Kidney Week

Abstract: FR-PO0008

Prediction of 30-Day Mortality After Kidney Replacement Therapy Initiation for AKI: An Analysis of the Medical Information Mart for Intensive Care IV Data

Session Information

• Artificial Intelligence: Clinical Prediction, Decision Support, and Digital Kidney Care
  October 23, 2026 | Location: Exhibit Hall A, Convention Center
  Abstract Time: 10:00 AM - 12:00 PM

Category: Artificial Intelligence, Digital Health, and Data Science

• 300 Artificial Intelligence, Digital Health, and Data Science

Authors

• Mathew, Roy O., University of California Los Angeles, Los Angeles, California, United States

Roy O. Mathew, MD, FASN

• Zaychikov, Michael Edward, University of California Los Angeles, Los Angeles, California, United States

Michael Edward Zaychikov, MS

• Kostelny, Matthew, University of California Los Angeles, Los Angeles, California, United States

Matthew Kostelny

• Yin, Sopyda, University of California Los Angeles, Los Angeles, California, United States

Sopyda Yin

• Ojukwu, Nena A., University of California Los Angeles, Los Angeles, California, United States

Nena A. Ojukwu

• Nicholas, Susanne B., University of California Los Angeles David Geffen School of Medicine, Los Angeles, California, United States

Susanne B. Nicholas, MD, PhD, MPH

• Rhee, Connie, VA Greater Los Angeles Healthcare System, Los Angeles, California, United States

Connie Rhee, MD, MS

• Kalantar-Zadeh, Kamyar, The Lundquist Institute, Torrance, California, United States

Kamyar Kalantar-Zadeh, MD, DrMed, PhD, MPH, FASN

Background

Acute kidney injury (AKI) is a devastating complication in medical intensive care unit (ICU) admissions. The need for kidney replacement therapy (KRT) in severe AKI markedly increases mortality risk. Prediction markers for short term mortality following KRT initiation in severe AKI in medical ICU patients are not readily available.

Methods

Data were collected from MIMIC-IV. Patients were included if they had age between 18 and 80, were admitted to a medical intensive care unit (including cardiology and medical neurology), had an international Classification of Disease (ICD) code consistent with AKI, and underwent some form of KRT for AKI (hemodialysis or continuous renal replacement therapy). Patients were excluded from the cohort if they underwent KRT for end stage kidney disease, were discharged on Hospice or against medical advice. Classification modeling using logistic regression, random forest, and XGBoost were performed to predict death at 30 days following KRT initiation. Model performance was assessed using area under the receiver operating curve, accuracy, sensitivity, and specificity.

Results

A total of 2558 patients were included in the analysis. The highest AUC was demonstrated by the XGBoost model (0.874) followed by logistic regression (0.861) and random forest (0.847). The XGBoost model also had the highest accuracy (0.858) and sensitivity (0.496). Random Forest demonstrated the highest specificity at 0.984. All models demonstrated similar order of preference for variable importance but the most parsimonious model was suggested by XGBoost (Figure).

Conclusion

A predictive model for 30-day mortality following KRT initiation for severe AKI in medical ICU patients was developed with excellent discrimination and accuracy. Such a model can be used to inform patients and caregivers regarding the utility of starting aggressive KRT in the setting of severe AKI. Further studies are needed to validate this model in external cohorts.