Comparison of pre-endoscopic risk scores for identifying high-risk Forrest lesions in emergency department patients with suspected upper gastrointestinal bleeding.

BACKGROUND: Early risk stratification in suspected upper gastrointestinal bleeding may help guide admission decisions, monitoring, and timing of endoscopy in the emergency department. Although several pre-endoscopic scores are widely used, their ability to predict high-risk endoscopic findings remains incompletely defined. This study aimed to compare the performance of established pre-endoscopic upper gastrointestinal bleeding scores and anticoagulation-related bleeding risk scores for predicting high-risk Forrest findings. METHODS: This prospective multicentre cohort study included adult patients who underwent upper gastrointestinal endoscopy for suspected upper gastrointestinal bleeding. The primary outcome was high-risk endoscopic findings, defined as Forrest Ia-IIb. Low-risk findings were defined as Forrest IIc-III. We compared the performance of CHAMPS, Glasgow-Blatchford score, AIMS65, pre-endoscopic Rockall, HAS-BLED, ATRIA, ORBIT, VTE-BLEED, and HEMORR₂HAGES using receiver operating characteristic analysis. Multivariable logistic regression, calibration assessment, 10-fold cross-validated modelling, and Brier score were used to evaluate incremental discriminatory contribution beyond a baseline clinical model. RESULTS: A total of 571 patients were included, of whom 199 patients (34.9%) had high-risk Forrest findings. Patients with high-risk findings were older and had greater haemodynamic instability, lower haemoglobin, higher blood urea nitrogen, greater transfusion requirement, higher intensive care unit admission, and higher mortality than those with low-risk findings. Among all evaluated scores, CHAMPS had the best overall performance for predicting high-risk endoscopic findings, with an AUC of 0.889 (95% CI: 0.860-0.916). At the optimal cut-off of ≥2, CHAMPS achieved 72.9% sensitivity, 94.9% specificity, 88.4% positive predictive value, and 86.7% negative predictive value. CHAMPS significantly outperformed the Glasgow-Blatchford score (AUC 0.831; p = 0.001). In multivariable analysis, lower haemoglobin, higher blood urea nitrogen, lower systolic blood pressure, higher heart rate, and older age were independently associated with high-risk Forrest findings. Adding CHAMPS to the baseline clinical model produced the greatest incremental improvement, increasing the cross-validated AUC from 0.804 to 0.898 and reducing the Brier score from 0.165 to 0.109. CHAMPS remained the best-performing score in both anticoagulated and non-anticoagulated patients. CONCLUSION: In patients undergoing endoscopy for suspected upper gastrointestinal bleeding, CHAMPS showed stronger discrimination for high-risk Forrest findings than the other evaluated scores. Prediction of high-risk endoscopic stigmata should be considered complementary to existing pre-endoscopic UGIB risk stratification rather than a replacement for established risk scores. Conventional upper gastrointestinal bleeding scores remained clinically useful, whereas anticoagulation-related bleeding risk scores showed more limited value for acute endoscopic risk prediction. Additional validation in different emergency department settings is needed before broader clinical use.