Identifying Diabetic Kidney Disease in Type 2 Diabetes Patients Using Explainable Machine Learning: A Case-Control...

Area of Science:

• Nephrology
• Medical Informatics
• Machine Learning

Background:

• Diabetic kidney disease (DKD) is a major complication of Type 2 diabetes.
• Early identification of DKD is crucial for effective management and prevention of progression.

Purpose of the Study:

• To develop and validate a machine learning-driven predictive tool for identifying diabetic kidney disease (DKD).
• To assess the clinical utility and key predictors of DKD using machine learning models.

Main Methods:

• Developed and validated prediction models using data from 1463 patients.
• Employed Least Absolute Shrinkage and Selection Operator regression for feature selection.
• Compared Random Forest (RF), extreme gradient boosting, support vector machine, and logistic regression using AUC-ROC, AUC-PR, accuracy, and F1-score.
• Evaluated clinical utility with decision curve and calibration analyses; interpreted feature importance using SHAP and LIME.

Main Results:

• The full RF model achieved superior performance in screening for DKD (AUC-ROC = 0.906, AUC-PR = 0.902, accuracy = 0.830, F1 = 0.847).
• The RF model demonstrated favorable clinical net benefit and good calibration.
• Key predictors included urine α1-microglobulin, hypertension, 24-h urinary total protein, diabetes duration, systolic blood pressure, serum retinol-binding protein, complement C1q, and 25-hydroxyvitamin D.

Conclusions:

• A Random Forest prediction model was successfully developed for early DKD screening.
• The model highlights the significant roles of specific clinical and laboratory factors in predicting DKD.
• This tool can assist in the early identification and management of diabetic kidney disease.