Using Bayesian network modelling to untangle farm management risk factors for bovine viral diarrhoea virus infection

Abstract

Understanding risk factors for bovine viral diarrhoea (BVD) transmission is important for planning national disease control programmes. However, traditional statistical approaches may miss important features of BVD epidemiology due to the highly correlated nature of many farm-level risk factors. In this cross-sectional study, we used data collected from 304 cattle herds in New Zealand during 2015/2016 to compare the results from multivariable logistic regression with Bayesian network (BN) analysis. Blood samples from 15 heifers from each farm were pooled and analysed with an antibody ELISA test to classify BVD virus exposure status. Farmers were surveyed about their general management practices, knowledge about BVD, and risk factors for disease transmission, including onto- and off-farm movements, within- and between-farm contacts, and whether they implemented BVD control measures for their service bulls. Multiple imputation was used to infer missing values in the dataset prior to statistical analysis. The results showed that 57/116 (49.1%) beef farms and 95/188 (50.5%) dairy farms were likely to be actively infected with BVD virus. Almost 60% of farms had movements of heifers/cows onto the premises and 13.8% of farmers reported contact with cattle from other farms. The results of the multivariable logistic regression showed that farms where heifers/cows had been moved onto the premises during all or most of the past five years were at higher risk of being BVD seropositive than farms without those movements (OR 2.21, 95% CI 1.29-4.24). Farms where cattle had occasional or rare contacts with cattle on other farms were also at increased risk compared with farms without any animal contacts between farms (OR 2.63, 95% CI 1.33-5.41) although this association was not frequency-dependent. Only close animal contacts between farms was directly associated with BVD status in the BN model, however, this approach further untangled other complex associations between correlated management factors, and provided additional important insights into BVD epidemiology. Compared to other countries with intensive production systems, over the fence contact appeared to play a more important role in New Zealand pastoral-based production systems and should be considered when developing strategies for a national BVD control programme.

Keywords: Bayesian network analysis (BN); Bovine viral diarrhoea; Disease transmission; Farm management; Logistic regression; Risk factor analysis.