Independent prognostic value of the blood urea nitrogen-to-albumin ratio beyond the creatinine-to-albumin ratio among eICU patients with recorded sepsis and acute respiratory failure: a retrospective cohort study

Patients recorded as having both sepsis and acute respiratory failure (ARF) constitute a high-risk critically ill population, although historical database-based sepsis classifications may not correspond directly to contemporary Sepsis-3 criteria. The blood urea nitrogen-to-albumin ratio (BAR) and the creatinine-to-albumin ratio (CAR) are readily available laboratory-derived indices that integrate renal dysfunction, catabolic stress, and nutritional status. Whether BAR and CAR provide independent or overlapping prognostic information in this operationally defined population remains unclear. We conducted a retrospective cohort study using the eICU Collaborative Research Database. Adult patients with eICU-derived admission flags for both sepsis and acute respiratory failure were included; Sepsis-3 criteria were not retrospectively reconstructed. BAR and CAR were calculated from the first available measurements within 24 h of ICU admission. The primary outcome was in-hospital mortality. Multivariable logistic regression was used to evaluate independent associations, including mutual adjustment between the two ratios and residual analyses. Restricted cubic splines examined dose-response relationships. Incremental predictive performance was quantified by AUC, NRI, and IDI. Subgroup and sensitivity analyses were performed to assess robustness. A total of 6,316 patients were included, of whom 1,742 (27.6%) died in hospital. Both BAR and CAR were independently associated with in-hospital mortality in separate models. After mutual adjustment, BAR retained a statistically significant association, whereas the association for CAR was substantially attenuated and no longer significant. Both ratios similarly improved discrimination beyond a clinical base model (ΔAUC ≈ 0.028–0.029), and the combined model achieved the highest overall performance. Findings were consistent across prespecified subgroups and multiple sensitivity analyses. Among patients identified using eICU-derived admission flags for sepsis and acute respiratory failure, BAR demonstrated a more robust independent prognostic association with in-hospital mortality than CAR after mutual adjustment. Both ratios may provide readily available prognostic information within this operationally defined cohort. External validation in contemporary cohorts explicitly defined according to Sepsis-3 is required before these findings can be generalised to current sepsis populations or applied in clinical practice.