Overview

• Researchers linked clinical electronic records with insurance claims to train machine learning and deep learning models that outperformed single-source approaches.
• Explainable AI techniques supported interpretability of predictions and helped address bias for potential clinical decision support.
• A 24-month observation window provided the strongest balance between early detection and accuracy when forecasting progression to end-stage renal disease.
• Adopting the 2021 eGFR equation improved predictive accuracy and reduced racial bias, with gains reported for African American patients.
• The team emphasized limits from single-institution data and EHR biases and urged external validation before wider clinical use.