PropertyValue
is nif:broaderContext of
nif:broaderContext
is schema:hasPart of
schema:isPartOf
nif:isString
  • Block Island is a 25.2 km2 landmass located in Washington County, Rhode Island, 23 km south of mainland Rhode Island [34]. The population of permanent residents is around 1,000, which increases during the summer months to approximately 12,000 with the influx of summer residents [31]. Deciduous forest, the most suitable habitat type for Ixodes scapularis in the mainland, is limited on the island to a 4 ha site [35], so most tick habitat on the Island is restricted to shrublands and shrub edges with sufficient leaf litter accumulation for tick survival. A longitudinal study was established in 1991 on Block Island, RI, by inviting all island residents to take part in serological surveys conducted every year in the spring and fall. The study population was restricted to residents who spent more than one month on the island during the May through September Lyme disease transmission period. The serosurvey was announced in the local newspaper, on television, and via flyers at local businesses and the Block Island Medical Center [31], [36]. All subjects were asked to provide blood samples for B. burgdorferi and B. microti serological analyses and to complete a questionnaire. Written informed consent was obtained from all study participants in accordance with the human investigation committees at the University of Connecticut School of Medicine and the Yale School of Public Health. They each provide ethical review and oversight of human research endeavors and approved this study. We restricted our analysis to subjects who participated in serosurveys from 2005 to 2011 because environmental exposure was assessed based on a 2010 satellite image. Landscape metrics were assumed to be minimally changed during that period. To minimize the influence of a recent Lyme disease diagnosis on an individual’s behavior, we excluded from the study individuals reporting a Lyme disease diagnoses within two years of their initial serosurvey visit and data from all subsequent visits after an individual developed positive B. burgdorferi serology. We defined a person with Lyme disease exposure as an individual who tested positive for B. burgdorferi antibody using a standard two tier ELISA and western blot antibody approach [36]. A positive ELISA result consisted of an IgM or IgG response at ≥1∶320 dilution. Positive or equivocal ELISA results were confirmed by western blotting. Specimens were considered positive if 5 or more bands of the ten most prevalent B. burgdorferi-specific bands were present in the immunoblot [36]–[38]. All antibody assays prior to the fall of 2008 were carried out at the University of Connecticut Health Center. Assays from the spring survey of 2009 until the fall of 2011 were performed by commercial laboratories in New England using standard Lyme serodiagnostic assays. All subjects were asked to complete a questionnaire assessing the history of previous tick borne illnesses, peridomestic factors potentially linked to tick exposure, their age, protective behaviors and outdoor activities (Table 1, Figure S1). The questionnaire was administered at the time of the blood draw and included questions about regularly performed behaviors related to tick exposure. Table data removed from full text. Table identifier and caption: 10.1371/journal.pone.0084758.t001 Behavioral and demographic characteristics of survey responders in relation to their serological status. Percent positive (or average) responses over the total responses for each question for behaviors and age reported by B. burgdorferi seropositive and seronegative participants in serological surveys between 2005 and 2011.Use of any protective measure = use of either protective clothing, tick checking, repellent or avoiding brush. We developed remotely sensed landscape metrics that quantified the amount of edge between lawn and shrub vegetation at all subject residences. We hypothesized that these sites would be areas of high tick density and increase human contact with ticks. We first generated a high resolution land cover classification of Block Island, and then calculated the composition and configuration (landscape metrics) of shrub and lawn cover characteristics for each individual property. Finally, we assessed the association between the landscape metrics and the density of I. scapularis nymphs in a representative subset of properties, as described below. We generated a land cover classification of Block Island using WorldView2 satellite sensor data acquired on September 10, 2010. WorldView2 has a spatial resolution of 1.82 m that allows detection of fine scale peridomestic landscape patterns. It also has high spectral resolution (8 bands), resulting in greater ability to discriminate among land cover types than the four bands typically available for other high spatial resolution sensors. We converted the data to top of the atmosphere radiance [39]. We performed a maximum likelihood land cover classification using ENVI (Environment for Visualizing Images) software (ITT 2011) [40]. To improve the accuracy of the classification, the image was stratified into vegetated and non-vegetated areas based on a threshold of the normalized difference vegetation index = 0 [40] and the classification was performed independently for each stratum. Training and testing pixels were obtained by collecting ground information for all vegetation classes and by visual inspection of a 2010 orthophoto for the water and urban-associated classes, which easily could be distinguished. A randomly selected subset of 80% of the pixels was used to train the classification and 500 pixels were randomly selected from the remaining 20% for each of the classes for testing. After the classification, the two strata were combined in one raster layer and a 3×3 pixel median filter was applied to remove the salt and pepper effect (Figure 1). Figure data removed from full text. Figure identifier and caption: 10.1371/journal.pone.0084758.g001 Land cover classification of Block Island, Rhode Island. Examples of properties with a) low shrub edge density and b) high shrub edge density.Map shows a Worldview2 image acquired on Sept. 4, 2010 from Digital Globe, Inc. Spatial analysis was performed utilizing geo-spatial modeling environment (GME) software version 0.5.8 beta [41] and Fragstats software version 3.3 [42]. We used a municipal parcel layer (Town of New Shoreham) to calculate the following landscape metrics for shrubs and lawns in each property parcel: area of the landscape class, largest patch index, total edge, edge density and landscape shape index (calculations described in Table S1). Landscape metrics were normalized using the Z-scale ([X-mean]/standard deviation) prior to use in statistical analyses. Association between landscape metrics and the density of I. scapularis nymphs: During 2012, I. scapularis nymphs were collected from 105 properties of serosurvey participants from May 15th to August 23rd. The property surveys consisted of dragging 1 m2 corduroy cloths along the edge of the lawn and shrub vegetation as outlined in previous studies [43]–[45]. Between 2 and 5 transects of approximately 100 meters in length were completed at each property, proportionally to the size of the property. Most properties were repeatedly sampled during the season, resulting in a total of 258 samples. Attached I. scapularis nymphs were counted, placed in 70% ethanol, and species confirmed using taxonomic keys [46]. We based our measure of risk only on the density of host-seeking I. scapularis nymphs (hereafter density of nymphs) without calculating the proportion infected with Borrelia burgdorferi because the small number of nymphs collected on most properties prevented an accurate estimate of infection prevalence. Identification of environmental risk factors: Negative binomial regression was used to assess the association between landscape metrics and the density of nymphs. Only those landscape metrics found to be significantly associated with the density of nymphs on a property were considered biologically relevant and thus included as potential risk factors in further analyses. Individual and environmental risk factors for Lyme disease: We used general estimating equation models (XTGEE) in STATA/SE, version 12.0 (STATA Corporation, College Station, TX) to assess the association between personal protective behaviors, age, landscape metrics and individual serological status. These models fit generalized linear models that yield logistic regression models via a Bernoulli distribution of the dependent variable and a logit link function. The models accounted for potential autocorrelation among observations in a time series - in this case serological tests at different time periods on the same subject. We performed univariate analyses for all variables and then examined multivariate models including all possible combinations of variables found to be significant in univariate analyses. We assessed two groups of models: one including self-reported Lyme disease diagnosis and one excluding this variable. Including self-reported Lyme disease is informative in terms of the consistency between previous and current risk; excluding this variable allowed for identification of current risk factors for Lyme disease infection. The maximum model size was reached when larger models resulted in all non-significant variables. We included age in all models to control for confounding. No more than one landscape metric was included in a model because these variables were highly collinear. Pairwise correlation among all variables was assessed and only variables with Pearson correlation coefficient lower than 0.2 were included in the same model. Models were compared by the QIC criterion, which an extension of the Akaike information criterion (AIC) [47]–[48] used for generalized estimating equation models [49]. The QIC is a measure of the relative quality of a statistical model for a given set of data. Similar to AIC, QIC not only rewards goodness of fit, but also includes a penalty that is an increasing function of the number of estimated parameters, resulting in the most parsimonious model [50]. We additionally assessed whether inclusion of variable interactions improved model fit and assessed the goodness-of-fit of the final model using the Hosmer-Lemeshow test [51], [52]. Finally, to determine whether there were spatial relationships among the properties not captured by the measured variables, we evaluated whether the residuals of the model were significantly autocorrelated using the Moran’s I test included in the ArcGIS Spatial Statistics toolbox [53], [54].
rdf:type