Influence of Language Discordance on Diagnosis of Delirium Among Critically Ill Adults

Tuesday, March 24, 2026

11:15 AM - 12:15 PM Central Time

Location: Connections Central - RST 09

BZ

Baihui Zhao, MA, MPH Candidate

University of Texas Southwestern Medical Center

Disclosure(s): No relevant financial relationship(s) to disclose.
VC

Vaishvik Chaudhari, MS

Data Scientist
Rush University Medical Center

Disclosure information not submitted.
CC

Catherine Chen, MD, MSHI

Dr
Parkland Health and Hospital System

Disclosure information not submitted.
Kelly Feldman, MD

UCLA Mattel Children's Hospital

Disclosure(s): No relevant financial relationship(s) to disclose.
AK

Arav Kumar, N/A

Research Assistant
MIT’s Laboratory for Computational Physiology

Disclosure information not submitted.
AS

Ali Sipra, MS

Data Analyst
University of Delaware

Disclosure information not submitted.
KW

Kyle Weant, PharmD, BCPS, BCCCP, BCEMP, FCCP

Clinical Assistant Professor
University of South Carolina College of Pharmacy

Disclosure information not submitted.

Introduction: Social determinants of health have been found to be independent risk factors for developing intensive care unit (ICU) delirium. Previous data have shown an association between language discordance and increased restraint use in adults receiving mechanical ventilation. However, the association between language discordance and delirium diagnosis has not been previously evaluated.

Methods: Primary data sources were Medical Information Mart for Intensive Care IV (MIMIC-IV) and the Korean Medical Information Mart (K-MIMIC). Adult patients admitted to the ICU for ≥48 hours were included. Patients who received neuromuscular blockade were excluded. Delirium was defined as ≥1 positive CAM-ICU assessment. Language discordance (LD) was defined as a non-English preferred language in the MIMIC-IV dataset. LD patients were 1:1 matched to language concordant (LC) patients using nearest-neighbor matching without replacement. A logistic regression model was created with delirium as the dependent variable and LD as the exposure. Covariates included patient age group, vasopressor use, preexisting substance use disorder or dementia, and ICU length of stay (LOS). A secondary linear regression analysis was performed with ICU LOS as the outcome.

Results: Of the 47,048 patients from MIMIC-IV who met inclusion criteria, 42,223 were LC and 4,825 were LD. While LD was negatively associated with delirium diagnosis (OR 0.86 [95% CI: 0.79, 0.94]), LD patients were more likely to have “unable to assess” (UTA) CAM-ICU assessments (OR 3.96 [95% CI: 3.57, 4.40]). Both LD (10.0 hours [95% CI: 4.9, 15.1]) and delirium (71.8 hours [95% CI: 66.5, 77.2]) were associated with increased ICU LOS. Of note, a much lower incidence of delirium was found in the K-MIMIC dataset (3.4%) as compared to the MIMIC-IV dataset (37.2%).

Conclusions: While LD is negatively associated with delirium diagnosis, there is a strong positive association with UTA CAM-ICU assessments. Increased UTA CAM-ICU assessments among LD patients may result in systematic under-diagnosis of ICU delirium. Both delirium and LD are associated with increased ICU LOS. UTA CAM-ICU documentation is common and its effect on the empiric treatment of delirium remains unknown, and further investigation is required.