A risk assessment model for esophageal varices occurrence based on endoscopic ultrasonography

Abstract

Objective To identify the independent risk factors of esophageal varices (EV) in cirrhosis by endoscopic ultrasonography (EUS), and further to establish a risk assessment model for predicting EV occurrence and evaluate the clinical predictive value of the model. Methods A retrospective cohort study was used in this study. Data of patients with cirrhosis without varicosity, who were hospitalized in Tianjin Second People's Hospital from September 2014 to March 2017 were collected. The location, diameter, and number of esophageal collateral circulation were measured by EUS. The non-selective beta blocker (NSBB) medication history and antiviral therapy were recorded. The time of the first EUS examination was taken as the starting point and the follow-up period was set up as 18 months. The end point was the occurrence of EV or the end of follow-up. The independent risk factors of EV occurrence were determined by univariate and multivariate logistic regression analysis, and the risk assessment model of EV occurrence was constructed. The predictive value of evaluation model for disease was studied by ROC analysis. Hosmer-Lemeshow goodness of fit was used to test the fitting efficiency of the evaluation model. Results A total of 638 subjects were recruited initially, 13 of them were lost in the course of the study. Finally, 625 cases were included in the study. Among them, 369 cases did not develop EV (the non-progress group) and 256 cases developed EV (the progress group). (1) Multivariate logistic regression analysis showed that 7 independent risk factors were selected into the risk assessment model of EV occurrence, and were assigned corresponding scores: no NSBB (3 points), no antiviral treatment (2 points), Child-Pugh stage B (1 point), the diameter of peri-ECV>2 mm (1 point), the number of peri-ECV≥5 (3 points), the diameter of para-ECV≥5 mm (4 points), and the number of para-ECV≥5 (4 points). (2) In the risk assessment model, the risk factor scores ranged from 1 to 4 with a total score of 0-18. The predicted incidence of EV increased from 0.003 to 1.000 with the increase of the score. (3) In the risk assessment model, the total risk score ≤2 was assigned into low-risk group, 3-5 into medium-risk group, and ≥6 into high-risk group. The actual EV incidence of each risk stratification was 2.78% in the low-risk group, 36.36% in the medium-risk group and 93.91% in the high-risk group, respectively. (4) The ROC analysis showed that area under curve (AUC) was 0.947 (P<0.05), suggesting that the risk assessment model had a good effect on predicting disease progression. Hosmer-Lemeshow test showed that P was 0.450, suggesting that the model fitted well. Conclusion The risk assessment model based on EUS can accurately predict the occurrence of EV, and is simple and easy to use. The model can provide scientific basis for the prevention and rational treatment of EV in liver cirrhosis. Key words: Liver cirrhosis; Varicose veins; Endosonography; Logistic models