U-shaped association between left ventricular ejection fraction and short-term mortality in adults with sepsis: a retrospective cohort study using MIMIC-IV and MIMIC-IV-ECHO

Background Cardiac dysfunction is common in sepsis, but the prognostic meaning of the left ventricular ejection fraction (LVEF) is uncertain and may be non-linear. We used newly released open echocardiographic data to characterise that shape. Methods Retrospective cohort study of adults meeting Sepsis-3 criteria in the linked MIMIC-IV and MIMIC-IV-ECHO databases (single centre, 2008–2022), restricted to intensive care unit (ICU) admissions with a resting transthoracic echocardiogram reporting an LVEF within 72 h of ICU admission. Follow-up for the 28-day and 90-day outcomes began at the echocardiogram, avoiding immortal-time bias. LVEF was modelled as a restricted cubic spline and in six categories (reference 50–59%). Outcomes were 28-day (primary), 90-day and in-hospital mortality. Because several acute covariates are measured after the echocardiogram and may be mediators, estimates come from three nested models. Blood pressure, heart rate, vasopressor dose, fluid balance and echocardiographic geometry characterised each stratum, and the incremental value of LVEF was quantified. Results Among 8792 admissions (8079 patients), 28-day mortality was 26.1%. The adjusted association was U-shaped ( P < 0.001 for non-linearity): relative to 50–59%, both a depressed LVEF (< 30%; odds ratio 1.41, 95% CI 1.18–1.68) and a hyperdynamic LVEF (≥ 70%; 1.41, 1.20–1.66) independently marked death, consistently at 90 days and in hospital. Demography and documented comorbidity attenuated the depressed limb (crude 1.97 to 1.80) but left the hyperdynamic limb unchanged (1.61). Against that reference stratum, the hyperdynamic stratum had the lowest estimated systemic vascular resistance (824 versus 1003 dyn·s·cm⁻⁵), the highest cardiac index, and the smallest end-diastolic dimension (4.10 versus 4.40 cm); filling pressures, where recorded, were not elevated. The non-linear relationship persisted in all eleven sensitivity analyses (six prespecified); adding LVEF raised the C-statistic only from 0.719 to 0.721. Conclusions Both depressed and hyperdynamic LVEF independently mark higher short-term mortality, so a normal-to-high ejection fraction is not reassuring; the hyperdynamic limb is accompanied by a vasodilated, small-cavity, high-output state. Because LVEF adds almost nothing to established clinical variables, and because the cohort—selected for early echocardiography—carried more cardiac comorbidity and higher mortality than the wider sepsis population, these findings are hypothesis-generating rather than a basis for risk stratification. Clinical trial number Not applicable.