Abstract: SA-PO0286
Use of Artificial Intelligence for Detection of AKI: A Scoping Review
Session Information
- AKI: Epidemiology and Risk Factors
October 24, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 101 AKI: Epidemiology, Risk Factors, and Prevention
Authors
- Acharya, Dilaram, The University of Alabama at Birmingham, Birmingham, Alabama, United States
- Ekrikpo, Udeme E., The University of Alabama at Birmingham, Birmingham, Alabama, United States
- Chen, Jin, The University of Alabama at Birmingham, Birmingham, Alabama, United States
- Neyra, Javier A., The University of Alabama at Birmingham, Birmingham, Alabama, United States
- Ghazi, Lama, The University of Alabama at Birmingham, Birmingham, Alabama, United States
Background
Acute kidney injury (AKI) is a frequent complication across inpatient settings. AKI is associated with significant morbidity, mortality, progression to chronic kidney disease, and increased healthcare costs. Traditional diagnostic criteria relying on serum creatinine and urine output are limited by delayed detection and susceptibility to non-renal confounders. Artificial intelligence (AI) methods leveraging electronic health record (EHR) data have emerged as promising tools for earlier AKI identification and risk stratification.
Methods
We performed a scoping review following PRISMA-ScR guidance. PubMed and Embase were searched for English-language studies published from January 2020 through February 2026. We included clinical studies with sample size ≥1,000 that used advanced AI methods for AKI detection, early prediction, or risk stratification in ICU/critical care, cardiac surgery/perioperative, or contrast-exposure settings. Studies using only conventional statistical models as the primary method, non-clinical studies, reviews, editorials, and conference abstracts were excluded.
Results
Of 2,372 records identified, 103 studies met inclusion criteria. Publication volume increased markedly, with 29 studies in 2025 alone. Studies were predominantly single-center, concentrated in China and the United States, and focused on ICU/critical care (n=56), cardiac surgery/perioperative setting (n=32), and post-contrast AKI (n=15). Most models used structured EHR data (96.1%), with AKI defined by KDIGO criteria in 85.4% of studies. Tree-based ensemble methods dominated: random forest in 53.4%, XGBoost appeared in 50.5% of studies, and LightGBM in 13.6%. Deep learning and transformer-based approaches remained uncommon (≤5% each). Model discrimination was reported in 97.1% of studies and generally strong (median of 0.84), though calibration and external validation was performed in only 37.9% and 34.0% of studies, respectively.
Conclusion
AI-driven models demonstrate strong discriminative performance for early AKI detection across diverse clinical settings. Critical gaps such as lack of external validation, reliance on single-center structured data, and inadequate sociodemographic reporting limit the clinical utility of these models. Prospective, multicenter studies with diverse populations and standardized outcome definitions are needed to support equitable clinical translation.
Funding
- NIDDK Support