Identifying Extreme Behaviour and Fitting Empirical Models for Dengue Incidents of Selected Regions in Sri Lanka

Abstract

Dengue fever is one of the most significant mosquito-borne diseases caused by a virus. Numerous methods available to predict dengue incidents are mainly focused on the mean features of events. However, understanding the extreme behaviour of dengue incidents is important, and that will allow sufficient time to take the necessary decisions and actions to safeguard the situation for local authorities. Therefore, this study mainly focuses to model the risk of rare dengue events, that is, extreme dengue events, and to identify the best-fitted distributions for the study areas. Further, the weather-based dengue empirical models for dengue incidents were fitted using climatological factors to forecast potential outbreaks. The weekly dengue incidents and climatology data (rainfall, temperature, and relative humidity) from January 2010 to December 2018 for seven administrative districts were collected from the Epidemiology Unit of the Ministry of Health (MoH), and the Meteorology Department of Sri Lanka, respectively. The Extreme value theory (EVT) was used to analyse the extreme dengue incidents, and the negative binomial generalized linear model was used to fit weather-based dengue empirical models. Various lag times between dengue and weather variables were analysed to identify the optimal dengue forecasting period. The best fitted empirical models for dengue incidents were identified for the selected districts. The Generalized Linear Negative Binomial (GLNB) models with monsoon season as a covariate, lag 0 model is the suitable model for Colombo and Gampaha districts, and lag 1 model is the suitable for Kurunegala whereas lag 2 model is the best for Anuradhapura with highest prediction accuracy. For Badulla district, lag 2 model without having monsoon season as a covariate shows highest prediction accuracy. The prediction accuracy is the same for the models with or without having the monsoon season as a covariate for Kandy (lag 2) and Ratnapura (lag 3) districts.