Bayesian network to predict hepatitis B surface antigen seroclearance in chronic hepatitis B patients

Area of Science:

• Hepatology
• Biostatistics
• Predictive Modeling

Background:

• Chronic hepatitis B (CHB) infection affects millions globally.
• Predicting hepatitis B surface antigen (HBsAg) seroclearance is crucial for treatment management.
• Existing models often lack interpretability and interaction considerations.

Purpose of the Study:

• To develop and validate a Bayesian network (BN) model for predicting HBsAg seroclearance in CHB patients.
• To identify key predictors influencing HBsAg seroclearance.
• To assess the model's accuracy and clinical utility.

Main Methods:

• A case-control study involving 1966 CHB patients.
• Utilized demographic, clinical, laboratory, and imaging data.
• Constructed a BN model to estimate HBsAg seroclearance probability.

Main Results:

• Baseline HBsAg and HBeAg levels, virological response, and HBeAg seroclearance were significant predictors.
• Patients with low baseline HBsAg, negative HBeAg, and initial virological response showed higher seroclearance probability.
• The BN model achieved an AUC of 0.896, sensitivity of 0.840, specificity of 0.880, and accuracy of 0.878.

Conclusions:

• The developed BN model provides an accurate and interpretable tool for predicting HBsAg seroclearance.
• The model can aid clinicians in making informed decisions for CHB management.
• Further validation in diverse patient populations is warranted.