Prediction and early warning of daily number of childhood common respiratory diseases based on weighted combination model and CUSUM

Background: Childhood common respiratory diseases are the most common diseases that could cause great harm to children’s health. Constructing a weighted combination and early warning models of the daily number of childhood common respiratory diseases is of great significance in analyzing the trend of the daily number of childhood common respiratory diseases in the future, and providing data support for hospital personnel regulation and resource planning. Methods: Due to the comprehensive characteristics of the trend, periodicity, autocorrelation and weekday effect of the daily number of childhood common respiratory diseases, a seasonal autoregressive integrated moving average (SARIMA) model was selected to capture the trend, periodicity and autocorrelation and a Holt-Winters model was constructed to consider the weekday effect. In addition, a multivariate long short-term memory (LSTM) model, combined with the lag effects of temperature and pollutant concentration on the daily number of childhood common respiratory diseases, was established.Based on the residual correction, a weighted combination model with the prediction results of the three models was constructed; meanwhile, a cumulative sum (CUSUM) early warning model was built. The mean absolute percentage error (MAPE) was used to evaluate the performance of the models. Results: The models were applied to forecast and warn the short-termly daily number of childhood common respiratory diseases based on the recent 2 years of daily visit data of a large-scale comprehensive hospital for children in Tianjin.The MAPE of the weighted combination model was 3.01%, which was the smallest overall. At the same time, the early warning situation of the predictive values and the real values were consistent. Both of them indicated that the weighted combination model had good forecasting performance. Conclusions: The weighted combination and early warning models can accurately predict the daily number of childhood common respiratory diseases, and send out the corresponding early warning signals, which will provide a scientific basis for the prevention and control of childhood common respiratory diseases.