A validated multivariable machine learning model to predict cardio-kidney risk in diabetic kidney disease

Area of Science:

• Cardiology
• Nephrology
• Biomarkers
• Machine Learning

Background:

• Diabetic kidney disease (DKD) patients face high risks of cardiac and kidney events.
• Accurate risk stratification is crucial for managing DKD.
• Existing methods may not fully capture the complex risk profile.

Purpose of the Study:

• To develop and validate a precise risk algorithm for predicting cardio-kidney events in DKD.
• To identify key clinical variables and biomarkers for risk assessment.
• To enable better patient stratification and personalized management.

Main Methods:

• Machine learning techniques were employed to develop a risk algorithm.
• Clinical variables and biomarkers were evaluated for predictive ability.
• The algorithm was validated in independent cohorts from CREDENCE and CANVAS trials.

Main Results:

• The final model integrated age, BMI, systolic blood pressure, and NT-proBNP, hs-cTnT, IGFBP-7, and GDF-15.
• The model demonstrated strong predictive performance (C-statistic 0.80) and significant risk stratification.
• Canagliflozin treatment was associated with reduced risk scores and events across risk levels.

Conclusions:

• A validated, parsimonious risk algorithm accurately predicts cardio-kidney outcomes in DKD.
• The algorithm effectively stratifies patients across a broad range of baseline risks.
• This tool can aid in clinical decision-making for DKD management.