A modified risk model for upper gastrointestinal bleeding: comparative evaluation of the H3B2A1 score

1. Introduction

Upper gastrointestinal bleeding (UGIB), one of the gastrointestinal emergencies, is the most important cause of mortality and morbidity worldwide. Despite advancements in diagnostic and treatment, the mortality rate remains between 5% and 15% because of the growing elderly population and comorbid conditions [1, 2]. UGIB accounts for more than 50% of all gastrointestinal bleeding cases and hospitalizations [3, 4]. Accurate early risk stratification of patients with UGIB is essential for guiding clinical decision-making, including the need for intensive care unit (ICU) admission, urgent endoscopic intervention, or safe discharge. Over the years, various prognostic scoring systems have been developed to assess the severity of UGIB and predict outcomes, including mortality, the need for transfusion, rebleeding, and the length of hospital stay [4]. These scoring systems are typically classified into two categories: pre-endoscopy and post-endoscopy models. Commonly used pre-endoscopy scores include the Glasgow-Blatchford Score (GBS), AIMS65, the Assessment of Blood Consumption (ABC) score, and the pre-endoscopy Rockall score (pRS) [5]. However, despite their widespread use, many scoring systems remain underutilized in emergency settings due to complexity and poor memorability, defined as the clinician’s ability to recall and apply the tool rapidly under time constraints. Tools requiring numerous inputs or subjective assessments may delay critical decision-making in acute care. Specifically, the GBS and modified GBS (mGBS) require multiple clinical and laboratory inputs, which may hinder their rapid application at the bedside [6].

Sasaki et al. [3] (2022) introduced the H3B2 score, a novel risk stratification model based on simple and objective clinical parameters. It is recommended for high-risk patients with UGIB who require emergency hemostatic interventions, including emergency endoscopy [3]. In line with prior efforts to simplify risk stratification, the H3B2A1 score incorporates serum albumin (ALB) into the original H3B2 model, aiming to enhance prognostic utility in UGIB.

This study aimed to evaluate the predictive performance of the H3B2A1 score in estimating in-hospital mortality among patients with UGIB. Additionally, we compared the prognostic accuracy of the H3B2A1 score with established risk stratification tools, including AIMS65, GBS, and mGBS, in predicting secondary clinical outcomes, such as ICU admission and the need for early endoscopic intervention. The ultimate goal was to validate the H3B2A1 score as a clinically practical and easy-to-use tool for early risk assessment in emergency settings.

2. Materials and methods

This retrospective single-center study was conducted at a tertiary care hospital, between 01 June 2022 and 31 July 2023. Patients aged 18 years and older who presented to the emergency department with gastrointestinal bleeding symptoms (hematemesis, hematochezia, melena, syncope, lethargy, confusion) and were diagnosed with upper gastrointestinal bleeding (UGIB) following endoscopic evaluation were included in the study.

Demographic data, presenting symptoms, comorbidities (e.g., liver disease, heart disease, diabetes mellitus, chronic kidney disease, cerebrovascular disease, Alzheimer’s disease, hypertension, malignancy, and hyperlipidemia), medication history (e.g., proton pump inhibitors, warfarin, aspirin, clopidogrel, rivaroxaban, ticagrelor, nonsteroidal anti-inflammatory drugs (NSAIDs), vital signs, level of consciousness, laboratory results, endoscopic findings, need for blood transfusion, hospitalization status, and in-hospital mortality were collected from electronic medical records and emergency department forms. When multiple measurements were recorded before endoscopy, the most extreme values were used for analysis.

Exclusion criteria included variceal bleeding, lack of endoscopic evaluation, trauma-related bleeding, age under 18 years, and inaccessible or incomplete medical records. Patients were categorized into two groups: survivors and non-survivors.

3. Results

A total of 649 patients admitted to the emergency department with suspected gastrointestinal bleeding (GIB) were retrospectively analyzed. Among them, 233 patients who underwent endoscopic evaluation and were diagnosed with UGIB were included in the final analysis. Patients were divided into two groups based on in-hospital outcomes: survivors and non-survivors. The overall in-hospital mortality rate was 10.7% (n = 25). The mean age of the patients was 70.28 ± 14.96 years, with 67.38% of the cohort aged 66 years or older. Demographic and clinical characteristics of the study population are summarized in Table 1.

The mean hemoglobin (Hb) level was 8.89 ± 2.32 g/dL in the total cohort, with no statistically significant difference between survivors (8.99 ± 2.34) and non-survivors (8.10 ± 1.98) (p = 0.068, independent samples t-test). Platelet counts were similar between survivors (266.09 ± 118.94 × 10³/μL) and non-survivors (273.04 ± 225.58 × 10³/μL), with no statistically significant difference (p = 0.881). However, serum albumin levels were significantly lower in non-survivors (26.84 ± 4.45 g/L) than in survivors (31.84 ± 5.43 g/L) (p < 0.001, t-test). A comparison of laboratory parameters is presented in Table 2.

To assess the prognostic performance of each scoring system, ROC curve analyses were conducted. The AUC, 95% CI, cutoff points, sensitivity, specificity, Youden’s J index, and p-values were calculated for in-hospital mortality and ICU admission. These results are summarized in Table 3.

All five scoring systems demonstrated statistically significant performance in predicting in-hospital mortality (p < 0.001 for all). The H3B2A1 score yielded an AUC of 0.750 (95% CI: 0.653–0.846), comparable to the AIMS65 score (AUC: 0.754; 95% CI: 0.655–0.852), and showed 88.0% sensitivity. H3B2 and mGBS also performed moderately well (AUC = 0.725 and 0.731, respectively). The GBS showed the lowest AUC (0.706), although it was still statistically significant.

Regarding the prediction of ICU admission, AIMS65 had the highest predictive accuracy (AUC = 0.844; 95% CI: 0.790–0.899), with a sensitivity of 96.1% and specificity of 56.0%. The H3B2A1 model demonstrated moderate accuracy (AUC = 0.645; 95% CI: 0.563–0.727), while H3B2, GBS, and mGBS had lower AUC values (0.624, 0.606, and 0.613, respectively). The complete results are presented in Table 3, with ROC curves illustrated in Figs. 1,2.

Fig. 1.ROC curve comparison of five scoring systems in predicting in-hospital mortality. ROC: Receiver operating characteristic; GBS: Glasgow-Blatchford Score; mGBS: modified Glasgow-Blatchford Score.

Fig. 2.ROC curve comparison of five scoring systems in predicting ICU admission. ROC: Receiver operating characteristic; GBS: Glasgow-Blatchford Score; mGBS: modified Glasgow-Blatchford Score.

4. Discussion

Gastrointestinal bleeding (GIB) is one of the gastroenterology emergencies requiring medical intervention and one of the most common causes of hospitalization [6]. Although the incidence varies according to age, elderly patients are more frequently affected, and the frequency of hospitalization increases with age. The most common cause of UGIB in the elderly population is gastric and duodenal ulcers or esophagitis, which is observed in 80% of cases [7]. In a 2007 study in the UK, 63% of patients diagnosed with UGIB were over 60 years of age [8]. Consistent with these findings, our study showed that the mean age of patients was 70.28 ± 14.96 years, with 67.38% of patients being over 65 years old. The in-hospital mortality rate was 10.7%, and mortality increased with age.

Given the substantial morbidity and mortality associated with acute UGIB, early assessment and risk prediction are critically important for optimizing patient outcomes. The severity of acute GIB ranges from mild to life-threatening. Therefore, various scoring systems are used to evaluate the condition and prognosis of patients [9]. However, in emergency department settings, the practicality and accessibility of parameters are crucial for the applicability of these scores. Although scores like GBS and mGBS are validated, their complexity limits routine use in high-pressure emergency settings [4, 5, 10]. Especially, commonly used risk scores, such as GBS, AIMS65, and Rockall, may be impractical in emergency clinical settings due to their reliance on laboratory and endoscopic data. In response, recent research has emphasized the need for simpler and more rapidly calculable indices [10]. Kocaoğlu and Çetinkaya (2021) reported the Age Shock Index, a metric based on vital signs and age, which showed moderate predictive ability and outperformed both the Shock Index (SI) and Modified Shock Index (MSI) in patients with GIB. Similarly, our study suggests that incorporating accessible parameters, such as serum albumin, may enhance early risk stratification and support prompt, bedside clinical decision-making.

In our study, the AIMS65 (AUC: 0.754) and the H3B2A1 (AUC: 0.750) scores were particularly effective in predicting in-hospital mortality. mGBS (AUC: 0.731), H3B2 (AUC: 0.725), and GBS (AUC: 0.706) also showed fair performance. These findings align with those of Sasaki et al. [3], who reported comparable predictive power for AIMS65 (AUC: 0.707) and H3B2 (AUC: 0.685) in UGIB, outperforming more complex scores, such as GBS (AUC: 0.587) and mGBS (AUC: 0.594). Similarly, Cazacu et al. [4] found that all scores had predictive value for mortality even in patients who did not undergo endoscopy, with AIMS65 and INBS performing particularly well.

As highlighted by Sasaki et al. [3], the H3B2 score is a more practical alternative for predicting mortality in UGIB patients. While the AIMS65 score includes easily obtainable parameters, such as age and mental status, the Glasgow-Blatchford Score (GBS) requires the evaluation of multiple variables, including the presence of hepatic and cardiac diseases, which may not always be readily accessible in emergency settings. Furthermore, in patients presenting with altered mental status, it is essential to consider other potential causes before applying scoring criteria. Taken collectively, these considerations emphasize the importance of promptly identifying high-risk UGIB patients, planning timely endoscopies, and accurately predicting in-hospital mortality to guide effective clinical management [11].

The literature shows that early risk classification indices are inadequate in the clinical decision-making process before endoscopy in high-risk patients, particularly when determining ICU admission or hospitalization in the relevant service. Generally, risk stratification indices have been developed against the possibility of mortality and rebleeding [12, 13]. Kherad et al. [12] (2023) evaluated ICU admission rates with the AIMS65, GBS, ABC, and Rockall scores (RS). They reported that ICU admission rates increased significantly for all high-risk patient scores predicted for all scores except the Rockall score, but AUC curves showed poor discriminatory ability for all scores (ABC: AUC: 0.55 (51%–60%); AIMS65: AUC: 0.61 (56%–66%); GBS: AUC: 0.61 (55%–66%), RS: 0.51 (53%–64%)). In our study, in patients admitted to the ICU, in addition to AIMS65 (AUC: 0.844 (95% CI: 0.790–0.899)), other scoring systems (GBS: AUC: 0.606 (0.519–0.69); mGBS: AUC: 0.613 (0.525–0.701); H3B2: AUC: 0.624 (0.535–0.713); H3B2A1: AUC: 0.645 (0.563–0.727)) showed similar characteristics in predicting ICU admission (p < 0.001). This is especially important because recent studies have questioned the clinical effectiveness of urgent endoscopy in all UGIB patients, particularly those without variceal bleeding.

Jeon et al. [14] (2024) reported that delayed endoscopy might be clinically acceptable in low-risk non-variceal upper gastrointestinal bleeding (NVUGIB) patients, whereas early intervention could decrease in-hospital complications in high-risk individuals. Consistent with these findings, our study highlights the importance of early risk stratification to support personalized patient management and enable timely clinical decisions. Additionally, recent studies have underscored the prognostic value of serum albumin in UGIB. Li et al. [15] (2023) identified hypoalbuminemia (<3.0 g/dL) as an independent predictor of in-hospital mortality, while Vara-Luiz et al. [16] (2025) associated it with increased hemodynamic instability and ICU admissions in elderly patients with non-cirrhotic UGIB. These findings strengthen the classification of UGIB as a geriatric syndrome, suggesting that including albumin in risk-stratification models may improve early prognostication and guide personalized management in high-risk populations.

In conclusion, all scoring systems are designed to predict the need for early endoscopic treatment, mortality, and transfusion [13]. However, the primary goal of risk stratification in emergency departments is to identify patients suitable for discharge [17]. Our results support using simplified models, such as H3B2A1, alongside validated scores like AIMS65, especially in elderly patients who might present with atypical symptoms or multiple comorbidities. More research is needed to externally validate these tools and evaluate their influence on clinical decisions and outcomes in various UGIB populations.

This study has several limitations. First, it was carried out at a single tertiary care center using a retrospective design, which may limit the generalizability of the findings. Second, the sample size was relatively small, especially in the non-survivor group, which could affect the statistical power of the analyses. Finally, external validation of the H3B2A1 score in various clinical settings and patient populations was not performed. Therefore, future prospective multicenter studies are needed to confirm and build upon these findings.

5. Conclusions

In this study, we evaluated the clinical utility of the H3B2A1 score—a modified version of the H3B2 model that incorporates serum albumin—for predicting in-hospital mortality and ICU admission in patients with upper gastrointestinal bleeding (UGIB). Although H3B2A1 is not yet a widely recognized score in the literature, our findings show that its predictive performance is similar to that of the AIMS65 score. The inclusion of serum albumin, a biomarker often linked to poor prognosis in various clinical settings, may improve the score’s effectiveness, especially in elderly or frail populations. Because of its simplicity and use of objective, routinely available clinical parameters, H3B2A1 could be a practical tool for early risk stratification in emergency departments. Nevertheless, further prospective, multicenter studies are warranted to validate its performance and clinical applicability across varied UGIB populations.

Availability of data and materials

The data supporting this study’s findings are available from the corresponding author upon reasonable request.

Author contributions

MF, MÇ and RK—designed the research; software; writing, review, and editing; supervision. MF, MÇ, ABÖ, MB, SG and RK—conceptualization. MB, MÇ, ABÖ and SG—methodology, data curation. MF and RK—validation; formal analysis; investigation. MF, MÇ and MB—resources; visualization; writing–original draft preparation, project administration. All authors have read and agreed to the published version of the manuscript.

Ethics approval and consent to participate

This study was approved by the Clinical Research Ethics Committee of Balikesir University (Approval No: 2023/66). All procedures were conducted according to the ethical rules and principles of the Declaration of Helsinki. Consent for participation was obtained from all subjects.

Acknowledgment

Not applicable.

Funding

This research received no external funding.

Conflict of interest

The authors declare no conflict of interest.