Joint Modeling of Clinical and Biomarker Data in Acute Kidney Injury Defines Unique Subphenotypes with Differing Outcomes

Background AKI is a heterogeneous syndrome. Current subphenotyping approaches have only used limited laboratory data to understand a much more complex condition. Methods We focused on patients with AKI from the Assessment, Serial Evaluation, and Subsequent Sequelae in AKI (ASSESS-AKI). We used hierarchical clustering with Ward linkage on biomarkers of inflammation, injury, and repair/health. We then evaluated clinical differences between subphenotypes and examined their associations with cardiorenal events and death using Cox proportional hazard models. Results We included 748 patients with AKI: 543 (73%) of them had AKI stage 1, 112 (15%) had AKI stage 2, and 93 (12%) had AKI stage 3. The mean age (+/- SD) was 64 (13) years; 508 (68%) were men; and the median follow- up was 4.7 (Q1: 2.9, Q3: 5.7) years. Patients with AKI subphenotype 1 (N=181) had the highest kidney injury molecule (KIM-1) and troponin T levels. Subphenotype 2 (N=250) had the highest levels of uromodulin. AKI subphenotype 3 (N=159) comprised patients with markedly high pro-brain natriuretic peptide and plasma tumor necrosis factor receptor-1 and -2 and low concentrations of KIM-1 and neutrophil gelatinase- associated lipocalin. Finally, patients with subphenotype 4 (N=158) predominantly had sepsisAKI and the highest levels of vascular/kidney inflammation (YKL-40, MCP-1) and injury (neutrophil gelatinase- associated lipocalin, KIM-1). AKI subphenotypes 3 and 4 were independently associated with a higher risk of death compared with subphenotype 2 and had adjusted hazard ratios of 2.9 (95% confidence interval, 1.8 to 4.6) and 1.6 ( 95% confidence interval, 1.01 to 2.6, P = 0.04), respectively. Subphenotype 3 was also independently associated with a three-fold risk of CKD and cardiovascular events. Conclusions We discovered four AKI subphenotypes with differing clinical features and biomarker profiles that are associated with longitudinal clinical outcomes.