Study design
The cross-sectional study was conducted on blood samples collected within the Swedish Bovine Viral Diarrhoea (BVD) surveillance program. Within this program, all Swedish herds are required to be tested on a regular basis to maintain their BVD free status [18]. For beef cattle herds, depending on the number of dams present in the herd, five to ten blood samples are taken in young stock over 12 months of age per herd-year and sent to the National Veterinary Institute where they are analyzed for presence of BVDV antibodies [19]. In total, approximately 45,000 blood samples are collected annually from beef herds.
Between November 2006 and May 2007, every 12th blood sample was systematically selected for an investigation of Neospora caninum in Swedish beef cattle [20]. The same study sample was used here; it consisted of 2,763 serum samples originating from 2,137 herds, corresponding to approximately 20% of all beef herds present in the country at this time. The sample was considered to be representative of the Swedish beef cattle population, as it was issued from a procedure functionally similar to a random sampling. The number of blood samples taken per herd ranged from 1 to 8, but most herds were represented by one or two samples (81 and 14%, respectively).
Diagnostic tests
The samples were analysed for presence of immunoglobulin G antibodies to BCV [4] and BRSV [21] by commercially available indirect enzyme-linked immunosorbent assays (ELISA; SVANOVA Biotech, Uppsala, Sweden). The optical density (OD) at 450 nm was corrected by subtraction of the negative control antigen OD. Cut-off was set to a corrected OD of 0.20, which is recommended by the manufacturer for individual samples. At this cut-off, the sensitivity is estimated to 84.6% for BCV and 94.6% for BRSV and specificity to 100% for both tests (SVANOVA manual). A sample was considered test positive if its corrected OD was >0.20, and test negative otherwise.
Location data
The locations of all Swedish beef herds, including the herds where the blood samples were collected, were specified by three-digit postal codes. Postal codes were retrieved from the database of the organization responsible for the BVD surveillance program, i.e. the Swedish Dairy Association (year 2007). Applicable postal codes were available for 2,757 samples from 2,131 beef herds in the study population.
Spatial analyses
For BCV and BRSV infections separately, the spatial distribution of samples and possible clustering were investigated by following the same procedure using data aggregated by postal code area (PCA).
The crude prevalence was defined as the number of positive samples over the total number of samples and was calculated for each PCA. This raw prevalence was adjusted by applying a Spatial Empirical Bayes smoothing (SEB), i.e. adjusted (i) for the potential biasing effects of variance instability due to differences in the size of the population at risk, and (ii) considering the estimates from neighboring areas [22]. Presence of global spatial autocorrelation was tested using the Moran's I test for SEB rates [23]. Its significance was calculated by Monte-Carlo simulation. All smoothing and testing for spatial associations of area aggregated data was performed using the GeoDa software version 0.9.5-i5 http://geodacenter.asu.edu/.
Identification of potential clusters of positive samples was based on location determined by PCA centroids, and using the spatial scan statistic (M. Kulldorff and Information Management Services, Inc. SatScan version 8.0, http://www.satscan.org, 2009). The method is based on either circles or ellipses centered on each PCA centroid; a Relative Risk can be estimated which compares the risk of being a case inside the circle/ellipse to the risk of being a case outside the circle/ellipse [24]. A circle/ellipse is considered a cluster if the Relative Risk is significantly higher or lower than one, when significance was tested using Monte-Carlo simulation. In this study, Poisson models applying both different cluster shapes (circular and elliptic) and sizes (maximum cluster sizes of 50, 20 and 10% of the total population at risk) were built to identify both high-risk and low-risk clusters. No overlapping of the circles/ellipses was allowed.
BCV and BRSV are contagious, and test positive animals are expected to be grouped in herds. As the likelihood of detecting at least one test positive animal increases with the number of individual samples collected per herd, identification of clusters and their spatial location might be biased if the herds from which two or more blood samples (19% of the studied herds) were collected are not evenly geographically-distributed. To investigate this potential bias in relation to the sampling strategy, possible clustering of herds with more than one sampled animal was explored in a preliminary step by using Moran's I. This test indicated that these herds were proportionally distributed over Sweden (data not shown).
To test whether or not a high beef herd-density was a risk factor for significantly higher seroprevalences of BCV and BRSV than expected, univariate logistic regression analysis was performed at PCA level, where the binary outcome was "PCA with a significantly higher number of positive-tested samples vs. without" and the putative explanatory variable was the PCA herd-density (in herds/100 km2) in 3 classes [<5; 5-10; >10].
Data management, statistics and creation of map shape-files were performed using SAS 9.2 (SAS Institute, Inc., Cary, NC, USA) and ArcGIS 9.1 (ESRI Inc., Redlands, CA, USA).