Liang LX, Liang X, Zeng Y, Wang F, Yu XK. Establishment and validation of a nomogram for predicting esophagogastric variceal bleeding in patients with liver cirrhosis. World J Gastroenterol 2025; 31(9): 102714 [DOI: 10.3748/wjg.v31.i9.102714]
Corresponding Author of This Article
Xue-Ke Yu, MD, Department of Gastroenterology, The Affiliated Changsha Central Hospital, Hengyang Medical School, University of South China, No. 161 Shaoshan South Road, Changsha 410008, Hunan Province, China. yuxueke2023@163.com
Research Domain of This Article
Gastroenterology & Hepatology
Article-Type of This Article
Retrospective Study
Open-Access Policy of This Article
This article is an open-access article which was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/
Lun-Xi Liang, Fen Wang, Department of Gastroenterology, The Third Xiangya Hospital, Central South University, Changsha 410013, Hunan Province, China
Lun-Xi Liang, Ya Zeng, Xue-Ke Yu, Department of Gastroenterology, The Affiliated Changsha Central Hospital, Hengyang Medical School, University of South China, Changsha 410008, Hunan Province, China
Lun-Xi Liang, Fen Wang, Hunan Key Laboratory of Nonresolving Inflammation and Cancer, The Third Xiangya Hospital, Central South University, Changsha 410006, Hunan Province, China
Xiao Liang, School of Clinical Medicine, Changsha Medical University, Changsha 410200, Hunan Province, China
Author contributions: Liang LX, Wang F and Yu XK designed the research study; Liang LX, Liang X and Zeng Y contributed data collection and analysis; Liang LX wrote the first draft of the manuscript; Wang F and Yu XK conceived and supervised the manuscript; All authors have read and approved the final manuscript.
Supported by the National Natural Science Foundation of China, No. 82270594; the National Natural Science Foundation for Youths of China, No. 82103151; the Fundamental Research Funds for the Central Universities of Central South University, No. 2022ZZTS0265; and the Graduate Research Innovation Project of Hunan Province, No. CX20220347.
Institutional review board statement: The study protocol was approved by the Ethics Committee of Changsha Central Hospital (Changsha Central Hospital Affiliated with the University of South China, approval No. 2022-S0019).
Informed consent statement: The necessity for obtaining informed written consent was waived due to the retrospective nature of the study, which did not implicate the privacy or commercial interests of the patients. Additionally, measures were implemented to anonymize biological samples, and a stringent data security management and technical protection system was established for the storage, utilization, and dissemination of biological samples and data, thereby ensuring the security of data and personal information.
Conflict-of-interest statement: The authors declare that they have no conflict of interest.
Data sharing statement: Data used and/or analyzed in the current study could be acquired from the corresponding author upon reasonable request.
Open Access: This article is an open-access article that was selected by an in-house editor and fully peer-reviewed by external reviewers. It is distributed in accordance with the Creative Commons Attribution NonCommercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: https://creativecommons.org/Licenses/by-nc/4.0/
Corresponding author: Xue-Ke Yu, MD, Department of Gastroenterology, The Affiliated Changsha Central Hospital, Hengyang Medical School, University of South China, No. 161 Shaoshan South Road, Changsha 410008, Hunan Province, China. yuxueke2023@163.com
Received: October 28, 2024 Revised: January 6, 2025 Accepted: January 18, 2025 Published online: March 7, 2025 Processing time: 114 Days and 16.7 Hours
Abstract
BACKGROUND
Patients with decompensated liver cirrhosis suffering from esophagogastric variceal bleeding (EGVB) face high mortality.
AIM
To investigate the risk factors for EGVB in patients with liver cirrhosis and establish a diagnostic nomogram.
METHODS
Patients with liver cirrhosis who met the inclusion criteria were randomly divided into training and validation cohorts in a 6:4 ratio in this retrospective research. Univariate analysis, least absolute shrinkage and selection operator regression, and multivariate analysis were employed to establish the nomogram model. Calibration curve, the area under the receiver operating characteristic curve (AUC), and decision curve analysis (DCA) were applied to assess the discrimination, accuracy, and clinical practicability of the nomogram, respectively.
RESULTS
A total of 1115 patients were enrolled in this study. The nomogram was established based on white blood cells (P < 0.001), hemoglobin (P < 0.001), fibrinogen (P < 0.001), total bilirubin (P = 0.007), activated partial thromboplastin time (P = 0.002), total bile acid (P = 0.012), and ascites (P = 0.006). The calibration curve indicated that the actual observation results were in good agreement with the prediction results of the model. The AUC values of the diagnostic model were 0.861 and 0.859 in the training and validation cohorts, respectively, which were higher than that of the aspartate aminotransferase-to-platelet ratio index, fibrosis index based on 4 factors, and aspartate aminotransferase-to-alanine aminotransferase ratio. Additionally, DCA indicated that the net benefit value of the model was higher than that of the other models.
CONCLUSION
This research constructed and validated a nomogram with perfect performance for predicting EGVB events in patients with liver cirrhosis, which could help clinicians with timely diagnosis, individualized treatment, and follow-up.
Core Tip: This research retrospectively investigated the risk factors for esophagogastric variceal bleeding in patients with liver cirrhosis and established a diagnostic nomogram based on white blood cells, hemoglobin, fibrinogen, total bilirubin, activated partial thromboplastin time, total bile acid, and ascites. This model was further verified in the validation cohort and proved to perform well. The factors included in this model are non-invasive, inexpensive, and readily available on admission, which has excellent clinical promotion value.
Citation: Liang LX, Liang X, Zeng Y, Wang F, Yu XK. Establishment and validation of a nomogram for predicting esophagogastric variceal bleeding in patients with liver cirrhosis. World J Gastroenterol 2025; 31(9): 102714
Liver disease causes two million deaths worldwide each year, accounting for 4% of all deaths. Most of these fatalities are attributed to the complications of liver cirrhosis and hepatic cancer[1]. Liver cirrhosis is a chronic, progressive disease in which normal liver tissue is replaced by hepatic nodules and fibrous tissue due to various reasons, potentially leading to hepatocellular carcinoma. In the compensated stage, there are non-significant clinical symptoms. However, in decompensated liver cirrhosis, the characteristic complications include esophagogastric variceal bleeding (EGVB), ascites, and hepatic encephalopathy. The EGVB is one of the most dangerous complications of decompensated liver cirrhosis, with a mortality rate of 20%-25%[2]. Furthermore, the rebleeding rate of EGVB could be up to 60% within the first year[3]. Patients with cirrhosis who develop EGVB are at greater risk of severe complications such as hepatic encephalopathy, acute kidney injury, and infection. These complications lead to prolonged hospitalization and increased medical costs. Currently, endoscopic therapy is the primary treatment method recommended by guidelines for EGVB[4]. Endoscopic ligation therapy and endoscopic sclerotherapy are highly effective for acute variceal bleeding, achieving hemostasis rates of 90%-95% and significantly reducing the risk of rebleeding[5]. Therefore, accurately predicting EGVB in patients with liver cirrhosis is crucial for timely clinical intervention and follow-up, which can help reduce both mortality and economic burden.
Currently, methods for assessing the occurrence of EGVB are primarily divided into two categories: “Invasive” and “non-invasive” methods. The hepatic venous pressure gradient is an essential indicator for risk stratification in patients with cirrhosis and can predict EGVB related to portal hypertension[6]. Nonetheless, the invasive properties limit its use to a few hospitals and scientific research institutes. Besides, gastroscopy is also helpful for assessing bleeding risk, allowing for early warning and timely endoscopic intervention. Unfortunately, as an invasive examination, gastroscopy is challenging to use as a routine prediction and follow-up method due to concerns of patients and poor compliance. Moreover, gastroscopy has specific technical requirements for doctors, particularly in communities and remote areas. The model for end-stage liver disease, Child-Pugh score and other non-invasive methods are the classic non-invasive tools used to evaluate the prognosis of liver cirrhosis[7]. In addition, many researchers have focused on studying acute variceal bleeding models from different perspectives, such as imaging and serology[8-10]. For example, Luo et al[11] established a clinical-radiomics nomogram to predict the risk of EGVB based on radiomic features and clinical information, which performed well in training and validation cohorts. Liu et al[12] retrospectively collected the clinical manifestations, computed tomography and ultrasonic information of hepatitis B-related liver cirrhosis and conducted a nomogram for EGVB, which was well verified in the validation cohort. However, there remain some shortcomings, such as low accuracy, limited applicability, incomplete serological and clinical indicators, and high cost, which cannot meet the diverse needs of clinical practice. Accordingly, new, convenient, and effective models for predicting EGVB are urgently needed.
In this study, we investigated the risk factors for EGVB in patients with liver cirrhosis and established a simple and effective nomogram model that can predict the occurrence of EGVB, which is expected to help clinical staff assess the risk of EGVB and further provide individualized treatment.
MATERIALS AND METHODS
Patient selection
This was a single-center retrospective study. Patients with liver cirrhosis aged over 18 years old who visited Changsha Central Hospital from January 2015 to March 2024 were enrolled in this research. Liver cirrhosis was diagnosed according to clinical features, laboratory examinations, imaging evidence, or liver biopsy. Moreover, patients with cirrhosis complicated with EGVB were defined as having hematemesis and melena, which was confirmed by gastroscopy, without bleeding events at other sites. The exclusion criteria were as follows: (1) Liver and extrahepatic malignant tumors; (2) Patients undergoing liver transplantation or splenectomy; (3) Gastrointestinal bleeding at other sites; (4) Patients requiring anticoagulant or anti-platelet therapy; and (5) Patients with missing information. In all included populations, patients with cirrhosis complicated with EGVB were regarded as the experimental group, while non-EGVB patients were regarded as the control group.
Data collection
The clinical information of patients admitted within 24 hours was recorded using an electronic medical record system, including age, sex, diabetes mellitus, etiology (viral, alcoholic, and mixed), history of EGVB within six months, history of propranolol intake within six months, ascites, hepatic encephalopathy, portal vein thrombosis (PVT), white blood cell (WBC), hemoglobin (HGB), platelet, neutrophil (NEUT), neutrophil/lymphocyte ratio (NLR), activated partial thromboplastin time (APTT), prothrombin time, fibrinogen (Fbg), international normalized ratio (INR), prothrombin activity, albumin (ALB), total bilirubin (TBIL), aspartate aminotransferase (AST), alanine aminotransferase, total bile acid (TBA), blood urine nitrogen (BUN), serum creatinine, serum sodium, Child-Pugh score and so on. Other models such as the aspartate aminotransferase-to-platelet ratio index (APRI), fibrosis index based on 4 factors (FIB-4), and aspartate aminotransferase-to-alanine aminotransferase ratio (AAR) were calculated[13].
Establishment of the diagnostic model
Initially, patients were selected based on the exclusion and inclusion criteria. Subsequently, selected patients were separated into the EGVB and the control non-EGVB groups. Univariate analysis was performed between the two groups to identify the risk factors related to the occurrence of EGVB (P < 0.1). Second, these patients were randomly separated into the training and validation cohorts using a 6:4 ratio. In the training cohort, least absolute shrinkage and selection operator (LASSO) regression analysis was performed to screen the variables based on the results of univariate analysis using the R package “glmnet”. The LASSO regression avoids over-fitting by imposing a penalty on the size of the model coefficients, estimating the weaker variable towards zero as the penalty increases. Third, the R package “glm” was applied to conduct stepwise multivariate logistic regression analysis, and then the diagnostic model for patients with liver cirrhosis complicated with EGVB was established. Finally, to visualize the weight of each risk factor and make the model more convenient for clinical application, a nomogram was plotted by using the “rms” package of R software.
Evaluation and validation of the diagnostic model
A nomogram model for EGVB was established in the training cohort, and its discrimination power, consistency test, and clinical benefit were evaluated. This model was further verified in the validation cohort. Conversely, the receiver operator characteristics (ROC) curve and area under the curve (AUC) were used to evaluate the model. The AUC represents the discriminative power of this diagnostic model. If the AUC is greater than 0.8, this model is considered to have good discriminative power. In contrast, if the AUC was approximately 0.5, the diagnostic model’s diagnostic value was considered low. In addition, we used the Hosmer-Lemeshow (HL) test, calibration curve, ROC curve and AUC to measure and validate the model. Decision curve analysis (DCA) was conducted to evaluate the clinical usefulness of the diagnostic model using the R package “rmda”. Moreover, in the training and validation cohorts, this diagnostic model was compared with the current common liver cirrhosis diagnostic models, such as APRI, FIB-4, and AAR, to analyze the model’s advantages.
Ethics statement
The study protocol was approved by the Ethics Committee of Changsha Central Hospital (Changsha Central Hospital Affiliated with the University of South China; Approval No. 2022-S0019). The necessity for obtaining informed written consent was waived due to the retrospective nature of the study, which did not implicate the privacy or commercial interests of the patients. Additionally, measures were implemented to anonymize biological samples, and a stringent data security management and technical protection system was established for the storage, utilization, and dissemination of biological samples and data, thereby ensuring the security of data and personal information.
Statistical analysis
The R software (version 4.0.2) was employed for statistical analyses. Non-normal continuous variables are expressed using median and interquartile range (IQR). The differences between the two groups were analyzed using the t-test or Wilcoxon rank-sum test. Categorical variables are described with frequencies and percentages, and χ2 tests were used to identify group differences. The HL test (P > 0.05 reflects a non-significant difference between the expected and observed outcomes, and larger P values represent better agreement) was used to analyze the calibration curves, and dot plots were used to plot the calibration curves (points closer to the diagonal indicate better calibration)[14]. All P values are two-sided tests, and P < 0.05 was considered statistically significant.
RESULTS
Baseline characteristics between cirrhosis patients with and without EGVB
A total of 1115 patients diagnosed with liver cirrhosis were enrolled in this research (Figure 1). There were 251 patients (22.5%) in the EGVB group and 864 patients (76.5%) in the non-EGVB group. The baseline characteristics of the two groups are shown in Table 1. The median age of patients with liver cirrhosis in the EGVB group was 57 years (IQR: 50-67 years), compared with 62 years (IQR: 54-71 years) in the non-EGVB group. Most patients in the EGVB and non-EGVB groups were males (67% and 75%), and the difference in gender composition ratio was remarkably significant (P = 0.034). Regarding the etiology of liver cirrhosis, viral hepatitis was the leading cause of liver cirrhosis in both groups (43% and 43%). The incidence of EGVB in the past six months was higher in the EGVB group than in the non-EGVB group (36% vs 21%, P < 0.001), indicating the possibility of recurrence. Some of the patients were hospitalized for other decompensated events of cirrhosis such as ascites, infection, and hepatic encephalopathy. Compared with the non-EGVB group, individuals in the EGVB group exhibited a lower probability of ascites (P < 0.001) and more PVT (P = 0.014). Interestingly, most patients in the EGVB group were Child-Pugh grade A and B, while the non-EGVB group was Child-Pugh grade C (P = 0.018). Furthermore, differences in laboratory tests were observed between the two groups. Compared with the non-EGVB group, the EGVB group exhibited significantly higher INR (1.30 vs 1.28, P = 0.041), WBC (6.3 × 109/L vs 4.1 × 109/L, P < 0.001), platelets (91 × 109/L vs 80 × 109/L, P < 0.001), NEUT (4.41 × 109/L vs 2.63 × 109/L, P < 0.001), NLR (3.9 vs 3.3, P = 0.006), ALB (32.0 U/L vs 31.0 U/L, P = 0.046) and BUN (8.4 μmol/L vs 5.5 μmol/L, P < 0.001). In contrast, TBA (19 μmol/L vs 35 μmol/L, P < 0.001), HGB (80 × 109/L vs 102 × 109/L, P < 0.001), APTT (28 seconds vs 31 seconds, P < 0.001), Fbg (1.53 g/L vs 1.80 g/L, P < 0.001), TBIL (18 μmol/L vs 26 μmol/L, P < 0.001), and AST (35 μmol/L vs 42 μmol/L, P < 0.001) were lower. The differences were statistically significant (P < 0.05).
Figure 1 Flow diagram of the participants selection and grouping.
GIB: Gastrointestinal bleeding; EGVB: Esophagogastric variceal bleeding.
Table 1 Baseline characteristics of patients with cirrhosis in the control and esophagogastric variceal bleeding groups, n (%).
Characteristics
non-EGVB group (n = 864)
EGVB group (n = 251)
P value
Age (years), median (IQR)
62 (54, 71)
57 (50, 67)
< 0.001
Gender
0.034
Female
281 (33)
64 (25)
Male
583 (67)
187 (75)
WBC (× 109/L), median (IQR)
4.1 (2.9, 5.9)
6.3 (4.4, 9.2)
< 0.001
HGB (g/L), median (IQR)
102 (81, 122)
80 (61, 101)
< 0.001
PLT (× 109/L), median (IQR)
80 (53, 112)
91 (68, 115)
0.001
NEUT (× 109/L), median (IQR)
2.63 (1.83, 4.11)
4.41 (2.85, 6.70)
< 0.001
NLR, median (IQR)
3.3 (2.1, 5.4)
3.9 (2.5, 6.3)
0.006
APTT (second), median (IQR)
31 (27, 37)
28 (25, 33)
< 0.001
PTA (%), median (IQR)
68 (54, 82)
66 (55, 79)
0.162
INR, median (IQR)
1.28 (1.13, 1.40)
1.30 (1.20, 1.47)
0.041
TT, median (IQR)
19.00 (17.48, 21.00)
19.00 (18.00, 21.00)
0.984
PT, median (IQR)
13.90 (12.50, 15.70)
14.10 (12.85, 15.50)
0.194
Fbg (g/L), median (IQR)
1.80 (1.40, 2.30)
1.53 (1.20, 1.95)
< 0.001
ALB (g/L), median (IQR)
31.0 (27.0, 35.0)
32.0 (28.5, 36.0)
0.046
TBIL (μmol/L), median (IQR)
26 (16, 44)
18 (12, 28)
< 0.001
ALT (U/L), median (IQR)
33 (21, 55)
30 (21, 43)
0.08
AST (U/L), median (IQR)
42 (27, 72)
35 (24, 55)
< 0.001
TBA (μmol/L), median (IQR)
35 (17, 69)
19 (9, 37)
< 0.001
BUN (μmol/L), median (IQR)
5.5 (4.1, 7.7)
8.4 (6.1, 11.7)
< 0.001
CREA (μmol/L), median (IQR)
66 (54, 83)
68 (54, 84)
0.96
Na (mmol/L), median (IQR)
139.7 (136.9, 142.0)
139.4 (137.0, 142.0)
0.838
Ascites
< 0.001
No
348 (40)
145 (58)
Yes
516 (60)
106 (42)
Hepatic encephalopathy
0.378
No
783 (90.6)
232 (92.4)
Yes
81 (9.4)
19 (7.6)
Child-Pugh grade
0.018
A and B
647 (75)
206 (82)
C
217 (25)
45 (18)
Propranolol
0.212
No
802 (92.8)
227 (90.4)
Yes
62 (7.2)
24 (9.6)
Bleeding history
< 0.001
No
684 (79)
160 (64)
Yes
180 (21)
91 (36)
PVT
0.014
No
800 (92.6)
220 (88)
Yes
64 (7.4)
31 (12)
Diabetes mellitus
0.669
No
532 (73)
187 (75)
Yes
232 (27)
64 (25)
Etiology
0.729
Viral
371 (43)
109 (43)
Alcoholic
196 (23)
64 (25)
Mixed
73 (8.4)
19 (7.6)
Others
224 (26)
59 (24)
Comparison of baseline characteristics in the training cohort and validation cohort
All included patients (n = 1115) were randomly separated into the training and validation cohorts in a 6:4 ratio. Therefore, there were 669 patients in the training cohort and 446 in the validation cohort. In this study, the training cohort was used to create the diagnostic model, while the validation cohort was for verification. The proportions of patients with liver cirrhosis who suffered from EGVB in the training cohort was 23% and 22% in validation cohort. Based on the findings shown in Table 2, the characteristics of all included patients in the training and validation cohorts were further analyzed. The results demonstrated the non-significant differences between the two cohorts (P > 0.05) except for the etiology of liver cirrhosis (P = 0.035), implying the uniform distribution of the significant variables between the two cohorts, which could effectively avoid conclusion bias.
Table 2 Characteristics of patients with cirrhosis in the training and validation cohorts, n (%).
Characteristics
Training cohort (n = 669)
Validation cohort (n = 446)
P value
EGVB
0.953
No
518 (77)
346 (78)
Yes
151 (23)
100 (22)
Age (years), median (IQR)
60 (52, 70)
60 (52, 70)
0.814
Gender
0.508
Female
202 (30)
143 (32)
Male
467 (70)
303 (68)
WBC (× 109/L), median (IQR)
4.5 (3.2, 6.8)
4.4 (3.0, 6.4)
0.106
HGB (g/L), median (IQR)
97 (76, 119)
97 (75, 118)
0.799
PLT (× 109/L), median (IQR)
84 (58, 115)
78 (53, 112)
0.097
NEUT (× 109/L), median (IQR)
3.00 (2.00, 4.88)
2.90 (1.90, 4.49)
0.19
NLR, median (IQR)
3.4 (2.2, 5.5)
3.3 (2.1, 5.7)
0.761
APTT (second), median (IQR)
30 (27, 36)
30 (27, 37)
0.695
PTA (%), median (IQR)
68 (55, 82)
66 (52, 81)
0.197
INR, median (IQR)
1.28 (1.14, 1.40)
1.30 (1.17, 1.47)
0.156
TT, median (IQR)
19.00 (17.30, 20.90)
19.00 (18.00, 21.00)
0.112
PT, median (IQR)
13.80 (12.60, 15.50)
14.00 (12.50, 15.90)
0.244
Fbg (g/L), median (IQR)
1.76 (1.36, 2.24)
1.70 (1.35, 2.20)
0.589
ALB (g/L), median (IQR)
31.1 (27.0, 35.0)
31.0 (27.0, 35.0)
0.409
TBIL (μmol/L), median (IQR)
24 (14, 40)
23 (15, 43)
0.592
ALT (U/L), median (IQR)
31 (20, 52)
33 (22, 53)
0.346
AST (U/L), median (IQR)
40 (26, 69)
41 (28, 64)
0.289
TBA (μmol/L), median (IQR)
29 (14, 58)
31 (15, 70)
0.083
BUN (μmol/L), median (IQR)
6.0 (4.3, 9.0)
5.9 (4.5, 8.6)
0.898
CREA (μmol/L), median (IQR)
66 (54, 82)
67 (54, 84)
0.963
Na (mmol/L), median (IQR)
140.0 (137.0, 142.0)
139.0 (136.0, 141.7)
0.323
Ascites
0.445
No
302 (45)
191 (43)
Yes
367 (55)
255 (57)
Hepatic encephalopathy
0.392
No
605 (90.4)
410 (91.9)
Yes
64 (9.6)
36 (8.1)
Child-Pugh grade
0.751
A and B
514 (77)
339 (76)
C
155 (23)
107 (24)
Propranolol
0.714
No
619 (92.5)
410 (91.9)
Yes
50 (7.5)
36 (8.1)
Bleeding history
0.732
No
504 (75)
340 (76)
Yes
165 (25)
106 (24)
PVT
0.511
No
609 (91)
411 (92.2)
Yes
60 (9.0)
35 (7.8)
Diabetes mellitus
0.293
No
499 (75)
320 (72)
Yes
170 (25)
126 (28)
Etiology
0.035
Viral
290 (43)
190 (43)
Alcoholic
172 (26)
88 (20)
Mixed
47 (7.0)
45 (10)
Others
160 (24)
123 (28)
Identification of significant factors for EGVB in patients with liver cirrhosis
In univariate analyses, variables with P < 0.1 were included in subsequent analyses. We further performed LASSO regression on these variables using the R software “glmnet” package. The adjustment parameter λ was verified using the tenfold cross method. The two dashed lines indicated: Lambda minimum (this reflects the value of λ, which means the smallest error and the smallest number of predictor variables) and lambda.1se (it demonstrates the value of λ corresponding to the most economical model within one standard error of lambda minimum). To obtain the least number of variables for further research, we performed the optimal λ value of the model (0.02169) when the cross-validation error was the minimum λ value + 1 standard error, and finally screened out 10 variables, including age, WBC, HGB, APTT, Fbg, TBA, TBIL, AST, BUN, and ascites (Figure 2A and B). Subsequently, multivariate regression analysis was performed. The results revealed that seven variables screened were associated with EGVB (P < 0.05). The results indicated that WBC [odds ratio (OR) = 1.25, 95% confidence interval (CI): 1.16-1.34, P < 0.001], HGB (OR = 0.98, 95%CI: 0.98-0.99, P < 0.001), APTT (OR = 0.95, 95%CI: 0.91-0.98, P = 0.002), Fbg (OR = 0.33, 95%CI: 0.22-0.47, P < 0.001), TBIL (OR = 0.98, 95%CI: 0.97-0.99, P = 0.007), TBA (OR = 0.99, 95%CI: 0.98-1.00, P = 0.012), and ascites (OR = 0.52, 95%CI: 0.32-0.83, P = 0.006) were independent influencing factors for EGVB events (Table 3).
Figure 2 Risk factors screened by least absolute shrinkage and selection operator regression analysis.
A: The alteration trajectory between the coefficient of the independent variable and the log value of lambda (λ); B: The confidence interval under each lambda. The first dashed line corresponds to the minimum λ value and the second to minimum λ + 1 standard error.
According to the results of multivariate analysis, the seven independent factors with P < 0.05 were selected to construct the model (Figure 3A). The final model consists of TBA, WBC, HGB, APTT, Fbg, TBA, and ascites, and the model formula is as follows:
To visualize the weight of each predictor in this diagnostic model, we applied R software to plot a nomogram for clinical application. Each predictor variable was given a point and then summed to compute the total score. A higher total score for each individual corresponds to a higher likelihood of EGVB (Figure 3B).
Evaluation and validation of the nomogram model
The Bootstrap repeated sampling method was applied to generate the calibration curve. During the process, the training cohort (n = 669) and validation cohort (n = 446) were sampled 1000 times (Figure 4A and B). The abscissa represents the likelihood of EGVB in patients with cirrhosis, while the ordinate represents the actual event. Specifically, there is a direct relationship between the error rate and the deviation from the diagonal of the correction curve. The HL test was applied for correction analysis (P > 0.05 indicated a non-significant difference between the expected and the observed results). The two calibration curves suggested that the bar charts of the two groups were close to the diagonal line, indicating a good fit. The HL test demonstrated that the predicted value was consistent with the actual observed value (training cohort, χ2 = 9.5565, P = 0.2975; validation cohort, χ2 = 8.1433, P = 0.4196), suggesting the reliable fitting result and the robust model.
Figure 4 Evaluation and validation of the nomogram model.
A and B: Standard curve of nomogram for the training cohort and validation cohort. Red represents the original calibration curve; green represents the corrected calibration curve; C and D: Receiver operator characteristics curves were plotted in the training and validation cohorts to compare the nomogram with other models (aspartate aminotransferase-to-platelet ratio index, aspartate aminotransferase-to-alanine aminotransferase ratio, and fibrosis index based on 4 factors); E and F: Decision curve analysis curves were plotted in the training cohort and validation cohort to compare the nomogram with other models. The gray oblique line represents the benefit of clinical intervention for all patients with esophagogastric variceal bleeding. The horizontal black line represents the benefit of all esophagogastric variceal bleeding patients without clinical intervention. AUC: Area under the curve; APRI: Aspartate aminotransferase-to-platelet ratio index; FIB-4: Fibrosis index based on 4 factors; AAR: Aspartate aminotransferase-to-alanine aminotransferase ratio.
The ROC curve was plotted to compare the prediction performance of the nomogram model (Figure 4C and D). The results indicated that the AUC of this model in the training cohort was 0.861, which was better than the APRI score (AUC: 0.616), FIB-4 index (AUC: 0.626), and AAR score (AUC: 0.575). Moreover, the AUC of this diagnostic model in the validation cohort was 0.859, which was better than the APRI score (AUC: 0.616), FIB-4 index (AUC: 0.597), and AAR score (AUC: 0.542). The AUC of the two cohorts indicated that the diagnostic model exhibited a good prediction performance. According to the Jorden index, when the cut-off value of the training cohort was 0.185, the sensitivity and specificity of the model were 73.9% and 85.4%, respectively. When the cut-off value of the validation cohort was 0.245, the sensitivity and specificity of the model were 83.2% and 79.0%, respectively.
Finally, clinical DCA was applied to evaluate the net benefit of this nomogram model in clinical application. The horizontal axis represents the scenario where no patients receive the intervention (net benefit of zero), while the slash line indicates all patients receiving the intervention. As depicted in Figure 4E and F, the results demonstrated that under two extreme clinical scenarios (all patients treated or all patients untreated), this model was higher than the APRI score, FIB-4 index, and AAR score in both the training and validation cohorts, suggesting that this nomogram model has high clinical value.
DISCUSSION
Current studies indicate that nearly 50% of patients with liver cirrhosis have esophageal and gastric varices, with an annual incidence of variceal bleeding between 10% and 15%[15]. Once EGVB occurs, the risk of hepatic encephalopathy, infection, and acute liver failure significantly increases[16]. For patients with stable circulation, the European Society of Gastrointestinal Endoscopy guidelines strongly recommend that endoscopy should be performed within 12 hours for suspected EGVB cases[17]. However, due to the invasiveness and high cost of gastroscopy, its limited accessibility in remote areas, and the inability of individuals, timely diagnosis and follow-up of EGVB are often challenging. Consequently, it is essential to develop a simple and non-invasive model for diagnosing EGVB events. In the current research, patients with liver cirrhosis were randomly separated into training and validation cohorts. The training cohort was employed to construct the model and the validation cohort was used for verification. The LASSO regression and multivariate logistic regression analysis were employed to identify risk factors of EGVB and establish a nomogram model using common clinical indicators. Our model demonstrated excellent performance in both the training and validation cohorts.
We evaluated the model’s accuracy using a calibration curve. The HL test displays that the predicted value was consistent with the measured values, suggesting the model’s robustness. Additionally, the ROC curve was employed to assess the model’s predictive ability, and the AUC was calculated to assess its performance. The closer the ROC curve is to the upper left corner of the plot, the more accurate the test is. A diagnostic method with an AUC greater than 0.5 is meaningful, while an AUC over 0.8 is considered acceptable[18]. In our study, the AUC of the model was 0.861 in the training cohort and 0.859 in the validation cohort, suggesting strong prediction performance. In addition, DCA was employed to evaluate the model’s clinical utility. The DCA is a method for estimating predictive models by calculating net clinical benefit[19]. The DCA results indicated that the risk prediction model developed exhibited good clinical benefits compared to two extreme clinical scenarios (all patients treated or untreated) in both the training and validation cohorts. It further validates the model’s excellent performance and high value in practical clinical work. Some studies have found that non-invasive evaluation indicators, such as APRI and FIB-4, perform well in diagnosing liver fibrosis[20]. Liver fibrosis and cirrhosis gradually increase intrahepatic vascular resistance, sclerotic vasodilation, and increased portal venous blood flow, which contribute to elevated portal pressure and the development of collateral circulation and eventually increase the risk of EGVB[21]. It has been reported that APRI, FIB-4, and AAR are closely correlated with EGVB[13]. Interestingly, the results revealed that APRI, FIB-4, and AAR were inferior to the diagnostic nomogram model regarding discriminatory power, calibration ability and clinical usefulness.
In our study, univariate analysis demonstrated that the EGVB group, in comparison to the non-EGVB group, exhibited significantly elevated levels of WBC, NEUT, NLR, platelets, INR, ALB, and BUN. Conversely, levels of TBA, HGB, APTT, Fbg, TBIL, and AST were significantly lower. These findings illustrated the alterations in corresponding serological biomarkers associated with EGVB events in patients with cirrhosis. For instance, a stress response following bleeding leads to an increased WBC count. Furthermore, the compromised immune function in cirrhotic patients, particularly following EGVB events, exacerbates inflammation[22-24]. This might partially elucidate why inflammatory markers, such as WBC, NEUT, and NLR, were elevated in cirrhotic patients following EGVB events compared to the control group. Additionally, the occurrence of EGVB in cirrhotic patients could lead to the further breakdown and absorption of HGB in the intestine, which may result in enterogenous azotemia. Moreover, renal hypoperfusion could cause an elevation in the level of BUN due to blood loss and associated neurohumoral regulatory mechanisms[25,26]. Consequently, BUN had been considered a reliable indicator for the clinical diagnosis of gastrointestinal bleeding. Notably, individuals in the EGVB group demonstrated a higher likelihood of PVT (P = 0.014) compared to the non-EGVB group. PVT is a significant complication in cirrhosis, with an incidence ranging from 2% to 40%, increasing with the severity of liver disease[27,28]. In the context of cirrhosis, elevated portal vein pressure, reduced portal vein flow velocity, and the disruption of coagulation function balance substantially heighten the risk of PVT formation[29,30]. Furthermore, the presence of PVT can exacerbate portal hypertension and elevate the risk of EGVB or rebleeding, thereby decreasing the survival rates of patients[31].
In this research, patients with liver cirrhosis combined with EGVB accounted for 22.5% of the total enrolled patients. Among the seven significant variables finally identified in the final model, Fbg (OR = 0.33, 95%CI: 0.22-0.47), ascites (OR = 0.52, 95%CI: 0.32-0.83), and WBC (OR = 1.25, 95%CI: 1.16-1.34) had the highest weight in the simplified model. Impaired liver synthesis in patients with cirrhosis disrupts the balance between procoagulant (Fbg) and anticoagulant blood, leading to coagulopathy. The Fbg promotes platelet aggregation by binding to glycoprotein IIb/IIIa, which initiates a cascade of amplified activation of coagulation factors to exert coagulation function. A decrease in Fbg might lead to inefficient removal of activated coagulation factors, consequently increasing the bleeding risk[32]. Liu et al[33] revealed that Fbg activity and antigen levels decrease with the severity of cirrhosis. Hyperfibrinolysis has also suggested a mechanism for increased bleeding risk in patients with cirrhosis, which has also established a correlation between reduced Fbg levels and prolonged bleeding time[34]. Luo et al[11] analyzed the risk factors for EGVB events by collecting clinical information from 211 hospitalized patients with cirrhosis. The ALB, Fbg, PVT, AST, and spleen thickness were identified as independent influencing factors of EGVB. The results implied that Fbg was a protective factor (OR = 0.385, 95%CI: -0.17 to 0.939, P = 0.001), consistent with our research conclusions. Giannini et al[35] investigated rebleeding in 109 patients with cirrhosis who preventatively underwent endoscopic variceal band ligation (EVBL) and found that low Fbg levels were related to an increased risk of bleeding after EVBL prophylaxis in patients with liver cirrhosis. Moreover, the negative predictive value of an Fbg level of ≤ 179 mg/dL for variceal bleeding was 98.6%[35].
Furthermore, ascites, one of the common complications of liver cirrhosis and portal hypertension, was found to be a protective factor against EGVB events in our research, consistent with previous reports. El Sheref et al[36] investigated the risk factors for EGVB and found that ascites, Helicobacter pylori infection, and Child-Pugh grades B and C were independent risk factors for EGVB. Among them, ascites were considered a protective factor (OR = 0.036, 95%CI: 0.0004-0.36). This may be due to the reduction of adequate circulating blood volume, which decreases blood pressure, thus reducing the possibility of EGVB in patients with liver cirrhosis. Consequently, it suggested that careful consideration should be given to the extent of ALB supplementation in patients with cirrhosis with ascites and EGVB. The WBC count is often a marker of bacterial infection. The intestines of patients with cirrhosis are prone to bacterial overgrowth, intestinal motility disorders, and increased intestinal permeability, which increase bacterial translocation. Several common infections in cirrhosis can be caused by bacterial translocation, including spontaneous bacterial peritonitis. Moreover, it is a source of bacterial by-products such as endotoxin, which can elevate portal pressure, impair liver function, cause abnormal coagulation function, and eventually induce variceal bleeding[37]. Patients experiencing variceal bleeding, when complicated by concurrent bacterial infections, exhibit an increased susceptibility to inadequate bleeding control, early rebleeding, and mortality. Accordingly, the prophylactic use of antibiotics reduces the incidence of bacterial infection and early rebleeding and significantly reduces mortality[38].
Our study has several strengths. First, we developed the diagnostic model in a large cohort of 1115 patients with univariate analysis, LASSO, and multivariate regression analysis. Second, this nomogram model was further verified in the validation cohort and proved to perform well. Moreover, our model is more robust and accurate than previous routine prediction models. Finally, the factors included in this model are objective, non-invasive, inexpensive, and readily available on admission, which is suitable for different levels of hospitals and has excellent clinical promotion value.
However, our research has several limitations. First, as this research was conducted at a single clinical center, specific sensitive serological markers such as soluble cluster of differentiation 163[39,40] and von Willebrand factor-antigen[41] were excluded, which may have introduced a certain degree of bias. Considering the retrospective design of this study, we plan to investigate the role of specific sensitive biomarkers in diagnosing EGVB through prospective multicenter analyses in future research. Second, our study mainly focused on patients with cirrhosis with or without EGVB. Furthermore, we intend to explore other bleeding types in patients with liver cirrhosis in future investigations. Third, this study is based on a retrospective cohort. Although we applied a validation cohort for internal validation of the prediction model, further prospective validation is necessary before this nomogram can be considered for clinical use.
CONCLUSION
This research established a non-invasive nomogram model using LASSO regression and multivariate regression analysis. This model can accurately identify EGVB events in patients with liver cirrhosis based on simple laboratory indicators. Compared to invasive gastroscopy, this model is non-invasive, simple, and safe, making it suitable for patients who cannot undergo gastroscopy and for routine follow-up after an initial EGVB event.
Footnotes
Provenance and peer review: Unsolicited article; Externally peer reviewed.
Peer-review model: Single blind
Specialty type: Gastroenterology and hepatology
Country of origin: China
Peer-review report’s classification
Scientific Quality: Grade B, Grade B, Grade B, Grade C
Novelty: Grade B, Grade B, Grade B, Grade C
Creativity or Innovation: Grade A, Grade B, Grade B, Grade C
Scientific Significance: Grade B, Grade B, Grade B, Grade B
P-Reviewer: Cao QG; Oyesanmi O; Tang J S-Editor: Fan M L-Editor: Webster JR P-Editor: Yu HG
Liu R, Sun Y, Xu K, Shi H, Sheng S, Kong D. A Histogram Model to Predict the Risk of Bleeding from Oesophageal and Gastric Variceal Rupture in Cirrhosis.J Coll Physicians Surg Pak. 2022;32:586-590.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Tan BG, Tang Z, Ou J, Zhou HY, Li R, Chen TW, Zhang XM, Li HJ, Hu J. A novel model based on liver/spleen volumes and portal vein diameter on MRI to predict variceal bleeding in HBV cirrhosis.Eur Radiol. 2023;33:1378-1387.
[PubMed] [DOI][Cited in This Article: ][Cited by in RCA: 1][Reference Citation Analysis (0)]
Hou Y, Yu H, Zhang Q, Yang Y, Liu X, Wang X, Jiang Y. Machine learning-based model for predicting the esophagogastric variceal bleeding risk in liver cirrhosis patients.Diagn Pathol. 2023;18:29.
[PubMed] [DOI][Cited in This Article: ][Cited by in RCA: 6][Reference Citation Analysis (0)]
Liu CH, Liu S, Zhao YB, Liao Y, Zhao GC, Lin H, Yang SM, Xu ZG, Wu H, Liu E. Development and validation of a nomogram for esophagogastric variceal bleeding in liver cirrhosis: A cohort study in 1099 cases.J Dig Dis. 2022;23:597-609.
[PubMed] [DOI][Cited in This Article: ][Cited by in RCA: 5][Reference Citation Analysis (0)]
Surjanovic N, Lockhart RA, Loughin TM. A generalized Hosmer-Lemeshow goodness-of-fit test for a family of generalized linear models.Test (Madr). 2024;33:589-608.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Garcia-Tsao G, Abraldes JG, Berzigotti A, Bosch J. Portal hypertensive bleeding in cirrhosis: Risk stratification, diagnosis, and management: 2016 practice guidance by the American Association for the study of liver diseases.Hepatology. 2017;65:310-335.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 1108][Cited by in RCA: 1367][Article Influence: 170.9][Reference Citation Analysis (3)]
Shiha G, Ibrahim A, Helmy A, Sarin SK, Omata M, Kumar A, Bernstien D, Maruyama H, Saraswat V, Chawla Y, Hamid S, Abbas Z, Bedossa P, Sakhuja P, Elmahatab M, Lim SG, Lesmana L, Sollano J, Jia JD, Abbas B, Omar A, Sharma B, Payawal D, Abdallah A, Serwah A, Hamed A, Elsayed A, AbdelMaqsod A, Hassanein T, Ihab A, GHaziuan H, Zein N, Kumar M. Asian-Pacific Association for the Study of the Liver (APASL) consensus guidelines on invasive and non-invasive assessment of hepatic fibrosis: a 2016 update.Hepatol Int. 2017;11:1-30.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 120][Cited by in RCA: 150][Article Influence: 16.7][Reference Citation Analysis (0)]
Zhao L, Leng Y, Hu Y, Xiao J, Li Q, Liu C, Mao Y. Understanding decision curve analysis in clinical prediction model research.Postgrad Med J. 2024;100:512-515.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 1][Reference Citation Analysis (0)]
Ballestri S, Mantovani A, Baldelli E, Lugari S, Maurantonio M, Nascimbeni F, Marrazzo A, Romagnoli D, Targher G, Lonardo A. Liver Fibrosis Biomarkers Accurately Exclude Advanced Fibrosis and Are Associated with Higher Cardiovascular Risk Scores in Patients with NAFLD or Viral Chronic Liver Disease.Diagnostics (Basel). 2021;11.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 31][Cited by in RCA: 49][Article Influence: 12.3][Reference Citation Analysis (0)]
Dos Santos Pinheiro C, de Oliveira Gomes CG, Ribeiro Lima Machado C, Guedes LR, Rocha HC, Guimarães RG, Carvalho FAC, Saturnino SF, do Nascimento VC, de Andrade MVM, Vilela EG. Performance of High Mobility Protein Group 1 and Interleukin-6 as Predictors of Outcomes Resulting from Variceal Bleeding in Patients with Advanced Chronic Liver Disease.Inflammation. 2022;45:544-553.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Matsue Y, van der Meer P, Damman K, Metra M, O'Connor CM, Ponikowski P, Teerlink JR, Cotter G, Davison B, Cleland JG, Givertz MM, Bloomfield DM, Dittrich HC, Gansevoort RT, Bakker SJ, van der Harst P, Hillege HL, van Veldhuisen DJ, Voors AA. Blood urea nitrogen-to-creatinine ratio in the general population and in patients with acute heart failure.Heart. 2017;103:407-413.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 41][Cited by in RCA: 45][Article Influence: 5.0][Reference Citation Analysis (0)]
Zhang Y, Xu BY, Wang XB, Zheng X, Huang Y, Chen J, Meng ZJ, Gao YH, Qian ZP, Liu F, Lu XB, Shi Y, Shang J, Li H, Wang SY, Yin S, Sun SN, Hou YX, Xiong Y, Chen J, Li BL, Lei Q, Gao N, Ji LJ, Li J, Jie FR, Zhao RH, Liu JP, Lin TF, Chen LY, Tan WT, Zhang Q, Zou CC, Huang ZB, Jiang XH, Luo S, Liu CY, Zhang YY, Li T, Ren HT, Wang SJ, Deng GH, Xiong SE, Liu XX, Wang C, Yuan W, Gu WY, Qiao L, Wang TY, Wu DD, Dong FC, Li H, Hua J. Prevalence and Clinical Significance of Portal Vein Thrombosis in Patients With Cirrhosis and Acute Decompensation.Clin Gastroenterol Hepatol. 2020;18:2564-2572.e1.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 38][Cited by in RCA: 47][Article Influence: 9.4][Reference Citation Analysis (0)]
Stine JG, Wang J, Shah PM, Argo CK, Intagliata N, Uflacker A, Caldwell SH, Northup PG. Decreased portal vein velocity is predictive of the development of portal vein thrombosis: A matched case-control study.Liver Int. 2018;38:94-101.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 85][Cited by in RCA: 93][Article Influence: 13.3][Reference Citation Analysis (0)]
Northup PG, Garcia-Pagan JC, Garcia-Tsao G, Intagliata NM, Superina RA, Roberts LN, Lisman T, Valla DC. Vascular Liver Disorders, Portal Vein Thrombosis, and Procedural Bleeding in Patients With Liver Disease: 2020 Practice Guidance by the American Association for the Study of Liver Diseases.Hepatology. 2021;73:366-413.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 171][Cited by in RCA: 330][Article Influence: 82.5][Reference Citation Analysis (1)]
Liu Y, Zhuang Y, Xu G, Wang X, Lin L, Ding Q. Fibrinogen dysfunction and fibrinolysis state in patients with hepatitis B-related cirrhosis.Hematology. 2024;29:2392028.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Drolz A, Horvatits T, Roedl K, Rutter K, Staufer K, Kneidinger N, Holzinger U, Zauner C, Schellongowski P, Heinz G, Perkmann T, Kluge S, Trauner M, Fuhrmann V. Coagulation parameters and major bleeding in critically ill patients with cirrhosis.Hepatology. 2016;64:556-568.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 122][Cited by in RCA: 139][Article Influence: 15.4][Reference Citation Analysis (0)]
Giannini EG, Giambruno E, Brunacci M, Plaz Torres MC, Furnari M, Bodini G, Zentilin P, Savarino V. Low Fibrinogen Levels Are Associated with Bleeding After Varices Ligation in Thrombocytopenic Cirrhotic Patients.Ann Hepatol. 2018;17:830-835.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 17][Cited by in RCA: 20][Article Influence: 2.9][Reference Citation Analysis (0)]
El Sheref SEDM, Afify S, Berengy MS. Clinical characteristics and predictors of esophagogastric variceal bleeding among patients with HCV-induced liver cirrhosis: An observational comparative study.PLoS One. 2022;17:e0275373.
[PubMed] [DOI][Cited in This Article: ][Reference Citation Analysis (0)]
Kalambokis GN, Oikonomou A, Christou L, Kolaitis NI, Tsianos EV, Christodoulou D, Baltayiannis G. von Willebrand factor and procoagulant imbalance predict outcome in patients with cirrhosis and thrombocytopenia.J Hepatol. 2016;65:921-928.
[PubMed] [DOI][Cited in This Article: ][Cited by in Crossref: 57][Cited by in RCA: 61][Article Influence: 6.8][Reference Citation Analysis (0)]
