Machine learning in risk prediction of continuous renal replacement therapy after coronary artery bypass grafting

Qian Zhang1, Peng Zheng1, Zhou Hong2

  • 1Department of Cardiology, The First Affiliated Hospital of Nanjing Medical University, Nanjing, Jiangsu, China.

Insights

Machine learning models predict continuous renal replacement therapy (CRRT) risk after coronary artery bypass grafting (CABG) surgery. The Gaussian Naïve Bayes model identified key predictors, aiding in optimizing patient care.

Area of Science:

  • Medical Informatics
  • Cardiovascular Surgery
  • Nephrology

Background:

  • Continuous renal replacement therapy (CRRT) is a critical intervention for intensive care unit (ICU) patients post-coronary artery bypass grafting (CABG).
  • Predicting the need for CRRT after CABG is essential for timely intervention and improved patient outcomes.
  • Existing risk prediction methods may not fully capture the complexity of CRRT requirement post-CABG.

Purpose of the Study:

  • To develop and validate machine learning models for predicting CRRT risk in ICU patients undergoing CABG surgery.
  • To identify significant risk factors associated with CRRT requirement post-CABG.
  • To enhance perioperative management strategies through accurate risk stratification.

Main Methods:

  • Utilized electronic medical record data from CABG patients.
  • Employed the Boruta method for feature selection and developed seven machine learning algorithms.
  • Validated models using 10-fold cross-validation, assessing discrimination with AUC and calibration plots.
  • Applied SHapley Additive exPlanations (SHAP) for feature interpretability.

Main Results:

  • 72 (37.89%) patients required CRRT post-CABG, exhibiting higher mortality.
  • The Gaussian Naïve Bayes (GNB) model demonstrated the highest predictive performance (AUC).
  • Key predictors included cardiac troponin T, creatine kinase isoenzyme, albumin, LDL cholesterol, NYHA class, serum creatinine, and age.
  • SHAP analysis provided individualized risk predictions.

Conclusions:

  • Machine learning models effectively predict CRRT risk in the post-CABG ICU population.
  • Identified crucial risk variables for CRRT, facilitating early intervention.
  • The developed models can aid in optimizing perioperative care for high-risk patients.
Abstract