Development, Validation, and Comparative Assessment of an International Scoring System to Determine Risk of Upper Gastrointestinal Bleeding

Abstract

The Glasgow-Blatchford score (GBS) and pre-endoscopy Rockall score (pRS) are used in determining prognoses of patients with acute upper gastrointestinal bleeding, but neither predicts outcomes of patients with a high level of accuracy. A scoring system is needed to identify patients at risk of adverse outcomes and patients at low risk of harm. We pooled data from 5 data sets in Canada, the United Kingdom, and Australia on 12,711 patients with acute upper gastrointestinal bleeding. The GBS and pRS were calculated for each patient. We performed multivariable logistic regression modeling of data from 10,639 cases to develop the new scoring system Canada - United Kingdom - Adelaide (CANUKA). We performed area under the receiver operating characteristic analyses to test the ability of CANUKA to identify patients who died or had rebleeding within 30 days, surgical or radiologic intervention to control bleeding, need for therapeutic endoscopy, and transfusion-a poor outcome was defined as 1 or more of these outcomes. Patients at low risk of a poor outcome (safe for management as an outpatient) were identified based on lack of transfusion, rebleeding, therapeutic endoscopy, interventional radiology or surgery, or death. We validated in 2072 patients from a separate cohort compiled from 2 datasets. In the development data set there was no difference between GBS and pRS in identifying patients who died without 30 days of bleeding (area under the receiver operating characteristic curve [AUROC], 0.67; 95% CI, 0.62-0.72 for GBS; AUROC, 0.70; 95% CI, 0.66-0.74 for pRS; P=.21). The GBS was superior to the pRS in identifying patients with rebleeding, hemostatic interventions, and transfusions. In the validation data set, CANUKA had higher accuracy than the GBS in identifying patients who died within 30 days of bleeding (AUROC, 0.77 vs 0.74; P=.047), but there was no significant difference in the accuracy of these scoring systems in identifying patients who required hemostatic intervention. The GBS more accurately identified patients who required therapeutic endoscopy (AUROC, 0.78; 95% CI, 0.76-0.81 for GBS; AUROC, 0.77; 95% CI, 0.74-0.79 for CANUKA; P= .47). For patients classified as low-risk patients by CANUKA (score ≤1), 96.3% were safely discharged, whereas 16 patients with a GBS ≤1 had an adverse outcome (a 95.3% probability of safe discharge). In an international validation analysis of the GBS and pRS for patients with acute upper gastrointestinal bleeding, we found the GBS to more accurately identify those who later required hemostatic interventions and transfusions; the scoring systems identified 30-day mortality or rebleeding with equal levels of accuracy. We developed a scoring system (CANUKA) that had similar performance to the GBS in predicting patient outcomes and it more accurately identifies patients at low risk for adverse outcomes.