Alcohol use disorder and persistent intensive care unit delirium among critically ill adults: a retrospective cohort study

Alcohol-related ICU studies have emphasized delirium occurrence, although course after onset also matters. We examined whether refined index-hospitalization alcohol use disorder (AUD) coding was associated with a more persistent documented delirium course among adults with ICU delirium. In this retrospective cohort study, we analyzed adults in Medical Information Mart for Intensive Care IV (MIMIC-IV) version 3.1 who had at least one valid Confusion Assessment Method for the Intensive Care Unit (CAM-ICU)-positive assessment. Refined AUD coding required an index-hospitalization abuse/dependence-family code without explicit alcohol-related delirium or psychosis wording; reference patients had no broad alcohol-related code. Persistence required a valid CAM-ICU-positive assessment on at least three consecutive calendar days. Logistic regression estimated the full-cohort covariate-conditional association; overlap weighting estimated marginal contrasts in the empirical overlap population. A multicenter analysis in the eICU Collaborative Research Database (eICU-CRD), version 2.0, separately reconstructed the same endpoint from strictly paired CAM-ICU records. The primary MIMIC-IV cohort included 14,089 patients (1,971 with refined AUD coding; 12,118 reference patients). Persistent delirium occurred in 732/1,971 (37.1%) and 3,492/12,118 (28.8%), respectively, with an adjusted OR of 1.26 (95% CI, 1.13–1.41). In the empirical overlap population, weighted probabilities were 37.05% and 31.89%, yielding an absolute probability difference of 5.16% (bootstrap 95% CI, 2.78%-7.49%). An endpoint-aligned competing-event model yielded a subdistribution hazard ratio of 1.19 (95% CI, 1.10–1.30). In the endpoint-harmonized eICU-CRD cohort of 3,604 patients across 45 hospitals, the age- and sex-adjusted mixed-effects OR was 1.51 (95% CI, 0.98–2.31). Among critically ill adults with documented ICU delirium, AUD ascertained by refined index-hospitalization coding was associated with a more persistent documented delirium course in MIMIC-IV. The separately reconstructed eICU-CRD analysis yielded a directionally aligned but imprecise estimate in a narrower withdrawal-coded subset. This course-focused evidence complements delirium-incidence research and supports prospective evaluation of alcohol-informed assessment and monitoring pathways.