Predicting within-herd prevalence of infection with bovine leukemia virus using bulk-tank milk antibody levels

Enzootic bovine leukosis (EBL) is an economically important infection of dairy cattle caused by bovine leukemia virus (BLV). Estimating the prevalence of BLV within dairy herds is a fundamental step towards pursuing efficient control programs. The objectives of this study were: (1) to determine the prevalence of BLV infection at the herd level using a bulk-tank milk (BTM) antibody ELISA in the Maritime region of Canada (3 provinces); and (2) to develop appropriate statistical models for predicting within-herd prevalence of BLV infection using BTM antibody ELISA titers. During 2013, three monthly BTM samples were collected from all dairy farms in the Maritime region of Canada (n=623) and tested for BLV milk antibodies using a commercial indirect ELISA. Based on the mean of the 3 BTM titers, 15 strata of herds (5 per province) were defined. From each stratum, 6 herds were randomly selected for a total of 90 farms. Within every selected herd, an additional BTM sample was taken (round 4), approximately 2 months after the third round. On the same day of BTM sampling, all cows that contributed milk to the fourth BTM sample were individually tested for BLV milk antibodies (n=6111) to estimate the true within-herd prevalence for the 90 herds. The association between true within-herd prevalence of BLV and means of various combinations of the BTM titers was assessed using linear regression models, adjusting for the stratified random sampling design. Herd level prevalence of BLV in the region was 90.8%. In the individual testing, 30.4% of cows were positive. True within-herd prevalences ranged from 0 to 94%. All linear regression models were able to predict the true within-herd prevalence of BLV reasonably well (R(2)>0.69). Predictions from the models were particularly accurate for low-to-medium spectrums of the BTM titers. In general, as a greater number of the four repeated BTM titers were incorporated in the models, narrower confidence intervals around the prediction lines were achieved. The model including all 4 BTM tests as the predictor had the best fit, although the models using 2 and 3 BTM tests provided similar results to 4 repeated tests. Therefore, testing two or three BTM samples with approximately two-month intervals would provide relatively precise estimates for the potential number of infected cows in a herd. The developed models in this study could be applied to control and eradication programs for BLV as cost-effective tools.