Abstract

Our objective was to estimate, using a generalised linear mixed-model approach, the sensitivity and specificity of an indirect ELISA when used to identify dams pregnant with persistently bovine viral-diarrhoea virus (BVDV)-infected foetuses. Cows that had been tested for antibodies to BVDV with a positive result during their pregnancy and where the offspring had been tested for both antibody and virus were identified by accessing the Swedish BVD database and the official pedigree records. The resulting data set consisted of 2162 cow-calf pairs in 126 herds, of which 281 included virus-positive calves. The sensitivities and specificities at 12 different decision thresholds (corresponding to optical densities (ODs) between 0.5 and 1.6) were estimated using generalised linear mixed models (binomial error, logit link), in which the gold standard (the BVDV status of the calf) was included as a covariate. In each model, the dependent variable was the dichotomous test result at the decision threshold in question. There was a significant positive interaction between the calf's status and gestational stage in all 12 models - indicating that the sensitivity and specificity at any given decision threshold was improved when the the test was performed later in pregnancy. The test should be applied only when samples have been taken in late gestation - not before the seventh month in pregnancy. If applied during the last months of pregnancy, the point estimate of the sensitivity ranges between 0.94 and LO as the decision threshold is moved from 1.0 and downwards to 0.7. Similarly, the specificity ranges between 0.39 and 0.67 as the decision threshold is moved from 0.8 and upwards to I. I. (C) 2001 Elsevier Science B.V. All rights reserved.

Keywords

BVDV; sensitivity; specificity; logistic regression; random effects

Published in

Preventive Veterinary Medicine
2001, volume: 51, number: 3-4, pages: 199-214
Publisher: ELSEVIER SCIENCE BV

Authors' information