Initially detected in Uganda in 1937, West Nile virus (WNV) spread across Africa to the Middle East, West Asia, and eastern Europe [1
]. The first occurrence of WNV in the Western Hemisphere was in the New York City area during 1999, where 59 patients were hospitalized with WNV infection during August and September [2
]. West Nile virus spread rapidly across North America and by 2002, was reported in 40 states and the District of Columbia of the United States (US) with 4,156 human and 14,539 equine cases of infection [3
]. West Nile virus infection can cause neuroinvasive diseases (e.g., encephalitis) and even human fatalities [3
]. Additionally, an epidemic of WNV can impose enormous impacts on local economies. For instance, the estimated short-term economic cost incurred from the 2002 WNV epidemic in Louisiana was $20.1 million [4
]. West Nile virus disease will continue to be a public health concern in the foreseeable future; therefore, the assessment and prediction of human WNV risk within an administrative unit (e.g., county) is critical for effective WNV control and prevention and resource allocation [5
Understanding ecological factors influencing the transmission of WNV can help predict human WNV risk and improve effectiveness of control measures [1
]. Birds are the predominant hosts of WNV, whereas mammals (including humans and equines) are accidental, dead-end hosts [6
]. West Nile virus is transmitted between birds by enzootic, ornithophilic mosquito vectors and to humans and horses primarily by bird-to-mammal bridge vectors, such as Culex
]. Several environmental factors including landuse, climate, and host community composition can influence the abundance of WNV hosts and vectors, and subsequently, affect WNV transmission rates. For example, increased temperatures from April through October and increased agricultural activities enhanced human WNV disease prevalence in Colorado, Louisiana, Nebraska, and Pennsylvania [1
]. Also, increased percentage of urban land or decreased percentage of forest land increases the human WNV incidence in the northeastern United States [8
]. Finally, increased avian species diversity was associated with reduced WNV host prevalence and consequently lower WNV transmission rates throughout the United States [9
]. However, Yiannakoulia and Svenson [10
] found that inclusion of data on corvids did not improve predictive powers of their models for WNV spread in Alberta, Canada.
The role of precipitation in the WNV transmission is inconclusive. It is generally thought that precipitation limits mosquito abundance and that mosquito populations are positively correlated with precipitation. However, Miramontes et al. [1
] found no association between annual precipitation and human WNV incidence. Surprisingly, Chase et al. [11
] found that mosquito outbreaks were associated with droughts during the previous year, suggesting that droughts reduce mosquito predators and competitors, allowing mosquito abundance to increase the following year. Although it seems counterintuitive, it is plausible to extend this reasoning and hypothesize that dry weather or low precipitation would increase the number of the human WNV incidences the following year due to increased mosquito abundance. It is known that periodic droughts are associated with certain climate regions and with global teleconnection phenomena such as the El Nino-Southern Oscillation; thus, knowledge of relationships between precipitation and WNV transmission is crucial for predicting the risk of the human WNV disease outbreaks under predicted global climate change scenarios. However, the hypothesis regarding the effects of dry weather on WNV transmission has not been tested empirically.
Spatial heterogeneity exists in the risk of human exposure to infectious disease vectors [8
]. For example, occurrences of infectious diseases often are spatially (auto-) correlated; human disease incidence at a location is positively related to incidences at neighboring locations. Therefore, models using geospatial statistics and disease mapping methods are necessary to explicitly estimate the influence of neighboring site's risks on the risk at a specific site [13
We developed a Bayesian hierarchical spatial model with conditional autocorrelation (CAR) distributions to estimate the relative risk of human WNV infection in Mississippi, the United States of America (USA), and account for spatial autocorrelations. Mississippi had the highest human incidence rate (57 cases per million people) in the 2002 WNV epidemic in the US [3
]. Our objective was to test the hypothesis that dry weather would induce outbreaks of human WNV the following year, assuming that dry weather-induced increases in mosquito abundance would increase WNV transmission from birds to humans.