This study included 2491 untreated participants with genotype B or C HBV infection, who were HBsAg-seropositive, HBeAg-seronegative, anti-HCV-seronegative, and cirrhosis free at study entry. Regression coefficients of predictors in Cox Regression models were converted into integer scores for predicting HBsAg seroclearance. Predictive accuracy was assessed with area under the receiver operating characteristic curves (AUROC), and predictive accuracies of models with and without serum HBsAg levels were compared.
Low serum levels of both HBsAg and HBV DNA were the strongest predictors of spontaneous HBsAg seroclearance. Compared to baseline serum HBsAg levels ?1000 IU/ml, the multivariate adjusted rate ratio of spontaneous HBsAg seroclearance was 10.96 (7.92-15.16) for those with baseline serum HBsAg levels <100 IU/ml. The predictive ability of HBsAg levels was modified by HBV viral load, showing a weaker effect in those with higher viral loads, and the strongest effect among those with undetectable viral loads. The inclusion of serum HBsAg levels greatly improved the AUROC for predicting HBsAg seroclearance at the fifth (from 0.79 [0.787-0.792] to 0.89 [0.889-0.891]) and tenth year (from 0.73 [0.728-0.732] to 0.84 [0.839-0.841]) after study entry.
Incorporated into an easy-to-use scoring system, HBV viral load and quantitative serum HBsAg levels can accurately predict HBsAg seroclearance.