Albumin-corrected anion gap–chloride index for Mortality risk in ICU-managed Gastrointestinal bleeding: a multicentre cohort analysis

Gastrointestinal bleeding is frequent in the ICU and is associated with poor outcomes. Acid–base and electrolyte abnormalities may indicate early physiological instability; however, the prognostic relevance of the albumin-corrected anion gap-to-chloride ratio (ACAG/Cl) has not been well defined. We examined the relationship between ACAG/Cl and short-term mortality in this population. Adult ICU patients with gastrointestinal bleeding were identified from two public critical care databases for derivation and external validation. ACAG/Cl was calculated from laboratory data collected within 24 h of ICU admission. The primary endpoint was 28-day all-cause mortality. Cox regression, Kaplan–Meier (K-M) analysis, restricted cubic splines (RCS), subgroup analyses, ROC curves, and external validation were used. A total of 1740 patients were included in the MIMIC-IV cohort, of whom 302 died within 28 days. Higher ACAG/Cl was independently associated with increased 28-day mortality after full adjustment (HR 1.06, 95% CI 1.03–1.08). Patients with ACAG/Cl ≥ 0.16 had a significantly higher risk of 28-day mortality than those with ACAG/Cl < 0.16 (HR 1.68, 95% CI 1.28–2.22). K-M and RCS analyses supported a significant and approximately linear association. ACAG/Cl provided a modest capacity to identify patients at higher risk of short-term death, with an area under the curve of 0.689. The result was confirmed in the eICU-CRD cohort, where patients with ACAG/Cl values of 0.16 or above had a persistently higher risk of short-term mortality after adjustment for confounders (HR 2.16, 95% CI 1.74–2.70). ACAG/Cl was independently associated with 28-day mortality in critically ill patients with gastrointestinal bleeding and showed external reproducibility in the eICU-CRD cohort. As a simple laboratory-derived marker, ACAG/Cl may help support early short-term risk stratification in this high-risk population. Not applicable.