Forecasting monthly hepatitis B cases in China: a nationwide comparative study based on surveillance data

Area of Science:

• Epidemiology
• Biostatistics
• Public Health

Background:

• Accurate forecasting of hepatitis B trends is crucial for China's national surveillance and public health strategies.
• Limited research exists on the comparative performance of various forecasting models for long-term hepatitis B surveillance data in China.

Purpose of the Study:

• To evaluate and compare the performance of ten different forecasting models for national hepatitis B surveillance data in China.
• To identify the most effective forecasting framework for predicting hepatitis B case counts.

Main Methods:

• Utilized monthly reported hepatitis B cases from China (2004-2024), with 2004-2023 as the training set and 2024 for validation.
• Assessed ten models: SARIMA, BSTS, four machine-learning models (RF, SVM, XGBoost, LightGBM), and four hybrid SARIMA-ML models.
• Performance metrics included MAE, NRMSE, MAPE, and NSE; AAPC estimated long-term trends.

Main Results:

• The annual hepatitis B case count showed no significant long-term trend (AAPC = -0.561%, P = 0.12), but a stable seasonal pattern was observed monthly.
• Bayesian structural time series (BSTS) demonstrated the best validation performance across all evaluated metrics.
• Among hybrid models, SARIMA-LightGBM showed the strongest predictive capability.

Conclusions:

• Forecasting model performance varied significantly for national hepatitis B surveillance.
• BSTS emerged as the superior model for overall predictive accuracy.
• BSTS-based forecasting can support early warning systems, resource allocation, and long-term hepatitis B surveillance planning in China.