Abstract: TH-PO0053
International Delphi Consensus on AKI: Foundations for Artificial Intelligence (AI)-Driven Digital Twin Development in Critical Care Nephrology
Session Information
- AKI: Pathogenesis and Disease Mechanisms
November 06, 2025 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 102 AKI: Clinical, Outcomes, and Trials
Authors
- Kashani, Mehdi, Mayo Clinic Research Rochester, Rochester, Minnesota, United States
- Ninan, Jacob, MultiCare Capital Medical Center, Olympia, Washington, United States
- Wei, Lifang, Mayo Clinic Research Rochester, Rochester, Minnesota, United States
- Gajic, Ognjen, Mayo Clinic Research Rochester, Rochester, Minnesota, United States
- Cheungpasitporn, Wisit, Mayo Clinic Research Rochester, Rochester, Minnesota, United States
- Kashani, Kianoush, Mayo Clinic Research Rochester, Rochester, Minnesota, United States
Background
Acute Kidney Injury (AKI) is a complex and variable condition in critically ill patients. To support AI-driven digital twin applications in precision nephrology, an internationally validated expert consensus on AKI pathophysiology and management is needed.
Methods
We conducted a modified Delphi study to establish expert consensus on key aspects of AKI. Although three rounds were planned, consensus was reached after two. A total of 49 global nephrology and critical care experts, identified via literature and networks, reviewed 94 statements related to AKI pathophysiology, biomarkers, and ICU management. Quantitative thresholds and qualitative feedback informed the final results. Expert-endorsed statements were synthesized into Directed Acyclic Graphs (DAGs) to model AKI progression and inform digital twin development.
Results
Consensus was reached on 82% of the 94 statements. Hemodynamic instability and nephrotoxicity were most frequently identified as leading AKI causes in ICU patients. The resulting DAGs modeled multifactorial AKI pathways, integrating clinical risks, interventions, and physiologic variables. These structures support dynamic, data-driven simulations for patient-specific AKI prediction and decision support.
Conclusion
This study provides a foundational expert consensus to guide digital twin modeling for AKI. By translating clinical expertise into structured causal frameworks, we support the development of interpretable, AI-based tools for real-time diagnosis and personalized AKI management in critical care settings.