Elevated blood urea nitrogen-to-albumin ratio is associated with incident sepsis-associated delirium: evidence from a weighted competing-risk analysis in a multicenter cohort study

Abstract

Background Sepsis-associated delirium (SAD) is a frequent manifestation of acute brain dysfunction in sepsis. Although the blood urea nitrogen-to-albumin ratio (BAR) has been associated with prognosis in critically ill patients, its relationship with incident SAD remains uncertain. We evaluated the association between early BAR and incident SAD. Methods This multicenter retrospective cohort study included patients with sepsis from MIMIC-IV as the primary cohort and eICU-CRD as an independent replication cohort. BAR was calculated from eligible blood urea nitrogen and serum albumin measurements obtained within 24 h after the index time. The primary outcome was 7-day incident delirium; 3-day delirium was additionally evaluated in MIMIC-IV. Death before delirium was treated as a competing event. Covariates were selected using Boruta and LASSO regression. CBPS-based inverse probability weighting and Fine–Gray competing-risk regression were applied, with restricted cubic spline, subgroup, and exploratory XGBoost/SHAP analyses. Results A total of 10,565 patients were included (9,612 MIMIC-IV; 953 eICU-CRD). In MIMIC-IV, the highest BAR quartile was associated with a greater subdistribution hazard of 7-day delirium vs. the lowest quartile (fully adjusted sHR 1.36, 95% CI 1.13–1.64; P = 0.001; P for trend = 0.002). Continuous BAR was also associated with 7-day delirium (sHR per 1-unit increase 1.00, 95% CI 1.00–1.01; P = 0.013), with similar findings for 3-day delirium. In eICU-CRD, the highest BAR quartile was likewise associated with 7-day delirium (sHR 2.20, 95% CI 1.13–4.29; P = 0.020; P for trend = 0.012), whereas the continuous association was not statistically significant (sHR 1.01, 95% CI 1.00–1.02; P = 0.075). Restricted cubic spline analyses suggested potential nonlinearity, particularly in MIMIC-IV. Exploratory SHAP analyses ranked BAR among the most influential model-based predictors in both cohorts. Conclusion Higher early BAR, particularly values in the highest quartile, was associated with a greater subdistribution hazard of incident SAD, and the highest-quartile association was independently reproduced in eICU-CRD. BAR may serve as a simple risk-associated marker for identifying patients with sepsis at higher risk of incident delirium.