Unsupervised clustering of hemodynamic and echocardiography parameters identified four cardiovascular subphenotypes in sepsis with distinct 90-day mortality rates ranging from 18% to 58% (p<0.0001).
Cohort (n=1,799)
No
Unsupervised clustering of echocardiographic and hemodynamic parameters identifies four distinct cardiovascular subphenotypes in sepsis with significantly different 90-day mortality risks.
p-value: p=<0.0001
Objectives: To apply unsupervised clustering methods to hemodynamic and transthoracic echocardiography (TTE) parameters to identify cardiovascular subphenotypes in ICU patients with sepsis. To examine subphenotype association with mortality and determine how differing hemodynamic management strategies influence these associations. Design: Retrospective, single-center cohort study. Setting: University Hospital ICU, Birmingham, United Kingdom. Patients: ICU patients that received TTE within 7 days of sepsis onset between April 2016 and December 2019 (derivation cohort) and January 2020 and December 2021 (validation cohort). Interventions: None. Measurements and Main Results: Nine hundred ninety-five patients were included in the derivation cohort and 804 patients in the validation cohort. A four-class model best fit both cohorts: class 1 (51% in derivation cohort, 56% in validation cohort; mostly normal left ventricular LV and right ventricular RV function), class 2 (30% in derivation cohort, 22% in validation cohort; mostly high cardiac index, hyperdynamic LV ejection fraction), class 3 (10% in derivation cohort, 12% in validation cohort, mostly dilated RV with impaired systolic function), and class 4 (9% in derivation cohort, 10% in validation cohort; mostly low cardiac output, with depressed LV ejection fraction). The four subphenotypes differed in their characteristics and outcomes, with 90-day mortality rates of classes 1–4 of 20%, 46%, 47%, and 41% in the derivation cohort and 18%, 45%, 57%, and 58% in the validation cohort, respectively ( p < 0.0001 for both cohorts). Following multivariable logistic regression analysis, classes 2–4 were independently associated with mortality. Three-variable models had high diagnostic accuracy in identifying all subphenotypes in both cohorts. The association with mortality of classes varied according to differing vasoactive agent and fluid administration strategies. Conclusions: Clustering analysis identified four cardiovascular subphenotypes in sepsis that reflected distinct circulatory failure mechanisms, were identifiable using simple models, and were associated with differing mortality risks and response to hemodynamic therapies. These classes may represent treatable traits to personalize shock management in sepsis.
Chotalia et al. (Mon,) conducted a cohort in Sepsis (n=1,799). Unsupervised clustering of hemodynamic and TTE parameters was evaluated on 90-day mortality (p=<0.0001). Unsupervised clustering of hemodynamic and echocardiography parameters identified four cardiovascular subphenotypes in sepsis with distinct 90-day mortality rates ranging from 18% to 58% (p<0.0001).