Estimation of the effect of misclassifications on diagnostic test performance in two persistent bovine viral infections

Abstract

Validation of diagnostic assay performance is hampered where no gold standard exists. Bayesian estimates of the sensitivity and specificity were applied to studies of two persistent bovine infections: bovine immunodeficiency virus (BIV), and bovine viral diarrhea virus (BVDV). BVDV pathogenesis is well described. Losses of $20 million in 1992 alone led the Danish government to mandate eradication. Two enzyme-linked immunosorbent assay (ELISA) tests were developed; one antibody-based to detect exposure, one antigen-based to detect persistent infection. Bayesian estimates of the sensitivity and specificity were 96.6% and 99.2%, and 98.2% and 99.8% for the antibody and antigen ELISA, respectively. Maximum Likelihood estimates of the sensitivity and specificity were 96.3% and 100% and 97.9% and 99.9% for the antibody and antigen ELISA, respectively. Estimates of diagnostic test performance by quarter are reported. A high prevalence of BIV has been reported in the southern U.S., but BIV ‘s effect on production and present assay reliability is still undetermined. Using Bayesian estimation, the performance of an immunofluorescence (IFA) assay (sensitivity = 60%, specificity = 88%) and a polymerase chain reaction (PCR) (sensitivity = 80%, specificity = 86%) for BIV detection in two herds with varying estimated BIV prevalence (20%, 71%) was estimated. Although PCR was more sensitive for diagnosing BIV infection, substantial misclassification of infection is possible regardless of which assay was used. Past research into the pathogenesis of BIV is shrouded in misclassification errors. To evaluate the ability of BIV to shift immune responses to a type 2 cytokine response similar to HIV, we measured the effects of bovine herpes virus 1 (BHV1) vaccination on a cohort of 89 lactating cows, all infected with bovine leukemia virus. BIV negative animals had a more appropriate immune response with decreasing IgG1:IgG2 when compared with BIV positive animals. Using Bayesian estimates of IFA performance, the effect of possible misclassification of BIV serostatus on study results was evaluated. Bayesian estimates are useful where no gold standard exists. More sensitive and specific diagnostic tests for BIV need to be developed. Methods for modeling epidemiological relationships that can adjust for misclassifcation errors are needed.