Conventional and unsupervised artificial intelligence analyses identified risk factors for antimicrobial resistance on dairy farms in the province of Québec, Canada.

Antimicrobial resistance (AMR) is one of the greatest threats to global health worldwide and is threatening not only humans, but also animal production systems, including dairy farms. The objective of this paper was to describe risks factors associated with AMR on dairy farms in Québec, Canada. This observational cross-sectional study included 101 commercial dairy farms and took place over a one-year period between the spring of 2017 and the spring of 2018. We explored risk factors such as farm practices and producer knowledge (measured using a questionnaire), antimicrobial use (quantified using veterinary invoices), and the presence of Salmonella Dublin (tested by serology). We evaluated AMR with fecal Escherichia coli retrieved from preweaning calves and lactating cows using the following outcomes: the presence of extended-spectrum β-lactamase/AmpC resistance and the number of resistances to antimicrobial classes. We used logistic regression models to evaluate the association between each risk factor and the 2 outcomes for the 2 types of samples (preweaning calves and lactating cows). Furthermore, we explored the relationships between these risk factors utilizing data dimensionality reduction and hierarchical clustering. Outputs of these analyses were used as regressors for AMR in regression models. Although the results for univariate analyses were ambiguous, the unsupervised analysis naturally categorized the sample of farms according to their health and treatment status (dimension 1, explaining 12.9% of the variance) and herd size (dimension 2, explaining 7.8%). Three clusters of farms were identified (cluster 1: mainly healthy herds and low ceftiofur users, cluster 2: relatively high ceftiofur users, cluster 3: farms with a higher incidence of diseases and higher antimicrobial treatment rates). Dimension 1 and cluster membership were statistically associated with the presence of extended-spectrum β-lactamase/AmpC resistance in lactating cows and in preweaning calves. Dimension 1 was also statistically associated with the number of resistances in lactating cows and in preweaning calves. This study highlights the complexity of analyzing risk factors associated with AMR. Our results suggest that the herd health status and the antimicrobial use-related practices used are associated with AMR in dairy farms. However, prospective studies are needed to confirm a causal relation.