Validation of the C-WATCH score for predicting severity in patients with upper gastrointestinal bleeding at a tertiary public hospital in Lima, Peru Authors: Jorge Espinoza-Ríos, Renatto Ruiz-Cortez, Brendy Santiago, Jorge Huerta-Mercado Tenorio Published: 2026 ABSTRACT AbstractBackgroundUpper gastrointestinal bleeding (UGIB) remains one of the most common emergencies in gastroenterology, with mortality rates reaching 14% in some settings. Despite advances in management, mortality has remained relatively unchanged. Several risk stratification tools, including Glasgow-Blatchford, AIMS65, and Rockall scores, are used to predict adverse outcomes. More recently, the C-WATCH score has been proposed as a simplified alternative. We aimed to validate the ability of the C-WATCH score to predict severity in patients with UGIB in a tertiary public hospital in Lima, Peru.MethodsWe conducted a prospective diagnostic accuracy study comparing the discriminative performance of C-WATCH, AIMS65, Glasgow-Blatchford, and Rockall (pre- and post-endoscopic) scores. The outcomes assessed included 30-day mortality, the need for transfusion of ≥ 2 units of packed red blood cells and rebleeding. Discrimination was assessed using area under the receiver operating characteristic curve (AUROC) and pairwise comparisons were performed using the DeLong test.ResultsA total of 235 patients were included; 59.6% (n = 140) were male, with a mean age of 61.0 ± 16.5 years. The thirty-day mortality rate was 4.7% (n = 11). The most frequent etiologies were gastric ulcers (31.1%, n = 73), esophageal varices (25.1%, n = 59), and duodenal ulcers (16.6%, n = 39). For mortality prediction, the AUROC was highest for C-WATCH (0.83, 95% CI 0.72–0.94), followed by AIMS65 (0.81, 95% CI 0.70–0.93), post-endoscopic Rockall (0.80, 95% CI 0.72–0.88), Glasgow-Blatchford (0.69, 95% CI 0.53–0.85) and pre-endoscopic Rockall (0.63, 95% CI 0.47–0.79). However, pairwise comparisons using the DeLong test showed no statistically significant differences between C-WATCH and comparator scores after Bonferroni correction. Similar findings were obtained for predicting transfusion ≥ 2 units. All scores demonstrated poor performance for rebleeding prediction (AUROC < 0.58) and no statistically significant differences.ConclusionsC-WATCH score demonstrated comparable performance for predicting 30-day mortality and transfusion requirements compared with AIMS65, Glasgow-Blatchford, and Rockall scores. All scores showed limited ability to predict rebleeding.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12876-026-04963-7.Keywords: Gastrointestinal hemorrhage, Risk assessment, Prognosis, Mortality, Receiver operating characteristic curve. Introduction Upper gastrointestinal bleeding (UGIB) remains one of the most common emergencies in gastroenterology worldwide, with reported mortality rates ranging from 3% to 14%, depending on etiology, patient age, and comorbidities. In the United States alone, nearly half a million hospital admissions annually are attributed to this condition [1, 2]. In Peru, a prospective study conducted in 2016 at Hospital Cayetano Heredia reported a mortality rate of 7.8% and a rebleeding rate of 3.9% [3]. These findings are consistent with previous institutional reports documenting mortality rates between 7.7% and 12.9% and rebleeding rates between 3.8% and 14.8% [4, 5]. Despite advances in endoscopic and pharmacological therapies, the overall mortality associated with UGIB has not declined substantially. This trend may be explained by the increasing age and comorbidity burden of affected patients [6–8]. Therefore, early risk stratification is essential to identify patients at high risk of adverse outcomes, such as mortality, rebleeding, and the need for blood transfusion, allowing for optimized resource allocation and timely intervention [9]. Current clinical practice guidelines from the American College of Gastroenterology (ACG), the European Society of Gastrointestinal Endoscopy (ESGE), and the national Peruvian guidelines recommend the use of validated risk stratification tools as part of appropriate triage and early clinical decision-making [2, 10, 11]. Several prognostic scores have been developed to predict the clinical severity in UGIB, including the Glasgow-Blatchford score, Rockall score, AIMS65, and more recently, the C-WATCH score [9, 12]. The Glasgow-Blatchford score incorporates clinical and laboratory parameters and ranges from 0 to 23 points [11, 13]. The Rockall score evaluates age, hemodynamic status, comorbidities, and endoscopic findings, with both pre-endoscopic and post-endoscopic components; the post-endoscopic version (0–11 points) provides improved risk discrimination [4, 11]. The AIMS65 score (0–5 points) includes serum albumin, international normalized ratio (INR), altered mental status, systolic blood pressure, and age ≥ 65 years [3, 14]. More recently, the C-WATCH score (0–8 points) was developed using admission laboratory parameters, including C-reactive protein, leukocyte count, alanine aminotransferase (ALT), platelet count, creatinine, and hemoglobin. The original study, conducted in Cologne, Germany, included 586 patients with UGIB and used multivariable logistic regression to predict rebleeding, the need for embolization or surgery, and 30-day mortality. A score ≥ 2 was used to identify high-risk patients, with an area under the receiver operating characteristic curve (AUROC) of 0.73 [12]. A subsequent retrospective study in 2022 involving 252 patients reported AUROC values of 0.68 for mortality, 0.66 for complications, and 0.69 for the need for endoscopic intervention [15]. Given the emergence of C-WATCH as a novel prognostic tool, external validation in different clinical settings is necessary. Therefore, the primary aim of this study was to validate the C-WATCH score for predicting 30-day mortality in patients with variceal and non-variceal UGIB at a tertiary public hospital in Lima, Peru. Secondary objectives included determining the optimal discrimination thresholds for the C-WATCH, Glasgow-Blatchford, Rockall, and AIMS65 scores; evaluating their performance in predicting severity outcomes (mortality, rebleeding, and transfusion requirements); and calculating the sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for each outcome. Methods Study design and setting This was a prospective, descriptive diagnostic accuracy study comparing the C-WATCH, post-endoscopic Rockall, AIMS65, and Glasgow-Blatchford scores for predicting mortality, rebleeding, and the need for transfusion of packed red blood cells in patients with upper gastrointestinal bleeding (UGIB) at a tertiary-level public hospital in Lima, Peru. All risk scores were calculated at the time of initial presentation using clinical and laboratory data obtained at admission. The calculation of the C-WATCH score is described in the Supplementary Appendix S1. Participants We included patients aged ≥ 18 years with a diagnosis of UGIB, defined as melena, hematemesis, coffee-ground emesis, or hematochezia with a confirmed upper gastrointestinal source. Patients were enrolled if they presented to the emergency department or developed UGIB during hospitalization for another condition. Exclusion criteria included the following: incomplete clinical records or missing data necessary to calculate the study variables; death prior to data collection; and absence of follow-up within 30 days after the bleeding episode. Sample size calculation Sample size was calculated using OpenEpi version 3, based on the original C-WATCH validation study and prior comparative studies of Glasgow-Blatchford, Rockall, and AIMS65 scores [3, 12]. A minimum sample of 160 patients was required to achieve 80% statistical power with a 95% confidence level. Data collection and statistical analysis Data were collected between June 2019 and September 2025 and recorded in Microsoft Excel. Statistical analysis was performed using Stata version 19 (StataCorp, College Station, TX, USA). The discriminative ability of each scoring system was evaluated using AUROC analysis. Optimal cut-off values for predicting severity outcomes (30-day mortality, rebleeding, and need for transfusion of ≥ 2 units of packed red blood cells) were determined using the Youden index. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated for each score and outcome. Pairwise comparisons between scores were performed using the DeLong test for correlated AUROCs. Bonferroni correction was applied to account for multiple comparisons and statistical significance was defined as p-value < 0.0125. Only patients with complete clinical and laboratory analysis required for scores calculation were included in the final analysis. Missing data were minimal and handled using a complete case approach. Loss to follow-up was defined as the inability to determine outcome status after 30 days after index event despite review of clinical record and follow-up contact. Ethics approval and consent to participate Patient data were extracted from medical records and coded sequentially to ensure confidentiality. All participants provided written informed consent to participate and to authorize data collection, including telephone follow-up at 7 and 30 days after hospital discharge. The study was conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Boards of Universidad Peruana Cayetano Heredia and Hospital Cayetano Heredia. Results A total of 235 patients were included, of whom 59.6% were male. The mean age was 61.0 ± 16.5 years, and the overall 30-day mortality rate was 4.7% (n = 11). No patients were lost to follow-up during the 30-day period. The general characteristics of the study population are presented in Table 1. Table 1Characteristics of patients includedVariablePatients n (%)Sex Male140 (59.6) Female95 (40.4)Comorbidity Cirrhosis50 (21.3) Heart failure7 (3.0) Myocardial infarction6 (2.5) Renal failure56 (23.8) Stroke8 (3.4) COPD2 (0.8) Hypertension72 (30.6) Type 2 DM55 (23.4) Malignancy17 (7.2) Alcoholism15 (6.4) Others76 (32.2) History of upper gastrointestinal bleeding58 (24.7)Clinical presentation Melena147 (62.5) Coffee-ground emesis75 (31.9) Hematochezia13 (5.5)Charlson Comorbidity Index 0 points7 (2.9) 1 point12 (5.1) 2 points20 (8.5) 3 points31 (13.2) 4 points38 (16.2) 5 points41 (17.4) 6 points28 (11.9) 7 points27 (11.5) 8 points15 (6.5) 9 points9 (3.9) 10 points7 (2.9)COPD Chronic obstructive pulmonary disease, DM Diabetes mellitus Characteristics of patients included COPD Chronic obstructive pulmonary disease, DM Diabetes mellitus The most common comorbidities were hypertension (30.6%), chronic kidney disease (23.8%), and liver cirrhosis (21.3%). Additionally, 24.7% of patients had a prior history of upper gastrointestinal bleeding. The most frequent causes of bleeding are shown in Table 2, with gastric ulcers (31.1%), esophageal varices (25.1%), and duodenal ulcers (16.6%) being the most common etiologies. Table 2Etiology of upper gastrointestinal bleedingEndoscopic diagnosisPatients n (%)Peptic ulcer disease112 (47.7)Gastric ulcers73 (31.1)Duodenal ulcers39 (16.6)Forrest classification Forrest IA2 (1.8) Forrest IB11 (9.8) Forrest IIA29 (25.9) Forrest IIB11 (9.8) Forrest IIC18 (16.1) Forrest III35 (31.3)No data available6 (5.4)Esophageal varices59 (25.1)Gastric varices12 (5.1)Gastric cancer16 (6.8)Esophagitis16 (6.8)Mallory-Weiss4 (1.7)Dieulafoy1 (0.4)Others15 (6.3)Total235 (100) Etiology of upper gastrointestinal bleeding Table 3 details the causes of bleeding among patients who died, with gastric and esophageal varices accounting for 54.5% of fatal cases. Table 3StableEtiology of upper gastrointestinal bleeding in deceased patientsEndoscopic diagnosisPatients n (%)Esophageal/gastric varices6 (54.5)Gastric ulcers3 (27.3)Duodenal ulcers2 (18.2)Total11 (100) StableEtiology of upper gastrointestinal bleeding in deceased patients For mortality prediction, the C-WATCH score demonstrated the numerically highest AUROC value (0.83), followed by AIMS65 (0.81), the post-endoscopic Rockall score (0.80), the Glasgow-Blatchford score (0.69) and the pre-endoscopic Rockall score (0.63). However, pairwise comparisons did not demonstrate consistent statistically significant differences after Bonferroni correction, except for the comparison between C-WATCH and the Pre-endoscopic Rockall score (p = 0.001), as shown in Table 4. Table 4Discriminative performance of risk scores for clinical outcomesOutcomeComparisonAUROC C-WATCH95% CIComparator AUROC95% CIp-valueMortalityC-WATCH vs. AIMS650.830.72–0.940.820.70–0.930.786C-WATCH vs. Glasgow-Blatchford0.830.72–0.940.690.53–0.850.122C-WATCH vs. Pre-Rockall0.830.72–0.940.630.47–0.790.001C-WATCH vs. Post-Rockall0.830.72–0.940.800.72–0.880.500Transfusion ≥ 2 unitsC-WATCH vs. AIMS650.660.59–0.730.540.47–0.620.010C-WATCH vs. Glasgow-Blatchford0.660.59–0.730.620.55–0.700.338C-WATCH vs. Pre-Rockall0.660.59–0.730.540.46–0.620.016C-WATCH vs. Post-Rockall0.660.59–0.730.590.51–0.660.119RebleedingC-WATCH vs. AIMS650.490.32–0.660.540.33–0.750.693C-WATCH vs. Glasgow-Blatchford0.490.32–0.660.540.32–0.770.725C-WATCH vs. Pre-Rockall0.490.32–0.660.570.38–0.770.533C-WATCH vs. Post-Rockall0.490.32–0.660.470.29–0.650.887Pairwise comparisons were performed using the DeLong test. After Bonferroni correction (p < 0.0125), most differences were not statistically significantPre-Rockall Pre-endoscopic Rockall score, Post-Rockall Post-endoscopic Rockall score Discriminative performance of risk scores for clinical outcomes Pairwise comparisons were performed using the DeLong test. After Bonferroni correction (p < 0.0125), most differences were not statistically significant Pre-Rockall Pre-endoscopic Rockall score, Post-Rockall Post-endoscopic Rockall score For predicting the need for transfusion of ≥ 2 units of packed red blood cells, the AUROC was 0.66 for C-WATCH, 0.62 for Glasgow-Blatchford, 0.59 for post-endoscopic Rockall, and 0.54 for both pre-endoscopic Rockall and AIMS65. Although some pairwise comparisons yielded p-values < 0.05, none remained statistically significant after Bonferroni correction. Regarding rebleeding, all scores demonstrated limited discriminative performance. The AUROC was 0.57 for the pre-endoscopic Rockall score, 0.54 for both Glasgow-Blatchford and AIMS65, 0.49 for C-WATCH, and 0.47 for the post-endoscopic Rockall score. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) for each score and outcome are summarized in Table 5. Optimal cut-off values were explored using receiver operating characteristic curve analysis; detailed results are provided in the Supplementary Appendix S2. Table 5Comparison of sensitivity and specificity with different scores to predict mortality, transfusion and rebleedingOutcomeScoreSensitivity (%)Specificity (%)PPV (%)NPV (%)MortalityC-WATCH ≥ 490.958.91099.2AIMS65 ≥ 372.778.614.398.3Glasgow-Blatchford ≥ 1281.850.97.698.2Pre-Rockall ≥ 390.929.95.998.5Post-Rockall ≥ 681.870.111.898.7Transfusion of ≥ 2 unitsC-WATCH ≥ 459.264.144.176.7AIMS65 ≥ 256.654.137.172.3Glasgow-Blatchford ≥ 1176.345.940.380.2Pre-Rockall ≥ 376.331.434.773.5Post-Rockall ≥ 643.472.943.472.9RebleedingC-WATCH ≥ 377.826.1496.7AIMS65 ≥ 255.650.94.396.6Glasgow-Blatchford ≥ 610010.24.2100Pre-Rockall ≥ 466.747.34.897.3Post-Rockall ≥ 388.916.44.197.4NPV Negative predictive value, PPV Positive predictive value, Pre-Rockall Pre-endoscopic Rockall score, Post-Rockall Post-endoscopic Rockall score Comparison of sensitivity and specificity with different scores to predict mortality, transfusion and rebleeding NPV Negative predictive value, PPV Positive predictive value, Pre-Rockall Pre-endoscopic Rockall score, Post-Rockall Post-endoscopic Rockall score The corresponding ROC curves for mortality, transfusion requirements, and rebleeding are shown in Figs. 1 and 2, and 3, respectively. Missing data were minimal and unlikely to have significantly influenced the results. ROC curves predicting 30-day mortality in upper gastrointestinal bleeding ROC curves predicting transfusion ≥ 2 units of packed red blood cells ROC curves predicting 30-day rebleeding in upper gastrointestinal bleeding Discussion In this prospective external validation study conducted in a Latin American cohort, the C-WATCH score demonstrated comparable, rather than clearly superior, discriminative performance for prediction of 30-day mortality and transfusion when compared with established scores such as AIMS65, Glasgow-Blatchford and Rockall. All evaluated scores showed poor performance for rebleeding prediction. These findings should be interpreted in the context of clinical practice, where accurate risk assessment is essential for optimized resource allocation and timely clinical decision-making. Accordingly, several prognostic scores have been developed to identify high-risk patients and anticipate adverse outcomes [16, 17]. Arıkoglu et al. conducted a retrospective study including 311 patients and evaluated the predictive performance of the Glasgow-Blatchford score (GBS), AIMS65, Rockall score (RS), and the International Bleeding Risk Score (INBS/ABC). They reported that all scores were capable of predicting hospitalization, transfusion requirements, and high-risk mortality, although the Rockall score was recommended for determining the need for endoscopic intervention [18]. Similar to our findings, melena was the most common clinical presentation. Blood transfusion remains a key component of UGIB management. They also reported transfusion rates of 61.4%, comparable to those described by Choe et al. (60.8%) and Marmo et al. (59.5%) ([19–20]). In our study, 54.5% of patients required transfusion of at least one unit of packed red blood cells. A retrospective study from India comparing AIMS65, GBS, and the modified Glasgow-Blatchford score reported AUROC values for mortality prediction of 0.77, 0.73, and 0.70, respectively. In that cohort, variceal bleeding accounted for 49% of cases, whereas peptic ulcer disease represented only 11% [21]. These findings differ from our results, in which peptic ulcer disease accounted for 47.7% of cases. Similar proportions of peptic ulcer bleeding have been reported in South Korea (43.3%), China (48.2%), and in a large multicenter study involving 3,012 patients across the United States, Scotland, England, Denmark, Singapore, and New Zealand (28%) [22–24]. At Hospital Cayetano Heredia, a previous comparative study of the Glasgow-Blatchford, post-endoscopic Rockall, and AIMS65 scores found that AIMS65 demonstrated superior performance for mortality prediction (AUROC 0.90) and for predicting transfusion of ≥ 3 units of packed red blood cells [3]. In contrast, our findings demonstrated numerically higher but no statistically significant discriminative performance of C-WATCH for 30-day mortality compared with AIMS65, Rockall, and Glasgow-Blatchford scores. To our knowledge, this study is the first prospective study conducted in Peru and Latin America evaluating the performance of the C-WATCH score in comparison with conventional risk scores. C-WATCH demonstrated numerically higher, but not statistically significant predictive accuracy for 30-day mortality. These findings are similar to those reported by Allo et al., who found no significant differences between C-WATCH, Glasgow-Blatchford, and Rockall for mortality prediction [15]. In contrast, Thiebaud et al. reported better performance of the Glasgow-Blatchford score (AUROC 0.87) compared with C-WATCH (AUROC 0.75) for predicting mortality and/or therapeutic intervention [25]. Although AIMS65 has traditionally been considered one of the strongest predictors of mortality in UGIB, our findings suggest that C-WATCH may represent a comparable alternative for mortality risk assessment, particularly given that it incorporates hemoglobin as a direct marker of bleeding severity [3, 5, 26]. Regarding transfusion requirements, prior systematic reviews and meta-analyses, particularly in patients with non-variceal UGIB, have suggested that the Glasgow-Blatchford score performs best for predicting the need for transfusion [27]. In our study, C-WATCH demonstrated numerically higher AUROC values compared with Glasgow-Blatchford for predicting transfusion of ≥ 2 units, however, the difference was modest and not statistically significant. From a practical perspective, C-WATCH relies exclusively on laboratory parameters available at admission and does not require endoscopic findings, which may facilitate rapid risk assessment in resource-limited settings. With respect to rebleeding, all evaluated scores, including C-WATCH, Glasgow-Blatchford, AIMS65 and both pre- and post-endoscopic Rockall, demonstrated limited predictive ability (AUROC < 0.58). This is consistent with findings reported by Nik Pa et al., who observed similarly poor performance of these scores for predicting rebleeding in both variceal and non-variceal UGIB [28]. However, other studies, such as that by Han et al., have suggested that a post-endoscopic Rockall score ≥ 5 may predict rebleeding in non-variceal UGIB, highlighting the variability across populations [29]. Although the C-WATCH score demonstrated numerically higher AUC values for predicting mortality and transfusion requirements, these findings should be interpreted with caution. On the one hand, the study was conducted at a single tertiary center, added to the relatively low number of mortality events that may affect the precision of the estimates. In addition, differences in patient characteristics, including younger age, comorbidity burden, and higher prevalence of variceal bleeding, may influence score performance and limit generalizability. From a clinical perspective, these findings suggest that C-WATCH may perform similarly to existing scores; however, the modest differences observed are unlikely to substantially influence clinical decision-making, such as discharge planning or timing of endoscopy. On the other hand, the prospective design, systematic 30-day follow-up, and direct comparison with established risk scores strengthen the clinical relevance of the findings. Further multicenter prospective studies are needed to confirm these results before broader implementation in clinical practice. Conclusion In this prospective study, the C-WATCH score demonstrated comparable performance to established scores, with no significant difference in pairwise comparisons for predicting 30-day mortality and the need for transfusion of ≥ 2 units of packed red blood cells in patients with upper gastrointestinal bleeding. All scores showed poor discrimination for rebleeding. Supplementary Information Supplementary Material 1. Authors’ contributions J.E. conceptualized and designed the study and revised the manuscript. R.R conducted the statistical analysis, drafted the manuscript and contributed with data interpretation. B.S. participated in patient recruitment, data acquisition and manuscript review. J.H. supervised and critically reviewed the manuscript. All authors read and approved the final manuscript. Funding Self-funding. Data availability The datasets generated an analyzed during the present study are available in the figshare repository: https://figshare.com/s/08f95201da57f8f0836a. Declarations Competing interests The authors declare no competing interests. Footnotes The original online version of this article was revised: there was a spelling error in the name of Brendy Santiago. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Change history 6/30/2026 A Correction to this paper has been published: 10.1186/s12876-026-05065-0 References Associated Data Supplementary Materials Supplementary Material 1. Data Availability Statement The datasets generated an analyzed during the present study are available in the figshare repository: https://figshare.com/s/08f95201da57f8f0836a.