Only data from Cx. quinquefasciatus
collected from gravid traps (38,120 of the total 71,824 mosquitoes collected) were used in this study [see Supplemental Material (doi:10.1289/ehp.1001939)]. Yearly averages of Cx. quinquefasciatus
ML WNV infection rates varied between 2.6 and 16.3 per 1,000 mosquitoes, with no clear temporal pattern (). Data were available for 321 WNV-positive corvids, which accounted for 98.2% of the WNV-infected dead birds collected in 2001–2006 [see Supplemental Material, Table 1 (doi:10.1289/ehp.1001939)]. The declining temporal trend in WNV-positive corvid death ratios (from 28.0 per 100,000 persons in 2002 to 0.00 per 100,000 in 2006; ) is most likely explained by the decreasing number of dead birds submitted by Atlanta residents to the FCDHW. Of the 50 confirmed cases among Fulton Country residents in 2001–2007, 26 (52%) had symptoms of neuroinvasive WNV. The 2001–2007 average WNV incidence rate among Fulton County residents (0.8 cases per 100,000) was two times higher than the 1999–2008 national average (0.4 cases per 100,000) (Lindsey et al. 2010
). With the exception of the years 2001 and 2006, human incidence rates showed little temporal variation, ranging between 0.93 and 1.0 cases per 100,000 ().
WNV infection in Cx. quinquefasciatus mosquitoes, corvid carcasses, and humans, Fulton County, Georgia, USA, 2001–2007.
was heterogeneously distributed, with the highest abundances (number per trap-night) and densities (numbers per hectare) concentrated within the boundaries of the city of Atlanta (). Traps with elevated abundance of Cx. quinquefasciatus
mosquitoes clustered in three locations within an average radius of 1,305 m (range, 200–2,000) from each trap () [Gi
) > 3.7; p
< 0.05]. The three high mosquito abundance clusters encompassed segments of five of the seven CSO streams in Atlanta (). High WNV infection intensity in Cx. quinquefasciatus
(WNV-positive mosquitoes/1,000) also clustered in three locations within Atlanta (, Gi
) > 3.7; p
< 0.05). The largest cluster [average radius = 1,042 m (range, 400–2,000)] surrounded part of the North Avenue CSO creek and encompassed 41 trapping locations (). The remaining two clusters encompassed only two trap locations each, with clustering radii of 320 and 93 m, respectively (). Clusters of WNV infection presence in Cx. quinquefasciatus
(presence/absence of WNV infection at a trapping location) also occurred within the city of Atlanta, in proximity [average = 988 m (range, 10–3,118)] to CSO streams () (Kulldorff Bernoulli test, p
< 0.05). K-function analysis (Ripley 1976
) confirmed that the observed clustering patterns were not the result of heterogeneous trap distribution, because traps were randomly distributed within the city of Atlanta [Supplemental Material, Figure 2B (doi:10.1289/ehp.1001939)].
Figure 1 (A) Abundance (mosquitoes/trap-night), density distribution (mosquitoes/ha), and local spatial clustering of Cx. quinquefasciatus abundance, 2001–2007. (B) Distribution of WNV infections in Cx. quinquefasciatus and location of clusters of high (more ...)
The proportion of WNV-infected Cx. quinquefasciatus pools (mosquitoes grouped by date and location) was significantly higher within 1 km of a CSO stream (125/1,215) than within 1 km of a non-CSO stream (77/1,044) (p = 0.027). Furthermore, 9 of 12 (75%) trapping locations where WNV-infected Cx. quinquefasciatus mosquitoes were detected during early spring were located within 1 km of a CSO stream. The pattern of ML WNV infection clustering in Cx. quinquefasciatus was consistent across years, with most positive pools located in close proximity to CSO streams [see Supplemental Material, Figure 3 (doi:10.1289/ehp.1001939)].
A regression model that included the average distance between each trapping location to the nearest CSO stream and the range of tree canopy cover within 1 km of each trap location was best supported by the data (ωi, 65.8%) (). Distance to CSO streams was the best predictor of the abundance of Cx. quinquefasciatus mosquitoes (∑ωi = 100%, by being statistically significant in the best models), followed by tree canopy coverage range (∑ωi = 66.9%) (). The remaining environmental factors were not statistically significant. Traps with a high density of Cx. quinquefasciatus (> 20/trap-night) were located at an average distance of 1,156 m (95% CI, 552–1,759 m) from a CSO stream and had an average tree canopy range of 98% (95% CI, 95–99%). Tree canopy range can be interpreted as the degree of canopy variability around a trap location, as high values include highly forested and deforested areas close to each other (within 1 km).
Table 2 Summary of multiple logistic regression models used to evaluate the average abundance of Cx. quinquefasciatus (mosquitoes/trap-night) and WNV infection presence (presence/absence of WNV-positive pools in a trapping location), urban Atlanta, Georgia, USA, (more ...)
The presence of WNV infection in Cx. quinquefasciatus was also predicted by the average distance to CSO streams and by the canopy cover range (). The two best candidate models included distance to CSO streams as the sole significant predictor of Cx. quinquefasciatus WNV infection presence. Distance to CSO streams (∑ωi, 80.2%) followed by canopy cover range (∑ωi, 15.2%) were the most important predictors in these models.
Human WNV incidence rates (EB) and WNV-positive corvid death ratios also clustered within the city of Atlanta (). One of the two clusters of high human WNV incidence (identified by the Local Moran’s I LISA test) was located near three of the seven CSO streams (). This cluster included six census tracts and six reported cases and had an average annual incidence (± SD) of 48.6 ± 37.4 cases per 100,000 persons. The second (northern) cluster of human WNV incidence was not directly associated with any CSO stream (). This cluster included 10 tracts and 10 cases and had an average annual incidence of 42.6 ± 31.3 cases per 100,000 persons. In contrast, Atlanta census tracts outside the clustering areas had an average annual incidence of 7.5 ± 11.0 human WNV cases per 100,000 persons. Clustering zones were similar but encompassed fewer tracts when we repeated the analysis using unsmoothed (versus EB) human WNV incidence rates [see Supplemental Material, Figure 4 (doi:10.1289/ehp.1001939)]. Spatial clustering of WNV-positive corvid death ratios overlapped with human WNV incidence clusters and Cx. quinquefasciatus infection clusters in the east and north of Atlanta (). Four of the seven CSO streams were partly encompassed by WNV-positive dead corvid ratio clusters ().
Figure 2 Distribution and spatial clustering of (A) EB-smoothed WNV human incidence rate estimates (cases/100,000 persons) and (B) WNV-positive corvid death ratios (number of dead corvids/100,000 persons) in Fulton County. Inset shows a detailed view of the city (more ...)
The human WNV incidence rate per census tract (2001–2007, log10 + 1 transformed EB estimates) was best predicted by a model that included the average distance from each tract to the nearest CSO stream, mean tree canopy cover, the proportion of houses built in the 1950s–1960s, median household income, and number of WNV-positive dead corvids identified during 2001–2004 () (R2 = 0.21; p < 0.001; ωi = 81.7%). When all variables with p < 0.07 were considered, distance to CSO streams was the best predictor of human WNV incidence (∑ωi = 100%), followed by the proportion of houses built in the 1950s–1960s (∑ωi = 94.0%) and the median household income (∑ωi = 92.0%) (). The remaining factors were poorly supported by the data. Similarly, log10 + 1 transformed WNV-positive corvid death ratios were best predicted by a model including the average distance to the nearest CSO stream, mean tree canopy cover, proportion of houses built in the 1950s–1960s, and median household income () (R2 = 0.31; p < 0.001; ωi = 91.3%). Again, mean distance to CSO streams and mean tree cover (both ∑ωi = 100%), followed by median household income (∑ωi = 91.0%), were the best predictors of high WNV-positive corvid death ratios ().
Summary of linear regression models evaluated for human WNV incidence rates and WNV-positive corvid death ratios per census tract in Atlanta, 2001–2007.a