Artificial Intelligence Across the Kidney Disease Continuum: A Systematic Review of Applications from Acute Kidney Injury to Precision Nephrology

Abstract

Background: Artificial intelligence (AI), including machine learning and deep learning (DL), is increasingly applied in nephrology, but evidence remains fragmented across clinical domains. Objectives: To systematically identify, appraise and synthesise original studies evaluating AI applications across the kidney disease continuum. Methods: We searched PubMed/MEDLINE, Scopus, Web of Science and Google Scholar for studies published from January 2015 to December 2024. Original studies developing, validating or implementing AI models in kidney disease were eligible. Two reviewers independently screened records and extracted data. Risk of bias was assessed with PROBAST, and certainty of evidence with GRADE principles. Owing to heterogeneity, findings were synthesised narratively by clinical domain following SWiM guidance. Results: Of 3,200 records, 142 studies met the inclusion criteria: acute kidney injury (AKI; n = 24), chronic kidney disease (CKD; n = 51), renal histopathology (n = 7), renal imaging (n = 5), dialysis (n = 25), transplantation (n = 3), diabetic kidney disease (n = 3), urolithiasis (n = 6), renal cell carcinoma (n = 6) and other renal conditions (n = 12). Reported discrimination was generally high (most AUCs 0.78–0.96). DL achieved more than 90% accuracy in imaging and histopathology tasks. Most studies were retrospective and single-centre, few were externally validated, calibration was seldom reported, and 20 CKD studies reused one 400-record public dataset. Few evaluated prospective deployment. Conclusions: AI models show promising discrimination throughout nephrology, but the certainty of evidence is low to moderate. External validation, calibration reporting, adherence to TRIPOD+AI and prospective impact trials are needed before routine clinical adoption. Registration: Not prospectively registered.

Keywords: artificial intelligence; machine learning; deep learning; acute kidney injury; chronic kidney disease; dialysis; kidney transplantation; systematic review