Abstract: FR-PO0386
Three-Metabolite Admission Signature Predicts Severe AKI in Patients in the Intensive Care Unit (ICU): Beyond Sequential Organ Failure Assessment (SOFA) and Neutrophil Gelatinase-Associated Lipocalin (NGAL)
Session Information
- AKI: Biomarkers, Diagnostics, and Risk Prediction
October 23, 2026 | Location: Exhibit Hall A, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Acute Kidney Injury
- 102 AKI: Clinical, Outcomes, and Trials
Authors
- Zuñiga Gonzalez, Erick Yasar, Instituto Nacional de Ciencias Medicas y Nutricion Salvador Zubiran, Mexico City, CDMX, Mexico
- Mercado Hernández, Yazmin Alejandra, Instituto Nacional de Ciencias Medicas y Nutricion Salvador Zubiran, Mexico City, CDMX, Mexico
- Del Toro-Cisneros, Noemi, Instituto Nacional de Ciencias Medicas y Nutricion Salvador Zubiran, Mexico City, CDMX, Mexico
- Rincon-Pedrero, Rodolfo, Instituto Nacional de Ciencias Medicas y Nutricion Salvador Zubiran, Mexico City, CDMX, Mexico
- Vega, Olynka, Instituto Nacional de Ciencias Medicas y Nutricion Salvador Zubiran, Mexico City, CDMX, Mexico
Background
Severe acute kidney injury (AKI) in the ICU is frequently recognized after functional decline. We hypothesized that a metabolic signature could predict creatinine-defined severe AKI within 72 hours and add discriminatory value beyond established clinical and biomarker models.
Methods
We analyzed 124 critically ill patients without AKI at ICU addmision. Untargeted serum metabolomics and urinary biomarkers were measured at admission. Primary endpoint was severe AKI (SCr-KDIGO stage 2-3) within 72 hours. Metabolomics was assessed by LC/MS using PCA, volcano plots and LASSO with cross-validation for metabolite selection. A clinically interpretable top-3 model was compared with SOFA, TIMP2/IGFBP7, and NGAL. Incremental value was assessed by bootstrap/DeLong AUC comparison, and metabolite–biomarker associations by Spearman correlation.
Results
Severe AKI occurred in 18 patients (14.5%). Discriminatory feature analyses identified gluconic acid, cysteine, 3,4-dihydroxybutanoic acid, and leucine, while LASSO with cross-validation selected an 18-metabolite panel and prioritized a three-metabolite signature: 3,4-dihydroxybutanoic acid, cysteine, and 3-hydroxybutyric acid (Figure 1A). The Top-3 model achieved an AUC of 0.73, outperforming NGAL and SOFA. At a 0.427 threshold, sensitivity was 0.89, specificity 0.64, and NPV 0.97. Adding the signature improved discrimination for SOFA (AUC 0.69→0.83; p=0.0036) and NGAL (0.62→0.80; p=0.0106), but not TIMP2/IGFBP7 (0.76→0.80; p=0.66) (Figure 1C,D). Selected metabolites showed modest correlations with urine biomarkers and SOFA (Figure 1B).
Conclusion
A three-metabolite admission signature predicted severe AKI within 72h (NPV 0.97) and improved discrimination beyond SOFA and NGAL, supporting its use as an early rule-out tool and rationale for external validation in larger ICU cohorts.
Funding
- Other NIH Support