Abstract
Rationale: Biological heterogeneity in critical illness syndromes, such as sepsis and acute kidney injury (AKI), has hindered development of effective therapies. In sepsis-associated AKI, two molecular subphenotypes (SP1 and SP2) have been identified with differing characteristics, outcomes, and response to vasopressor treatment. It is unknown if these subphenotypes extend to all patients with sepsis and whether they respond differently to fluid resuscitation strategy. Methods: Patients enrolled in the CLOVERS clinical trial with plasma collected at study enrollment were classified into two subphenotypes using a previously validated parsimonious prediction model that included angiopoietin-1 and -2 and soluble tumor necrosis factor receptor-1. Kaplan-Meier estimates were used to test differences in outcomes and subphenotype by treatment interaction. Measurements and Main Results: Among 1,289 patients, we identified 1,016 as SP1 and 273 as SP2. The risk of poor clinical outcomes was greater in SP2 than in SP1 independent of demographics, comorbidities, and illness severity scores. Patients with SP2, characterized by more severe endothelial injury and inflammation, had higher 28-day mortality with a liberal versus restrictive fluid strategy (41% vs. 27%), whereas patients with SP1 had no difference (9% vs. 9%) (P value for interaction = 0.02). Furthermore, SP2 had fewer days free from ventilation, renal replacement therapy, and vasopressors with a liberal compared with a restrictive resuscitation strategy. Conclusions: Molecular subphenotypes previously identified in AKI are also identifiable in sepsis and respond differently to fluid resuscitation strategy. Future prospective identification of these subphenotypes could inform a precision-guided therapeutic approach in sepsis.