Abstract: TH-PO1186
Randomized Trial of an Early, Standardized Nephrology Consult Triggered by a Machine Learning Model
Session Information
- Late-Breaking Research Posters
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
- 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
- Churpek, Matthew M., University of Wisconsin-Madison, Madison, Wisconsin, United States
Group or Team Name
- ESTOP Investigators.
Background
Early detection of impending AKI may improve outcomes. We previously developed a machine learning model to detect AKI (ESTOP). We aimed to determine if earlier detection using our model combined with structured early nephrology consultation (ENC) improves outcomes
Methods
We conducted a single-center randomized controlled trial in hospitalized patients without serum creatinine-based (SCr) AKI and an elevated ESTOP score(NCT03590028). Patients were randomized to receive a structured ENC from a nephrology attending within 8 hours of their elevated score or usual care (UC). Consults included recommendations about volume status, renal perfusion, medications, electrolytes, and nutrition. Those in the UC arm only received a nephrology consult when requested. The primary outcome was peak change in SCr during the first 7 days (ΔSCr). Secondary outcomes included development of AKI, need for renal replacement therapy (RRT), length of stay (LOS), and inpatient and 90-day mortality
Results
We randomized 180 patients, 86 of whom received an ENC. Median(IQR) enrollment SCr was 1.04(0.75-1.25) mg/dl (ENC) and 1.0(0.72-1.31) (UC). There was no significant difference in the ΔSCr over the first 7 days (0.08 mg/dl (ENC) vs. 0.07 (UC)), and no difference in the development of AKI across the 2 arms, with 67(38%) subjects developing ≥ Stage 1 AKI (39.5 vs 36.5%;p=0.77). Stage ≥2 AKI developed in 15(17%) of the ENC and 12(13%) in the UC (p=0.56). There was also no difference in median LOS (9 vs 9,p=0.8), receipt of RRT (2 vs 3%,p=0.73), or inpatient mortality (9.3 vs 6.6%,p=0.7). There were 114 (ENC) and 18 (UC) consults during the first 7 days, containing 258 and 31 recommendations. Medication dosage and discontinuation, diuretics/fluids, and vasopressor recommendations were more likely to be completely followed in UC (75%) compared to ENC subjects (40%) (p=0.025). While 69(39%) subjects were re-admitted over the 90-day follow-up, there was no difference in readmission rates, recurrent AKI, major adverse cardiac event rates, or 90-day mortality (15.1 vs 18.7%,p=0.49)
Conclusion
Structured ENC triggered by a machine learning AKI risk score is feasible, but did not decrease AKI rates. Consult recommendations were less likely to be followed for patients in the ENC arm. Whether increasing compliance with recommendations could improve outcomes deserves further study
Funding
- NIDDK Support