Abstract
AIM
To build a diagnostic non-invasive model for screening of large varices in cirrhotic hepatitis C virus (HCV) patients.
METHODS
This study was conducted on 124 post-HCV cirrhotic patients presenting to the clinics of the Endemic Medicine Department at Mansoura University Hospital for evaluation before HCV antiviral therapy: 78 were Child A and 46 were Child B (score <= 8). Inclusion criteria for patients enrolled in this study was presence of cirrhotic HCV (diagnosed by either biopsy or fulfillment of clinical basis). Exclusion criteria consisted of patients with other etiologies of liver cirrhosis, e.g., hepatitis B virus and patients with high MELD score on transplant list. All patients were subjected to full medical record, full basic investigations, endoscopy, and computed tomography CT), and then divided into groups with no varices, small varices, or large risky varices. In addition, values of Fibrosis-4 score (FIB-4), aminotransferase-to-platelet ratio index (APRI), and platelet count/splenic diameter ratio (PC/SD) were also calculated.
RESULTS
Detection of large varies is a multi-factorial process, affected by many variables. Choosing binary logistic regression, dependent factors were either large or small varices while independent factors included CT variables such coronary vein diameter, portal vein (PV) diameter, lieno-renal shunt and other laboratory noninvasive variables namely FIB-4, APRI, and platelet count/splenic diameter. Receiver operating characteristic (ROC) curve was plotted to determine the accuracy of non-invasive parameters for predicting the presence of large esophageal varices and the area under the ROC curve for each one of these parameters was obtained. A model was established and the best model for prediction of large risky esophageal varices used both PC/SD and PV diameter (75% accuracy), while the logistic model equation was shown to be (PV diameter x -0.256) plus (PC/SD x -0.006) plus (8.155). Values nearing 2 or more denote large varices.
CONCLUSION
This model equation has 86.9% sensitivity and 57.1% specificity, and would be of clinical applicability with 75% accuracy.