Early on-treatment dynamics predicting hepatitis B e antigen seroconversion in chronic hepatitis B.

BACKGROUND

Hepatitis B e antigen (HBeAg) loss with anti-HBe appearance marks the first durable immunological milestone during nucleos(t)ide analogues (NAs) therapy, yet static baseline or single time-point biomarkers anticipate this event poorly. We hypothesized that early dynamic parameters reflecting viral, antigenic and hepatic function during the first 12 weeks of therapy contain richer prognostic information than baseline values alone.

AIMS: This study aimed to construct a model that fuses week-12 changes in serum biochemical and virological parameters and tested its performance in predicting week-60 HBeAg seroconversion (SR).

METHODS

Serum biochemical and virological parameters of HBeAg-positive patients receiving NAs treatment between January and December 2023 were collected. Difference, ratio and absolute week-12 values of variables were derived. After ten-fold cross-validation, a multi-variable signature was selected to build a least absolute shrinkage and selection operator (LASSO) logistic regression model.

RESULTS

Overall, 38.24% (26/68) of patients achieved HBeAg SR. Using LASSO regression with ten-fold cross-validation, an eight-variable signature related to the dynamic changes of HBeAg, hepatitis B surface antigen (HBsAg), gamma-glutamyl transferase (GGT), and albumin (ALB) were selected. In the training cohort (n=68), the multi- variable model demonstrated an area under the curve (AUC) of 0.975 (sensitivity, 96.2%; specificity, 95.2%), significantly outperforming single-variable models. In the test cohort (n=34), the model achieved an AUC of 0.721 with maintained high specificity (95%) and positive predictive value (PPV, 90%).

CONCLUSIONS

Early on-treatment integrated dynamics of HBeAg, HBsAg, GGT and ALB showed predictive value for subsequent HBeAg SR. The eight-variable model may assist monitoring or alternative strategies in patients receiving treatment.