Comparison of ARIMA and Bayesian Structural Time Series Models for Predicting the Trend of Syphilis Epidemic in Jiangsu Province.

Purpose: This study sets out to explore the forecasting value in syphilis incidence of the Bayesian structural time series (BSTS) model in Jiangsu Province.

Methods: The seasonal autoregressive integrated moving average (ARIMA) and BSTS models were constructed using the series from January 2017 to December 2021, and the prediction accuracy of both models was tested using the series from January 2022 to November 2022.

Results: From January 2017 to November 2022, the total number of syphilis cases in Jiangsu Province was 170629, with an average monthly notification cases of 2403. The optimal model was ARIMA (1,0,0) (0,1,1) 12 (AIC = 663.12, AICc = 664.05, and BIC = 670.60). The model coefficients were further tested: AR1 = 0.48 (t = 3.46, P < 0.001), and SMA1 =-0.48 (t =-2.32, P = 0.01). The mean absolute deviation, mean absolute percentage error, root mean square error, and root mean square percentage error from the BSTS model were smaller than those from the ARIMA model. The total number of syphilis cases predicted by the BSTS model from December 2022 to December 2023 in Jiangsu Province was 29902 (95% CI: 16553 ~ 42,401), with a monthly average of 2300 (95% CI: 1273 ~ 3262) cases.

Conclusion: Syphilis is a seasonal disease in Jiangsu Province, and its incidence is still at a high level. The BSTS model is superior to the ARIMA model in dynamically predicting the incidence trend of syphilis in Jiangsu Province and has better application value.