Validating the CHAMPS Score: A Novel and Reliable Prognostic Score of Non-Variceal Upper Gastrointestinal Bleeding

Introduction

Acute non-variceal upper gastrointestinal bleeding (NVUGIB) is a common emergency. Despite significant advancements in endoscopic techniques and proton pump inhibitor (PPI) development, the mortality rate remains between 2% and 14%, and is frequently associated with severe comorbidities.^1–3 Therefore, current guidelines recommend concurrent initial evaluation and resuscitation, advocating for risk stratification to guide treatment strategies.^4–6 Based on current evidence, only the Glasgow-Blatchford score (GBS) is recommended for identifying the very low-risk group for safe discharge.^4–6

However, risk stratification has further potential advantages, supporting post-hospital admission decisions, such as intensive care unit necessity, endoscopy timing, rebleeding risk post-intervention, and discharge timing. Consequently, the risk prediction scores were designed for primary outcomes and expanded to address various needs. However, to date, many outcomes still lack a score with a reliable predictive value, such as the timing of endoscopy and length of hospital stay. Regarding transfusion requirements, almost all scores lack predictive value, and guidelines are often based on the subjective Forrest classification during endoscopy.^7–9 Regarding mortality, the recently introduced Age, Blood tests, and Comorbidities (ABC) score outperformed other scores in various studies; however, the area under the receiver operating characteristic (AUROC) values varied, and identifying the low-risk group was a limitation of this score, with a sensitivity of only 60% and a mortality rate of up to 2.3%.^10,11

Recently, the Charlson comorbidity Index ≥2, in-Hospital onset, Albumin <2.5 g/dL, altered Mental status, Eastern Cooperative Oncology Group Performance status ≥2, Steroid use (CHAMPS) score has been developed and validated by incorporating comorbidity and performance status assessments, thereby enhancing its predictive capabilities. The CHAMPS score comprises six variables of equal weight, including the Charlson Comorbidity Index, in-hospital onset, albumin level, altered mental status, Eastern Cooperative Oncology Group performance status, and steroid use.¹² Through various studies, the CHAMPS score has consistently demonstrated good predictive ability for mortality (AUROC > 0.8), better than four other commonly used scores, namely GBS, Rockall score, Albumin level <3.0 mg/dL, International normalized ratio >1.5, altered Mental status, Systolic blood pressure ≤90 mm Hg, and age >65 years (AIMS65), and the ABC score, in predicting mortality (p < 0.05 for all). Furthermore, CHAMPS identified 15% of patients in the high-risk group with a significantly higher mortality rate of 25–26.5%. We therefore recommend early and aggressive intervention for this group. Furthermore, CHAMPS identified 40% of the low-risk patients with a mortality rate of 0.2–0.7%, outperforming other systems. The GBS, although showing a zero mortality rate, identified only 0.8–2% of the low-risk patients, limiting its practical utility.^12,13 Despite its significant potential, the paucity of research on the CHAMPS score and its unexplored extension to other outcomes prompted us to validate its predictive capability across multiple in-hospital outcomes, helping clinical decision-making.

Methods

Study Design

A prospective cohort study was conducted on adult patients with acute NVUGIB admitted to the Department of Gastroenterology at Cho Ray Hospital, Ho Chi Minh City, Vietnam, between November 2022 and June 2023. This study was approved by the Ethics Committee of the University of Medicine and Pharmacy, Ho Chi Minh City, Vietnam (Ethical Approval Number: 814/HDDD-DHYD).

Sample Size

The sample size was calculated using the formula: ; ; a = 1.414 × Z_AUC. Where: α, type 1 error rate; Z_(1-α/2), corresponding z-score for the desired confidence level (1-α); d, margin of error; V_AUC, Variance of the Area Under the Curve (AUC); n_disease, number of high-risk patients to be estimated; n_non-disease, number of low-risk patients to be estimated. We chose: Z_(1-α/2)=1.96 for α=0.05 (95% confidence); d = 0.1; AUC of the CHAMPS score, according to Tamotsu Matsuhashi’s study, is 0.812.¹² Therefore, the minimum sample size is 90 patients, including at least 45 high-risk patients.

Study Population

The inclusion criteria consisted of evidence of NVUGIB on esophagogastroduodenoscopy, along with one of the following symptoms: (a) hematemesis, (b) melena, and (c) a decrease in hemoglobin level >2 g/dL compared to the previous examination. Patients with bleeding from the tumor or post-procedural complications were excluded.^12,13

Data on CHAMPS, ABC, AIMS65, full Rockall score (fRS), clinical Rockall score (cRS) and GBS scores were collected. Subsequently, patient outcomes, including mortality (bleeding- and non-bleeding-related), rebleeding, blood transfusion, and endoscopic and surgical interventions, were recorded until discharge. Rebleeding was defined as the recurrence of bleeding during hospitalization, manifested through a new episode of hematemesis, melena, or unstable hemodynamics, and reconfirmed by identification of the same source on esophagogastroduodenoscopy.¹² Intervention was defined as a composite endpoint of the need for intervention during hospitalization (blood transfusion, endoscopic or surgical intervention). Based on the collected data, we validated the CHAMPS score for each specific outcome and determined the optimal cutoff thresholds. We compared the CHAMPS scores with other scores based on various outcomes. The cutoff points for the other scores were similar to those used in previous studies. Specifically, the cutoffs of low-risk groups for ABC, AIMS65, cRS, fRS, and GBS were set at ≤3, ≤1, ≤2, ≤2, and ≤1, respectively. In parallel, the high-risk thresholds for the aforementioned scores were established at ≥8, ≥2, ≥3, ≥8, and ≥5, respectively.^12–14

Statistical Analysis

The software used for data analysis included Stata 16.0 and Excel Office 365 (facilitating chart creation).

Continuous variables with a normal distribution were presented as mean ± standard deviation (mean ± SD), while continuous variables without a normal distribution were presented as median and interquartile range (IQR). Categorical variables were presented as counts and percentages.

Interpretation of score values was performed using the AUROC curve, and the DeLong test was used to compare different AUROC curves. The optimal cutoff threshold was calculated using the Youden index and the distance “d” (the shortest distance from the point (0, 1) to the ROC curve).

Statistically significant results were defined as those having a p-value of less than 0.05.

Results

Patient Characteristics and Outcomes

The present study included 140 patients (70% men, with a mean age of 63.5 ± 15.5 years). Table 1 presents a detailed description of the baseline characteristics of the study population. Regarding in-hospital outcomes, 15 deaths (10.7%) were recorded, with non-bleeding-related deaths being the most frequent (73.3%). Of all patients, 103 (73.6%) required intervention and three (2.1%) experienced rebleeding.

Table 1 Patient Baseline Characteristics and Outcomes

CHAMPS Performance

The CHAMPS score showed robust predictive accuracy for mortality (AUROC 0.89; 95% Confidence Interval (CI): 0.83–0.94), with AUROC values consistently exceeding 0.8 even within subgroups of bleeding-related and non-bleeding-related mortality. Concerning rebleeding and intervention, the CHAMPS score showed no predictive value in our study, with AUROC values (95% CI) of 0.43 (0.09–0.76) and 0.55 (0.44–0.66), respectively.

The optimal cutoff threshold for mortality was determined to be 3 points, yielding the highest Youden index and a minimal distance “d”, achieving a sensitivity of 100% and a specificity of 71.2%. This threshold was also identified as optimal for bleeding- and non-bleeding-related mortality. Owing to the suboptimal predictive performance of the intervention and rebleeding endpoints, we refrained from identifying an optimal cutoff threshold for these outcomes. At a cutoff threshold of 3, the sensitivity and specificity for both intervention and bleeding were suboptimal at 38.8% and 70.3%, respectively, for intervention and 33.3% and 63.5%, respectively, for rebleeding.

A threshold of 0 served as the optimal cutoff for the low-risk mortality group, with a sensitivity of 100%, specificity of 13.6%, positive predictive value of 12.2%, and a negative predictive value of 100%. With this cutoff threshold, the low-risk group, according to CHAMPS, showed mortality and rebleeding rates equal to zero.

Comparison with Five Pre-Existing Scores

Subsequently, we compared the performance of the CHAMPS score with that of other scores by (1) evaluating their predictive abilities for various outcomes, (2) differentiating high-risk individuals, and (3) distinguishing low-risk individuals suitable for discharge or localized treatment.

For mortality prediction, the CHAMPS score exhibited superior performance (AUROC = 0.89) that was significantly better than that of AIMS65 (AUROC = 0.71; p = 0.02) and GBS (AUROC = 0.72; p < 0.05) scores. Regarding intervention, only the GBS score demonstrated a predictive value with an AUROC of 0.75, while the others lacked predictive value. Specifically, for blood transfusion, endoscopic intervention, and surgical intervention, the AUROC values were either average or lacked a predictive value. Concerning rebleeding, the AIMS65 score exhibited good predictive ability, the GBS and Rockall scores provided moderate predictive ability, and the CHAMPS score had no predictive value (Table 2 and Figure 1).

Table 2 Discriminative Ability of the Scores by Outcome

Figure 1 Comparison of scores in prediction of (A) mortality, (B) bleeding-related mortality, (C) non-bleeding-related mortality, (D) intervention, and (E) rebleeding. Abbreviations: AUROC, area under receiver operating characteristic; CHAMPS, The Charlson comorbidity Index ≥2, in-Hospital onset, Albumin <2.5 g/dL, altered Mental status, Eastern Cooperative Oncology Group Performance status ≥2, Steroid use; ABC, Age, Blood tests, and Comorbidities; AIMS65, Albumin level <3.0 mg/dL, International normalized ratio >1.5, altered Mental status, Systolic blood pressure ≤90 mm Hg, and age >65 years; cRS, clinical Rockall score; fRS, full Rockall score; GBS, Glasgow Blatchford score.

In the high-risk cohort, CHAMPS identified 36.4% of patients classified as high-risk, with a mortality rate of 29.4%. The mortality rate in the high-risk group according to the CHAMPS score was higher than that in the other groups. With a threshold of ≥5, the GBS score recognized 92.9% of patients as high-risk, but the mortality rate within this group was only 10.8%, resulting in a notably low specificity of 7.2%, indicating poor GBS performance in identifying the high-risk group for mortality (Table 3).

Table 3 The Predictive Performance and Discriminative Abilities for Identification of Patients at High-Risk of Mortality

Sensitivity is of the utmost importance for low-risk patient groups. Both the CHAMPS and GBS scores achieved 100% sensitivity for outcomes related to mortality and rebleeding. Regarding intervention, the GBS cutoff maintained 100% sensitivity, while that of CHAMPS was only 90.3% (Table 4). Although low-risk individuals identified by both CHAMPS and GBS scores had rebleeding and mortality rates of 0%, a more detailed analysis revealed that the CHAMPS score outperformed other scoring systems by identifying as many as 12.1% of patients in the low-risk category, whereas GBS only identified 1.4%. However, a disadvantage of the CHAMPS scoring system was its recognition of up to 58.8% of patients as low-risk cases who may require intervention, while patients with a GBS score ≤1 had a 0% intervention rate (Table 5).

Table 4 The Predictive Performance for Different Outcomes in the Low-Risk Group Among Six Scores

Table 5 The Distribution of Outcomes Within the Low-Risk Group

Discussion

The CHAMPS score, initially designed to assess in-hospital mortality, was validated in our study for this outcome and extended to assess intervention and rebleeding. Regarding mortality, our results indicated that the CHAMPS score (AUROC: 0.91) outperformed ABC, AIMS65, GBS, and cRS scores both in the derivation and validation cohorts (all p-values <0.05).¹² Another validation study by Hakan Aydin reported that CHAMPS robustly predicted mortality with an AUROC value of 0.812 (95% CI: 0.78–0.84), outperforming GBS (AUROC: 0.683; p = 0.008), and showed no significant difference compared to ABC and AIMS65 scores.¹³ In our study, CHAMPS was the only score showing good overall mortality prediction (AUROC: 0.89; 95% CI: 0.83–0.94), while other scores exhibited only moderate predictive ability. The CHAMPS score significantly outperformed AIMS65 (AUROC: 0.71) and GBS (AUROC: 0.72) (p < 0.05) scores, whereas no significant differences were observed compared to ABC (AUROC: 0.79), cRS (AUROC: 0.76), and fRS (AUROC: 0.74) scores (Figure 1). Upon detailed analysis, CHAMPS demonstrated significantly superior mortality prediction compared to the ABC score for bleeding-related mortality (AUROCs of CHAMPS and ABC were 0.88 and 0.66, respectively; p < 0.05). Additionally, it performed significantly better than AIMS65 and GBS for predicting non-bleeding-related mortality (AUROC values of CHAMPS, AIMS65, and GBS were 0.87, 0.64, and 0.65, respectively; all p < 0.05). Overall, CHAMPS appears to have a better mortality prediction score than the others, with AUROC values consistently exceeding 0.8 in various studies.

Rebleeding requires continued treatment and intervention, leading to prolonged hospitalization and increased costs. Identifying patients at high risk of rebleeding allows stratified care and discharge decisions. To date, the Forrest classification has been exclusively employed in current guidelines,^4–6 however, subjectivity in endoscopist interpretation remains a limitation. Concerning pre-endoscopic risk scores for rebleeding, several previous studies favored the Progetto Nazionale Emorragia Digestiva (PNED) score (AUROC: 0.85–0.87) over other pre-endoscopic risk scores in predicting rebleeding.^8,15 In our study, the AIMS65 score exhibited good predictive ability for rebleeding (AUROC: 0.84), whereas the fRS (AUROC: 0.77) and GBS (AUROC: 0.75) scores displayed moderate predictive capabilities, and the CHAMPS score demonstrated no predictive ability for this outcome (AUROC: 0.43). Additional research is required to further investigate the application of risk scores to predict rebleeding outcomes.

Regarding the prediction of interventions, the findings of our study closely parallel those of previous reports. According to previous studies, the GBS score exhibits superior predictive performance compared to other risk scores in predicting the need for either endoscopic or surgical intervention. With regard to predicting the need for blood transfusion, the GBS score consistently outperformed other risk scores in multiple comparative studies.⁸ Using a composite endpoint of the need for intervention during hospitalization, the GBS score demonstrated a moderate level of predictive efficacy in the present study (AUROC: 0.75), while fRS and AIMS65 showed limited predictive capabilities, recording AUROC values of 0.68 and 0.66, respectively. However, the CHAMPS score failed to predict this outcome. The GBS score maintained its standing as the most reliable predictive score for blood transfusion, despite a modest AUROC value of 0.75, whereas the CHAMPS score demonstrated no predictive value for this outcome.

With the global concern over healthcare system overload, the identification of very low-risk patients for early discharge or outpatient treatment has gained attention. Prioritizing patient safety, the decision to discharge emphasizes high sensitivity and specificity. Based on recent evidence, a GBS score of ≤1 is still chosen for safe outpatient treatment.^4–6 In our study, considering mortality and rebleeding outcomes, low-risk patients identified by CHAMPS and GBS scores had a 0% mortality and rebleeding rate, with 100% sensitivity and negative predictive value. Upon more detailed analysis, we observed a notable advantage of the CHAMPS score in identifying 17 cases (12.1%) in the low-risk group, whereas the GBS score identified only two cases (1.4%). However, regarding intervention prediction, the low-risk group predicted by CHAMPS had a 58.8% intervention rate, while the low-risk group defined via GBS (GBS ≤ 1) had a 0% intervention rate. Thus, while the low-risk threshold of CHAMPS may not determine outpatient treatment, its 0% mortality and rebleeding rates suggest treatment at lower-level hospitals with facilities for endoscopic and surgical interventions to avoid transfer to facilities with better intensive care capabilities. Additionally, for low-risk patients who have undergone successful intervention for bleeding, the CHAMPS score may assist in making safe and early discharge decisions.

Limitations

Our study was conducted at a single center over a short period of time. Therefore, further multicenter studies with larger sample sizes are required to validate the performance of the CHAMPS risk prediction scoring system.

Conclusion

The CHAMPS score is a simple mnemonic score (with each letter corresponding to a variable), accessible on mobile phones, cost-effective with only one albumin test, and has consistently shown good predictive value for mortality in various studies. In practice, the CHAMPS score can be used to distinguish patients with a high-risk of mortality, leading to early aggressive treatment and reduced mortality rates. The low-risk group identified using the CHAMPS score may only require localized treatment, avoiding the need for advanced intensive care facilities and allowing for safe early discharge after successful bleeding control.