Study On The Influencing Factors Of Hand Foot And Mouth Disease Prevalence And Establishment Of Prediction Models

ObjectivesTo discuss the period and pattern of Hand Foot and Mouth Disease（HFMD）epidemic,to study the influencing factors of HFMD prevalence,and to build a prediction model based on wavelet analysis to predict its epidemic.Materials and methodsThe weekly number of HFMD cases from 2009 to 2016 were collected from the Zhengzhou Center for Disease Control and Prevention.The climatological data of the same period were collected from Zhengzhou Meteorological Bureau,which main variables were temperature,sunshine duration,air pressure and humidity.Wavelet transform and cross-wavelet analysis were used to study periodicity of HFMD and its cohenrence with climatic factors.A wavelet-based SARIMA（seasonal autoregressive integrated moving average）-NNAR（neural network nonlinear autoregressive）hybrid model was fitted with HFMD weekly cases from 2009 to 2016 in Zhengzhou,China.Autocorrelation function（ACF）and Box-Ljung test were used to check the residual sequences.Then,use this model to forecast HFMD cases in 2016 and estimate it’s performance with actual data.At the same time,single SARIMA model,simplex NNAR model and pure SARIMA-NNAR hybrid model were established as well for comparison and estimation.The indices are root mean square error（RMSE）,mean absolute error（MAE）and mean absolute percentage error（MAPE）.ResultsThe mean air temperature,sunshine duration,average air pressure and average relative humidity and HFMD in Zhengzhou all have one-year period and they share high power with HFMD in a one-year cycle.While the sunshine duration,average relative humidity and HFMD in Zhengzhou also have half-year period and they share high power with HFMD in a half-year cycle.The wavelet-based SARIMA-NNAR combination model we fitted is composed of one-layer wavelet decomposition,SARIMA（4,1,0）（1,1,0）₅₂（P value of Box-Ljung test of residual sequences is 0.1476,ACF shows no autocorrelation for residuals）and NNAR（16,1,9）₅₂（P value of Box-Ljung test of residual sequences is 0.0546,ACF shows no autocorrelation for residuals）.This model is well fitted and can be used for prediction.The RMSE,MAE and MAPE of the wavelet-based SARIMA-NNAR hybrid model are 71.29,37.45 and 21.81 in training set,while 296.18,227.25 and61.32 in validation set.Compared with the other three models,its RMSE in the validation set is the smallest.For RMSE,MAE,MAPE in training set and MAE,MAPE in validation set,the indices of single NNAR model are the smallest,followed by wavelet-based SARIMA-NNAR hybrid model.The graph of fitting and prediction shows that the fitting and prediction values of the wavelet-based SARIMA-NNAR hybrid model are close to the actual values.ConclusionsMean air temperature,sunshine duration,mean air pressure and mean relative humidity are closely related to HFMD.Wavelet analysis can be used to study the relationship between HFMD prevalence and climatic factors.The wavelet-based SARIMA-NNAR hybrid model has good performance and is suitable for predicting the number of HFMD cases.This kind of model is helpful for the prevention and control of HFMD.