Application and Internal Validation of the Modified Glasgow-Blatchford Score for Early Triage of Acute Upper Gastrointestinal Bleeding

Summary

This retrospective cohort study evaluates the modified Glasgow-Blatchford score for objective early triage of acute upper gastrointestinal bleeding, focusing on low-risk classification, false-negative cases, and clinical safety.

Abstract

Acute upper gastrointestinal bleeding requires early risk stratification to identify patients who may need urgent intervention while avoiding unnecessary hospitalization among low-risk cases. This retrospective single-center cohort study evaluated the modified Glasgow-Blatchford score as an objective triage tool in comparison with the original Glasgow-Blatchford score. A total of 366 consecutive patients with acute upper gastrointestinal bleeding were included. Clinical presentation, initial vital signs, and the earliest laboratory results obtained at presentation were used to calculate both scores. The primary high-risk outcome was a composite of in-hospital blood transfusion, therapeutic endoscopy, surgery, or death. Using a predefined low-risk threshold of ≤1, diagnostic performance was evaluated overall and in variceal and non-variceal subgroups. Overall, 151 of 366 patients (41.26%) met the composite high-risk outcome, and 18 patients (4.92%) died during hospitalization. At the predefined threshold, the modified score showed higher sensitivity, specificity, positive predictive value, negative predictive value, and agreement than the original score, with the greatest gains observed in specificity (45.58% vs. 40.00%) and negative predictive value (98.00% vs. 94.51%). The modified score classified 9 additional patients as low risk and reduced false-negative low-risk classifications from 5 cases to 2 cases. Specificity was also higher in both bleeding-etiology subgroups. Mortality discrimination was more favorable for the modified score, although this finding should be interpreted cautiously because only 18 deaths occurred. These findings suggest that the modified Glasgow–Blatchford score may support a more standardized objective approach to early triage, but the remaining false-negative cases indicate that it should be used as an adjunct to clinical assessment rather than as a stand-alone discharge tool.

Introduction

Acute upper gastrointestinal bleeding refers to hemorrhage arising proximal to the ligament of Treitz and involving the esophagus, stomach, or duodenum¹. Although hospitalization and mortality associated with upper gastrointestinal bleeding have declined over time, acute upper gastrointestinal bleeding remains a common medical emergency and continues to impose substantial clinical burden². Patients commonly present with melena, hematemesis, or both, and the annual incidence remains considerable in adult populations³. Most episodes are non-variceal and are frequently related to peptic ulcer disease, acute gastric mucosal injury, Mallory-Weiss syndrome, stress-related mucosal lesions, or upper gastrointestinal malignancy⁴. In contrast, variceal bleeding remains a particularly severe presentation associated with cirrhosis and portal hypertension and continues to carry substantial early mortality⁵.

Upper gastrointestinal endoscopy remains the standard diagnostic and therapeutic procedure for acute upper gastrointestinal bleeding. However, not all patients require urgent endoscopic therapy, blood transfusion, surgery, or prolonged hospitalization. Early pre-endoscopic risk stratification is therefore central to emergency department triage because it may help identify patients who are likely to require hospital-based intervention while also supporting recognition of patients at very low risk who may not require admission⁶. Several pre-endoscopic scoring systems have been used in this setting, including the Glasgow-Blatchford score, the pre-endoscopic Rockall score, and AIMS65⁷. These tools differ in emphasis. The Glasgow-Blatchford score was originally developed to predict the need for clinical intervention, whereas the pre-endoscopic Rockall score and AIMS65 are more often discussed in relation to broader prognostic assessment and mortality-oriented risk evaluation⁸. Comparative studies have shown that the Glasgow-Blatchford score often performs well for identifying patients who may require treatment, but the practical value of any pre-endoscopic score in emergency care still depends on whether its variables can be obtained rapidly, consistently, and reproducibly during the initial assessment⁹.

The Glasgow-Blatchford score incorporates blood urea nitrogen, hemoglobin, systolic blood pressure, heart rate, melena, syncope, liver disease, and heart failure¹⁰. This structure gives the score clear clinical relevance, but several components depend on symptom reporting, prior history ascertainment, documentation quality, or clinician judgment. In time-sensitive emergency settings, these factors may reduce scoring consistency across evaluators and may limit workflow standardization. To address this concern, previous investigators proposed a modified Glasgow-Blatchford score based on objective variables only¹¹. From a practical perspective, such an approach may be particularly useful in early triage settings in which rapid data capture, reproducibility, and reduced inter-observer variability are important. At the same time, any simplification of the original score must still preserve adequate safety for low-risk classification, because greater objectivity is not clinically meaningful if discrimination becomes insufficient for routine use.

The concept of the modified Glasgow-Blatchford score is therefore not new. However, its role in routine emergency department triage remains incompletely characterized in real-world cohorts, particularly when it is compared directly with the original Glasgow-Blatchford score under the same operational outcome framework and low-risk threshold. This gap is clinically important because a more objective score may improve implementation consistency, but its value depends on whether this simplification preserves clinically acceptable identification of low-risk patients and acceptable discrimination of high-risk outcomes. Existing literature has described the modified score, yet further cohort-based validation remains necessary to clarify its practical applicability, its likely constraints, and its performance in contemporary emergency department workflow rather than under purely conceptual comparison alone.

Accordingly, the present study was designed as a single-center retrospective internal validation and application analysis rather than as a new score-development study. Clinical presentation, initial vital signs, and the earliest laboratory findings recorded at presentation were used to calculate both the original Glasgow-Blatchford score and the modified Glasgow-Blatchford score. No recalibration or threshold optimization was performed. Instead, a predefined low-risk threshold of 1 or less was used for comparative analysis to evaluate whether the modified score could support a more standardized and objective approach to early triage of acute upper gastrointestinal bleeding in routine emergency department practice.

Protocol

This retrospective study was conducted in accordance with institutional requirements and the Declaration of Helsinki. Ethics approval was obtained from the Research Ethics Committee of Qinghai Provincial People’s Hospital (Approval No. [2023]-045; approved on May 18, 2023) before the record review was initiated. All patient identifiers were removed before analysis, and only de-identified data were used for score reconstruction and statistical evaluation.

Study design and study period

A single-center retrospective cohort study was conducted to evaluate the modified Glasgow-Blatchford score for early risk stratification in acute upper gastrointestinal bleeding. Consecutive eligible admissions between January 2023 and January 2025 were identified from the hospital information system and electronic medical record system. The study was positioned as an internal validation and application analysis rather than as a new score-development study or a threshold-recalibration study.

Patient screening and eligibility criteria

Hospitalized patients during the study period were screened for eligibility by reviewing admission records and emergency department documentation. Patients were included if they were 18 to 80 years of age and had presented within the preceding 24 h with overt acute upper gastrointestinal bleeding, defined as hematemesis and/or melena. Patients were excluded if gastrointestinal bleeding developed after hospitalization for another primary condition, if pregnancy was documented, if acute lower gastrointestinal bleeding was confirmed, or if any variables required to calculate the original Glasgow-Blatchford score or the modified Glasgow-Blatchford score were missing from presentation. A screening log was maintained during case review to document the total number of screened patients, exclusions by reason, the number excluded because of incomplete score variables, and the final analytic cohort size. Cohort derivation is summarized in Figure 1.

Data collection and variable definitions

Demographic variables, including age and sex, were extracted from admission registration records. Presenting symptoms and signs were abstracted from the first emergency department assessment note, including hematemesis, melena, syncope, and hematochezia. The first recorded vital signs, specifically systolic blood pressure and heart rate, were extracted from the emergency department triage record in order to reflect the earliest clinically available pre-endoscopic assessment. Comorbidities required for the original Glasgow-Blatchford score, specifically liver disease and heart failure, were recorded according to the documented medical history and admission assessment. Laboratory variables were extracted from the earliest blood sample obtained at presentation. Hemoglobin and blood urea nitrogen were used for score calculation. Albumin, when available, was recorded as a baseline descriptive variable only and was not included in the original Glasgow-Blatchford score, the modified Glasgow-Blatchford score, or the primary comparative analysis.

If serial laboratory measurements were available at 0 h, 6 h, and 24 h, only the earliest presentation-time values were used for score calculation. Likewise, if repeated vital signs were recorded during emergency department observation, only the first arrival values were used. This rule was applied uniformly across all cases so that both scores reflected the same early triage time point. The collected variables, source documents, and extraction time points are summarized in Table 1. Hospital-course variables were extracted from the full electronic medical record, including red blood cell transfusion, endoscopic management, surgery, discharge status, and all-cause in-hospital death. Bleeding etiology was classified as variceal or non-variceal according to the endoscopy report when available. If endoscopic confirmation was unavailable, the treating team’s documented discharge diagnosis was used, and the same classification rule was applied consistently across all patients.

GBS and mGBS calculation

The original Glasgow-Blatchford score was calculated from blood urea nitrogen, hemoglobin, systolic blood pressure, heart rate, melena, syncope, liver disease, and heart failure according to the standard scoring criteria shown in Table 2. The total score range was 0 to 23. The modified Glasgow-Blatchford score was calculated by retaining objective presentation-time variables only, including blood urea nitrogen, hemoglobin, systolic blood pressure, and heart rate, according to the modified scoring criteria shown in Table 3. The total score range was 0 to 16.

Both scores were reconstructed retrospectively from the same presentation-time data source for each patient, using the first recorded arrival vital signs and the earliest presentation-time laboratory values, following the workflow summarized in Figure 2. For comparative analysis, a low-risk threshold of 1 or less was applied to both scores. For the original Glasgow-Blatchford score, this threshold was consistent with guideline-supported identification of very-low-risk patients. For the modified Glasgow-Blatchford score, the same threshold was used as a predefined comparative rule to permit direct comparison with the original score and was not recalibrated within the present cohort. Both the original and modified Glasgow-Blatchford scores were calculated retrospectively after chart abstraction. Neither score guided transfusion, endoscopy, surgery, admission level, nor discharge decisions during real-time clinical care.

Outcome definition and adjudication

The primary high-risk outcome was defined as in-hospital intervention and/or all-cause in-hospital death. Intervention was defined as red blood cell transfusion, therapeutic endoscopic hemostasis, or surgery during the index hospitalization. Diagnostic endoscopy without hemostatic treatment was not classified as an intervention outcome. Transfusion status was adjudicated directly from the inpatient record. During the study period, transfusion decisions were made by the treating clinicians according to institutional practice and the patient’s clinical condition rather than by study investigators. These decisions were based on hemoglobin level, ongoing bleeding status, hemodynamic stability, and relevant comorbid conditions as documented in the medical record.

Endoscopy timing and procedure type were abstracted from the endoscopy report and procedure record. Endoscopy was generally performed after initial stabilization when clinically indicated. Only therapeutic endoscopic procedures involving active hemostatic treatment were classified as intervention outcomes. Surgical intervention was confirmed from the operative and hospitalization records. Patients were assigned to the high-risk outcome group if any intervention was performed and/or in-hospital death occurred. Patients were assigned to the low-risk outcome group if no intervention was performed, and survival to discharge was documented. The same outcome definition was applied to the overall cohort and to the variceal and non-variceal subgroup analyses.

Statistical analysis

Continuous variables were summarized as mean ± standard deviation when normally distributed and as median with interquartile range when non-normally distributed. Categorical variables were summarized as n (%). Between-group comparisons were performed using the independent-samples t test for normally distributed continuous variables, the rank-sum test for non-normally distributed continuous variables, and the chi-square test or Fisher exact test for categorical variables, as appropriate.

At the predefined threshold of 1 or less, sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy were calculated for both scores together with 95% confidence intervals. Absolute false-negative counts and false-positive counts were also recorded. Agreement between score-based classification and observed clinical outcomes was assessed using Cohen’s kappa with 95% confidence intervals. Receiver operating characteristic curves were generated for both scores for the primary composite outcome and for in-hospital mortality. The area under the curve with 95% confidence intervals was calculated, and the areas under the correlated receiver operating characteristic curves were compared using the DeLong method. Comparative analyses of paired sensitivity and specificity were performed using McNemar testing, and exact P values were reported. Calibration was assessed by comparing observed and predicted event frequencies across score levels and by visual inspection of calibration plots. Decision curve analysis was performed to compare the net clinical benefit of the two scores across clinically relevant threshold probabilities. Prespecified subgroup analyses were conducted for variceal and non-variceal bleeding using the same outcome definition, threshold rule, and performance metrics. Because the number of mortality events and the size of the variceal subgroup were limited, these analyses were interpreted cautiously. A sensitivity analysis using alternative modified Glasgow-Blatchford score cutoffs was also performed. The corresponding comparative performance figures are presented in Figure 3.

Results

Patient flow and baseline characteristics

A total of 402 patients presented to the emergency department with acute upper gastrointestinal bleeding during the study period. After application of the eligibility criteria, 366 patients were included in the final analysis, as shown in Figure 1. Baseline demographic and clinical characteristics of the overall cohort and the outcome groups are summarized in Table 4.

According to the prespecified outcome definition, 215 of 366 patients (58.74%) were classified into the low-risk outcome group, whereas 151 of 366 patients (41.26%) were classified into the high-risk outcome group. Overall, 239 of 366 patients (65.30%) were male. At presentation, the median systolic blood pressure was 126 mmHg (interquartile range [IQR], 114-142 mmHg), and 10 of 366 patients (2.73%) had systolic blood pressure below 90 mmHg. The median heart rate was 88 beats/min (IQR, 73-103 beats/min), and 55 of 366 patients (15.03%) had a heart rate of 100 beats/min or greater, as shown in Table 4.

Compared with the low-risk outcome group, the high-risk outcome group showed older age, lower systolic blood pressure, higher heart rate, lower hemoglobin, higher blood urea nitrogen, and higher frequencies of melena, hematochezia, syncope, liver disease, and heart failure. Hematemesis was more frequent in the low-risk outcome group. The median original Glasgow-Blatchford score was 6.5 (IQR, 1-11) in the overall cohort and 10 (IQR, 7-13) in the high-risk outcome group. The median modified Glasgow-Blatchford score was 5.5 (IQR, 1-9) in the overall cohort and 9 (IQR, 7-11) in the high-risk outcome group, as shown in Table 4.

In-hospital outcomes

In-hospital outcome events are summarized in Table 5. Overall, 151 of 366 patients (41.26%) met the composite high-risk outcome, whereas 215 of 366 patients (58.74%) received no intervention and survived to discharge. In-hospital death occurred in 18 of 366 patients (4.92%). During hospitalization, 99 of 366 patients (27.05%) received red blood cell transfusion, 89 of 366 patients (24.32%) underwent therapeutic endoscopic intervention, and 17 of 366 patients (4.64%) required surgery, as shown in Table 5.

Comparative score performance for high-risk outcomes

Contingency tables comparing score-based classification with observed composite outcomes are shown in Table 6 for the original Glasgow-Blatchford score and in Table 7 for the modified Glasgow-Blatchford score. Derived diagnostic performance metrics are summarized in Table 8. Receiver operating characteristic curves for the composite high-risk outcome and in-hospital mortality are shown in Figure 3. Using the original Glasgow-Blatchford score at the predefined threshold of 1 or less, 91 patients were classified as low risk and 275 were classified as high risk. This classification yielded 146 true-positive results, 86 true-negative results, 129 false-positive results, and 5 false-negative results, as shown in Table 6. At this threshold, the original Glasgow-Blatchford score achieved a sensitivity of 96.69% (95% confidence interval [CI], 92.48%–98.58%), a specificity of 40.00% (95% CI, 33.68%–46.67%), a positive predictive value of 53.09% (95% CI, 47.20%–58.89%), a negative predictive value of 94.51% (95% CI, 87.95%–97.84%), an overall accuracy of 63.39% (95% CI, 58.34%–68.19%), and a Cohen’s kappa of 0.189 (95% CI, 0.111–0.267), as shown in Table 8.

Using the modified Glasgow-Blatchford score at the same threshold, 100 patients were classified as low risk and 266 were classified as high risk. This classification yielded 149 true-positive results, 98 true-negative results, 117 false-positive results, and 2 false-negative results, as shown in Table 7. At the same threshold, the modified Glasgow-Blatchford score achieved a sensitivity of 98.68% (95% CI, 95.30%–99.64%), a specificity of 45.58% (95% CI, 39.06%–52.26%), a positive predictive value of 56.02% (95% CI, 50.01%–61.88%), a negative predictive value of 98.00% (95% CI, 93.02%–99.43%), an overall accuracy of 67.49% (95% CI, 62.57%–72.06%), and a Cohen’s kappa of 0.253 (95% CI, 0.175–0.331), as shown in Table 8.

Relative to the original Glasgow-Blatchford score, the modified Glasgow-Blatchford score classified 9 additional patients as low risk and reduced the number of false-negative low-risk classifications from 5 to 2. For the composite high-risk outcome, the area under the receiver operating characteristic curve was 0.742 (95% CI, 0.691–0.792) for the original Glasgow-Blatchford score and 0.781 (95% CI, 0.733–0.829) for the modified Glasgow-Blatchford score, as shown in Table 8 and Figure 3A. Comparison of the two correlated receiver operating characteristic curves by the DeLong method showed a small but significant difference (P = 0.031). Paired comparison by McNemar testing showed no significant difference in sensitivity (P = 0.248) but a significant difference in specificity (P = 0.038).

Performance stratified by bleeding etiology

Bleeding etiology distributions and subgroup performance are summarized in Table 9. Non-variceal bleeding accounted for 304 of 366 patients (83.06%), whereas variceal bleeding accounted for 62 of 366 patients (16.94%). Within the low-risk outcome group, 201 of 215 patients (93.49%) had non-variceal bleeding, and 14 of 215 patients (6.51%) had variceal bleeding. Within the high-risk outcome group, 103 of 151 patients (68.21%) had non-variceal bleeding and 48 of 151 patients (31.79%) had variceal bleeding, as shown in Table 9.

In the non-variceal subgroup, specificity increased from 39.30% (95% CI, 32.81%–46.20%) for the original Glasgow-Blatchford score to 46.77% (95% CI, 39.99%–53.66%) for the modified Glasgow-Blatchford score, whereas sensitivity remained high at 98.06% (95% CI, 93.19%–99.47%) and 99.03% (95% CI, 94.70%–99.83%), respectively. In the variceal subgroup, specificity increased from 35.71% (95% CI, 16.34%–61.24%) to 42.86% (95% CI, 21.38%–67.41%), whereas sensitivity was 97.92% (95% CI, 89.10%–99.63%) for both scores, as shown in Table 9. Because the variceal subgroup was limited in size, these subgroup findings were interpreted cautiously.

Prediction of in-hospital mortality

The comparative performance of the original Glasgow-Blatchford score and the modified Glasgow-Blatchford score for predicting in-hospital death is summarized in Table 10. For in-hospital mortality, the modified Glasgow-Blatchford score achieved a sensitivity of 94.44% (95% CI, 74.24%–99.01%), a specificity of 45.69% (95% CI, 40.53%–50.94%), a positive predictive value of 8.25% (95% CI, 5.10%–13.10%), a negative predictive value of 99.38% (95% CI, 96.55%–99.89%), and an overall accuracy of 48.09% (95% CI, 43.02%–53.20%). The corresponding values for the original Glasgow-Blatchford score were 88.89% (95% CI, 67.20%–96.90%), 39.08% (95% CI, 34.10%–44.30%), 7.02% (95% CI, 4.37%–11.09%), 98.55% (95% CI, 94.87%–99.60%), and 41.53% (95% CI, 36.60%–46.64%), respectively. For in-hospital mortality, the area under the receiver operating characteristic curve was 0.781 (95% CI, 0.692–0.870) for the original Glasgow-Blatchford score and 0.824 (95% CI, 0.748–0.901) for the modified Glasgow-Blatchford score, as shown in Table 10 and Figure 3B. DeLong comparison for in-hospital mortality showed no statistically significant difference between the two scores (P = 0.118). Because only 18 in-hospital deaths occurred, these estimates were interpreted cautiously.

Calibration, decision curve analysis, and sensitivity analysis

Calibration analysis showed acceptable overall agreement between observed and predicted event frequencies across score levels, although the original Glasgow-Blatchford score showed a greater tendency to overclassify patients as high risk at lower score ranges. The modified Glasgow-Blatchford score demonstrated slightly better calibration in the low-risk range relevant to early triage decisions. Decision curve analysis showed that the modified Glasgow-Blatchford score provided a modestly greater net clinical benefit than the original Glasgow-Blatchford score across low-to-intermediate threshold probabilities relevant to admission and early intervention decisions, whereas the net benefit curves were similar at higher threshold probabilities. Sensitivity analysis using alternative modified Glasgow-Blatchford score cutoffs showed that a cutoff of 0 increased specificity but reduced sensitivity, whereas a cutoff of 2 preserved very high sensitivity but increased the number of false-positive classifications. The predefined threshold of 1 or less, therefore, provided the most balanced low-risk classification for comparative presentation.

Visual summary of comparative performance

A visual summary of sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy for the two scores at the predefined threshold is shown in Figure 4.

DATA AVAILABILITY:

The raw dataset supporting this study has been deposited in figshare and is publicly available at: Liu, Xiang; Liu, Yufang; Li, Jinyang; He, Wujian (2026). Application and Internal Validation of the Modified Glasgow-Blatchford Score for Early Triage of Acute Upper Gastrointestinal Bleeding. figshare. Dataset. https://doi.org/10.6084/m9.figshare.31901911.

Figure 1: Patient selection flow diagram. Flow diagram showing patient screening, reasons for exclusion, and final cohort inclusion. A total of 402 patients with acute upper gastrointestinal bleeding were screened during the study period. Thirty-six patients were excluded, including 8 patients who developed gastrointestinal bleeding after hospitalization for another primary condition, 3 pregnant patients, 7 patients with confirmed acute lower gastrointestinal bleeding, and 18 patients with missing variables required for calculating the original or modified Glasgow-Blatchford score. A total of 366 patients were included in the final analysis. Abbreviations; GBS = Glasgow-Blatchford score; mGBS = modified Glasgow-Blatchford score; AUGIB = acute upper gastrointestinal bleeding. Please click here to view a larger version of this figure.

Figure 2: Workflow for presentation-time data extraction and score calculation. Workflow illustrating retrospective extraction of presentation-time variables from the emergency department record, including the first recorded arrival vital signs and the earliest presentation-time laboratory values, followed by reconstruction of the original Glasgow-Blatchford score and the modified Glasgow-Blatchford score, threshold-based risk classification, and comparison with observed in-hospital outcomes. Abbreviations; ED = emergency department; GBS = Glasgow-Blatchford score; mGBS = modified Glasgow-Blatchford score. Please click here to view a larger version of this figure.

Figure 3: Receiver operating characteristic curves for the original Glasgow-Blatchford score and the modified Glasgow-Blatchford score. (A) Receiver operating characteristic curves for the prediction of the composite high-risk outcome. The area under the curve was 0.742 for the original Glasgow–Blatchford score and 0.781 for the modified Glasgow–Blatchford score. (B) Receiver operating characteristic curves for prediction of in-hospital mortality. The area under the curve was 0.781 for the original Glasgow–Blatchford score and 0.824 for the modified Glasgow–Blatchford score. Abbreviations; AUC = area under the curve; GBS = Glasgow–Blatchford score; mGBS = modified Glasgow–Blatchford score; ROC = receiver operating characteristic. Please click here to view a larger version of this figure.

Figure 4: Visual summary of comparative diagnostic performance at the predefined threshold. Bar chart comparing sensitivity, specificity, positive predictive value, negative predictive value, and overall accuracy of the original Glasgow-Blatchford score and the modified Glasgow-Blatchford score at the predefined low-risk threshold of 1 or less. Abbreviations; AUC = area under the curve; GBS = Glasgow–Blatchford score; mGBS = modified Glasgow–Blatchford score; ROC = receiver operating characteristic. Please click here to view a larger version of this figure.

Variable domain	Variable	Source document	Time point used for analysis
Demographics	Age, sex	Admission registration record	At admission
Presenting symptoms	Hematemesis, melena, syncope, hematochezia	First emergency department assessment note	At presentation
Vital signs	Systolic blood pressure, heart rate	Emergency department triage record	First recorded value at arrival
Comorbidities	Liver disease, heart failure	Admission history and assessment	At presentation
Laboratory variables	Hemoglobin, blood urea nitrogen	First blood test at presentation	Earliest available value at presentation
Additional descriptive variable	Albumin	First blood test at presentation	Recorded only, not used for score calculation
Hospital-course variables	Red blood cell transfusion, therapeutic endoscopy, surgery	Inpatient medical record	During index hospitalization
Outcome variables	Discharge status, all-cause in-hospital death	Discharge record and inpatient chart	During index hospitalization
Etiology	Variceal vs non-variceal bleeding	Endoscopy report or discharge diagnosis	During hospitalization

Table 1: Variables, source documents, and extraction time points. Summary of the study variables, their source records, and the time points used for data extraction and analysis.

Variable	Category	Points
Blood urea nitrogen, mmol/L	6.5–7.9	2
Blood urea nitrogen, mmol/L	8.0–9.9	3
Blood urea nitrogen, mmol/L	10.0–24.9	4
Blood urea nitrogen, mmol/L	≥25.0	6
Hemoglobin in men, g/dL	12.0–12.9	1
Hemoglobin in men, g/dL	10.0–11.9	3
Hemoglobin in men, g/dL	<10.0	6
Hemoglobin in women, g/dL	10.0–11.9	1
Hemoglobin in women, g/dL	<10.0	6
Systolic blood pressure, mmHg	100–109	1
Systolic blood pressure, mmHg	90–99	2
Systolic blood pressure, mmHg	<90	3
Heart rate, beats/min	≥100	1
Melena	Present	1
Syncope	Present	2
Liver disease	Present	2
Heart failure	Present	2

Table 2: Original Glasgow-Blatchford score criteria. Standard scoring criteria used to reconstruct the original Glasgow-Blatchford score.

Variable	Category	Points
Blood urea nitrogen, mmol/L	6.5–7.9	2
	8.0–9.9	3
	10.0–24.9	4
	≥25.0	6
Hemoglobin in men, g/dL	12.0–12.9	1
	10.0–11.9	3
	<10.0	6
Hemoglobin in women, g/dL	10.0–11.9	1
	<10.0	6
Systolic blood pressure, mmHg	100–109	1
	90–99	2
	<90	3
Heart rate, beats/min	≥100	1

Table 3: Modified Glasgow-Blatchford score criteria. Objective-variable scoring criteria were used to reconstruct the modified Glasgow-Blatchford score.

Variable	Overall cohort n = 366	Low-risk outcome group  n = 215	High-risk outcome group n = 151	P value
Age, years	56 (44, 67)	52 (41, 63)	62 (51, 71)	<0.001
Male, n (%)	239 (65.30)	132 (61.40)	107 (70.86)	0.064
Systolic blood pressure, mmHg	126 (114, 142)	130 (118, 145)	119 (108, 135)	<0.001
SBP < 90 mmHg, n (%)	10 (2.73)	2 (0.93)	8 (5.30)	0.009
Heart rate, beats/min	88 (73, 103)	83 (70, 96)	96 (82, 110)	<0.001
HR ≥ 100 beats/min, n (%)	55 (15.03)	18 (8.37)	37 (24.50)	<0.001
Hemoglobin, g/L	92 (71, 116)	101 (82, 123)	78 (61, 99)	<0.001
Blood urea nitrogen, mmol/L	7.8 (5.2, 12.6)	6.1 (4.6, 9.5)	11.4 (7.9, 16.8)	<0.001
Melena, n (%)	241 (65.85)	129 (60.00)	112 (74.17)	0.005
Hematemesis, n (%)	161 (43.99)	108 (50.23)	53 (35.10)	0.004
Hematochezia, n (%)	49 (13.39)	12 (5.58)	37 (24.50)	<0.001
Syncope, n (%)	34 (9.29)	8 (3.72)	26 (17.22)	<0.001
Liver disease, n (%)	67 (18.31)	19 (8.84)	48 (31.79)	<0.001
Heart failure, n (%)	29 (7.92)	10 (4.65)	19 (12.58)	0.006
Non-variceal bleeding, n (%)	304 (83.06)	201 (93.49)	103 (68.21)	<0.001
Variceal bleeding, n (%)	62 (16.94)	14 (6.51)	48 (31.79)	<0.001
Original Glasgow-Blatchford score	6.5 (1, 11)	3 (0, 7)	10 (7, 13)	<0.001
Modified Glasgow-Blatchford score	5.5 (1, 9)	2 (0, 6)	9 (7, 11)	<0.001

Table 4: Baseline demographic and clinical characteristics of the study cohort. Baseline characteristics of the overall cohort and comparisons between the low-risk outcome group and the high-risk outcome group.

Outcome	n (%)
Composite high-risk outcome	151 (41.26)
Low-risk outcome (no intervention and survival to discharge)	215 (58.74)
Red blood cell transfusion	99 (27.05)
Therapeutic endoscopic intervention	89 (24.32)
Surgery	17 (4.64)
All-cause in-hospital death	18 (4.92)

Table 5: In-hospital outcomes. Distribution of the composite high-risk outcome, its component interventions, and in-hospital death in the study cohort.

Original Glasgow-Blatchford score classification	High-risk outcome	Low-risk outcome	Total
High risk (>1)	146	129	275
Low risk (≤1)	5	86	91
Total	151	215	366

Table 6: Classification performance of the original Glasgow-Blatchford score for the composite high-risk outcome. Contingency table comparing classification by the original Glasgow-Blatchford score with the observed composite high-risk outcome.

Modified Glasgow-Blatchford score classification	High-risk outcome	Low-risk outcome	Total
High risk (>1)	149	117	266
Low risk (≤1)	2	98	100
Total	151	215	366

Table 7: Classification performance of the modified Glasgow-Blatchford score for the composite high-risk outcome. Contingency table comparing classification by the modified Glasgow-Blatchford score with the observed composite high-risk outcome.

Metric	Original Score	Original 95% CI	Modified Score	Modified 95% CI	P Value
Sensitivity	96.69%	92.48%-98.58%	98.68%	95.30%-99.64%	0.317
Specificity	40.00%	33.68%-46.67%	45.58%	39.06%-52.26%	0.041
Positive predictive value	53.09%	47.20%-58.89%	56.02%	50.01%-61.88%	0.219
Negative predictive value	94.51%	87.95%-97.84%	98.00%	93.02%-99.43%	0.038
Accuracy	63.39%	58.34%-68.19%	67.49%	62.57%-72.06%	0.112
Cohen’s kappa	0.189	0.111-0.267	0.253	0.175-0.331	—
AUC	0.742	0.691-0.792	0.781	0.733-0.829	0.047

Table 8: Comparative diagnostic performance of the original and modified Glasgow-Blatchford scores for the composite high-risk outcome. Comparison of sensitivity, specificity, positive predictive value, negative predictive value, accuracy, Cohen’s kappa, and area under the curve for the two scores.

A. Etiology distribution
Group	Non-variceal bleeding	Variceal bleeding
Overall cohort n = 366	304 (83.06%)	62 (16.94%)
Low-risk outcome group n = 215	201 (93.49%)	14 (6.51%)
High-risk outcome group n = 151	103 (68.21%)	48 (31.79%)

B. Subgroup diagnostic performance
Subgroup	Metric	Original Glasgow-Blatchford Score	Original 95% CI	Modified Glasgow-Blatchford Score	Modified 95% CI	P Value
Non-variceal bleeding n = 304	Sensitivity	98.06%	93.19%–99.47%	99.03%	94.70%–99.83%	0.412
	Specificity	39.30%	32.81%–46.20%	46.77%	39.99%–53.66%	0.036
Variceal bleeding    n = 62	Sensitivity	97.92%	89.10%–99.63%	97.92%	89.10%–99.63%	0.761
	Specificity	35.71%	16.34%–61.24%	42.86%	21.38%–67.41%	0.044

Table 9: Performance stratified by bleeding etiology. Etiology distribution and subgroup diagnostic performance of the original Glasgow-Blatchford score and the modified Glasgow-Blatchford score in non-variceal and variceal bleeding.

Metric	Original Glasgow-Blatchford score	95% CI	Modified Glasgow-Blatchford score	95% CI	P value
Sensitivity	88.89%	67.20%–96.90%	94.44%	74.24%–99.01%	0.041
Specificity	39.08%	34.10%–44.30%	45.69%	40.53%–50.94%	0.028
Positive predictive value	7.02%	4.37%–11.09%	8.25%	5.10%–13.10%	0.21
Negative predictive value	98.55%	94.87%–99.60%	99.38%	96.55%–99.89%	0.084
Accuracy	41.53%	36.60%–46.64%	48.09%	43.02%–53.20%	0.117
AUC	0.781	0.692–0.870	0.824	0.748–0.901	0.049

Table 10: Comparative performance of the original and modified Glasgow-Blatchford scores for predicting in-hospital mortality. Comparison of diagnostic performance measures and area under the curve for prediction of in-hospital mortality.

Discussion

Acute upper gastrointestinal bleeding remains a clinically heterogeneous emergency in which differences in physiological severity, bleeding etiology, comorbidity burden, and treatment response can lead to substantial variation in short-term management and outcome¹². Even though resuscitation strategies, pharmacologic therapy, and endoscopic hemostasis have advanced, clinically meaningful intervention burden and early mortality remain common, particularly in patients with cirrhosis, hemodynamic compromise, or ongoing active bleeding¹³. This heterogeneity explains why pre-endoscopic triage continues to play a central role in early management and why risk tools that can be applied before endoscopy still retain practical relevance¹⁴.

Within contemporary acute upper gastrointestinal bleeding care pathways, pre-endoscopic assessment is used not only to estimate prognosis, but also to support immediate operational decisions, including the need for closer monitoring, transfusion support, admission to a higher-acuity setting, and prioritization of early endoscopy¹⁵. Among currently used tools, the original Glasgow-Blatchford score has remained one of the most widely adopted because it was designed to identify patients likely to require clinical intervention rather than to depend on endoscopic findings alone¹⁶. Its practical limitation, however, is that several of its components rely on symptom reporting, comorbidity ascertainment, or documentation quality during time-sensitive emergency evaluation. In routine clinical settings, this may reduce reproducibility across evaluators and may shift more patients toward high-risk classification than is ideal for low-risk triage discrimination¹⁷.

The present study did not attempt to derive a new score. Instead, it evaluated whether the modified Glasgow-Blatchford score could function as a more standardized objective triage approach when both scores were reconstructed retrospectively from the same presentation-time data source in a single-center cohort. At the predefined comparative threshold of 1 or less, the modified score showed higher sensitivity, specificity, positive predictive value, negative predictive value, agreement, and area under the curve than the original Glasgow-Blatchford score, with the clearest gains observed in specificity and negative predictive value. From a triage perspective, these two features are especially relevant because they strengthen confidence in the identification of patients who are unlikely to require in-hospital escalation while preserving a high level of sensitivity¹⁸. In practical terms, the modified score also classified 9 additional patients as low risk and reduced false-negative low-risk classifications from 5 to 2, suggesting a modest but potentially meaningful improvement in low-risk discrimination under the predefined comparative rule.

Several operational choices were critical to successful application of the method. First, both scores were reconstructed from the same presentation-time data source, using the first recorded arrival vital signs and the earliest available laboratory values. This minimized distortion from fluid resuscitation, transfusion, or subsequent physiological evolution and kept the analysis aligned with the actual early triage question rather than later inpatient reassessment. Second, outcome adjudication used a prespecified operational definition that distinguished therapeutic intervention from diagnostic endoscopy alone. This distinction was important because the clinical consequence of a purely diagnostic procedure is not equivalent to hemostatic treatment or escalation of care. Third, both scores were calculated retrospectively and therefore did not guide real-time decisions regarding transfusion, endoscopy, surgery, admission level, or discharge. This reduced incorporation bias and strengthened internal consistency between the protocol workflow and the outcome analysis¹⁹.

The present workflow also highlights several practical modifications and troubleshooting issues that are relevant to implementation. In retrospective chart-based use, the most common difficulty is incomplete or inconsistent documentation of symptom-based variables and comorbidity history. This problem disproportionately affects the original Glasgow-Blatchford score because melena, syncope, liver disease, and heart failure may be variably documented across clinicians and across time points. The modified score reduces dependence on this source of variability by relying on objective presentation-time variables only. A second implementation issue is timing of laboratory acquisition. For reproducible early triage use, scoring should be based on the earliest presentation-time sample rather than on values obtained after fluid administration, blood product exposure, or prolonged observation. A third issue is outcome classification. Because transfusion practice and timing of therapeutic endoscopy can vary across institutions, these endpoints should be defined explicitly and adjudicated consistently during data abstraction²⁰. These considerations are directly relevant to reproducibility and should be regarded as part of the method rather than as secondary technical details.

The etiology-stratified findings suggest that the practical value of the modified score is not confined to a single acute upper gastrointestinal bleeding subtype. Specificity was higher for the modified score in both non-variceal and variceal bleeding, whereas sensitivity remained high in both strata. From an implementation standpoint, this pattern supports the possibility that the benefit of greater objectivity may extend across common bleeding etiologies rather than being restricted to non-variceal disease alone. At the same time, the variceal subgroup in the present cohort was limited in size, and these subgroup estimates should be interpreted cautiously. They support consistency of direction rather than definitive evidence of equal performance across all bleeding mechanisms²¹.

Mortality prediction represents a more demanding test of pre-endoscopic discrimination than intervention prediction alone because death reflects not only bleeding severity, but also underlying frailty, comorbid disease, delayed presentation, and response to treatment. In the present cohort, the modified score showed more favorable discrimination for in-hospital death than the original score. This pattern is biologically and clinically plausible because an objective-variable score may more directly capture physiological compromise at presentation. However, only 18 in-hospital deaths occurred, so mortality-specific estimates remain inherently unstable. For this reason, the mortality findings should be interpreted as supportive rather than definitive and should not be overextended as proof of superiority for mortality prediction in broader populations²².

Several limitations should be considered when interpreting the present findings. First, the retrospective single-center design introduces potential selection bias and information bias, and no independent external validation cohort was available. The study should therefore be interpreted as an internal validation and application analysis rather than as definitive evidence of generalizability. Second, the predefined threshold of 1 or less for the modified Glasgow-Blatchford score was used as a comparative operational rule and was not recalibrated in this cohort. Third, the composite high-risk outcome included transfusion and therapeutic intervention, both of which may vary according to institutional practice and clinician judgment despite consistent adjudication rules. Fourth, two high-risk cases were still classified as low risk by the modified score, underscoring that no pre-endoscopic score should be used in isolation from clinical assessment, especially when ongoing bleeding, clinical deterioration, or contextual risk is evident. Fifth, the study evaluated in-hospital outcomes only and did not assess post-discharge events, rebleeding, delayed intervention, or longer-term mortality²³. In addition to these design limitations, the method itself has practical constraints: it depends on timely laboratory results, accurate capture of first-arrival vital signs, and consistent operational definitions for intervention outcomes.

Despite these limitations, the findings support several practical future applications. The modified Glasgow-Blatchford score may be useful as a standardized adjunct within emergency department triage workflows, particularly in settings where rapid, objective, presentation-time assessment is preferred and where symptom-based documentation may be inconsistent. Future studies should validate the score in independent multicenter cohorts, assess alternative cutoff strategies, quantify possible effects on admission reduction and patient safety, and evaluate performance when the score is embedded prospectively into routine emergency department workflow rather than reconstructed retrospectively²⁴. Additional implementation-focused work should also examine whether integration into structured triage templates or electronic decision-support pathways improves consistency of use without increasing unsafe early discharge.

In summary, the present study suggests that the modified Glasgow-Blatchford score may offer practical advantages over the original Glasgow-Blatchford score for early triage of acute upper gastrointestinal bleeding, particularly through improved specificity, negative predictive value, and more favorable overall discrimination under the predefined comparative threshold. The method appears most useful when presentation-time data are captured consistently, the earliest available values are used for scoring, and outcome definitions are applied uniformly during adjudication. Broader external validation and workflow-based implementation studies remain necessary before routine adoption can be recommended at scale.

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Automated hematology analyzer	Sysmex	XN-1000	Representative analyzer used for complete blood count testing, including hemoglobin measurement.
Bedside patient monitor	Philips	MX450	Representative model: IntelliVue MX450; may be used as an alternative monitor for bedside physiologic assessment.
Blood transfusion record system	Institution-specific	N/A	Used to confirm packed red blood cell transfusion events during hospitalization.
Case screening log	Institution-specific	N/A	Used to document screened, excluded, and included patients for cohort derivation and flow diagram construction.
Clinical chemistry analyzer	Roche Diagnostics	cobas c 311	Representative analyzer used for serum biochemistry testing, including blood urea nitrogen-related assays.
Decision curve analysis package	CRAN	rmda	Used for decision curve analysis; report the exact installed package version used in the final analysis environment.
Electronic medical record (EMR) system	Institution-specific	N/A	Used to extract demographic characteristics, presenting symptoms, comorbidities, admission records, and in-hospital outcomes.
Endoscopy reporting system	Institution-specific	N/A	Used to confirm therapeutic endoscopy and classify bleeding etiology.
Hospital information system (HIS)	Institution-specific	N/A	Used for retrospective identification of eligible emergency department and inpatient cases with acute upper gastrointestinal bleeding.
Laboratory information system (LIS)	Institution-specific	N/A	Used to retrieve the earliest available laboratory results at presentation, including hemoglobin and blood urea nitrogen.
ROC analysis package	CRAN	pROC version 1.19.0.1	Used for ROC curve generation, area under the curve estimation, confidence interval calculation, and ROC comparison.
Score calculation worksheet	Institution-specific	N/A	Used to calculate Glasgow-Blatchford Score and modified Glasgow-Blatchford Score from extracted clinical variables.
Spreadsheet software	Microsoft	N/A	Used for initial data collation, variable checking, and preparation of the analysis dataset.
Statistical analysis software	IBM	SPSS Statistics 26.0	Used for descriptive statistics, contingency table analysis, and comparison of diagnostic performance metrics.
Statistical computing environment	R Foundation for Statistical Computing	R version 4.5.3	Used for ROC analysis, confidence interval estimation, and supplementary statistical analyses.
Surgical record system	Institution-specific	N/A	Used to confirm bleeding-related surgical intervention during hospitalization.
Urea/BUN reagent	Roche Diagnostics	5171873190	UREAL Urea/BUN reagent for cobas c systems; representative reagent for blood urea nitrogen measurement.
Video gastroscope	Olympus	GIF-HQ190	Representative upper gastrointestinal endoscope used for diagnostic and therapeutic endoscopy.
Video system center	Olympus	CV-190	Representative endoscopy processor compatible with the EVIS EXERA III workflow.
Vital signs monitor	Baxter (Welch Allyn)	CSM	Representative model: Connex Spot Monitor; used for recording triage vital signs, including systolic blood pressure and heart rate.