Second, most studies validated selleck their results in an internal validation cohort from the same population with the training cohort. We validated our results in an external validation cohort that included chronic HBV carriers from Shanghai, Fujian Province, and Jiangsu Province, China. The geographic diversity of the training and validation cohort helped us to

find out models of stable accuracy irrespective of where the patient comes from. We notice that the diagnostic indices of the S index in the training cohort (Table 4) are slightly lower than those in the validation cohort (Table 5). These might be due to higher S1-2 prevalence in the training cohort, which is more difficult for a noninvasive predictive model to give a correct classification. Third,

our predictive model was based only on routine laboratory markers. GGT, PLT and ALB are all routine tests readily available to most clinicians managing patients with chronic buy Vemurafenib HBV infection, so no additional tests are needed. The diagnostic accuracy of models consisting of simple routine tests was compared with models introducing special tests such as HA and A2M. To our knowledge, such validation and comparison were not carried out in chronic HBV carriers before. We noticed that the SLFG model and Hepascore performed better in identifying significant fibrosis than the Forns score and APRI, but the superiority was not significant in identifying selleckchem advanced fibrosis or cirrhosis. The result was similar to a validation study in CHC patients,10 indicating that such special tests might improve the sensitivity of a diagnostic model in predicting early fibrosis. But including tests unavailable in daily practice makes standardization, validation and routine bedside use difficult. Fourth, the S index is easily calculated. Most of the previous models, except the APRI, involved complex formulas, which require a logarithmic calculator for calculations. The simplicity of the S index and APRI allows them to be determined in the clinic or bedside easily. But the APRI

was conducted in CHC patients and one of its two parameters is AST, which did not show a significant correlation with liver fibrosis staging of CHB patients in our study. This may explain the low AUROC of APRI compared with other models. The S index consisted of the most significant predictors of fibrosis among routine markers (GGT, PLT and ALB) and was simplified from three complicated regression functions. Despite a slightly lower AUROC than the respective function in each histological endpoint, the S index allows both significant fibrosis and cirrhosis to be identified using one simple formula. There are some limitations in our study. An incorrigible defect in studies of diagnostic models is the questionable gold standard we have to use. Liver biopsy is not a perfect gold standard for fibrosis evaluation due to sampling error and observer variability.