Current and Future Applications of AI-Driven Predictive Modeling and a Proposed Framework for AI-Bioprognostics in Kidney Care.

Abstract

Artificial intelligence (AI) is reshaping kidney care by integrating electronic health records (EHRs), imaging, digital pathology, genomics, wearable biosensors, and longitudinal physiologic data into dynamic risk-prediction systems. Traditional nephrology risk stratification has relied on static variables and regression-based models such as estimated glomerular filtration rate (eGFR), albuminuria, and composite clinical scores, which often inadequately capture nonlinear disease trajectories and phenotypic heterogeneity. Advances in machine learning, deep learning, and multimodal foundation models are accelerating the shift toward predictive, preventive, and precision nephrology. This narrative review evaluated AI applications across acute kidney injury (AKI), chronic kidney disease (CKD), dialysis, and transplantation. PubMed, Embase, Scopus, Web of Science, and IEEE Xplore were searched for English-language studies published from January 2015 through July 2026, prioritizing those reporting discrimination, calibration, external validation, or implementation outcomes. Across AKI, CKD, dialysis, and kidney transplantation, AI-based models have frequently demonstrated improved discrimination relative to conventional approaches, although the magnitude of improvement varies substantially across populations, prediction targets, comparators, and validation settings. In transplantation, multimodal systems integrating histopathology, donor-derived biomarkers, and clinical variables have improved prediction of rejection and graft failure. Emerging bioprognostic frameworks incorporating wearables, dialysis telemetry, and molecular biomarkers may support dynamic or near-real-time risk estimation of hyperkalemia, intradialytic hypotension, and cardiovascular instability. Prospective validation, assessment of calibration drift, and formal fairness analyses remain limited across the published literature, while dataset shift, restricted generalizability, algorithmic opacity, and workflow integration continue to constrain clinical adoption. AI-driven prediction and bioprognostics have the potential to support a transition from predominantly reactive kidney care toward more anticipatory and continuously informed precision care, although improved predictive performance has not yet been consistently shown to translate into improved patient outcomes. Realizing this potential will require rigorous external validation, prospective clinical-impact evaluation, equitable deployment, interoperable infrastructure, and human-in-the-loop oversight.