Predicting CKD progression using time-series clustering and light gradient boosting machines.

Area of Science:

• Nephrology
• Data Science
• Biostatistics

Background:

• Chronic kidney disease (CKD) progression is difficult to predict due to subtle symptoms and complicating factors.
• Early identification of CKD patient trajectories is crucial for timely intervention and management.

Purpose of the Study:

• To predict kidney function trajectories in non-dialysis CKD patients using advanced machine learning techniques.
• To identify key parameters influencing the estimated Glomerular Filtration Rate (GFR) decline.

Main Methods:

• Applied time-series cluster analysis to stratify 780 CKD patients into groups based on 5-year estimated GFR changes.
• Utilized a light gradient boosting machine algorithm and Shapley Additive Explanation for prediction model development and parameter importance analysis.
• Defined five distinct patient clusters with varying baseline GFRs and observed GFR decline rates.

Main Results:

• Classified participants into five clusters, with GFR decline rates ranging from 4.9% to 45.1% over five years.
• Achieved a prediction accuracy of 0.675 for estimated GFR trajectories.
• Identified baseline estimated GFR (1.61), hemoglobin (0.12), and body mass index (0.11) as the most significant predictors.

Conclusions:

• Baseline estimated GFR is the primary determinant of kidney function transition in CKD patients.
• A threshold of approximately 50 mL/min/1.73 m² for estimated GFR appears critical for trajectory prediction.
• Machine learning models offer a promising approach for personalized CKD progression prediction.