Clinical predictors of outcome in acute upper gastrointestinal bleeding.

OBJECTIVE Endoscopy has traditionally been used to risk-stratify patients with upper gastrointestinal bleeding (UGIB). This is problematic in resource-poor environments. The study aimed to identify patients who would not require urgent endoscopy by identifying clinical variables before endoscopy that predict uneventful recovery. DESIGN Prospective, descriptive cross-sectional study. SETTING Groote Schuur Hospital, Cape Town. SUBJECTS Two hundred consecutive patients aged over 12 years, presenting with haematemesis and/or melaena. OUTCOME MEASURES Good outcome, i.e. no blood transfusion, endotherapy or surgery, and alive at 1 month following presentation. RESULTS Eighty patients (40%) had a good outcome. Haemoglobin > 10 g/dl (odds ratio (OR) 25.5, 95% confidence interval (CI): 8.9-74.8; p < 0.001), absence of melaena (OR 4.8, 95% CI: 1.79-12.94, p = 0.002) and absence of syncope (OR 4.0, 95% CI: 1.67-9.48; p = 0.002) were independent predictors of good outcome. The three variables combined as a positive test had the best association with good outcome when compared with a single variable or a combination of two variables. The three-variable model had sensitivity for good outcome of 34%, specificity of 98%, and likelihood ratio for a positive test of 13.5 and for a negative test of 0.68. Thirty patients (15%) had the combination for the prediction rule, i.e. haemoglobin > 10 g/dl, no melaena and no syncope; 3 (10%) had a poor outcome (required endotherapy). CONCLUSION The prediction rule accurately excluded poor outcome, a priority in the clinical context, but did not predict good outcome. Clinical implications are a 15% reduction in unnecessary urgent endoscopies, with less than 5% of patients with poor outcome not undergoing urgent endoscopy. These findings may have particular clinical relevance in under-resourced health care environments.