and

Salmonella Dublin ( S. Dublin) is an emerging patho-gen on dairy farms in Canada. In Ontario, S. Dublin has been increasingly isolated from diagnostic laboratory samples. The objective of this observational cross-sectional study was to identify management practices associated with herd positivity for S. Dublin. A convenience sample of 100 dairy farms was visited in Ontario, Canada from April to August 2022. Farms were visited once to collect blood samples from 20 heifers between 4 and 24 mo old, sample bulk tank milk, and administer an in-person questionnaire on management practices. An additional bulk tank milk sample was collected before the visit by milk transporters. All bulk tank and serum samples underwent ELISA testing to determine Salmonella Dublin positivity (≥35% percent positivity on ELISA). Of the 1,990 heifers sampled, 44 (2.2%) animals were seropositive for S. Dublin. At least one seropositive heifer was identified on 24% of participating farms. Based on the bulk tank milk samples collected during both sampling periods, 4% of farms were positive for S. Dublin. Overall, of the 100 farms visited, 25% were classified as S. Dublin positive, meaning at least one serum or bulk tank sample was interpreted as positive. A multivariable logistic regression model identified 5 factors associated with herd-level positivity for S . Dublin. Specifically, introducing purchased animals within the last 2 years increased the likelihood that farms were positive for S. Dublin (odds ratio (OR) = 4.6). Farms that had at least one animal leave the premises for a cattle show, embryo collection center, or loan to another farm and return within the last 2 years were also at a higher risk for S. Dublin (OR = 4.9). Farms that removed manure from the surface of bedding in calving pens twice per month or after every calving were at greater risk for S. Dublin than farms that removed manure less frequently (OR = 8.5). Farms that added bedding material to calving areas once or twice weekly were at lower risk for S.


ABSTRACT
Salmonella Dublin (S.Dublin) is an emerging pathogen on dairy farms in Canada.In Ontario, S. Dublin has been increasingly isolated from diagnostic laboratory samples.The objective of this observational crosssectional study was to identify management practices associated with herd positivity for S. Dublin.A convenience sample of 100 dairy farms was visited in Ontario, Canada from April to August 2022.Farms were visited once to collect blood samples from 20 heifers between 4 and 24 mo old, sample bulk tank milk, and administer an in-person questionnaire on management practices.An additional bulk tank milk sample was collected before the visit by milk transporters.All bulk tank and serum samples underwent ELISA testing to determine Salmonella Dublin positivity (≥35% percent positivity on ELISA).Of the 1,990 heifers sampled, 44 (2.2%) animals were seropositive for S. Dublin.At least one seropositive heifer was identified on 24% of participating farms.Based on the bulk tank milk samples collected during both sampling periods, 4% of farms were positive for S. Dublin.Overall, of the 100 farms visited, 25% were classified as S. Dublin positive, meaning at least one serum or bulk tank sample was interpreted as positive.A multivariable logistic regression model identified 5 factors associated with herdlevel positivity for S. Dublin.Specifically, introducing purchased animals within the last 2 years increased the likelihood that farms were positive for S. Dublin (odds ratio (OR) = 4.6).Farms that had at least one animal leave the premises for a cattle show, embryo collection center, or loan to another farm and return within the last 2 years were also at a higher risk for S. Dublin (OR = 4.9).Farms that removed manure from the surface of bedding in calving pens twice per month or after every calving were at greater risk for S. Dublin than farms that removed manure less frequently (OR = 8.5).Farms that added bedding material to calving areas once or twice weekly were at lower risk for S.

INTRODUCTION
Salmonella enterica ssp.enterica serovar Dublin (S.Dublin) is an emerging pathogen in Canada (Nielsen, 2013a;Uyama et al., 2022;Mangat et al., 2019).This serotype is host-adapted to cattle and its associated outbreaks can lead to high levels of morbidity, mortality, and economic losses (McDonough et al., 1999;Nielsen et al., 2013).Infections can take various forms, primarily manifesting as septicaemia and respiratory disease in younger animals, or reproductive disease in older animals (McDonough et al., 1999;Holschbach and Peek, 2018).Infections can lead to high rates of mortality, especially in young calves (McDonough et al., 1999).Surviving animals can become chronically infected, continuously or intermittently shedding bacteria into the environment via feces, colostrum, or milk (Veling et al., 2000;Counter and Gibson, 1980;Holschbach and Peek, 2018).
In the 1960s, S. Dublin was the most frequently isolated Salmonella serotype in the United Kingdom and was initially identified in North America on dairy farms in California and other western United States, with more recent outbreaks occurring in the north-eastern United States (Pacer et al., 1989;Davidson et al., 2018;Afema et al., 2015;Cummings et al., 2018).In Canada, S. Dublin was first isolated in Quebec in 2011, and it is currently estimated that 6.8% of dairy farms in Quebec have been exposed to this bacterium (Um et al., 2022).The province of Ontario saw S. Dublin first isolated in 2012 and it has since been increasingly detected in diagnostic laboratory submissions and recently purchased dairy cattle (Uyama et al., 2022;Renaud et al., 2022).It is now one of the most commonly isolated serotypes in Ontario (Uyama et al., 2022) and circulating strains are largely multi-antimicrobial resistant in both cattle and humans (Uyama et al., 2022;Mangat et al., 2019).
Management practices identified to increase the risk of S. Dublin infection on dairy farms in predominantly European countries include larger herd sizes, introduction of purchased cattle, allowing cattle to return to the farm when not successfully sold, grazing cattle at other farms, presence of neighboring herds positive for S. Dublin, feeding waste milk to calves, organic certification, and frequent sighting of rodents (Nielsen and Dohoo, 2012;Nielsen and Dohoo, 2013;Ågren et al., 2017;van Schaik et al., 2002;Nielsen, 2007).Management factors associated with control of this pathogen in endemically test-positive herds include allowing a maximum of 4 cows in the calving area at one time, avoiding the use of the calving pen for recovery of sick animals, cleaning the calving area at least twice monthly, providing new bedding at least once per week, rearing calves in individual housing with non-permeable barriers between calves, and avoiding pooling of colostrum (Nielsen et al., 2012).Currently, there is a lack of data available to describe the prevalence of S. Dublin and its associated risk factors on dairy farms in Canada.
The objective of this cross-sectional observational study was to describe management practices that are associated with increased risk of S. Dublin herd positivity.We hypothesized that previous identification of S. Dublin on-farm, recent introduction of animals to the herd, return of cattle that temporarily left the premises, frequent visitors (e.g., artificial insemination technologists) and group calving pens would increase the risk of S. Dublin on-farm, whereas adequate disinfection procedures within-and between-farms, removal of calves from dam within 1-h after calving, regular removal of manure and re-bedding of calving area, and adequate quarantine of new or returning animals would be protective against S. Dublin.

MATERIALS AND METHODS
A convenience sample of dairy farms in Ontario, Canada were recruited through contact with herd veterinarians.Farms (n = 100) were visited once between April 13, 2022 and August 26, 2022 to collect serum and bulk tank milk samples, as well as verbally administer a questionnaire on management practices.An additional bulk tank milk sample was collected between November 11 and December 13, 2021 by milk transporters/graders, as part of another Ontario-wide bulk tank milk sampling study at the University of Guelph.Sampling and herd visit scheme is outlined as a Supplemental Fig-ure (https: / / doi .org/ 10 .5683/SP3/ 729ZW9).Human ethics (REB#22-02-029) and animal use (AUP#4770) approvals were received from the University of Guelph Research Ethics Board and Animal Care Committee, respectively, before beginning the study.This study is reported according to the guidelines of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE)-Vet statement (Sargeant et al., 2016).

Serum sample collection and testing
A total of 20 animals were sampled per farm, with 10 calves sampled that were between 4 and 12 mo old and 10 replacement heifers sampled that were between 1 and 2 years old.If more than 20 eligible animals were present, heifers were selected by sampling every other animal locked in head gates, or every other animal approached in loose housing systems.A 20-gauge, 1-inch hypodermic needle was used to draw 10 mL of blood from each animal into a sterile blood collection tube without anticoagulant (BD Vacutainer; Becton, Dickinson and Co., Franklin Lakes, NJ).Blood samples were allowed to clot, centrifuged at 3000 rpm for 15 min at approximately 20°C, and then frozen at −20°C.Serum was tested at the University of Guelph Animal Health Laboratory (Guelph, ON, Canada) using PrioCHECK S. Dublin Ab Strip Kit (Prionics AG, Switzerland).This assay has high correlation with the Danish inhouse Salmonella Dublin ELISA kit on bulk tank milk (r = 0.96) and serum (r = 0.85 -0.90) (Applied Biosystems, 2017).Serum was interpreted as positive for S. Dublin if percent positivity (PP) was ≥35%.The threshold was set at PP ≥35% following the manufacturer's (Prionics, Lelystad, Netherlands) recommended cut-off (Nyman et al., 2013).

Bulk Tank Milk sample collection and testing
Bulk tank milk samples were collected over 2 sampling periods (Supplemental Figure ; (https: / / doi .org/ 10 .5683/SP3/ 729ZW9).The first sampling period was from November 11, 2021 to December 13, 2021.Due to issues with sample quality, 8/100 (8%) of herds were resampled between June 4, 2022 and June 6, 2022.Samples were collected by the milk transporter using a sterile container containing one bronopol preservative tablet.Bulk tank contents were agitated before milk samples were collected.Bulk tank milk from the first sampling period were shipped fresh to the Quebec Lactanet Laboratory (Anne-de-Bellevue, QC, Canada) and underwent ELISA testing using PrioCHECK S. Dublin Ab Strip Kit without removal of the fat layer before testing, within a median of 18 d (range = 7-36) from collection.Bulk tank milk samples for the second sampling period were collected at the time of the farm visits between April 2022 to August 2022.Bulk tank milk samples were collected by a trained research technician using a sterile container containing one bronopol preservative tablet and transported to the University of Guelph on ice.Samples were frozen for a maximum of 25 d before testing at the Quebec Lactanet Laboratory for ELISA testing using PrioCHECK S. Dublin Ab Strip Kit (Prionics AG, Switzerland).A maximum of 25 d for storage was based on validation by Um et al. (2020), as this allowed for a testing window maximum of 28 d, with 3 d for the samples to be shipped and tested by technicians at the laboratory.If farms had multiple bulk tanks, we planned to collect a composite sample from both bulk tanks; however, all participating herds had a single bulk tank.Bulk tank milk samples from both sampling periods were interpreted as positive for S. Dublin if PP ≥35% following the manufacturer's recommended cut-off for individual milk samples (Prionics, Lelystad, Netherlands).A cut-off of PP% ≥ 35% was used for both sampling periods as similar measures of agreement have been observed between fresh bulk tank milk samples and samples frozen for 25 to 28 d (Um et al., 2020).

Questionnaire
During the farm visit, the person responsible for calf care was asked to verbally answer questions posed by a research technician, and their responses were entered into an online survey software (Qualtrics; https: / / www .qualtrics.com/).No advice or opinions were offered by technicians while the questions were answered to avoid influencing the respondent.The questionnaire consisted of 114 multiple-choice and fill-in-the-blank questions focusing on herd demographics, within-herd and between-herd biosecurity, calving management, calf care, colostrum management, calf housing and feeding, as well as dry and milking cow management.The questionnaire was developed through a literature review of main risk factors associated with S. Dublin infection, as well as overall calf morbidity and mortality.The questionnaire was pretested on 4 dairy farms in the Guelph area before beginning data collection.A full version of the questionnaire is available as a Supplemental File (https: / / doi .org/ 10 .5683/SP3/ 7D1YP9).

Sample Size Calculation
To study the effects of herd-level factors associated with the positivity for S. Dublin, it was determined that a sample of 100 herds would need to be recruited to allow for the estimation of a difference between the proportion of farms test-positive for S. Dublin compared with the proportion of farms test-negative for S. Dublin of 0.25 at a 5% significance level with 90% power.From each herd, 10 animals needed to be sampled to estimate the prevalence of S. Dublin in youngstock stratified by category (calves and heifers) at the 95% confidence level with 5% error rate.This was calculated based on the assumption that 6.5% of Ontario dairy herds will be test-positive for S. Dublin, with a within-herd prevalence of 25% seropositive youngstock in infected herds (MAPAQ, 2015;Nielsen, 2013b).

Statistical Analysis
Data from surveys were collected on the Qualtrics Surveys app, exported to a Microsoft Excel spreadsheet (Microsoft Corp., Redmond, WA) for collation and then moved into Stata 17 (Stata/BE Version 17.0 for Mac, StatCorp, College Station, TX) for analysis.Descriptive statistics were run for each independent and dependent variable to ensure adequate variation.Normality was assessed using the Shapiro-Wilk test for continuous variables and categorical variables were evaluated to ensure that each category contained at least 5 observations.Herds were classified as positive for S. Dublin if at least one serum sample was PP ≥35%, and/or if at least one bulk tank milk sample from either sampling period was PP ≥35%.A causal diagram was used to identify potential intervening, confounding, or distorter variables and is available as a Supplemental Figure (https: / / doi .org/ 10 .5683/SP3/ B5YVZM).Potential associations between herd-level factors and herd positivity were tested using a logistic regression model.The assumption of linearity of continuous variables was assessed by a locally weighted regression of the herd positive outcome on predictor variables.If a variable failed to meet the linearity assumption, the variable was categorized.The maximum number of cows per pen in the calving area was the only variable that failed to meet the linearity assumption and it was categorized into quartiles (1-3, 4-6, 7-15, ≥ 16).Correlation among continuous explanatory variables was tested using Pearson correlation coefficients.Correlation among categorical explanatory variables was examined using χ2 tests, with an additional Fisher's exact test if one or more categories had an expected frequency ≤5.If the correlation coefficient was ≥0.7, or the χ2 or Fisher's exact p-value ≤0.05,only one variable was retained based on reliability of measurement and biological plausibility.Univariable models were built for each explanatory variable and if statistically significant using a liberal p-value (P < 0.20), the variable was offered to a multivariable model (Dohoo et al., 2010b).There were 13 variables offered to the multivariable logistic model (Table 1), which was built using manual backward selection.Only herds with complete information on variables in the final model were included.Two-way interactions were assessed between all predictors that were found to have a statistically significant unconditional association with the outcome (P < 0.20), and those interaction terms remained in the final model if P-value <0.05 (Dohoo et al., 2010a;Dohoo et al., 2010b).No statistically significant interactions of the main effects and herd size were identified, and herd size was not statistically significant in any model.Covariates were kept in the model if P < 0.05, or if covariates were considered as potential confounders as their removal changed the coefficients by more than 20%.Model fit was assessed using a deviance χ2 test, whereas deviance residuals were used to evaluate potential outliers.No outliers were identified.Predicted probabilities were generated using the logit postestimation command in Stata 17 (Stata).

RESULTS
From April 13, 2022 to August 26, 2022, 100 Ontario dairy farms were visited where blood from 1,990 dairy heifers (median = 20 samples per farm, range = 17 to 20; Figure 1) and 100 bulk tank milk samples were collected.In addition, a questionnaire was completed at each of the dairy farms, where each question was read verbally to the farmers by a single graduate student (K.V.P.) with responses recorded in a Qualtrics form.An additional bulk tank milk sample was collected from each of the participating farms previously from November 2021 to December 2021, with 8/100 of these resampled in June 2022 due to issues with sample quality.Dairy farms were distributed across Ontario, with 79/100 (79%), 19/100 (19%), and 2/100 (2%) located in western, eastern, and central Ontario respectively.The farms milked, on average, 126 dairy cows (median = 96, range = 37 to 770) and had an average of 281 total dairy animals (median = 214, range = 67 to 1842), including cows, heifers, steer, and bulls, housed on the premises at time of visit.The average herd milk production was 3595.7 L per day (median = 2957.3,range = 888.7 to 16387.4) and the average somatic cell count was 165,700 cells/ml (median = 148,000, range = 0 to 399,000) during the month of the farm visit.An anonymized distribution of responses to the questionnaire is available as a Supplemental File (https: / / doi .org/ 10 .5683/SP3/ KUFAMC).

Serum testing
Of the 1,990 heifers sampled, 44 (2.2%) animals were seropositive for S. Dublin.The median age of sampling was 255 d (range = 94-364) and 469 d (range = 365-932) for calves and replacement heifers, respectively.At least one seropositive heifer was identified on 24/100 (24%) of participating farms.On farms where at least one serum sample tested positive for S. Dublin, an average of 2 animals were positive per farm (median = 1, range = 1 to 6).Of the 100 farms visited, fewer than 20 animals were sampled for blood collection at 8 farms (median = 19, range = 17-19) due to a lack of availability of dairy heifers within the required age range (4 mo -2 years) on-site, or samples that were discarded due to issues in processing (Figure 1).

Bulk tank milk testing
At least one of the bulk tank milk samples from either sampling period was interpreted as positive for S. Dublin for 4/100 (4%) of participating farms (Figure 2).During the November 2021 to June 2022 sampling period, 4/100 (4%) bulk tanks were positive for S. Dublin.The mean percent positivity was 1.8% (median = 0.2, range = −6.3 to 25.5) for negative farms and 58.3% (median = 56.8,range = 39.0 to 80.5) for positive farms.During the April 2022 to August 2022 sampling period, 3/100 (3%) of bulk tank samples were positive, with the mean percent positivity being 0.7% (median = −0.9,range = −6.7 to 24.5) for negative farms and 72.3% (median = 88.3,range = 35.6 to 93.0) for positive farms.Of the 4 farms considered positive for S. Dublin from the first sampling period, 3 of the same farms were considered positive in the second sampling period.

Overall positivity for Salmonella Dublin
Overall, 25/100 (25%) of participating farms had at least one serum or bulk tank milk sample test positive for S. Dublin.Of the 4 farms that had a bulk tank milk sample positive for S. Dublin, 4/4 (100%) had at least 1 heifer identified as seropositive.Of the farms that did not have a positive bulk tank milk sample, at least 1 heifer was identified as seropositive for S. Dublin at 21.9% (21/96) of these farms (Figure 3).

Associations with herd positivity for Salmonella Dublin
The variables unconditionally associated with herd positivity using a P-value <0.20 are outlined in Table 1.The variables that remained in the final model included history of S. Dublin on the farm, presence of animal introductions within the last 2 years, presence of animals that left the farm and returned within the last 2 years, frequency of manure removal from the surface of bedding in calving areas, frequency of bedding material addition to calving area, and maximum number of cows kept in the calving area per pen at a time.There were 14 herds with missing data; hence, 86/100 were included in the final model.The estimated sensitivity and specificity of the model was calculated to be 52.4% and 93.9%, respectively, with a positive predictive value of 73.3% and a negative predictive value of 85.9%.Specifically, the odds of S. Dublin identification on farms that had purchased or introduced animals within 2 years relative to the day of sampling were higher compared with farms that had no introductions (OR = 4.6; 95% CI = 1.1 -19.7;P = 0.04) (Figure 4).Farms that had at least one animal leave the premises for a cattle show, embryo collection center, or loan to another farm (i.e., temporary housing due to renovation, reproduction, or need for quota) and return within the last 2 years had higher odds of S. Dublin identification compared with farms that had no animals temporarily leave the premises (OR = 4.9; 95% CI = 1.1 -21.5;P = 0.04) (Figure 5).History of S. Dublin on the farm (i.e., the producer having knowledge of previous identification or isolation of S. Dublin from an animal on the farm) was retained as a confounder because upon removal, it changed the coefficient for presence of animals that left the farm and returned within the last 2 years in the model by greater than 20% (Dohoo et al., 2010a).The odds of S. Dublin identification were higher for farms that removed manure from the surface of bedding in the calving area at least twice per month or after every calving, compared with farms that removed manure once per month or less often (OR = 8.5; 95% CI = 1.8 -39.5;P = 0.006) (Figure 6).Farms that added, without removing, bedding material to the calving area one to 2 times per week had lower odds of S. Dublin identification, compared with farms that added bedding material less than once per week (OR = 0.1; 95% CI = 0.0-0.9;P = 0.03) (Figure 7).The odds of S. Dublin identification were higher for farms that kept 4 or more cows per pen in the calving area, compared with farms that kept a maximum of 3 cows per pen (Table 2; Figure 8).Frequency of adding bedding material to calving area confounded the association between the maximum number of cows kept in the calving area per pen and herd positivity.Hence, it was retained in the model.

DISCUSSION
This study indicates that test positivity for S. Dublin among this sample of Ontario dairy farms is high.In addition, several management practices were associated with an increased risk of S. Dublin infection.This included introduction of animals, leaving and re-introduction of cattle, removing manure from the surface of bedding in calving areas at least twice per month or after every calving, addition of bedding material to calving areas less than once per week, and keeping more than 3 cows per pen in the calving area.
The herd-level prevalence of S. Dublin identified among a convenience sample of 100 Ontario dairy farms visited in this study was 25%; however, the prevalence estimated by this study is not equivalent to a herdlevel prevalence estimate for Ontario dairy producers.In Quebec, Canada, a bulk tank milk survey estimated 6.8% of dairy farms to be positive for S. Dublin (MAPAQ, 2015; Um et al., 2022).There are limited data available regarding the prevalence of S. Dublin in other Canadian provinces.In other countries where S.
Dublin is known to be present, it is relatively common in dairy herds.In Denmark, a study of dairy cattle from 2001 to 2004 reported an overall apparent prevalence of 21-26% of herds positive for S. Dublin based on an average of quarterly bulk tank milk antibody concentrations (Nielsen et al., 2007).Most recently, a sample of dairy farms from Britain resulted in an estimated true prevalence of 40% using the Danish eradication scheme classification for herds on quarterly bulk tank milk samples (Henderson et al., 2022).It is important to note that studies and surveillance programs from other countries can vary greatly in their methodology, so international prevalence estimates can be difficult to compare.This study identified at least one heifer that was seropositive via serum sampling at 24% of participating herds, however, only 4% of participating herds were interpreted as positive on bulk tank milk.Bulk tank milk ELISA testing produces varied results which depend on a variety of considerations, such as herd-level prevalence or cow-level factors (Warnick et al., 2006;Nielsen et al., 2005).With at least one heifer identified as seropositive at 22% of farms that would have otherwise been interpreted as negative on bulk tank samples alone, this suggests that other studies are likely underestimating the number of S. Dublin positive herds if they rely solely on bulk tank milk testing.
Farms that had purchased or introduced an animal to the dairy herd within the last 2 years were more likely to be S. Dublin positive in this study.Infectious diseases are frequently introduced to naïve dairy herds through animal additions, increasing the risk of high calf mortality as dairy herds introduce more animals (Shortall et al., 2017;van Schaik et al., 2001;Johnson et al., 2011;Nielsen et al., 2010).Previous studies have also shown that purchase of cows from other herds is an important risk factor for introduction and persistence of S. Dublin on dairy farms (Trueman et al., 1996;Nielsen and Dohoo, 2012;Nielsen and Dohoo, 2013).One Ontario study found that 1% of newly arrived cattle at dairy farms were seropositive for S. Dublin, suggesting that while this prevalence may be low, all incoming cattle should be tested to avoid introduction of S. Dublin to the herd (Renaud at al., 2022).Although there are no regulations in Canada to test herds or animals at auction for S. Dublin, Denmark has operated a national surveillance scheme structured to discourage the purchase of animals from herds at high risk for S. Dublin infection since 2002 (Jordan et al., 2008).In this system, herds are classified by the probability that they are test-positive for S. Dublin based on quarterly measurements of S. Dublin antibodies in bulk tank milk, with penalties aimed at producers that purchase from high-risk herds (Jordan et al., 2008).Multiple studies have demonstrated that this system has reduced the risk of S. Dublin spread (Nielsen et al., 2011).As these systems do not exist outside of Europe, producers in Ontario should consider adopting a closed herd and eliminate future introduction of animals as this is an important risk factor for S. Dublin introduction.However, farms that need to purchase animals should test arriving cattle before introduction.
Herds that had at least one animal temporarily leave the premises and return within the last 2 years were more likely to be positive for S. Dublin.In this study, leaving the premises included visiting cattle shows, embryo collection centers, or loans to other farms.Sites such as these tend to allow for direct contact with cattle from other herds, which is known to be a risk factor for infectious disease spread (van Schaik et al., 2002).Other studies have identified grazing with cattle from other farms as a risk factor for S. Dublin infection, likely due to the fecal-oral nature of transmission (van Schaik et al., 2002;Vaessen et al., 1998).Since group housing and shared feeding areas are typically present at Ontario cattle shows, embryo collection centers, and dairy farms, the mechanism behind the association observed in this study is likely manure cross-contamination or contamination of feed with fecal matter.Allowing animals to temporarily leave the premises and isolating upon their return has been investigated as a way to mitigate the risk of infectious disease spread, but this was shown to be ineffective in preventing general infectious disease spread (Shortall et al., 2017).This study and others suggest that allowing animals to temporarily leave the premises is an important risk factor for S. Dublin infection on dairy farms, and   maintaining a truly closed herd is likely best practice to prevent S. Dublin infection (Holschbach and Peek, 2018;van Schaik et al., 2001;Johnson et al., 2011).However, farms with animals that need to temporarily leave the herd should isolate and test all cattle before re-introduction.
In this study, farms that kept a maximum of 3 cows per pen in the calving area were less likely to be positive for S. Dublin than herds that kept 4 or more cows per calving area pen.Cows are more likely to become S. Dublin carriers around the calving period, which is exacerbated by crowded maternity areas as it increases the likelihood of animals becoming exposed to S. Dublin contaminated manure (Holschbach and Peek, 2018;Nielsen et al., 2004).Farms that use group calving pens have also been found to be at higher risk for Salmonella infection than those using only individual calving pens (Losinger et al., 1995).Large dairy herds are more likely to manage group calving areas over individual calving pens, which is thought to contribute to why herd size has been identified as an important risk factor for herd-level Salmonella infection (Ågren et al.,    Adding bedding material, without removing existing bedding, to calving areas one to 2 times weekly decreased the likelihood that a farm would be positive for S. Dublin, compared with farms that added bedding material less often than once per week.This is consistent with Nielsen et al. (2012), which demonstrated that good calving area hygiene, such as adding bedding material to calving pens weekly, aids in controlling S. Dublin infection in endemically infected herds.Adding bedding material to calving areas may reduce exposure of animals by covering contaminated material, as S. Dublin can survive in fecal matter or bedding material for varying periods of time (Nielsen et al., 2012;Pilch et al., 2022;Wray and Sojka, 1981).
This study also identified that removing manure from the calving area at least twice per month or after every calving increased the likelihood that a farm would be positive for S. Dublin, compared with farms that removed manure less often.This association contradicts what we hypothesized, as maintaining good hygiene in the calving area has been shown to prevent the spread of fecal-oral pathogens within dairy farms, as is the case with Mycobacterium avium subspecies paratuberculosis, the causative agent of Johne's disease (Nielsen et al., 2012;Donat et al., 2016).Removal of manure from calving areas may increase the spread of bacteria throughout the barn, and this observed association may suggest that manure could be left in calving areas for longer periods of time, if management practices focus on keeping the surface dry and consistently covered with fresh straw.This association could also be because farms that added bedding material to calving areas, without removing existing bedding, also removed manure from calving pens less often, or farms that were already aware of S. Dublin presence on their farm may have been more likely to have received recommendation from their herd veterinarian to increase manure removal in the calving area.Maintaining good calving area hygiene to reduce animal contact with contaminated manure continues to be best practice in reducing the risk of S. Dublin infection, however, more research is needed to better understand Ontario producers' perceived barriers to adequate calving area hygiene (Nielsen et al., 2012).Other risk factors identi- fied in this study are similar to those that are known to increase the risk of Johne's disease, as well as other infectious diseases that are fecal-orally transmitted in calves.If the risk factors described are indeed causally related to S. Dublin infection on dairy farms, then changes to management practices to reduce S. Dublin infection could have broader benefits for producers.
There are some limitations in how this study was conducted.First, herds were not randomly selected, and dairy farms from Central and Northern Ontario were less represented in this study.This study visited only 100 of Ontario's 3,298 dairy farms that were actively shipping milk in 2022 (Statistics Canada, 2022).As well, herd size was above the Ontario average, with a higher number of milking cows (n = 126) than the average Ontario dairy farm (n = 75 to 95; DFC 2021).Another limitation to this study is that serum ELISA testing demonstrates good diagnostic performance in animals 100-300 d old, but only moderately good in animals greater than 300 d old (Nielsen and Ersbøll, 2004).Blood was collected from animals 100 to 730 d old for this study, which means that serum ELISA test sensitivity and specificity was not as high for animals over 300 d old.As well, there is likely some misclassification bias in this study as some samples may be incorrectly classified as seropositive for S. Dublin because cross-reactivity with other Salmonellae serotypes can occur with indirect serum ELISA testing (Nielsen 2003).Another limitation is that the test-positivity resulting from indirect ELISA methodology used in this study can indicate new or persistent S. Dublin infections (Nielsen and Dohoo 2013).The risk factors described in this study could therefore affect the probability of introduction to a herd, or the duration of recovery from an existing infection.The final limitation to consider is that bulk tank milk from both sampling periods was collected and tested using different sampling protocols.Bulk tank milk from the first sampling period was tested fresh, and milk from the second sampling period was frozen for 25-28 d before testing, however, excellent concordance has been observed between fresh bulk tank milk samples and those frozen for 25-28 d (Um et al., 2020).
This study was conducted using a convenience sample.Because of this, future research to establish the apparent and true prevalence of S. Dublin in Ontario is recommended to better understand the burden of this pathogen.In addition, this study was conducted on dairy farms exclusively; however, further research exploring the burden of and risks factors associated with S. Dublin on beef, dairy-beef, and veal farms in Ontario would be beneficial.This is especially important for the dairy-beef and veal industries, as these industries are closely linked to the dairy industry, with presumably different risk factors for S. Dublin infection as management practices vary greatly.

CONCLUSIONS
It was frequent for farms to test positive for Salmonella Dublin on a convenience sample of dairy farms in Ontario, Canada.This study also found that the risk of S. Dublin infection on dairy farms can be reduced by avoiding purchase or introduction of animals, avoiding temporary leave of cattle to shows, embryo collection centers, or loan to other farms, and keeping a maximum of 3 cows per pen in the calving area.
Perry et al.: Risk factors for…

Figure 1 .
Figure 1.Distribution of serum test results and number of animals tested for each participating farm.Red bars indicate the number of animals that were seropositive for S. Dublin, using a percent positivity cut-off of 35%.Grey bars indicate the number of seronegative animals tested.
Figure 2. Boxplots displaying ELISA values for individual serum samples (n = 1,990, left) and bulk tank milk samples (n = 200, right), by test interpretation using a percent positivity cut-off of 35%.Whiskers indicate upper and lower adjacent values.

Figure 3 .
Figure 3. Distribution of the percent positivity (PP) measured by individual serum samples from heifers (black dots) and the bulk tank milk percent positivity (PP) measured during the farm visit (lavender bars).Each gridline represents a participating, anonymized herd.The red reference line represents the percent positivity cut-off (35%) for serum and bulk tank milk; percent positivity equal to or greater than 35% was classified as positive.

Figure 4 .
Figure 4. Predicted probability of Salmonella Dublin herd positivity by history of purchased or introduced animals within 2 years before date of study visit.Whiskers indicate 95% confidence interval.

Figure 5 .
Figure 5. Predicted probability of Salmonella Dublin herd positivity by history of animals temporarily leaving the farm (i.e., to cattle shows, embryo collection centers, or loan to another farm) within 2 years before date of study visit.Whiskers indicate 95% confidence interval.
Figure 6.Predicted probability of Salmonella Dublin herd positivity by frequency of removal of manure from surface of bedding in calving areas.Whiskers indicate 95% confidence interval.

Figure 7 .
Figure 7. Predicted probability of Salmonella Dublin herd positivity by frequency of addition of bedding material to calving areas.Whiskers indicate 95% confidence interval.
2017).Previously,Nielsen et al. (2012)  showed that a maximum of 4 cows in the calving area was associated with successful control of S. Dublin in endemically infected herds, which is similar to what was observed in this study.Based on this study and previous literature, keeping more than 3-4 cows per pen in the calving area could be an important risk factor for S. Dublin infection.

Figure 8 .
Figure 8. Boxplot of predicted probability of Salmonella Dublin positivity for farms that allowed a maximum of 1-3, 4-6, 7-15, or ≥16 cows per calving pen, showing the median and interquartile range in the colored boxes and the 95% confidence interval in the whiskers.

Table 1 .
Perry et al.: Risk factors for… Results of unconditional logistic regression models evaluating risk factors for Salmonella Dublin from a single interview of 100 Ontario dairy farms, visited from April 13 to August 26, 2022, based on serum samples collected at the time of visit and bulk tank milk samples collected from November 11, 2021 to August 26, 2022

Table 2 .
Results of final multivariable logistic regression model evaluating risk factors for Salmonella Dublin from a single interview of 100 Ontario dairy farms, visited from April 13 to August 26, 2022, based on serum samples collected at the time of visit and bulk tank milk samples collected from November 11, 2021 to August 26, 2022