Irrigated Agriculture Is an Important Risk Factor for West Nile Virus Disease in the Hyperendemic Larimer-Boulder-Weld Area of North Central Colorado

Abstract

This study focused on two West Nile virus (WNV) disease outbreak years, 2003 and 2007, and included a three-county area (Larimer, Boulder, and Weld) in North Central Colorado that is hyperendemic for WNV disease. We used epidemiological data for reported WNV disease cases at the census tract scale to: 1) elucidate whether WNV disease incidence differs between census tracts classified as having high versus lower human population density (based on a threshold value of 580 persons/km²) and 2) determine associations between WNV disease incidence and habitat types suitable as development sites for the larval stage of Culex mosquito vectors. WNV disease incidence was significantly elevated in census tracts with lower human population density, compared with those with high density of human population, in both 2003 (median per census tract of 223 and 143 cases per 100,000 population, respectively) and 2007 (median per census tract of 46 and 19 cases per 100,000 population). This is most likely related, in large part, to greater percentages of coverage in less densely populated census tracts by habitats suitable as development sites for Culex larvae (open water, developed open space, pasture/hay, cultivated crops, woody wetlands, and emergent herbaceous wetlands) and, especially, for the subset of these habitats made up by irrigated agricultural land (pasture/hay and cultivated crops) that presumably serve as major producers of the locally most important vector of WNV to humans: Culex tarsalis. A series of analyses produced significant positive associations between greater coverage of or shorter distance to irrigated agricultural land and elevated WNV disease incidence. As an exercise to produce data with potential to inform spatial implementation schemes for prevention and control measures within the study area, we mapped the spatial patterns, by census tract, of WNV disease incidence in 2003 and 2007 as well as the locations of census tracts that had either low (<25th percentile) or high (>75th percentile) WNV disease incidence in both outbreak years (relative to the incidence for each year). This revealed substantial changes from 2003 to 2007 in the spatial pattern for census tracts within the study area with high WNV disease incidence and suggests a dynamic and evolving scenario of WNV transmission to humans that needs to be taken into account for prevention and control measures to stay current and represent the most effective use of available resources.

Colorado, Culex, irrigated agriculture, mosquito larval habitat, West Nile virus disease

After the emergence of West Nile virus (WNV) in the United States in 1999, WNV disease has become a serious problem in the Great Plains and Rocky Mountains (Hayes et al. 2005). Colorado experienced a dramatic outbreak in 2003, with 2,325 reported West Nile fever (WNF) cases, 622 reported West Nile neuroinvasive disease (WNND) cases, and 63 deaths, and a second major outbreak in 2007 with 478 reported WNF cases, 100 reported WNND cases, and seven deaths (http://www.cdphe.state.co.us/dc/Zoonosis/wnv/index.html). Three neighboring counties in the north central part of the state, Larimer, Boulder, and Weld, consistently report the highest case numbers. For example, this three-county area accounted for 46% of total reported WNV disease cases in Colorado in 2003 and 50% in 2007. To cost effectively reduce the burden of disease in this WNV hot spot, we need a solid understanding of environmental, demographic, and behavioral factors associated with elevated risk of WNV exposure among the human population. This will allow us to most effectively target limited resources for prevention and control efforts, such as mosquito vector control and education campaigns, to local areas, specific environments, and time periods with the most severe risk of human exposure to mosquito vectors and WNV.

Since the emergence of WNV in Colorado in 2002, several studies on mosquito vectors or WNV disease epidemiology have concentrated on or included the Larimer-Boulder-Weld WNV disease focus. These studies have revealed that Culex mosquitoes, especially Culex tarsalis, are important local WNV vectors (Bolling et al. 2007, 2009; Gujral et al. 2007; Barker et al. 2009a; Kent et al. 2009) and that the abundance of Cx. tarsalis is high in the plains portion of the Larimer-Boulder-Weld area, but decreases dramatically with increasing elevation on the eastern slope of the Rocky Mountains in the western portions of Larimer and Boulder Counties (Eisen et al. 2008, Winters et al. 2008a, Barker et al. 2009a). Furthermore, studies based on epidemiological WNV disease data have revealed that climate and landscape factors (e.g., temperature, precipitation, and vegetation indices) are associated with elevated WNV disease incidence at the zip code scale in the eastern Colorado plains (Winters et al. 2008b), and that WNV exposure risk is influenced by behavioral factors such as use of repellents and time spent outdoors during the time of day (dusk to dawn) when Culex vectors are most active (Gujral et al. 2007). Another study direction has been to evaluate the potential for WNV surveillance in birds or mosquitoes to predict human disease cases in space or over time (Patnaik et al. 2007, Bolling et al. 2009). However, fine-scale associations between land cover and abundance of Culex vectors or occurrence of WNV disease have received only limited attention in Colorado (Barker et al. 2009b).

Our aim in this study was to determine whether land cover or demographic characteristics of the human population are associated with elevated incidence of reported WNV disease at the census tract scale within the hyperendemic Larimer-Boulder-Weld area. Specifically, we used epidemiological data for reported human WNV disease cases at the census tract scale during the outbreak years of 2003 and 2007 to: 1) elucidate whether WNV disease incidence differs between census tracts classified as having high versus lower human population density (based on a threshold value of 580 persons/km²), and 2) determine associations between WNV disease incidence and selected demographic factors (household income, owner occupancy, and population age structure) or percentage coverage by or distance to land cover types serving as potential development sites for Culex larvae. This was accompanied by mapping of spatial WNV disease incidence patterns at the census tract scale.

Materials and Methods

Study Area.

The study focused on the Larimer-Boulder-Weld area in North Central Colorado (Fig. 1). Larimer and Boulder Counties include the eastern edge of the Rocky Mountains and the western edge of the Great Plains, and Weld County borders these counties to the east and extends into the plains. The climate of the study area is characterized by cold winters and hot summers (mean daily minimum/maximum temperatures for December–February, ≈-10/6°C; for June–August, ≈13/30°C) with low humidity. The average annual rainfall in Fort Collins in Larimer County from 1971 to 2000 was 393 mm (Mountain States Weather Services, Fort Collins, CO). The plains portion of the study area is characterized by prairie landscapes with grasses and shrubs interspersed with riparian corridors lined by cottonwood (Populus spp.) and willow (Salix spp.). These riparian areas generally are surrounded by irrigated agricultural land. Population centers typically occur along waterways that emerge from the Rocky Mountains and flow eastward into the plains. The western parts of Larimer and Boulder Counties include foothills habitats dominated by shrubs and montane habitats dominated by conifers, especially Ponderosa pine (Pinus ponderosa), and aspen (Populus tremuloides).

Fig. 1

Location of census tracts with high human population density (≥580/km²) versus lower human population density (<580/km²), and names of major population centers within Larimer, Boulder, and Weld Counties. The location of these counties in Colorado is shown in the inset figure.

Classification of Census Tracts by Human Population Density.

We selected a threshold of 580 persons/km² in 2004 to separate census tracts with high human population density (≥580/km²) from census tracts with lower human population density (<580/km²) within the Larimer-Boulder-Weld area (Fig. 1). This resulted in the clean delineation of five large population centers (Fort Collins and Loveland in Larimer County, Greeley in Weld County, and Longmont and Boulder in Boulder County) and two smaller population centers in Boulder County (Niwot and Lafayette) from surrounding less densely populated areas of a more rural nature (Fig. 1).

Epidemiological Data.

The study was based on WNV disease cases (combining WNF and WNND) reported from Larimer, Boulder, and Weld Counties to the Colorado Department of Public Health and Environment during the 2003 and 2007 outbreak years. The epidemiological database provided by Colorado Department of Public Health and Environment included, for each case, information for county, zip code, and census tract of residence, date of onset of symptoms, and whether the case was classified as WNF or WNND. No personal identifiers were included in the database. Of the 1,369 cases from 2003 and 287 cases from 2007 included in the overall epidemiological database, 1,349 cases from 2003 (>98%) and 284 cases from 2007 (>98%) could be assigned to a census tract of patient residence. WNV disease incidence was calculated for 2003 or 2007 at the census tract scale based on human population data from 2004 (Figs. 2–3). The average population size for a census tract located within the study area was 5,150 (range, of 790–19,014 with 90% of census tracts having population sizes of 2,000–10,000).

Fig. 2

Incidence of reported WNV disease per 100,000 population in 2003 for census tracts within Larimer, Boulder, and Weld Counties in Colorado. Quartiles were used to define the incidence classes. The full extent of the study area is shown in the inset map. (Online figure in color.)

Fig. 3

Incidence of reported WNV disease per 100,000 population in 2007 for census tracts within Larimer, Boulder, and Weld Counties in Colorado. Quartiles were used to define the incidence classes (all values of 0 are included in the first quartile). The full extent of the study area is shown in the inset map. (Online figure in color.)

The epidemiological database has two notable shortcomings. First, the aggregation of cases to census tract is based on the assumption that the likely WNV exposure site was located within the specific census tract where the patient's residence was located. This introduces some degree of error because of occupational or recreational exposures, where individuals are exposed outside of their census tract of residence. However, in the absence of reliable information for probable WNV exposure sites from patient case files, use of residence as the assumed exposure location is the best available solution. Second, the database includes information only for reported cases of WNV disease. Data for true infection rates for WNV are not available because many infections in the United States are inapparent or cause only mild symptoms that are unlikely to result in visits to physicians, laboratory confirmation of WNV exposure, and case reporting (Mostashari et al. 2001). This introduces some degree of error as a result of our assumption that WNV disease incidences based on reported cases are representative of WNV attack rates among the human population.

Geographic Information System (GIS)-Based Boundary and Demographic Data.

GIS-based data for census tract boundaries, 2004 census tract population, and demographic factors (based on the 2000 Census) were obtained from the United States Census Bureau provided by the Environmental Systems Research Institute (Redlands, CA). Demographic characteristics selected for inclusion in the study related to household income, occupancy by owner versus renter, and population age structure. The rationale for examining associations between household income or housing occupancy and WNV disease incidence was a speculation that more affluent households, or owner-occupied ones, are more likely to be situated in close proximity to Culex vector breeding habitat in the naturally semiarid Colorado plains landscape where waterfront acreage is prime real estate. Our inclusion of a population age factor, using a threshold of 50 yr, was based on a previous study from Colorado showing that age >50 yr is a risk factor for WNND (Bode et al. 2006).

Annual household income data were broken down into 16 income categories: <$10,000, $10,000–14,999, $15,000–19,999, $20,000–24,999, $25,000–29,999, $30,000–34,999, $35,000–39,999, $40,000-44,999, $45,000-49,999, $50,000-59,999, $60,000-74,999, $75,000–99,999, $100,000–124,999, $125,000–149,999, $150,000–199,999, and >$200,000. Percentages of households falling within each of these income classes were calculated for each census tract. After preliminary data exploration, we stratified the census tracts by an annual household income threshold of $50,000. A census tract was classified as falling below the $50,000 annual income threshold if the majority of households in that census tract had incomes <$50,000. For housing occupancy (renter- versus owner-occupied housing units), we classified census tracts as being predominantly renter occupied (>50% of housing units being renter occupied) or predominantly owner occupied (>50% of housing units being owner occupied). For age of the human population, we classified census tracts based on a threshold of 20% of the population being >50 yr of age.

GIS-Based Land Cover Data.

GIS-based data for land cover were obtained from the National Land Cover Database 2001 (MRLC 2001 Consortium; http://www.mrlc.gov/nlcd.php). The resolution of this raster data layer is 30 × 30 m. Land cover classes occurring within the study area include the following: 11, open water; 12, perennial ice/snow; 21, developed open space; 22, developed low intensity; 23, developed medium intensity; 24, developed high intensity; 31, barren land; 41, deciduous forest; 42, evergreen forest; 43, mixed forest; 52, shrub/scrub; 71, grassland/herbaceous; 81, pasture/hay; 82, cultivated crops; 90, woody wetlands; and 95, emergent herbaceous wetlands. Based on their perceived suitability as development sites for larvae of Culex mosquito WNV vectors, especially Cx. tarsalis, land cover classes were grouped as suitable development sites (open water, developed open space [which includes heavily irrigated habitats such as golf courses and parks], pasture/hay, cultivated crops, woody wetlands, and emergent herbaceous wetlands) versus less suitable development sites (perennial ice/snow, developed low intensity, developed medium intensity, developed high intensity, barren land, deciduous forest, evergreen forest, mixed forest, shrub/scrub, and grassland/herbaceous).

It should be noted that open water is a "problem class" in this respect because it includes both suitable development sites for Culex larvae along the edges of the open water that are intermittently flooded and unsuitable water surface where the water is too deep (Zou et al. 2006). For the purpose of this study, we chose to include open water in the habitat grouping classified as suitable for Culex larvae. We also created a separate grouping for irrigated agricultural lands (pasture/hay and cultivated crops), which are presumed to serve as major larval development sites for the key local WNV vector, Cx. tarsalis, within the study area. We then calculated, for each census tract, the percentages of the census tract area covered by the following: 1) suitable development sites for Culex larvae or 2) the subset of these habitats made up by irrigated agricultural lands. Finally, we calculated, for each census tract, the average distance to the nearest pixel in the 30 × 30-m raster National Land Cover Data Layer classified as either: 1) suitable development site for Culex larvae or 2) irrigated agricultural land. For a pixel falling within these classifications, the distance = 0 m. The output distance values for a census tract represents the average distance from a 30 × 30-m pixel contained within the census tract to the nearest pixel classified as suitable development habitat for Culex larvae or irrigated agricultural land.

Notes on Statistical Analyses and Map Development.

All data analyses excluded six census tracts located in the western, mountainous parts of Larimer or Boulder Counties (Fig. 1). This was done because previous studies indicated that they are located at elevations where Culex WNV vectors are very rare or absent (Eisen et al. 2008, Winters et al. 2008a, Barker et al. 2009a). WNV infections reported from these census tracts therefore were very likely related to activities taking place outside the census tract of residence at lower elevations in the Front Range area. An additional three census tracts (one in each county) were excluded because data for one or more covariates were lacking.

Odds ratios (OR) for associations between demographic or land cover factors and elevated WNV disease incidence were calculated to determine the likelihood of the incidence exceeding the median census tract incidence value for the census tract grouping examined. Comparisons between or among groups were done using Wilcoxon or Kruskal-Wallis rank sum tests. Other statistical tests are indicated in the text. Statistical analyses were conducted using the JMP 8.0.1 statistical package (SAS Institute, Cary, NC), and results were considered significant when P < 0.05. Maps were developed using ArcGIS 9.2 (Environmental Systems Research Institute).

Results

WNV Disease Incidence in Relation to Human Population Density.

Annual incidences of reported WNV disease per 100,000 population among census tracts within the Larimer-Boulder-Weld area ranged from 0 to 508 in 2003 and from 0 to 215 in 2007. WNV disease incidences appeared to generally be elevated in census tracts with lower human population density compared with those with high human population density (Figs. 1, 2, and 3). The overall incidence of reported WNV disease per 100,000 population (overall incidence: total cases/total population × 100,000) for census tracts with lower human population density was 226 in 2003 and 57 in 2007, compared with 140 in 2003 and 23 in 2007 for census tracts with high human population density.

Statistical analyses based on census tract level data confirmed that WNV disease incidence was significantly elevated in census tracts with lower, compared with high, human population density in both 2003 and 2007 (Table 1; rank sum test with χ² approximation: χ² ≥ 16.51, df = 1, P < 0.001 in both cases). This is most likely related, in part, to greater percentages of coverage in less densely populated census tracts by: 1) habitats suitable as development sites for Culex larvae (median coverage of 72% for census tracts with lower human population density versus 25% for census tracts with high human population density: χ² = 60.89, df = 1, P < 0.001), and 2) irrigated agricultural land, which is a subset of suitable Culex habitat presumed to be an important producer of the key local WNV vector Cx. tarsalis (median coverage of 56% for census tracts with lower human population density versus 0% for census tracts with high human population density: χ² = 83.89, df = 1, P < 0.001).

Table 1

Incidence of reported WNV disease in census tracts with high vs lower human population density within the Larimer-Boulder-Weld study area in Colorado in 2003 and 2007

For census tracts with lower human population density, WNV disease incidence was higher in Larimer County than in Boulder or Weld Counties in 2003 (Table 1; χ² ≥ 4.86, df = 1, P < 0.05 in both cases), whereas there were no significant differences among counties in 2007 (χ² = 0.90, df = 2, P = 0.63). There were significant differences for WNV disease incidence among five cities located within the study area (Fig. 1; Fort Collins, Loveland, Longmont, Boulder, and Greeley) in 2003 (χ² = 21.71, df = 4, P < 0.001), but not in 2007 (χ² = 6.90, df = 4, P = 0.14). Pairwise comparisons for 2003 revealed that WNV disease incidence was highest in Loveland (Table 1; χ² ≥ 5.85, df = 1, P < 0.05 for all four pairwise comparisons with other cities) and lowest in Boulder (χ² ≥ 5.14, df = 1, P < 0.05 for all four pairwise comparisons with other cities). The most notable trend for 2007 was low WNV disease incidences in Boulder and Greeley (Table 1).

WNV Disease Incidence in 2003 Versus 2007.

Statistical analyses based on census tract level data showed that incidences were significantly elevated in 2003, compared with 2007, for both census tract population density classifications (Table 1; χ² ≥ 63.10, df = 1, P < 0.001 in both cases). We also determined the strength of the correlation for census tract WNV disease incidences in 2003 and 2007. There were significant positive correlations between census tract WNV disease incidences in 2003 and 2007 for census tracts with lower human population density (Spearman's coefficient of rank correlation; ρ_s = 0.322, n = 54, P = 0.02) and high human population density (ρ_s = 0.220, n = 94, P = 0.03). As a final exercise, to produce data with potential to inform spatial implementation schemes for prevention and control efforts within the study area, we determined which census tracts had either low WNV disease incidence (<25th percentile) or high WNV disease incidence (>75th percentile) in both 2003 and 2007. This revealed that 19 census tracts with high population density and four census tracts with lower population density had incidences <25th percentile in both outbreak years, whereas 14 census tracts with lower population density and four census tracts with high population density had incidences >75th percentile in both outbreak years (Fig. 4).

Fig. 4

Spatial distributions of census tracts with low incidence of reported WNV disease in both 2003 and 2007, falling in the first quartile (<25th percentile) for incidence in both years, and of census tracts with high incidence of reported WNV disease in both 2003 and 2007, falling in the fourth quartile (>75th percentile) for incidence in both years. The full extent of the study area is shown in the inset map. (Online figure in color.)

Associations Between Land Cover Characteristics and WNV Disease Incidence.

We found that >25% coverage of a census tract by suitable habitats for Culex larvae (open water, developed open space, pasture/hay, cultivated crops, woody wetlands, and emergent herbaceous wetlands) was a statistically significant risk factor for elevated WNV disease incidence for all census tracts combined (in both 2003 and 2007; OR = 2.06 and 3.43, respectively) as well as for census tracts with lower human population density (in 2003; OR = 10.00) or high human population density (in 2007; OR = 2.92) (Table 2). Furthermore, presence (>0% coverage) of irrigated agricultural land (pasture/hay and cultivated crops) was a significant risk factor for all census tracts combined (in 2003 and 2007; OR = 2.79 and 3.18, respectively) and for census tracts with high human population density (in 2007; OR = 2.71). We also examined the association between distance to irrigated agricultural land and WNV disease incidence. This showed that an average distance from pixels within a census tract to irrigated agricultural land of <1 km was a statistically significant risk factor for elevated WNV disease incidence for all census tracts combined (in both 2003 and 2007; OR = 3.14 and 2.32, respectively) (Table 2). There also were strong trends toward significance for census tracts with high human population density both in 2003 (OR = 2.43; 95% confidence interval [CI] = 0.99-5.43) and 2007 (OR = 2.27; 95% CI = 0.95-5.46). The distributions of irrigated agricultural land and other habitats suitable for Culex larvae are shown in Fig. 5.

Table 2

Odds ratios and 95% confidence intervals for associations between selected land cover or demographic factors and elevated incidence of reported WNV disease in 2003 and 2007 at the census tract scale within the Larimer-Boulder-Weld study area in Colorado

Fig. 5

Distributions of irrigated agricultural land (pasture/hay or cultivated crops) and other habitat suitable for development of Culex mosquito larvae (open water, developed open space, woody wetlands, and emergent herbaceous wetlands) within the section of the study area that contains major population centers. (Online figure in color.)

We also determined in more detail how overall WNV disease incidence changed within the study area with the following: 1) overall percentage of suitable habitats for Culex larvae within a census tract; 2) percentage of irrigated agricultural land within a census tract; or 3) average distance from pixels within a census tract to irrigated agricultural land. Overall, WNV disease incidence was found to increase with increasing coverage by suitable habitats for Culex larvae (Fig. 6), and subsequent analyses of census tract-level data showed that the observed differences were statistically significant in both 2003 (χ² = 20.48, df = 3, P < 0.001) and 2007 (χ² = 25.18, df = 3, P < 0.001). Similarly, overall WNV disease incidence increased with increasing coverage by the subset of suitable habitats made up by irrigated agricultural land (Fig. 7; χ² = 32.33, df = 3, P < 0.001 for 2003, and χ² = 26.43, df = 3, P < 0.001 for 2007). Finally, overall WNV disease incidence was found to decrease with increasing distance to irrigated agricultural land in both years (Fig. 8; χ² = 27.26, df = 2, P < 0.001 for 2003, and χ² = 12.67, df = 2, P = 0.002 for 2007).

Fig. 6

Association between percentage area of a census tract covered by suitable habitats for Culex larvae (open water, developed open space, pasture/hay, cultivated crops, woody wetlands, and emergent herbaceous wetlands) and WNV disease for all census tracts within the three-county study area.

Fig. 7

Association between percentage area of a census tract covered by irrigated agricultural land (pasture/hay and cultivated crops) and WNV disease incidence for all census tracts within the three-county study area.

Fig. 8

Association between distance to irrigated agricultural land and WNV disease incidence for all census tracts within the three-county study area. The range for the >1,000-m class was 1,002–6,205 m.

The association between irrigated agricultural land and WNV disease incidence also was examined at the city level, including Fort Collins, Loveland, Longmont, Boulder, and Greeley. In 2003, we found strong associations between overall WNV disease incidence for a city and the following: 1) the average percentage of coverage by irrigated agricultural land within the census tracts of the city (linear regression; WNV disease incidence per 100,000 population in 2003 = 63.20 + 9.96 × average percentage of coverage of census tracts by irrigated agricultural land; F_1,3 = 23.15, r² = 0.885, P = 0.02), or 2) the average distance from the census tracts of the city to irrigated agricultural land (WNV disease incidence per 100,000 population in 2003 = 261.58 - 0.12 × average distance to irrigated agricultural land in meters; F_1,3 = 12.56, r² = 0.807, P = 0.04). However, neither of these associations was statistically significant in 2007.

Associations Between Demographic Factors and WNV Disease Incidence.

When all census tracts within the study area were included in the analysis, we found that elevated WNV disease incidence was significantly associated with the majority of households having an annual income >$50,000 (in 2003; OR = 2.71), >50% of housing units being owner occupied (in 2003, OR = 7.89, and with a strong trend in 2007: OR = 2.32; 95% CI = 0.97-5.58), and >20% of the population being older than 50 yr (in 2003; OR = 4.38) (Table 2). Conducting the same type of analysis separately for census tracts by human population density classification produced different results; the only significant risk factors were >50% of housing units in census tracts with high human population density being owner occupied (in 2003, OR = 5.92) and >20% of population in census tracts with high human population density being older than 50 yr (in 2003; OR = 4.33).

Discussion

Within the hyperendemic Larimer-Boulder-Weld area in North Central Colorado, which accounts for approximately half of the WNV disease cases reported from the state, we found that: 1) incidence of reported WNV disease consistently was elevated in census tracts with lower, compared with high, human population density, and 2) elevated incidence of reported WNV disease was associated with percentage of coverage by or distance to irrigated agricultural land, which helps to explain both the elevated incidence in census tracts with lower human population density (where irrigated agricultural land is more common than in densely populated census tracts) and incidence patterns among and within population centers. Taken together with entomological data showing high abundances of WNV-infected Cx. tarsalis in both urban and rural settings within the study area (Bolling et al. 2007, 2009; Barker et al. 2009a), this implicates Cx. tarsalis, rather than Culex pipiens, which is more of an urban mosquito in the western United States, as the primary local vector of WNV to humans. We also recorded changes from 2003 to 2007 in the spatial pattern for census tracts within the study area with high (>75th percentile) WNV disease incidence. This suggests a dynamic and evolving scenario of WNV transmission to humans that needs to be taken into account for prevention and control measures to stay current and represent the most effective use of available resources.

Our findings help to build a more complete picture of risk factors for WNV transmission in the Colorado plains landscape and also suggest some important future research directions. Most notably, to address the primary weakness of this study (the assumption that the census tract of residence also was the census tract of WNV exposure) and to increase the spatial precision of future studies that include epidemiological WNV disease data, we need to define the spatial dimensions of WNV transmission to humans in different settings, for example, rural versus urban areas. Prospective studies to determine probable WNV exposure locations for WNV disease patients, similar to the case investigations routinely conducted for plague cases in the United States (Wong et al. 2009), are needed to close this knowledge gap and set the stage for more spatially explicit analyses of environmental factors associated with elevated risk of WNV exposure.

Human Population Density, Land Cover, and WNV Disease Incidence.

In addition to our previous study from Colorado (Winters et al. 2008b), studies from other parts of the United States or from Canada have used GIS-based data for vegetation or land cover to examine associations with elevated incidence of human WNV disease (Brownstein et al. 2002; Ruiz et al. 2004, 2007; Cooke et al. 2006; Yiannakoulias et al. 2006; Ezenwa et al. 2007; Brown et al. 2008; DeGroote et al. 2008; Liu et al. 2008). Of these studies, two are of direct relevance to the discussion of our results because they also were conducted within areas (Iowa and Alberta, Canada) where Cx. tarsalis is present. Similar to the findings from Iowa and Alberta (Yiannakoulias et al. 2006, DeGroote et al. 2008), we provide clear evidence for elevated incidence of reported WNV disease in less densely populated areas than within urban population centers in North Central Colorado. This finding is also supported by entomological data; in 2007, we found that the vector index for abundance of WNV-infected Cx. tarsalis females along the South Platte River was generally higher in sites located in rural areas outside of population centers (Barker et al. 2009a). Furthermore, higher catches of Cx. tarsalis for trap locations characterized as rural or intermediate, compared with urban, were reported from Grand Junction in western Colorado during a St. Louis encephalitis outbreak in the late 1980s (Tsai et al. 1988).

Elevated incidence of reported WNV disease in less densely populated settings most likely results, in large part, from suitable habitats for Cx. tarsalis larvae, especially irrigated agricultural lands, being more common in these areas, resulting in higher vector mosquito abundance, more intensive enzootic WNV transmission, and elevated risk for human WNV exposure. The importance of irrigated agriculture for production of Cx. tarsalis is well established in California (Brookman 1950, Bohart and Washino 1978, Reisen et al. 1992, Lawler and Lanzaro 2005), and the previously mentioned studies from Iowa and Alberta reported positive associations between irrigated agriculture and occurrence of WNV disease (Yiannakoulias et al. 2006, DeGroote et al. 2008). We present a series of analyses indicating that elevated WNV disease incidence within a hyperendemic area in Colorado is positively associated with percentage of coverage by and negatively associated with distance to irrigated agricultural land.

Armed with the knowledge of a strong link between irrigated agricultural land and WNV disease incidence, it is illuminating to map the distribution of such land in relation to population centers (Figs. 1 and 5). Within the study area, some population centers are surrounded on all sides by irrigated agricultural land (e.g., Greeley in Weld County and Longmont in Boulder County), whereas others are bordered by irrigated agricultural land to the north, east, and south, but not to the west, where they border on the foothills of the Rocky Mountains (e.g., Fort Collins in Larimer County and Boulder in Boulder County). This helps to explain fine-scale spatial WNV disease incidence patterns such as those seen in Fort Collins and Boulder in 2003, with higher incidences in census tracts located in the eastern compared with the western parts of the cities, which are more distant from irrigated agriculture (Figs. 2 and 5). Research in 2003–2004 and entomological surveillance data from Colorado Mosquito Control further support this linkage; for example, WNV-infected Culex pools typically are found more commonly along the southern and eastern edges of Fort Collins compared with the central and western parts of the city (Bolling et al. 2007; Colorado Mosquito Control, Inc. 2006, 2007). The scarcity of irrigated agricultural land in the areas immediately surrounding the city of Boulder (Fig. 5) also provides a clue to the relatively low WNV disease incidence in this city during the 2003 outbreak.

From the perspective of mosquito control, it is important to understand not only which habitat types are associated with elevated risk of vector and WNV exposure, but also how vector mosquitoes emerging from such habitat types may disperse into nearby population centers. For example, we have demonstrated significant gene flow in Cx. tarsalis along the Big Thompson-South Platte River corridor, which extends from Loveland in Larimer County through irrigated agricultural land to reach Greeley in Weld County (Barker et al. 2009a). Taken together with the findings in this study, a picture is emerging in which river corridors in the study area may serve as "highways" for dispersal of Cx. tarsalis into nearby population centers from irrigated agricultural lands, where they presumably are produced in great numbers. In California, it has been demonstrated through mark-release-recapture studies that this mosquito species can disperse over long distances, >10 km, along favorable habitats such as riparian corridors (Bailey et al. 1965, Reisen et al. 1992, Reisen and Lothrop 1995). In Larimer County, we found that Cx. tarsalis also dispersed effectively for several hundred meters away from larval development sites into prairie landscapes with scattered trees (Barker et al. 2009b). Future studies are needed to determine whether Cx. tarsalis disperses from irrigated agricultural land into nearby population centers in the plains landscape of eastern Colorado along distinct landscape corridors, which could be exploited in highly targeted mosquito control efforts, or whether they disperse in a more diffuse manner. A related research need is to determine the importance of different types of irrigation, for example, flood versus pivot irrigation, for production of Cx. tarsalis on irrigated agricultural lands in the Great Plains.

The observed pattern of elevated WNV disease incidence in less densely populated census tracts may to some extent also be influenced by the mosquito control activities carried out by Colorado Mosquito Control under contract from individual cities and typically including the city itself and a buffer area surrounding the city limits. Similar activities are minimal in distinctly rural census tracts within the study area. Recent studies from California provide strong evidence that intensive and timely mosquito control during WNV disease outbreaks can suppress Culex vector populations, decrease the intensity of WNV transmission, and reduce human disease cases (Carney et al. 2008, Elnaiem et al. 2008, Lothrop et al. 2008).

Demographic Factors and Spatial Changes in WNV Disease Incidence Patterns Over Time.

The most extensive previous studies on spatial associations between demographic factors and WNV disease come from the Chicago area, which was severely afflicted by WNV in 2002. Those studies revealed stratification in WNV disease incidence for different types of urban settings within Chicago (Ruiz et al. 2007) and positive associations between WNV disease incidence and an older age population, housing built during the 1950s, and higher median household income (Ruiz et al. 2004). The findings from Chicago, where Cx. pipiens is the primary WNV vector, influenced our selection of demographic factors to examine, but cannot be assumed to be directly applicable to Colorado, where Cx. tarsalis appears to be the primary vector of WNV to humans in most settings.

We found that selected demographic factors were associated with elevated WNV disease incidence in North Central Colorado, but also that these associations need to be interpreted with great care because they were not consistent across settings with different human population densities or, in some cases, for different outbreak years. For example, the positive association between majority of households with income >$50,000 and elevated WNV disease incidence observed for the entire study area in 2003 did not hold true either for census tracts examined by human population density classification or for the entire study area in 2007 (Table 2). Finding a positive association between >20% of population being older than 50 yr and elevated WNV disease incidence was not surprising because of the following: 1) WNV attack rates can be higher in older age groups (Mostashari et al. 2001); 2) older persons are more likely to suffer severe manifestations of WNV infection and thus are more likely to seek medical care and be recognized as being infected with the virus (Hayes et al. 2005, Huhn et al. 2005, Bode et al. 2006, Reimann et al. 2008); and 3) older age of the population was a risk factor for occurrence of WNV disease in Chicago (Ruiz et al. 2004). However, when census tracts were examined by human population density classification, we found this association to be statistically significant only for 2003 in census tracts with high human population density. The lack of a similar association in 2007 may have several causes, including age-related changes in care-seeking behavior with WNV disease now being a more familiar and perhaps less threatening disease or, speculatively, the older age group engaging more frequently during 2003–2006 in nighttime outdoor activities (placing them at risk for Culex vector bites), and thus having a higher prevalence of protective antibodies by 2007 compared with younger age groups. Renewed studies to determine behavioral risks for WNV exposure in an endemic situation, such as conducted in Larimer County in 2003 during the initial outbreak (Gujral et al. 2007), and to assess changes over time for age-specific prevalence of WNV antibodies among the population are needed to shed light on the dynamics underlying potential changes in WNV disease incidence for different segments of the human population over time. We also need to gain a better understanding of potential biases in health care-seeking behavior in the study area, which for example could be related to age, financial status, or distance to health facility, which tends to be longer in rural settings.

With regard to the positive association between elevated WNV disease incidence and >50% occupancy by owners (rather than renters), we speculate that this may be related to the following: 1) desirable property locations, which are less likely to be rented, being located close to habitats producing Cx. tarsalis, such as small reservoirs or lakes, ponds, and riparian areas, and 2) fewer rental apartment buildings in rural areas.

Our WNV disease incidence mapping exercises (Figs. 2–4) and the results from analyses for 2003 versus 2007 indicate a spatially dynamic and evolving scenario of WNV transmission to humans within the study area from the initial 2003 outbreak to the subsequent 2007 outbreak year. Although the analyses revealed that some census tracts consistently had low or high WNV disease incidence, the correlation of census tract incidences in 2003 and 2007 was relatively weak (correlation coefficients of 0.322 for census tracts with lower human population density, and 0.220 for census tracts with high human population density). Similarly changing spatial patterns for WNV disease incidence over time have been reported from Indiana and the northeastern United States (Brown et al. 2008, Liu et al. 2008).

The changes observed for spatial patterns of WNV disease incidence from 2003 to 2007 in our study area could result from multiple factors, including shifts in spatial patterns for the following: 1) abundance of vector mosquitoes, 2) abundance of WNV-susceptible birds, or 3) levels of susceptibility to WNV in the human population. Based on the intensive transmission of WNV in the study area, it is not difficult to envision a scenario in which the prevalence of protective antibodies becomes spatially stratified among the human population over time, and higher in areas with elevated WNV attack rates. To ultimately explain these changing patterns, we need to conduct prospective studies that account for both entomological risk of exposure to WNV-infected vectors and spatial stratification in susceptibility among the human population to WNV infection.

Acknowledgements

We thank Rebecca J. Eisen of the Centers for Disease Control and Prevention for helpful discussions. This work was supported, in part, by Contract N01-(AI)-25489 from the National Institutes of Allergy and Infectious Diseases.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Allergy and Infectious Diseases or the National Institutes of Health.