• View in gallery

    Semivariogram of the occurrence of Ehrlichia chaffeensis. Lag = 50 km; nugget variance c0 = 0.13014; sill c = 0.21846; range a = 463.3 km.

  • View in gallery

    Map of endemnicity probabilities for Ehrlichia chaffeensis based on kriging. Solid circles represent positive E. chaffeensis serologic data and open circles represent negative serologic data obtained from a previous white-tailed deer surveillance study.9 This figure appears in color at www.ajtmh.org.

  • View in gallery

    Map of endemnicity probabilities for Ehrlichia chaffeensis based on a region-wide logistic regression model derived for the entire 19-state region. Solid circles represent positive E. chaffeensis serologic data and open circles represent negative serologic data obtained from a previous white-tailed deer surveillance study.9 This figure appears in color at www.ajtmh.org.

  • View in gallery

    Maps of endemnicity probabilities for Ehrlichia chaffeensis based on separate logistic regression models for the eastern, central, and western subregions. Projections for Florida are from the original region-wide model. Solid circles represent positive E. chaffeensis serologic data and open circles represent negative serologic data obtained from a previous white-tailed deer surveillance study.9 This figure appears in color at www.ajtmh.org.

  • View in gallery

    County-level average annual incidence rates for human monocytotrophic ehrlichiosis (HME) reported to the Centers for Disease Control and Prevention through the National Electronic Telecommunications System for Surveillance. Data were obtained from references 12 and 13.

  • View in gallery

    Comparison of human monocytotrophic ehrlichiosis (HME) case data with the predicted presence of Ehrlichia chaffeensis based on kriging and logistic regression analyses.

  • View in gallery

    Geographic information system overlay maps for selected key explanatory variables used in development of logistic regression models. A, White-tailed deer density; B, Elevation; C, Percentage of all crops; D, Annual minimum temperature. This figure appears in color at www.ajtmh.org.

  • 1

    Dawson JE, Stallknecht DE, Howerth EW, Warner C, Biggie K, Davidson WR, Lockhart JM, Nettles VF, Olson JG, Childs JE, 1994. Susceptibility of white-tailed deer (Odocoileus virginianus) to infection with Ehrlichia chaffeensis, the etiologic agent of human ehrlichiosis. J Clin Microbiol 32 :2725–2728.

    • Search Google Scholar
    • Export Citation
  • 2

    Dawson JE, Childs JE, Biggie KL, Moore C, Stallknecht D, Shaddock J, Bouseman J, Hofmeister E, Olson JG, 1994. White-tailed deer as a potential reservoir of Ehrlichia spp. J Wildl Dis 30 :162–168.

    • Search Google Scholar
    • Export Citation
  • 3

    Ewing SA, Dawson JE, Kocan AA, Barker RW, Warner CK, Panciera RJ, Fox JC, Kocan KM, Blouin EF, 1995. Experimental transmission of Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) among white-tailed deer by Amblyomma americanum (Acari: Ixodidae). J Med Entomol 32 :368–374.

    • Search Google Scholar
    • Export Citation
  • 4

    Lockhart JM, Davidson WR, Stallknecht DE, Dawson DE, Howerth EW, 1997. Isolation of Ehrlichia chaffeensis from wild white-tailed deer (Odocoileus virginianus) confirms their role as natural reservoir hosts. J Clin Microbiol 35 :1681–1686.

    • Search Google Scholar
    • Export Citation
  • 5

    Lockhart JM, Davidson WR, Stallknecht DE, Dawson JE, Little SE, 1997. Natural history of Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) in the Piedmont physiographic province of Georgia. J Parasitol 83 :887–894.

    • Search Google Scholar
    • Export Citation
  • 6

    Childs JE, Paddock CD, 2002. The ascendancy of Amblyomma americanum as a vector of pathogens affecting humans in the United States. Annu Rev Entomol 48 :307–337.

    • Search Google Scholar
    • Export Citation
  • 7

    Paddock CD, Childs JE, 2003. Ehrlichia chaffeensis: a prototypical emerging pathogen. Clin Microbiol Rev 16 :37–64.

  • 8

    Yabsley MJ, Varela AS, Tate CM, Dugan VG, Stallknecht DE, Little SE, Davidson WR, 2002. Ehrlichia ewingii infection in white-tailed deer (Odocoileus virginianus). Emerg Infect Dis 8 :668–671.

    • Search Google Scholar
    • Export Citation
  • 9

    Yabsley MJ, Dugan VG, Stallknecht DE, Little SE, Lockhart JM, Dawson JE, Davidson WR, 2003. Evaluation of a prototype Ehrlichia chaffeensis surveillance system using white-tailed deer (Odocoileus virginianus) as natural sentinels. Vector Borne Zoonotic Dis 3 :195–207.

    • Search Google Scholar
    • Export Citation
  • 10

    Irving RP, Pinger RR, Vann CN, Olesen JB, Steiner FE, 2000. Distribution of Ehrlichia chaffeensis (Rickettsiales: Rickettsiaeceae) in Amblyomma americanum in southern Indiana and prevalence of E. chaffeensis–reactive antibodies in white-tailed deer in Indiana and Ohio in 1998. J Med Entomol 37 :595–600.

    • Search Google Scholar
    • Export Citation
  • 11

    Mueller-Anneling L, Gilchrist MJ, Thorne PS, 2000. Ehrlichia chaffeensis antibodies in white-tailed deer, Iowa, 1994 and 1996. Emerg Infect Dis 6 :397–400.

    • Search Google Scholar
    • Export Citation
  • 12

    McQuiston JH, Paddock CD, Holman RC, Childs JE, 1999. Human ehrlichioses in the United States. Emerg Infect Dis 5 :635–642.

  • 13

    Gardner SL, Holman RC, Krebs JW, Berkelman R, Childs JE, 2003. National surveillance for the human ehrlichioses in the United States, 1997–2001, and proposed methods for evaluation of data quality. Ann N Y Acad Sci 990 :80–89.

    • Search Google Scholar
    • Export Citation
  • 14

    Isaaks EH, Srivastava RM, 1989. An Introduction to Applied Geostatistics. New York: Oxford University Press.

  • 15

    Rossi RE, Mulla DJ, Journel AG, Franz EH, 1992. Geostatistical tools for modeling and interpreting ecological spatial dependence. Ecol Monogr 62 :277–314.

    • Search Google Scholar
    • Export Citation
  • 16

    Goovaerts P, 1997. Geostatistics for Natural Resources Evaluation. New York: Oxford University Press.

  • 17

    Kleinschmidt I, Bagayoko M, Clarke GPY, Craig M, Le Sueur D, 2000. A spatial statistical approach to malaria mapping. Int J Epidemiol 29 :355–361.

    • Search Google Scholar
    • Export Citation
  • 18

    Brownstein JS, Holford TR, Fish D, 2003. A climate-based model predicts the spatial distribution of the Lyme disease vector Ixodes scapularis in the United States. Environ Health Perspect 111 :1152–1157.

    • Search Google Scholar
    • Export Citation
  • 19

    Guerra M, Walker E, Jones C, Paskewitz S, Cortinas MR, Stancil A, Beck L, Bobo M, Kitron U, 2002. Predicting the risk of Lyme disease: Habitat suitability for Ixodes scapularis in the north central United States. Emerg Infect Dis 8 :289–297.

    • Search Google Scholar
    • Export Citation
  • 20

    Lockhart JM, Davidson WR, Stallknecht DE, Dawson JE, 1996. Site-specific geographic association between Amblyomma americanum (Acari: Ixodidae) infestations and Ehrlichia chaffeensis-reactive (Rickettsiales: Ehrlichieae) antibodies in white-tailed deer. J Med Entomol 33 :153–158.

    • Search Google Scholar
    • Export Citation
  • 21

    Little SE, Stallknecht DE, Lockhart JM, Dawson JE, Davidson WR, 1998. Natural coinfection of a white-tailed deer (Odocoileus virginianus) population with three Ehrlichia spp. J Parasitol 84 :897–901.

    • Search Google Scholar
    • Export Citation
  • 22

    Thornton PE, Running SW, White MA, 1997. Generating surfaces of daily meteorological variables over large regions of complex terrain. J Hydrol 190 :214–251.

    • Search Google Scholar
    • Export Citation
  • 23

    Thornton PE, Hasenauer H, White MA, 2000. Simultaneous estimation of daily solar radiation and humidity from observed temperature and precipitation: an application over complex terrain in Austria. Agric Forest Meterorol 104 :255–271.

    • Search Google Scholar
    • Export Citation
  • 24

    Loveland TR, Reed BC, Brown JF, Ohlen DO, Zhu J, Yang L, Merchant JW, 2000. Development of a Global Land Cover Characteristics Database and IGBP DISCover from 1-km AVHRR Data. Int J Remote Sens 21 :1303–1330.

    • Search Google Scholar
    • Export Citation
  • 25

    Fielding AH, Bell JF, 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv 24 :38–49.

    • Search Google Scholar
    • Export Citation
  • 26

    Nagelkerke NJD, 1991. A note on a general definition of the coefficient of determination. Biometrika 78 :691–692.

  • 27

    Matthews SA, 1990. Epidemiology using a GIS: the need for caution. Comput Environ Urban Sys 14 :213–221.

  • 28

    Haile DG, Mount GA, 1987. Computer simulation of population dynamics of the lone star tick, Amblyomma americanum (Acari: Ixodidae). J Med Entomol 24 :356–369.

    • Search Google Scholar
    • Export Citation
  • 29

    Mount GA, Haile DG, Barnard DR, Daniels E, 1993. New version of LSTSIM for computer simulation of Amblyomma americanum (Acari:Ixodidae) population dynamics. J Med Entomol 30 :843–857.

    • Search Google Scholar
    • Export Citation
  • 30

    Augustin NH, Mugglestone MA, Buckland ST, 1996. An autologistic model for the spatial distribution of wildlife. J Appl Ecol 33 :339–347.

    • Search Google Scholar
    • Export Citation
  • 31

    Fotheringham AS, Brunsdon C, Charlton ME, 2002. Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. Chichester, United Kingdom: John Wiley.

  • 32

    Hair JA, Howell DE, 1970. Lone Star Ticks - Their Biology and Control in Ozark Recreation Areas. Stillwater, OK: Oklahoma State University Agriculture Experimental Station Bulletin. B-679.

  • 33

    Lancaster JL Jr, 1957. Control of the Lone Star Ticks. Fayetteville, AR. Agricultural Experimental Station. College of Agriculture and Horticultural Ecology, University of Arkansas, Representative Series 67.

  • 34

    Soneshine DE, Atwood EL, Lamb JT, 1966. The ecology of ticks transmitting Rocky Mountain spotted fever in a study area in Virginia. Ann Entomol Soc Am 59 :1234–1262.

    • Search Google Scholar
    • Export Citation
  • 35

    Hooker WA, Bishopp FC, Wood HP, 1912. Some North American Ticks. Washington, DC: United States Bureau of Entomology Bulletin No. 106.

  • 36

    Robertson AS, Patrick CD, Semtner PJ, Hair JA, 1975. The ecology and behavior of the lone star tick (Acarine:Ixodidae): VI. Response of unfed adults to certain environmental parameters. J Med Entomol 12 :525–529.

    • Search Google Scholar
    • Export Citation
  • 37

    Koch HG, Dunn JC, 1980. Oviposition, egg hatch, and larval survival of the lone star ticks (Amblyomma americanum) held at different temperatures and humidities. Southwestern Entomol 5 :169–174.

    • Search Google Scholar
    • Export Citation
  • 38

    Bunnell JE, Price SD, Das A, Shields TM, Glass GE, 2003. Geographic information systems and spatial analysis of adult Ixodes scapularis (Acari:Ixodidae) in the Middle Atlantic Region of the U.S.A. J Med Entomol 40 :570–576.

    • Search Google Scholar
    • Export Citation
  • 39

    Lockhart JM, Davidson WR, Dawson JE, Stallknecht DE, 1995. Temporal association of Amblyomma americanum with the presence of Ehrlichia chaffeensis reactive antibodies in white-tailed deer. J Wildl Dis 31 :119–124.

    • Search Google Scholar
    • Export Citation
  • 40

    Anderson BE, Sims KG, Olson JG, Childs JE, Piesman JF, Happ CM, Maupin GO, Johnson BJ, 1993. Amblyomma americanum: a potential vector of human ehrlichiosis. Am J Trop Med Hyg 49 :239–244.

    • Search Google Scholar
    • Export Citation
  • 41

    Burket CT, Vann CN, Pinger RR, Chatot CL, Steiner FE, 1998. Minimum infection rate of Ambylomma americanum (Acari: Ixodidae) by Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) in southern Indiana. J Med Entomol 35 :653–659.

    • Search Google Scholar
    • Export Citation
  • 42

    Ijdo JW, Wu C, Magnarelli LA, Stafford KC 3rd, Anderson JF, Fikrig E, 2000. Detection of Ehrlichia chaffeensis DNA in Amblyomma americanum ticks in Connecticut and Rhode Island. J Clin Microbiol 38 :4655–4656.

    • Search Google Scholar
    • Export Citation
  • 43

    Roland WE, Everett ED, Cyr TL, Hasan SZ, Dommaraju CB, McDonald GA, 1998. Ehrlichia chaffeensis in Missouri ticks. Am J Trop Med Hyg 59 :641–643.

    • Search Google Scholar
    • Export Citation
  • 44

    Steiert JB, Gilfoy F, 2002. Infection rates of Amblyomma americanum and Dermacentor variabilis by Ehrlichia chaffeensis and Ehrlichia ewingii in Southwest Missouri. Vector Borne Zoonotic Dis 2 :53–60.

    • Search Google Scholar
    • Export Citation
  • 45

    Steiner FE, Pinger RR, Vann CN, 1999. Infection rates of Amblyomma americanum (Acari: Ixodidae) by Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) and prevalence of E. chaffeensis-reactive antibodies in white-tailed deer in southern Indiana, 1997. J Med Entomol 36 :715–719.

    • Search Google Scholar
    • Export Citation
  • 46

    Stromdahl EY, Evans SR, O’Brien JJ, Gutierrez AG, 2001. Prevalence of infection in ticks submitted to the Human Tick Test Kit Program of the U.S. Army Center for Health Promotion and Preventive Medicine. J Med Entomol 38 :67–74.

    • Search Google Scholar
    • Export Citation
  • 47

    Whitlock JE, Fang QQ, Durden LA, Oliver JH Jr, 2000. Prevalence of Ehrlichia chaffeensis (Rickettsiales: Rickettsiaceae) in Amblyomma americanum (Acari: Ixodidae) from the Georgia coast and Barrier Islands. J Med Entomol 37 :276–280.

    • Search Google Scholar
    • Export Citation

 

 

 

 

SPATIAL ANALYSIS OF THE DISTRIBUTION OF EHRLICHIA CHAFFEENSIS, CAUSATIVE AGENT OF HUMAN MONOCYTOTROPIC EHRLICHIOSIS, ACROSS A MULTI-STATE REGION

View More View Less
  • 1 Southeastern Cooperative Wildlife Disease Study, D. B. Warnell School of Forest Resources, and Department of Infectious Diseases, College of Veterinary Medicine, University of Georgia, Athens, Georgia

Ehrlichia chaffeensis, which causes human monocytotrophic ehrlichiosis (HME), is an important emerging tick-borne pathogen in the southeastern and southcentral United States. The endemnicity probability of E. chaffeensis and, by implication, locations with risk for HME, was predicted by using two modeling methods. This is first large-scale study to use geospatial analyses to estimate the distribution of E. chaffeensis, and it was conducted using data from a prototypic surveillance system that used white-tailed deer as natural sentinels. Analyses included the E. chaffeensis serostatus for 563 counties from 18 states. Both kriging and logistic regression models provided very reliable portrayals of E. chaffeensis occurrence and predicted that E. chaffeensis distribution had good concordance with human case data. The integration of a deer surveillance system with geospatial analyses was useful in developing HME risk maps that will be useful for identifying high-risk areas for public health interventions such as prevention and control efforts.

INTRODUCTION

Ehrlichia chaffeensis, which causes human monocytotrophic ehrlichiosis (HME), is an important emerging tick-borne pathogen in the southeastern and southcentral United States. It is maintained in a zoonotic cycle involving white-tailed deer (Odocoileus virginianus) as principal vertebrate reservoirs and lone star ticks (Amblyomma americanum) as biologic vectors.15 The lone star tick ranges from mid-Texas north to Iowa and east to the Atlantic coast and recently has been detected as far north as Maine.6 Evidence of E. chaffeensis has been detected in white-tailed deer and/or ticks throughout the range of the lone star tick; however, the distribution of E. chaffeensis in these states is not continuous.7 Some known distributional limits of E. chaffeensis in white-tailed deer include a western boundary extending across Kansas, Oklahoma, and Texas, a southern boundary across peninsular Florida, and a cluster of many seronegative populations centered along the Appalachian Mountains.2,8,9 Other studies of white-tailed deer have documented an apparent northern range limit for E. chaffeensis in southern portions of Iowa, Illinois, Indiana, and Ohio.1,10,11

Currently the only estimate of HME risk is based on human case data reported to the Centers for Disease Control and Prevention through the National Electronic Telecommunications System for Surveillance.12,13 Although the Centers for Disease Control and Prevention and the Council of State and Territorial Epidemiologists recommended that human ehrlichioses be nationally reportable in 1998, currently 11 states do not report ehrlichiosis cases.13 The incidence of HME in three of these states (Louisiana, Maryland, and Mississippi) is suspected to be high based on numbers of cases in neighboring states. Even for states that consider HME reportable, inconsistencies in diagnosis and reporting of HME limits the usefulness of these data for risk mapping.13

Recently, an extensive relatively fine-scale (county) serologic surveillance system using white-tailed deer as natural sentinels was evaluated as a means of discerning the distribution of E. chaffeensis.9 This alternative surveillance system included 563 populations of white-tailed deer throughout 18 states in the southcentral and southeastern United States. The goals of the present study were to evaluate alternative methodologies for interpolating between the sampled counties to generate continuous maps of predicted serologic status, and to determine whether data from these maps reflected HME risk by examining its spatial correlation with the distribution of reported HME cases. To accomplish this goal, we conducted two analyses 1) geostatistical analyses (kriging) to estimate the spatial pattern of E. chaffeensis and 2) logistic regression modeling to identify ecologic predictor variables that could be useful in estimating distributional limits of locales suitable for E. chaffeensis transmission. The distributions of E. chaffeensis projected by kriging and logistic regression were then compared with Centers for Disease Control and Prevention HME case data extracted from McQuiston and others12 and Gardner and others.13

Because distribution data of diseases or disease agents are rarely available for all possible locations, geographic information systems (GISs) and geospatial analyses are being used with increasing frequency to estimate geographic distributions. Geostatistics provides a set of methods for explicitly incorporating the spatial pattern data into analysis (geospatial analysis). Spatial autocorrelation among the observations are characterized across multiple scales using semivariograms, and these spatial dependencies can be used to interpolate between the sampled locations by kriging.14 The predictions obtained using this technique are based on the assumptions that spatial variability in the data is related to the distance between observations, and that values in unsampled areas can be predicted as a weighted average of observations at nearby locations. Geostatistics have been widely used in the fields of ecology, physical geography, and natural resource management, but have only recently been applied to the problem of disease mapping.1517

An alternative method of spatial interpolation is based on ecologic relationships between the variable of interest (the presence of E. chaffeensis) and one or more independent variables.18,19 The variable of interest is typically available at only a limited number of spatial locations, whereas the independent variables are mapped across the entire landscape. Statistical methods such as logistic regression are used to develop predictive equations, and these equations are then applied to the unsampled areas to generate a predictive map. This approach differs from kriging in that predictions of E. chaffeensis serostatus are based on relationships with ecologic variables sampled at the same location, whereas kriging predictions are based on spatial patterns of E. chaffeensis serostatus across multiple locations.

MATERIALS AND METHODS

Ehrlichia chaffeensis database.

The white-tailed deer E. chaffeensis serologic database used9 was derived from samples collected by Southeastern Cooperative Wildlife Disease Study researchers.1,8,20,21 The database comprised a total of 563 white-tailed deer populations tested for antibodies reactive to E. chaffeensis by the indirect immunofluorescent antibody test as previously described.1,9,20 Serologic data for each population were categorized by county or parish and linked to a map of county boundaries in a GIS. If one or more deer with antibodies reactive to E. chaffeensis was detected, that county or parish was classified as positive.

Geostatistical modeling (kriging).

Because county serostatus was coded as an indicator variable (present = 1 and absent = 0), the spatial distribution of E. chaffeensis was interpolated using indicator kriging.14 The predictions obtained by this technique are based on the assumptions that 1) spatial variability in E. chaffeensis serostatus is related to the distance between counties and 2) the presence of E. chaffeensis in an unsampled county can be predicted based on the status of nearby sample locations. An empirical semivariogram was calculated over twelve 50-km lag intervals using data for the sampled counties (n = 563), with the spatial location of each county represented by its centroid. A spherical model was fitted to this semivariogram and used to generate a kriged map of the probability of E. chaffeensis presence or absence.

Kriging accuracy was assessed using subsets of data. The 563 data points were randomly divided into a training set (75%, 422 data points) and a test set (25%, 141 data points) using ArcGIS 8.3 (Environmental Systems Research Institute, Redlands, CA). The training set was used to develop the kriging model that was then used to predict the values of the 141 test data points. The subset analysis was independently conducted 10 times. Counties were classified as positive for E. chaffeensis if the predicted probability was > 0.5, and negative otherwise. Accuracy (percent of counties correctly classified), sensitivity (percent of positive counties correctly classified), and specificity (percent of negative counties correctly classified) were computed. Semivariogram analysis, kriging, and validation were conducted using ArcGIS 8.3.

Logistic regression modeling.

Explanatory variables.

Spatial predictor variables for the logistic regression model were obtained as GIS datasets from a variety of sources and a full description of variables is given in Table 1. Climate variables included average monthly and annual minimum temperature, maximum temperature, number of frost days, precipitation, and water vapor pressure summarized for 1980–1997. These data were obtained as 1-km grids generated using the DayMet model.22 Seasonal data were summed from average monthly data (winter, December–February; spring, March–May; summer, June–August; fall, September–November). Relative humidity was calculated using water vapor pressure and minimum and maximum temperatures using a method previously described.23 Land cover data were obtained from the Global Land Cover Characteristics Database and were derived from 1-km Advanced Very High Resolution Radiometer data collected from April 1992 to March 1993.24 Elevation data was obtained as a 1-km digital elevation model. A soil drainage index was computed for each county using the State Soil Geographic (STATSGO) database (1:250,000). Low values characterized frequently saturated soils, whereas high values characterized well-drained soils. For each mapped soil polygon, a composite index for each polygon was computed based on the area-weighted average of drainage indices of the associated soil series. Deer density data from 1999 were obtained as a paper map from the Quality Deer Management Association (Watkinsville, GA). These data were mapped as polygons with five density levels: 1) deer absent, rare, or urban with unknown population; 2) < 15 deer/km2; 3) 15–30 deer/km2; 4) 31–45 deer/km2; and 5) > 45 deer/km2. The states included in this study were digitized, georeferenced, and converted to a 1-km grid.

These data sets were overlaid on the county boundaries to compute summary values for each county. Climate variables and soil drainage were summarized as mean values for each county. Land cover was summarized as the percentage of each county occupied by various land cover classes. Elevation was summarized as mean value for each county, and a ruggedness index was computed by subtracting the lowest elevation from the highest elevation within each county. Deer density was summarized by assigning the density class that covered the majority of each county to represent the entire county.

Logistic regression models.

Logistic regression analysis was carried out in SAS 8.2 (SAS Institute, Inc., Cary, NC) using the serostatus of white-tailed deer populations as the dependent variable. We used stepwise regression with a P < 0.05 cutoff to reduce the large set of ecologic predictor variables to a more parsimonious subset. Because we had a priori reasons (e.g., differing climates in western versus eastern states) to believe that different predictive variables were important in different portions of the 19-state study region, we analyzed the region as a whole and then analyzed three separate subregions containing 18 of the 19 original states. The western subregion included Kansas, Oklahoma, and Texas (n = 110 of 563 counties), the central subregion included Alabama, Arkansas, Louisiana, Mississippi, and Missouri (n = 179), and the eastern subregion included Georgia, Delaware, Kentucky, New Jersey, North Carolina, South Carolina, Tennessee, Virginia, and West Virginia (n = 256). Florida was excluded from other subregions because of its semi-isolated geography and unique climate. Because of small sample size (n = 18), data for Florida could not be analyzed separately.

The logistic regression models were applied in ArcView 3.2 (Environmental Systems Research Institute) using spatial predictor variables from all of the counties in the 19-state region to create a map of the probability of E. chaffeensis occurrence (hereafter termed endemnicity probability). The overall fit of the models was measured by comparing observed and predicted values for the sampled counties, with E. chaffeensis classified as present when the predicted probability of occurrence was greater than 0.5, and absent otherwise. Accuracy, sensitivity, and specificity were computed as described previously. We also calculated the area under the receiver operating characteristic curve (AUC ROC) and the pseudo-R2 as overall indices of model fit.25,26

Comparison of predicted Ehrlichia chaffeensis distribution with HME case data.

Incidence rates of HME reported to the Centers for Disease Control and Prevention were obtained by combining data from McQuiston and others12 and Gardner and others.13 In both studies, annual incidence rates (per million human population) had been calculated using the corresponding yearly census estimate. Incidence groups were low (< 2.5), low-medium (2.5 to < 5), medium-high (5 to < 15), and high (> 15) by McQuiston and others12 and low (< 3.5), low-medium (3.5 to < 8.3), medium-high, (8.3 to < 17), and high (> 17) by Gardner and others.13 Although category limits differed slightly in the two studies, the four levels of incidence (low to high) were combined to obtain unified sequential incidence estimates. If a county was positive in both studies but classified in different incidence categories, the higher of the two was used in this study. These incidence estimates per county were entered into ArcView 3.2 and compared with the endemnicity probability maps generated using our geostatistical and logistic regression modeling.

RESULTS

Geostatistical modeling (kriging).

The fitted semivariogram had a range of 463.3 km, a nugget of 0.13014, and a sill of 0.21846 (Figure 1). The range identified the maximum distance at which spatial autocorrelation was detected among the sampled counties. The nugget quantified the minimum variability at a lag distance of zero, whereas the sill quantified maximum variability among spatially independent samples. The high ratio of nugget to sill values (0.60) indicated that a considerable amount of fine-scale (< 50 km) spatial variability in E. chaffeensis distribution was not accounted for in the kriging model. With a cutoff value of 0.5, the mean accuracy of the model was 85.4%, the mean sensitivity was 96.2%, and the mean specificity was 60.4% (Table 2).

Large irregularly shaped clusters of both positive and negative counties were detected in the study region (Figure 2). Positive counties were clustered in Atlantic coastal states from Florida to New Jersey and in a broad area extending north from Texas to Missouri east to Kentucky and south to Mississippi. Two large negative clusters were apparent. One was along the western edge of the region through western portions of Kansas, Oklahoma, and Texas and a second was centered along the Appalachians in Virginia, Kentucky, West Virginia, Tennessee, and Georgia. Two smaller negative clusters were observed: one in southern Louisiana and Mississippi and one in southern Florida. Sixty counties, both scattered and clustered within the study region, were misclassified as positive when they were observed to be negative (Figure 2). These misclassified counties occurred in close proximity to several positive counties.

Logistic regression models.

For the entire 19-state region-wide model, five variables were significantly related to E. chaffeensis serostatus in white-tailed deer (Table 3 and Figure 3). The E. chaffeensis endemnicity probability decreased with elevation, increased with percent savannah, increased with percent mixed deciduous forest (combination of deciduous and mixed), decreased with percent wooded wetlands, and increased with summer maximum temperature. With a cutoff value of 0.5, this model accurately predicted the E. chaffeensis serostatus of 421 (74.8%) of 563 counties (Table 2). Only 30 counties were misclassified as negative (sensitivity = 92.1%). The predicted distribution of seropositive white-tailed deer populations in some areas (southern Louisiana, Mississippi, and along the Mississippi River alluvial basin) was overestimated using this region-wide model and the predicted distribution in eastern Kansas and northern Missouri was underestimated (Figure 3).

In the eastern subregion model, five variables were significantly related to E. chaffeensis serostatus (Table 3 and Figure 4). Predicted distribution decreased with elevation, and increased with percent total forest, soil moisture content, annual minimum temperature, and summer relative humidity. Three explanatory variables were associated with serostatus of white-tailed deer populations in the central subregion model (Table 3). The predicted distribution decreased with percentage cropland and annual precipitation, and increased with the density of white-tailed deer (Table 3 and Figure 4). In the western subregion model, predicted positive counties were associated with increasing annual precipitation and increasing summer minimum temperature (Table 3 and Figure 4).

The overall fit of the three subregion models was better than the region-wide model. The accuracy of the eastern subregion model compared with the region-wide model increased by only 3.3%, but the specificity increased by 15.1% (Table 2). The accuracy of the central subregion model was greater than the overall model (79.3% versus 74.8%, Table 2). The sensitivity was slightly lower than the overall model, but the specificity was higher (47.7% versus 38.5%). The western subregion model correctly classified the serostatus of 84.5% counties, resulting in a 9.7% increase in predictive power over the overall model (Table 2). The sensitivity of the region-wide and western subregion models was the same, but the specificity of the western model was 29.1% higher than that of the region-wide model. All three subregion models had higher max-rescaled r2 values and AUC ROC (Table 2) indicating that they better fit the data.

Comparison with HME case data.

Within our 19 state study region, 271 counties in 15 states (Figure 5) had reported cases of HME during the years 1997–2002.12,13 The highest correlation with human case data was observed for the kriging analysis, with a total of 251 counties (92.6%) with human cases being classified as positive by our analysis (Figure 6). Overall, a high correlation was noted for the three subregion logistic models; 244 counties (90%) with reported HME cases were predicted to be positive. As HME incidence increased, there was higher correlation with correctly classified positive counties and nearly 100% of high incidence counties were classified as positive by our analyses (Figure 6).

DISCUSSION

We used two different modeling methods to predict the endemnicity probability of E. chaffeensis within the southeastern and southcentral United States and, by implication, identify locations with risk for HME. This is first large-scale study to use geospatial analyses to estimate the geographic distribution of E. chaffeensis, and it was conducted using data from a prototypic surveillance system that used white-tailed deer as natural sentinels.9 Data for this deer sentinel system was derived by serologically testing deer from 563 populations (counties) in 18 southcentral and southeastern states.9 Because geographic gaps existed since deer were not tested from every county, the present study used kriging and logistic regression modeling to predict the distribution of E. chaffeensis across all counties in these states. Logistic regression analyses also identified several climatic and landcover variables useful as predictors of E. chaffeensis occurrence. To relate these geospatial analyses to the occurrence of HME, comparisons were made with the geographic distribution of laboratory-confirmed or suspected cases of HME for the period 1997–2002 compiled by public health agencies as previously published.12,13 These comparisons disclosed spatial concordance between HME cases and E. chaffeensis endemnicity projected by both geospatial models.

The usefulness of modeling is dependent upon the reliability of input data, how well model predictions fit the actual status of variables being modeled, and in regard to geospatial modeling, especially the issues of scale and scope of sampling.27 Input data from the white-tailed deer surveillance sentinel system used in this study has previously been demonstrated to meet critical reliability criteria including diagnostic accuracy, adequacy of sample sizes and sampling intensity, key epidemiologic associations with the lone star tick vector, and ability to detect spread of E. chaffeensis.9 Data for the present study included more than one-third (33.8%) of counties in the 19-state study region, and measures of reliability for both kriging and logistic regression models disclosed high accuracy, sensitivity, and somewhat lower specificity (Table 2). Furthermore, analyses confirm that the logistic regression models fit the data well and that variables included in the models had biologic relevance based on previous studies.28,29

Kriging accurately estimated the distribution of E. chaffeensis across the entire study region. This procedure investigates and projects the spatial pattern of E. chaffeensis based on spatial autocorrelation among sample populations, but does not consider relationships with secondary variables such as landcover and climate. This purely spatial analysis is dependent on a sufficient number of populations being sampled because an increase in distances between sampled populations leads to increased uncertainty. Large clusters of counties with similar serologic status were accurately detected; however, when variability in E. chaffeensis serostatus occurred among counties in close proximity, this model tended to over predict the presence of E. chaffeensis. Two prominent clusters of seronegative populations in the Mississippi River alluvial basin were misclassified as seropositive, in this case, small numbers of seronegative counties were encircled by seropositive counties. Three smaller negative clusters in central Alabama, southcentral Georgia, and southeastern Texas also were projected to have moderate risk (0.5 < P < 0.7). Similarly, two positive counties in western Kansas, one in southern Louisiana, and one in western North Carolina fell well within areas projected to have a low probability of E. chaffeensis occurrence. The failure to correctly classify individual or small clusters of counties reflects the high nugget value of our semivariogram, which indicated that some fine scale variability (< 50 km) was present but not being detected.

Logistic regression analyses also accurately classified the E. chaffeensis status within the study region and identified several ecologic variables associated with the distribution of E. chaffeensis. In contrast to kriging, logistic regression modeling relies on correlations with ecologic variables that are then used to estimate the distribution of E. chaffeensis. The region-wide model had good accuracy (75%) and high specificity (92%), but low sensitivity (39%). Distributions of E. chaffeensis were accurately predicted in many states, but the region-wide logistic regression model produced results that conflicted with observed distributions in northern portions of Kansas and Missouri, central South Carolina, New Jersey, and portions of the Mississippi River alluvial basin and Gulf Coast. Partitioning into three subregions improved the accuracy, sensitivity, and specificity of the logistic regression modeling method. The most significant changes were the increased ability of subregion models to accurately predict positive status in northern Kansas/Missouri and New Jersey and to predict negative status for areas in eastern Kentucky, within the Mississippi River alluvial basin, and along the central Gulf coast. Both region-wide and subregional logistic regression models projected limited positive areas in north central Kentucky, northeastern Alabama, northwestern Georgia, and central South Carolina that were outside the current distribution based on observed serostatus and kriging analysis.

As with most predictive disease mapping studies, we did not explicitly account for spatial autocorrelation in our logistic regression models. Thus, our comparisons between spatial modeling methods highlight the differences between a purely spatial approach (kriging) and a purely ecologic approach (logistic regression). Both kriging and subregion logistic regression models provided very reliable portrayals of E. chaffeensis occurrence, and the similarity of maps produced by these two modeling processes tends to substantiate the projected distributions. Because kriging was more accurate than logistic regression, models the observed distribution of serostatus, and does not require collection of additional ecologic variables for use in a GIS, kriging analysis probably provides the most accurate estimate of current E. chaffeensis distribution and would be more useful in monitoring spread to new locations. Conversely, because logistic regression analysis is based on association with ecologic variables, the projected extensions of E. chaffeensis into currently negative areas suggests these areas probably are suitable for maintenance of E. chaffeensis and that they are likely locations of future expansion. Alternative statistical methods such as kriging with varying local means, autologistic regression, and geographically weighted regression have the potential to integrate spatial and ecologic information into a unified modeling framework.16,30,31 These approaches hold promise for improving prediction accuracy and for elucidating stronger relationships with ecologic predictor variables, and we plan to examine them in future studies.

The distribution of E. chaffeensis is highly dependent on the presence of white-tailed deer to serve as reservoirs,4,5 and the presence of the lone star tick to serve as a vector,3 and indirectly on any factors that limit the distribution of white-tailed deer or the lone star tick. Based on extensive biologic data relevant to the lone star tick, five critical variables, and interactions thereof, have been identified as being important to lone star tick population dynamics.28,29 These variables consist of measures of moisture, temperature, habitat type, day length, and host density (white-tailed deer and/or cattle). During warm months, temperature and moisture are positively correlated with survival of the lone star tick and activity, while day length is negatively correlated with lone star tick activity. Abundance and survival of the lone star tick are positively associated with wooded habitats, particularly young second-growth forests containing thick underbrush that provides proper microhabitat, but are negatively associated with grass or crop land habitats.3234 The landcover variables in our models (Table 3) conformed in general terms with these previously identified critical variables important to lone star tick population dynamics. The presence of E. chaffeensis was correlated with higher soil moistures and higher proportions of forest/savannah, which are habitats that provide better microhabitat for the lone star tick, while wooded wetlands, cropland (Figure 7), and high elevations (probably to some extent reflecting temperature) (Figure 7) were negatively correlated with E. chaffeensis.

Regarding climatic variables, our logistic regression models were in agreement with previous laboratory and field studies on the lone star tick in which high annual precipitation and high relative humidity prevent desiccation and higher average temperatures (Figure 7) help prolong questing, all of which conspire to increase lone star tick survival and densities.29,33,3537 Of interest was that the presence of E. chaffeensis was negatively correlated with increasing annual precipitation in the central subregion model, possibly because high precipitation in this low-lying region may lead to flooding or excessive wetness. The abundance of another ixodid tick, Ixodes scapularis, is negatively impacted by excessively wet soils,38 and our negative correlation with wooded wetlands suggests that A. americanum also may not be well suited for habitats subject to standing ground water.

The presence of white-tailed deer is clearly important to maintenance of high densities of lone star ticks;29; however, interestingly, we found that the density of white-tailed deer was not a significant predictor variable in the region-wide model or two of the three subregion models. Although deer were important to modeling population dynamics of lone star ticks,29 there are critical differences between modeling lone star ticks and modeling the probability of E. chaffeensis infection in white-tailed deer. Because the force of transmission is so heavily focused on white-tailed deer, low numbers of lone star ticks can quickly introduce and maintain E. chaffeensis within white-tailed deer populations.9,39 Furthermore, based on almost 100% concordance between the presence of lone star ticks and the presence of E. chaffeensis in white-tailed deer,8,9,20 and detection of E. chaffeensis by a polymerase chain reaction in nearly all surveys of white-tailed deer,6,7,10,4047 it is reasonable to conclude that a very high proportion of lone star tick populations are infected. White-tailed deer were present throughout our study area (Figure 7), frequently at high densities. Only the central subregion logistic regression model included deer density as a variable, and this appeared attributable to sizable highly agriculturalized areas within the Mississippi River alluvial basin where deer are rare, absent, or exist at low densities or in semi-isolated populations.

Comparison of the predicted distribution for E. chaffeensis with HME case data showed that > 90% of counties with HME cases were predicted positive for E. chaffeensis by both kriging and logistic regression. In fact, counties in the high HME incidence category were nearly always (> 98%) identified as positive for E. chaffeensis based on our analyses. Several counties with a low incidence of HME were predicted to be negative. This could be because our analyses misclassified the county status or that the HME cases were acquired in another locale but reported in the county of residence. This comparison provides evidence that use of white-tailed deer as sentinels is an efficient alternative to human surveillance for predicting E. chaffeensis distribution and HME risk.

The distribution of E. chaffeensis projected in the current study should be validated by testing additional white-tailed deer populations from unsampled counties, especially in areas where the prediction data were inconsistent between the two models. Although highly correlated with HME case data, a better means for validation of the sentinel deer surveillance system is needed because HME case data are known to contain a variety of biases and under-represent the distribution and magnitude of HME.12,13 An extensive cross-sectional serosurvey of humans for E. chaffeensis antibodies, which would provide a less biased and more accurate estimate of exposure to E. chaffeensis than does HME case data, should be conducted to validate the ability of the sentinel white-tailed deer surveillance system to accurately predict HME risk.

Table 1

Description of variables used in logistic regression modeling

VariableVariable descriptionUnitsRange
ElevationAverage elevation per countyMeters1–1,501
RuggednessDifference in maximum and minimum elevation valuesMeters1–1,843
Soil drainage valueRepresents the natural drainage condition of the soil and refers to the frequency and duration of periods when the soil is free of saturationNone1.9–80.7
TemperatureAverage monthly and annual minimum and maximum temperature from 1980 to 1997°CMonthly −9.9–29.6
 Annual 2.3–29.6
PrecipitationAverage monthly and annual precipitation from 1980 to 1997CentimetersMonthly 0.6–24.6
 Annual 23.5–189.6
Frost daysAverage number of frost days per year from 1980 to 1997Days0.4–164.5
Relative humidityAverage relative humidity calculated from the monthly and annual water vapor pressure and the monthly and annual minimum and maximum temperatures%Monthly 20.5–82.5
 Annual 35.7–85.8
Deer densityCategories of deer density for each county based on 1999 population estimatesNo. of deer/km25 categories for densities of 0–≥45 deer/km2
Land coverUrbanPercentage of urban and built-up land% cover0–100
Land coverCropsPercentage of dryland and irrigated cropland, grassland and woodland mosaic, and pasture% cover0–100
Land coverSavannaPercentage of savanna% cover0–95
Land coverDeciduous broadleaf and mixed forestPercentage of deciduous broadleaf forest and mixed forest (no type of tree exceeds 50% of tree types)% cover0–100
Land coverEvergreen needleleaf and mixed forestPercentage of evergreen needleleaf forest and mixed forest (no type of tree exceeds 50% of tree types)% cover0–98
Land coverTotal forestPercentage of all forest types% cover0–100
Land coverWdwetPercentage of wooded wetlands% cover0–63
Table 2

Comparison of four models for predicting distribution of Ehrlichia chaffeensis based on a white-tailed deer surveillance system*

Analysis†Kriging‡ mean (range)19-State regionEastern subregionCentral subregionWestern subregion
* NA = not applicable; AUC ROC = area under the receiver operating characteristic curve.
† A county was classified as positive for E. chaffeensis if P > 0.5 and compared to the actual serologic status of that county.
‡ 75% of the values (n = 422) were used to develop a kriging model and the remaining 25% were used to validate the model.
Accuracy (%)85.4 (81.6–87.9)74.878.179.384.5
Sensitivity (%)96.2 (93.2–98.9)92.193.189.692.1
Specificity (%)60.4 (45.2–68.6)38.553.647.767.6
Maximum-rescaled R2NA0.240.450.310.43
AUC ROCNA0.7240.8340.8040.845
Table 3

Logistic regression models for predicting distribution of Ehrlichia chaffeensis

RegionVariableCoefficientχ2P
19-state regionIntercept−11.5114.730.0001
Elevation−0.0935.45< 0.0001
Savanna3.784.670.0308
Deciduous forest mix1.1810.790.0010
Wooded wetlands−214.95.060.0244
Summer maximum temperature0.4120.62< 0.0001
Eastern subregionIntercept−36.7211.060.0009
Elevation−0.3128.61< 0.0001
Forest1.888.360.0038
Soils0.0814.300.0002
Annual minimum temperature0.205.620.0178
Summer relative humidity0.5211.630.0006
Central subregionIntercept18.9923.80< 0.0001
Annual precipitation−1.5521.69< 0.0001
Crops−2.6911.470.0007
Deer density0.597.650.0057
Western subregionIntercept−13.9519.45< 0.0001
Annual precipitation0.8516.66< 0.0001
Summer minimum temperature0.377.160.0074
Figure 1.
Figure 1.

Semivariogram of the occurrence of Ehrlichia chaffeensis. Lag = 50 km; nugget variance c0 = 0.13014; sill c = 0.21846; range a = 463.3 km.

Citation: The American Journal of Tropical Medicine and Hygiene Am J Trop Med Hyg 72, 6; 10.4269/ajtmh.2005.72.840

Figure 2.
Figure 2.

Map of endemnicity probabilities for Ehrlichia chaffeensis based on kriging. Solid circles represent positive E. chaffeensis serologic data and open circles represent negative serologic data obtained from a previous white-tailed deer surveillance study.9 This figure appears in color at www.ajtmh.org.

Citation: The American Journal of Tropical Medicine and Hygiene Am J Trop Med Hyg 72, 6; 10.4269/ajtmh.2005.72.840

Figure 3.
Figure 3.

Map of endemnicity probabilities for Ehrlichia chaffeensis based on a region-wide logistic regression model derived for the entire 19-state region. Solid circles represent positive E. chaffeensis serologic data and open circles represent negative serologic data obtained from a previous white-tailed deer surveillance study.9 This figure appears in color at www.ajtmh.org.

Citation: The American Journal of Tropical Medicine and Hygiene Am J Trop Med Hyg 72, 6; 10.4269/ajtmh.2005.72.840

Figure 4.
Figure 4.

Maps of endemnicity probabilities for Ehrlichia chaffeensis based on separate logistic regression models for the eastern, central, and western subregions. Projections for Florida are from the original region-wide model. Solid circles represent positive E. chaffeensis serologic data and open circles represent negative serologic data obtained from a previous white-tailed deer surveillance study.9 This figure appears in color at www.ajtmh.org.

Citation: The American Journal of Tropical Medicine and Hygiene Am J Trop Med Hyg 72, 6; 10.4269/ajtmh.2005.72.840

Figure 5.
Figure 5.

County-level average annual incidence rates for human monocytotrophic ehrlichiosis (HME) reported to the Centers for Disease Control and Prevention through the National Electronic Telecommunications System for Surveillance. Data were obtained from references 12 and 13.

Citation: The American Journal of Tropical Medicine and Hygiene Am J Trop Med Hyg 72, 6; 10.4269/ajtmh.2005.72.840

Figure 6.
Figure 6.

Comparison of human monocytotrophic ehrlichiosis (HME) case data with the predicted presence of Ehrlichia chaffeensis based on kriging and logistic regression analyses.

Citation: The American Journal of Tropical Medicine and Hygiene Am J Trop Med Hyg 72, 6; 10.4269/ajtmh.2005.72.840

Figure 7.
Figure 7.

Geographic information system overlay maps for selected key explanatory variables used in development of logistic regression models. A, White-tailed deer density; B, Elevation; C, Percentage of all crops; D, Annual minimum temperature. This figure appears in color at www.ajtmh.org.

Citation: The American Journal of Tropical Medicine and Hygiene Am J Trop Med Hyg 72, 6; 10.4269/ajtmh.2005.72.840

Authors’ addresses: Michael J. Yabsley, Southeastern Cooperative Wildlife Disease Study, College of Veterinary Medicine, University of Georgia, Athens, GA 30602. Michael C. Wimberly, D. B. Warnell School of Forest Resources, The University of Georgia, Athens, GA 30602. David E. Stallknecht and Susan E. Little, Department of Infectious Diseases, College of Veterinary Medicine, The University of Georgia, Athens, GA 30602. William R. Davidson, Southeastern Cooperative Wildlife Disease Study, College of Veterinary Medicine, University of Georgia, Athens, GA 30602 and D. B. Warnell School of Forest Resources, The University of Georgia, Athens, GA 30602.

Acknowledgments: We thank Scott Danskin and Sarah Cross for excellent help with building the GIS databases used in this study. We also thank the numerous staff at federal, state, and private wildlife agencies and the Southeastern Cooperative Wildlife Disease Study for collecting samples used in this study.

Financial support: This work was supported primarily by the National Institutes of Allergy and Infectious Diseases (grant 5 R01 AI044235). Additional support was provided by the Federal Aid to Wildlife Restoration Act (50 Stat. 917) and through sponsorship from fish and wildlife agencies in Alabama, Arkansas, Florida, Georgia, Kansas, Kentucky, Louisiana, Maryland, Mississippi, Missouri, North Carolina, Oklahoma, Puerto Rico, South Carolina, Tennessee, Virginia, and West Virginia.

REFERENCES

  • 1

    Dawson JE, Stallknecht DE, Howerth EW, Warner C, Biggie K, Davidson WR, Lockhart JM, Nettles VF, Olson JG, Childs JE, 1994. Susceptibility of white-tailed deer (Odocoileus virginianus) to infection with Ehrlichia chaffeensis, the etiologic agent of human ehrlichiosis. J Clin Microbiol 32 :2725–2728.

    • Search Google Scholar
    • Export Citation
  • 2

    Dawson JE, Childs JE, Biggie KL, Moore C, Stallknecht D, Shaddock J, Bouseman J, Hofmeister E, Olson JG, 1994. White-tailed deer as a potential reservoir of Ehrlichia spp. J Wildl Dis 30 :162–168.

    • Search Google Scholar
    • Export Citation
  • 3

    Ewing SA, Dawson JE, Kocan AA, Barker RW, Warner CK, Panciera RJ, Fox JC, Kocan KM, Blouin EF, 1995. Experimental transmission of Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) among white-tailed deer by Amblyomma americanum (Acari: Ixodidae). J Med Entomol 32 :368–374.

    • Search Google Scholar
    • Export Citation
  • 4

    Lockhart JM, Davidson WR, Stallknecht DE, Dawson DE, Howerth EW, 1997. Isolation of Ehrlichia chaffeensis from wild white-tailed deer (Odocoileus virginianus) confirms their role as natural reservoir hosts. J Clin Microbiol 35 :1681–1686.

    • Search Google Scholar
    • Export Citation
  • 5

    Lockhart JM, Davidson WR, Stallknecht DE, Dawson JE, Little SE, 1997. Natural history of Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) in the Piedmont physiographic province of Georgia. J Parasitol 83 :887–894.

    • Search Google Scholar
    • Export Citation
  • 6

    Childs JE, Paddock CD, 2002. The ascendancy of Amblyomma americanum as a vector of pathogens affecting humans in the United States. Annu Rev Entomol 48 :307–337.

    • Search Google Scholar
    • Export Citation
  • 7

    Paddock CD, Childs JE, 2003. Ehrlichia chaffeensis: a prototypical emerging pathogen. Clin Microbiol Rev 16 :37–64.

  • 8

    Yabsley MJ, Varela AS, Tate CM, Dugan VG, Stallknecht DE, Little SE, Davidson WR, 2002. Ehrlichia ewingii infection in white-tailed deer (Odocoileus virginianus). Emerg Infect Dis 8 :668–671.

    • Search Google Scholar
    • Export Citation
  • 9

    Yabsley MJ, Dugan VG, Stallknecht DE, Little SE, Lockhart JM, Dawson JE, Davidson WR, 2003. Evaluation of a prototype Ehrlichia chaffeensis surveillance system using white-tailed deer (Odocoileus virginianus) as natural sentinels. Vector Borne Zoonotic Dis 3 :195–207.

    • Search Google Scholar
    • Export Citation
  • 10

    Irving RP, Pinger RR, Vann CN, Olesen JB, Steiner FE, 2000. Distribution of Ehrlichia chaffeensis (Rickettsiales: Rickettsiaeceae) in Amblyomma americanum in southern Indiana and prevalence of E. chaffeensis–reactive antibodies in white-tailed deer in Indiana and Ohio in 1998. J Med Entomol 37 :595–600.

    • Search Google Scholar
    • Export Citation
  • 11

    Mueller-Anneling L, Gilchrist MJ, Thorne PS, 2000. Ehrlichia chaffeensis antibodies in white-tailed deer, Iowa, 1994 and 1996. Emerg Infect Dis 6 :397–400.

    • Search Google Scholar
    • Export Citation
  • 12

    McQuiston JH, Paddock CD, Holman RC, Childs JE, 1999. Human ehrlichioses in the United States. Emerg Infect Dis 5 :635–642.

  • 13

    Gardner SL, Holman RC, Krebs JW, Berkelman R, Childs JE, 2003. National surveillance for the human ehrlichioses in the United States, 1997–2001, and proposed methods for evaluation of data quality. Ann N Y Acad Sci 990 :80–89.

    • Search Google Scholar
    • Export Citation
  • 14

    Isaaks EH, Srivastava RM, 1989. An Introduction to Applied Geostatistics. New York: Oxford University Press.

  • 15

    Rossi RE, Mulla DJ, Journel AG, Franz EH, 1992. Geostatistical tools for modeling and interpreting ecological spatial dependence. Ecol Monogr 62 :277–314.

    • Search Google Scholar
    • Export Citation
  • 16

    Goovaerts P, 1997. Geostatistics for Natural Resources Evaluation. New York: Oxford University Press.

  • 17

    Kleinschmidt I, Bagayoko M, Clarke GPY, Craig M, Le Sueur D, 2000. A spatial statistical approach to malaria mapping. Int J Epidemiol 29 :355–361.

    • Search Google Scholar
    • Export Citation
  • 18

    Brownstein JS, Holford TR, Fish D, 2003. A climate-based model predicts the spatial distribution of the Lyme disease vector Ixodes scapularis in the United States. Environ Health Perspect 111 :1152–1157.

    • Search Google Scholar
    • Export Citation
  • 19

    Guerra M, Walker E, Jones C, Paskewitz S, Cortinas MR, Stancil A, Beck L, Bobo M, Kitron U, 2002. Predicting the risk of Lyme disease: Habitat suitability for Ixodes scapularis in the north central United States. Emerg Infect Dis 8 :289–297.

    • Search Google Scholar
    • Export Citation
  • 20

    Lockhart JM, Davidson WR, Stallknecht DE, Dawson JE, 1996. Site-specific geographic association between Amblyomma americanum (Acari: Ixodidae) infestations and Ehrlichia chaffeensis-reactive (Rickettsiales: Ehrlichieae) antibodies in white-tailed deer. J Med Entomol 33 :153–158.

    • Search Google Scholar
    • Export Citation
  • 21

    Little SE, Stallknecht DE, Lockhart JM, Dawson JE, Davidson WR, 1998. Natural coinfection of a white-tailed deer (Odocoileus virginianus) population with three Ehrlichia spp. J Parasitol 84 :897–901.

    • Search Google Scholar
    • Export Citation
  • 22

    Thornton PE, Running SW, White MA, 1997. Generating surfaces of daily meteorological variables over large regions of complex terrain. J Hydrol 190 :214–251.

    • Search Google Scholar
    • Export Citation
  • 23

    Thornton PE, Hasenauer H, White MA, 2000. Simultaneous estimation of daily solar radiation and humidity from observed temperature and precipitation: an application over complex terrain in Austria. Agric Forest Meterorol 104 :255–271.

    • Search Google Scholar
    • Export Citation
  • 24

    Loveland TR, Reed BC, Brown JF, Ohlen DO, Zhu J, Yang L, Merchant JW, 2000. Development of a Global Land Cover Characteristics Database and IGBP DISCover from 1-km AVHRR Data. Int J Remote Sens 21 :1303–1330.

    • Search Google Scholar
    • Export Citation
  • 25

    Fielding AH, Bell JF, 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environ Conserv 24 :38–49.

    • Search Google Scholar
    • Export Citation
  • 26

    Nagelkerke NJD, 1991. A note on a general definition of the coefficient of determination. Biometrika 78 :691–692.

  • 27

    Matthews SA, 1990. Epidemiology using a GIS: the need for caution. Comput Environ Urban Sys 14 :213–221.

  • 28

    Haile DG, Mount GA, 1987. Computer simulation of population dynamics of the lone star tick, Amblyomma americanum (Acari: Ixodidae). J Med Entomol 24 :356–369.

    • Search Google Scholar
    • Export Citation
  • 29

    Mount GA, Haile DG, Barnard DR, Daniels E, 1993. New version of LSTSIM for computer simulation of Amblyomma americanum (Acari:Ixodidae) population dynamics. J Med Entomol 30 :843–857.

    • Search Google Scholar
    • Export Citation
  • 30

    Augustin NH, Mugglestone MA, Buckland ST, 1996. An autologistic model for the spatial distribution of wildlife. J Appl Ecol 33 :339–347.

    • Search Google Scholar
    • Export Citation
  • 31

    Fotheringham AS, Brunsdon C, Charlton ME, 2002. Geographically Weighted Regression: The Analysis of Spatially Varying Relationships. Chichester, United Kingdom: John Wiley.

  • 32

    Hair JA, Howell DE, 1970. Lone Star Ticks - Their Biology and Control in Ozark Recreation Areas. Stillwater, OK: Oklahoma State University Agriculture Experimental Station Bulletin. B-679.

  • 33

    Lancaster JL Jr, 1957. Control of the Lone Star Ticks. Fayetteville, AR. Agricultural Experimental Station. College of Agriculture and Horticultural Ecology, University of Arkansas, Representative Series 67.

  • 34

    Soneshine DE, Atwood EL, Lamb JT, 1966. The ecology of ticks transmitting Rocky Mountain spotted fever in a study area in Virginia. Ann Entomol Soc Am 59 :1234–1262.

    • Search Google Scholar
    • Export Citation
  • 35

    Hooker WA, Bishopp FC, Wood HP, 1912. Some North American Ticks. Washington, DC: United States Bureau of Entomology Bulletin No. 106.

  • 36

    Robertson AS, Patrick CD, Semtner PJ, Hair JA, 1975. The ecology and behavior of the lone star tick (Acarine:Ixodidae): VI. Response of unfed adults to certain environmental parameters. J Med Entomol 12 :525–529.

    • Search Google Scholar
    • Export Citation
  • 37

    Koch HG, Dunn JC, 1980. Oviposition, egg hatch, and larval survival of the lone star ticks (Amblyomma americanum) held at different temperatures and humidities. Southwestern Entomol 5 :169–174.

    • Search Google Scholar
    • Export Citation
  • 38

    Bunnell JE, Price SD, Das A, Shields TM, Glass GE, 2003. Geographic information systems and spatial analysis of adult Ixodes scapularis (Acari:Ixodidae) in the Middle Atlantic Region of the U.S.A. J Med Entomol 40 :570–576.

    • Search Google Scholar
    • Export Citation
  • 39

    Lockhart JM, Davidson WR, Dawson JE, Stallknecht DE, 1995. Temporal association of Amblyomma americanum with the presence of Ehrlichia chaffeensis reactive antibodies in white-tailed deer. J Wildl Dis 31 :119–124.

    • Search Google Scholar
    • Export Citation
  • 40

    Anderson BE, Sims KG, Olson JG, Childs JE, Piesman JF, Happ CM, Maupin GO, Johnson BJ, 1993. Amblyomma americanum: a potential vector of human ehrlichiosis. Am J Trop Med Hyg 49 :239–244.

    • Search Google Scholar
    • Export Citation
  • 41

    Burket CT, Vann CN, Pinger RR, Chatot CL, Steiner FE, 1998. Minimum infection rate of Ambylomma americanum (Acari: Ixodidae) by Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) in southern Indiana. J Med Entomol 35 :653–659.

    • Search Google Scholar
    • Export Citation
  • 42

    Ijdo JW, Wu C, Magnarelli LA, Stafford KC 3rd, Anderson JF, Fikrig E, 2000. Detection of Ehrlichia chaffeensis DNA in Amblyomma americanum ticks in Connecticut and Rhode Island. J Clin Microbiol 38 :4655–4656.

    • Search Google Scholar
    • Export Citation
  • 43

    Roland WE, Everett ED, Cyr TL, Hasan SZ, Dommaraju CB, McDonald GA, 1998. Ehrlichia chaffeensis in Missouri ticks. Am J Trop Med Hyg 59 :641–643.

    • Search Google Scholar
    • Export Citation
  • 44

    Steiert JB, Gilfoy F, 2002. Infection rates of Amblyomma americanum and Dermacentor variabilis by Ehrlichia chaffeensis and Ehrlichia ewingii in Southwest Missouri. Vector Borne Zoonotic Dis 2 :53–60.

    • Search Google Scholar
    • Export Citation
  • 45

    Steiner FE, Pinger RR, Vann CN, 1999. Infection rates of Amblyomma americanum (Acari: Ixodidae) by Ehrlichia chaffeensis (Rickettsiales: Ehrlichieae) and prevalence of E. chaffeensis-reactive antibodies in white-tailed deer in southern Indiana, 1997. J Med Entomol 36 :715–719.

    • Search Google Scholar
    • Export Citation
  • 46

    Stromdahl EY, Evans SR, O’Brien JJ, Gutierrez AG, 2001. Prevalence of infection in ticks submitted to the Human Tick Test Kit Program of the U.S. Army Center for Health Promotion and Preventive Medicine. J Med Entomol 38 :67–74.

    • Search Google Scholar
    • Export Citation
  • 47

    Whitlock JE, Fang QQ, Durden LA, Oliver JH Jr, 2000. Prevalence of Ehrlichia chaffeensis (Rickettsiales: Rickettsiaceae) in Amblyomma americanum (Acari: Ixodidae) from the Georgia coast and Barrier Islands. J Med Entomol 37 :276–280.

    • Search Google Scholar
    • Export Citation

Author Notes

Reprint requests: Michael J. Yabsley, Wildlife Health Building, Southeastern Cooperative Wildlife Disease Study, College of Veterinary Medicine; University of Georgia, Athens, GA 30602, Telephone: 706-542-1741, Fax: 706-542-5865, E-mail: myabsley@vet.uga.edu.
Save