Towards Interpretable End-Stage Renal Disease (ESRD) Prediction: Utilizing Administrative Claims Data with Explainable AI Techniques

Administrative claims data can effectively predict serious kidney disease progression: The study demonstrates that insurance billing records—data already routinely collected by healthcare systems—contain sufficient information to predict which chronic kidney disease patients will progress to end-stage renal disease, potentially enabling earlier interventions.

Deep learning models outperformed traditional approaches for this medical prediction task: LSTM (Long Short-Term Memory) networks, a type of artificial intelligence designed to recognize patterns over time, showed superior performance compared to conventional machine learning methods like Random Forest and XGBoost, suggesting that tracking how patient conditions change over months matters for accurate predictions.

Explainability is achievable even with complex AI models: The researchers used SHAP analysis to make their deep learning model's decisions transparent, allowing clinicians to understand which specific patient factors drove individual predictions rather than treating the model as a "black box"—an important requirement for responsible deployment in healthcare.

A 24-month observation window proved optimal for predictions: The study found that examining patient data over two years provided the best balance for predicting ESRD progression, offering practical guidance for healthcare systems implementing such tools in real-world settings.

Existing healthcare data can be repurposed for disease management: Rather than requiring new data collection, this research shows that administrative claims data already held by insurance companies represents an underutilized resource for improving patient outcomes in chronic kidney disease management.