Highlights
- •
Weaned pig isolates had lesser associations with MIC values and both enrofloxacin or ceftiofur compared to older pig isolates.
- •
Isolates from farms with greater antibiotic purchases had greater associations with MIC values than farms with lesser purchases.
- •
Age, time since last antibiotic treatment, and disease status are important confounders that must be assessed.
Abstract
Antimicrobial resistance is considered a global One Health threat. Controlling selection pressure by reducing antibiotic use in livestock is a significant component of the response to this threat. The science concerning use and resistance is complicated and affected by time from antibiotic exposure, changing bacterial fitness, and varies by drug and pathogen. From May 2020 through October 2023, we collected intestinal (substandard and sick pigs) and fecal swab (healthy pig) samples at breed-to-wean (BTW) and wean-to-market (WTM) swine production sites and isolated E. coli bacteria. Antibiotic susceptibility testing was performed on these isolates to determine minimum inhibitory concentrations (MIC) for ceftiofur and enrofloxacin. Monthly antibiotic purchase data were used to calculate the active milligrams of drug purchased and these were divided by the kilograms of pigs produced from a farm site to provide a mass-adjusted proxy metric for farm-level antibiotic use. The relationship between use and MIC was then evaluated using a variety of multivariable statistical models. Across multiple modeling approaches, both farm type (i.e., BTW versus WTM) and farm-level antibiotic use maintained statistically significant relationships relative to E. coli MIC values for each respective drug. Use of ceftiofur and enrofloxacin can lead to increased MIC values among E. coli over time. The reasons for antibiotic purchases were not tracked as part of this project. Future work should evaluate the age of the individual pig and the time from last treatment when sampling these animals to separate out the group from individual-level effects of antibiotic use.
Keywords
Swine
Phenotypic
Antibiotic resistance
E. coli
Antibiotic use
Statistical modeling
1. Introduction
Antimicrobial resistance (AMR) is considered a threat to human and animal health on a global scale (World Health Organization, 2015). Numerous studies have investigated the relationship between antibiotic use and resistance in pathogens that could be transmitted between animals and humans (Chadwick et al., 1996, Hummel et al., 1986, Klare et al., 1995a, Klare et al., 1995b, Schouten et al., 1997, Witte, 2000). The suggestions to manage AMR through reduction of antibiotic use in food-producing animals is premised on the concept that harboring antimicrobial resistance genes exacts a fitness cost to the bacteria in the absence of antibiotic use; therefore, by reducing or removing antibiotic selection pressure, bacteria will gain a competitive advantage by losing resistance genes (Andersson, 2003, Penkova and Raymond, 2024). However, this fitness and resistance relationship is complex, and there are other factors that influence whether or not resistance is related to bacterial fitness (Anderson et al., 2023). Certainly, the relationship between antibiotic use and resistance is complicated and is expected to vary by drug class, resistance mechanism, and bacterial pathogen.
Escherichia coli bacteria are a highly diverse species of gram-negative enteric bacteria that can act as commensal, pathogenic, or zoonotic organisms under different contexts. E. coli and other Enterobacteriaceae can be associated with extraintestinal infections in humans and it is believed that at least some of these strains originally colonized the human host from a foodborne source (Manges and Johnson, 2012), even though the link between AMR genes in bacterial strains from livestock and pathogenic human strains remains unclear (Ludden et al., 2019). Nonetheless, E. coli harboring resistance against critically important antibiotics such as third-generation cephalosporins and fluoroquinolones were identified as a human health threat for extra-intestinal E. coli infections (World Health Organization, 2019). Furthermore, since non-pathogenic E. coli also carry circulating AMR genes, they are considered both a reservoir of, and an indicator for, AMR resistance in a given bacterial population (European Food Safety Authority and European Centre for Disease Prevention and Control, 2012). Therefore, in this paper we evaluated E. coli isolates as indicators of resistance in commercially raised pigs.
Antibiotic use and its relation to resistance is a highly researched topic without consistent results across all antimicrobial drugs or classes, including in many studies involving swine. One study evaluated the impact of injectable ceftiofur and penicillin use in pigs and found that use during the finishing period did not impact AMR gene distribution (Gaire et al., 2022). Another study showed that pigs seeded with an E. coli strain that had a plasmid bearing a β-lactamase resistance gene would shed more resistant E. coli strains after treatment with a cephalosporin (Fleury et al., 2015). Other studies showed use of antibiotics could increase resistance to some antibiotic classes but not others (Alali et al., 2009, Burow et al., 2019), or increase resistance in antibiotic classes not used, suggesting co-selection in the animal, but no difference in the AMR in isolates from the pork harvested (Lugsomya et al., 2018). Other population-based studies, captured a trend of decreasing antibiotic use and resistance in E. coli, but these authors also discuss that the data sets were small and that studies should be performed at the species level (Callens et al., 2018). Our study focuses on farm-level data rather than aggregate national or regional level data and will add considerable granularity to this discussion.
Antibiotic sensitivity tests that use broth microdilution returns the bacteria’s minimum inhibitory concentration (MIC) for an antibiotic. MIC data is hierarchical categorical data that do not necessarily have an underlying distribution. This makes choosing the appropriate statistical model challenging. Each model has various strengths and weaknesses in the generated results, such as misspecification bias and poor model fit (Michael et al., 2020, Aerts et al., 2022). Another study that used multiple statistical models to assess use and resistance, but that dichotomized isolates into resistant versus susceptible, could not define an ideal model and compared the various model results to identify trends (Noyes et al., 2016). An additional study evaluating multi-drug resistance in Salmonella isolates from poultry used generalized linear mixed models and accelerated failure time frailty models and found that the models returned different levels of sensitivity on the characteristics of the temporal trends associated with resistance and that censoring could make the models unstable (Bjork et al., 2015). Choice of statistical models is challenging when evaluating MIC data.
Globally, swine production has been estimated to represent the greatest mass of antimicrobial consumption among livestock species (Mulchandani et al., 2023), and also has the highest apparent consumption in the United States (Food and Drug Administration (FDA), 2022). Understanding the relationship between antimicrobial use (AMU) in animal and human hosts and resistance among bacteria is important and is shown to vary across bacterial species and by antibiotic class. There are few longitudinal studies in the literature evaluating the relationship and trends between AMR and AMU in commercially raised swine. The objective of this study was to assess the AMU and AMR relationship separately for ceftiofur and enrofloxacin, antibiotics commonly used in the US swine industry. This was done by evaluating multiple statistical models for agreement on significant explanatory variables of milligrams of purchased antibiotic over kilograms of pig produced, pig type (substandard versus sick), season, and farm type evaluated against the minimum inhibitory concentrations of enrofloxacin and ceftiofur from E. coli isolated from pooled pig samples.
2. Methods
2.1. Farm enrollment
Farm enterprises already enrolled in the Pipestone Antibiotic Resistance Tracker (PART) program were eligible for study inclusion. PART is an antibiotic stewardship program that tracks antibiotic purchases for swine producers and allows them to benchmark their purchases against those of other swine producers. In total, 155 farm enterprises were enrolled over the course of the study, which ran from May 2020 through October 2023. The farm enterprises included breed-to-wean (BTW), wean-to-market (WTM), and breed-to-market (BTM) enterprises. BTW farm enterprises were single sites where female pigs give birth to piglets, and piglets were raised until weaning, at which point they were sold and transported to other enterprises to be raised to harvest. WTM farm enterprises were comprised of farms where weaned piglets were raised to harvest. WTM farm enterprises often represented multiple sites, and for this project, different sites from each enterprise may have been sampled at consecutive sampling time points. BTM enterprises had both BTW and WTM sites, but only their WTM sites were sampled and included in this analysis. Between May 2020 and August 2022, two substandard pigs and two sick pigs from each site were humanely euthanized and sampled. Substandard pigs were smaller than their peers, did not have an obvious illness, and were not being treated with antibiotics at the time of euthanasia. Sick pigs were identified based on clinical signs and treatment history by the sampling veterinarian. Samples were pooled within each pig type. Intestine and cecum were collected into a new, resealable plastic bag. From July 2022 through October 2023, four healthy pigs per visit had fecal swabs collected using a BBL culture swab (Becton, Dickinson, and Company, Franklin Lakes, New Jersey). This change of samples from tissues to fecal swabs was made for ease of collection, based on veterinary feedback, and to prevent unnecessary pig losses, based on farmer feedback. Pigs were selected at the convenience of the veterinarian doing the sampling. Enterprises were sampled twice a year.
2.2. Laboratory testing
All samples were sent to the South Dakota State University (SDSU) Animal Disease Research and Diagnostic Laboratory (ADRDL). Internal diagnostic protocols were used by the ADRDL to process tissue and fecal swab samples. Samples were cultured onto tryptic soy agar with sheep’s blood (blood agar) and tergitol-7 agar for isolation of E. coli (Pollard, 1946, Chapman, 1947, MacFaddin, 1985). If cultures had multiple colony morphologies, one colony representing each of the morphologies was tested for antibiotic susceptibility testing (AST) using broth microdilution. Presumptive E. coli isolates were randomly selected, and their genus and species confirmed with matrix-assisted laser desorption/ionization-time of flight (MALDI-TOF) technology via the MALDI Biotyper® (Bruker, Billerica, Massachusetts). The isolates then had their antibiotic MIC determined for a set of antibiotics using the Sensititre® system (Thermofisher Scientific, Waltham, Massachusetts) with results read on the SWIN platform and data recorded to the associated software system (Thermofisher Scientific, Waltham, Massachusetts). Bovine/Porcine version 7 F veterinary plates were used (Thermofisher Scientific, Waltham, Massachusetts). The quality control strain used was the ATCC 25922 E. coli. This paper summarizes results from E. coli isolated from fecal swabs and intestine/cecum samples. For ceftiofur, the MIC values tested were 0.25, 0.5, 1, 2, 4, and 8 µg/ml and for enrofloxacin the MIC values tested were 0.125, 0.25, 0.5, 1 and 2 µg/ml.
2.3. Data management and analysis
Through a collaboration between Norsoft (Mankato, Minnesota) and SDSU ADRDL, a Javascript application was built to collate and pull laboratory data into a structured query language (SQL) database housed on the Pipestone Veterinary Services (Pipestone, Minnesota) server. A second SQL statement was used in JMP 16.2 (SAS, Cary, North Carolina) to extract data specific to this project. These data were further formatted using JMP to include farm type, farm information, and the farm identification codes used in PART. The latter allowed the laboratory data to be linked to the PART antibiotic purchase data. The data were then downloaded into Microsoft Excel (Microsoft Corporation, Redmond, Washington) and then checked against each laboratory report to ensure data accuracy and completeness. Any inconsistencies between the laboratory report and the database were discussed with SDSU ADRDL and corrected in the data set.
PART antibiotic purchase data were used to compile 12-month averages for enrofloxacin and of ceftiofur purchases by farm enterprise from monthly data collected from May 2020 through May 2021, and June 2021 through June 2022, for the first two periods, respectively. Likewise, a monthly average was compiled for the 15-month period from July 2022 through October 2023. Milligrams of active drug ingredient was summed from drug labels on the inventory for each site for each respective period. To estimate kilograms of live weight of the pigs produced at each site, we utilized average weights as follows: for BTW sites, weaned pigs were produced and each pig weighed approximately 5.44 kg; for WTM sites, finished hogs were produced with an approximate weight of 127.01 kg. Then the amount of purchased antibiotic per month was divided by the mass of pigs produced per month to create a proxy use metric.
Data were analyzed in two groups. Initially, the intestinal E. coli isolates from May 2020 through August 2022 were summarized with the fecal swab E. coli isolates from July 2022 through October 2023 summarized separately. Each data set had descriptive statistics determined for the confounding variables of pig type (substandard versus sick), farm type (BTW versus WTM), and season. Seasons were coded by three-month periods with December, January, and February as winter; March, April, and May as spring; June, July, and August as summer; and September, October, and November as fall. E. coli isolates were summarized by MIC value for ceftiofur and enrofloxacin using both frequency and percentages. These summary statistics were stratified by sample type (intestinal sample and fecal swab) and farm type (BTW and WTM). The median and range of milligrams of ceftiofur and enrofloxacin per kg of pigs produced were calculated; further, the frequency of E. coli isolated at each quartile of mg of antibiotic purchased per kg of pig produced (mg/kg) was calculated and stratified by farm type.
Univariable Kaplan Meier graphs (Pol and Ruegg, 2007, Cortinhas et al., 2013) with the MIC50 and MIC90 were created that compared the proportion of surviving E. coli isolates at each drug concentration tested in the broth microdilution for both ceftiofur and enrofloxacin. This comparison was completed among farms within a quartile of antibiotic purchased and the graphs were further stratified by farm type (BTW and WTM) and sample type (intestinal samples and fecal swabs). These relationships were assessed for significance using the log rank test. Next, a series of multivariable model structures were deployed and the agreement between models of the statistically significant explanatory variables were assessed. The models evaluated the relationship between the mg/kg of enrofloxacin and ceftiofur purchased, separately, and the MIC value for ceftiofur and enrofloxacin of E. coli isolates from the same farm enterprise while controlling for pig type, season, and farm type. Since there were no swine-specific enrofloxacin and ceftiofur breakpoints available for E. coli isolates from the intestinal tract, the data could not be reliably dichotomized for swine isolates into susceptible and resistant. MIC data were reported based on two-fold dilutions of an antibiotic concentration.
For the statistical analyses, right-censored data were defined as any MIC values greater than the highest measured MIC value for that antibiotic (for ceftiofur > 8 µg/ml and for enrofloxacin > 2 µg/ml). Left-censored data were treated as an interval between zero and the lowest MIC measured; therefore, these values were included in the lowest MIC value (for ceftiofur = 0.25 µg/ml and for enrofloxacin = 0.12 µg/ml). Interval censoring was not addressed, and each MIC value was considered as a category. Stata BE version 18.0 (Stata Corp, College Station, Texas, USA) was used in all analyses. The Stata software commands used are listed in parentheses after each model type below. The variable mg/kg was forced into the model even if non-significant (P ≥ 0.05) when confounders were shown to be significant (P < 0.05). Farm type, season, and pig type (for period 1 (May 2020 through August 2022): substandard or sick) were evaluated as confounders. Multiple model types were used to address the unique nature of MIC data, as a result, there is not an underlying distribution that defines this data well and the different models use different distributions to assess the MIC relationship to purchased antibiotic amount with different strengths and weaknesses. This results in an inability to choose the most efficient models as Akaike and Bayesian Information Criteria are only appropriately used when the underlying model structure, and thus log likelihood calculation, being compared is the same (Dohoo et al., 2012). The models used were the following (note: for the hazard models, the MIC was used as a time-related failure would normally be used).
- •
Model 1: Cox proportional hazard models (stcox) without clustered standard errors
- •
Model 2: Cox proportional hazard model with frailty at the farm-level (stcox)
- •
Model 3: Parametric proportional hazard models (streg) with a generalized gamma (ggamma) distribution without clustered standard errors
- •
Model 4: Parametric proportional hazard models (streg) with a generalized gamma (ggamma) distribution with clustered standard errors
- •
Model 5: Mixed effects parametric proportional hazard models with a ggamma distribution (mestreg),
- •
Model 6: Ordinal regression (ologit)
- •
Model 7: Continuation-ratio ordinal regression models (ocratio)
- •
Model 8: Multinomial regression (mlogit) with clustered standard errors
- •
Model 9: Multinomial regression (mlogit) without clustered standard errors
- •
Model 10: Multinomial mixed effects (xtmlogit) and
- •
Model 11: Generalized ordinal regression models (gologit2)
Any models that did not use a random effect (mixed effects models) were computed with normal standard errors and standard errors calculated using a cluster sandwich variance calculated at the farm-level. For the Cox proportional hazard models, proportional hazards were evaluated using the Schoenfeld test for proportional hazards as well as a proportional hazard plot and visually via the Kaplan-Meier plot of the model probabilities. For the ordinal regression models, proportional odds assumptions were evaluated using the Brant test. For the continuation-ratio ordinal regression the likelihood ratio test was used to assess proportional odds. Generalized ordinal and mixed effects multinomial regression models required that the MIC values be rescaled to whole numbers and MICs were scaled so that the lowest MIC was 0. Only generalized ordinal models with < 15 negative predicted probabilities were reported. Groups of dummy variables were assessed using an omnibus Wald test, and a likelihood ratio test was used for the mixed effects multinomial model a likelihood ratio test was used. These variables were included if the omnibus test was significant and excluded if it was non-significant at a level of significance of 0.05. All variables were included in the initial model and backwards selection was used to identify a final model with a level of significance of 0.05. Significant variables between models that converged and that did not have any proportional hazard or odds violations were compared to identify agreement.
3. Results
There were 1223 intestinal samples collected, and of these 1201 (98.2 %) had E. coli isolated from the sample. All but three of these samples (1198) had MIC data available for analysis. In total, 1208 fecal swab samples were taken and 1186 (98.2 %) grew E. coli. All these samples had MIC data available for analysis. Table 1 summarizes the distribution of E. coli isolates with MIC data across farm type, season, and pig type (May 2020 through August 2022 only). Of the 1198 E. coli isolates from intestinal samples, 49.8 % (597) of isolates that came from substandard pigs and 50.2 % (601) came from sick pigs with the largest proportion collected in the summer (30.5 %, 365) and the smallest proportion in the fall (16.8 %, 201). WTM sites had more isolates (59.0 %, 707) compared to BTW (41.0 %, 491). From July 2022 through October 2023, there were 1186 E. coli isolates collected from fecal swabs with 44.2 % (524) from BTW sites and 55.8 % (662) from WTM sites. Most of these isolates were collected in the spring (52.1 %, 618) and the fewest were collected in the winter (9.6 %, 114).
Table 1. Descriptive statistics of E. coli isolate distributions across confounders assessed in a model that evaluated separately the relationship between milligrams of ceftiofur and enrofloxacin purchased by the kilograms of pig produced by farm enterprises in the mid-West United States.
| Empty Cell | Intestinal sample E. coli isolates, n = 1198 | Total Intestinal Samples, n = 1223 | ||
|---|---|---|---|---|
| Empty Cell | # | % | # | % E. coli detected |
| Pig Type | ||||
| Substandard | 597 | 49.8 | 607 | 98.4 % |
| Sick | 601 | 50.2 | 616 | 97.6 % |
| Farm type | ||||
| Breed-to-wean | 491 | 41 | 495 | 99.2 % |
| Wean-to-market | 707 | 59 | 728 | 97.1 % |
| Seasons | ||||
| Winter | 352 | 29.4 | 362 | 97.2 % |
| Spring | 280 | 23.4 | 284 | 98.6 % |
| Summer | 365 | 30.5 | 370 | 98.6 % |
| Fall | 201 | 16.8 | 207 | 97.1 % |
| Empty Cell | Fecal swab E. coli isolates, n = 1186 | Total Fecal Swab Samples, n = 1208 | ||
| Empty Cell | # | % | # | % E. coli detected |
| Farm type | ||||
| Breed-to-wean | 524 | 44.2 | 529 | 99.1 % |
| Wean-to-market | 662 | 55.8 | 679 | 97.5 % |
| Seasons | ||||
| Winter | 114 | 9.6 | 121 | 94.2 % |
| Spring | 618 | 52.1 | 631 | 97.9 % |
| Summer | 282 | 23.8 | 282 | 100.0 % |
| Fall | 172 | 14.5 | 174 | 98.9 % |
Table 2 summarized data for the mg of ceftiofur and enrofloxacin purchased per kg of pigs produced from May 2020 through August 2022 (period 1) and from July 2022 through October 2023 (period 2). There were fluctuations in the use of ceftiofur between the two time periods on BTW sites; meanwhile, there was little variability elsewhere. The median use on BTW sites in period 1 was 0.72 mg of ceftiofur per kg of pig produced and in period 2 it was 2.42, representing nearly a threefold increase. Overall, the isolates were evenly distributed across the quartiles of enrofloxacin and ceftiofur purchase data, as would be expected. Differences were revealed when the isolates were stratified by farm type. On BTW sites in period 1, a plurality of isolates came from sites that had the amount of purchased antibiotics in the highest quartile for ceftiofur (58.9 %, 289) and enrofloxacin (47.9 %, 235). The same trend was observed in period 2 on BTW sites, i.e., the plurality of isolates came from sites that had antibiotic purchases in the highest quartile for ceftiofur (56.5 %, 296) and enrofloxacin (45.4 %, 238). By contrast, for WTM sites in period 1, the plurality of isolates came from sites with antibiotic purchases in the second quartile for ceftiofur (37.5 %, 265) and enrofloxacin (33.5 %, 237). In period 2 on WTM sites, isolates were commonly from sites with antibiotic purchases in the second quartile for ceftiofur (42.0 %, 278) and the third quartile for enrofloxacin (34.4 %, 228).
Table 2. Descriptive statistics of milligrams of ceftiofur and enrofloxacin purchased by the kilograms of pig and distribution of E. coli isolates across quartiles and farm type in the Midwest United States.
| May 2020 through August 2022 | ||||||
|---|---|---|---|---|---|---|
| Empty Cell | Overall | Breed-to-Wean | Wean-to-Market | |||
| Empty Cell | Median | Range | Median | Range | Median | Range |
| Ceftiofur (mg/kg) | 0.06 | 0, 21.31 | 0.72 | 0, 21.31 | 0.03 | 0, 0.42 |
| Enrofloxacin (mg/kg) | 0.33 | 0, 11.20 | 0.66 | 0, 11.2 | 0.24 | 0, 1.82 |
| Empty Cell | # | % | # | % | # | % |
| Milligrams ceftiofur purchased versus kilogram pig produced | ||||||
| ≤ 25th percentile (≤ 0.005 mg/kg) | 300 | 25 | 77 | 15.7 | 223 | 31.5 |
| > 25th to 50th percentile (≤ 0.059 mg/kg) | 301 | 25.1 | 36 | 7.3 | 265 | 37.5 |
| > 50th to 75th percentile (≤ 0.333 mg/kg) | 300 | 25 | 89 | 18.1 | 211 | 29.8 |
| > 75th to 100th percentile (≤ 21.311 mg/kg) | 297 | 24.8 | 289 | 58.9 | 8 | 1.1 |
| Milligrams enrofloxacin purchased versus kilogram pig produced | ||||||
| ≤ 25th percentile (≤ 0.116 mg/kg) | 303 | 25.3 | 112 | 22.8 | 191 | 27 |
| > 25th to 50th percentile (0.330 mg/kg) | 296 | 24.7 | 59 | 12.0 | 237 | 33.5 |
| > 50th to 75th percentile (0.765 mg/kg) | 301 | 25.1 | 85 | 17.3 | 216 | 30.6 |
| > 75th to 100th percentile (≤ 11.205 mg/kg) | 298 | 24.9 | 235 | 47.9 | 63 | 8.9 |
| July 2022 through October 2023 | ||||||
| Empty Cell | Overall | Breed-to-Wean | Wean-to-Market | |||
| Empty Cell | Median | Range | Median | Range | Median | Range |
| Ceftiofur (mg/kg) | 0.09 | 0, 9.00 | 2.42 | 0, 9.00 | 0.03 | 0, 1.05 |
| Enrofloxacin (mg.kg) | 0.38 | 0, 8.85 | 0.68 | 0, 8.85 | 0.28 | 0, 3.51 |
| Empty Cell | # | % | # | % | # | % |
| Milligrams ceftiofur purchased versus kilogram pig produced | ||||||
| ≤ 25th percentile (≤ 0.017 mg/kg) | 295 | 24.9 | 51 | 9.7 | 244 | 36.9 |
| > 25th to 50th percentile (≤ 0.087 mg/kg) | 294 | 24.8 | 16 | 3.1 | 278 | 42 |
| > 50th to 75th percentile (≤ 2.110 mg/kg) | 301 | 25.4 | 161 | 30.7 | 140 | 21.1 |
| > 75th to 100th percentile (≤ 9.002 mg/kg) | 296 | 25 | 296 | 56.5 | 0 | 0 |
| Milligrams enrofloxacin purchased versus kilogram pig produced | ||||||
| ≤ 25th percentile (≤ 0.164 mg/kg) | 299 | 25.2 | 129 | 24.6 | 170 | 25.7 |
| > 25th to 50th percentile (0.385 mg/kg) | 294 | 24.8 | 91 | 17.4 | 203 | 30.7 |
| > 50th to 75th percentile (0.988 mg/kg) | 294 | 24.8 | 66 | 12.6 | 228 | 34.4 |
| > 75th to 100th percentile (8.851 mg/kg) | 299 | 25.2 | 238 | 45.4 | 61 | 9.2 |
Table 3 summarizes the enrofloxacin and ceftiofur MIC distributions for isolates evaluated in this study stratified by farm type, pig type, and period. For ceftiofur on BTW sites in period 1, substandard pigs had a higher percentage of isolates at the highest MIC value (28.5 %) compared to sick pigs (24.9 %). On WTM sites, the reverse was true, 25.2 % of isolates at the highest MIC value were from sick pigs compared to 23.7 % from substandard pigs. BTW sites had a higher proportion of isolates (28.5 %) at the highest MIC value among all substandard pigs, and WTM sites had highest proportion of isolates (27.3 %) among all sick pigs in period 1. In period 2, the opposite trend was seen between farm site types with the ceftiofur MIC value distribution. Only 16.2 % of BTW isolates had the highest MIC value, while 20.2 % of isolates had the highest MIC value on WTM sites. For enrofloxacin on BTW sites in period 1, sick pigs had a higher percentage of isolates at the highest MIC value (16.7 %) compared to substandard pigs (13.0 %). On WTM sites, the same was true, 40.7 % of isolates from WTM sites had the highest MIC value compared to 28.8 % from substandard pigs. WTM sites had a higher proportion of isolates (28.8 %) at the highest MIC value among all substandard pigs, and the highest proportion of isolates (40.7 %) among all sick pigs. In period 2, the same trend was seen with the enrofloxacin MIC value distribution. The WTM site isolates had 17.5 % of isolates with the highest MIC value, while 8.6 % of isolates had the highest MIC value on BTW sites.
Table 3. The ceftiofur and enrofloxacin minimum inhibitory concentration (µg/ml) distribution of intestinal and fecal swab E. coli isolates collected from June 2022 through October 2023 stratified by farm type and pig type.
| Empty Cell | May 2020 through August 2022 | July 2022 through October 2023 | May 2020 through August 2022 | July 2022 through October 2023 | ||||
|---|---|---|---|---|---|---|---|---|
| Empty Cell | Breed-to-Wean (intestinal isolates) | Wean-to-Market (intestinal isolates) | ||||||
| Empty Cell | Substandard | Sick | Substandard | Sick | ||||
| Empty Cell | Ceftiofur | Enrofloxacin | Ceftiofur | Enrofloxacin | Ceftiofur | Enrofloxacin | Ceftiofur | Enrofloxacin |
| MIC values | # (%) | # (%) | # (%) | # (%) | # (%) | # (%) | # (%) | # (%) |
| 0.12 | — | 195 (72.5) | — | 192 (73.0) | — | 242 (51.6) | — | 239 (49.0) |
| 0.25 | 30 (11.2) | 5 (1.9) | 36 (13.7) | 3 (1.1) | 48 (10.2) | 16 (3.4) | 35 (7.2) | 17 (3.5) |
| 0.5 | 115 (42.8) | 14 (5.2) | 112 (42.6) | 16 (6.1) | 158 (33.7) | 39 (8.3) | 154 (31.6) | 29 (5.9) |
| 1 | 27 (10.0) | 16 (6.0) | 22 (8.4) | 8 (3.0) | 74 (15.8) | 39 (8.3) | 96 (19.7) | 24 (4.9) |
| 2 | 4 (1.5) | 7 (2.7) | 2 (0.8) | 2 (0.8) | 38 (8.1) | 19 (4.1) | 40 (8.2) | 21 (4.3) |
| > 2 | — | 32 (11.9) | — | 42 (16.0) | — | 114 (24.3) | — | 158 (32.4) |
| 4 | 6 (2.2) | — | 3 (1.1) | — | 8 (1.7) | — | 11 (2.3) | — |
| 8 | 16 (6.0) | — | 25 (9.5) | — | 35 (7.5) | — | 29 (5.9) | — |
| > 8 | 71 (26.4) | — | 63 (24.0) | — | 108 (23.0) | — | 123 (25.2) | — |
| Empty Cell | Breed-to-Wean (fecal swab isolates) | Wean-to-Market (fecal swab isolates) | ||||||
| Empty Cell | Ceftiofur | Enrofloxacin | Ceftiofur | Enrofloxacin | ||||
| MIC values | # (%) | # (%) | # (%) | # (%) | ||||
| 0.12 | — | 387 (72.7) | — | 482 (50.3) | ||||
| 0.25 | 66 (12.4) | 8 (1.5) | 83 (8.7) | 33 (3.4) | ||||
| 0.5 | 227 (42.7) | 30 (5.6) | 312 (32.6) | 68 (7.1) | ||||
| 1 | 49 (9.2) | 24 (4.5) | 171 (17.9) | 63 (6.6) | ||||
| 2 | 6 (1.1) | 9 (1.7) | 78 (8.1) | 40 (4.2) | ||||
| > 2 | — | 74 (13.9) | — | 272 (28.4) | ||||
| 4 | 9 (1.7) | — | 19 (2.0) | — | ||||
| 8 | 41 (7.7) | — | 64 (6.7) | — | ||||
| > 8 | 134 (25.2) | — | 231 (24.1) | — | ||||
Fig. 1, Fig. 2 summarize the relationship between survival of E. coli isolates at each MIC value across the different quartiles of purchased ceftiofur and enrofloxacin from the farms enrolled in the study. The relationship of quartile of antibiotic purchased and MIC values were statistically significant among BTW for isolates from both intestinal samples and fecal swabs and among WTM for isolates from intestinal samples. For BTW sites, sites that purchased the most ceftiofur (4th quartile) had isolates that had greater survival across all MIC values for both sample types (p-value<0.0001), although the MIC50 (MIC where 50 % of the isolates’ growth was inhibited) was the same for all quartiles. For WTM sites, the intestinal isolates revealed that the isolates from farms that had ceftiofur purchases in the third quartile had the greatest survival (p-value=<0.0001), and isolates from the farms that had purchases in the 3rd quartile the MIC50= 2 while all others had an MIC50= 0.5. There were only six isolates from WTM farms in the 4th quartile for isolates from intestinal isolates and there were zero isolates from WTM farms with 4th quartile purchases for isolates from fecal swabs (See Table 2). Only fecal swabs from BTW sites had a Kaplan Meier graph for enrofloxacin MIC values that was statistically significant (p-value=0.014). Isolates from farms that purchased enrofloxacin in the second and fourth quartile had greater survival than farms that had purchases in the third and first quartile. Furthermore, the isolates from farms that had purchases in the first quartile had an MIC50 = 0.5 µg/ml while all other isolates had an MIC50 = 1 µg/ml. Isolates from intestinal samples collected from BTW sites had a p-value= 0.062 and WTM sites had a p-value= 0.051, and isolates from fecal swabs collected on WTM sites had a p-value of 0.4526 when evaluated against enrofloxacin.
Fig. 1. Univariable Kaplan-Meier graphs comparing the growth of E. coli isolates from a farm and the quartile of ceftiofur antibiotics purchased among farms involved in the study. 1a: E. coli isolated from intestinal samples collected on breed-to-wean (BTW) farms (log rank test p-value<0.0001); 1b: E. coli isolated from fecal swabs collected on breed-to-wean (BTW) farms (log rank test p-value=0.0011); 1c: E. coli isolated from intestinal samples collected on wean-to-market (WTM) farms (log rank test p-value<0.0001); 1d: E. coli isolated from fecal swabs collected on wean-to-market (WTM) farms (log rank test p-value=0.9274).
Fig. 2. Univariable Kaplan-Meier graphs comparing the growth of E. coli isolates from a farm and the quartile of enrofloxacin antibiotics purchased among farms involved in the study. 1a: E. coli isolated from intestinal samples collected on breed-to-wean (BTW) farms (log rank test p-value=0.062); 1b: E. coli isolated from fecal swabs collected on breed-to-wean (BTW) farms (log rank test p-value=0.014); 1c: E. coli isolated from intestinal samples collected on wean-to-market (WTM) farms (log rank test p-value=0.051); 1d: E. coli isolated from fecal swabs collected on wean-to-market (WTM) farms (log rank test p-value=0.4526).
Across all modeling strategies, two variables were consistently significant (P < 0.05) with a consistent relationship between them and the MIC of the E. coli isolates tested (See Table 4, Table 5). These two variables were farm type (BTW versus WTM) and mg/kg of purchased ceftiofur. The only model in which mg/kg was not statistically significant was in the period 2 multinomial regression models. All other models that converged or that did not have assumptions violated had both variables as significant.
Table 4. Summary table of models’ results for the comparison of milligrams of ceftiofur purchased by the kilograms of pig produced to E. coli isolate minimum inhibitory concentrations determined by broth microdilution.
| May 2020 through August 2022 – Intestinal E. coli isolates | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Empty Cell | Cox Prop Hazard | Cox Prop Hazard Frailty | Para Survival | Para Survival Clst SE | Para Survival ME | Ord | Ord Cont Ratio | Multi-nomial | Multi- nomial Clst SE | Multi-nomial ME | Gen Ord | Gen Ord Clst SE |
| Empty Cell | P-values and model outcomes | |||||||||||
| BTW versus WTM | 0.003 | 0.005 | 0.002 | 0.002 | 0.013 | Viol-ated | Viol-ated | 0.001 | < 0.001 | NC | < 0.001 | < 0.001 |
| Season (df = 3) | Not sig | Not sig | Not sig | Not sig | Not sig | Viol-ated | Viol-ated | Not sig | Not sig | NC | Scarce data | Scarce data |
| mg/kg | < 0.001 | < 0.001 | < 0.001 | 0.002 | < 0.001 | Viol-ated | Viol-ated | 0.016 | 0.008 | NC | 0.017 | 0.008 |
| Substandard pig versus sick pig | Not sig | Not sig | Not sig | Not sig | Not sig | Viol-ated | Viol-ated | Not sig | Not sig | NC | Not sig | Not sig |
| Empty Cell | Hazard ratio (95 % CI) | Hazard ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Odds Ratio | Odds Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Odds ratio | Odds ratio |
| BTW versus WTM | 1.25 (1.08, 1.46) | 1.25 (1.07, 1.46) | 0.88 (0.81, 0.95) | 0.88 (0.81, 0.96) | 0.82 (0.70, 0.96) | — | — | Table 6 | Table 6 | — | Table 6 | Table 6 |
| Spring versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| Summer versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| Fall versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| mg/kg | 0.87 (0.83, 0.92) | 0.87 (0.83, 0.92) | 1.06 (1.04, 1.09) | 1.06 (1.02, 1.11) | 1.14 (1.09, 1.19) | — | — | Table 6 | Table 6 | — | Table 6 | Table 6 |
| Substandard pig versus sick pig | — | — | — | — | — | — | — | — | — | — | — | — |
| Empty Cell | July 2022 through October 2023 – Fecal swab E. coli isolates | |||||||||||
| Empty Cell | Cox Prop Hazard | Cox Prop Hazard Frailty | Para Surv | Para Surv Clst SE | Para Survival ME | Ord | Ord Cont Ratio | Multi-nomial | Multi-nomial Clst SE | Multi-nomial ME | Gen Ord ** | Gen Ord Clst SE** |
| Empty Cell | P-values and model outcomes | |||||||||||
| BTW versus WTM | 0.008 | 0.011 | NC | NC | 0.005 | Viol-ated | Viol-ated | 0.043 | 0.041 | Not sig | XS NPP | XS NPP |
| Season (df = 3) | Not sig | Scarce data | NC | NC | Not sig | Viol-ated | Viol-ated | Not sig | Scarce data | Not sig | XS NPP | XS NPP |
| mg/kg | 0.019 | 0.022 | NC | NC | 0.008 | Viol-ated | Viol-ated | 0.128§ | 0.127§ | Not sig | XS NPP | XS NPP |
| Empty Cell | Hazard ratio (95 % CI) | Hazard ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Odds Ratio | Odds Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Odds ratio | Odds ratio |
| BTW versus WTM | 1.26 (1.06, 1.50) | 1.26 (1.06, 1.51) | — | — | 0.76 (0.62, 0.92) | — | — | Table 6 | Table 6 | — | — | — |
| Spring versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| Summer versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| Fall versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| mg/kg | 0.95 (0.90, 0.99) | 0.95 (0.90, 0.99) | — | — | 1.07 (1.02, 1.13) | — | — | — | — | — | — | — |
The ordinal and continuation ratio ordinal regression model violated proportional odds assumptions in both periods and was not included in the table. Abbreviations were used to describe the models. “BTW” for breed-to-wean farm sites, “WTM” for wean-to-market farm sites, “Prop” for proportional, “Para” for parametric, “Clst SE” for clustered standard errors, “ME” for mixed effects, “Gen Ord” for generalized ordinal, “Rel” for relative, “Not Sig” for not significant, “XS NPP” for excess negative predicted probabilities, and “NC” for no convergence. *Final model had < 15 negative predicted probabilities. **Final model after backwards selection had 119 negative predicted probabilities. “Scarce data” indicated that there was not enough data across all levels of MIC for the model to prevent separation or model convergence. § The mg/kg variable was forced into the model but was not significant overall.
Table 5. Multinomial and generalized ordinal statistical model significant independent variable results for all levels of ceftiofur minimum inhibitory concentrations from E. coli isolates determined by broth microdilution.
| Empty Cell | Multinomial | Multinomial Clst SE | Gen Ordinal* | Gen Ord Clst SE* | ||||
|---|---|---|---|---|---|---|---|---|
| MIC | Relative risk ratio (95 % CI) | p-value | Relative risk ratio (95 % CI) | p-value | Odds ratio (95 % CI) | p-value | Odds ratio (95 % CI) | p-value |
| INTESTINAL SAMPLES | ||||||||
| 0.25 | Reference | |||||||
| 0.5 | ||||||||
| mg/kg | 1.09 (0.92, 1.29) | 0.34 | 1.09 (0.92, 1.29) | 0.342 | 1.14 (0.97, 1.34) | 0.105 | 1.14 (0.97, 1.34) | 0.101 |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.85 (0.56, 1.27) | 0.421 | 0.85 (0.56, 1.28) | 0.429 | 0.70 (0.47, 1.04) | 0.077 | 0.70 (0.47, 1.04) | 0.075 |
| 1 | ||||||||
| mg/kg | 1.25 (0.95, 1.64) | 0.115 | 1.25 (0.98, 1.59) | 0.077 | 1.23 (1.09, 1.39) | 0.001 | 1.23 (1.10, 1.38) | < 0.001 |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.31 (0.14, 0.70) | 0.005 | 0.31 (0.15, 0.67) | 0.003 | 0.45 (0.31, 0.67) | < 0.001 | 0.45 (0.29, 0.70) | < 0.001 |
| 2 | ||||||||
| mg/kg | 1.20 (0.69, 2.09) | 0.523 | 1.20 (0.79, 1.81) | 0.391 | 1.24 (1.08. 1.41) | 0.002 | 1.24 (1.09, 1.40) | 0.001 |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.09 (0.02, 0.40) | 0.002 | 0.09 (0.02, 0.37) | 0.001 | 0.53 (0.34, 0.83) | 0.006 | 0.53 (0.33, 0.87) | 0.012 |
| 4 | ||||||||
| mg/kg | 1.51 (1.13, 2.00) | 0.005 | 1.51 (1.18, 1.92) | 0.001 | 1.26 (1.10, 1.44) | 0.001 | 1.26 (1.11, 1.43) | < 0.001 |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.31 (0.10, 0.98) | 0.045 | 0.31 (0.10, 0.97) | 0.045 | 0.78 (0.48, 1.27) | 0.321 | 0.78 (0.46, 1.33) | 0.362 |
| 8 | ||||||||
| mg/kg | 1.31 (1.07, 1.60) | 0.01 | 1.31 (1.07, 1.59) | 0.008 | 1.20 (1.04, 1.39) | 0.013 | 1.20 (1.07, 1.35) | 0.002 |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.72 (0.39, 1.33) | 0.295 | 0.72 (0.40, 1.30) | 0.276 | 0.94 (0.56, 1.60) | 0.825 | 0.94 (0.56, 1.59) | 0.823 |
| FECAL SWABS | ||||||||
| 0.25 | Reference | |||||||
| 0.5 | ||||||||
| mg/kg | 1.05 (0.94, 1.16) | 0.405 | 1.05 (0.95, 1.15) | 0.337 | — | — | — | — |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.97 (0.66, 1.42) | 0.884 | 0.97 (0.66, 1.42) | 0.884 | — | — | — | — |
| 1 | ||||||||
| mg/kg | 1.15 (0.86, 1.54) | 0.34 | 1.15 (0.91, 1.47) | 0.249 | — | — | — | — |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.24 (0.08, 0.74) | 0.013 | 0.24 (0.09, 0.67) | 0.006 | — | — | — | — |
| 2 | ||||||||
| mg/kg | 1.30 (0.86, 1.98) | 0.212 | 1.30 (1.05, 1.62) | 0.017 | — | — | — | — |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.18 (0.03, 1.20) | 0.076 | 0.18 (0.05, 0.71) | 0.014 | — | — | — | — |
| 4 | ||||||||
| mg/kg | 1.26 (0.98, 1.63) | 0.073 | 1.26 (1.04, 1.54) | 0.018 | — | — | — | — |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.49 (0.15, 1.57) | 0.23 | 0.49 (0.17, 1.42) | 0.189 | — | — | — | — |
| 8 | ||||||||
| mg/kg | 1.24 (1.03, 1.50) | 0.024 | 1.24 (0.98, 1.57) | 0.071 | — | — | — | — |
| Farm type | ||||||||
| Wean-to-market | Reference | |||||||
| Breed-to-wean | 0.61 (0.27, 1.39) | 0.236 | 0.61 (0.23, 1.63) | 0.322 | — | — | — | — |
“SE” is an abbreviation for standard error. “NA” is an abbreviation for not applicable. “Gen” is for generalized. * The model had excess negative probabilities for the fecal swab samples and results were not reported.
In all models completed with ceftiofur purchases, the MIC of 0.25 µg/ml was the referent category. Although the different models assessed risk using different methods, thus resulting in different ratios or risks, the trends were similar. For the Cox proportional hazard model, the hazard ratios estimated the hazard of being susceptible (no growth), while all other models assessed the odds or risk of having an MIC value in a higher category (less susceptible to the antibiotic) than the referent category. As a result, the hazard ratios were inverted compared to all other models’ ratios and risks. For isolates from intestinal samples, in the Cox proportional hazard models with and without frailty, the BTW enterprises had a 1.25 (without frailty 95 % confidence interval (95 % CI): 1.08, 1.46; with frailty 95 % CI: 1.07, 1.46) greater hazard of susceptibility than WTM sites, and a one-unit increase in the mg/kg of purchased ceftiofur resulted in a decreased hazard ratio (HR; HR=0.87; with and without frailty 95 % CI: 0.83, 0.92) or decreased susceptibility. For isolates from fecal swabs, the same trend was seen for both models, with the BTW enterprise exhibiting a 1.26 (without frailty 95 % CI: 1.06, 1.50; with frailty 95 % CI: 1.06, 1.51) greater hazard of susceptibility than WTM, and a one-unit increase in mg/kg of purchased ceftiofur showing a lower hazard of susceptibility (HR=0.95; without and with frailty 95 % CI: 0.90, 0.99). The parametric survival models returned time ratios (TR), though MIC values were assessed rather than time. For both the parametric survival model with and without clustered standard errors (SE) on isolates from intestinal samples, BTW sites had median MIC values lower than WTM sites (TR=0.88; without clustered SE 95 % CI: 0.81, 0.95; with clustered SE 95 % CI: 0.81, 0.96). For a one unit increase in mg/kg of purchased ceftiofur, the MIC increased by 1.06 (without clustered SE 95 % CI: 1.04, 1.09; with clustered SE 95 % CI: 1.02, 1.11) as well. For the mixed effects parametric survival model, the TR for farm type (BTW versus WTM) was 0.82 (95 % CI: 0.70, 0.96) and for mg/kg of purchased ceftiofur the TR was 1.14 (95 % CI: 1.09, 1.19). For the fecal swab isolates, none of the parametric survival models without mixed effects had convergence. The mixed effects model had a TR of 0.76 (95 % CI: 0.62, 0.92) for the farm type comparison and 1.07 (95 % CI: 1.02, 1.13) for the mg/kg of purchased ceftiofur variable. For multinomial models, each level of the MIC from 0.5 to 8 µg/ml for ceftiofur was assessed against the 0.25 µg/ml MIC concentration category using a model with and without clustered standard errors. The range of significant relative risk ratios (RRR) is provided in Table 5. Among intestinal isolates, a one-unit increase in mg/kg of purchased ceftiofur resulted in higher MIC values when the MIC was 4 µg/ml (RRR=1.51; 95 % CI: 1.13, 2.00)) and the MIC was 8 µg/ml (RRR=1.31; 95 % CI: 1.07, 1.60), when compared to the referent 0.25 µg/ml MIC value. Farm type was also significantly associated with MIC values, with the model indicating a lower RRR for BTW relative to WTM enterprises at MIC values of 1 µg/ml, 2 µg/ml, and 4 µg/ml with RRRs of 0.31 (95 % CI: 0.14, 0.70), 0.09 (95 % CI: 0.02, 0.40), and 0.31 (95 % CI: 0.10, 0.98), respectively. Among fecal swab isolates, mg/kg of purchased ceftiofur was only statistically significant at the MIC of 8 µg/ml in the multinomial model without clustered SE (RRR=1.24; 95 % CI: 1.03, 1.50). The RRR comparing BTW to WTM sites was significant for both models for an MIC of 1 µg/ml (RRR=0.24; without clustered SE 95 % CI: 0.08, 0.74, with clustered SE 95 % SE: 0.09,0.67), and for an MIC of 2 µg/ml (RRR=0.18; 95 % CI: 0.05, 0.71) in the model with clustered SE. The mixed effects model did not converge when isolates from intestinal samples were assessed and no variables were statistically significant when the fecal swab isolates were assessed.
Similar patterns were seen when the relationship between mg/kg of purchased enrofloxacin was compared to enrofloxacin MIC values from isolates obtained from the same farms (See Table 6, Table 7). However, in contrast to the ceftiofur models, pig type (substandard versus sick) was significant in many models for isolates from intestinal samples and indicated that sick animals were more likely to have higher MICs. In all enrofloxacin models, the MIC of 0.12 µg/ml was the referent category. For intestinal isolates evaluated using the Cox proportional hazard model, with and without frailty, which evaluated the hazard of susceptibility, farm type (BTW compared to WTM), mg/kg of purchased enrofloxacin, and pig type (substandard compared to sick) were statistically significant in both models. Isolates from BTW versus WTM farms had HRs with and without frailty of 1.72 (95 % CI: 1.23, 2.41) and 1.70 (95 % CI: 1.17, 2.46), respectively; a one-unit increase in mg/kg of purchased enrofloxacin had HRs of 0.85 (95 % CI: 0.75, 0.97) and 0.86 (95 % CI: 0.75, 0.98), respectively; and both models had an HR of 0.64 (without and with frailty 95 % CI: 0.49, 0.84) when comparing substandard pigs to sick pigs. When analyzing fecal swab isolates, neither of the Cox proportional hazard indicated any statistically significant associations between any of the variables and MIC outcomes. For the parametric survival analysis with and without clustered SEs, the intestinal isolates had a TR of 0.76 (without clustered SE 95 % CI: 0.62, 0.93; with clustered SE 95 % CI: 0.59, 0.98) for farm type (BTW versus WTM), 1.08 without clustered SE 95 % CI: 1.01, 1.15; with clustered SE 95 % CI: 1.00, 1.16)for every one-unit increase in mg/kg of purchased enrofloxacin, and 1.27 (without clustered SE 95 % CI: 1.09, 1.48; with clustered SE 95 % CI: 1.11, 1.45) for pig type (substandard vs sick). The mixed effects parametric survival model had a TR of 0.75 (95 % CI: 0.60, 0.92) for farm type (BTW vs WTM), 1.29 (95 % CI: 1.05, 1.58) for spring compared to winter, and 1.27 (95 % CI: 1.11, 1.45) for pig type (substandard vs sick). For fecal swab isolates, only mg/kg of purchased enrofloxacin was significant (TR=1.05; 95 % CI: 1.01, 1.09) in the model without clustered SE. The model with clustered SE and the mixed effects model had no statistically significant findings.
Table 6. Summary table of model results for the comparison of milligrams of enrofloxacin purchased by the kilograms of pig produced to E. coli isolate minimum inhibitory concentrations determined by broth microdilution.
| May 2020 through August 2022 – Intestinal E. coli isolates | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Empty Cell | Cox Prop Hazard | Cox Prop Hazard Frailty | Para Survival | Para Survival Clust SE | Para Survival ME | Ord | Ord Cont Ratio | Multi-nomial | Multi- nomial Clst SE | Multi- nomial ME | Gen Ord | Gen Ord Clst SE |
| BTW versus WTM | 0.002 | 0.005 | 0.009 | 0.038 | 0.007 | < 0.001 | Vio-lated | < 0.001 | < 0.001 | NC | < 0.001 | < 0.001 |
| Season (df = 3) | Not sig | Not sig | Not sig | Not sig | 0.038 | Not sig | Vio-lated | Not sig | Not sig | NC | Not sig | Not sig |
| mg/kg | 0.015 | 0.029 | 0.028 | 0.045 | 0.055§ | 0.002 | Vio-lated | 0.015 | 0.003 | NC | 0.018 | 0.02 |
| Substandard pigs versus sick pigs | 0.001 | 0.001 | 0.002 | 0.001 | < 0.001 | Not sig | Vio-lated | Not sig | Not sig | NC | Not sig | Not sig |
| Empty Cell | Hazard ratio (95 % CI) | Hazard ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Odds Ratio (95 % CI) | Odds Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Odds ratio | Odds ratio |
| BTW versus WTM | 1.72 (1.23, 2.41) | 1.70. (1.17, 2.46) | 0.76 (0.62, 0.93) | 0.76. (0.59, 0.98) | 0.75 (0.60, 0.92) | 0.34 (0.24, 0.50) | — | Table 7 | Table 7 | — | Table 7 | Table 7 |
| Spring versus winter | — | — | — | — | 1.29. (1.05, 1.58) | — | — | — | — | — | — | — |
| Summer versus winter | — | — | — | — | Not Sig | — | — | — | — | — | — | — |
| Fall versus winter | — | — | — | — | Not Sig | — | — | — | — | — | — | — |
| mg/kg | 0.85 (0.75, 0.97) | 0.86 (0.75, 0.98) | 1.08 (1.01, 1.15) | 1.08 (1.00, 1.16) | Not Sig | 1.22 (1.08, 1.39) | — | Table 7 | Table 7 | — | Table 7 | Table 7 |
| Substandard compared to sick | 0.64 (0.49, 0.84) | 0.64 (0.49, 0.84) | 1.27 (1.09, 1.48) | 1.27 (1.11, 1.46) | 1.27. (1.11, 1.45) | NA | — | — | — | — | — | — |
| July 2022 through October 2023 – Fecal swab E. coli isolates | ||||||||||||
| Empty Cell | Cox Prop Hazard | Cox Prop Hazard Frailty | Para Survival | Para Survival Clust SE | Para Survival ME | Ord | Ord Cont Ratio | Multi-nomial | Multi- nomial Clst SE | Multi- nomial ME | Gen Ord | Gen Ord Clst SE |
| BTW versus WTM | Not sig | Not sig | Not sig | Not sig | Not sig | < 0.001 | Vio-lated | < 0.001 | < 0.001 | < 0.001 | < 0.001 | < 0.001 |
| Season (df = 3) | Not sig | Not sig | Not sig | Not sig | Not sig | Not sig | Vio-lated | Not sig | Not sig | Not sig | Not sig | Not sig |
| mg/kg | Not sig | Not sig | 0.018 | Not sig | Not sig | < 0.001 | Vio-lated | < 0.001 | 0.012 | 0.332§ | < 0.001 | 0.019 |
| Empty Cell | Hazard ratio (95 % CI) | Hazard ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Time Ratio (95 % CI) | Odds Ratio (95 % CI) | Odds Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Rel. Risk Ratio | Odds ratio | Odds ratio |
| BTW versus WTM | — | — | — | — | — | 0.26 (0.18, 0.37) | NA | Table 7 | Table 7 | Table 7 | Table 7 | Table 7 |
| Spring versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| Summer versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| Fall versus winter | — | — | — | — | — | — | — | — | — | — | — | — |
| mg/kg | — | — | 1.05 (1.01, 1.09) | — | — | 1.27 (1.15, 1.39) | — | Table 7 | Table 7 | — | Table 7 | Table 7 |
The continuation ratio ordinal regression model violated the proportional odds assumption in both periods and was not included in the table. Abbreviations were used to describe the models. “BTW” for breed-to-wean farm sites, “WTM” for wean-to-market farm sites, “Prop” for proportional, “Para” is an abbreviation for parametric, “Clst SE” for clustered standard errors, “ME” for mixed effects, “Gen ord” for generalized ordinal, “Rel” for relative, “Not Sig” for not significant, “Excess Neg Pred Prob” for excess negative predicted probabilities, and “NA” for not applicable. * Was an insignificant dummy variable, but overall omnibus variable was significant. † Final model had < 15 negative predicted probabilities. § The mg/kg variable was forced to stay in the model, but it was not significant overall.
Table 7. Multinomial logistic and generalized ordinal statistical models results for enrofloxacin minimum inhibitory concentrations from E. coli isolates determined by broth microdilution.
| Empty Cell | Multinomial | Multinomial Clst SE | Multinomial ME | Gen Ord | Gen Ord Clst SE | |||||
|---|---|---|---|---|---|---|---|---|---|---|
| MIC | Relative risk ratio (95 % CI) | p-value | Relative risk ratio (95 % CI) | p-value | Relative risk ratio (95 % CI) | p-value | Odds ratio | p-value | Odds ratio | p-value |
| INTESTINAL SAMPLES | ||||||||||
| 0.12 | Reference | |||||||||
| 0.25 | ||||||||||
| mg/kg | 1.24 (0.89, 1.74) | 0.208 | 1.24 (0.88, 1.75) | 0.217 | — | — | 1.25 (1.09, 1.42) | 0.001 | 1.25 (1.06, 1.46) | 0.006 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.27 (0.10, 0.76) | 0.013 | 0.27 (0.08, 0.90) | 0.032 | — | — | 0.33 (0.23, 0.48) | < 0.001 | 0.33 (0.20, 0.54) | < 0.001 |
| 0.5 | ||||||||||
| mg/kg | 1.33 (1.12, 1.58) | 0.001 | 1.33 ( 1.14, 1.56) | < 0.001 | — | — | 1.24 (1.08, 1.42) | 0.003 | 1.24 (1.06, 1.44) | 0.006 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.29 (0.17, 0.51) | < 0.001 | 0.29 (0.15, 0.56) | < 0.001 | — | — | 0.35 (0.24, 0.52) | < 0.001 | 0.35 (0.22, 0.57) | < 0.001 |
| 1 | ||||||||||
| mg/kg | 1.18 (0.95, 1.46) | 0.127 | 1.18 (0.95, 1.47) | 0.14 | — | — | 1.08 (0.89, 1.32) | 0.433 | 1.08 ( 0.87, 1.35) | 0.471 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.38 (0.21, 0.68) | 0.001 | 0.38 (0.20, 0.74) | 0.004 | — | — | 0.46 (0.28, 0.75) | 0.002 | 0.46 (0.26, 0.81) | 0.007 |
| 2 | ||||||||||
| mg/kg | 1.09 (0.72, 1.63) | 0.689 | 1.09 (0.70, 1.68) | 0.708 | — | — | 0.98 (0.63, 1.54) | 0.947 | 0.98 (0.59, 1.65) | 0.954 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.39 (0.15, 1.02) | 0.054 | 0.39 (0.13, 1.18) | 0.095 | — | — | 0.51 (0.20, 1.30) | 0.156 | 0.51 (0.18, 1.46) | 0.209 |
| FECAL SWABS | ||||||||||
| 0.12 | Reference | |||||||||
| 0.25 | ||||||||||
| mg/kg | 1.04 (0.79, 1.36) | 0.79 | 1.04 (0.73, 1.48) | 0.837 | 1.09 (0.75, 1.58) | 0.658 | 1.26 (1.14, 1.39) | < 0.001 | 1.26 (1.08, 1.47) | 0.004 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.24 (0.11, 0.52) | < 0.001 | 0.24 (0.08, 0.66) | 0.006 | 0.17 (0.05, 0.56) | 0.004 | 0.25 (0.18, 0.36) | < 0.001 | 0.25 (0.16, 0.41) | < 0.001 |
| 0.5 | ||||||||||
| mg/kg | 1.26 (1.10, 1.45) | 0.001 | 1.26 (1.07, 1.48) | 0.005 | 1.32 (1.07, 1.62) | 0.009 | 1.30 (1.17, 1.44) | < 0.001 | 1.30 (1.12, 1.51) | 0.001 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.26 (0.15, 0.44) | < 0.001 | 0.26 (0.14, 0.47) | < 0.001 | 0.22 (0.11, 0.47) | < 0.001 | 0.27 (0.18, 0.40) | < 0.001 | 0.27 (0.16, 0.45) | < 0.001 |
| 1 | ||||||||||
| mg/kg | 1.30 (1.14, 1.49) | < 0.001 | 1.30 (1.10, 1.54) | 0.002 | 1.37 (1.12, 1.67) | 0.002 | 1.29 (1.12, 1.48) | < 0.001 | 1.29 (1.06, 1.57) | 0.01 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.32 (0.19, 0.56) | < 0.001 | 0.32 (0.16, 0.65) | 0.002 | 0.27 (0.13, 0.58) | 0.001 | 0.30 (0.18, 0.51) | < 0.001 | 0.30 (0.15, 0.59) | < 0.001 |
| 2 | ||||||||||
| mg/kg | 1.28 (0.96, 1.71) | 0.094 | 1.28 (0.90, 1.82) | 0.173 | 1.37 (0.94, 1.98) | 0.098 | 1.26 (0.90, 1.76) | 0.182 | 1.26 (0.82, 1.93) | 0.301 |
| Farm type | ||||||||||
| Wean-to-market | Reference | |||||||||
| Breed-to-wean | 0.12 (0.04, 0.37) | < 0.001 | 0.12 (0.03, 0.39) | < 0.001 | 0.09 (0.02, 0.38) | 0.001 | 0.14 (0.04, 0.51) | 0.003 | 0.14 (0.03, 0.62) | 0.01 |
“SE” is an abbreviation for standard error. “NA” is an abbreviation for not applicable. “Gen” is for generalized. * The model for intestinal samples did not converge.
Among intestinal isolates, the ordinal regression model resulted in statistically significant results for BTW versus WTM farms (OR=0.34; 95 % CI: 0.24, 0.50)) and a one-unit increase in mg/kg of purchased enrofloxacin (OR=1.22; 95 % CI: 1.08, 1.39). For the fecal swab isolates, similar results were seen with BTW versus WTM farms (OR=0.26; 95 % CI: 0.18, 0.37) and a one-unit increase in mg/kg of purchased enrofloxacin (OR=1.27; 95 % CI: 1.15, 1.39). The proportional odds were violated when using both intestinal and fecal swab isolates for the ordinal continuous ratio model. Among the multinomial models, with and without clustered SEs used to assess intestinal isolates, farm type and mg/kg of purchased enrofloxacin were both statistically significant (P < 0.05). For both models, the MIC value of 0.12 µg/ml was the referent category. For farm type, the RRRs were the same for both models. BTW sites had a lower risk than WTM sites for MIC values of 0.25 µg/ml (RRR=0.27; without clustered SE 95 % CI: 0.10, 0.76; with clustered SE 95 % CI: 0.08, 0.90), 0.5 µg/ml (RRR=0.29; without clustered SE 95 % CI: 0.17, 0.51; with clustered SE 95 % CI: 0.15, 0.56), and 1 µg/ml (RRR=0.38; without clustered SE 95 % CI: 0.21, 0.68; with clustered SE 95 % CI: 0.20, 0.74). The mg/kg of purchased enrofloxacin variable was statistically significant only at the MIC value of 0.5 µg/ml and had an RRR of 1.33 (without clustered SE 95 % CI: 1.12, 1.58; with clustered SE 95 % CI: 1.14, 1.56) for both models. The mixed effects model did not converge when used to assess isolates from intestinal samples, and no results were available. Among the multinomial models, with and without clustered SEs used to assess fecal swab isolates, farm type and mg/kg of purchased enrofloxacin were also statistically significant both models yielded the same RRRs. Specifically, BTW farm sites compared to WTM sites had a lesser relative risk ratio than WTM sites at all MIC values, as follows: at an MIC of 0.25 µg/ml the RRR was 0.24 (without clustered SE 95 % CI: 0.11, 0.52; with clustered SE 95 % CI: 0.08, 0.66), at an MIC value of 0.5 µg/ml the RRR was 0.26 (without clustered SE 95 % CI: 0.15; 0.44, with clustered SE 95 % CI: 0.14, 0.47), at an MIC value of 1 µg/ml the RRR was 0.32 (without clustered SE 95 % CI: 0.19, 0.56; with clustered SE 95 % CI: 0.16, 0.65), and at an MIC value of 2 µg/ml the RRR was 0.12 (without clustered SE 95 % CI: 0.04, 0.37; with clustered SE 95 % CI: 0.03, 0.39). The mg/kg of purchased enrofloxacin variable was significant at an MIC of 0.5 µg/ml and an MIC of 1 µg/ml with RRRs of 1.26 (without clustered SE 95 % CI: 1.10, 1.45; with clustered SE 95 % CI: 1.07, 1.48) and 1.30 (without clustered SE 95 % CI: 1.14, 1.49; with clustered SE 95 % CI: 1.10, 1.54), respectively. The mixed effects model returned similar statistically significant results for isolates from fecal swabs, with farm type exhibiting statistically significant associations across all MIC values: an RRR of 0.17 (95 % CI: 0.05, 0.45; MIC=0.25 µg/ml), an RRR of 0.22 (95 % CI: 0.11, 0.47; MIC=0.5 µg/ml), an RRR of 0.27 (95 % CI: 0.13, 0.58; MIC=1 µg/ml), and an RRR of 0.09 (95 % CI: 0.02, 0.38; MIC=2 µg/ml). The omnibus test of mg/kg or purchased enrofloxacin was not statistically significant in the mixed effects model, but the variable was forced into the model and was significant at an MIC of 0.5 µg/ml (RRR=1.32; 95 % CI: 1.07, 1.62) and an MIC of 1 µg/ml (RRR=1.37; 95 % CI: 1.12, 1.67). Finally, the generalized ordinal regression model with and without clustered standard errors also indicated that farm type and mg/kg of purchased enrofloxacin were statistically significantly associated with MIC values, The ORs were the same for each model (with and without clustered SEs) for both the intestinal isolates and the fecal swab isolates. For the intestinal isolates, the BTW versus WTM comparison was statistically significant at the MIC values of 0.25, 0.5 and 1 µg/ml with ORs of 0.33 (without clustered SE 95 % CI: 0.23, 0.48; with clustered SE 95 % CI: 0.20, 0.54), 0.35 (without clustered SE 95 % CI: 0.24, 0.52; with clustered SE 95 % CI: 0.22, 0.57), and 0.46 (without clustered SE 95 % CI: 0.28, 0.75; with clustered SE 95 % CI: 0.26, 0.81), respectively. The mg/kg of purchased enrofloxacin variable was significant at MIC values of 0.25 and 0.5 µg/ml with ORs of 1.25 (without clustered SE 95 % CI: 1.09, 1.42; with clustered SE 95 % CI: 1.06, 1.46) and 1.24 (without clustered SE 95 % CI: 1.08, 1.42; with clustered SE 95 % CI: 1.06, 1.44), respectively. For the fecal swab isolates, the farm type variable (BTW versus WTM) was statistically significant at all levels of MIC assessed (0.25, 0.5, 1, and 2 µg/ml) with ORs of 0.25 (without clustered SE 95 % CI: 0.18, 0.36; with clustered SE 95 % CI: 0.16, 0.41), 0.27 (without clustered SE 95 % CI: 0.18, 0.40; with clustered SE 95 % CI: 0.16, 0.45), 0.30 (without clustered SE 95 % CI: 0.18, 0.51; with clustered SE 95 % CI: 0.15, 0.59), and 0.14 (without clustered SE 95 % CI: 0.04, 0.51; with clustered SE 95 % CI: 0.03, 0.62), respectively. The mg/kg variable was statistically significant at all values except MIC of 2 µg/ml. For an MIC of 0.25 µg/ml the OR was 1.26 (without clustered SE 95 % CI: 1.14, 1.39; with clustered SE 95 % CI: 1.08, 1.47), an MIC of 0.5 µg/ml the OR was 1.30 (without clustered SE 95 % CI: 1.17, 1.44; with clustered SE 95 % CI: 1.12, 1.51), and an MIC of 1 µg/ml the OR was 1.29 (without clustered SE 95 % CI: 1.12, 1.48; with clustered SE 95 % CI: 1.06, 1.57).
4. Discussion
In this study, we summarized the relationship of presumptive antibiotic use, based on purchase records and number of pigs produced, to antibiotic MIC values for enrofloxacin and ceftiofur from pig E. coli isolates while controlling for farm type, season, and pig health status through evaluation of the agreement between different multivariable modeling approaches.
By evaluating multiple statistical models, we discovered a consistent association between farm type and the amount of ceftiofur and enrofloxacin purchased at an enterprise level, and phenotypic E. coli resistance as measured via the MIC. Interestingly, BTW isolates had a lower odds, hazard, or risk of having higher MIC values compared to WTM isolates, despite the fact that BTW sites had higher median mg/kg of antibiotic purchases as well as more isolates from farms in the fourth quartile of antibiotic purchases per kg of pig produced.
The relationship detected between mg/kg of antibiotic purchased and MIC was similar to another experimental study that evaluated therapeutic and subtherapeutic doses of enrofloxacin and ceftiofur on E. coli resistance 10 days after the last antibiotic exposure. The study reported that resistant E. coli populations were more prevalent in the subtherapeutic group and therapeutic enrofloxacin and ceftiofur exposure groups relative to controls that were not exposed (Lin et al., 2017). Yet, that study only looked at a short-time period of antibiotic exposure, with a short follow-up period. Another study evaluating the effect of ceftiofur treatment on fecal E. coli recovered from cattle revealed a temporal effect on the likelihood of recovering a resistant isolate, which lasted from 13 to 15 days post-exposure (Lowrance et al., 2007). The study described herein did not capture any data about the time lapse between antibiotic exposure and sample collection, which would have improved results interpretation. More robust longitudinal sampling could improve understanding of antibiotic exposure effects on bacteria isolated from commercial pigs.
Clearly, antimicrobial resistance is complex with many variables to consider. Our study looked at weaning-aged pigs (i.e., 21–28 days of age) at BTW sites and in older-aged pigs at WTM sites. WTM sites have the widest potential range of pig ages from 21 days to 6 months; however, pig age was not consistently captured at the time of sampling. It is well-documented that neonatal, suckling pigs had the highest E. coli resistant to antibiotics across age groups in longitudinal studies (Cameron-Veas et al., 2016, Dohmen et al., 2017, Gaire et al., 2021). This study evaluated pigs at the weaning age only (BTW sites) and that may be one reason the WTM sites resulted in having higher MIC values overall. The trend of having higher level of antimicrobial resistance in neonatal pigs may still exist, but for these reasons, were not seen here. In addition, recent and current antibiotic treatments that shortly preceded or occurred contemporaneously with sampling were not captured, which could also potentially bias the results, particularly if treatments have a short-term effect on resistance prevalence. Another potentially confounding factor was the inclusion of sick and substandard pigs in the May 2020 through August 2022 period (period 1), but only healthy pigs in the July 2022 through October 2023 period (period 2). The inclusion of sick pigs in period 1 could have made pathogenic E. coli more likely to be recovered in the corresponding isolates, which could have skewed the resistance profile of the isolate set as compared to period 2. The strain-level profile of recovered E. coli is likely an important factor in phenotypic resistance profiles and could explain the significant difference in MIC values observed between intestinal isolates recovered from sick versus substandard pigs. However, commensal E. coli are considered an indicator for AMR gene presence, so the findings here likely represent the bacterial populations present (European Food Safety Authority and European Centre for Disease Prevention and Control, 2012). Our sampling design did not equally space sampling over time; instead, it was at the convenience of the veterinarian with encouragement to sample every six months. The practicality of needed veterinary services resulted in sites being sampled at intervals greater than or less than six months. Future studies would do well to capture farm type, use, age, and duration since last treatment at sampling to better understand these relationships. Standard sampling time points should also be set to ensure representative sampling over time.
Furthermore, using purchased data as a proxy for antibiotic use contained challenges. Purchase data is an aggregate metric at the enterprise or barn-level, but not at the pig-level. Many antibiotics are given as water medication to the entire barn, which can make it difficult to quantify the final amount a pig received without further assumptions and estimations. and are Injectable antibiotics, such as enrofloxacin and ceftiofur, are typically given to sick pigs. Finally, in BTW sites, the antibiotics account for the antibiotics given to sows and piglets (only piglets were sampled) while in WTM sites they account for the antibiotics given solely to the pigs sampled. This is another reason that farm type was likely a significant confounder in the model as well. Isolates were taken from pooled samples of pigs, but a limitation to this study is that it assumed a shared environment impacted pig-level resistance patterns.
This study provides further evidence of the relationship between use and resistance. Yet, the relationship is complex and may have been confounded in this study by missing data on pig age and time since last treatment, as discussed above. It is more likely that sick pigs received treatments of enrofloxacin and ceftiofur as these drugs are given by injection in swine medicine and not in the feed or water. Healthy pigs would not receive these treatments. During period 1, substandard pigs were compared to sick pigs in this study, substandard pigs were not treated, yet, they had a lesser hazard of being susceptible (0.64) and an increased time ratio (1.27) for MIC values when enrofloxacin purchases were assessed compared to sick pigs when using the Cox proportional hazard model and the parametric proportional hazard model, but not for other models. There was no such relationship when ceftiofur was evaluated. It is also important to note that BTW sites had greater mg/kg purchases of antibiotics relative to WTM sites, and that BTW sites still had lower MIC values compared to WTM sites, which initially implied that higher use was not related to MIC values. Clearly, the production stage matters and needs to be included in statistical models as a confounder. An additional challenge with these findings is that mg/kg antibiotic metric itself. On WTM sites, the population of pigs being treated with the purchased drug in the numerator is the same as the group being marketed in the denominator. On BTW sites, a larger population is treated with the purchased drugs (female pigs being raised for breeding, breeding female pigs, and male pigs used to detect estrus) than are in the denominator (weaned pigs produced). It is very difficult to separate out what animals were treated off paper records and, since piglets are not marketed for food, antibiotic use in that age group is not recorded. Yet, as breeding pigs are treated, the piglets that become weaned pigs are also exposed either through nursing or the placenta, so excluding sow treatments can also be troublesome. Veterinarians should be aware of the evidence that there is a relationship between use and resistance for enrofloxacin and ceftiofur, but the research community should continue to clarify the details of the relationship.
5. Conclusion
Antibiotic resistance among enteric bacteria to ceftiofur and enrofloxacin is of concern as these drugs are members of the highest priority and critically important drug classes used to treat humans with E. coli and other gram-negative bacterial infections such as salmonellosis (World Health Organization, 2019). The results of this study suggest that enterprises that purchase and, thus, potentially use, more ceftiofur and enrofloxacin have higher MIC values for ceftiofur and enrofloxacin among fecal E. coli isolates obtained from pigs raised in commercial barns, which could be suggestive of higher resistance. Future work should control for time from treatment and the age of the animal as well to ensure the observed associations are consistent across the production stages, and especially at slaughter when any risk to the food supply would be greatest.
Funding sources
Funding for this work was provided by the US Department of Agriculture through a cooperative agreement (Funding opportunity number: USDA-APHIS-10025-VSSP0000–22–0025) with the SDSU ADRDL, the Foundation for Food and Agriculture (Grant IDs: ICASA-0000000016, RDS-0000000001, and 22–000515), Pipestone Veterinary Services, and the National Pork Board (Grant numbers 19–239 and 21–018).
Ethical statement
This study was approved by the Pipestone Research Institutional Animal Care and Use Committee. Protocol #2020–01.
CRediT authorship contribution statement
Taylor Spronk: Project administration, Methodology, Investigation, Data curation, Conceptualization. Scott Dee: Writing – review & editing, Project administration, Methodology, Investigation, Funding acquisition, Conceptualization. Noelle Noyes: Writing – review & editing, Methodology, Conceptualization. Laura B Goodman: Writing – review & editing, Methodology, Conceptualization. H Morgan Scott: Writing – review & editing, Methodology, Conceptualization. Karyn A Havas: Writing – original draft, Validation, Project administration, Formal analysis, Data curation. Laura Ruesch: Writing – review & editing, Project administration, Methodology, Investigation, Data curation. Roy Edler: Writing – review & editing, Project administration, Methodology, Data curation. Joel Nerem: Writing – review & editing, Supervision, Methodology, Conceptualization. Marlee Braun: Writing – review & editing, Investigation.
Declaration of Competing Interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: All authors reports financial support was provided by National Pork Board. All authors reports financial support was provided by Foundation for Food and Agriculture Research. Laura Ruesch, Marlee Braun, Karyn Havas, Roy Edler, Joel Nerem, Scott Dee, Taylor Spronk reports financial support was provided by United States Department of Agriculture. Karyn A Havas, Roy Edler, Joel Nerem, Scott Dee, Taylor Spronk reports a relationship with Pipestone managed sow farms and independent family farmers that includes: consulting or advisory and employment. Authors from Pipestone Research and Pipestone Veterinary Services provide veterinary and research support to the farmers’ pigs that were sampled as part of the study. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgements
We would like to thank all the swine enterprises that enrolled in this study and allowed for sample collection at their farm sites. We would also like to thank the Pipestone veterinary staff that conducted the sampling for this project. Finally, we would like to thank Dr. Alice Green, Dr. Charles Haley, Dr. Chelsey Shively, and Dr. Matthew Vuolo for their assistance on this project. We would also like to thank Pipestone Veterinary Services for their work in collecting samples.
Data availability
Data may be available upon reasonable request.
References
- Aerts et al., 2022
A multicategory logit model detecting temporal changes in antimicrobial resistancePLoS ONE, 17 (2022), Article e0277866, 10.1371/journal.pone.0277866
View in ScopusGoogle ScholarAlali et al., 2009
Relationship between level of antibiotic use and resistance among Escherichia coli isolates from integrated multi-site cohorts of humans and swine
Prev. Vet. Med., 90 (2009), pp. 160-167, 10.1016/j.prevetmed.2009.05.018
View PDFView articleView in ScopusGoogle ScholarAnderson et al., 2023
Evaluating natural planned exposure protocols on rotavirus shedding patterns in gilts and the impact on their suckling pigs
JSHAP, 31 (2023), pp. 10-19, 10.54846/jshap/1294
View in ScopusGoogle ScholarAndersson, 2003
Persistence of antibiotic resistant bacteria
Curr. Opin. Microbiol., 6 (2003), pp. 452-456, 10.1016/j.mib.2003.09.001
View PDFView articleView in ScopusGoogle ScholarBjork et al., 2015
Comparison of mixed effects models of antimicrobial resistance metrics of livestock and poultry Salmonella isolates from a national monitoring system
Prev. Vet. Med., 122 (2015), pp. 265-272, 10.1016/j.prevetmed.2015.10.010
View PDFView articleView in ScopusGoogle ScholarBurow et al., 2019
Antibiotic resistance in Escherichia coli from pigs from birth to slaughter and its association with antibiotic treatment
Prev. Vet. Med., 165 (2019), pp. 52-62, 10.1016/j.prevetmed.2019.02.008
View PDFView articleView in ScopusGoogle ScholarCallens et al., 2018
Associations between a decreased veterinary antimicrobial use and resistance in commensal Escherichia coli from Belgian livestock species (2011–2015)
Prev. Vet. Med., 157 (2018), pp. 50-58, 10.1016/j.prevetmed.2017.10.013
View PDFView articleView in ScopusGoogle ScholarCameron-Veas et al., 2016
Shedding of cephalosporin resistant Escherichia coli in pigs from conventional farms after early treatment with antimicrobials
Vet. J., 211 (2016), pp. 21-25, 10.1016/j.tvjl.2016.02.017
View PDFView articleView in ScopusGoogle ScholarChadwick et al., 1996
Glycopeptide-resistant enterococci isolated from uncooked meat
J. Antimicrob. Chemother., 38 (1996), pp. 908-909, 10.1093/jac/38.5.908
- Chapman, 1947
A superior culture medium for the enumeration and differentiation of coliformsJ. Bacteriol., 53 (1947), p. 504
Minimum inhibitory concentrations of cephalosporin compounds and their active metabolites for selected mastitis pathogens
ajvr, 74 (2013), pp. 683-690, 10.2460/ajvr.74.5.683
View in ScopusGoogle ScholarDohmen et al., 2017
Risk factors for ESBL-producing Escherichia coli on pig farms: a longitudinal study in the context of reduced use of antimicrobials
PLoS ONE, 12 (2017), Article e0174094, 10.1371/journal.pone.0174094
- Dohoo et al., 2012
Model-building strategiesin: Methods in Epidemiologic Research, VER Inc, Prince Edward Island, Canada (2012), pp. 401-428
European Food Safety Authority, European Centre for Disease Prevention and Control, 2012
The European Union Summary Report on antimicrobial resistance in zoonotic and indicator bacteria from humans, animals and food in 2010
EFS, 2 (2012), 10.2903/j.efsa.2012.2598
Impact of ceftiofur injection on gut microbiota and Escherichia coli resistance in pigs
Antimicrob. Agents Chemother., 59 (2015), pp. 5171-5180, 10.1128/AAC.00177-15
- Food and Drug Administration (FDA), 2022
Summary Report On Antimicrobials Sold or Distributed for Use in Food-Producing AnimalsU.S. Department of Health and Human Services, Rockville, MD (2022)
A longitudinal investigation of the effects of age, dietary fiber type and level, and injectable antimicrobials on the fecal microbiome and antimicrobial resistance of finisher pigs
J. Anim. Sci., 100 (2022), Article skac217, 10.1093/jas/skac217
View in ScopusGoogle ScholarGaire et al., 2021
Age dependence of antimicrobial resistance among fecal bacteria in animals: a scoping review
Front. Vet. Sci., 7 (2021), Article 622495, 10.3389/fvets.2020.622495
Spread of plasmid-mediated nourseothricin resistance due to antibiotic use in animal husbandry
J. Basic Microbiol., 26 (1986), pp. 461-466, 10.1002/jobm.3620260806
View in ScopusGoogle ScholarKlare et al., 1995a
Enterococcus faecium Strains with vanA -mediated high-level glycopeptide resistance isolated from animal foodstuffs and fecal samples of humans in the community
Microb. Drug Resist., 1 (1995), pp. 265-272, 10.1089/mdr.1995.1.265
View in ScopusGoogle ScholarKlare et al., 1995b
vanA -mediated high-level glycopeptide resistance in Enterococcus faecium from animal husbandry
FEMS Microbiol. Lett., 125 (1995), pp. 165-172, 10.1111/j.1574-6968.1995.tb07353.x
Effect of ceftiofur and enrofloxacin on E. coli sub-population in pig gastrointestinal tract
J. Glob. Antimicrob. Resist., 10 (2017), pp. 126-130, 10.1016/j.jgar.2017.05.010
View PDFView articleView in ScopusGoogle ScholarLowrance et al., 2007
Changes in antimicrobial susceptibility in a population of Escherichia coli isolated from feedlot cattle administered ceftiofur crystalline-free acid
ajvr, 68 (2007), pp. 501-507, 10.2460/ajvr.68.5.501
View in ScopusGoogle ScholarLudden et al., 2019
One health genomic surveillance of Escherichia coli demonstrates distinct lineages and mobile genetic elements in isolates from humans versus livestock
mBio, 10 (2019), Article e02693-18, 10.1128/mBio.02693-18
View in ScopusGoogle ScholarLugsomya et al., 2018
Antimicrobial resistance in commensal Escherichia coli Isolated from pigs and pork derived from farms either routinely using or not using in-feed antimicrobials
Microb. Drug Resist., 24 (2018), pp. 1054-1066, 10.1089/mdr.2018.0154
- MacFaddin, 1985
Media fo Isolation-Cultivation-Identification-Maintenance of Medical BacteriaWilliams and Wilkins, Baltimore, Maryland, USA (1985)
Food-borne origins of Escherichia coli causing extraintestinal infections
Clin. Infect. Dis., 55 (2012), pp. 712-719, 10.1093/cid/cis502
Overview of quantitative methodologies to understand antimicrobial resistance via minimum inhibitory concentration
Animals, 10 (2020), p. 1405, 10.3390/ani10081405
Global trends in antimicrobial use in food-producing animals: 2020 to 2030
PLOS Glob. Public Health, 3 (2023), Article e0001305, 10.1371/journal.pgph.0001305
View in ScopusGoogle ScholarNoyes et al., 2016
Modelling considerations in the analysis of associations between antimicrobial use and resistance in beef feedlot cattle
Epidemiol. Infect., 144 (2016), pp. 1313-1329, 10.1017/S0950268815002423
View in ScopusGoogle ScholarPenkova and Raymond, 2024
When does antimicrobial resistance increase bacterial fitness? Effects of dosing, social interactions, and frequency dependence on the benefits of AmpC β -lactamases in broth, biofilms, and a gut infection model
Evol. Lett. qrae015 (2024), 10.1093/evlett/qrae015
Relationship between antimicrobial drug usage and antimicrobial susceptibility of gram-positive mastitis pathogens
J. Dairy Sci., 90 (2007), pp. 262-273, 10.3168/jds.S0022-0302(07)72627-9
View PDFView articleView in ScopusGoogle ScholarPollard, 1946
A useful selective bactericidal property of tergitol-7
Science, 103 (1946), pp. 758-759, 10.1126/science.103.2687.758
View in ScopusGoogle ScholarSchouten et al., 1997
VRE and meat
Lancet, 349 (1997), p. 1258, 10.1016/S0140-6736(05)62461-0
View PDFView articleView in ScopusGoogle ScholarWitte, 2000
Selective pressure by antibiotic use in livestock
Int. J. Antimicrob. Agents, 16 (2000), pp. 19-24, 10.1016/S0924-8579(00)00301-0
View PDFView articleView in ScopusGoogle Scholar
- World Health Organization, 2015
World Health Organization, 2015. Global action plan on antimicrobial resistance. World Health Organization, Geneva.
- World Health Organization, 2019
World Health Organization, 2019. Critically important antimicrobials for human medicine, 6th rev. ed. World Health Organization, Geneva.
