%0 Journal Article
%T Spatially Explicit Nonlinear Models for Explaining the Occurrence of Infectious Zoonotic Diseases
%A Stephen Jones
%A William Conner
%A Bo Song
%J ISRN Biomathematics
%D 2012
%R 10.5402/2012/132342
%X Zoonotic diseases can be transmitted via an arthropod vector, and disease risk maps are often created based on underlying associative factors within the surrounding landscape of known occurrences. A limitation however is the ability to map disease risk at a meaningful geographic scale, and traditional regression modeling approaches may not always be appropriate. Our objective was to determine if nonlinear modeling could improve explanatory power in describing the occurrence of 2 tick-borne diseases (Lyme disease (LD) and Rocky Mountain spotted fever (RMSF)) known to occur in Tennessee. Medically diagnosed cases of LD (ICD-9: 088.81) and RMSF (ICD-9: 082.0) were extracted from a managed care organization data warehouse for the 2000¨C2009 time period. Four separate modeling techniques were constructed (logistic regression, classification and regression tree (CART), gradient boosted tree (GBT), and neural network (NNET)) and compared for accuracy. Results suggest that areas higher in disease prevalence were not necessarily the same areas having high predicted disease risk. GBT best explained LD occurrence (misclassification rate: 0.232; ROC: 0.789). RMSF prevalence was best explained with an NNET algorithm (misclassification rate: 0.288; ROC: 0.696). Covariates explaining disease risk included forested wetlands, urbanization, and median income. Nonlinear modeling may provide better results than traditional regression-based approaches. 1. Introduction Because zoonotic diseases are transmitted via an arthropod vector, it is often of interest to understand vector habitat in the epidemiologic study of diseases. It is common in spatial epidemiology to describe vector habitat and then create causal inference risk maps of potentially high-risk areas based on habitat preferences [1, 2]. These geospatial mapping exercises outline areas having high probabilities of vector prevalence and then infer disease risk based on probable presence or absence. For example, abundance of the tick genus Ixodes, one of which is the vector primarily responsible for the transmission of Lyme disease (LD), is associated with temperature, landscape slope [3], forested areas with sandy soils [4], and increasing residential development [5]. Tularemia prevalence is positively associated with dry forested habitat areas [6]. Human populations living within forested areas and on specific soils are at higher risk of contracting LD [7, 8]. Human monocytic ehrlichiosis (HME or Ehrlichia chaffeensis) is more associated with wooded habitats compared to neighboring grassy areas [9]. A major
%U http://www.hindawi.com/journals/isrn.biomathematics/2012/132342/