Introduction

Acute gastrointestinal bleeding (GIB) is an emergency and critical clinical event [1]. The mortality of acute GIB is 6–20% in patients with cirrhosis [2,3,4]. The prognosis seems to be similar between patients with cirrhosis and acute variceal bleeding and those with cirrhosis and peptic ulcer bleeding [5]. It is important to accurately evaluate the prognosis in patients with cirrhosis and acute GIB. Conventional scoring systems for assessing the prognosis of patients with acute GIB mainly include Rockall score [6], Glasgow–Blatchford score (GBS) [7], and AIMS65 score [8]. However, they are not specific for patients with cirrhosis in whom gastroesophageal varices are the most frequent sources of acute GIB [9] and the severity of liver dysfunction is closely associated with patients’ outcomes. On the other hand, Child–Pugh [10], model for end-stage liver disease (MELD) [11], MELD-Na [12], and neutrophil–lymphocyte ratio (NLR) [13] scores have been widely employed for prognostic assessment in general patients with liver cirrhosis. But their predictive performances remain suboptimal in patients with cirrhosis and acute GIB.

The present work aimed to develop and validate a novel model for assessing the prognosis of patients with cirrhosis and acute GIB on the basis of the data obtained from a multicenter study.

Methods

The present study was based on the TORCH study (NCT03846180), which was an investigator-initiated multicenter study across 13 centers from eight provinces or municipalities in China. It was carried out following the rules of the Declaration of Helsinki and was approved by the Medical Ethical Committee of the General Hospital of Northern Theater Command (formerly General Hospital of Shenyang Military Area), which is the principal affiliation of this study. The ethical approval number was k (2019) 20. The requirement for informed written consent was waived because of the nature of this study. Briefly, we enrolled the patients with cirrhosis who were admitted because of acute GIB from January 2010 to December 2018. Age, gender, and comorbidities were not limited.

The following data were collected: age; gender; etiology of liver cirrhosis; history of GIB, diabetes, and hepatocellular carcinoma (HCC); ascites; hepatic encephalopathy (HE); and laboratory tests at admission, mainly including hemoglobin (Hb), hematocrit (HCT), white blood cell (WBC), platelet (PLT), total bilirubin (TBIL), albumin (ALB), alanine aminotransferase (ALT), alkaline phosphatase (AKP), gamma-glutamyl transpeptidase (GGT), serum creatinine (Scr), potassium (K), sodium (Na), and international normalized ratio (INR); and in-hospital death. Child–Pugh [10], MELD [11], MELD-Na [12], and NLR [13] scores were calculated.

Random sampling was used to divide patients into training and validation cohorts with an approximate percentage of 50%. Continuous variables were expressed as mean ± standard deviation and median (range), and categorical variables were expressed as frequency (percentage). Difference between training and validation cohorts was compared by the non-parametric Mann–Whitney U test and the chi-square test. In the training cohort, logistic regression analyses were performed to identify the independent predictors associated with in-hospital death. Odds ratios (ORs) with 95% confidence intervals (CIs) were calculated. An equation for predicting the death of patients with cirrhosis and acute GIB was established by merging the independent predictors with their regression coefficients. Then, receiver operating characteristic curve (ROC) analysis was performed to evaluate the predictive performance of the new equation. The area under curve (AUC) and the best cutoff value with sensitivity and specificity were calculated. The predictive performance of the new equation was also compared with other established scores (Child–Pugh [10], MELD [11], MELD-Na [12], and NLR [13] scores). All statistical analyses were performed using SPSS software version 20.0 (IBM Corp, Armonk, NY, USA) and MedCalc software version 11.4.2.0 (MedCalc Software, Mariakerke, Belgium). P < 0.05 was considered statistically significant.

Results

Patient Selection

A total of 1682 patients with cirrhosis and acute GIB were included, of whom 865 and 817 patients were enrolled into the training and validation cohorts, respectively. Characteristics of patients are summarized in Table 1. All but the percentage of HCC were not statistically significantly different between the training and validation cohorts (Table 1).

Table 1 Characteristics of patients in training and validation cohorts
Full size table

Univariate and Multivariate Analyses in the Training Cohort

Univariate logistic regression analyses demonstrated that HCC, diabetes, hepatic C virus infection, ascites, HE, WBC, TBIL, ALB, ALT, Scr, and INR were significantly associated with in-hospital death (Table 2). Multivariate logistic regression analyses showed that HCC, diabetes, TBIL, ALB, ALT, and Scr were independently associated with in-hospital death (Table 2).

Table 2 Univariate and multivariate analyses of predictors associated with the in-hospital mortality of acute GIB in training cohort
Full size table

Development of CAGIB Score

A prognostic model called CAGIB (Cirrhosis Acute GastroIntestinal Bleeding) was established. CAGIB = Diabetes (yes = 1, no = 0) × 1.040 + HCC (yes = 1, no = 0) × 0.974 + TBIL (μmol/L) × 0.005 − ALB (g/L) × 0.091 + ALT (U/L) × 0.001 + Scr (μmol/L) × 0.012 − 3.964. It had an AUC of 0.829 (95% CI 0.801–0.854, P < 0.0001), and its best cutoff value was greater than − 4.6646 with a sensitivity of 78.57% and a specificity of 75.52% (Fig. 1). The AUCs of Child–Pugh, MELD, MELD-Na, and NLR scores were 0.762 (95% CI 0.732–0.791), 0.778 (95% CI 0.748–0.806), 0.765 (95% CI 0.735–0.793), and 0.587 (95% CI 0.553–0.620), respectively (Fig. 2). The difference was statistically significant between CAGIB and NLR score (P = 0.0001), but not between CAGIB and Child–Pugh, MELD, or MELD-Na score.

Fig. 1
figure 1

ROC curve of CAGIB score for predicting the in-hospital death of patients with cirrhosis and acute GIB in the training cohort

Full size image
Fig. 2
figure 2

Comparison of predictive performance of CAGIB score with Child–Pugh, MELD, MELD-Na and NLR scores in the training cohort. Brown line refers to the CAGIB score, red line refers to the Child–Pugh score, green line refers to the MELD score, purple line refers to the MELD-Na score, and orange line refers to the NLR score

Full size image

Validation of CAGIB Score

In the validation cohort, the CAGIB score had an AUC of 0.714 (95% CI 0.682–0.746, P = 0.0006) (Fig. 3). The AUCs of Child–Pugh, MELD, MELD-Na, and NLR scores were 0.693 (95% CI 0.659–0.725), 0.662 (95% CI 0.627–0.695), 0.660 (95% CI 0.626–0.694), and 0.538 (95% CI 0.503–0.574), respectively (Fig. 4). The difference was statistically significant between CAGIB and NLR score (P = 0.0165), but not between CAGIB and Child–Pugh, MELD, or MELD-Na score.

Fig. 3
figure 3

ROC curve of CAGIB score for predicting the in-hospital death of patients with cirrhosis and acute GIB in the validation cohort

Full size image
Fig. 4
figure 4

Comparison of predictive performance of CAGIB score with Child–Pugh, MELD, MELD-Na and NLR scores in the validation cohort. Brown line refers to the CAGIB score, red line refers to the Child–Pugh score, green line refers to the MELD score, purple line refers to the MELD-Na score, and orange line refers to the NLR score

Full size image

Discussion

Our study developed a new model (CAGIB score) for assessing the prognosis of patients with cirrhosis and acute GIB. Our study has the following notable features: (1) the data was obtained from multiple institutions in China; (2) a large number of patients were included; (3) the variables used for this model were readily available in clinical practice; (4) CAGIB score had a greater predictive performance than other conventional models in both training and validation cohorts; and (5) the predictive performance of CAGIB score was further validated.

CAGIB score includes two clinical variables (i.e., diabetes and HCC). Diabetes is a worldwide pandemic with a prevalence of 9.4% in the USA [14] and 11.6% in China [15]. Increasing evidence suggests a close relationship between diabetes and outcomes of liver disease. Diabetes increased the risks of liver cancer and chronic liver diseases [16,17,18] and was also associated with an increased risk of mortality in patients with cirrhosis [19]. Our previous single-center study also showed that diabetes was significantly associated with the prognosis of patients with cirrhosis and acute upper GIB, which is consistent with our current study [20]. On the other hand, HCC is one of the most common causes of cancer-related death [21]. And 80% of HCC patients have liver cirrhosis [22]. HCC can further aggravate portal pressure due to tumor compression and tumor thrombus formation and is considered as the independent predictor of death and re-bleeding in patients with cirrhosis and GIB [9, 23,24,25].

CAGIB score also includes four laboratory variables (i.e., TBIL, ALB, Scr, and ALT). Inclusion of TBIL, ALB, and Scr into this new model is easily understood, because they are important components of conventional scoring systems (i.e., MELD and Child–Pugh scores). Notably, a rapid increase in Scr level is often an acute critical condition indicating decreased kidney perfusion in patients with cirrhosis developing an acute GIB episode. Indeed, regardless of acute GIB, renal failure increases the mortality sevenfold in patients with cirrhosis [26]. In patients with cirrhosis and acute GIB, acute kidney injury is also an independent predictor for death [5, 27]. Besides, our study found that an increased ALT level was another independent predictor. In patients with cirrhosis and massive GIB, nearly all organs, including liver, are in an ischemic state after acute blood loss [28]. Hypoxic hepatitis, which is characterized by a rapid rise in serum aminotransferases due to liver cell necrosis by mitochondrial damage and DNA fragmentation [29], can be frequently observed in patients with cirrhosis and variceal bleeding [30] and negatively influence the patients’ outcomes [31].

A major limitation was that CAGIB score could not be compared with conventional scoring systems for GIB, such as Rockall, GBS, and AIMS65 scores, because not all patients underwent endoscopy. Second, for some patients, the source of GIB was unclear due to lack of the relevant endoscopy data. Thus, the association of sources of acute GIB with the mortality was not explored in the current study. Third, the stage of HCC was not extracted in our study. Fourth, the potential heterogeneity in the treatment selection among the participating centers should be acknowledged.

Conclusions

We developed and validated the CAGIB score to predict the in-hospital death of patients with cirrhosis and acute GIB. A CAGIB score of greater than − 4.6646 suggested a high risk of in-hospital death in liver cirrhosis with acute GIB. On the basis of the CAGIB score, physicians may also pay attention to the management of diabetes, improvement of liver and renal function, and supplementation of human albumin solution for patients with cirrhosis and acute GIB.