Abstract: TH-PO014
Artificial Intelligence/Machine Learning Externally Validated Models for AKI Risk-Classification: A Systematic Review and Meta-Analysis
Session Information
- Augmented Intelligence for Prediction and Image Analysis
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Cama-Olivares, Augusto, The University of Alabama at Birmingham Division of Nephrology, Birmingham, Alabama, United States
- Braun, Chloe Grace, The University of Alabama at Birmingham Department of Pediatrics, Birmingham, Alabama, United States
- Takeuchi, Tomonori, The University of Alabama at Birmingham Division of Nephrology, Birmingham, Alabama, United States
- Kaiser, Kathryn A., The University of Alabama at Birmingham Department of Health Behavior, Birmingham, Alabama, United States
- Ghazi, Lama, The University of Alabama at Birmingham Department of Epidemiology, Birmingham, Alabama, United States
- Chen, Jin, The University of Alabama at Birmingham Division of Nephrology, Birmingham, Alabama, United States
- Forni, Lui G., University of Surrey Department of Clinical and Experimental Medicine, Guildford, United Kingdom
- Kane-Gill, Sandra L., University of Pittsburgh Department of Pharmacy and Therapeutics, Pittsburgh, Pennsylvania, United States
- Ostermann, Marlies, Guy's and St Thomas' NHS Foundation Trust, London, London, United Kingdom
- Shickel, Benjamin, University of Florida Division of Nephrology Hypertension & Renal Transplantation, Gainesville, Florida, United States
- Ninan, Jacob, Mayo Clinic Minnesota, Rochester, Minnesota, United States
- Neyra, Javier A., The University of Alabama at Birmingham Division of Nephrology, Birmingham, Alabama, United States
Background
Artificial Intelligence (AI) through machine learning (ML) models can provide accurate and precise acute kidney injury (AKI) risk classification, but their extent and performance in real-world settings have not been established.
Methods
PubMed, EMBASE, Web of Science, and Scopus were searched until 08/2023. Articles reporting on externally validated models for prediction of AKI onset, AKI severity, and post-AKI complications in hospitalized adult and pediatric patients were searched using text words related to AKI, AI, and ML. Two independent reviewers screened article titles, abstracts, and full texts. Areas under the receiver operating characteristic curves (AUCs) were used to compare model discrimination and pooled using random-effects model.
Results
Of 4,280 articles initially identified and screened, 85 were included with 2.8 million admissions (study sample dates ranged from 1996 to 2022). The KDIGO criteria were the most frequently used to define AKI (72.9%). We identified 291 models, with the most commonly reported being logistic regression (35.1%), random forest (9.6%), and XGBoost (9.6%). The most frequently reported predictors of AKI onset were age, sex, diabetes, serum creatinine, and hemoglobin. The pooled AUC for AKI onset was 0.82 (95% CI, 0.80-0.84) and 0.78 (95% CI, 0.76-0.80) for internal and external validation, respectively. Pooled AUC across multiple clinical settings, AKI severities, and post-AKI complications ranged from 0.78 to 0.87 for internal validation and 0.75 to 0.84 for external validation. Although data were limited, results in the pediatric population aligned with those observed in adults. Between-study heterogeneity was high for all outcomes (I2 >90%), and most studies presented high-risk of bias (72.9%) according to the Prediction model Risk Of Bias ASsessment Tool (PROBAST).
Conclusion
Most externally validated models performed well in predicting AKI onset, AKI severity, and post-AKI complications in hospitalized adult and pediatric populations. However, heterogeneity in clinical settings, study populations, and predictors limits their generalizability and implementation.