Farm Animal Contact as Risk Factor for Transmission of Bovine-associated Salmonella Subtypes

Salmonellosis prevention should focus on safe animal contact as well as food safety.

with cattle specifi cally was also signifi cantly associated with being a case-patient (odds ratio 7.4, p = 0.0002), after food exposures were controlled for. More cases of bovineassociated salmonellosis in humans might result from direct contact with cattle, as opposed to ingestion of foods of bovine origin, than previously recognized. Efforts to control salmonellosis should include a focus on transmission routes other than foodborne.
S almonella enterica remains a formidable public health challenge, resulting in ≈1.2 million illnesses and 400 deaths annually in the United States alone (1). Disease manifestations include diarrhea, fever, anorexia, abdominal pain, vomiting, and malaise. Although clinical disease generally resolves within 3-7 days, Salmonella spp. can also produce potentially fatal invasive infections. The incidence of human salmonellosis has not declined over the past 15 years and is signifi cantly higher than it was during 2006-2008 (2). An estimated 94% of Salmonella infections are foodborne (1); common sources include undercooked eggs, poultry, beef, and pork; unpasteurized dairy products; and raw vegetables (3)(4)(5)(6)(7). Although some studies have shown that direct contact with infected animals is a risk factor for salmonellosis (8,9), the foodborne route is still regarded as the primary transmission route.
Dairy cattle are considered a key source of several Salmonella serovars that are a threat to human health, including multidrug-resistant S. enterica serovar Newport and S. enterica serovar Typhimurium (8)(9)(10)(11). Foodborne transmission can occur through fecal contamination of beef carcasses at the time of slaughter (12) or through contamination of crops, either by manure used as fertilizer or by manure-contaminated irrigation water (13). Milk and other dairy products pose less of a public health threat because of commercial pasteurization, although consumption of unpasteurized dairy products persists. Infection by direct contact is an occupational risk for dairy farm workers and veterinarians. The most recent National Animal Health Monitoring System Dairy Study reports that there were >75,000 dairy operations in the United States in 2006, and the American Veterinary Medical Association reports that there were >5,000 veterinarians engaged either predominantly or exclusively in food animal practice as of 2010. Persons who interact with dairy cattle in public settings, such as open farms, petting zoos, and county or state fairs, are also at risk for salmonellosis through direct exposure (8,9,14,15).
Our objective was to identify signifi cant risk factors for salmonellosis caused by bovine-associated Salmonella subtypes (including those within the Newport and Typhimurium serovars) by using the case-case study design (16). We specifi cally evaluated the role of direct animal contact as a potential route of transmission.

Study Population
This case-case study was conducted by using culture-confi rmed human salmonellosis cases reported in targeted geographic areas in the states of New York and Washington, USA. Specimens were collected from March 1, 2008, through March 1, 2010.

New York State
Public Health Law in New York requires laboratories and physicians to report all salmonellosis cases to local health departments and to submit isolates to the New York State Department of Health (NYSDOH) Wadsworth Center Public Health Laboratory for diagnostic confi rmation and speciation. The local health departments submit all case information, including laboratory data and questionnaire results, to the NYSDOH by a secure electronic data collection system. Within the FoodNet (www.cdc.gov/ foodnet/) catchment area of the Centers for Disease Control and Prevention (CDC) Emerging Infections Program, surveillance offi cers actively monitor clinical microbiology laboratories and contact local health departments to ascertain all laboratory-confi rmed salmonellosis cases, and they review case reports for accuracy and completeness. This catchment area includes 34 counties in the Albany, Buffalo, and Rochester areas of New York, representing ≈4.3 million residents (22% of the total state population).

Washington State
As in New York, salmonellosis is reportable in the state of Washington, and clinical laboratories are required to submit all isolates to the Washington State Department of Health (WSDOH) Public Health Laboratories for further characterization. County health departments submit all case information to the WSDOH Communicable Disease Epidemiology Unit. The 6 participating counties in Washington were King, Pierce, Snohomish, Spokane, Whatcom, and Yakima. These included 3 of the most populous counties (King, Pierce, and Snohomish) and 2 counties with the highest concentrations of dairy cattle (Whatcom and Yakima) in Washington. Spokane County comprises an urban population in addition to rural and farming communities. The 6 participating counties represent ≈4.3 million residents (65% of the total state population).

Laboratory Methods
In New York, serotyping and pulsed-fi eld gel electrophoresis (PFGE) were performed on all Salmonella FoodNet isolates received by the NYSDOH during the study period. Typing data were forwarded to Cornell University (Ithaca, NY, USA) for PFGE pattern comparison by BioNumerics software (Applied Maths Inc., Austin, TX, USA). Confi rmed Salmonella isolates of bovine origin, obtained either from clinical samples submitted to the Cornell University Animal Health Diagnostic Center or from fi eld study samples collected from clinically ill and asymptomatic dairy cattle, were sent to the US Department of Agriculture, Animal and Plant Health Inspection Service, National Veterinary Services Laboratories (Ames, IA, USA) for serotyping by standard protocols. PFGE subtyping of bovine isolates was performed in the Food Science Laboratory at Cornell University. The standard CDC PulseNet protocol (17) was used for subtyping all study isolates.
In Washington, serotyping and PFGE were performed on all human clinical isolates submitted to the WSDOH as described for New York. Bovine isolates, obtained either from clinical samples submitted to the Washington Animal Disease Diagnostic Laboratory or from samples collected during dairy cattle fi eld studies, were also sent to the National Veterinary Services Laboratories for serotyping. PFGE subtyping of bovine isolates was performed at Washington State University, again by using the standard CDC PulseNet protocol.

Questionnaire
As part of the routine investigation of foodborne Salmonella infections, trained interviewers from local health departments in both states administered a standardized questionnaire to each patient by telephone. The NYSDOH Salmonella questionnaire was adapted from a previous version used for investigating all cases of foodborne infection. The standard WSDOH questionnaire was supplemented with an additional set of questions to ensure completeness of exposure data collection and to better align Washington data with New York data. Patient identifi cation data were removed from each dataset before being transferred to the university research group in the respective state. Data collected in each state included demographic information, clinical features, and exposure history during the 5 days before disease onset. Exposure data included animal contacts, food history, food hygiene practices, water use for drinking and recreation, health care or daycare exposures, and travel history. After data collection had ended, both datasets were compiled at Cornell University for analysis.

Case-Patients and Control-Patients
Eligible cases included Salmonella-positive patients from the NYS FoodNet catchment area and the 6 participating Washington counties that were identifi ed during the study period. Patients were excluded if they were associated with an obvious outbreak (as noted by the state health departments) or if they had a typhoidal Salmonella infection (either Typhi or Paratyphi A). For the case-control analysis, case-patients were defi ned as patients infected with Salmonella isolates that matched contemporary bovine isolates from the respective state by serovar and PFGE pattern. Control-patients were defi ned as patients infected with Salmonella isolates that were not associated with cattle, according to those criteria. Specifi cally, all patients infected with S. enterica serovar Dublin were classifi ed as case-patients because this serovar is host-adapted to cattle (18). Patients infected with 6 other serovars (Newport, Typhimurium, Infantis, 4,5,12:i:-, Agona, and Montevideo) were classifi ed as potential casepatients because of the importance of these serovars in bovine and human hosts.
According to a recent comprehensive study on the incidence of salmonellosis among dairy herds in New York and other northeastern states, the fi rst 5 serovars just mentioned were among the leading serovars shed by dairy cattle with clinical Salmonella infections (19), and Montevideo is consistently one of the most prevalent serovars shed by asymptomatic cattle (20). All 6 are among the top 20 serovars isolated from human patients with laboratory-confi rmed salmonellosis in the United States (21). For patients infected with one of the aforementioned serovars, PFGE patterns from the human isolates were compared with those from cattle. To be considered bovine associated, an isolate had to have a PFGE pattern indistinguishable from that of isolates obtained from >2 cattle in the same state from March 1, 2007, through March 1, 2010; patients infected with such isolates were thus classifi ed as case-patients. Human isolates that differed from the most similar bovine isolate by 1-3 visible bands were excluded from the analysis, as were human isolates with a PFGE pattern matching that of just 1 bovine animal. Patients infected with isolates that differed from the most similar bovine isolate by >4 visible bands were classifi ed as control-patients.
Patients infected with Salmonella serovars other than those previously listed were classifi ed as control-patients if the serovar was not detected in cattle in the same state during that time frame. If the serovar was detected in cattle, the human isolate had to differ from the most similar bovine isolate by >4 visible bands in order for that patient to be considered a control-patient; otherwise, the human isolate was excluded from the analysis. A total of 422 bovine isolates from New York and 447 bovine isolates from Washington were used for PFGE pattern comparison.

Data Analysis
Data were imported into a commercially available statistical software program (SAS, version 9.2; SAS Institute Inc., Cary, NC, USA) for variable coding and analysis. Age was converted into a categorical variable (<5, 5-12, 13-20, 21-40, 41-60, and >60 years of age). Animal, food, and other exposures were analyzed as dichotomous variables (yes/no). The variables "farm animal contact" and "bovine contact" were created to most effectively capture data from 2 state health department questionnaires that were not identical. In the New York dataset, farm animal contact was considered "yes" if the patient reported an occupation of animal farming or a history of farm animal contact; bovine contact was considered "yes" if the patient specifi ed cattle as the type of farm animal. In the Washington dataset, farm animal contact was considered "yes" if the patient reported a history of living or working on a dairy or other farm type or reported a history of contact with cattle, sheep, goats, horses, or pigs; bovine contact was considered "yes" if the patient specifi ed cattle as the type of farm animal.
Analysis was performed to compare exposures between case-patients and control-patients. Univariable descriptive analysis was performed on all explanatory variables. Bivariable analysis with the χ 2 test was used to determine whether each variable was independently associated with case or control status. Multivariable logistic regression models were used to identify risk factors for infection with bovine-associated subtypes; case or control status was used as the dichotomous outcome variable. Initial selection of variables was based on the bivariable analysis screening (p<0.25), and a backward elimination approach was used to identify a fi nal multivariable model; values of p<0.05 were considered signifi cant. Relevant 2-way interaction terms (involving exposure variables retained in the fi nal model, demographic variables, and state) were also investigated for signifi cance within each model. Consumption of undercooked ground beef and unpasteurized milk in the 5 days before disease onset were included in each model as potential confounders. The population attributable fraction (PAF), defi ned as the proportion of disease in a population that can be attributed to a given exposure, was calculated for variables retained in each model by using the formula PAF = P(OR adj -1)/OR adj (where P = the proportion of casepatients exposed to the risk factor and OR adj = the adjusted odds ratio for that factor) (22).

Results
From March 1, 2008, though March 1, 2010, the NYSDOH received nontyphoidal Salmonella isolates from 835 patients within the NYS FoodNet catchment area. According to our criteria, 40 (4.8%) of these were classifi ed as case-patients and 356 (42.6%) as control-patients. Among case-patients, 20 (50.0%) were female; among control-patients, 215 (60.4%) were female. The median age among case-patients was 31.5 years, whereas that among control-patients was 31 years. Typhimurium was the most common serovar among case-patients (67.5%), and Enteritidis and Typhimurium were equally predominant among control-patients (10.1%; Table 1).
During the study period, 562 patients with nontyphoidal salmonellosis were identifi ed in the 6 participating Washington counties. According to our criteria, 87 (15.5%) of these were classifi ed as case-patients and 428 (76.2%) as control-patients. Among case-patients, 53 (60.9%) were female; among control-patients, 229 (53.5%) were female. The median age among case-patients was 28 years, whereas that among control-patients was 33 years. The most common serovar among case-patients was Typhimurium (51.7%), and the most common serovar among controlpatients was Enteritidis (40.4%, Table 2).
The datasets from each state were combined to yield a total of 127 case-patients and 784 control-patients. Bivariable analysis indicated that more case-patients (11.0%) than control-patients (3.8%) reported a history of farm animal contact during the 5 days before disease onset (p = 0.0004). More case-patients (6.3%) than control-patients (1.0%) also reported a specifi c history of bovine contact during the 5 days before illness (p<0.0001). Attendance at an open farm/petting zoo/fair was more common (p = 0.05) among case-patients (11.8%) than control-patients (7.0%), and more case-patients (12.6%) than control-patients (6.5%) reported a history of contact with animal manure (p = 0.01). Fewer case-patients (3.1%) than control-patients (13.9%) reported a history of international travel before illness (p = 0.0006). Case-patients and control-patients did not differ signifi cantly with respect to sex, age group, eating undercooked ground beef, or drinking unpasteurized milk.
Multivariable logistic regression analysis showed that a history of farm animal contact during the 5 days before disease onset was signifi cantly associated with being a case-patient (odds ratio [OR] 3.2, 95% CI 1.6-6.4, p = 0.0008), after consumption of undercooked ground beef and unpasteurized milk was accounted for (Table 3). A specifi c history of bovine contact during the 5 days before illness was also signifi cantly associated with being a casepatient (OR, 7.4, 95% CI 2.6-20.9, p = 0.0002), according to estimates from a separate logistic regression model that also controlled for those food exposures (Table 4). International travel was negatively associated with being a case-patient in each of the models. No signifi cant interaction was found between farm animal/bovine contact and state, sex, or age group in the respective models. The PAF, applied here as the proportion of Salmonella infections among the source population of laboratory-confi rmed cases that can be attributed to a certain exposure, was calculated to be 7.6% for farm animal contact and 5.4% for bovine contact in particular.
To perform a sensitivity analysis for testing the effect of our strict case defi nition, we repeated multivariable logistic regression models under 2 extreme scenarios; all potential case-patients that were excluded (because the isolate differed from the most similar bovine isolate by 1-3 visible bands or because its PFGE pattern matched that of just 1 bovine animal) served alternatively as casepatients (scenario 1) and control-patients (scenario 2). The parameter estimates and ORs from these hypothetical models were comparable to those obtained from our original analyses (the ORs under scenarios 1 and 2 were 3.2 and 2.3 for farm animal contact, 6.1 and 4.0 for bovine contact, respectively).

Discussion
The case-case study design proposed by McCarthy and Giesecke is an adaptation of the conventional casecontrol approach (16). It has been used to study risk factors and clinical features associated with particular subtypes of Salmonella spp. (8,(23)(24)(25)(26), Campylobacter spp. (27,28), and Clostridium diffi cile (29). One of its main advantages is the removal of selection bias imposed by the surveillance system; case-patients and control-patients were subjected to the same selection process in order to be detected by a state health department as a laboratory-confi rmed case. Another advantage is the negation of recall bias (a form of information bias); because case-patients and controlpatients had salmonellosis, their recall of exposures should have been similarly affected by attitudes regarding causation. A potential limitation of the case-case study design is that the control-patients might not represent the exposure prevalence in the source population on account of the unique exposures that led them to become infected. However, we believe that we addressed this issue by including a diverse array of serovars and PFGE types in the control group, assuming that their associated exposures were presumably also diverse and thus more representative of the total spectrum of exposures associated with nonbovine Salmonella strains. Another possible drawback of this study design is that case-patients and control-patients share a certain subset of exposures that pose a risk for Salmonella infections in general; such exposures will therefore remain unidentifi ed or at least be underestimated as risk factors. Although this study design precludes the study of general risk factors for salmonellosis, it is useful for investigating exposures that are serovar or subtype specifi c.
Other studies have found an association between salmonellosis and having previous contact with either cattle or a farm environment (8,9,30). Our study investigated this association with a case-case approach that used a strict case defi nition. Another strength of this study was the use of sporadic cases of salmonellosis rather than cases associated with outbreaks. Insight regarding the epidemiology of sporadic Salmonella infections has traditionally been limited because specifi c sources of enteric illness are seldom identifi ed when not occurring as part of an outbreak.
Direct contact with dairy cattle or their environment during the 5 days before illness onset was signifi cantly associated with salmonellosis caused by a bovine-matched subtype in New York and Washington. Because there was no interaction between state and the animal contact variables in the models, we concluded that the effect estimate was consistent across the 2 states. The ORs for farm animal contact and specifi c bovine contact in each state were also similar to those obtained from analysis of the combined dataset (data not shown). In addition, our sensitivity analysis led us to decide that we still would have reached the same conclusions with modifi ed case criteria. These results have important implications for dairy farm workers and their families, veterinarians and veterinary staff, and those who interact with dairy cattle in public settings.
Although attendance at an open farm/petting zoo/fair was not signifi cantly associated with being a case-patient in this study, it is logical to believe that visiting such a facility might increase the risk for salmonellosis, on the basis of our other fi ndings. S. enterica is transmitted primarily by the fecal-oral route. Direct contact with the feces of infected cattle can occur through feeding, petting, or otherwise handling them; contaminated clothing or footwear, animal bedding, barriers, or other environmental surfaces can also be sources of infection (15,31). This threat is underscored by the recent fi nding that the median duration of fecal Salmonella shedding following clinical disease among dairy cattle is 50 days (32). The negative association between recent international travel and salmonellosis caused by a bovine-matched subtype was anticipated. Although travel outside the United States is a well-known risk factor for Salmonella infections (33) (observed in a signifi cantly higher proportion of control-patients in this study), it would not be expected to have an association with salmonellosis caused specifi cally by subtypes shared by dairy cattle in New York and Washington.
The percentage of Salmonella infections in the United States that are foodborne was recently estimated at 94% (1). The results of our statistical analyses suggest that this percentage might be an overestimate, at least for bovine-associated Salmonella subtypes, although our PAF results (which also take into account the frequency of exposure) are more consistent with this estimate. It also must be noted that the effect of animal exposure observed in New York and Washington might not be representative of the rest of the country. Nevertheless, more human infections originating from bovine sources might result from direct contact with cattle (as opposed to foods of bovine origin) than previously recognized. Clear evidence for the role of direct farm animal contact as a source of human salmonellosis indicates that it is imperative for Salmonella control efforts to include a focus on transmission routes other than foodborne. The effi cacy and public health impact of addressing nonfoodborne transmission of Salmonella spp. have been demonstrated by studies of direct contact transmission from pet turtles to humans, particularly children. In response to studies that established turtles as an important source of human salmonellosis (34,35), federal legislation in 1975 prohibited the sale and distribution of turtles <4 inches in carapace length. Still in effect today, this ban coincided with an 18% reduction in Salmonella infections among children 1-9 years of age (36). A number of measures can be taken to minimize the likelihood of becoming infected with Salmonella spp. from direct contact with farm animals. Washing hands with soap and water is a simple yet highly protective step that can be taken after contact with animals or feces (15,37). Children <5 years of age, elderly adults, and immunocompromised persons are at increased risk for invasive salmonellosis (38,39) and thus should pay special attention to hygiene or avoid certain animal contacts altogether. Veterinarians should teach cattle owners and farm employees to wash well after work or before eating, to disinfect boots and equipment, and to keep coveralls out of the house. If treating infected cattle, veterinarians must specifi cally counsel their clients about the risk for zoonoses. In particular, high-risk groups should avoid contact with infected cattle. Veterinarians should instruct their staff members to protect themselves by using appropriate infection control procedures (40), especially if working with livestock. Increased physician awareness of the role of direct farm animal contact in transmitting Salmonella spp. and other enteric zoonotic pathogens is likewise needed. Physicians should educate their patients, particularly those at increased risk for severe disease, regarding the potential threat posed by animal contact and the importance of hand hygiene after such contact. Patients with diarrheal illness should be questioned about their exposures to cattle, farm environments, and other animal species. It is also essential that those who operate open farms and other animal exhibits adhere to current guidelines by equipping such areas with handwashing facilities, preventing food and drink in these areas, maintaining an adequate cleaning and disinfection protocol, and providing visitors with educational materials on disease prevention (37). In conclusion, prevention of salmonellosis should include a focus on safe animal contact in addition to food safety measures.