Epidemiological and clinical characteristics of influenza patients in respiratory department under the prediction of autoregressive integrated moving average model

This study was to explore the epidemiological distribution characteristics and future development trends of influenza-like illness (ILI) by autoregressive integrated moving average model (ARIMA). The information on ILI in the hospital from January 2016 to August 2020 was collected in this study. Firstly, the differences in distribution of different virus subtypes, the distribution of epidemic time, and the gender and age of susceptible groups were analyzed comprehensively. Secondly, the ARIMA model was constructed based on the number of weekly influenza cases and the percentage of visits to predict the percentage of ILI in the percentage of visits (ILI%). The optimized Elman neural network (ENN) model was applied to combine the ARIMA model into the ARIMA-ENN model, so as to improve the prediction effect of the ARIMA model. Then, the fitting prediction effect was evaluated with the ARIMA model. The results showed that there was a total of 11,293 suspected ILI cases in hospital, of which 773 were positive results, with the ILI% of 6.84%. There were obvious differences in ILI% among patients of different ages in various years (P < 0.0.5). The peak ILI% was reached in patients aged 60-80 in 2019 in both spring and winter. The positive rates of influenza viruses showed visible difference in various years (P < 0.0.5), and the distribution of influenza virus subtypes in different years and seasons was changeable. Finally, the ARIMA model and the ARIMA-ENN model were used for fitting prediction. Compared with the prediction results of the ARIMA model, the prediction accuracy of the ARIMA-ENN model was improved by 11%, while the mean squared error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percent error (MAPE) decreased by 4.8%, 13.8%, 9.8%, and 4.8%, respectively. It indicated that the peak of ILI in this region was mainly concentrated in spring and winter, and the predominant strains were changeable in different years. In addition, using ARIMA-ENN model to predict influenza monitoring data showed superior performance.