This article is more than 5 years old. Information may no longer be current.
Computer model predicts West Nile virus outbreaks
Click Here to Manage Email Alerts
Researchers said they developed a computer model that could create accurate retrospective forecasts for West Nile virus outbreaks in two U.S. counties, showing that real-time forecasts for the most common arbovirus in the country are possible.
“There is a great deal of variation in outbreak intensity and duration year to year,” Nicholas B. DeFelice, PhD, post-doctoral research scientist at the Columbia University Mailman School of Public Health, said in a news release. “Absent a computer model, it’s difficult to predict the impact of an outbreak, even once the outbreak is underway, and thus it is important that robust quantitative decision tools are developed to help guide control efforts.”
Since being introduced in the United States in 1999, West Nile virus has become the most common arbovirus in the country, according to DeFelice and colleagues. More than 2,000 cases of West Nile disease, including 94 deaths, were reported to the CDC for 2016, but national case numbers for the mosquito-borne illness have been much higher in other years.
Local outbreaks of West Nile virus remain unpredictable, according to DeFelice and colleagues, so they aimed to create a way to predict outbreaks by drawing on a method used previously for forecasting models of infectious diseases such as Ebola and influenza. Their approach first involved accounting for West Nile virus transmission between mosquitoes and birds while also accounting for spillover to humans. The model was then coupled with field collection data that documented mosquito infection rates and reported human cases of West Nile virus in Suffolk County, New York.
DeFelice and colleagues used an algorithm to create retrospective forecasts of West Nile virus in Suffolk County from 2001 to 2014, initiating the model 4 weeks before the first positive mosquito observation in each annual outbreak. They accurately forecasted outbreak peaks with 1 week’s notice more than 80% of the time and were also able to accurately predict the total number of human cases.
According to the news release, DeFelice and colleagues had similar success using the model to create retrospective forecasts for West Nile virus outbreaks in Cook County, Illinois, from 2007 to 2014.
“Reliable West Nile virus forecasts would give public health officials a leg up on efforts to control mosquito populations and reduce human West Nile virus cases, and could even help them refine these efforts,” Jeffrey Shaman, PhD, associate professor of environmental health sciences at the Columbia University Mailman School of Public Health and another author on the study, said in the release. “With weeks of advance notice, officials could better plan for spraying mosquito breeding grounds, alert the public, and determine if parks and camping grounds should be closed.” – by Gerard Gallagher
Disclosure: Shaman reports partial ownership of SK Analytics. The remaining authors report no relevant financial disclosures.
Perspective
Back to Top
The rapid development of West Nile virus outbreaks without forewarning creates a considerable public health challenge. Development of methods to accurately predict these outbreaks would enable a more robust public health response, and thus remains a priority. The marked ecologic heterogeneity across the United States and the outsize influence of weather, which in itself is very difficult to predict, are among the factors hampering development of these methods.
The CDC recommends the vector index, which simply is the infection rate in vector mosquitoes multiplied by the number of trapped mosquitoes, as a means to predict West Nile outbreaks ahead of human case onsets. However, virus monitoring in captured mosquitoes and relatively constant mosquito trapping efforts from year to year are required, so this measure, while effective, is rarely available. DeFelice and colleagues modeled data, including mosquito infection rates, to predict the timing and magnitude of human West Nile virus disease incidence in Suffolk County, New York. They faced a number of potential challenges, including the relatively low West Nile incidence in that county, inconsistencies in mosquito collections from year to year, and that outside of 2 years with very high mosquito infection rates, little correlation existed between human case incidence and mosquito infection rates. Their model could accurately predict human case incidence at the time of peak mosquito infection rates, but not before, which, of course, cannot be foretold during an outbreak. Planned refinements of the model to include weather and other data will hopefully enable the model to predict human incidence at its onset when public health interventions may be best mobilized.
Click Here to Manage Email Alerts