We collected AMU and AMR information at the farm level through a network of poultry veterinarians (n = 17) who were assigned to producers (n = 97–147, depending on the year) in the 5 major poultry-producing provinces of Canada: British Columbia, Alberta, Saskatchewan, Ontario, and Québec (14). Participating producers signed an informed consent form, which was administered by the veterinarian. We obtained information on farm-level AMU and farm demographics by using a questionnaire and collected fecal samples for bacterial recovery and antimicrobial-susceptibility testing. We collected samples according to the formula for detection of AMR in a population of >1,000 individuals (n = ln α / ln [1 – minimum expected prevalence]; α = 0.05) (15), according to the routine CIPARS/FoodNet Canada farm sampling strategy. We divided each barn from each farm in 4 quadrants, and we collected 10–15 fresh fecal droppings from each quadrant. We pooled the samples from each quadrant and selected randomly 1 isolate per pooled sample for all E. coli, Salmonella, and Campylobacter for further analysis. Each year, we sampled 1 flock of preharvest broilers (>30 days old) that had been randomly selected from each production unit. We administered questionnaires to record flock characteristics, including hatchery or province and country of origin of the hatching eggs or chicks, breed, production system (conventional or antimicrobial-free), age, and estimated weight of birds at preharvest sampling. We collected detailed AMU information, including the quantity of antimicrobial active ingredients administered, routes of administration (in ovo or subcutaneous injections at the hatchery, feed, and water) and primary reasons for use of antimicrobial (prophylaxis, growth promotion, or disease treatment). We also collected information on biosecurity, health status, and vaccination history (questionnaires were published elsewhere [16] as supplemental material).

When an isolate of each bacterial species of interest (Salmonella, E. coli and Campylobacter) was identified, we saved that isolate and tested it for susceptibility. We conducted antimicrobial-susceptibility testing by using routine CIPARS methodology (14). We performed automated broth microdilution by using Sensititre (ThermoFisher Scientific, https://www.thermofisher.com) using the CMV4AGNF panel for Salmonella and E. coli and the CAMPY plates for Campylobacter. Plate configurations were designed by the US National Antimicrobial Resistance Monitoring System. We applied Clinical and Laboratory Standards Institute breakpoint guidelines (17,18) (Appendix Table 1). According to routine CIPARS/National Antimicrobial Resistance Monitoring System methods, we classified isolates with intermediate susceptibility patterns as susceptible. According to CIPARS AMR testing methods, we sued no selective media in this study.

The number of antimicrobial classes each isolate was resistant to (nC) was the main outcome in the regression models. We evaluated the effect of covariates on the nC by using a 2-step procedure. First, we used a LASSO regression to select a subset of risk factors to be included in the generalized models (Appendix Table 2). Second, we ran a mixed-effect model with veterinarian and flock identification as random effects in all models. We cross-validated the models by dividing the dataset into 3 validation sets.

The term “ideal method for cleaning and disinfection” refers to the method recommended by the World Organisation for Animal Health (OIE) (19) aimed at reducing infectious pathogens in animal premises. This method consists of dry cleaning (i.e., removing of all equipment and brushing and scraping all surfaces), followed by a warm water (60°C) wash and application of a disinfectant to reduce microbial populations and carry over of pathogens to the next production cycle. For production system categories, the term “antimicrobial-free”(in contrast with “conventional” refers to farms that were not exposed to nationally defined medically important antimicrobials (20) or farms that have a reduced AMU program (i.e., one that may allow use of chemical coccidiostats, according to guidelines [21], or ionophores). We estimated AMU at the flock level in milligrams of antimicrobial active ingredient per kilogram broiler chicken biomass (mg/kg) by summing of all antimicrobials reportedly used in the flock from all routes of administration and dividing by the live animal biomass (e.g., birds at risk multiplied by the average preslaughter live weight) (22).

We compared the model fit between models by using the Akaike information criteria and the likelihood ratio test. We performed post hoc pairwise testing of mean flock differences in nC among groups of disinfection method, use of antimicrobials at the hatcheries, year, and province by using Tukey’s multiple comparison test.

We quantified the trends of antimicrobial use (Appendix Figures 2–4 for Salmonella, 8–10 for E. coli, and 14–16 for Campylobacter), and the association between resistance for individual antimicrobial classes (Appendix Figures 5–7 for Salmonella, 11–13 for E. coli, and 17–19 for Campylobacter) by using mixed-effect logistic regression models for each bacterial species. We conducted all statistical analysis in RStudio 1.2.5033 (https://www.rstudio.com) and defined statistical significance as p<0.05.

For Salmonella, the nC an isolate was resistant to in 2018 was 0.9 times lower than the nC an isolate was resistant to in 2013 (p<0.001); however, the nC an isolate was resistant to in 2019 was 1.6 times higher than in 2013 (p = 0.045), given that other variables were held constant in the model. In individual provinces, compared with the value for Alberta, the nC an isolate was resistant to was 1.7 times higher in British Columbia (p = 0.007), 1.8 times higher in Ontario (p = 0.002), 3.8 times higher in Québec (p<0.001), and 1.9 times higher in Saskatchewan (p = 0.009). For every 1-unit increase in antimicrobial injected in ovo (mg/kg) in the hatcheries, the national nC an isolate was resistant to increased by 3.4 (p = 0.02). Posthoc (Tukey test) showed that Ontario (p = 0.015) and Québec (p<0.001) had a significantly higher mean nC that an isolate was resistant to compared with Alberta; Québec also had a significantly higher mean nC that an isolate was resistant to than British Columbia and Ontario across all years (p<0.001 for both provinces) (Table 1). The antibiotic-free flocks (n = 286) were not different from conventional flocks (n = 1,612) in the nC an isolate was resistant to (Table 1). However, prevalence of Salmonella Heidelberg was statistically significantly higher at conventional farms (Appendix Figure 1). Using the ideal method of disinfection, which that entails dry and wet cleaning followed by the application of a disinfectant, was not a significant factor in the nC a Salmonella isolate was resistant to. However, significantly higher prevalence of Salmonella Heidelberg and Kentucky (Appendix Figure 1) was found in flocks that did not use the ideal method of disinfection.

For E. coli, nationally, during 2018 and 2019, the nC an isolate was resistant to was 0.9 (in 2018, p = 0.015) and 0.8 (in 2019, p<0.001) times lower than the nC an isolate was resistant to in 2013 after controlling for other variables (Table 2). The nC an isolate was resistant to was 1.2 times higher in British Columbia (p = 0.032) and 1.4 times higher in Québec (p<0.001) than the nC an isolate was resistant to in Alberta; in Saskatchewan, the nC an isolate was resistant to was 0.5 times lower than in Alberta (p<0.001). Posthoc (Tukey test) examination demonstrated that the provinces of British Columbia, Ontario, Québec, and Saskatchewan had a significantly higher mean nC an isolate was resistant to compared with Alberta; Québec also had a significantly higher mean nC an isolate was resistant to than the means for British Columbia and Ontario. In 2019, we observed a significantly lower nC an isolate was resistant to than in 2013 (p = 0.002), 2014 (p = 0.002), 2015 (p = 0.012), and 2016 (p = 0.014) (Table 2). The antibiotic-free status of the flock and ideal method of disinfection were not significant factors in the nC to which an E. coli isolate was resistant.

For Campylobacter, in 2016, the nC to which an isolate was resistant was 0.4 times lower than the nC for 2013, given that other variables were held constant in the model (p = 0.03). Posthoc (Tukey test) comparison shows that 2016 (p = 0.008) and 2018 (p = 0.037) had a significantly lower mean nC to which an isolate was resistant than the value for 2015 (Table 3). The antibiotic-free status of the flock and ideal method of disinfection were not significant factors in the nC to which a Campylobacter isolate was resistant.

Prevalence of resistance remained <15% (Appendix Table 1) for 10 of 13 tested antimicrobials for Salmonella isolates (n = 1,898), 7 of 13 tested antimicrobials for E. coli isolates (n = 3,671), and 5 of 8 tested antimicrobials for Campylobacter isolates (n = 769). The prevalence of Salmonella isolates resistant to tetracycline was 44.7% (95% CI 42.5%–46.9%) and to streptomycin was 43.6% (95% CI 41.3%–45.8%) (Appendix Table 1). Moreover, prevalence of E.coli isolates resistant to tetracycline was 46.8% (95% CI 45.2%–48.4%), to streptomycin was 46.3% (95% CI 44.7%–47.9%), to sulfisoxazole was 39.4% (95% CI 37.8%–41.0%), to ampicillin was 40.5% (95% CI 38.9%–42.1%), to gentamicin was 18.4% (95% CI 17.2%–19.7%), and to trimethoprim/sulfamethoxazole was 16.1% (95% CI 14.9%–17.3%) (Appendix Table 1). The prevalence of Campylobacter isolates resistant to tetracycline was 38.8% (95% CI 35.3%–42.2%), to ciprofloxacin was 16.5% (95% CI 13.9%–19.1%), and to nalidixic acid was 16.4% (95% CI 13.8%–19.0%) (Appendix Table 1).

For Salmonella, we observed a significant decrease in the mean resistance rates across all antimicrobial drugs included in the panel (1.8%), as well as individually to to cefoxitin (11.8%), amoxicillin/clavulanic acid (15.3%), ceftriaxone (15.3%), and ampicillin (15.9%) during 2013–2019. However, AMR rose significantly in streptomycin (18.8%) and tetracycline (19.7%) during the same period (Figures 1, 2). For E. coli, we observed a significant decrease in resistance overall (11.7%), as well as individually to tetracycline (11.4%), cefoxitin (25.4%), amoxicillin/clavulanic acid (25.7%), ceftriaxone (24.5%), and ampicillin (29.9%), whereas resistance to gentamicin (3.8%) and nalidixic acid (3.6%) increased (Figures 1, 2). For Campylobacter, we observed a significant decrease in overall (3.7%) resistance and to tetracycline (37.6%), but we observed a significant increase in nalidixic acid resistance (2.8%) (Figures 1, 2).

In flocks where Salmonella was isolated, we observed a significant decrease in overall AMU, use of lincosamide-aminocyclitol combinations, and use of third-generation cephalosporins through injection (in ovo or subcutaneous routes) during 2013–2019 (Figures 3, 4; Appendix Figures 3, 4). For feed, we observed a statistically significant decrease in the use of macrolides, penicillins, streptogramins, but we observed a significant increase in the use of orthosomycins (Figures 3, 4; Appendix Figure 4). In flocks where E. coli was isolated, we observed a significant decrease in injectable antimicrobials during 2013–2019 (Figures 3, 4; Appendix Figure 8). We observed a decrease in the use of penicillins and streptogramins and an increase in the use of bacitracins and orthosomycins through feed over time (Figures 3, 4; Appendix Figure 10). In flocks where Campylobacter was isolated, we observed a significant decrease in overall injectable antimicrobials during 2013–2019 (Figures 3, 4; Appendix Figure 14). For feed, we observed a decrease in the use of macrolides, penicillins, streptogramins, and a significant increase in the use of bacitracins and orthosomycins (Figures 3, 4; Appendix Figure 16).

Flocks from which multidrug-resistant (MDR) Salmonella was isolated (n = 79 of 604 total flocks) had significantly higher median overall AMU compared with flocks where no MDR Salmonella was identified. Specifically, MDR flocks had significantly higher use of injectable lincosamide-aminocyclitol combinations (Figure 5; Appendix Figure 5), penicillins through water (Figure 5; Appendix Figure 6), and penicillins and tetracyclines through feed (Figure 5; Appendix Figure 7). Flocks from which MDR E. coli was isolated (n = 444/928) also had significantly higher median overall AMU. Most important, these flocks had significantly higher use of lincosamide-aminocyclitol combinations in ovo or subcutaneously at the hatcheries (Figure 5; Appendix Figure 11); tetracyclines, aminoglycosides, and penicillins through water (Figure 5; Appendix Figure 12); and penicillins, trimethoprim/sulfonamide combinations, bacitracins, and tetracyclines through feed (Figure 5; Appendix Figure 13). Flocks from which MDR Campylobacter was isolated (n = 30/218) also had significantly higher median overall AMU. Specifically, these flocks had significantly higher use of injectable lincosamides (Figure 5; Appendix Figure 17); used significantly more aminoglycosides and penicillins through water (Figure 5; Appendix Figure 18); and used significantly more macrolides, penicillins, streptogramins, trimethoprim/sulfonamide combinations, and bacitracins through feed (Figure 5; Appendix Figure 19).