Abstract: Objective To investigate the indexes selection in non-invasive diagnostic model for liver fibrosis
in chronic hepatitis B patients with ALT below 2 times of the upper limit of normal (ULN). Methods Total
of 249 chronic hepatitis B patients with ALT ＜ 2 × ULN in the Third People’s Hospital of Shenzhen from
January 2007 to April 2014 were selected. Commonly clinical indexes were taken into consideration according
to the clinical experiences and literatures reviews. Spearman rank correlation statistics was used to analyze the
association between liver fibrosis and each index and screen the eligibility indexes (P ＜ 0.001, r ＞ 0.200).
Different models were constructed according to the size and numbers of relations. The areas under curve of
AUC in each model were compared, and the model which owns least indexes and best diagnosis abilities
was the one we need. Results Total of 249 cases were selected, including 204 cases in modeling group and
45 cases in validation group. Nine of the nineteen common indexes were correlated with liver fibrosis, which
included γ-glutamyl endopeptidase (GGT), platelet count (PLT), hyaluronic acid (HA), age, HBV DNA,
spleen thick, cholinesterase (CHE), the specific value of albumin and globulin (A/G), alkaline phosphatase

(ALP) (P ＜ 0.01, r ＞ 0.200). Comprehensive consideration of model with AUC, sensitivity, specificity,
positive and negative predicted values, DMFibroS was selected as the best non-invasive diagnostic model
for liver fibrosis in chronic hepatitis B patients with ALT ＜ 2 × ULN. The accuracy was 77.8% by other 45
patients for validation. Conclusion The construction of a model is a complex progress. The best non-invasive
diagnosis model not only own least indexes but also has best diagnosis abilities, which also needs example
verification.