VBM5 vs AIMS65: a comparative evaluation of prognostic scores in upper gastrointestinal bleeding
Abstract
Abstract Background Acute variceal bleeding is a life-threatening complication of cirrhosis in which early mortality risk stratification may support triage, monitoring intensity, and resource allocation. General upper gastrointestinal bleeding scores such as AIMS65 are widely used, but their performance may differ in variceal bleeding, where hepatic dysfunction and organ failure strongly influence outcomes. We aimed to derive and internally validate a parsimonious presentation-based mortality prediction model, VBM-5, and to compare its apparent performance with AIMS65. Material and Methods We performed a single-center retrospective cohort study including consecutive adult patients with cirrhosis and endoscopically confirmed acute variceal bleeding admitted between January 2024 and December 2025. The outcome was in-hospital mortality. VBM-5 was derived using logistic regression with five prespecified presentation variables: age, albumin, creatinine, international normalized ratio, and first laboratory hemoglobin. Internal validation used bootstrap optimism correction. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC), and calibration was evaluated using Brier score, calibration-in-the-large, calibration slope, and calibration plots. AIMS65, MELD-Na, Child–Pugh class, and an exploratory VBM-5 bedside points score were evaluated as comparators. Decision-curve analysis was performed as an exploratory assessment of clinical utility. Results The cohort included 224 patients, of whom 53 died during hospitalization (23.7%). The continuous VBM-5 model showed higher apparent discrimination than AIMS65 as captured in the available retrospective data (AUC 0.916 vs. 0.829; DeLong p=0.0012). Bootstrap optimism for VBM-5 AUC was 0.014, yielding an optimism-corrected AUC of 0.902; the out-of-bag mean AUC was 0.892. AIMS65 out-of-bag mean AUC was 0.828. The exploratory bedside VBM-5 points score showed a graded increase in observed mortality across score categories. However, comparison with AIMS65 was limited by incomplete standardized availability of systolic blood pressure and mental status, and calibration of mapped comparator scores was cohort-derived. Conclusions In this single-center derivation cohort, VBM-5 showed promising optimism-corrected discrimination for in-hospital mortality after acute variceal bleeding. Nevertheless, the positive hemoglobin coefficient, borderline events-per-variable ratio, treatment heterogeneity, and incomplete AIMS65 component ascertainment limit transportability. External multicenter validation, reassessment of coefficient stability, and recalibration are required before clinical implementation.