ARTIFICIAL INTELLIGENCE FOR PREDICTING CHRONIC KIDNEY DISEASE PROGRESSION: A REVIEW OF CURRENT EVIDENCE AND GLOBAL CLINICAL IMPLICATIONS

Abstract

Chronic Kidney Disease (CKD) affects approximately 800 million people globally and represents a growing public health and economic burden. Traditional prognostic tools, such as the Kidney Failure Risk Equation, demonstrate limited discriminative accuracy in early disease stages (G1–G3), where kidney failure events are inherently rare. Aim: The primary objective of this review is to evaluate the performance of artificial intelligence (AI) and machine learning (ML) models in predicting CKD progression. Methods: A narrative review approach was used to synthesise evidence published between January 2022 and June 2026, identified through searches of PubMed/MEDLINE, Embase, and Web of Science. Ten studies comprising diverse international cohorts were included. Key findings: AI/ML models demonstrated superior predictive accuracy compared with traditional benchmarks, achieving a pooled Area Under the Curve (AUC) of 0.87–0.89. Long Short-Term Memory recurrent neural networks, leveraging longitudinal electronic health record data, attained AUROC scores of 0.957–0.967. Tree-based ensemble methods, including Random Forest and XGBoost, showed robust performance across heterogeneous populations. However, a consistent imbalance between high model specificity (0.87–0.92) and relatively low sensitivity (0.43–0.68) was identified, primarily attributable to class imbalance in training datasets. Explainable AI frameworks, particularly SHAP and LIME, successfully addressed the interpretability barrier, identifying eGFR slope, proteinuria, and haemoglobin as dominant predictors of renal decline. Conclusions: These findings support a transition toward proactive, precision-based nephrology. Global clinical adoption, however, requires standardisation of predictor criteria, optimisation of model sensitivity, broader external validation across diverse ethnic populations, and targeted solutions for implementation in low- and middle-income healthcare settings.