AI in Kidney Disease and Nephrology

Kidney care uses repeated measurements and clinical context. Acute kidney injury can develop during hospitalization, while chronic kidney disease progression is assessed over time using measures such as estimated glomerular filtration rate (eGFR) and urine albumin. AI models may analyze laboratory trends, vital signs, medications and diagnoses in electronic records to estimate a future risk or identify a change that merits closer review. They are intended to support attention and planning, not to diagnose kidney disease from a score alone. One multicenter validation study of a machine-learning AKI risk model included nearly 496,000 hospital admissions across six hospitals in three health systems. It tested the model internally and at other sites; alert thresholds preceded the recorded event by nearly a day and a half. This was a retrospective study with defined inclusion and exclusion criteria, not proof that deploying the alert prevents injury or improves outcomes. Some models have limited evidence for patients with advanced kidney disease, incomplete lab histories or care outside the hospitals where they were developed. For CKD, KDIGO’s 2024 guideline emphasizes assessment of GFR and albuminuria to monitor progression and individualize the frequency of testing. A model may help organize trends or estimate risk, but it should be checked against reliable measurements and the patient’s history. A clinician should consider data gaps, changing conditions and whether an alert is calibrated for the local population. Overreacting to normal variation can trigger unnecessary testing, while a missed alert can create false reassurance. AI can make records easier to scan; the care team determines whether a result warrants action.