Interpretable machine learning model based on routine metabolic laboratory indices to identify advanced chronic...

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Early identification of advanced chronic kidney disease (CKD) is crucial for timely intervention.
• Current risk equations often rely on unavailable data like albuminuria.
• Metabolic and endocrine disturbances characterize advanced CKD.

Purpose of the Study:

• To develop and validate machine learning models for distinguishing advanced CKD (G3a-5) from preserved kidney function (G1-2).
• To identify key metabolic parameters predictive of advanced CKD.
• To assess the potential of these models for routine clinical use.

Main Methods:

• Retrospective analysis of adult patients from university-affiliated departments.
• Development and validation of five machine learning algorithms using demographic, clinical, and metabolic laboratory data.
• Performance evaluation using discrimination (AUC), calibration, and interpretability (feature importance, SHAP).

Main Results:

• Gradient Boosting classifier demonstrated superior performance (AUC = 0.972 internally, 0.965 externally) with good calibration.
• Key predictors included urea, kidney disease type, phosphorus, albumin, and lipid parameters.
• The model identified systemic metabolic dysregulation as a significant factor.

Conclusions:

• An interpretable Gradient Boosting model effectively identifies advanced CKD using routinely available metabolic data.
• The model captures clinically relevant metabolic patterns associated with CKD severity.
• This approach supports integration into electronic health records for CKD risk stratification.