Development and External Validation of a Machine Learning Model for Progression of CKD

Area of Science:

• Nephrology
• Data Science
• Machine Learning

Background:

• Chronic kidney disease (CKD) affects millions globally, with progression leading to kidney failure.
• Predicting CKD progression is crucial for timely intervention and improved patient outcomes.
• Current prediction methods may not fully leverage the wealth of routinely collected laboratory data.

Purpose of the Study:

• To develop and externally validate a machine learning model for predicting CKD progression.
• To utilize a comprehensive set of demographic and laboratory features for prediction.
• To assess the model's accuracy and generalizability in diverse patient cohorts.

Main Methods:

• A random forest model was developed using data from a population-based cohort in Manitoba, Canada.
• External validation was performed using data from Alberta, Canada.
• The model incorporated over 80 laboratory features and demographic data to predict a 40% decline in estimated glomerular filtration rate (eGFR) or kidney failure.

Main Results:

• The model demonstrated strong predictive performance with an area under the receiver operating characteristic curve (AUC) of 0.88 at 2 years and 0.84 at 5 years in internal testing.
• These results were validated externally, showing preserved discrimination and calibration (AUC 0.87 at 2 years, 0.84 at 5 years).
• The top 30% of individuals identified as high or intermediate risk accounted for 87% of CKD progression events within 2 years and 77% within 5 years.

Conclusions:

• Machine learning models utilizing routinely collected laboratory data can accurately predict eGFR decline or kidney failure.
• This predictive tool offers a promising approach for proactive CKD management.
• The model's accuracy and generalizability support its potential clinical application in identifying at-risk individuals.