Acute Kidney Injury in Critically Ill Patients Admitted to a Regional Tertiary ICU in South-West Romania: Prevalence, Clinical Phenotypes, and Outcomes During a High-Burden Period

Abstract

Although KDIGO criteria stratify acute kidney injury (AKI) severity, they treat AKI as a single biological entity and may obscure clinically meaningful heterogeneity in critically ill patients. We retrospectively analysed 389 adults admitted to a tertiary regional ICU in south-west Romania during a high-burden COVID-19 phase. AKI was classified per the 2012 KDIGO criteria using measured baseline creatinine (n = 292) or MDRD back-estimated values (n = 97). Within the AKI subset, k-means clustering was applied to twelve admission variables, with cluster number determined by silhouette, Calinski–Harabasz, Davies–Bouldin, gap statistic, and bootstrap Jaccard stability. AKI developed in 251 of 389 patients (64.5%); KDIGO stages 1, 2, and 3 represented 16.5%, 17.7%, and 30.3%, respectively. Mortality rose steeply with stage (26.8%, 57.8%, 92.8%, 100.0%; p < 0.001), and KDIGO remained an independent predictor after adjustment for age, sex, Charlson index, APACHE II, lactate, and oxygenation (Cox HR 1.25 per stage, 95% CI 1.08–1.43). Four reproducible phenotypes emerged (bootstrap Jaccard 0.81): P1 hyper-inflammatory shock (100% mortality), P2 inflammatory-respiratory (100%), P3 cardi-orenal/AKI-on-CKD (93.9%, RRT 38.8%), and P4 mild/recoverable (54.7%). P3 sponta-neously captured 75% of all AKI-on-CKD patients (45 of 60), a 5.4-fold enrichment over the cohort CKD prevalence of 17%, and was characterised by severe structural renal injury yet paradoxically attenuated inflammatory signalling. These findings support AKI-on-CKD as a mechanistically distinct entity and provide a reproducible bedside taxonomy with prognostic value beyond KDIGO staging.