Application of a semi-dependent latent model in the Bayesian estimation of the sensitivity and specificity of two faecal culture methods for diagnosis of paratuberculosis in sub-clinically infected Greek dairy sheep and goats

In this study, we compared the frequency of isolation of Mycobacterium avium subsp. paratuberculosis (MAP) from faecal samples grown on Herrold's egg-yolk medium (HEYM) or on Lowenstein–Jensen (LJ) medium and estimated the sensitivity (Se) and specificity (Sp) of the methods separately in sub-clinically infected Greek dairy sheep and goats, using latent-class models and Bayesian estimation procedures. Faecal and blood samples were collected from 400 animals ≥1 year old in April–May 2002. The HEYM supported growth of MAP better than the LJ method and their agreement was very poor (weighted kappa = 0.062 (95% CI: −0.098, 0.222)). There was no evidence of dependence between the Ses whereas the Sps were positively correlated. Thus, a semi-dependent model that assumed independence of Ses and accounted for the dependence of Sps was adopted. Under this model, the parallel interpretation of the results of the two methods gave median estimates and 95% credible intervals (CrIs) for Separ, Sppar of 15% (CrIs: 3, 45%), 96% (92, 98%) in sheep and 16% (6, 36%) and 97% (94, 99%) in goats.