Epidemiological characteristics and incidence prediction of varicella from 2014 to 2023 in Chongqing, China.

Objective: To characterize the epidemiology of varicella in Chongqing during 2014-2023, and establish the most suitable prediction model for the varicella incidence trends in the city, providing scientific support for early warning of the varicella incidence trend, the formulation and the optimizing of precise varicella preventive strategies. Methods: Varicella reported cases in Chongqing during 2014-2023 were collected to characterize the epidemiology, all the varicella cases were sourced from the "Information management system for infectious disease reporting." Seasonal autoregressive integrated moving average (SARIMA) model, long short-term memory (LSTM) model and SARIMA-LSTM hybrid models were established based on the surveillance data. The fitting effects and prediction performances of the established models in this study were evaluated through root mean squared error (RMSE) and mean absolute error (MAE). Results: In Chongqing, 265,824 varicella cases were reported during 2014-2023, the annual average reported incidence rate is 85.99/100,000. The incidence of varicella initially increased and then fluctuated with a downward trend, showing clear seasonality. The peak incidence periods occurred in May-June and October-December each year. The average incidence rates for males and females were 88.92/100,000 and 80.94/100,000, respectively. Children under 15 years old, particularly school-aged children and students, represented the main affected population. The annual incidence rates across districts ranged from 26.90/100,000 to 145.76/100,000. The global spatial autocorrelation analysis indicate that the varicella incidence rate in Chongqing does not exhibit spatial autocorrelation in each year, while the local spatial autocorrelation analysis identified "hotspot" areas primarily concentrated in the main urban metropolitan area. Among the three prediction models based on the monthly incidence rate of varicella from January 2023 to December 2023, LSTM model has the best prediction performance, with RMSE and MAE of 1.52 and 1.19, respectively. The RMSE and MAE of the SARIMA model are 1.91 and 1.49, respectively, while the RMSE and MAE of the SARIMA-LSTM model are 1.99 and 1.47, respectively. Conclusion: Sustained and effective measures need to be adopted to better curb the spread and prevalence of varicella, particularly among children and adolescents, as well as in the central urban areas and other high-incidence regions. The LSTM model can effectively predict varicella incidence trends, providing scientific evidence to assist relevant authorities in making decisions regarding varicella prevention and control.