M Siva Durga Prasad Nayak and KA Narayan
Background: Dengue is one of the most serious and fast emerging tropical diseases. In India, over the past decade, Dengue fever has increased in frequency and geographical extent. Detailed information about when and where DF/DHF outbreaks occurred in the past can be used for epidemiological modeling to predict future trends and impending outbreaks. Based on this background, an attempt was made to convert the available monthly data of dengue fever incidence in the Kerala state into seasonal ARIMA model to forecast disease burden.
Methods: The current study was retrospective analytical study using secondary data from department of Director of Public Health of Kerala state, India. The monthly reports of integrated disease surveillance project (IDSP) for a period of thirteen years from 2006 to 2018 were downloaded and data of dengue fever cases was extracted from the downloaded pdf files. Using SPSS trial version 21 and a sample data set, several ARIMA models were run and best suited seasonal ARIMA model was identified. The selected model was then used to forecast monthly dengue fever incidence from the next coming year i.e. from 2007 onwards. Monthly forecasted incidence and monthly real incidence of dengue fever cases from 2007 to 2018 were compared and the difference between them was tested using paired t test.
Results: Seasonal ARIMA (1, 0, 0) (0, 1, 1)12 model was found to be the best fitted model for the given data. Stationary R square value of selected models is 0.815. Ljung–Box test value is 11.271 and p value is 0.792, indicating that the selected model is adequate. Average number of forecasted incidence of dengue fever cases from January 2007 to December 2018 were nearer to the real incidence in every month, but the difference among them was not statistically significant, indicating that the model fit was good.
Conclusion: A Seasonal ARIMA (1, 0, 0) (0, 1, 1)12 was selected as best suited model to predict the future incidence of dengue fever cases in the forthcoming period. The technique would be useful for health care administrators for better preparedness. The model can be made dynamic to include the current data and for a more dynamic model.