Using machine learning algorithms to predict MACE in peritoneal dialysis patients.

Area of Science:

• Nephrology
• Cardiology
• Artificial Intelligence

Background:

• Major adverse cardiac events (MACE) are a significant concern for patients undergoing peritoneal dialysis (PD).
• Predicting MACE risk in PD patients is crucial for timely intervention and improved outcomes.
• Existing risk stratification methods may not fully capture the dynamic nature of MACE development over time.

Purpose of the Study:

• To develop and validate machine learning (ML) algorithms for predicting MACE risk in PD patients.
• To incorporate a time-dependent factor, predicting MACE risk at 1-year and 5-year follow-ups.
• To identify key clinical variables influencing MACE development in this population.

Main Methods:

• A retrospective study of 1006 PD patients from January 2010 to December 2016.
• Utilized XGBoost, Random Forest (RF), and Adaboost ML algorithms to train predictive models.
• Evaluated model performance using Area Under the Curve (AUC) for overall, 1-year, and 5-year MACE prediction.

Main Results:

• The RF model achieved an AUC of 0.80 for overall MACE prediction, with Parathyroid hormone, Congestive heart failure, and Age as top predictors.
• The XGBoost model (AUC=0.86) was optimal for 1-year MACE prediction, with High-Density Lipoprotein Cholesterol (HDL-C), Age, and Calcium as key factors.
• The RF model (AUC=0.75) performed best for 5-year MACE prediction, with Age, Creatinine, and estimated Glomerular Filtration Rate being most influential.

Conclusions:

• Novel ML algorithms were developed and validated to predict MACE risk in PD patients.
• The study highlights the importance of specific biomarkers and clinical factors in MACE prediction over different time horizons.
• These predictive models offer a promising tool for risk stratification and personalized management of cardiac health in PD patients.