CKD Progression Prediction in a Diverse US Population: A Machine-Learning Model.

Area of Science:

• Nephrology
• Data Science
• Biostatistics

Background:

• Chronic kidney disease (CKD) is a significant global health issue associated with high morbidity and mortality.
• Predictive models for CKD progression are lacking, especially for early disease stages.
• Early identification of CKD progression is crucial for timely intervention and management.

Purpose of the Study:

• To develop and validate machine learning models for predicting CKD progression using readily available demographic and laboratory data.
• To assess the performance of these models across the spectrum of CKD.
• To identify key predictors of accelerated kidney function decline.

Main Methods:

• A retrospective observational study utilizing deidentified laboratory data from 110,264 adult patients over 5 years.
• Machine learning models, specifically random forest survival methods, were employed.
• Predictors included demographic and laboratory characteristics, with a focus on estimated glomerular filtration rate (eGFR) slope and urine albumin-creatinine ratio.

Main Results:

• A 7-variable risk classifier achieved an area under the curve (AUC) of 0.85 for predicting >30% eGFR decline within 5 years.
• The most significant predictor of CKD progression was the eGFR slope.
• Other key predictors included initial eGFR, urine albumin-creatinine ratio, serum albumin (initial and slope), age, and sex.

Conclusions:

• The developed machine learning classifier accurately predicts significant eGFR decline in patients with CKD.
• The model effectively utilizes easily obtainable laboratory data for risk prediction.
• This tool has the potential to enhance early recognition and optimize management strategies for patients at risk of CKD progression.