Extended convolution model to bayesian spatio-temporal for diagnosing the DHF endemic locations

Abstract

We develop a Bayesian spatio-temporal model based on a spatial Poisson-Lognormal convolution model. The convolution model consists of two component, i.e.; heterogeneity (fixed effects) and uncertainty (local and global random effects). We introduce an additional component in the development model so-called temporal trend. The developed model is adjusted based on dengue hemorrhagic fever (DHF) characteristics data. The DHF cases are considered as two-level hierarchical data since the cases are measured at smallest locations (as level 1) and the locations are nested to larger area (as level 2). Because the data are dis- crete with non-normal distribution, the development model is expressed as a generalized linear model. The parameters of the model are estimated using its full conditional distribu- tion, respectively, by applying Bayesian Markov Chain Monte Carlo (MCMC) computational technique. We apply the model to analyze the endemic locations of DHF cases in the city of Kendari, Southeast Sulawesi, Indonesia. The developed model can be well identifying the DHF prone locations in the city more than the convolution model. In this model we use two predictors, i.e. rainfall and population density, and both of them are statistically significant. Applying the model to the real data, the results show that Kadia and Puwatu districts are endemic locations. The best time to intervene the DHF risk in the both locations to prevent the spread to other locations was in the month of January

Keywords:

• Bayesian spatio-temporal
• spatial convolution
• DHF
• Generalized linear model
• MCMC
• and Poisson-Lognormal