Development and Validation of a Machine Learning Model for Incident Heart Failure Prediction in Chronic Kidney...

Area of Science:

• Nephrology
• Cardiology
• Data Science

Background:

• Chronic kidney disease (CKD) and heart failure (HF) share common pathophysiological pathways.
• HF is a major cardiovascular complication in CKD patients.
• Existing HF prediction models lack accuracy in CKD populations.

Purpose of the Study:

• To develop and validate a CKD-specific risk prediction model for new-onset heart failure.
• To improve the accuracy of HF risk assessment in patients with chronic kidney disease.

Main Methods:

• Utilized a large development set of 52,251 CKD patients from the China Renal Data System for model training and internal validation.
• Externally validated the model in independent Chinese cohorts (21,798 patients) and the UK Biobank (3,323 participants).
• Developed five machine learning models, selecting the top-performing extreme gradient boosting model, simplified to nine key predictors.

Main Results:

• The simplified extreme gradient boosting model demonstrated high predictive performance (AUC 0.879 in Chinese cohort, 0.851 in UK Biobank).
• Key predictors identified include estimated glomerular filtration rate and albuminuria.
• The CKD-specific model significantly outperformed established risk scores like ARIC and CVD-PREDICT.

Conclusions:

• A 9-variable extreme gradient boosting model offers a tailored approach to predict heart failure risk in CKD patients.
• This model can aid clinicians in identifying high-risk individuals within the CKD population.
• Further validation in diverse global populations is recommended to confirm generalizability.