Incidence of Acute Hemorrhagic Conjunctivitis in Chongqing: A Forecasting Study Based on Mathematical Models

Xiaobing Xian¹Sitian Wu²Yandi Fu²Xiaoli Fan²Yan Chen³Zhangmei Hou³Yinzhi Chen^3*

• ¹The Thirteenth People's Hospital of Chongqing, Chongqing, China
• ²Chongqing Medical University, Chongqing, China
• ³Chongqing General Hospital, Chongqing, China

Background: Acute hemorrhagic conjunctivitis (AHC) is a highly infectious eye disease. It poses a significant threat to public health given its propensity for rapid transmission in densely populated areas. Recent epidemiological data have demonstrated a distinct seasonal outbreak pattern in Chongqing. However, conventional single prediction models exhibit limitations in accurately capturing the complex spatiotemporal transmission characteristics of AHC. This study endeavors to compare the performance of different mathematical models in forecasting AHC incidence in Chongqing. Through the investigation of optimal predictive methodologies, this study establishes a theoretical foundation for relevant department to formulate policies for preventing AHC. Methods: The monthly incidence data of AHC in Chongqing from March 2019 to October 2024 were collected from the official website of the Chongqing Municipal Health Commission. Five predictive models (SARIMA, KNN, Prophet model as well as SARIMA-KNN and SARIMA-Prophet model) were employed to fit the incidence data. The data from March 2019 to December 2023 was designated as the training set, while the data from January 2024 to October 2024 served as the test set. Model performance was evaluated through multiple metrics, including MSE, RMSE, MAE, and MAPE. Subsequently, the Diebold-Mariano test was implemented to statistically assess the significance of predictive performance differences among the five models. Results: During the period from March 2023 to October 2024, the incidence rate of AHC in Chongqing showed a pronounced seasonal fluctuation pattern, with the peak period consistently occurring between June and September annually. The comparative analysis of model performance revealed that the SARIMA-KNN hybrid model demonstrated optimal performance metrics in terms of MSE, MAE, RMSE, and MAPE. Furthermore, the predicted curve of the SARIMA-KNN model demonstrated superior fitting accuracy compared to the actual curve. The Diebold-Mariano statistical test confirmed that the SARIMA-KNN model's performance was significantly superior to other models. Conclusion: In comparison with the other four models, the SARIMA-KNN hybrid model effectively integrates the temporal characteristics of AHC incidence. It offers the technical support for the development of early warning systems and the formulation of prevention and control strategies in Chongqing. This approach holds substantial practical significance in the field of public health.