Machine-learning algorithms to identify key biosecurity practices and factors associated with breeding herds reporting PRRS outbreak.

Investments in biosecurity practices are made by producers to reduce the likelihood of introducing pathogens such as porcine reproductive and respiratory syndrome virus (PRRSv). The assessment of biosecurity practices in breeding herds is usually done through surveys. The objective of this study was to evaluate the use of machine-learning (ML) algorithms to identify key biosecurity practices and factors associated with breeding herds self-reporting (yes or no) a PRRS outbreak in the past 5 years. In addition, we explored the use of the positive predictive value (PPV) of these models as an indicator of risk for PRRSv introduction by comparing PPV and the frequency of PRRS outbreaks reported by the herds in the last 5 years. Data from a case control study that assessed biosecurity practices and factors using a survey in 84 breeding herds in U.S. from 14 production systems were used. Two methods were developed, method A identified 20 variables and accurately classified farms that had reported a PRRS outbreak in the previous 5 years 76% of the time. Method B identified six variables which 5 of these had already been selected by model A, although model B outperformed the former model with an accuracy of 80%. Selected variables were related to the frequency of risk events in the farm, swine density around the farm, farm characteristics, and operational connections to other farms. The PPVs for methods A and B were highly correlated to the frequency of PRRSv outbreaks reported by the farms in the last 5 years (Pearson r = 0.71 and 0.77, respectively). Our proposed methodology has the potential to facilitate producer's and veterinarian's decisions while enhancing biosecurity, benchmarking key biosecurity practices and factors, identifying sites at relatively higher risk of PRRSv introduction to better manage the risk of pathogen introduction.