Decisions on managing bleeders remain a critical issue due to the high death risk. The Glasgow-Blatchford score (GBS) correctly identifies low-risk patients but none of the existing scores precisely assess the high risk patients.

To evaluate bleeding scores’ prognostic performances in predicting mortality risk.

To compare the scores in low and high-risk patients and identify the “best performing cut-off” (if different from the standard one) in discriminating survivors from deceased.

prospective multicenter cohort study including consecutive UGIB patients admitted to 50 Italian hospitals. We collected information to calculate Rockall, PNED, AIMS65, GB, and ABC scores, together with demographic and clinical data, and outcomes.

for low-risk patients, the GB and the ABC are the best performing scores; for high-risk patients, all scores showed weak results, with the PNED score having the higher PPV. Searching for the “best performing cut-off”, we found different points that determined a relevant numerical gain in terms of patients correctly assessed.

we suggest using the GBS and the ABC score at admission, while the PNED appears to be more useful for high-risk. We also suggest using a new decisional cut-offs that, if validated, may increase the accuracy of current scores.