Predicting Positive ELISA Results in Dairy Herds with a Preferred Status in a Paratuberculosis Control Program

Abstract

Dairy herds participating in the Dutch milk quality assurance program for paratubercu-losis are assigned a herd status on the basis of herd examinations by ELISA of individual serum or milk samples, followed by an optional confirmatory fecal PCR. Test-negative herds are assigned Status A; the surveillance of these herds consists of biennial herd examinations. Farmers falsely believing that their Status A herds are Map-free may inadvertently refrain from preventive measures. Therefore, we aimed to develop a predictive model to alert Status A farmers at increased risk of future positive ELISA results. Using data of 8566 dairy herds with Status A in January 2016, two logistic regression models were built, with the probabilities of ≥1 or ≥2 positive samples from January 2017–June 2019 as dependent variables, and province, soil type, herd size, proportion of cattle born elsewhere, time since previous positive ELISA results, and the 95th percentile of the S/P ratios in 2015–2016, as explanatory variables. As internal validation, both models were applied to predict positive ELISA results from January 2019–June 2021, in 8026 herds with Status A in January 2019. The model predicting ≥1 positive sample had an area under the receiv-er-operating-characteristics curve of 0.76 (95% CI: 0.75, 0.77). At a cut-off predicted probability πc = 0.40, 25% of Status A herds would be alerted with positive and negative predictive values of 0.52 and 0.83, respectively. The model predicting ≥2 positive samples had lower positive, but higher negative, predictive values. This study indicates that discrimination of Status A herds with high and low risks of future positive ELISA results is feasible. This might stimulate farmers with the highest risks to take additional measures to control any undetected Map infections.