Creatinine-to-albumin ratio for early ICU-course (first 24 hours) risk stratification of mortality in critically ill patients with gastrointestinal bleeding: A retrospective cohort study

ObjectiveThe creatinine-to-albumin ratio (CAR) has recently been recognized as a composite biomarker reflecting both renal function and nutritional status, showing strong prognostic relevance across diverse diseases. However, its prognostic significance in gastrointestinal bleeding (GIB) remains unclear. This study aimed to investigate the association between CAR and mortality risk in patients with GIB and to explore potential non-linear risk patterns to identify a candidate inflection point for exploratory stratification.MethodsWe conducted a retrospective cohort study using the MIMIC-IV database, including adult patients with a first ICU admission for GIB, identified by ICD-9/10 codes. Patients with GIB were stratified into quartiles by CAR levels. Survival outcomes were evaluated using Kaplan-Meier curves, multivariable Cox regression, receiver operating characteristic (ROC) analysis, and restricted cubic spline (RCS) modeling. To develop a clinically applicable tool, the Boruta algorithm identified key prognostic variables, which were incorporated into a Cox-based nomogram. Model discrimination, calibration, and clinical utility were assessed using ROC, bootstrap-corrected calibration curves, concordance index (C-index), and decision curve analysis (DCA). External validation was performed in the eICU-CRD cohort. Predictors were derived from data recorded within the first 24 hours after ICU admission, consistent with an early ICU-course risk model.ResultsA total of 1,611 patients (median age 66 years) were included. All-cause mortality was 18.6% at 28 days and 26.8% at one year. Patients in the highest CAR quartile had significantly higher mortality than those in the lowest (31.3% vs. 7.2% at 28 days; 42.4% vs. 11.7% at 365 days; both p<0.001). In multivariable models, CAR independently predicted mortality (28-day HR 1.29, 95% CI 1.09-1.53; 365-day HR 1.24, 95% CI 1.07-1.43). ROC analysis showed moderate discrimination, with CAR outperforming creatinine or albumin and adding incremental value to ICU-window AIMS65 (operationalized using data recorded within the first 24 hours after ICU admission). RCS and subgroup analyses showed consistent results, and external validation supported the generalizability of the CAR-mortality association and the non-linear pattern. The final nomogram integrating five predictors (sepsis, APTT, CAR, TB, and BUN) demonstrated strong discrimination (28-day AUC 0.779, 95% CI 0.750-0.807), good calibration, and a corrected C-index of 0.774.ConclusionsElevated CAR independently predicts mortality in GIB and may aid early ICU-course (first-24-hour) risk stratification. Any candidate inflection point derived from spline analyses should be considered exploratory and requires further validation before clinical adoption.