Fluid Resuscitation Volume After Early Echocardiography and Mortality in Sepsis: Target Trial Emulations With Strictly Pre-Exposure Confounder Measurement in MIMIC-IV

Abstract Background Whether echocardiographic phenotypes modify the association between early fluid resuscitation volume and mortality in sepsis is unknown. We emulated three target trials within a single cohort of Sepsis-3 patients with early transthoracic echocardiography (TTE), measuring every confounder strictly before the start of the exposure window. Methods Using MIMIC-IV v3.1 (2008–2022), we identified Sepsis-3 stays with TTE within ± 24 h of ICU admission (first stay per patient; n = 7,819). Emulation 1 (primary) compared post-TTE 24-h fluid volume ≥ 30 vs < 30 mL/kg with a landmark at TTE + 24 h (n = 7,398). Emulation 2 (secondary) compared dobutamine new-users versus non-users within the LVEF < 40% stratum (n = 1,201; 100 new-users). Emulation 3 (secondary) compared vasopressin new-users versus non-users (n = 6,474; 360 new-users). All adjustment covariates—including SOFA components, vasopressor exposure with norepinephrine dose, Sepsis-3 septic shock status, fluid balance and urine output—were measured strictly before the echocardiogram. Overlap weighting was used throughout; propensity scores were re-estimated within phenotype strata. The outcome was 28-day all-cause mortality from each landmark. Effect modification was assessed with formal interaction tests on multiplicative and additive scales. Bias analyses included E-values with hazard-to-risk-ratio conversion, tipping-point grids, and negative control outcomes restricted to diagnoses from prior admissions. Results Post-echocardiography fluid volume ≥ 30 mL/kg was associated with higher 28-day mortality (HR 1.74, 95% CI 1.54–1.97; weighted risk 38.7% vs 24.8%; risk difference + 13.9%, 95% CI 10.8–17.0). The association was similar across LVEF strata (< 40%: HR 1.63, 1.26–2.10, n = 1,361; 40–54%: 2.02, 1.47–2.79, n = 1,180; 55–69%: 1.88, 1.51–2.34, n = 2,752; ≥70% hyperdynamic: 1.39, 1.02–1.89, n = 896) and across RV strata (normal: 1.67, 1.42–1.97, n = 4,814; dysfunction: 1.77, 1.43–2.18, n = 1,979). No formal interaction test supported effect modification (LVEF: multiplicative p = 0.44, additive p = 0.56; RV: multiplicative p = 0.65, additive p = 0.16; LVEF modelled continuously: p = 0.82). A previously observed protective association in hyperdynamic hearts without cardiac disease was not confirmed when cardiac disease was ascertained from pre-admission diagnoses (HR 1.33, 0.90–1.97). The association persisted within pre-TTE SOFA tertiles and septic-shock strata, at 20 and 40 mL/kg thresholds, per 10 mL/kg as a continuous exposure (HR 1.14, 1.11–1.17), with a 6-h exposure window (HR 1.43, 1.27–1.60), and after excluding pre-ICU echocardiograms (HR 1.64, 1.44–1.87). Dobutamine initiation (Emulation 2) was imprecisely estimated (HR 1.57, 0.95–2.60); vasopressin initiation (Emulation 3) was strongly associated with mortality (HR 2.55, 2.13–3.06) without phenotype interaction (multiplicative p = 0.23). The E-value was 2.60 (2.29 for the confidence bound); prior-admission negative control outcomes indicated residual bias acting, if anything, toward the null. Conclusions After strict temporal alignment of confounder measurement, exposure and follow-up, higher fluid volume in the 24 h after early echocardiography was consistently associated with higher 28-day mortality in sepsis, but we found no statistically supported evidence that LVEF or RV phenotype modifies this association. These results do not support phenotype-guided fluid restriction derived from observational subgroup patterns; adequately powered randomised trials with protocolised echocardiography remain necessary.