Machine Learning Based Prediction of Postoperative Acute Kidney Injury Risk in Coronary Artery Bypass Grafting...

Area of Science:

• Cardiology
• Nephrology
• Medical Informatics

Background:

• Postoperative acute kidney injury (AKI) is a significant complication following coronary artery bypass grafting (CABG).
• AKI after CABG is associated with increased mortality and prolonged hospital stays.
• Reliable predictive models for early AKI detection post-CABG are currently lacking.

Purpose of the Study:

• To develop and evaluate machine learning models for the early prediction of AKI in patients undergoing CABG.
• To identify key clinical variables that predict AKI development after CABG.

Main Methods:

• Data from 520 CABG patients were analyzed, split into training (70%) and validation (30%) sets.
• Six machine learning models (RF, XGBoost, LR, LightGBM, Softmax Regression, SVM) were constructed after variable screening using LASSO regression.
• SHapley Additive exPlanations (SHAP) was employed to determine feature importance.

Main Results:

• The incidence of post-CABG AKI was 25.96%.
• The XGBoost model demonstrated superior performance in the training group (AUC=0.89).
• Logistic Regression and Softmax Regression models showed high stability in the validation group (AUC=0.86).
• Estimated glomerular filtration rate (eGFR), intraoperative epinephrine, and calcium levels were identified as top predictors by SHAP analysis.

Conclusions:

• Machine learning models can effectively predict AKI following CABG.
• These models facilitate the early identification of patients at high risk for developing AKI.
• Early prediction allows for timely intervention to mitigate adverse outcomes.