Bayesian Spatial Modeling and Mapping of Dengue Fever: A Case Study of Dengue Feverin the City of Bandung, Indonesia

Abstract

Dengue Fever (DF) is an acute febrile disease caused by the dengue virus which is transmitted by the Aedes Aegypti mosquito. The World Health Organization (2009) noted that Asia has the highest incidence of dengue fever in the world and Indonesia is at the top of the list of the Southeast Asian countries having the highest dengue fever cases. Modeling and mapping of the spread of dengue fever are needed to monitor endemic regions. The modeling and mapping will eventually form the basis for the preventive actions taken to overcome the spread of the disease. The most common approach for disease modeling and mapping, i.e. the spatial one, is based on a log-linear relationship between the relative risk and the local variation without taking covariates into account. However, ignoring covariates results in bias and unreliable estimates of the relative risk. Being a spatial approach, there is a local variation of the relative risk of this model influenced by the environmental factors, such as climates and human behavior; while on the other hand, the general assumption in spatial modeling is stationarity of the mean and covariance. Stationarity assumption of the mean implies the associations between the relative risk and a set of covariates which is constant over regions. In actuality, the relative risk modeling usually violates the stationarity assumption because there are the spatial dependencies and unobserved factors that influence the relative risk. Non-stationarity of the mean can be accommodated by using a Spatially Varying Coefficients (SVC) model. The Generalized Linear Mixed Model (GLMM) is proposed as well and Bayesian inference with Integrated Nested Laplace Approximation (INLA) is applied to construct the SVC and compare with Fixed Coefficient model (FCM) or the global model. The SVC model generated is finally applied to dengue fever incidence in the city of Bandung, Indonesia. The covariates included in the model are population density, larva-free home index, healthy housing index and rainfall. The Deviance Information Criterion (DIC) is applied for the model selection. Based on the application of the DIC, it was found out that the SVC model results in a better estimation of the relative risk than the FCM, with DIC = 266.24. The research shows that the percentage of larva-free home index becomes the dominant effect on the relative risk and it is almost constant over regions. The dengue fever map is finally constructed from the posterior means of the relative risk. The resulting map can be used to guide disease spread assessment and to set up mitigation strategies, including those related to health impact