Abstract

Dengue and dengue hemorrhagic pose significant burdens in many tropical countries. Dengue incidences have perpetually increased, leading to an annual (uncertain) peak. Dengue cases cause an enormous public health problem in Thailand because there is no anti-viral drug against the dengue virus. Searching for means to reduce the dengue incidences is a challenging and appropriate strategy for primary prevention in a dengue outbreak. This study constructs the best predictive model from past statistical dengue incidences at the provincial level and studies the relationships among dengue incidences and weather variables. We conducted experiments for 65 provinces (out of 77 provinces) in Thailand since there is no dengue information for the remaining provinces. Predictive models were constructed using weekly data during 2001-2014. The training set are data during 2001-2013, and the test set is the data from 2014. Collected data were separated into two parts: current dengue cases as the dependent variable, and weather variables and previous dengue cases as the independent variables. Eight weather variables are used in our models: average pressure, maximum temperature, minimum temperature, average humidity, precipitation, vaporization, wind direction, wind power. Each weather variable includes the current week and one to three weeks of lag time. A total of 32 independent weather variables are used for each province. The previous one to three weeks of dengue cases are also used as independent variables. There is a total of 35 independent variables. Predictive models were constructed using five methods: Poisson regression, negative binomial regression, quasi-likelihood regression, ARIMA(3,1,4) and SARIMA(2,0,1)(0,2,0). The best model is determined by combinations of 1–12 variables, which are 232,989,800 models for each province. We construct a total of 15,144,337,000 models. The best model is selected by the average from high to low of the coefficient of determination (R2) and the lowest root mean square error (RMSE). From our results, the one-week lag previous case variable is the most frequent in 55 provinces out of a total of 65 provinces (coefficient of determinations with a minimum of 0.257 and a maximum of 0.954, average of 0.6383, 95% CI: 0.57313 to 0.70355). The most influential weather variable is precipitation, which is used in most of the provinces, followed by wind direction, wind power, and barometric pressure. The results confirm the common knowledge that dengue incidences occur most often during the rainy season. It also shows that wind direction, wind power, and barometric pressure also have influences on the number of dengue cases. These three weather variables may help adult mosquitos to survive longer and spread dengue. In conclusion, The most influential factor for further cases is the number of dengue cases. However, weather variables are also needed to obtain better results. Predictions of the number of dengue cases should be done locally, not at the national level. The best models of different provinces use different sets of weather variables. Our model has an accuracy that is sufficient for the real prediction of future dengue incidences, to prepare for and protect against severe dengue outbreaks.

Highlights

  • Dengue diseaseDengue is a mosquito-borne disease, in which Aedes aegypti is the main vector

  • The detailed results of each province are provided in S1 Appendix including: 1. Scatter plot for dengue cases vs. selected independent variables, using the weekly period during January 2001–December 2013, average pressure, maximum temperature, minimum temperature, average humidity, precipitation, vaporization of water, wind direction, and wind power

  • 4. (a) Two plots for lag-time of dengue incidences and Autocorrelation functions (ACF) vs. Partial Autocorrelation Functions (PACF), calculated from the Autoregressive Integrated Moving Average (ARIMA) model (b) Summary plots of time series analysis, multiple plots include a plot of the predicted model vs. time, plot of ACF residual vs. lag-time of dengue incidences, residual Q-Q plot of standard residual, and p-value for Ljung-Box statistics of PACF over the training data starting from January 2001 to December 2013

Read more

Summary

Introduction

Dengue diseaseDengue is a mosquito-borne disease, in which Aedes aegypti is the main vector. Dengue fever (DF) and dengue hemorrhagic fever (DHF), a severe form of the disease, are caused by four dengue serotypes represented by DEN 1, 2, 3, and 4 [1,2,3,4]. Reinfections of different serotypes can cause severe illnesses or deaths Once infected, it can take 3–14 days for the virus to incubate. Clinical manifestations of DHF include reduced blood pressure due to plasma leakage from capillaries and bleeding due to low platelet counts and impairment of platelet functions. If their blood pressure is extremely low, patients may enter the stage of dengue shock syndrome, lose consciousness, and pass into the last stage of shock

Methods
Results
Conclusion
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call