Analysis of risk factors and a prediction model for rebleeding in liver cirrhosis patients with esophagogastric variceal bleeding

Abstract

BACKGROUND

Liver cirrhosis with esophagogastric variceal bleeding (EGVB) carries a high oneyear rebleeding rate (30%-50%) despite current therapies. Existing predictive tools lack specificity or invasiveness, necessitating a simple noninvasive model for rebleeding risk stratification.

AIM

To identify rebleeding risk factors and develop a noninvasive nomogram for predicting oneyear EGVB rebleeding in cirrhotic patients.

METHODS

This retrospective study enrolled 130 patients with cirrhosis and EGVB who were treated at the Second Affiliated Hospital of Hainan Medical University in 2024. All patients were followed up by telephone for 1-year. The sample size was determined based on the eventspervariable principle commonly used in logistic regression, aiming for at least 10 events per candidate predictor, which is consistent with the number of events observed in this study. With rebleeding within 1 year as the study endpoint, patients were divided into a rebleeding group and a non-rebleeding group. Relevant clinical data were collected, including demographics, laboratory tests, imaging findings, and endoscopic features. Basic data and laboratory/imaging examination data were statistically analyzed. Based on LASSO regression and multivariate logistic regression analyses, a nomogram model was constructed to predict the risk of EGVB rebleeding. This his graphical calculating tool, based on a multivariate regression model, visually represents the predicted probability of a patient experiencing a specific clinical event. The model was evaluated using the receiver operating characteristic (ROC) curve.

RESULTS

Of the 130 patients included, the mean age was 54 years. Univariate analysis identified the following risk factors for EGVB rebleeding in cirrhosis: Age (P = 0.034), red blood cell (RBC) (P < 0.001), lymphocyte (LYM) (P = 0.031), red cell distribution width-coefficient of variation (P = 0.004), RBC distribution width coefficient of variation to LYM ratio (P = 0.005), neutrophil-to-lymphocyte ratio (P < 0.001), creatine kinase isoenzyme (P = 0.034), blood urea nitrogen (BUN) (P < 0.001), total cholesterol (P = 0.001), procalcitonin (P < 0.001), ascites (P < 0.001), and gallbladder wall thickening (P < 0.001). Indicators with statistically significant differences in univariate analysis were included in LASSO regression, and a total of four predictors were selected: BUN, ascites, RBC, and thickening of the gallbladder wall (TGBW). Multivariate logistic regression analysis showed that BUN [odds ratio (OR) = 1.116, P = 0.013], RBC (OR = 0.519, P = 0.009), ascites (OR = 0.235, P = 0.001), and TGBW (OR = 0.199, P < 0.001) were independent influencing factors for rebleeding. The nomogram model demonstrated good discrimination with an area under the ROC curve of 0.853 (95%CI: 0.791-0.915). The optimal cut-off value was 0.343, corresponding to a sensitivity of 0.871 and a specificity of 0.706. These findings indicate that elevated BUN, decreased RBC, presence of ascites, and TGBW increase the incidence of EGVB rebleeding.

CONCLUSION

RBC, BUN, ascites, and TGBW are independent risk factors for EGVB rebleeding in cirrhosis. A nomogram for predicting the risk of EGVB rebleeding was established.

Key Words: Liver cirrhosis; Esophageal and gastric varices; Esophagogastric variceal rebleeding; Prediction model

Core Tip: This single-center retrospective study enrolled 130 cirrhotic patients with esophagogastric variceal bleeding (EGVB) in 2024 and conducted a one-year follow-up, including 62 rebleeding cases and 68 non-rebleeding cases, aiming to address the high mortality of EGVB rebleeding and the lack of simple non-invasive predictive tools for Hainan population. The study identified four independent risk factors for rebleeding -decreased red blood cell, elevated blood urea nitrogen, ascites, and thickening of the gallbladder wall (TGBW); notably, TGBW was originally validated as a novel non-invasive imaging predictor with an odds ratio value of 0.199 (P < 0.001). A reliable nomogram prediction model was further constructed with an original area under the curve (AUC) of 0.853 and a bootstrap internal validation AUC of 0.855, presenting favorable calibration and clinical applicability. Since this model merely requires routine blood tests and ultrasonic indicators, it can realize convenient and non-invasive risk stratification to guide secondary clinical prevention for local patients. Nevertheless, this research is limited by its retrospective nature, single-center setting and relatively small sample size; and further multi-center prospective studies are required to verify the generalizability of the model.

INTRODUCTION

Cirrhosis is a histological condition characterized by regenerative nodules surrounded by fibrous bands, arising from chronic liver injury and potentially leading to portal hypertension and end-stage liver disease. In this pathological process, liver fibrosis play a key role, caused by persistent abnormalities in the normal wound healing response, resulting in continuous progression of fibrous tissue formation (connective tissue production and deposition). In cirrhosis, hepatic sinusoids become capillarized, the space of Disse is filled with scar tissue, and endothelial fenestrae disappear. These changes impair hepatocyte function and lead to portal hypertension and hepatocellular carcinoma[1,2]. Cirrhosis is widespread worldwide and can be caused by various etiologies, such as obesity, nonalcoholic fatty liver disease, high alcohol intake, hepatitis B or C, autoimmune diseases, cholestatic diseases, and iron or copper overload. Long-term inflammation causes healthy liver parenchyma to be replaced by fibrous tissue and regenerative nodules, leading to portal hypertension[3]. Chronic liver disease (CLD) cirrhosis is one of the leading causes of mortality, morbidity, disability-adjusted life years, and reduced quality of life[4]. Studies have shown that cirrhosis has become the world’s 11th leading cause of death, accounting for 1.2 million deaths per year (estimated in 2015)[5]. Another estimate indicates that in 2019, deaths caused by cirrhosis accounted for 2.4% of total global deaths. Globally, the most common causes of liver cirrhosis include chronic hepatitis B and C virus, metabolic dysfunction-associated steatotic liver disease (MASLD), and alcohol-associated liver disease[6]. Notably, the incidence of viral hepatitis cirrhosis in the Chinese population is significantly higher than in Western countries[7]. In China, hepatitis B virus (HBV) infection remains the main cause, but with the popularization of the hepatitis B vaccine, the number of patients with hepatitis B cirrhosis is gradually decreasing, and alcoholic cirrhosis has risen to become the second leading cause of end-stage liver disease. Undoubtedly, liver cirrhosis has become a major global public health problem.

Liver cirrhosis can lead to many adverse outcomes, such as portal hypertension, esophagogastric variceal bleeding (EGVB), ascites, hepatic encephalopathy, portal vein thrombosis, hepatorenal syndrome, hepatopulmonary syndrome, spontaneous bacterial peritonitis, and primary liver cancer. EGVB in cirrhosis refer to excessive pressure in the portal venous system, which forces blood to find alternative pathways, resulting in abnormal dilation, tortuosity, and thinning of the venous vessels in the esophagus and gastric fundus (i.e., varices), ultimately leading to rupture of these fragile vessel walls and potentially fatal bleeding. It is one of the most common and dangerous complications of liver cirrhosis. EGVB is a frequent cause of gastrointestinal bleeding and a common complication of liver cirrhosis, usually caused by tortuous dilation of esophageal and gastric veins due to portal hypertension. It is characterized by rapid onset, rapid progression, and high mortality. Portal hypertension is the main cause[8]. In compensated cirrhosis, EGVB develops at an annual rate of 8%, and the incidence is higher in decompensated cirrhosis. About 50% of patients with cirrhosis have EGVB, and the bleeding rate is 5%-15%[9]. Among them, 12% of patients with cirrhosis experience the first rupture of EGVB[10]. The risk of rebleeding and death after successful hemostasis in patients with cirrhosis and EGVB is very high. In recent years, with the continuous development of electronic gastroscopy technology, the acute-phase management of EGVB has become increasingly refined[11], but the risk of rebleeding after control of acute bleeding remains high. The latest guidelines indicate that the risk of rebleeding within 1 to 2 years after the initial EGVB episode is 60%-70%, with a mortality risk as high as 20%[12]. Therefore, after achieving initial hemostasis, secondary preventive measures must be taken, including pharmacological, endoscopic, and other treatment interventions. Without treatment, the rate of rebleeding within 2 years after the first EGVB is as high as 70%, making it a major cause of death in patients with cirrhosis. Therefore, preventing rebleeding is crucial. Once EGVB occurs, the patient’s quality of life declines sharply and the risk of rebleeding increases[13].

Current clinical treatments for EGVB in cirrhosis include drug therapy, interventional therapy, and endoscopic therapy. Endoscopic therapy can achieve rapid hemostasis by directly acting on the variceal veins, destroying their vascular structure, or forming an effective blood flow blocking mechanism[14,15]. However, even with active treatment and endoscopic intervention, the mortality and rebleeding rates of EGVB remain high. Therefore, analyzing the clinical characteristics of EGVB and determining its risk factors are very important for prevention, early intervention strategies, and individualized treatment of patients with cirrhosis.

Currently, gastroscopy remains the gold standard for diagnosing EGVB and bleeding. The hepatic venous pressure gradient (HVPG) obtained by measuring wedged and free hepatic venous pressure via jugular vein catheterization is the internationally recommended gold standard for diagnosing cirrhotic portal hypertension. Both gastroscopy and HVPG measurement have limitations such as invasiveness, patient intolerance, potential to induce bleeding, and high technical requirements. However, currently, most research focuses on initial EGVB bleeding, and non-invasive methods for predicting and evaluating rebleeding in patients are not well-established. The internationally recognized non-invasive method for predicting high-risk EGVB is the Baveno VII criteria, which include platelet count (PLT) and liver stiffness measured by transient elastography. However, this prediction method cannot directly predict the risk of gastrointestinal rebleeding[16,17]. Therefore, the purpose of this study is to analyze the risk factors for rebleeding in patients with cirrhosis and EGVB in Hainan; to screen for independent predictors associated with rebleeding; and to construct and validate a non-invasive and simple prediction model for rebleeding. Focusing on the core goal of early prediction of rebleeding in EGVB patients, through accurate risk identification and early intervention, we aim to reduce the incidence and mortality of rebleeding, thereby providing an important basis for improving clinical treatment efficacy and patient prognosis.

MATERIALS AND METHODS

Study population

This study was a retrospective analysis. Patients with cirrhosis and first-time endoscopically diagnosed EGVB admitted to the Department of Gastroenterology, the Second Affiliated Hospital of Hainan Medical University during 2024 were collected. After applying strict inclusion and exclusion criteria, a total of 130 EGVB patients were included.

The required sample size was prospectively estimated using the single-population proportion formula. Based on a previous Chinese single-center retrospective cohort study reporting a 1-year rebleeding rate of 18.0% among cirrhotic patients receiving secondary prevention (Liu et al[18]), the expected rebleeding rate was set at P = 0.18. With a two-sided significance level of α = 0.05, [Z (1-α/2) = 1.96] and a permissible absolute error of d = 0.07, the initial sample size was calculated as n₀ = [Z (1-α/2)]² × p (1-p)/d² = 1.96² × 0.18 × (1-0.18)/0.07² = 115.72, which was rounded up to 116 evaluable cases. To account for a 10% potential rate of ineligibility, incomplete data, or loss to follow-up, the final required sample size was determined as n = 116/(1-0.10) = 128.89, rounded up to 129 cases. Accordingly, we ultimately enrolled 130 patients, meeting this prespecified target.

Furthermore, to ensure the robustness of the multivariate logistic regression analysis, we also adhered to the events-per-variable (EPV) principle, which recommends at least 10 outcome events per candidate predictor. In our cohort, 62 rebleeding events were observed during the 1-year follow-up, and the final model incorporated 4 independent predictors, yielding an EPV of 15.5 (i.e., 62/4), which substantially exceeds the recommended minimum and supports reliable model development. These 130 patients were followed up by telephone for 1 year. They were divided into a rebleeding group (62 patients) and a non-rebleeding group (68 patients) based on whether rebleeding occurred within 1 year (Figure 1).

Figure 1 Data collection and classification flow chart. EGVB: Esophagogastric variceal bleeding.

Definitions

Criteria for diagnosing bleeding: Presence of portal hypertension due to cirrhosis, and esophagogastroduodenoscopy revealing direct signs of active bleeding: Active bleeding from varices, varices with “white nipple-like” material, or varices with adherent clots; endoscopic signs of direct bleeding are the “gold standard” for diagnosis. Diagnosis can be made when variceal bleeding or direct bleeding is observed endoscopically[19,20].

The core definition of EGVB rebleeding is the Recurrence of active bleeding after the acute bleeding episode caused by esophageal and gastric varices in cirrhotic patients has been controlled by treatment. Clinically, recurrence of active bleeding consistent with variceal bleeding is usually judged by the following indicators: Recurrence of hematemesis (bright red or coffee-ground), melena (tarry stool) or dark red bloody stool; a decrease in systolic blood pressure of more than 20 mmHg from a stable baseline, or an increase in heart rate of more than 20 beats/minutes from a stable baseline, accompanied by dizziness, palpitations, cold sweats, and other pre-shock or shock manifestations; in the absence of blood transfusion, a decrease in hemoglobin concentration of more than 30 g/L, or requiring more than 2 units of packed red blood cells (RBC) to maintain hemoglobin stability.

Diagnostic criteria

Diagnostic criteria for liver cirrhosis: (1) Liver histology meeting the criteria for cirrhosis; (2) Endoscopic evidence of esophagogastric or gastrointestinal ectopic varices, excluding non-cirrhotic portal hypertension; (3) Imaging evidence of cirrhosis or portal hypertension; and (4) For patients without histological, endoscopic, or imaging examination, meeting 2 of the following 4 items: (a) PLT < 100 × 10⁹/L, unexplained by other reasons; (b) Serum albumin < 35 g/L, excluding kidney disease; (c) International normalized ratio (INR) > 1.3 or prolonged prothrombin time (PT); and (d) Aspartate aminotransferase (AST) to PLT ratio index > 2[21].

Diagnostic criteria for esophagogastric varices: Endoscopy: This is the most valuable method for diagnosing EGVB. It allows direct observation of the presence of varices in the esophagus and gastric fundus and assessment of their degree and extent. Endoscopic EGVB can be graded based on location, form, fundamental tone, red color signs, etc. Gastric varices were classified according to the Sarin classification. EGVB bleeding can be diagnosed when active bleeding, “white nipple sign” on varices, clots on varices, or varices without other potential causes of bleeding are found on endoscopy[22].

Liver stiffness measurement (LSM) combined with PLT[23]: This is the most effective non-invasive method for diagnosing clinically significant portal hypertension (CSPH). CSPH can be diagnosed if one of the following conditions is met: Decompensated cirrhosis (e.g., significant ascites, variceal bleeding, or severe hepatic encephalopathy); ultrasound, computed tomography, or magnetic resonance imaging results suggesting the formation of portosystemic collateral circulation; endoscopic examination suggesting the presence of EGVB; LSM ≥ 25 kPa, or LSM 20-25 kPa with PLT < 150 × 10⁹/L, or LSM 15-20 kPa with PLT < 110 × 10⁹/L.

Inclusion criteria

(1) Age 18-75 years, male or female; (2) Liver cirrhosis due to various causes; (3) All subjects were clearly diagnosed with EGVB and underwent endoscopic treatment in our hospital; (4) Patients with complete relevant medical history data; and (5) Signed informed consent.

Exclusion criteria

(1) Advanced hepatocellular carcinoma with portal vein tumor thrombus formation; (2) Hepatic encephalopathy, coma, or other conditions where patients could not accept or cooperate with endoscopic treatment; (3) Possibility of ectopic embolism due to large gastrorenal or splenorenal shunts; (4) Severe cardiovascular diseases, including acute myocardial infarction, severe conduction block, and heart failure; (5) Severe renal insufficiency; (6) Patients with other end-stage diseases; (7) Pregnant or lactating women; (8) Surgical hemostasis treatment; (9) Prior portosystemic shunt and disconnection surgery, such as transjugular intrahepatic portosystemic shunt; and (10) Those lost to follow-up.

Collection of clinical data

Basic information: Gender, age, etiology of cirrhosis, smoking history, drinking history, hypertension, diabetes.

Blood routine: White blood cell count, RBC count, PLT, lymphocyte (LYM) count, RBC distribution width-standard deviation, RBC distribution width-coefficient of variation, ratios of RBC distribution width coefficient of variation to LYM count and to PLT, platelet-to-lymphocyte ratio (PLR), neutrophil-to-lymphocyte ratio (NLR), platelet hematocrit.

Liver function: Albumin, albumin-to-globulin ratio, direct bilirubin (DBIL), total bilirubin (TBIL), AST, alanine aminotransferase (ALT), total bile acid, creatine kinase isoenzyme (CK-MB).

Blood lipids: Triglyceride (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C).

Renal function: Blood urea nitrogen (BUN), creatinine (CR), uric acid (UA).

Coagulation function: PT, INR, activated partial thromboplastin time (APTT), thrombin time (TT), D-dimer, prothrombin activity (PTA), fibrinogen (FIB).

Imaging workup: Ascites, thickening of the gallbladder wall (TGBW).

Statistical analysis

SPSS 26.0 software was used for data analysis. Measurement data conforming to a normal distribution were expressed as mean ± SD and compared using the independent samples t-test. Measurement data not conforming to normal distribution were expressed as median and interquartile range and compared using the rank-sum test. Count data were analyzed using the χ² test. A logistic regression model was used to analyze the risk factors for rebleeding in patients with cirrhosis and EGVB. Indicators with P < 0.05 in the univariate analysis were included to screen for independent predictors. R software (version 3.6.1) was used to develop and visualize the nomogram prediction model for rebleeding. The receiver operating characteristic (ROC) curve was used to evaluate the model. The model’s discrimination was assessed by the area under the ROC curve (AUC), sensitivity, specificity, and optimal cut-off value determined by the Youden index. Calibration was evaluated using the Hosmer-Lemeshow test and calibration plots. Clinical utility was assessed via decision curve analysis. Internal validation was performed using the Bootstrap method with 1000 resamples. P < 0.05 indicated a statistically significant difference.

RESULTS

Comparison of basic data between non-rebleeding group and rebleeding group

Among the 130 patients included, the average age was 54 years, including 111 males and 19 females. There were 98 cases with smoking history, 98 with drinking history, 23 with hypertension, 53 with hepatitis B cirrhosis, 45 with alcoholic cirrhosis, 6 with hepatitis C cirrhosis, and 26 with other causes. In the univariate analysis of the samples, there was a significant difference in age (P = 0.034) between the non-rebleeding group and the rebleeding group (P < 0.05). Specific results are shown in Table 1.

Table 1 Comparison of basic data between non-rebleeding group and rebleeding group, n (%)/median (interquartile range).

Variable	Non-rebleeding group (n = 68)	Rebleeding group (n = 62)	Statistic	P value
Age (years)	51.50 (44.75-59.50)	54.00 (49.00-67.75)	Z = -2.120	0.034
Gender			χ² = 1.050	0.305
Male	56 (82.35)	55 (88.71)
Female	12 (17.65)	7 (11.29)
Relevant histories
Smoking history	50 (73.53)	48 (77.42)	χ² = 0.264	0.607
Drinking history	45 (66.18)	48 (77.42)	χ² = 2.013	0.156
Past medical history
History of hypertension	15 (22.06)	8 (12.90)	χ² = 1.867	0.172
History of diabetes	53 (77.94)	54 (87.10)	χ² = 2.816	0.093
Etiology of cirrhosis				0.863
HBV-related cirrhosis	30 (44.12)	23 (37.10)
Alcoholic cirrhosis	23 (33.82)	22 (35.48)
HCV-related cirrhosis	3 (4.41)	3 (4.84)
Other	12 (17.65)	14 (22.58)

HBV: Hepatitis B virus; HCV: Hepatitis C virus.

Comparison of laboratory tests between the non-rebleeding group and the rebleeding group

In the univariate analysis, there were significant differences in RBC (P < 0.001), LYM (P = 0.031), red cell distribution width-coefficient of variation (RDW-CV) (P = 0.004), RBC distribution width coefficient of variation to LYM ratio (P = 0.005), NLR (P < 0.001), CK-MB (P = 0.034), BUN (P < 0.001), TC (P = 0.001), procalcitonin (PCT) (P < 0.001) and other indicators between the non-rebleeding group and the rebleeding group (P < 0.05). The results showed that decreased RBC, decreased LYM, increased RDW-CV, increased RDWCVtolymphocyte ratio, increased NLR, increased CK-MB, decreased TC, and increased BUN increased the risk of EGVB rebleeding in cirrhosis. There were no significant differences in white blood cell, PLT, RBC distribution width standard deviation, RBC distribution width coefficient of variation to PLT ratio, PLR, A/G, DBIL, TBIL, AST, ALT, Platelet hematocrit, ALb, CR, UA, TG, HDL-C, LDL-C, PTA, PT, INR, APTT, TT, FIB, D-dimer and other variables (P > 0.05). Specific results are shown in Table 2.

Table 2 Comparison of laboratory tests between non-rebleeding group and rebleeding group, median (interquartile range).

Variable	Non-rebleeding group (n = 68)	Rebleeding group (n = 62)	Statistic	P value
WBC	4.44 (3.25-6.57)	5.57 (3.73-7.17)	Z = -1.522	0.128
RBC	3.75 (3.17-4.22)	2.75 (2.40-3.36)	Z = -4.247	< 0.001
PLT	85.00 (65.50-124.25)	84.50 (66.25-115.75)	Z = -0.261	0.794
LYM	1.15 (0.80-1.46)	0.95 (0.61-1.22)	Z = -2.156	0.031
RDW-SD	46.85 (43.25-53.25)	48.55 (44.27-54.35)	Z = -0.958	0.338
RDW-CV	14.85 (13.70-18.22)	16.55 (15.60-19.75)	Z = -2.862	0.004
RDW-CV/LYM	14.59 (9.89-22.41)	18.88 (13.54-27.71)	Z = -2.832	0.005
RDW-CV/PLT	0.17 (0.11-0.28)	0.20 (0.14-0.28)	Z = -1.369	0.171
PLR	74.05 (52.62-106.95)	92.80 (60.77-165.13)	Z = -1.834	0.067
NLR	2.44 (1.51-3.82)	4.01 (2.73-6.40)	Z = -4.093	< 0.001
Platelet hematocrit	0.09 (0.06-0.13)	0.08 (0.06-0.10)	Z = -0.507	0.612
ALb	33.00 (28.50-36.32)	31.80 (29.05-34.55)	Z = -0.848	0.396
A/G	1.11 (0.89-1.36)	1.27 (0.97-1.47)	Z = -1.834	0.067
DBIL	12.30 (7.33-34.57)	15.35 (9.80-28.15)	Z = -0.739	0.460
TBIL	28.05 (15.15-62.15)	27.05 (19.40-44.55)	Z = -0.019	0.985
AST	42.50 (29.00-59.50)	39.50 (28.25-90.50)	Z = -0.893	0.372
ALT	28.00 (18.00-45.25)	32.00 (19.25-45.00)	Z = -0.415	0.678
CK-MB	30.50 (20.00-41.25)	35.00 (23.00-51.25)	Z = -2.124	0.034
BUN	4.20 (3.25-5.41)	6.84 (4.81-10.10)	Z = -4.561	< 0.001
CR	69.00 (59.00-85.25)	67.00 (58.25-91.50)	Z = -0.049	0.961
UA	284.00 (241.00-366.50)	299.00 (228.50-386.50)	Z = -0.040	0.968
TG	0.87 (0.66-1.24)	0.85 (0.73-1.43)	Z = -0.566	0.571
TC	3.88 (3.21-4.41)	3.12 (2.53-4.01)	Z = -3.242	0.001
HDL-C	1.08 (0.76-1.31)	0.88 (0.62-1.21)	Z = -1.485	0.138
LDL-C	2.00 (1.39-2.48)	1.71 (1.38-2.58)	Z = -0.145	0.885
PTA	58.00 (48.15-70.20)	56.60 (48.70-70.20)	Z = -0.165	0.869
PT	14.25 (13.00-15.77)	14.80 (13.03-16.40)	Z = -0.783	0.433
INR	1.24 (1.13-1.44)	1.30 (1.13-1.45)	Z = -0.879	0.379
APTT	33.15 (28.35-40.58)	30.95 (26.10-36.05)	Z = -1.925	0.054
TT	19.05 (17.70-20.42)	18.80 (17.55-20.28)	Z = -0.737	0.461
FIB	1.79 (1.22-2.21)	1.83 (1.24-2.29)	Z = -0.056	0.955
D-D	1.10 (0.38-4.06)	1.12 (0.52-3.65)	Z = -0.012	0.991
PCT	0.07 (0.03-0.12)	0.21 (0.10-0.37)	Z = -4.541	< 0.001

WBC: White blood cell; RBC: Red blood cell; PLT: Platelet; LYM: Lymphocyte; RDW-SD: Red blood cell distribution width-standard deviation; RDW-CV: Red blood cell distribution width-coefficient of variation; RDW-CV/LYM: Red blood cell distribution width coefficient of variation to lymphocyte ratio; RDW-CV/PLT: Red blood cell distribution width coefficient of variation to platelet count ratio; PLR: Platelet-to-lymphocyte ratio; NLR: Neutrophil-to-lymphocyte ratio; ALb: Albumin; A/G: Albumin-to-globulin ratio; DBIL: Direct bilirubin; TBIL: Total bilirubin; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; CK-MB: Creatine kinase isoenzyme; BUN: Urea nitrogen; CR: Creatinine; UA: Uric acid; TG: Triglyceride; TC: Total cholesterol; HDL-C: High-density lipoprotein cholesterol; LDL-C: Low-density lipoprotein cholesterol; PTA: Prothrombin time activity; PT: Prothrombin time; INR: International normalized ratio; APTT: Activated partial thromboplastin time; TT: Thrombin time; FIB: Fibrinogen; D-D: D-dimer; PCT: Procalcitonin.

Imaging comparison between non-rebleeding group and rebleeding group

Univariate analysis found statistically significant differences in the presence of TGBW and ascites between the non-rebleeding group and the rebleeding group (P < 0.05), indicating that cirrhotic patients with EGVB bleeding were more likely to experience rebleeding if they had TGBW and ascites (Table 3).

Table 3 Imaging comparison between non-rebleeding group and rebleeding group, n (%).

Variable	Non-rebleeding group (n = 68)	Rebleeding group (n = 62)	Statistic	P value
TGBW			χ² = 25.655	< 0.001
Present	16 (23.53)	42 (67.74)
Absent	52 (76.47)	20 (32.26)
Ascites presence			χ² = 20.716	< 0.001
Present	20 (29.41)	43 (69.35)
Absent	48 (70.59)	19 (30.65)

TGBW: Thickening of the gallbladder wall.

LASSO regression screening of predictors

Indicators with statistically significant differences in the univariate analysis were included in the LASSO regression, total of 12 indicators: Age, RBC, LYM, RDW-CV, RBC distribution width coefficient of variation to LYM count ratio, NLR, TC, CK-MB, BUN, PCT, TGBW, and ascites. Ten-fold cross-validation was used to determine one standard error (1se) of the optimal value of the tuning parameter (λ). A total of 4 predictors were screened out: BUN, ascites, RBC, and TGBW, as shown in Figures 2 and 3.

Figure 2 LASSO regression cross validation.

Figure 3 LASSO regression path diagram.

Multivariate logistic regression modeling for non-rebleeding group and rebleeding group

The variables selected by LASSO were used as independent variables (BUN, RBC, ascites, TGBW), and the presence or absence of rebleeding was used as the dependent variable. Multivariate logistic regression analysis with stepwise backward regression was performed. Multivariate logistic regression analysis showed that BUN [odds ratio (OR) = 1.116, P = 0.013], RBC (OR = 0.519, P = 0.009), ascites (OR = 0.235, P = 0.001) and TGBW (OR = 0.199, P < 0.001) were independent risk factors for EGVB rebleeding. The results are shown in Table 4.

Table 4 Results of multivariate logistic regression analysis.

Variables	β	SE	Z	P value	OR (95%CI)
BUN	0.109	0.044	2.473	0.013	1.116 (1.023-1.217)
RBC	-0.656	0.251	-2.617	0.009	0.519 (0.317-0.848)
Ascites
Present					1.000 (Reference)
Absent	-1.446	0.455	-3.178	0.001	0.235 (0.097-0.574)
TGBW presence
Present					1.000 (Reference)
Absent	-1.617	0.457	-3.540	< 0.001	0.199 (0.081-0.486)

BUN: Blood urea nitrogen; RBC: Red blood cell; TGBW: Thickening of the gallbladder wall; OR: Odds ratio.

A nomogram was used to visualize the multivariate logistic regression model. The results are shown in the figure: For each unit increase in BUN, the score increased by approximately 11.11 points; ‘Yes’ for ascites scored 0 points, ‘No’ scored 29.35 points; for every unit increase in RBC, the score decreased by approximately 6.66 points; ‘Yes’ for TGBW scored 0 points, ‘No’ scored 32.81 points. The scores corresponding to all variables are summed to obtain a total score, and the corresponding risk of rebleeding can be queried using the nomogram, providing an intuitive basis for clinical prediction (Figure 4).

Figure 4 Risk prediction nomogram. BUN: Blood urea nitrogen; RBC: Red blood cell; TGBW: Thickening of the gallbladder wall.

Model evaluation

The performance of the nomogram model was evaluated by discrimination (ROC curve), calibration (calibration curve, Hosmer-Lemeshow test), and clinical utility (clinical decision curve). The results are as follows.

The model’s cut-off value was 0.343. The area under the curve (AUC) was 0.853 (95%CI: 0.791-0.915), sensitivity was 0.871 (0.788-0.954), specificity was 0.706 (0.598-0.814), accuracy was 0.785 (0.782-0.787), and the Kappa value was 0.572 (0.434-0.710). This indicates that the model has a high ability to distinguish between positive and negative samples, can effectively predict the occurrence of the target event, and the model has good consistency and clinical application value.

In terms of calibration, the calibration curve for the training set is shown in Figure 5. The predicted probability fitting curve for the training set closely follows the ideal curve, indicating that the model fitting result is close to the actual situation. The Hosmer-Lemeshow test (χ² = 10.341, P = 0.247) result shows a P-value greater than 0.05, indicating good calibration ability for the model.

Figure 5 Receiver operating characteristic curve of training set. AUC: Area under the curve.

In terms of clinical predictive benefit, the clinical decision curves for the training set are shown in Figures 6 and 7. In the training set, the net benefit rate for patients using the model was significantly higher than that of the two comparison curves, and the risk threshold ranged from 12% to 91%, indicating good clinical utility of the model.

Figure 6 Calibration curve of training set.

Figure 7 Clinical decision curve of training set.

The Bootstrap internal validation method was used to validate the model, shown in Figure 8. The number of resamples was 1000. The average AUC from the internal validation of the model was 0.855 (0.791, 0.912), indicating that the model has good discriminative ability.

Figure 8 Bootstrap receiver operating characteristic curve. ROC: Receiver operating characteristic.

DISCUSSION

EGVB is a serious and life-threatening complication in cirrhosis patients. Approximately 30% of patients with cirrhosis have esophageal varices at diagnosis[24], and this proportion can increase to 90% within 10 years. Variceal rupture is one of the most fatal complications of cirrhosis[25]. EGVB rebleeding is an important cause of death in cirrhosis patients, capable of causing massive hemorrhage in a short time, leading to hemorrhagic shock and multiple organ failure. Early identification of risk factors for rebleeding and timely intervention or treatment are of extremely important clinical value for improving patient prognosis. For cirrhosis patients, timely diagnosis and treatment are key to ensuring their safety. The goal of this study was to construct a non-invasive risk prediction model to assess the likelihood of EGVB rebleeding in cirrhosis.

Based on strict inclusion and exclusion criteria, the enrolled patients were categorized into a rebleeding group and a non-rebleeding group. Detailed statistical analysis was performed on the hematological indicators and imaging findings of the included patients. Univariate analysis revealed that the following serological data: RBC, LYM, RDW-CV, RDW-CV/LYM, NLR, CK-MB, BUN, TC, HDL-C, LDL-C, PCT-and imaging data-ascites, TGBW-showed statistically significant differences (P < 0.05). Other data showed no statistical differences. Multivariate Logistic regression analysis found that RBC, BUN, TGBW, and ascites were independent risk factors for rebleeding in patients with cirrhosis and EGVB.

The mechanistic basis for these associations warrants further discussion. Rupture of EGVB in cirrhosis leads to rapid, massive upper gastrointestinal bleeding, causing a sharp decrease in total RBC and resulting in hemorrhagic anemia. Consequently, a complete blood count typically reveals a decreased RBC count, and clinicians monitor RBC-related indicators to assess bleeding severity and treatment response. Chronic anemia reduces RBC count and blood viscosity, potentially increasing bleeding risk[26]. In this study, the rebleeding group exhibited a significantly lower RBC level compared to the non-rebleeding group (Z = -4.247, P < 0.001), corroborating the role of chronic anemia as a predisposing factor for variceal rupture. Decreased blood viscosity reduces the flow resistance within blood vessels, leading to a compensatory increase in blood flow velocity. In already fragile vessels with portal hypertension and varices, increased flow velocity exerts greater fluid shear stress on the vessel wall[27]. This increased stress may subject the already weak variceal wall to greater pressure, thereby increasing its risk of rupture. CLD and portal hypertension themselves lead to a systemic hyperdynamic circulatory state. Chronic anemia can further exacerbate this hyperdynamic state, leading to a relative increase in portal venous blood flow, potentially further increasing portal pressure and indirectly increasing the risk of variceal rupture[28].

The formation mechanism of ascites in cirrhosis is complex, wherein portal hypertension-induced increased hydrostatic pressure plays a decisive role. In addition, hypoalbuminemia, activation of the renin-angiotensin-aldosterone system, reduced hepatic inactivation of aldosterone and antidiuretic hormone, and impaired lymphatic also play important roles in ascites formation. This study identified ascites as an independent risk factor for EGVB rebleeding.

In patients with cirrhosis, BUN and blood ammonia levels are often pathophysiologically interrelated. Intestinal ammonia production is the main source of blood ammonia, and intestinal blood accumulation after bleeding greatly increases ammonia generation. Hyperammonemia has been shown to stimulate astrocytes and vascular endothelial cells to produce more nitric oxide. Nitric oxide is a potent vasodilator. Widespread vasodilation of splanchnic vessels can further increase portal venous blood flow, thereby exacerbating portal hypertension[29]. This is an important mechanism of the hyperdynamic circulatory state in cirrhotic portal hypertension. The higher the portal pressure, the more prone EGVB are to rupture and bleed.

This study found that TGBW visible on patient CT is an independent risk factor. The critical value for gallbladder wall thickness in the presence of EVs varies between 3.1 mm and 4.35 mm, with variable sensitivity of 46%-90.9%[30]. Patients with cirrhosis with a gallbladder wall thickness ≥ 3.4 mm have a significantly higher cumulative incidence of liver decompensation events (including variceal bleeding)[31]. This provides direct evidence for its role as a predictor of prognosis in cirrhosis. One study demonstrated that a gallbladder wall thickness > 4 mm predicted EGVB bleeding with a sensitivity of 90.9% and a specificity of 82.5%, outperforming traditional indicators like PLT[32]. That study validated the value of gallbladder wall thickness in risk stratification of portal hypertension patients using multicenter data. Gallbladder wall thickness is also related to the degree of varices. In viral cirrhosis, a gallbladder wall thickness > 4 mm can predict moderate to severe varices (sensitivity 62%, specificity 90%). Another study found that when gallbladder wall thickness > 3.5 mm, the accuracy for predicting high-grade varices (grade III-IV) reached 77.1%. In viral cirrhosis, a gallbladder wall thickness > 4 mm can predict moderate to severe varices (sensitivity 62%, specificity 90%)[33]. Gallbladder wall thickening (GBT) may be accompanied by chronic inflammation (e.g., cholecystitis). The release of inflammatory factors can further damage vascular endothelium, increasing the fragility of EGVB. Although existing studies have not clearly defined the independent role of inflammation, cirrhosis patients often have associated chronic cholecystitis, which may jointly promote bleeding risk along with portal hypertension.

The practical value of our nomogram lies in its reliance on universally available, cost-effective parameters that require no additional invasive procedures or specialized equipment. To contextualize our model, it is instructive to compare it with established prognostic scores. The Child-Pugh and model for end-stage liver disease (MELD) scores are the most widely used systems for assessing overall liver disease severity and predicting mortality[34,35]. However, both were primarily designed to estimate short-term survival rather than specifically to predict variceal rebleeding. The Child-Pugh score includes subjective components (ascites, encephalopathy), introducing inter-observer variability, whereas MELD, though objective, uses bilirubin, INR, and CR-markers of global function that may be less sensitive to local hemodynamic drivers of variceal rupture.

The AIMS65 score, which includes albumin, INR, altered mental status, systolic blood pressure, and age > 65 years, has been validated for predicting in-hospital mortality in upper GI bleeding[36,37]. However, it was not specifically designed for variceal bleeding and includes parameters that overlap with Child-Pugh and MELD. More importantly, none of these scores incorporate markers of portal hypertensive congestion, such as TGBW, or direct indicators of the hyperdynamic circulatory state, such as RBC and BUN. Our model’s strength lies in its pathophysiological focus: RBC reflects anemia-induced hyperdynamic flow; BUN captures the interplay between prerenal azotemia and hyperammonemia; ascites and TGBW directly signal advanced portal hypertension and its systemic manifestations. This mechanistic specificity may render our nomogram more sensitive to rebleeding risk than global liver function scores. Nevertheless, head-to-head prospective comparisons are needed to establish whether our model offers incremental predictive value over Child-Pugh, MELD, or AIMS65 in the specific context of EGVB rebleeding.

Several limitations of this study must be acknowledged. First, the retrospective, single-center design, with a relatively modest sample size of 130 patients, may limit the generalizability of our findings. While the sample size was adequate for the number of predictors analyzed (EPV > 15), the single-center setting introduces potential selection bias, and the modest sample size may have limited statistical power to detect smaller effect sizes or to perform extensive subgroup analyses. Second, the model requires external validation in an independent, geographically diverse cohort, as predictive models often exhibit performance degradation when applied to new populations. Third, the 1-year follow-up was conducted via telephone, which may be subject to recall bias, and we cannot definitively exclude the possibility that some rebleeding events were missed, particularly if patients sought care at other institutions. Fourth, TGBW measurement was not standardized across all imaging studies, as this was a retrospective analysis, potentially introducing variability in this predictor. Fifth, we did not collect data on LSM by transient elastography or the baseline grade and size of varices, both of which are known predictors of bleeding risk[15,16]. Incorporation of these parameters might have further improved model performance. Finally, our study population was predominantly from Hainan, China, with a high proportion of HBV-related cirrhosis. This etiological distribution may limit the applicability of our model to populations with alcohol-related or MASLD-related cirrhosis, which are increasingly prevalent in Western countries.

In conclusion, our study demonstrates that RBC, BUN, ascites, and TGBW are independent risk factors for EGVB rebleeding in cirrhotic patients. The nomogram constructed from these four parameters exhibits good discriminative ability (AUC = 0.853) and calibration, offering a practical, non-invasive tool for individualized risk stratification. While these findings are promising, external validation in larger, multi-center cohorts is essential to confirm the model’s robustness and to compare its performance against existing prognostic scores before it can be recommended for routine clinical use.

CONCLUSION

Univariate analysis identified multiple potential risk factors for EGVB rebleeding; however, multivariate analysis confirmed that RBC, BUN, ascites, and GBT are independent risk factors. RBC, BUN, TGBW, and ascites are independent risk factors for EGVB rebleeding in liver cirrhosis. A nomogram prediction model for EGVB risk was established.