[Noninvasive assessment of liver fibrosis in chronic hepatitis B using a predictive model]

To develop a diagnostic model comprising clinical and serum markers for assessing HBV-related liver fibrosis.

270 chronic hepatitis B patients were randomly allocated to either an estimation group (195 cases) or a validation group (75 cases). Liver biopsies were done and staging of fibrosis was assessed. Twenty-six common clinical and serum markers were analyzed initially in the estimation group to derive a predictive model to discriminate the stages of fibrosis. The model created was then assessed with ROC analysis. It was also applied to the validation group to test its accuracy.

Among 13 variables associated with liver fibrosis selected by univariate analysis, age, gamma glutamyltranspeptidase (GGT), hyaluronic acid (HA), and platelet count (PLT) were identified by multivariate logistic regression analysis as independent factors of fibrosis. A fibrosis index constructed from the above four markers was established. In ROC analysis, the AUC was 0.889 for the estimation group and 0.850 for the validation group for discriminating > or =S3 from < or=S2. Using the optimal cutoff score 3.0, the sensitivity of the index was 90.2%, the specificity 76.1%, and the accuracy was 82%. There was a positive linear relationship between the index scores and the fibrosis stages (r = 0.731, P<0.001). The AUC for identifying > or=S2 was 0.873 with sensitivity/specificity of 79%/82%, cutoff score 2.2; The AUC for identifying S4 was 0.872 with sensitivity/specificity of 83%/75%, cutoff score 5.4. There were no significant differences in diagnostic efficacy in the model between the estimation and the validation group (P>0.05).

A model for assessment of liver fibrosis was established with easily accessible markers. It appears to be sensitive, accurate and reproducible, suggesting it could be used to assist or replace liver biopsy to detect dynamic changes of HBV-related liver fibrosis.