Derivation and Validation of Pediatric Sepsis-Associated Acute Kidney Injury Subphenotypes With Prognostic Relevance

OBJECTIVES: Sepsis-associated acute kidney injury (SAKI) is a heterogeneous syndrome associated with poor outcomes. Subphenotypes of SAKI with prognostic and therapeutic relevance have been identified in adults, but not in children. We sought to identify reproducible and clinically relevant pediatric SAKI (pSAKI) subphenotypes using readily available clinical and laboratory data. DESIGN: Secondary analysis of a retrospective observational study of pediatric sepsis. SETTING: Thirteen PICUs in the United States from January 2012 to January 2018. PATIENTS: Patients aged 0-18 years with septic shock (sepsis and requiring vasoactive medications) and day 1-2 SAKI (≥ Kidney Disease Improving Global Outcomes stage 1 by serum creatinine). INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Fourteen hundred fifty-five patients were included after inclusion and exclusion criteria were applied: 873 (60%) in the derivation cohort and 582 (40%) in the external validation cohort. A two-subphenotype latent class analysis model had the best fit in both cohorts: pSAKI subphenotype 1 (pSAKI-1) and pSAKI subphenotype 2 (pSAKI-2). pSAKI-2 was characterized by younger age, more organ support, greater fluid accumulation, and laboratory evidence of inflammation, acid-base derangement, thrombocytopenia, and coagulopathy. pSAKI-2 had uniformly worse outcomes, including higher rates of severe and persistent AKI at days 3-4 (54% vs. 23%, p < 0.001) and day 7 (31% vs. 12%, p < 0.001), increased use of continuous renal replacement therapy (21% vs. 6%, p < 0.001), and independently increased odds of mortality after adjustment for potential confounders (adjusted odds ratio 1.59; 95% CI, 1.04-2.41; p = 0.03). A parsimonious classification model accurately identified pSAKI-2 membership (C-statistic 0.94 [95% CI, 0.92-0.95] and 0.85 [95% CI, 0.82-0.88], respectively, in the derivation and internal validation cohorts). CONCLUSIONS: We identified two distinct early pSAKI subphenotypes using readily available data that exhibit differential risk for poor outcomes and can be identified from a parsimonious set of variables. Pending external validation, operationalization of pSAKI subphenotypes may allow for prognostic enrichment to guide clinical care and inform clinical trial enrollment.