Using data on the incubation of West Nile virus, the life cycle of mosquitoes, and the flight of birds, DYCAST software successfully predicted where human infections would occur. Image: Centers for Disease Control |
A computerized epidemiological model of the spread of the mosquito-borne
West Nile virus in 17 counties of California
in 2005 successfully predicted where 81.6% of human cases of the disease would
arise and defined high-risk areas where the risk of infection turned out to be
39 times higher than in low-risk areas, according to newly published research.
The DYCAST software used in those predictions is now open-source and is being
applied to other diseases.
“One of the things that really differentiates DYCAST from other approaches
is that it’s based on biological parameters,” says Ryan Carney, a Brown University
graduate student who is the lead author on a paper about DYCAST’s
performance that appears in Emerging
Infectious Diseases. “All of the parameters in the model are based
on experimental data related to the biology and ecology of the virus, mosquito
vector, and bird host.”
For example, the spatial parameters of the model include how far mosquitoes
and infected birds are likely to fly. Key time parameters include how long the
virus needs to incubate in mosquitoes before they become infectious and the
lifespan of infected birds. Carney said that by using biology to define the
geographic and temporal attributes of the model rather than county or census
tract borders, which are convenient for humans but irrelevant to birds and
mosquitoes, the model allowed the California Department of Public Health to
provide early warnings to an area stretching from the Bay Area through
Sacramento to the Nevada line, as well as regions in southern California.
Carney implemented the software when he worked for the California department in 2005. (The software
was created by Constandinos Theophilides at the City University of New York.)
Feeding the model in 2005 were 109,358 dead bird reports phoned in or entered
by members of the public via a state hotline and Website.
As more dead birds were reported in close proximity, the software would
generate daily maps of areas at high risk for human infection, providing an
early warning to local public health officials. The software, for example,
predicted areas as high-risk more than a month before the first human cases
arose, on average.
In Sacramento County, location of the largest West Nile virus epidemic in
the United States that year, DYCAST helped mosquito control officials target
their testing and spraying resources—actions that ultimately reduced human
illness, Carney says.
After 2005, the department implemented the model throughout the state,
although the number of human cases and reported dead birds, along with the
model’s prediction rates, dropped sharply.
In 2007 Carney enrolled as a master’s student at Yale and adapted the
DYCAST model to track dengue fever in Brazil, using a version of the
software that his CUNY collaborators had converted to an open-source platform.
With the specific parameters of that disease, DYCAST was able to predict its
spread in the city of Riberão Preto in Brazil, Carney
says, citing unpublished data.
Carney has continued his analysis
and development of DYCAST and dengue at Brown, where he is a doctoral student
of ecology and evolutionary biology. He said the software at its core has
potential to be adapted as an early warning system for other infectious
diseases or even bioterrorism attacks.