The researchers harnessed MIMIC-IV, version 3.1, a widely used, de-identified critical care database drawn from the electronic health records of Beth Israel Deaconess Medical Center in Boston. From this resource they identified 14,089 adults who had at least one positive score on the Confusion Assessment Method for the Intensive Care Unit, or CAM-ICU, the bedside screening instrument that has become the de facto standard for detecting delirium in sedated and mechanically ventilated patients. The CAM-ICU works by testing attention, thought disorganization, and level of consciousness, and a positive assessment signals acute brain dysfunction rather than the sedation or dementia that can mimic it.

Exposure to alcohol was defined with deliberate caution. Rather than simply flagging any alcohol-related mention in the record, the investigators required a refined coding scheme: patients in the alcohol group carried an index-hospitalization diagnosis code from the abuse or dependence family, but without any explicit wording suggesting alcohol-related delirium or psychosis. This restriction matters because including patients coded with alcohol-induced delirium would blur the line between the exposure and the outcome, effectively counting delirium on both sides of the equation. The reference group comprised patients with no broad alcohol-related code at all. The result was 1,971 patients with refined alcohol use disorder coding and 12,118 reference patients, all of whom had documented delirium.

The primary outcome was persistence, defined as a valid CAM-ICU-positive assessment on at least three consecutive calendar days. That threshold converts a snapshot finding into a documented course of brain dysfunction, capturing the burden of delirium that clinicians and families experience day after day. The contrast between the two groups was striking. Persistent delirium occurred in 732 of the 1,971 patients with alcohol use disorder coding, or 37.1 percent, compared with 3,492 of the 12,118 reference patients, or 28.8 percent. In a fully covariate-adjusted logistic regression model spanning the entire cohort, this translated into an adjusted odds ratio of 1.26 with a 95 percent confidence interval of 1.13 to 1.41, a statistically robust elevation in the odds of a persistent course.

Because conventional regression on skewed populations can overstate contrasts, the team also deployed overlap weighting, a modern causal inference technique that reweights the sample toward patients whose baseline characteristics are shared by both groups. In this empirical overlap population, the weighted probability of persistent delirium was 37.05 percent in the alcohol group and 31.89 percent in the reference group, an absolute difference of 5.16 percentage points with a bootstrap 95 percent confidence interval of 2.78 to 7.49 percent. Put plainly, among comparable patients, the presence of an alcohol use disorder diagnosis was associated with roughly one additional case of persistent delirium for every nineteen or so patients treated, a magnitude that scales substantially across the millions of intensive care admissions worldwide each year.

The investigators went further, recognizing that persistent delirium must be measured against the possibility that a patient dies or is discharged before the outcome can even be observed. An endpoint-aligned competing-event model, which accounts for these interruptions to follow-up, produced a subdistribution hazard ratio of 1.19 with a 95 percent confidence interval of 1.10 to 1.30, again indicating a faster and more sustained accumulation of documented delirium days in the alcohol group. Triangulating across statistical frameworks that make different assumptions is a hallmark of careful observational research, and the consistency of these estimates strengthens the central finding.

External validation came from an entirely separate data ecosystem, the eICU Collaborative Research Database, version 2.0, which aggregates de-identified records from intensive care units across many United States hospitals. Here the team harmonized the endpoint definition to match the primary analysis, reconstructing persistent delirium from strictly paired CAM-ICU records in a cohort of 3,604 patients spread across 45 hospitals. The age- and sex-adjusted mixed-effects odds ratio was 1.51 with a 95 percent confidence interval of 0.98 to 2.31. The point estimate points in the same direction as the primary analysis, but the confidence interval crosses one, meaning the result is statistically inconclusive. The authors attribute this imprecision partly to a narrower withdrawal-coded subset in the multicenter data, and they present it honestly as directionally aligned but uncertain rather than as independent confirmation.

The biological plausibility of the association rests on well-characterized neurobiology. Chronic heavy alcohol exposure induces long-lasting adaptations in gamma-aminobutyric acid and glutamate signaling, dampening inhibitory tone while potentiation of excitatory N-methyl-D-aspartate receptor activity leaves the brain hyperexcitable when alcohol is withdrawn. Alcohol use disorder also promotes systemic inflammation, disrupts sleep architecture, depletes thiamine and other micronutrients, and accelerates cortical atrophy, all of which lower the threshold for the widespread network dysfunction that constitutes delirium. In the intensive care setting, where infections, sedatives, hypoxia, and metabolic derangements already assault the brain, these preexisting vulnerabilities may combine to keep patients delirious for longer stretches, extending the documented course captured by consecutive CAM-ICU positivity.

The methodological craftsmanship of the study deserves attention in its own right. Because it is a secondary analysis of de-identified data, it required no new ethics approval or participant consent; the underlying data-sharing initiatives were approved by the institutional review boards of the Massachusetts Institute of Technology and Beth Israel Deaconess Medical Center, and the eICU data met HIPAA safe-harbor certification standards. The authors followed established reporting guidelines for observational studies conducted with routinely collected health data and published an extensive suite of supplementary analyses covering exposure definitions, assessment opportunities, missingness, sensitivity analyses, matching diagnostics, and competing-event models. The research received no external funding, and the authors declare no competing interests. Received in May 2026 and published on 25 September 2026, the paper exemplifies the growing power of open critical care databases to answer clinical questions that would be impractical to address in prospective trials.

The clinical implications are forward-looking rather than immediate. As the authors conclude, this course-focused evidence complements the long-standing literature on delirium incidence and supports prospective evaluation of alcohol-informed assessment and monitoring pathways. In practical terms, it suggests that screening for alcohol use disorder at intensive care admission, perhaps with validated tools such as the revised Clinical Institute Withdrawal Assessment for Alcohol, could help clinicians identify patients who need intensified delirium monitoring, earlier sedation minimization, and structured withdrawal management. A simple diagnostic code, read at the bedside, may one day flag which delirious patients are most likely to remain trapped in confusion for days, giving families and care teams a clearer prognosis and researchers a concrete target for intervention trials. In a condition as murky as delirium, that kind of clarity is a genuine advance.

Subject of Research: The association between alcohol use disorder and persistent intensive care unit delirium in critically ill adults

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

Article References: Chen, M., & Huang, X. (2026). Alcohol use disorder and persistent intensive care unit delirium among critically ill adults: a retrospective cohort study. BMC Psychiatry. https://doi.org/10.1186/s12888-026-08641-6

Image Credits: AI Generated

DOI: 10.1186/s12888-026-08641-6

Keywords: alcohol use disorder, delirium, intensive care unit, CAM-ICU, persistent delirium, MIMIC-IV, eICU-CRD, retrospective cohort study, critical care, electronic health records, BMC Psychiatry, withdrawal