Abstract: FR-PO0392
Combining a Multimodal Deep Learning Risk Model with Biomarkers to Identify Patients at High Risk
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
- Koyner, Jay L., The University of Chicago Division of the Biological Sciences, Chicago, Illinois, United States
- Fatima, Aiman, The University of Chicago Division of the Biological Sciences, Chicago, Illinois, United States
- Anjorin, Ola, The University of Chicago Division of the Biological Sciences, Chicago, Illinois, United States
- Weber, Michael J., University of Wisconsin-Madison, Madison, Wisconsin, United States
- Oguss, Madeline K., University of Wisconsin-Madison, Madison, Wisconsin, United States
- Singh, Tripti, University of Wisconsin-Madison, Madison, Wisconsin, United States
- Carey, Kyle, The University of Chicago Division of the Biological Sciences, Chicago, Illinois, United States
- Churpek, Matthew M., University of Wisconsin-Madison, Madison, Wisconsin, United States
Background
AKI risk scores and urine and blood biomarkers can identify patients at high risk for AKI before changes in serum creatinine (SCr). Little is known about the ability of these biomarkers to predict adverse outcomes in a population already identified as high risk by a multimodal deep learning model
Methods
Across 2 academic centers, we identified hospitalized patients at high risk for severe AKI (≥KDIGO Stage 2) using a multimodal deep learning model (ESTOP2.0) with a score ≥0.091. Patients were enrolled within 8 hours of elevated risk, with blood and urine samples collected at enrollment (time 0) and every 12 hours for 36 hours. We measured urine albumin/creatinine, urine neutrophil gelatinase-associated lipocalin (uNGAL), and plasma NGAL (pNGAL). We excluded patients who had already developed ≥Stage 2 AKI at the time of the high ESTOP2.0 score, as well as those with ESKD and kidney transplants. We looked at the AUCs for these biomarkers to predict Stage 2 AKI, the receipt of RRT and inpatient mortality
Results
We enrolled 148 patients, 68(45%) of whom had SCr-based Stage 1 AKI at the time of consent, and 80(55%) with no AKI. The median SCr at enrollment was 1.30 (1.0-1.66) and 114(76%) were in the ICU. Twenty-nine patients (19%) developed SCr-based Stage ≥2 in the first 4 days, while 9(6%) received RRT and 29(19%) experienced inpatient mortality. The AUCs (95%CI) for each outcome and the performance of each biomarker at different time points are shown in the Table. Albuminuria had the highest discrimination for ≥Stage 2 AKI, and pNGAL had the highest discrimination for mortaltiy
Conclusion
In a cohort already enriched for severe AKI risk, albuminuria, uGNAL, and pNGAL measured within 36 hours of an elevated ESTOP 2.0 risk score reliably identified hospitalized patients at the highest risk for KDIGO stage ≥2 AKI, RRT, and inpatient mortality
Biomarker Perfromance In Predicting Patient Outcomes
| AUC(95%CI) | SCr based KDIGO AKI ≥ Stage 2 in 48 hours (n=29). | SCr and Urine Output based KDIGO AKI ≥ Stage 2 in the next 48 hours (n-41) | Recevied Dialysis (RRT) (n=9) | Inpatient Mortality (N=29) |
| Albuminuria (mg/g) at Enrollment (0 hour) | 0.55 (0.35 - 0.75) | 0.50 (0.38 - 0.63) | 0.57 (0.40 - 0.75) | 0.64 (0.53 - 0.75) ** |
| Albuminuria (mg/g) -12 hours | 0.69 (0.57 - 0.81) ** | 0.59 (0.31 - 0.89) | 0.61 (0.43 - 0.80) | 0.60 (0.47 - 0.73) |
| Albuminuria (mg/g) - 36 hours | 0.82 (0.73 - 0.91) ** | 0.80 (0.70 - 0.90) ** | 0.58 (0.42 -0.74) | 0.67 (0.53 - 0.80) ** |
| uNGAL at Enrollment (0 hour) | 0.60 (0.43 - 0.78) | 0.68 (0.54-0.81) ** | 0.58 (0.36 - 0.80) | 0.61 (0.49 - 0.73) |
| uNGAL - 12 hours | 0.50 (0.23 - 0.78) | 0.50 (0.29 - 0.72) | 0.75 (0.59 - 0.90) ** | 0.56 (0.41 -0.70) |
| uNGAL - 36 hours | 0.72 (0.21 - 0.99) | 0.45 (0.33 - 0.57) | 0.60 (0.30 -0.90) | 0.57 (0.39 -0.74) |
| pNGAL at Enrollment (0 hour) | 0.57 (0.41 - 0.73) | 0.60 (0.45 - 0.76) | 0.66 (0.27 - 0.90) | 0.69 (0.55 - 0.83) ** |
| pNGAL - 12 hours | 0.48 (0.31 - 0.64) | 0.50 (0.32 - 0.68) | 0.50 (0.34 - 0.67) | 0.68 (0.55 - 0.83) ** |
| pNGAL - 36 hours | 0.64 (0.10 - 0.90) | 0.27 (0.07 -0.56) | 0.54 (0.34 - 0.74) | 0.70 (0.54 - 0.85) ** |
** p < 0.05
Funding
- NIDDK Support