Objective To determine whether the Glasgow–Blatchford bleeding score (GBS) or pre-endoscopy Rockall score was better at accurately identifying patients with acute upper gastrointestinal bleeding (AUGIB) who were at low risk of the need for clinical intervention and death and therefore suitable for outpatient management.

Design Retrospective database and case note review of all patients admitted to the emergency departments with AUGIB from 1 January 2008 to 31 December 2009.

Setting Two tertiary centre teaching hospitals.

Patients 432 patients met the inclusion criteria.

Intervention None.

Main outcome measure Clinical interventions (blood transfusion, endoscopic therapy and surgery) and death.

Results Of 432 patients, 40 (9.3%) had a GBS of 0 and none required intervention or died. In contrast, 13/104 patients (12.5%) who had a pre-endoscopy Rockall score of 0 and 23/125 patients (18.4%) who had a complete Rockall score <3, required clinical intervention. The performance of the scores at these cut-offs were: GBS (sensitivity 100%, specificity 16.1%, positive predictive value (PPV) 37.8%, negative predictive value (NPV) 100%, accuracy 82.3%), pre-endoscopy Rockall (sensitivity 91.2%, specificity 32.0%, PPV 41.2%, NPV 87.5%, accuracy 70.9%) and complete Rockall (sensitivity 84.5%, specificity 50.7%, PPV 55.8%, NPV 81.6%, accuracy 76.2%). For prediction of the need for intervention or death, the accuracy of the GBS (0.82; 95% CI 0.78 to 0.86) was superior to the pre-endoscopy Rockall score (0.71; 95% CI 0.67 to 0.76).

Conclusion The GBS but not the pre-endoscopy Rockall score identifies patients with upper gastrointestinal bleeding who may be suitable for outpatient management, therefore potentially allowing for more efficient use of hospital resources.