Objective: Early prognostic stratification could optimize the management of patients with upper gastrointestinal bleeding (UGIB) and reduce unnecessary hospitalizations. The aim of this study was to assess and compare the performance of existing prognostic scores in predicting therapeutic intervention and/or death.

Methods: A systematic search of the literature identified existing prognostic scores. A multicenter retrospective cohort study included adult patients hospitalized for UGIB from January 1, 2019, to December 31, 2020. The primary outcome was a composite including therapeutic intervention within 7 days (blood transfusion, endoscopic, surgical, or interventional radiology hemostasis) and/or 30-day death. Discrimination performance was estimated by the area under the curve (AUC). The ability to identify low-risk patients was analyzed using sensitivity and negative predictive value (NPV), for defined thresholds.

Results: The systematic search identified 39 prognostic scores, 12 of which could be analyzed.

Among the 990 patients included, therapeutic intervention and/or death occurred in 755 (76.4%) patients. Scores with the highest discriminative performance to predict the primary composite outcome were Glasgow-Blatchford score (GBS) (AUC 0.869 [0.842-0.895]), modified GBS (AUC 0.872 [0.847-0.898]) and modified GBS 2 (AUC 0.855 [0.827-0.884]).

The best performance to identify low-risk patients was for GBS ≤ 1 (sensitivity 0.99 [0.99-1.00], NPV 0.89 [0.75-0.97]) and modified GBS = 0 (sensitivity 0.99 [0.98-1.00], NPV 0.84 [0.71-0.94]).

Conclusions: The GBS and modified GBS are the two most performing scores because they achieve both key objectives: stratifying patients based on their risk of therapeutic intervention and/or death, and identifying low-risk patients who may qualify for outpatient management.