Discovering Spatio‐Temporal Models of the Spread of West Nile Virus

Abstract

Emerging infectious diseases are characterized by complex interactions among disease agents, vectors, wildlife, humans, and the environment. Since the appearance of West Nile virus (WNV) in New York City in 1999, it has infected over 8,000 people in the United States, resulting in several hundred deaths in 46 contiguous states. The virus is transmitted by mosquitoes and maintained in various bird reservoir hosts. Its unexpected introduction, high morbidity, and rapid spread have left public health agencies facing severe time constraints in a theory-poor environment, dependent largely on observational data collected by independent survey efforts and much uncertainty. Current knowledge may be expressed as a priori constraints on models learned from data. Accordingly, we applied a Bayesian probabilistic relational approach to generate spatially and temporally linked models from heterogeneous data sources. Using data collected from multiple independent sources in Maryland, we discovered the integrated context in which infected birds are plausible indicators for positive mosquito pools and human cases for 2001 and 2002.

Delineating West Nile Virus Transmission Cycles at Various Scales: The Nearest Neighbor Distance–Time Model

“…Spider diagrams were only selected when they satisfied the temporal sequence of actual WNV transmission and when the reporting dates of the dead birds within the reach of a spider diagram (or ”web” ) were 10-15 days prior, and the testing dates of mosquito pools were 5-7 days before the onset date of human cases found in that zip code. These temporal thresholds are well documented components of the WNV disease cycle (Theophilides et al 2003; Orme-Zavaleta et al 2006; Theophilides et al 2006). The WNV research groups at the MMCD and the MDH confirmed that these temporal windows are appropriate for capturing bird-mosquito-human dynamics because infected birds generally tend to live 10 to 14 days after being infected by the virus.…”

“…Orme-Zavaleta et al (2006) applied a Bayesian probabilistic relational technique to generate spatio-temporal models from heterogeneous data sources including location and dates of dead birds, positive mosquito pools, and human cases in order to investigate the nature of WNV transmission dynamics in Maryland (Orme-Zavaleta et al 2006). The development of the NNDT model adds to the limited number of methodologies which consider the spatio-temporal information of all three components of the WNV transmission cycle.…”

Combining Mosquito Vector and Human Disease Data for Improved Assessment of Spatial West Nile Virus Disease Risk

“…40,41 Following earlier efforts for other mosquito-borne arboviruses, 42 several studies have attempted to address these issues by combining entomological risk measures (vector larval habitat, vector abundance, abundance of infected mosquito pools) with either avian WNV surveillance data 43 or human WNV disease data (case locations, disease incidence). [44][45][46][47][48][49] However, these studies typically either were restricted to risk assessments at a crude county spatial scale 44,47,48 or failed to generate continuous spatial risk surfaces. 43,46,49 Perhaps the most complete previous approach comes from a Mississippi study combining spatial continuous environmental data with zip-code-based incidence of WNV disease in humans.…”

“…[44][45][46][47][48][49] However, these studies typically either were restricted to risk assessments at a crude county spatial scale 44,47,48 or failed to generate continuous spatial risk surfaces. 43,46,49 Perhaps the most complete previous approach comes from a Mississippi study combining spatial continuous environmental data with zip-code-based incidence of WNV disease in humans. 45 This approach was, however, to some extent impeded by a low case load of human WNV disease (276 cases were reported from Mississippi during the 2002-2003 study period).…”

A Bayesian Belief Network Model Assessing the Risk to Wastewater Workers of Contracting Ebola Virus Disease During an Outbreak

“…BBNs have been used in multiple previous studies of health and environmental risk. For example, BBNs have mapped the spread of West Nile virus, combining human epidemiologic data with information on changes in environmental conditions to pinpoint indicators of change in disease vector populations and the disease patterns that result . BBNs have also combined data on pathogen behavior with supply chain structures to model risk of Staphylococcus aureus in milk sold as “pasteurized.” A full review of BBN studies in the context of environmental risk assessment can be found elsewhere …”