[Epidemiological characteristics of chronic hepatitis B and establishment of prediction model based on socio-demographic index in Shenzhen, 2005-2023].

Objectives: To analyze the epidemiological characteristics and incidence trends of chronic hepatitis B in Shenzhen from 2005 to 2023, develop a prediction models with performance evaluation, explore its associations with social demographic index (SDI) and inform targeted prevention strategy development. Methods: Based on surveillance data of infectious diseases, descriptive epidemiological methods were applied to analyze the spatiotemporal and population distribution characteristics. A multifactorial prediction model integrating the SDI was established, and its predictive performance was evaluated by using data from 2020-2023. Model accuracy was evaluated by using root mean square error and mean absolute percentage error (MAPE). The association between SDI and incidence rates was assessed through generalized linear models. Results: A total of 235 703 chronic hepatitis B cases were reported cumulatively in Shenzhen from 2005-2023, with an annual average incidence rate of 98.84/100 000. Long-term trends revealed a significant increase in the incidence from 2005 to 2019. The incidence rate was 2.48 times higher in men than in women, and the majority of cases occurred in age group 20-50 years. The cases were mainly workers in manufacturing and services. Seasonal incidence peaks were observed in March and during May to November. The overall SDI exhibited a consistent upward trend, and the positive correlation between SDI and incidence rate was observed in central urban districts (Futian and Nanshan). In contrast, industrial zones (Guangming and Bao'an) saw a significant decline in incidence rates due to intensified prevention interventions despite the increase of SDI level. Model predictions indicated that the multivariate long short-term memory (LSTM) deep learning model integrating SDI parameters outperformed both the spatiotemporal covariate- enhanced model and the augmented Bayesian structural time series model, with MAPE of 4.71%, 7.66% and 10.30%, respectively. Conclusion: SDI is a key social determinant associated with hepatitis B transmission risks, and dynamic thresholds can be established to develop tiered early warning mechanisms. It is suggested to integrate multisource SDI data into the LSTM framework, implement targeted interventions such as "rapid antibody screening in key areas + vaccination boosters for high-risk populations" and improve the timeliness of epidemic response through hybrid models to reduce disease burden level.