Comparison of predictive and prognostic risk stratification scores for non-variceal upper gastrointestinal bleeding: a retrospective cohort analysis in a tertiary center

Aims Several risk stratification scores have been developed for predicting endoscopic treatment or patient survival in non-variceal upper gastrointestinal bleeding (NVUGIB). Our aim was to compare the performance of some previously described scores (Canada-United Kingdom-Adelaide (CANUKA) score, Glasgow-Blatchford score (GBS), HARBINGER score, neutrophil-to-lymphocyte ratio (NLR), red cell distribution width (RDW), Rockall pre-endoscopy and complete scores) in predicting the following outcomes in patients with NVUGIB: need for endoscopic intervention, in-hospital rebleeding rate, need for red blood cell (RBC) transfusion, need for surgical or interventional radiological procedure, 30-day mortality.

Methods Patients undergoing emergency endoscopy for upper GI bleeding at the Department of Emergency Medicine at Semmelweis University between January 1, 2022 and December 31, 2023 were included in a retrospective cohort analysis. Endoscopic treatment was applied to lesions with high-risk stigmata (HRS) defined by ESGE, based on the Forrest classification. Statistical methods included ROC analysis, through which we obtained the AUROC values.

Results There were 528 patients with NVUGIB, including 99 patients (61 males, 38 females, mean age 72) with high-risk stigmata (HRS) and 429 patients (246 males, 183 females, mean age 70) with low-risk stigmata (LRS). In-hospital rebleeding rate was 3.6% (n=19/528), all-caused 30-day mortality was 17.80% (n=94/528). Due to missing data in 51 patients, scoring systems were analyzed in 477 patients. Area under receiver operator characteristic curves (AUROC) were generated for predicting need for endoscopic intervention: Forrest: 0.91, CANUKA: 0.57, GBS: 0.58, NLR: 0.55, RDW: 0.46, Rockall (pre-endoscopy): 0.53, Rockall (complete): 0.69, HARBINGER: 0.57; in-hospital rebleeding rate: Forrest: 0.65, CANUKA: 0.67, GBS: 0.55, NLR: 0.50, RDW: 0.51, Rockall (pre-endoscopy): 0.54, Rockall (complete): 0.70, HARBINGER: 0.67; need for RBC transfusion: Forrest: 0.64, CANUKA: 0.79, GBS: 0.80, NLR: 0.51, RDW: 0.68, Rockall (pre-endoscopy): 0.62, Rockall (complete): 0.66, HARBINGER: 0.64; need for surgical or interventional radiological procedure: Forrest: 0.77, CANUKA: 0.62, GBS: 0.56, NLR: 0.64, RDW: 0.46, Rockall (pre-endoscopy): 0.57, Rockall (complete): 0.72; 30-day mortality: Forrest: 0.53, CANUKA: 0.70, GBS: 0.65, NLR: 0.67, RDW: 0.60, Rockall (pre-endoscopy): 0.70, Rockall (complete): 0.70, HARBINGER: 0.62.

Conclusions Currently used risk stratification scores have a suboptimal performance (AUROC≤0.8) as per ESGE guideline in predicting the need for endoscopic therapy, in-hospital rebleeding, need for surgical or interventional radiological procedure and 30-day mortality. GBS and CANUKA scores proved to be effective in estimating the need for RBC transfusion, whereas CANUKA and Rockall scores seem to be acceptable in estimating 30-day mortality. Further prospective studies on larger patient cohorts exploiting artificial intelligence are needed to analyze and optimize risk stratification scores for NVUGIB.

Publication History

Article published online:
27 March 2025