Enhancing end-stage renal disease outcome prediction: a multisourced data-driven approach

Area of Science:

• Nephrology
• Biomedical Informatics
• Artificial Intelligence in Healthcare

Background:

• Chronic kidney disease (CKD) poses a significant public health challenge, with progression to end-stage renal disease (ESRD) leading to substantial morbidity and mortality.
• Accurate prediction of CKD progression is crucial for timely intervention and improved patient outcomes.
• Existing prediction models often lack interpretability and may be subject to biases.

Purpose of the Study:

• To enhance the prediction of CKD progression to ESRD using machine learning (ML) and deep learning (DL) models.
• To evaluate the impact of integrated clinical and claims data with varying observation windows on prediction accuracy.
• To leverage explainable artificial intelligence (AI) for model interpretability and bias reduction.

Main Methods:

• Utilized integrated clinical and claims data from 10,326 CKD patients (2009-2018).
• Evaluated multiple ML and DL models across five distinct observation windows.
• Employed feature importance and SHapley Additive exPlanations (SHAP) for interpretability and bias assessment.

Main Results:

• Integrated data models significantly outperformed single-source models, with Long Short-Term Memory (LSTM) achieving an AUROC of 0.93.
• A 24-month observation window demonstrated an optimal balance between early detection and prediction accuracy.
• The 2021 eGFR equation improved prediction and reduced racial bias, particularly for African American patients.

Conclusions:

• Advanced ML/DL models utilizing integrated data offer improved accuracy and interpretability for predicting CKD progression.
• Explainable AI techniques are vital for understanding model predictions and mitigating biases.
• This framework has the potential to enhance CKD management, guide interventions, and reduce healthcare disparities.