The effectiveness of biosecurity interventions in reducing the transmission of bacteria from livestock to humans at the farm level: A systematic literature review

Abstract Zoonotic bacterial infections are a health hazard for people who are in regular contact with livestock at the farm level. Improved biosecurity can limit zoonotic pathogen transmission within farms. The aim of this review was to summarize the effectiveness of farm‐level biosecurity interventions in reducing bacterial transmission from animals to people who lived, worked in or visited farms. A systematic literature review was conducted using Embase, Ovid Medline and Agris databases, which were searched on 7th of July 2019, limited to English language papers but with no time exclusion criteria. A narrative synthesis was undertaken utilizing the Centre for Reviews and Dissemination approach, reported in accordance with Preferred Reporting Items for Systematic Reviews and Meta‐Analyses guidelines. Risk of bias within and across the included studies was performed using established checklists. Out of 869 studies retrieved through database searches, 11 studies were selected. In addition, three studies were found through study reference lists. Fourteen studies were therefore included in this review. Biosecurity interventions were grouped into five categories: hand washing, sanitization and hygienic measures (six studies); personal protective equipment (five studies); vaccination (two studies); other interventions (e.g. air ventilation flap) (four studies); and routine farm activities (two studies). Across studies that investigated odds of human colonization or infection (three studies), odds were seen to both be increased and decreased through use of tested biosecurity measures. Large confidence intervals that often crossed the threshold of an odds ratio equal to 1 were found. Most of the studies' overall risk of bias was ‘medium risk’ (11 studies), with selection bias domains generally being scored ‘medium risk.’ Biosecurity interventions are potentially beneficial in reducing bacterial transmission from animals to humans. However, more high‐quality evidence is needed to increase certainty in which interventions, in which contexts, are most effective from the human health perspective.


| Rationale
Zoonotic bacterial infections are a serious public health hazard, as well as an occupational risk for farm workers who are in regular contact with livestock. Previous research indicates that there is transmission of methicillin-resistant Staphylococcus aureus (MRSA) from animals to farmers or those in direct contact with livestock (Graveland et al., 2011;Pletinckx et al., 2013;Van den Broek et al., 2009). In addition, risk of infections from microbes such as Campylobacter and Brucella species has been found to be increased in people who are in direct contact with animals (El-Tras et al., 2015;Te-Chaniyom et al., 2016). Increased zoonotic infections, especially antibiotic resistant ones such as MRSA, may increase the burden on healthcare systems and reduce agricultural labour force productivity (Grace et al., 2012). Therefore, understanding how to reduce their emergence and transmission is important from a public health and economic perspective.
Biosecurity is defined by the Irish Department of Agriculture, Environment and Rural Affairs as 'The prevention of disease causing agents entering or leaving any place where they can pose a risk to farm animals, other animals, humans, or the safety and quality of a food product' (Department of Agriculture, Environment, & Rural Affairs, 2019). Improved biosecurity is an often-applied intervention to limit pathogen transmission within and between farms. These interventions can also affect transmission of zoonotic pathogens to humans, thus having a beneficial impact on public health. Types of biosecurity interventions and their effectiveness to reduce transmission of bacterial infections from animal to farmers are available (Schimmer et al., 2012). However, the effectiveness of these interventions in terms of the reduced transmission of bacteria to humans has not been systematically analysed or assessed in terms of risk of bias. Understanding the effectiveness of such interventions, alongside the strength of evidence, is important for selecting efficient interventions from a One Health perspective.

| Aim and objectives
The aim of this review was to summarize the effectiveness of biosecurity interventions in reducing the transmission of bacterial infections from livestock to humans at the farm level.
To meet the aim of this review, the following objectives were set: 1. To collate and describe evidence that assesses the effectiveness of biosecurity interventions in reducing bacterial transmission from animals to people at the farm level. Effectiveness was defined in terms of bacterial (colonization or infection) transmission, prevalence, incidence, intervention cost-effectiveness and/or cost benefit.
2. To assess the risk of bias within and across the current evidence on the effectiveness of biosecurity interventions on the transmission of bacteria from animals to people, within farms.

| Study design
This is a systematic review guided by the Center for Reviews and Dissemination (CRD) report on conducting systematic reviews (Centre for Reviews & Dissemination, 2009), in particular applying its narrative synthesis framework. Reporting is in line with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist (Moher et al., 2009) (see Appendix S1 Table 1.1 for the completed checklist).

| Eligibility criteria
The inclusion and exclusion criteria (Table 1)

Impact
• Understanding the effectiveness of farm-level biosecurity interventions from a human health perspective allows farmers and policymakers to select the most effective biosecurity measures, not only in terms of animal health but also in terms of public health.
• The majority of studies that were found estimated the impact of interventions directly targeting the humananimal interface (including hygiene and personal protective equipment), and were based in Europe.
• Across all intervention groups, there was evidence of reduced human bacterial colonization and/or infection outcomes, but alongside high uncertainty in effect size and direction in studies investigating odds ratios.

| Search strategy
There were five concepts of search terms used in the search strategy: biosecurity, interventions, livestock, farm workers and effectiveness. A comprehensive list of relevant synonyms and mesh terms (subheadings) was used for each term. Boolean operators such as AND and OR were used to build the search. In addition, truncation signs and wild cards were used to broaden the search. All search terms used were consulted with an information specialist of London School of Hygiene and Tropical Medicine library for any recommendations. Search terms for AGRIS database were (Biosecurity) AND (Intervention* OR polic* OR measure* OR strateg* OR methods OR procedures OR techniques) AND (Farmer OR farm* OR farm-level OR livestock OR "Farm animal*") AND (Effectiveness OR incidence OR prevalence OR cost-effectiveness OR cost-benefit OR outcome* OR "bacterial infection" OR "bacterial colonization" OR "bacterial colonisation" OR colonization OR colonisation OR "microbial colonization" OR "microbial colonisation" OR "cost-effectiveness analysis" OR "cost-effectiveness ratio" OR "cost-benefit ratio" OR "cost-benefit analysis" OR "cost analysis"). A full list of search terms for each database used can be found in Appendix S2.
English language filters were utilized. No search limits were applied on publication dates or quality of studies. Relevant references from the selected studies were identified, consulted and included in the review if suitable.

| Study selection
All studies were exported from Web interfaces of the databases into an Excel sheet using the 'CSV' options. Removal of duplicates was conducted using the Microsoft Excel 'Remove Duplicates' function (Microsoft Corporation, 2010) based on author and title, then manually for any remaining duplicates. The inclusion and exclusion criteria (Table 1) were then applied at two stages ('title and abstract' screening, then 'full-text' reviewing) by two independent reviewers. The first reviewer reviewed 100% of the retrievals at each stage. The second reviewer reviewed 25% for the 'title and abstract' stage, and 27% for the 'full-text stage.' This is due to articles that could not be accessed being re-reviewed through titles and abstracts, changing the denominator of full texts. Agreement at the first stage was (81.75%). Studies in which disagreement occurred were discussed. It was agreed that all studies that at least one reviewer had included should enter the next stage of review to reduce the likelihood of wrong exclusion. Agreement was (100%) for the full-text stage.

| Data collection process
A data extraction table was constructed by the first reviewer.
It was then refined and agreed upon between the first and second reviewer. It was developed in Microsoft Excel (Microsoft TA B L E 1 Inclusion/exclusion criteria applied using the 'Population, Intervention, Comparison, Outcome, and Study design (PICOS)' framework Corporation, 2010), with data items as described in Section 2.6.
The first reviewer extracted the data from the included studies. Any studies with unclear results (as determined by the first reviewer) were checked by the second reviewer (four out of 14 studies).

| Data items
In line with the agreed data extraction

| Risk of bias in individual studies
Risk of bias assessment of the included studies was conducted using the National Institute of Health tool for cross-sectional, cohort, case-control and pre-post study designs (National Heart, Lung, & Blood Institute, 2019 Risk of bias was assessed for each study according to the presence of selection bias, information bias (measurement bias) and confounding as explained by the tool guidance (National Heart, Lung, & Blood Institute, 2019). For each criterion of bias, a scoring of low, medium or high risk was given according to how many biases were found for the same criteria (see Appendix S6). For example, if two types of selection bias were assessed (such as responder bias and selection of participants from different populations), studies that had only one type were scored as 'medium risk of selection bias' and those with more than one type were scored as 'high risk of selection bias.' A high-risk score was also given for studies with no control of confounding.
This review assessed overall study risk of bias as: 1. A study including two or more high risk of bias indicators was considered to have an overall high risk of bias.

A study including one or more medium risk of bias indicators was
considered to have an overall medium risk of bias.

A study including two low-risk and one high-risk indicators was
considered as overall medium risk.

A study including only low risk of bias indicators was considered
to have an overall low risk of bias.
The CASP tool had no scoring system; however, the appraisal depended on three main questions, which were addressed whilst completing the checklist for economic evaluation studies: (a) Is the economic evaluation valid? (b) How were consequences and costs assessed and compared? and (c) Will the results help in purchasing for local people? (Critical Appraisal Skills Programme, 2018).

| Summary measures
The primary outcome was the reduction of bacterial infection/ colonization transmission from animal to people with odds ratios (ORs), relative risks (RRs), prevalence, colonization of the infection as principal summary effect measures. In addition, the costeffectiveness ratio and cost-benefit ratio of the intervention in reducing bacterial infection transmission were used as summary effect measures.

| Synthesis of results
Data were combined and compared using a narrative synthesis approach, in line with the CRD guidance (Centre for Reviews & Dissemination, 2009). No meta-analysis was conducted due to heterogeneity in bacteria and measures of impact used in each study. First, an initial description of the studies' characteristics was done, alongside a preliminary synthesis of the results of the included studies. Studies were grouped by intervention type, and results were tabulated for visual presentation. Interventions were considered protective or not based on the stated effect sizes. For OR and RR, ORs < 1 and RRs < 1 were interpreted as protective effects, whilst ORs > 1 and RRs > 1 were interpreted as harmful effects. For prevalence measures (such as proportions), a decrease in prevalence of the outcome was interpreted as protective. p-values (p < .05) and 95% confidence intervals (CIs) were used to describe statistical significance. As the narrative synthesis is an iterative process (Centre for Reviews & Dissemination, 2009), a theory of change linking the discussed biosecurity interventions to the outcomes of interest was then drawn out. To explore the relationship within and between studies, forest plots of available ORs were created, grouped by intervention type. No singular summary measure per intervention was given, as a meta-analysis was not performed due to study characteristic heterogeneity. In addition, studies for interventions and infection types were compared per type of farm.

| Risk of bias across studies
Risk of bias across studies was assessed through grouping studies by intervention type and outcome type (Cochrane, 2011). Across each of these groups, general trends were highlighted in overall study bias and individual criterion of bias.

| Study selection
Initially, 869 studies were identified from the three databases. After deduplication, 550 unique studies were retrieved, which were then screened by title and abstracts. Three hundred seventy-five studies were excluded after title/abstract screening, and 175 studies were full text reviewed ( Figure 1). Of these, 11 studies were selected to be included. In addition, three eligible references were identified from reference lists of these studies. Therefore, a total of 14 studies were included in this review.

| Study characteristics
Study characteristics such as year, aims and objectives for each study are represented in Table 3.1 in Appendix S3.
The study locations were classified according to World Bank regional groupings (The World Bank, 2019). From the Europe and central Asia region, two studies were from the United Kingdom (UK) (Ridley et al., 2011;Williams et al., 2013), two from the Netherlands (Schimmer et al., 2012(Schimmer et al., , 2014 and one from Belgium (Pletinckx et al., 2013). Whilst for the Middle East and North Africa region, two studies were from Egypt (Elmonir et al., 2019;El-Tras et al., 2015).
Nine cross-sectional studies, one cohort study, one case-control study and two pre-post studies were included. Additionally, one economic evaluation study was included.

| Risk of bias within studies
Out of the 14 studies included in this review, 13 used the National Institute of Health risk of bias tool, whilst one study was assessed using CASP tool for economic evaluation study design (Zinsstag et al., 2007). Please refer to Appendix S5 for individual criterion of bias and overall risk of bias, and Appendix S6 for individual study risk of bias checklist results.

| Results of individual studies
Among the 14 studies reviewed, five main types of biosecurity interventions were identified, which were as follows: (a)  Appendix S4 provides a detailed table with individual study results.

| Hand washing, sanitization and hygienic measures
Bacterial count, prevalence and cases were found to be significantly lower from hand washing in studies based in Canada (Racicot et al., 2013) and the UK (Ridley et al., 2011). One Canadian study considered four different protocols, where all had a statistically significant, positive effect on initial hand contamination for total coliform count and total aerobic bacterial count with p < .0001 (Racicot et al., 2013).
All hands washed were found negative for Salmonella with each of the protocols tested (Racicot et al., 2013). One UK-based study found a reduction of Campylobacter prevalence from 14% to 10% in the hands of poultry catching members was found to be significantly associated with hand washing, p = .002 (Ridley et al., 2011), whilst another poultrybased study found a 33% prevalence of Salmonella in hand swabs from farm attendants. Poor hand washing practices by attendants were considered as a risk factor for Salmonella prevalence (Abdi et al., 2017).
On the other hand, a different Canadian study did not find significant effects of hand washing on reducing C. burnetii infection/colonization (p = .19) (Meadows et al., 2016). This paper found results that were intuitively hard to accept, with 'frequent and infrequent hand-washing behaviour' and 'always hand washing only after assisting with normal birth' being shown as having a positive odds of C. burnetii seropositivity, with very wide CIs indicating uncertainty.
However, these factors were not explored in the multivariate analysis (see Table 4.1 Appendix S4 for full results) (Meadows et al., 2016).
Other hygienic measures were found to reduce the prevalence of bacterial infections. In poultry farms, poor cleaning and disinfection was found to be associated with a higher prevalence of In the UK, a reduction of Campylobacter prevalence from 41% to 19% in the footwear of poultry catching members was found to be significantly associated with footwear disinfection (p = .002) (Ridley et al., 2011). Whilst, in the USA, a study found that work laundry separation was found to be a possible risk factor for MRSA infection prevalence in pork producers, however this result was not statistically significant (p = .11) (Leedom Larson et al., 2010).

| Personal protective equipment
In relation to gloves, two studies were found. Meadows
In a Q fever outbreak in a goat farm in Australia, human Q fever

| Other specific interventions
Three studies investigated interventions linked to air flows and filtration (Bond et al., 2016;Pletinckx et al., 2013;Schimmer et al., 2012). A HEPA filter was assessed in an Australian goat farm (Bond et al., 2016). The risk of C. burnetii infection among administrative staff in unfiltered adjoining offices and among workers regularly handling goat and kids was around five times the risk of infection among workers in a HEPA-filtered factory (Bond et al., 2016).
Another study showed that dairy goat farmers using air ventilation flaps had lower odds of C. burnetii infection compared to those not using air ventilation flaps in univariate analysis (OR: 0.52, 95% CI: [0.5-1.11]); however, evidence of association was deemed not statistically significant (p = .10) (Schimmer et al., 2012). In Belgium, a study looking at pigs, mixed broiler-pig farms and mixed dairy-pig farms found that a pig farm with separate buildings and separate air flow seemed to have lower prevalence of MRSA in farmers compared to farms without, though formal associations were not tested (Pletinckx et al., 2013).
Farmers using screen/gauze in stables and those using wind-

| Routine farm activities
This type of intervention included changes to routine farmer practices that may affect bacterial transmission and thus could act as a biosecurity measure. In the Netherlands, dairy goat farmers' odds of C. burnetii infection was estimated to be significantly associated with some farm activities (Schimmer et al., 2012). This included the use of silage, use of maize, milking goats, feeding goats, supply and removal of dairy goats or bucks, care for general animal health, birth assistance, spread manure and clean the stables (p < .01), use of artificial insemination (p = .04) and remove manure (p = .01

| Synthesis of results
The Of the studies included in this review, three studies reported results using ORs. These were extracted and visually represented in Figure 4, a forest plot to explore the relationship within and between studies. If multivariate analyses were available, those were chosen above univariate analyses values. One study (discussed in detail in Section 3.4; PPE) presents ORs but based on PPE being an effect modifier, and so results are not directly comparable with the ORs presented in Figure 4; hence, it's exclusion from the plot (Williams et al., 2013). The figure shows many large CIs that indicates some uncertainty and many of these CIs are crossing the threshold of OR = 1.
There is less uncertainty in the 'farm activities' category; however, all these results come from one study (Schimmer et al., 2012).

| Risk of bias across studies
Risk of bias across studies for each outcome can be found in Table 2 (see Appendix S5 were looking at the proportion of colonization and were 'medium risk'. In general, these studies were scored 'medium risk' on the selection bias domain of the checklist. For PPE, the majority of studies were looking at the OR of colonization and were generally scored medium risk of bias. Most of these studies scored 'high risk' on the information bias domain and low risk on the 'confounding' domain. For the 'other interventions category,' the studies were generally medium risk of bias for all of the outcomes measures, in addition to being all scored low in the 'confounding' domain. Only one study per type of outcome measure (one proportion-estimate and one oddsbased estimate) was found in 'farm activities,' see Table 2 for their risk of bias. where (Grace, 2015). More global evidence on the effectiveness of biosecurity interventions in terms of human health and productivity will help the international agenda of antibiotic stewardship from a One Health perspective.

| Summary of evidence
Only one economic evaluation study was found (Zinsstag et al., 2007), with such information being crucial for decisionmakers to consider the cost and utility impacts of different   (Dixon et al., 1999;Morwal & Sharma, 2017;Theon et al., 2006).
Our results show the potential importance of hand washing and hygienic measures as a biosecurity intervention. These results agree with other studies of interventions from an animal perspective where poor cleaning and disinfection was associated with higher risks of poultry flock infections (Cardinale et al., 2004;Johnsen et al., 2006;Newell et al., 2011). Also, results for PPE as biosecurity interventions agreed with other studies where wearing clothes and boots was associated with a risk reduction of infections in poultry flocks (Bouwknegt et al., 2004;Newell et al., 2011). However, more robust evidence is needed before we can recommend such interventions on the basis of human health impact.

| Strengths and limitations
This is the first systematic review summarizing and quality assess- Though a systematic review protocol was used, only around 25% of title/abstract and full articles reviewing was done by the 2nd reviewer, which may add some risk of bias in study selection. However, inter-rater agreement was above 80% over both the abstract and fulltext stages, indicating overall low selection bias. Given the hypothesized lack of studies that would meet the criteria, the reviewers erred on the side of caution for any subjective cases, reducing the likelihood of wrongly excluded studies at review stage. In addition, studies were not excluded based on quality; however, this was due to the scarcity of studies found about effectiveness of biosecurity interventions.

| CON CLUS IONS
Zoonotic bacterial infections are a serious health risk for people in contact with livestock farms. This review suggests that biosecurity interventions may help in reducing bacterial transmission from livestock to humans, with studies found investigating the impact on hand hygiene, PPE, vaccinations, farm factors and farm activities. However, high-quality evidence at a global level is needed before strong conclusions can be drawn on which of these intervention types is the most effective for improving human health outcomes, particularly in low-and middle-income countries.
Additionally, as the economic evidence is currently lacking, recommendations based on cost-effectiveness or cost-benefit results cannot be given. Therefore, we recommend that future studies on biosecurity interventions at the farm level should include human health and economic outcomes in addition to more traditional animal colonization-based outcomes.

ACK N OWLED G EM ENTS
The authors would like to thank the consulted information specialist of London School of Hygiene and Tropical Medicine, Russell Burke, for his advice on building the search strategy.

CO N FLI C T O F I NTE R E S T
No conflict of interest to declare.

AUTH O R CO NTR I B UTI O N S
DY was the lead literature reviewer and developed the study protocol. NN was the second reviewer and helped in developing the protocol. All authors contributed to the drafting of the manuscript, interpretation and communication of the results.