Clinical Indicators Associated with Heterogeneity in the Association Between Early Antibiotic Administration and Mortality in Critically Ill Patients with Sepsis and Cirrhosis: A Multicenter Retrospective Cohort Study

Background: Early antibiotic administration is a cornerstone of sepsis management, but its survival benefit in septic patients with pre-existing cirrhosis remains unclear. This study evaluated the association between antibiotic timing and short-term mortality, and explored clinical indicators that modify this therapeutic effect. Methods: This multicenter retrospective cohort study used MIMIC-IV as the primary cohort (n = 1660) and subsequently evaluated consistency in MIMIC-III (n = 1776), Chengdu First People’s Hospital (n = 176), and NWICU (n = 88). Adult patients with sepsis and cirrhosis who reached the ICU without documented prior antibiotic administration were included (excluding those with pre-ICU antibiotics or ICU stay < 24 h). Antibiotic timing was dichotomized as early (≤2 h) vs. delayed (>2 h) from ICU admission. Propensity score matching balanced confounders. The primary outcome was 30-day mortality. Results: In the matched primary cohort, early antibiotics showed no overall survival benefit. However, machine learning prioritized INR, PT, and MELD-Na as potential modifiers. In exploratory analyses, lower adjusted odds of mortality were observed within specific data-derived exploratory ranges: INR 0.9–2.3 (adjusted risk difference [ARD] −11.62%, 95% CI −19.11% to −4.13%), PT 9.4–25.0 s (ARD −12.21%, 95% CI −19.70% to −4.73%), and MELD-Na 6.0–28.3 (ARD −10.88%, 95% CI −19.26% to −2.49%). A composite score (0–100) was derived; lower adjusted odds were observed in patients with score < 23.7 (OR 0.51, 0.300–0.820), whereas no association was observed in those with higher scores. Directional consistency for the exploratory composite score threshold was observed in MIMIC-III and the CD-FH external cohort, but not in the smaller NWICU cohort, underscoring the exploratory nature of these external assessments. Conclusions: Early antibiotic administration was not associated with improved short-term survival in the overall cohort of septic patients with cirrhosis, though exploratory analyses suggest that the association of timing may vary with underlying liver disease severity and coagulation abnormalities. Importantly, our findings are strictly confined to patients surviving the initial hyperacute ICU resuscitation phase (≥24 h). Given the observational design, these results remain hypothesis-generating and warrant prospective validation prior to clinical translation.