Artificial intelligence at the bedside for the prevention, detection, and management of acute kidney injury.

Abstract

Purpose of review: Artificial intelligence is continuously and rapidly evolving. Artificial intelligence has the potential to address several clinical challenges associated with the prevention, detection, and management of acute kidney injury (AKI). This review provides an overview of the state of artificial intelligence for AKI decision-making, highlighting key recent developments, trends, and innovations towards real-world bedside deployment.

Recent findings: External validation of supervised artificial intelligence models for predicting AKI outcomes is now common, with numerous retrospective studies demonstrating strong performance across institutions, patient populations, and international borders. Explainability and transportability of AKI prediction models have become increasingly prioritized, and many recent models use a smaller set of the most widely collected EHR variables with tree-based classifiers. New potential applications focused on supporting bedside AKI decision-making have emerged based on reinforcement learning and causal inference algorithms.

Summary: Although consistency among externally validated AKI models is promising for eventual deployment at the bedside, few have undergone prospective validation, and the real-world clinical impact of artificial intelligence systems for AKI at the bedside remains unclear. Future work should focus on recent advances in artificial intelligence techniques and implementation studies, which assess overall clinical applicability.