Predictive Modeling of West Nile Virus Outbreaks Using RemotelySensed Data

1
Predictive Modeling of West Nile Virus Outbreaks
Using Remotely-Sensed Data
Dr. Michael Ward Professor of Epidemiology College
of Veterinary Medicine Texas AM University
James Steele Conference on Diseases in Nature
Transmissible to Man, Austin, 11 June 2007
2

• James Schuermann
• Zoonosis Control Group
• Texas Department of State Health Services, Austin
  TX
• Linda Highfield
• Department of Veterinary Integrative Biosciences
• Texas AM University, College Station TX
• partial funding provided by the
• Texas Equine Research Advisory Committee

3
Outline

• Background
• Methods
• Results
• Discussion
• Conclusions

4
1. Background
5
(No Transcript)
6
West Nile Virus

• family Flaviviridae genus Flavivirus
• Japanese Encephalitis serocomplex, includes
• Japanese encephalitis
• Murray Valley encephalitis
• St. Louis encephalitis
• Kunjin
• antigenically, all closely related

7
WNV History

• first occurrence in U.S. 1999 (
  Bronx Zoo, New York )
• by 2001 extension of range to include Florida
• 2002 large equine epidemic
• by 2003 46 states, 7 Canadian provinces, 5
  Mexican states
• only states WNV not detected
• Alaska, Hawaii

8
WNV Life Cycle

• Vector
• Mosquito
• Reservoir
• Wild birds
• Dead end host
• Horses and humans

9
WNV Mosquito Vectors

• biological and mechanical vectors
• 14 species identified
• Culex spp. most likely in the U.S.
• breed in standing water
• Cx. pipiens, quiquefasciatus, tarsalis
• Aedes spp. may spread disease to horses
• breed in locations where water will be present

10
WNV Avian Reservoirs

• responsible for distribution
• gt110 species of birds
• most susceptible species include American
  crows, fish jays, blue jays
• game species (wild ducks, geese, pheasants,
  turkeys, pigeons, doves)
• raptors (owls, hawks, eagles)

11
(No Transcript)
12
First indicators of WNV activity
13
WNV Surveillance Programs

• avian mortality surveillance tracking system
• mosquito trapping and testing
• testing wild birds, sentinel chickens,
  horses and humans with neurologic disease
• forecasting systems environmental variables
• temperature
• precipitation
• remotely-sensed data

14
2. Methods
15

• reported cases of equine WNV encephalomyelitis
  2002, 2003 and 2004
• time series of case reports, 2-week window
• image data 2-week 1km2 resolution rasters of the
  Normalized Difference Vegetation Index (NDVI)
• mean NDVI for each 2-week period
• periods with versus without reported cases
• autoregressive model NDVI as a predictor of
  equine WNV cases (scaled, ? transform)

16

• What is the NDVI?
• Advanced Very High Resolution
• Radiometer (AVHRR) sensor,
• NOAA polar-orbiting satellite
• Normalized Difference Vegetation Index
• visible and near-infrared data
• daily observations ? biweekly 1km2 resolution
  raster based on daily maximum observed NDVI value
• resulting 1x1 km pixel represents maximum scaled
  NDVI value during each 2 weeks of the study period

17
(No Transcript)
18
(No Transcript)
19
3. Results
20
(No Transcript)
21
(Plt0.001)

• correlation, number of cases reported versus
  NDVI 45

22

• cases 0.9102 8.5762 (casesweeks 12)
  5.6137 (casesweeks 34)
• 0.9262 (NDVIweeks 12) 0.2661 (NDVIweeks
  34)
• no. observed versus predicted cases highly
  correlated (rSP 83, Plt0.001)

23

• mean difference, observed versus predicted cases,
  P 0.973

24
4. Discussion
25
Prevention and Control

• reduce exposure
• indoor housing, repellants?
• mosquito control
• larvicides, adulticides, environment
• vaccination
• killed or recombinant canarypox-vectored
• 2 doses, 3-6 weeks apart annual booster

26
Forecasting Systems

• anticipate increases in risk
• optimize control strategies
• increased awareness
• identify hotspots
• sentinel warning for zoonotic disease

27
5. Conclusion
28

• remotely-sensed data
• availability
• low-cost
• coverage
• could be used to
• enhanced WNV surveillance
• provide early warning of increased risk
• identify hotspots
• warn of potential zoonotic transmission of WNV