Perioperative RRT/CRRT and mechanical ventilation support intensity and 365 day mortality after liver transplantation in MIMIC-IV v3.0: an EHR-based cohort study

Abstract Renal replacement therapy/continuous renal replacement therapy (RRT/CRRT) and mechanical ventilation are central intensive care unit (ICU) support processes before and after liver transplantation (LT). In electronic health record (EHR) data, these variables are difficult to interpret because they reflect both treatment decisions and evolving perioperative severity. We aimed to characterize perioperative ICU support intensity in MIMIC-IV v3.0 and explore its association with 365 day all-cause mortality after LT. We conducted a retrospective EHR-based cohort study of adult LT recipients with peri-transplant ICU exposure in MIMIC-IV v3.0. The primary endpoint was 365 day all-cause mortality after LT. The prespecified model included age, Model for End-Stage Liver Disease (MELD) score, and pre-LT RRT/CRRT. Sensitivity and exploratory models evaluated pre-LT mechanical ventilation, post-LT RRT/CRRT, post-LT mechanical ventilation, recorded support duration, vasopressor use, postoperative infection, a sepsis-3 timing proxy, and candidate mediation-style pathways. Among 324 unique ICU-exposed LT recipients, 20 (6.2%) died within 365 days after LT and 13 (4.0%) died in hospital. In the prespecified model, MELD score was associated with 365 day mortality (OR 1.97 per 5 points, 95% CI 1.30–2.98), whereas pre-LT RRT/CRRT was not (OR 1.06, 95% CI 0.29–3.80). Pre-LT mechanical ventilation showed an unstable borderline association in primary sensitivity analysis (OR 4.22, 95% CI 0.95–18.77), which attenuated in duplicate-aggregation sensitivity analysis. Post-LT RRT/CRRT and post-LT mechanical ventilation were not independently associated with mortality in adjusted exploratory models. Recorded ventilation time showed an exploratory association, but recorded duration variables were treated as treatment-context markers because they are vulnerable to indication, practice, and survival-time bias. Sepsis-3, postoperative infection, and positive sputum/BAL culture did not show compatible exploratory mediation patterns through RRT/CRRT or mechanical ventilation. In this EHR-based LT-ICU cohort, MELD score was the most stable correlate of 365 day mortality. RRT/CRRT, mechanical ventilation, and recorded support duration before or after LT should be interpreted as indication-dependent ICU treatment-context markers rather than causal treatment-effect estimates. The available data did not support infection/sepsis mediation pathways through RRT/CRRT or mechanical ventilation.