Associations of antimicrobial uses with antimicrobial resistance of fecal Escherichia coli from pigs on 47 farrow-to-finish farms in Ontario and British Columbia

Abstract

This study assessed the associations between antimicrobial use and other management practices in pigs and antimicrobial resistance in generic Escherichia coli recovered from feces of weaner and finisher pigs on 39 purposefully selected farrow-to-finish farms in Ontario and 8 in British Columbia. Antimicrobials (n = 13), most frequently penicillins and tetracycline, were administered to different age groups of pigs on study farms through various routes of administration. Logistic regression was used to model risk factors to antimicrobial resistance in fecal E. coli of pigs for the following antimicrobials: ampicillin, apramycin, carbadox, cephalothin, chloramphenicol, kanamycin, neomycin, nitrofurantoin, spectinomycin, streptomycin, sulfamethoxazole, tetracycline, and cotrimoxazole (trimethoprim and sulfamethoxazole). Use of antimicrobials in weaner pigs compared with use in finisher pigs was associated with resistance in most models. There was phenotypic evidence of different mechanisms of resistance selection, including direct selection [use of carbadox was associated with carbadox resistance (OR = 6.48)]; cross-resistance [use of spectinomycin was associated with streptomycin resistance (OR = 2.29)]; and possible co-selection [ceftiofur use was associated with tetracycline resistance (OR = 6.12)]. These results provide further evidence that use of antimicrobials in pigs selects for resistance among fecal E. coli within and between classes of antimicrobials.

Introduction

While antimicrobial use in animals is believed to contribute to the selection of resistance among some bacteria that can be transmitted to humans through contaminated food or other means, the extent of this selection is inadequately understood (1). Antimicrobial use may select for resistance in pathogenic Escherichia coli, as well as nonpathogenic or commensal E. coli (2,3). Antimicrobials may be administered to animals through a variety of routes; however, in many countries, the largest volumes are administered orally through feed. Addition of antimicrobials to food animal rations in order to promote growth and feed efficiency has been practiced for decades and is estimated to account for more than half the total antimicrobial use globally (4).

In North America, both narrow-spectrum and broad-spectrum antimicrobials may be added to pig feed without prescription, and this in-feed medication is used for growth promotion, disease prophylaxis, and therapy (5,6). A number of epidemiological studies have focused on resistance in intestinal commensals in pigs and some have described the association between antimicrobial use in pigs under commercial production conditions and resistance in fecal E. coli (7–9). There is, however, a pressing need for more information on risk factors for resistance in food animals to assist in making evidence-based decisions concerning use of antimicrobials in these animals. This is particularly important in pig production where antimicrobial use is comparatively common (10).

The objective of this study was to determine potential risk factors associated with antimicrobial resistance in fecal E. coli of pigs, mainly focusing on in-feed antimicrobial use practices at the farm level.

Materials and methods

Sample collection, laboratory testing, and questionnaire

The methods used for farm selection, fecal sample collection, E. coli culture and susceptibility testing are described in detail elsewhere (11). In brief, study farms, [n = 47; 39 and 8 farms from Ontario (ON) and British Columbia (BC), respectively] were purposefully selected from farrow-to-finish operations that had at least 50 sows. The basis of purposeful selection was in-feed antimicrobial use practices; use of in-feed antimicrobials, or no in-feed antimicrobials. Eligible farms within these parameters were selected at random.

During each farm visit, 10 fecal samples were collected from weaners and pooled; the same was done for finishers. Thus, a total of 188 pooled fecal samples (94 from weaners and 94 from finishers) was obtained. From each pooled fecal sample, E. coli were recovered using conventional culture methods and 5 isolates were subjected to susceptibility testing to 21 antimicrobials of importance to animal and human health, using Clinical and Laboratory Standards Institute (CLSI) breakpoint concentrations, where available (11).

Data on antimicrobial use and other management factors were obtained using a questionnaire during farm visits (2 per farm) that took place from March to September, 1999 (ON) and May to August, 2000 (BC). The questionnaire captured information on the antimicrobials that were used or present on the farm at the time of the visits, the pig category treated, and the route of medication (in-feed, in water, injectable, topical). Information on potential confounders such as farm size, farm, visit number, and province were also collected simultaneously (Table I).

Table I.

Description and codes for independent variables collected and used in model building

Description of independent variables	Code	Description of independent variables	Code
Potential confounders		Antimicrobial use by pig category variables
Weaner versus finisher pigs	AGE	Apramycin used in piglets	Aprup
In-feed antimicrobial use farms	MED1	Apramycin used in weaners	Apruw
No antimicrobials in-feed farms	MED2	Carbadox used in piglets	Carup
Ontario versus British Columbia	PROV	Carbadox used in weaners	Caruw
Number of pigs on farm	SIZE	Carbadox used in finishers	Caruf
First versus second visit	VIS	Ceftiofur used in piglets	Cefup
General antimicrobial use variables by pig category		Ceftiofur used in weaners	Cefuw
In-feed antimicrobials used in piglets	FEEDP	Gentamicin used in weaners	Genuw
In-feed antimicrobials used in weaner pigs	FEEDW	Lincomycin used in finishers	Linuf
In-feed antimicrobials used in finisher pigs	FEEDF	Lincomycin used in weaners	Linuw
In-feed antimicrobials used in sows	FEEDS	Neomycin used in piglets	Neoup
Injectable antimicrobials used in piglets	INJP	Neomycin used in weaners	Neouw
Injectable antimicrobials used in weaner pigs	INJW	Penicillins used in finishers	Penuf
Injectable antimicrobials used in finisher pigs	INJF	Penicillins used in sows	Penus
Injectable antimicrobials used in sows	INJS	Penicillins used in weaners	Penuw
Antimicrobial use variables at farm level		Sulfonamides/trimethoprim used in piglets	Sulup
Apramycin used on farms	APR	Sulfonamides/trimethoprim used in finishers	Suluf
Carbadox used on farms	CAR	Sulfonamides/trimethoprim used in weaners	Suluw
Ceftiofur used on farms	CEF	Tetracycline used in finishers	Tetuf
Gentamicin used on farms	GEN	Tetracycline used in weaners	Tetuw
Lincomycin used on farms	LIN	Tiamulin used in finishers	Tiauf
Neomycin used on farms	NEO	Tiamulin used in piglets	Tiaup
Penicillin used on farms	PEN	Tylosin used in finishers	Tyluf
Spectinomycin used on farms	SPC	Tylosin used in sows	Tylus
Streptomycin used on farms	STR	Tylosin used in weaners	Tyluw
Sulphonamides used on farms	SUL	Routes of medication variables
Tetracycline used on farms	TET	No antimicrobials used	0
Tiamulin used on farms	TIA	Injectable individual treatment + oral individual treatment	1
Tylosin used on farms	TYL	In-feed group medication	2
		Water group medication	3
		Injectable individual + in-feed group treatment	4
		Injectable individual + water group treatment	5
		In-feed group + water group treatment	6
		Injectable individual + in-feed group + water group treatment	7

Data management and analysis

Data were manipulated and analyzed in SAS version 6.12 (SAS Institute, Cary, North Carolina, USA). The data were checked for accuracy and completeness using manual and computer software techniques. For analytical purposes, the unit of analysis was resistance (yes or no) at the level of the individual isolate. The antimicrobial resistance dependent variables (ampicillin resistance of the E. coli isolate) were expressed as binary outcomes (resistant or susceptible). All independent variables (antimicrobial use or other factors; Table I) were coded and screened with univariate analysis using the Likelihood Ratio Test Statistic. Variables significant at P ≤ 0.20 were eligible for inclusion in multivariable models and were retained in models only if significant at P ≤ 0.05 unless exclusion resulted in a significant change in deviance. Antimicrobial use variables (in most cases describing antimicrobial classes) were refined progressively in stages until the final model was obtained. For example, modeling started with coding tetracycline use on the farm; yes or no. If these data were significant, the tetracycline use on farms was refined further and recoded as tetracycline used on farms in weaners; yes or no. If these data were significant, the tetracycline use on farms in weaners was again refined and recoded as tetracycline used on farms in weaners via in-feed medication; yes or no (Figure 1).

Figure 1.

Approach taken to refine antimicrobial use variables in the final models and other management variables against antimicrobial resistance of fecal E. coli from pigs on 47 farrow-to-finish swine farms.

For highly correlated independent variables, preference was given to variables most highly significant in the models. In the case of correlated antimicrobial use variables, preference was given to the variable describing antimicrobial in the same class as the type of antimicrobial resistance being modeled.

Backwards elimination was used to build the models and two-way interaction terms were tested and when significant, stratified analyses were performed to be able to interpret the variables involved in the interaction. Variables not significant on initial univariate screening (P ≤ 0.20) were individually re-entered into final models to assess for significance, which may have indicated the presence of distortion effects in the model. Confounders were retained only if significant (P ≤ 0.05) in final models or if their exclusion resulted in a significant change in deviance.

Models were built for the following antimicrobials: ampicillin (32 μg/mL), apramycin (32 μg/mL), carbadox (30 μg/mL), cephalothin (32 μg/mL), chloramphenicol (32 μg/mL), cotrimoxazole (80 μg/mL, consisting of 4 μg/mL of trimethoprim and 76 μg/mL of sulfamethoxazole), kanamycin (64 μg/mL), neomycin (16 μg/mL), spectinomycin (64 μg/mL), streptomycin (64 μg/mL), sulfamethoxazole (512 μg/mL), and tetracycline (16 μg/mL). Antimicrobials that were not modeled because of low resistance prevalence (11) or inability for the model to converge included: amikacin (64 μg/mL) 0.00%, ceftriaxone (64 μg/mL) 0.32%, ciprofloxacin (4 μg/mL) 0.0%, ceftiofur (8 μg/mL) 0.00%, florfenicol (16 μg/mL) 0.32%, gentamicin (16 μg/mL) 0.74%, nalidixic acid (32 μg/mL) 0.43%, nitrofurantoin (32 μg/mL) 4.15%, and tobramycin (8 μg/mL) 3.40%.

Multiple E. coli isolates were harvested from pooled fecal samples and multiple fecal samples were collected from farms; circumstances that could predispose to clustering in the data. Different methods (d-scale option, generalized estimation equation, and Glimmix macro) were explored to account for potential clustering in the models (12), and the Glimmix macro with the restricted maximum likelihood option was selected as the most effective in minimizing clustering effects. In these mixed models, “farm” was treated as a random effect, and “province” and “visit” were treated as fixed effects.

Results

Antimicrobial use on the 47 farrow-to-finish swine farms

Table II shows the distribution of antimicrobials used on study farms. Penicillins, sulfonamides, tetracycline, and tylosin were among antimicrobials used by most producers (≥ 55%). Most “in-feed medication farms” commonly used penicillins, sulfonamides, tetracycline, and tylosin; and “no in-feed medication farms” commonly used penicillins, tetracycline, and sulfonamides (for individual treatments). No farms reported using antimicrobials in water.

Table II.

Types and usage patterns of antimicrobials among 47 farrow-to-finish pig farms

Antimicrobials used by any route of administration on study farms	Total number of farms that used the antimicrobial (n = 47) (%)	Use of the antimicrobial on farms that did not use in-feed medication (n = 13)	Use of antimicrobial on farms that did use in-feed medication (n = 34)
Apramycin	3 (6.38)	0	3
Carbadox	9 (19.15)	0	9
Ceftiofur	6 (12.77)	0	6
Gentamicin	8 (17.02)	2	6
Lincomycin	13 (27.66)	0	13
Neomycin	7 (14.89)	3	4
Penicillins	40 (85.11)	6	34
Spectinomycin	11 (23.40)	1	10
Streptomycin	1 (2.13)	0	1
Sulfonamides	27 (57.45)	1	26
Tetracycline	33 (70.21)	4	29
Tiamulin	6 (12.77)	0	6
Tylosin	26 (55.32)	2a	24

^aAlthough some farms did not use tylosin as in-feed medication, the antimicrobial was used for individual treatment of piglets or sows.

Table III shows patterns of antimicrobials used by stage of pig production on study farms. Most farms (59.6%) used in-feed medication in one or more stages of pig production (piglets, weaners, finishers, and sows), 10.6% administered in-feed medication to all 4 stages, and 29.8% of the farms did not use any in-feed medication. Most farms (89.4%) administered individual treatments to pigs at one or more stages of production.

Table III.

Antimicrobial use practices by pig stage categories on 47 farrow-to-finish swine farms

		Number of farms that used antimicrobial in treatment category
Antimicrobial used	Pig age category	Group treatment only	Individual treatment only	Group and individual treatments	Antimicrobial not used on farm (%)
Apramycin	Piglets	0	2	0	45 (95.75)
	Weaners	1	1	0	45 (95.75)
	Finishers	1	0	0	46 (97.87)
Carbadox	Piglets	3	0	0	44 (93.62)
	Weaners	9	0	0	38 (80.85)
	Finishers	1	0	0	46 (97.87)
Ceftiofur	Piglets	0	4	0	43 (91.49)
	Weaners	0	4	0	43 (91.49)
	Finishers	0	2	0	45 (95.75)
	Sows	0	2	0	45 (95.75)
Gentamicin	Piglets	0	8	0	39 (82.98)
	Weaners	0	2	0	45 (95.75)
Lincomycin	Piglets	6	3	0	38 (80.85)
	Weaners	7	1	1	38 (80.85)
	Finishers	4	0	0	43 (91.49)
	Sows	3	2	0	42 (89.36)
Neomycin	Piglets	2	0	0	45 (95.75)
	Weaners	3	0	0	44 (93.62)
	Finishers	1	0	0	46 (97.87)
	Sows	2	0	0	45 (95.75)
Penicillins	Piglets	2	23	0	22 (46.81)
	Weaners	8	16	6	17 (36.17)
	Finishers	0	23	2	22 (46.81)
	Sows	1	31	1	14 (29.79)
Spectinomycin	Piglets	4	3	0	40 (85.11)
	Weaners	7	0	0	40 (85.11)
	Finishers	1	0	0	46 (97.87)
	Sows	1	0	0	46 (97.87)
Streptomycin	Piglets	0	1	0	46 (97.87)
	Weaners	0	1	0	46 (97.87)
	Finishers	0	1	0	46 (97.87)
	Sows	0	1	0	46 (97.87)
Sulfonamides	Piglets	2	14	0	31 (65.96)
	Weaners	9	13	0	25 (53.19)
	Finishers	1	6	0	40 (85.11)
	Sows	1	9	0	37 (78.72)
Tetracyclines	Piglets	6	5	0	36 (76.60)
	Weaners	16	5	3	23 (48.94)
	Finishers	1	9	1	36 (76.60)
	Sows	1	12	1	33 (70.21)
Tiamulin	Piglets	4	1	0	42 (89.36)
	Weaners	3	0	1	43 (91.49)
	Finishers	2	0	0	45 (95.75)
	Sows	1	0	0	46 (97.87)
Tylosin	Piglets	1	7	0	39 (82.98)
	Weaners	0	10	2	35 (74.47)
	Finishers	10	5	3	29 (61.70)
	Sows	0	4	0	43 (91.49)

As shown in Table IV, group medication of pig rations by farmers was highest in weaners (63.8% of farms) and lowest in sows (14.9%). About equal numbers of farms used group medication in rations of piglets and finishers; 34% and 38.3%, respectively. About 2/3 of the farms used antimicrobials for individual treatments in all 4 stages of pig production.

Table IV.

General medication usage patterns in the various stages of pig production in 47 farrow-to-finish swine farms

	Number of farms (%)
Age categories of pigs	Group medication of ration	Individual-pig medication
Piglets	16 (34.0)	33 (70.2)
Weaners	30 (63.8)	26 (55.3)
Finishers	18 (38.3)	30 (63.8)
Sows	7 (14.9)	33 (70.2)

Multivariable modeling

The final logistic regression models for resistance to antimicrobials are presented in Table V. In the ampicillin model, an interaction between tiamulin use and AGE (weaners vs finishers) was significant; therefore, further analysis stratified by AGE was done (data not shown), which revealed that tiamulin use was associated [odds ratio (OR) = 4.59] with ampicillin resistance only in E. coli from finisher pigs.

Table V.

Final logistic regression models of risk factors for antimicrobial resistance among fecal Escherichia coli of pigs on 47 farrow-to-finish farms

						95% Confidence limits (CI)
Model of resistance outcome (% resistance)	Management and antimicrobial use variables	Pig stage category	Route of medication	P	Odds ratio	Lower CI	Upper CI
Ampicillin (35.21)	Age group of pigs	N/Aa	N/A	0	3.21	2.33	4.43
	Tiamulin use in pigs	N/Sb	N/S	0.002	4.87	1.82	13.01
	Interaction: age group of pigsc tiamulin use on farms	N/S	N/S	0.0242	0.41	0.19	0.89
Carbadox (10.11)	Carbadox use on farms	N/A	N/S	0.009	6.48	1.6	26.23
	Injection of sows	N/A	N/A	0.008	5.72	1.57	20.83
	Lincomycin use on farms	N/A	N/S	0.0137	0.17	0.04	0.69
Cephalothin (3.19)	Age group of pigs	N/A	N/A	0.6298	0.89	0.54	1.46
	Tylosin use on farms	N/S	N/S	0.023	0.04	0.01	0.32
	Interaction: age group of pigsc tylosin use on farms	N/A	N/A	0	8.47	2.44	29.37
Chloramphenicol (10.96)	Age group of pigs	N/A	N/A	0	3.77	2.36	6.03
	Tetracycline use on farms	N/S	N/S	0.0493	0.33	0.11	1
	Tylosin use on farms	Finisher pigs	In-feed	< 0.0001	9.29	2.84	30.37
		Finisher pigs	In-feed and injection	< 0.0001	19.9	3.63	109.1
Kanamycin (9.57)	Use of penicillins on farms	Weaner pigs	Injection	0.0141	10.41	1.61	67.38
Neomycin (9.79)	Age group of pigs	N/A	N/A	0	2.29	1.51	3.46
	Use of penicillins on farms	Piglets	Injection	0.0340d	3.5	1.1	11.17
	Farm size	N/A	N/A	0.008	1	1	1
Spectinomycin (55.00)	Age group of pigs	N/A	N/A	0	3.76	2.71	5.23
	Use of penicillins on farms	N/S	N/S	0.3426d	1.71	−1.77	5.16
	Spectinomycin use on farms	N/S	N/S	0.0539	2.39	−1.01	5.82
	Interaction: age group of pigsc use of penicillins on farms	N/A	N/A	0.004	3.08	1.43	6.6
Streptomycin (32.02)	Age group of pigs	N/A	N/A	0.001	1.62	1.22	2.16
	Spectinomycin use on farms	N/S	N/S	0.0168	2.29	1.15	4.16
	Farm size	N/A	N/A	0.002	1	1	1
Sulfamethoxazole (58.51)	Age group of pigs	N/A	N/A	< 0.0001	3.47	2.45	4.88
	Use of sulfonamides on farms	Weaner pigs	In-feed	0.0344	2.95	1.08	7.11
Tetracycline (81.28)	Age group of pigs	N/A	N/A	< 0.0001	3.47	2.46	4.89
	Ceftiofur use on farms	N/S	N/S	0	6.12	2.14	17.51
	Tiamulin use on farms	Weaner pigs	In-feed	0.0342	3.45	1.1	10.86
Trimethoprim/	Age group of pigs	N/A	N/A	0	4.02	2.52	6.4
sulfamethoxazole (5.53)	Tiamulin use on farms	N/S	N/S	0.028	7.75	1.25	48.11

^aNot applicable.

^bNot significant.

^cDenotes interaction between explanatory variables.

^dWald’s standard errors used.

Isolates from weaner pigs were approximately 6 times more likely to be resistant to carbadox compared to those from finisher pigs. Spectinomycin use was dropped from the model because of high correlation with lincomycin use.

An interaction between tylosin use and AGE was significant in the cephalothin model; therefore, further analysis stratified by age was done (data not shown), which revealed that tylosin use was associated (OR = 12.88) with cephalothin resistance only in E. coli from finisher pigs.

Tylosin use as in-feed medication in finisher pigs was associated (OR = 9.29) with chloramphenicol resistance. The odds of chloramphenicol resistance was further increased when tylosin was used as both an in-feed and injectable medication in finisher pigs (OR = 19.90).

An interaction between penicillin use and AGE was significant in the kanamycin model. Further analysis stratified by AGE (data not shown), revealed that penicillin use was associated (OR = 10.41) with kanamycin resistance only in E. coli from weaner pigs. When the age categories of pigs receiving the antimicrobial and route of administration were considered in the model, penicillin use by injection in weaner pigs was significantly associated with kanamycin resistance.

Use of penicillins in piglets through injection was significantly associated (OR = 3.5) with neomycin resistance. There was a high and significant correlation between use of penicillins in piglets and use of penicillins in finisher pigs; therefore, only the former was retained in the model. Also, E. coli from weaner pigs were approximately twice as likely to express neomycin resistance compared to those from finisher pigs. Farm size was a significant confounder in the model.

Use of spectinomycin on farms was associated (OR = 2.39) with spectinomycin resistance (P = 0.054). Use of penicillins on farms was significant in the spectinomycin resistance model but an interaction between use of penicillins and AGE was also significant in the model. Upon stratified analysis (data not shown), use of penicillins was associated with spectinomycin resistance (OR = 2.39) only in E. coli from weaner pigs.

Escherichia coli from farms that used spectinomycin were twice as likely to show streptomycin resistance compared to E. coli from farms that did not use this antimicrobial. Also, the odds of streptomycin resistance was approximately doubled in E. coli from weaner pigs compared to those from finisher pigs. Farm size again was a significant confounder in this model as was the case with the neomycin model.

Use of sulfonamides in weaner pigs through in-feed medication was associated (OR = 2.95) with sulfamethoxazole resistance. Use of penicillins and lincomycin was dropped from this model at the initial stages because of high correlation (57.8% and 45.3%, respectively) with sulfonamide use. As with most models in this study, E. coli of weaner pigs were approximately 3 times more likely to show sulfamethozaxole resistance compared to E. coli of finisher pigs.

Escherichia coli of weaners were approximately 3 times more likely to express tetracycline resistance than isolates from finisher pigs. Also, ceftiofur use by injection (OR = 6.12) and tiamulin use in weaner pigs through in-feed medication (OR = 3.45) were associated with tetracycline resistance.

Escherichia coli from farms that used tiamulin were approximately 8 times more likely to show resistance to trimethoprim/sulfamethoxazole (cotrimoxazole) compared to E. coli from farms that did not. Also, E. coli from weaner pigs were about 4 times more likely to show resistance to cotrimoxazole than those from finisher pigs.

Discussion

Risk factors were identified for all the resistance models and with a few exceptions, antimicrobial use variables when significant in regression models indicated a positive association with resistance. This suggests that some types of antimicrobial use under field conditions increases the odds of E. coli resistance in pigs.

The use of in-feed medication was associated with antimicrobial resistance in many models, and more consistently with weaners than other age categories of pigs (Table IV). This increased risk of resistance among E. coli from weaners was probably due in part to the more extensive use of antimicrobials in this phase of pig production, as reported previously for E. coli and other bacteria (7,13).

In-feed medication was more consistently associated with resistance (chloramphenicol, sulfamethoxazole, and tetracycline models) than injectable use (kanamycin and neomycin models). This makes biological sense since it would be expected that more animals would be exposed to antimicrobials for longer periods of time with in-feed medication than with injectable antimicrobial use. While neither route-specific treatment rates, nor quantities of antimicrobials used by various routes were measured in the present study, the results are consistent with those of a previous study in pigs under similar conditions, which showed that exposure to antimicrobials was higher with in-feed medication (7).

The results of multivariable modeling (Table V) provided evidence for direct selection of resistance (antimicrobial use selects for resistance to the same antimicrobial) as well as cross-resistance (antimicrobial use selecting for resistance to another antimicrobial in the same class), and co-selection (antimicrobial use selecting for resistance to an antimicrobial in another class). The results indicate that the use of carbadox was associated with carbadox resistance; spectinomycin use was associated with spectinomycin resistance; consistent with direct selection of resistance. It is worth noting that this study was conducted before carbadox was withdrawn from use in food animals in 2001. These findings are consistent with other studies that have shown that use of antimicrobials can select for resistance to those same antimicrobials (7,14,15).

This study also revealed evidence of cross-resistance in many models. For example, spectinomycin use was associated with streptomycin resistance. Although spectinomycin belongs to the aminocyclitol class of antimicrobials and streptomycin to the aminoglycoside class, these antimicrobials have similar basic amino-ring structures allowing cross-resistance to occur (16). Spectinomycin was approved for use in pigs in Canada during the entire study period, and streptomycin (reportedly used on 1 study farm) was withdrawn from use in food producing animals in Canada in July 2000; however, this was after the data had been collected from most study farms. Genes have been found that encode resistance to both streptomycin and spectinomycin in Gram-negative bacteria (17). Another example of possible cross-resistance is evident among the sulfonamides; that is, the use of sulfonamides (for example, sulfamethazine) was associated with sulfamethoxazole resistance. Sulfamethoxazole is not approved for use in pigs in Canada (18), and is only used in human medicine. The association seen here, therefore, is indicative of cross-resistance. Such an association was not observed with cotrimoxazole (trimethoprim/sulfonamide combination) resistance, thus illustrating the benefit of using synergistic antimicrobial combinations to overcome resistance.

Evidence of co-selection of resistance was also observed in this study. For example, the use of penicillins was associated with kanamycin and neomycin resistance. Penicillins are often used together with aminoglycosides. Co-selection may be an important mechanism for increased resistance of foodborne bacteria in pig populations (7,19). For example, it is believed that resistance to chloramphenicol among enteric bacteria is maintained from generation to generation through this process of co-selection (20) as chloramphenicol has not been approved for use in food animals in Canada since 1984, and there was no evidence that it was used on study farms.

Tylosin use was associated with cephalothin and chloramphenicol resistance. The latter association was also reported in another study (7). Tylosin is an antimicrobial to which E. coli are intrinsically resistant because of the nature of their cell wall and enzymatic activities (17). Thus, the associations seen here may be spurious or caused through some rare causal pathway.

Tetracycline use was not statistically associated with tetracycline resistance in this study. This is somewhat surprising, even though an earlier and similar study on swine in Ontario (7) made a similar observation. Other studies on pigs have shown that tetracycline use is associated with resistance in E. coli (15,21). Perhaps our study lacked statistical power, because both the prevalence of resistance to tetracycline, and the frequency of tetracycline use among study farms were high.

Some negative coefficients of antimicrobial use variables were observed on occasion (for example, lincomycin use negatively associated with carbadox resistance). This implies that in some cases antimicrobial use may have actually decreased, rather than increased the risk of resistance to certain antimicrobials. This seems implausible, and the observations may therefore be spurious. However, it is still possible that these findings may reflect incompatibility among resistance determinants, or that such factors may be acting as surrogates to some other risk factors yet to be identified.

Comparatively few individual-animal treatments with antimicrobials were significant in regression models. For example, injectable use of penicillins in piglets was significant in the neomycin model. Individual animal treatment with antimicrobials not approved for in-feed medication in swine in Canada had an approximately even distribution among the farm groups (Table III). The injectable use of penicillins in piglets, however, may actually be a “group” treatment since on many farms piglets are administered pencillin at processing (7).

This study had some limitations and attempts were made to mitigate the effects. Farms were purposively sampled in order to provide a better assessment of in-feed antimicrobial use; small farms that had < 50 sows were excluded. This may have introduced some level of selection bias into the study. In an attempt to minimize additional selection bias, random selections were made from eligible farms in the various categories. It is believed that the excluded farms constituted a very small fraction of the pork industry in terms of market hog production; their impact on antimicrobial resistance was proportionately small. The results are reasonably representative of the target population and meaningful inferences can be drawn from this study. Also, the antimicrobial use information provided by the farmers could have lead to misclassification bias if the information provided was not accurate. To reduce the potential for such bias, the claims of farmers concerning their antimicrobial use practices were validated by making a manual check of the feed additive records and existing drug stocks in their refrigerators and on shelves.

In conclusion, the results from this study provide further evidence that antimicrobial use in pigs, particularly use of in-feed medication, is significantly associated with resistance to a wide variety of antimicrobials. These include antimicrobials of the same and different classes.