardiac surgery.
• To stratify risk and improve clinical decision-making for better patient outcomes.

Main Methods:

• A retrospective study of 8309 elderly patients undergoing noncardiac surgery.
• Development of logistic regression and machine learning models using clinical data, biomarkers, and risk factors.
• Evaluation of models using ROC curves, decision curves, calibration curves, sensitivity, specificity, and F1-score.

Main Results:

• A logistic regression model achieved an AUC of 0.895, identifying Pro-BNP, cardiac function, and creatinine as key predictors (PCC indicator).
• Machine learning models accurately predicted PACEs across risk strata (low-risk precision 0.86, high-risk precision 0.970) with high sensitivity (0.736) and specificity (0.973).
• The models demonstrated strong predictive performance and clinical value in identifying high-risk patients.

Conclusions:

• Established risk prediction models for PACEs in elderly patients undergoing noncardiac surgery exhibit good predictive accuracy.
• These models provide a scientific basis for timely diagnostic and treatment adjustments, potentially reducing PACEs-related mortality.