Toward the standardization of big datasets of urine output for AKI analysis: a multicenter validation study

Abstract

Abstract Acute kidney injury (AKI) is a prevalent condition in ICU patients. However, inconsistencies in urine charting and guideline interpretations hinder accurate diagnosis and research. This study aimed to derive and validate a standardization for the processing of big urine output datasets to improve consistency in AKI diagnosis and staging. Using a derivation cohort from 14 ICUs at Beth Israel Deaconess Medical Center (2008–2019) and a validation cohort from an academic center in Amsterdam (2003–2016), we developed and validated an algorithm for computing hourly urine output rates and identifying oliguric AKI across its definitions. Peak AKI stages computed using the method were significantly associated with all clinical outcomes, including severity scores, serum creatinine levels, ICU and hospital lengths of stay, renal replacement therapy requirements, and hospital mortality (all p < 0.001). Adjusted 30-day mortality odds ratios for AKI stages 1–3 were 1.58, 2.93, and 5.24 in the derivation cohort and 2.91, 5.16, and 13.59 in the validation cohort (all p < 0.001). Tested on over 85,000 multinational ICU admissions, this approach demonstrated robust performance and consistent results across diverse settings; it has the potential to enhance clinical practice through e-alerts and support future AKI and fluid balance research, including ML model training and inference. Supported by open-source code, the proposed method advances the standardization of AKI diagnostic criteria and can be applied to other EHR-based databases.