Introduction Antithrombotic drugs (ATDs) commonly cause non-variceal upper gastrointestinal bleeding (NVUGIB). Established risk scoring systems have not been validated in users of these drugs. We aimed to compare Blatchford, Rockall, and Charlson scores in predicting the outcomes of NVUGIB in ATD users and controls.

Methods A total of 2071 patients with NVUGIB, 2005–2011, were grouped into ATD users (n = 851) and controls (n = 1220). ATDs included low-dose aspirin, clopidogrel, dipyridamole, warfarin, and low-molecular weight heparin. Outcomes included length of hospital admission, the need for blood transfusion, re-bleeding requiring surgery, and 30-day mortality. Results were expressed as Spearman rank correlation coefficients (Rs) for length of admission and area-under-the-curve (AUC) values for the receiver opating characteristic curves (ROC) for binary outcomes, and were compared using z-tests, after Fisher’s transformation in the case of Rs values.

Results (1) The LENGTH OF ADMISSION correlated with all three scores in non-ATD patients (controls), but these correlations were significantly weaker in ATD users. Rs in control vs. ATD: 0.45 vs. 0.20 for Blatchford; 0.48 vs. 0.32 for Rockall; and 0.42 vs. 0.26 for Charlson, all P < 0.001. Rockall had the strongest correlation with duration of admission and Blatchford the weakest (P < 0.01 vs. Rockall in ATD users). (2) The NEED FOR TRANSFUSION was best predicted by Blatchford (P < 0.001 vs. Rockall and Charlson in both ATD users and controls) followed by Rockall (P < 0.001 vs. Charlson in controls). All scores performed less well in ATD users than controls. AUC in control vs. ATD: 0.90 vs. 0.85 for Blatchford; 0.74 vs. 0.59 for Rockall; and 0.64 vs. 0.54 for Charlson, all P < 0.005. (3) In predicting the NEED FOR SURGERY, only Rockall performed significantly better than by chance. AUC in control vs. ATD: 0.62 vs. 0.59 for Blatchford; 0.73 vs. 0.74 for Rockall; and 0.53 vs. 0.50 for Charlson. (4) In predicting MORTALITY, the Charlson score performed best by a small margin, and there was a trend towards weaker relationships in ATD users. AUC in control vs. ATD: 0.71 vs. 0.61 for Blatchford; 0.74 vs. 0.71 for Rockall; and 0.81 vs. 0.72 for Charlson.

Conclusion (1) In both ATD users and controls, the Blatchford score was the strongest predictor of the need for blood transfusion, Rockall had the strongest correlation with duration of admission and with re-bleeding requiring surgery, and Charlson was best in predicting 30-day mortality. (2) There was a consistent tendency for all scoring systems to be less effective in predicting outcomes in ATD users than in controls. (3) Modifications of risk scoring systems should be explored to improve their efficiency in users of antithrombotic drugs.

Disclosure of Interest A. Taha Consultant for: Almiral UK, Vifor UK, Horizon Pharma USA, C. McCloskey: None Declared, T. Craigen: None Declared, W. Angerson: None Declared.