Developing a machine learning model to predict renal function decline within three years using one-year longitudinal...

Area of Science:

• Nephrology
• Endocrinology
• Data Science

Background:

• Diabetic kidney disease (DKD) is a major complication of type 2 diabetes mellitus (T2DM), leading to dialysis.
• Predicting renal function decline is crucial for timely intervention.
• Existing prediction models often require advanced stages of DKD or pre-SGLT2 inhibitor data.

Purpose of the Study:

• To evaluate machine learning's ability to predict renal outcomes in T2DM patients.
• To assess renal decline risk using one-year fluctuations in estimated glomerular filtration rate (eGFR).
• To analyze post-SGLT2 inhibitor data and predict risk at any clinical time point.

Main Methods:

• Utilized retrospective outpatient T2DM data with mean eGFR ≥45 mL/min/1.73 m².
• Predicted ≥30% eGFR decline over 3 years using semiannually extracted data.
• Developed machine learning models incorporating demographic, laboratory, and variability data.

Main Results:

• A baseline model achieved an AUC of 0.77.
• Incorporating features like proteinuria, HbA1c range, and eGFR variability improved AUC to 0.82.
• The enhanced model demonstrated strong predictive performance for renal outcomes.

Conclusions:

• Machine learning accurately predicts renal outcomes in T2DM patients with preserved renal function.
• Routinely collected clinical data can facilitate early identification of high-risk patients.
• Timely therapeutic interventions can be enabled through early risk stratification.