Development and validation of a nomogram based on clinical factors and standard laboratory tests for prediction of clinically significant liver fibrosis in chronic hepatitis C virus infection

Abstract

Staging liver fibrosis in chronic viral hepatitis C (HCV) patients is essential for prompting surveillance and treatment. The aim of this study was to develop a nomogram, on the basis of simple clinical and laboratory variables, to predict three clinically significant stages of fibrosis (nil-mild, moderate, advanced/cirrhosis), using histology as reference, and to compare its performance with that of FibroTest, a widely used noninvasive fibrosis score. Nomograms are graphical representations of a mathematical formula, used as predictive tools. The study retrospectively recruited 406 HCV patients undergoing liver biopsy. Nomogram was developed in a training set of 252 patients and tested in a validation set of 154 patients. Histology was staged according to the Metavir system. Fibrosis stages were subgrouped as follows: advanced fibrosis/cirrhosis (F3/F4, 24%), nil-mild (F0/F1, 36%), and moderate (F2, 40%). Age at biopsy, aspartate aminotransferase, γ-glutamyl transpeptidase, albumin, platelet count, and prothrombin activity formed the basis for the so-called Fibro-Nomogram, which, in one graphical representation, estimates probability for different stages of fibrosis. Areas under the receiver-operating characteristic curves for advanced fibrosis/cirrhosis were similar for training (0.86) and validation sets (0.87). For nil-mild fibrosis, area under the receiver-operating characteristics were 0.81 and 0.79. Compared with FibroTest, Fibro-Nomogram performed slightly better at predicting severe fibrosis (F3/F4) with positive likelihood ratio (LR+) 5.07 (95% confidence interval 3.08-8.37) versus LR+ 3.82 (95% confidence interval 2.56-5.71) for FibroTest. For nil-mild fibrosis, the two tests showed limited but comparable performances. In HCV patients, Fibro-Nomogram, an inexpensive and readily available predictive tool, could enable clinicians to interpret patients' profile, concurrently stratifying patients into three clinically relevant probability categories with good overall performance.