External validation and comparative performance of four risk scores for acute kidney injury in critically ill adults

Background Several parsimonious scores predict acute kidney injury (AKI), but their comparative performance in critically ill adults is uncertain. We hypothesized that the ICU-derived UCSD-Mayo score would show higher discrimination than three scores developed in broader acute-care populations.Methods This single-center retrospective validation study included 1,723 emergency ICU patients admitted from 2014 to 2021. We compared the UCSD-Mayo, AKI-RiSc, Bell, and APS scores for in-hospital AKI, AKI within 7 days, and stages 2 and 3 AKI. Analyses included AUROC with paired DeLong comparisons, cross-validated local recalibration, calibration plots, IDI, categorical NRI at the event rate, decision-curve analysis, and prespecified sensitivity analyses.Results In-hospital AKI occurred in 406 patients (23.6%), 281 (16.3%) developed AKI within 7 days, and 169 (9.8%) developed stages 2 and 3 AKI. UCSD-Mayo had the highest observed AUROC for in-hospital AKI (0.713, 95%CI 0.685–0.740), 7-day AKI (0.690, 0.658–0.723), and stages 2 and 3 AKI (0.706, 0.665–0.746). Its 7-day AUROC did not differ significantly from AKI-RiSc (difference 0.029, 95%CI −0.006 to 0.064; p = 0.103). Reclassification and net benefit varied by outcome and threshold; sensitivity analyses did not materially change the overall pattern.Conclusions In this single-center Chinese emergency ICU cohort, UCSD-Mayo showed the highest observed discrimination, but improvements were modest and were not uniform across all outcomes, metrics, or decision thresholds. These scores may support risk awareness and monitoring but should not be used as stand-alone treatment rules.