Statistical analysis

All statistical analyses were performed using R v3.2.2 (https://www.r-project.org/). Values were presented as either mean with standard deviation (SD), median with 25–75% percentiles, or numbers with percentages where appropriate. Chi-squared tests were used to test the significant differences of categorical variables between groups. Kruskal-Wallis or Mann Whitney Wilcoxon tests were used to compare non-parametric data of quantitative variables between different groups where appropriate. Logistic regression models were built to establish the potent biomarkers-based panels (AFP, miR-122 expression and circulating TERT promoter mutations) to predict HCC from individuals with chronic hepatitis B who did not have cancer at the time of the study. Receiver operating characteristic (ROC) curves were generated and the diagnostic value of different panels was assessed by computation of the area under the ROC curve (AUC). The level of significance was set at a two-sided P value of <0.05.