Developing the prediction model of esophagogastric variceal rebleeding in patients with liver cirrhosis based on artificial neural network
10.3969/j.issn.1001-5256.2022.11.011
- VernacularTitle:肝硬化食管胃底静脉曲张破裂出血患者再出血预测模型的建立
- Author:
Qun ZHANG
1
;
Ke SHI
1
;
Xianbo WANG
1
Author Information
1. Center of Integrative Medicine, Beijing Ditan Hospital, Capital Medical University, Beijing 100015, China
- Publication Type:Original Articles_Liver Fibrosis and Liver Cirrhosis
- Keywords:
Esophageal and Gastric Varices;
Hemorrhage;
Neural Networks(Computer)
- From:
Journal of Clinical Hepatology
2022;38(11):2493-2498
- CountryChina
- Language:Chinese
-
Abstract:
Objective To develop an artificial neural network model to predict the risk of rebleeding within one year in cirrhotic patients with esophagogastric variceal bleeding. Methods We retrospectively collected 441 cirrhotic patients with esophagogastric variceal bleeding hospitalized at Beijing Ditan Hospital, Capital Medical University, from August 2008 to October 2017. The enrolled patients were followed up for one year. According to the primary endpoint which was rebleeding within one year, patients were divided into rebleeding (249 cases) and non-rebleeding (192 cases) groups. Fisher exact test or chi-square test were used for comparison of categorical data. Comparison of continuous data with normal distribution between groups was performed using t test, while comparison of non-normally distributed data was performed by Mann-Whitney U test. Cox univariate and multivariate regression were used to identify independent factors affecting rebleeding within one year in cirrhotic patients with esophagogastric variceal bleeding, and then an artificial neural network prediction model was constructed using identified factors. The predictive performance of the model was evaluated using the area under the receiver operating characteristic curve (AUC). Results In total, 249 (56.5%) patients developed esophagogastric variceal rebleeding within one year. Cox multivariate regression showed INR (AHR=1.566, 95%CI: 1.023~2.398, P=0.039) and NLR (AHR=1.033, 95%CI: 1.009~1.058, P=0.006) were risk factors for 1-year rebleeding, while CHB (AHR=0.769, 95%CI: 0.597~0.991, P=0.042), Na (AHR=0.967, 95%CI: 0.936~0.999, P=0.044), endoscopic (AHR=0.829, 95%CI: 0.743~0.926, P=0.001) and surgical treatment (AHR=0.246, 95%CI: 0.120~0.504, P < 0.001) were protective factors. Using the above six independent influence factors, we successfully constructed an artificial neural network model (https://lixuan.me/annmodel/myg-v3/). The model's ability to predict 1-year rebleeding had an AUC of 0.782 (95%CI: 0.740-0.825), which was higher than 0.672 (95%CI: 0.622-0.722, P < 0.001) of Cox regression model, 0.557 (95%CI: 0.504-0.610, P < 0.001) of Child-Pugh and 0.562 (95%CI: 0.509-0.616, P < 0.001) of MELD scores. Conclusion The artificial neural network model has good individualized prediction performance and can be used as a risk assessment tool for esophagogastric variceal rebleeding.
