Troponin-echocardiography prognostic discordance in selected ICU patients with sepsis: A MIMIC-IV-ECHO cohort study

In selected ICU patients with sepsis who undergo both troponin testing and echocardiography, the two tests may capture different domains of cardiovascular injury. We examined whether concordance and discordance between pre-index troponin elevation and biventricular systolic dysfunction identify distinct mortality trajectories. We conducted a retrospective cohort study linking MIMIC-IV v3.1 and MIMIC-IV-ECHO v1.0. Adults meeting Sepsis-3 criteria with a first transthoracic echocardiogram (TTE) 0–48 h after ICU admission and troponin T measured during the preceding 24 h through the TTE were eligible. The primary cohort required evaluable left-ventricular ejection fraction (LVEF) and qualitative right-ventricular (RV) function. Troponin-high was >5 times the assay upper reference limit (ULN); echo-positive was LVEF < 50% or structured qualitative RV systolic dysfunction. Kaplan–Meier curves, subject-clustered Cox models using covariates available at or before TTE, restricted cubic splines, exploratory subgroup interactions, and sensitivity analyses evaluated 28- and 90-day mortality from the index TTE. The primary cohort included 2,523 admissions from 2,445 patients: 538 troponin-low/echo-negative, 681 troponin-high/echo-negative, 345 troponin-low/echo-positive, and 959 double-positive. Ninety-day mortality was 36.8%, 40.7%, 41.2%, and 51.7%, respectively (overall log-rank P < 0.001). Relative to double-negative status, adjusted 90-day hazard ratios were 1.18 (95% CI 0.98–1.42) for troponin-high/echo-negative, 1.33 (1.06–1.66) for troponin-low/echo-positive, and 1.68 (1.41–2.01) for double-positive. Continuous pre-index troponin was associated with mortality (overall P < 0.001), without evidence of nonlinearity (P = 0.302) or troponin-by-echo interaction (P = 0.136). The double-positive association persisted across timing, threshold, repeated-admission, missing-data, chamber-evaluability, and renal-function analyses. No exploratory subgroup interaction remained significant after false-discovery-rate correction. Among selected ICU patients with sepsis who underwent both tests, pre-index troponin elevation and structured LV/RV systolic dysfunction provided complementary risk information. Their co-occurrence identified the highest-risk phenotype; isolated echo positivity was also associated with higher 90-day mortality, whereas the isolated troponin association was less precise. This framework supports prognostic stratification, not phenotype-directed treatment or deployment as a prediction model.