Vaccinating Meat Chickens against Campylobacter and Salmonella: A Systematic Review and Meta-Analysis

Foodborne enteritis is a major disease burden globally. Two of the most common causative bacterial enteropathogens in humans are Campylobacter and Salmonella species which are strongly associated with the consumption of raw or contaminated chicken. The poultry industry has approached this issue by use of a multi-hurdle method across the production chain to reduce or eliminate this risk. The use of poultry vaccines is one of these control methods. A systematic review and meta-analysis of vaccination effects against caecal Campylobacter and Salmonella were performed on primary research published between 2009 and 2022. Screening was conducted by three reviewers with one reviewer performing subsequent data extraction and one reviewer performing the risk of bias assessment. The confidence in cumulative evidence was evaluated based on the GRADE method. Meta-analyses were performed using standardised mean differences (SMDs) with additional analyses and random effects regression models on intervention effects grouped by the vaccine type. A total of 13 Campylobacter and 19 Salmonella studies satisfied the eligibility criteria for this review. Many studies included multi-arm interventions, resulting in a total of 25 Campylobacter and 34 Salmonella comparators which were synthesised. The analyses revealed a large reduction in pathogen levels; however, many effects required statistical adjustment due to unit of analysis errors. There was a moderate level of confidence in the reduction of Campylobacter by 0.93 SMD units (95% CI: −1.275 to −0.585; p value < 0.001) and a very low level of confidence in the reduction of Salmonella by 1.10 SMD units (95% CI: −1.419 to −0.776; p value < 0.001). The Chi2 test for heterogeneity (p value 0.001 and <0.001 for Campylobacter and Salmonella, respectively) and the I2 statistic (52.4% and 77.5% for Campylobacter and Salmonella, respectively) indicated high levels of heterogeneity in the SMDs across the comparators. The certainty of gathered evidence was also affected by a high risk of study bias mostly due to a lack of detailed reporting and, additionally for Salmonella, the presence of publication bias. Further research is recommended to source areas of heterogeneity, and a conscious effort to follow reporting guidelines and consider units of analysis can improve the strength of evidence gathered to provide recommendations to the industry.


Introduction
Foodborne enteritis is a major disease burden globally and can be caused by a variety of pathogens and pollutants [1]. Of the bacterial infectors, two of the most common human enteropathogens are Campylobacter and Salmonella with case rates (per 100,000 people) of: USA 14.4 and 13.3 [2], EU 64.8 and 19.7 [3] and Australia 146.9 and 74.7 [4], respectively. Campylobacter and Salmonella food poisoning are mostly associated with the consumption or cross-contamination of chicken meat products [5,6].
Campylobacter is Gram-negative, micro-aerophilic, motile, curved rod bacteria and comprises several species with jejuni and coli being zoonotic causes of gastroenteritis in humans [7]. The chicken gut, especially the caeca, is the ideal growth environment for high levels of Campylobacter; however, Campylobacter does not remain viable outside its specific growth conditions, making it difficult to culture from the environment and to implement

Scope
This study is an updated sub-component of a larger project that explored the effects of interventions delivered at the primary production, processing or distribution channel stages in reducing Campylobacter and Salmonella levels in meat chickens. This review was performed on studies that used vaccines to address Campylobacter and Salmonella colonisation in meat chickens.

Eligibility Criteria
Studies that were included in the broad review were limited to meat chickens (live or in ovo) or products as well as surfaces exposed to meat chicken/product handling. Eligible studies applied controlled interventions at any stage during meat chicken production or processing prior to mincing, marination or any other value-adding step and measured outcomes as Campylobacter and Salmonella levels or prevalence in detail. The studies had a randomised trial design that described intervention methods in sufficient detail for implementation. Studies were primary research, published within 2009 to 2019 in English (due to insufficient budget for translation), and were not limited by region.
Data extracted from the broad review were used to further screen for studies that used vaccines or immunisation methods on meat chickens and measured endpoint caecal Campylobacter and Salmonella Additional screening was also performed on studies published within 2019 to 2022 to form this updated review.
The additional search for literature published within November 2019 to 13 July 2022 was performed on 13 July 2022 using Informit, Web of Science, USDA Agricola and AGRIS. The search domain was limited to the abstract field using search terms (broiler* OR chicken* OR gallus* OR poultry* OR "meat bird*") AND (vaccin* OR immuni* OR antibod*) AND (salmonell* OR campy*).

Screening Procedures
Two reviewers independently performed relevance screening (title and abstract screening and full-text screening) using Covidence software [21], and duplicate studies were removed at both stages.
Titles and abstracts were screened against the eligibility criteria for the population, intervention, comparator and outcome, prior to progression to full-text screening. Abstracts where both reviewers scored YES, or at least one reviewer scored UNCLEAR, moved to full-text screening. Disagreements were resolved by a third reviewer, and where consensus could not be reached, the abstract advanced to full-text screening.
At the full-text screening stage, studies were screened against all the eligibility criteria and excluded if the full text was unobtainable. Studies progressed to the data extraction/risk of bias assessment stage when both reviewers scored YES. Disagreements were resolved by discussion.

Data Extraction
Data were extracted from the included studies, and results presented in graphs were extracted using the free-to-use software ycasd version 3 as described by Gross and colleagues [22]. Authors were not contacted for missing data or for clarification of published results. Data extracted included: (1) bibliographic information: study name, authors, publication title and year, country, institution and design (i.e., level of allocating units to treatment groups); (2) population characteristics: chicken breed, age and sex; (3) intervention protocols: type, administration method, dose, frequency, duration and control type (i.e., non-treated/placebo/standard treatment); (4) challenge details (if applicable): level, time and type of challenge; (5) outcome details: method of outcome assessment, method of adjustment for non-independence in outcome data (if applicable) and reported point estimates with variability measures.
Assessment began at HIGH certainty as studies were randomised controlled tri certainty of evidence was rated down accordingly by one or two levels if the p limitations lowered the confidence in the estimated effect.

Search Result
The search and screening results are summarised in the Preferred Reportin for Systematic Reviews and Meta-Analyses (PRISMA) [34] flow diagram (Figure and abstract screening were performed on 8210 studies, of which, 404 were from to 2022 updated screening (Figure 1). A total of 1107 full texts were screened, inclu from the updated screening. Of these studies, 579 were excluded due to not satisf eligibility criteria. Reasons for exclusion of the full-text studies are summarised 1, with ineligible population (n = 143) being the most common reason for exclus lowed by inaccessible full texts (n = 121). A further 496 full-texts were excluded du satisfying the eligibility criteria for this review, resulting in a final 32 included in view, 14 of which were studies from the updated screening.   Table 1. Number of studies excluded at the full-text screening stage with justification. Note: one reason of exclusion was assigned to each excluded study. Values include studies from the updated vaccine search excluded due to article type (n = 1), irrelevant population (n = 18), wrong intervention (n = 3), irrelevant outcome (n = 9) and unclear intervention (n = 1).

Number of Studies Excluded
Justification for Exclusion of Study

Characteristics of the Included Studies
The effect of vaccinating meat chickens against Campylobacter and Salmonella was analysed in 13 and 19 studies, respectively. The characteristics of these studies are summarised in Tables 2 and 3 for Campylobacter and Salmonella. A majority of the studies investigating Campylobacter vaccines were Canadian (n = 4), whereas the majority for Salmonella were from the USA (n = 8). The remaining studies were from Australia, Belgium, Brazil, Egypt, France, India, Iran, Netherlands and Poland.
All studies used a challenged model experimental design. Studies that investigated Campylobacter vaccines used birds challenged with C. jejuni. In contrast, different Salmonella serovars (Enteritidis, n = 9, Typhimurium, n = 3, Infantis, n = 2, Heidelberg, n = 2, or more than one serovar, n = 3) were used in Salmonella vaccine trials.
Of the 13 Campylobacter studies, 7 had a multi-arm treatment experimental design, resulting in a total of 25 outcomes for analysis. Eleven of the nineteen Salmonella studies had a multi-arm treatment experimental design, leading to a total of 34 outcomes for analysis. Units of analysis across the studies were either birds or pens. All studies had a cluster-randomised experimental design, approximately half of which had a unit of analysis error.
The most common intervention used across the Campylobacter outcomes were subunit vaccines (n = 21), followed by inactivated vaccines (n = 3) and passive immunisation (n = 1). Whereas for Salmonella, 17 outcomes used live vaccines, with the remaining interventions being subunit vaccines (n = 8), passive immunisation (n = 7), an inactivated vaccine (n = 1) and a combination of a subunit and live vaccine (n = 1).
Bird age at the time of intervention varied across both Campylobacter and Salmonella outcomes, ranging from being treated in ovo to 3 weeks old. The most common age of intervention for Campylobacter outcomes was 6 or 7 days (n = 15), whereas for Salmonella outcomes, most were first treated at day of hatch or 1 day old (n = 25). Birds at the time of outcome assessment were older in the Campylobacter outcomes (starting during their third week of age) compared to Salmonella outcomes (starting during their first week of age) with 6 and 15 respective outcomes measured at multiple time points.                 (high risk of bias). 2 Studies that did not have a unit of analysis error. Note: Population age refers to the bird's age at the beginning of the intervention.

Risk of Bias
The risk of bias assessed for the eight domains was used to give an overall judgment for each outcome resulting in most being unclear (Tables 2 and 3). A noticeable lack of detailed methods and results was the main contributor for the unclear assessment across various domains, specifically, allocation concealment, blinding of personnel administering interventions and blinding of personnel assessing outcomes. No studies addressed any method of blinding or concealment in their methods. No outcomes were judged to have an overall low risk of bias.
Campylobacter had 10 outcomes judged with an overall high risk of bias (Table 2). This was due to visibly different methods of administrating treatments in a multi-arm study, such as administering one treatment via oral gavage vs. subcutaneous injection [36,39]. Another cause was not giving the control group a sham treatment to replicate a visible method of administering treatment, such as administering treatments via subcutaneous injection and not injecting birds with a diluent (e.g., phosphate-buffered saline) [40,44].
Similar reasons for high risk of bias were found in five Salmonella outcomes (Table 3) where control groups were not treated with a sham treatment method [61] or multi-arm studies administered treatments via different visible methods [54,60]. An additional study, however, was judged to have a high risk of bias for two outcomes due to selective outcome reporting [48]. Although the study measured the effects of both treatments on caecal Salmonella at days 16 and 18, the outcomes were only reported at day 16.
Overall, selective outcome reporting was the most common domain judged as having a low risk of bias across both Campylobacter and Salmonella outcomes, followed by incomplete outcome data. This indicated a common absence of detailed intervention methods.

Intervention Effects
The random effects meta-analyses for both vaccine syntheses revealed an overall reduction of Campylobacter and Salmonella in the caeca of meat chickens. Forest plots of the Campylobacter and Salmonella meta-analyses are shown in Figures 2 and 3, respectively, and summarised in Table 4.

Heterogeneity
Although the confidence intervals of the pooled effects were negative for both syntheses, significant levels of heterogeneity were present (Chi 2 p values = 0.001 and <0.001 for Campylobacter and Salmonella, respectively). The I 2 statistic was 52.4% for the Campylobacter synthesis and 77.5% for the Salmonella synthesis, interpreted as substantial heterogeneity of effects for both syntheses. The variability of effects contributed to wide prediction intervals of effects in similar studies, some of which may increase Campylobacter (−2.23 to 0.37) or Salmonella (−2.62 to 0.43).

Reporting Bias
The standardised effects were plotted against their standard errors in inversed funnel plots to identify the presence of reporting bias (Figures 4 and 5). Funnel plots for both analyses were asymmetrical, leaning towards the upper left region, most obvious in the Salmonella analysis ( Figure 5). The contour-enhanced funnel plot for the Campylobacter analysis ( Figure 4B) showed studies to have an approximately equal distribution in all levels of statistical significance. However, the gap in the middle to right region of the funnel plot for Salmonella and the missing smaller studies in regions of statistical non-significance (  (B) white/no shading: p > 10%, dark grey: 5% < p < 10%, grey: 1% < p < 5%, light grey: p < 1%.

Subgroup Analysis
Random effects subgroup meta-analyses and meta-regression were performed to investigate whether the different intervention types were a contributing factor causing the observed statistical heterogeneity. Comparators across both the Campylobacter and Salmonella analyses were grouped by the vaccination or immunisation type, and results are summarised in Table 4.
The SMDs used for the Campylobacter synthesis were categorised into three intervention types: subunit vaccine (n = 21), inactivated vaccine (n = 3) and passive immunisation (n = 1). The inconsistency of comparators, represented by the I 2 statistic, was substantial for subunit vaccines (I 2 = 57.8%) and minor for inactivated vaccines (I 2 < 0.1%).

Subgroup Analysis
Random effects subgroup meta-analyses and meta-regression were performed to investigate whether the different intervention types were a contributing factor causing the observed statistical heterogeneity. Comparators across both the Campylobacter and Salmonella analyses were grouped by the vaccination or immunisation type, and results are summarised in Table 4.
The SMDs used for the Campylobacter synthesis were categorised into three intervention types: subunit vaccine (n = 21), inactivated vaccine (n = 3) and passive immunisation (n = 1). The inconsistency of comparators, represented by the I 2 statistic, was substantial for subunit vaccines (I 2 = 57.8%) and minor for inactivated vaccines (I 2 < 0.1%).

Subgroup Analysis
Random effects subgroup meta-analyses and meta-regression were performed to investigate whether the different intervention types were a contributing factor causing the observed statistical heterogeneity. Comparators across both the Campylobacter and Salmonella analyses were grouped by the vaccination or immunisation type, and results are summarised in Table 4.
The SMDs used for the Campylobacter synthesis were categorised into three intervention types: subunit vaccine (n = 21), inactivated vaccine (n = 3) and passive immunisation (n = 1). The inconsistency of comparators, represented by the I 2 statistic, was substantial for subunit vaccines (I 2 = 57.8%) and minor for inactivated vaccines (I 2 < 0.1%).
The SMDs used for the Salmonella synthesis were categorised into five intervention types: subunit vaccine (n = 8), inactivated vaccine (n = 1), passive immunisation (n = 7), live vaccine (n = 17) and subunit with live vaccine (n = 1). A considerably high level of heterogeneity of effects was observed within the live vaccine group (I 2 = 85.5%). The inconsistency was moderate for the effects of passive immunisation (I 2 = 48.1%) and minor for the effects of subunit vaccines (I 2 < 0.1%).
The difference in SMDs across the subgroups was estimated in a random effects metaregression (Table 5). A large amount of variability remaining in the meta-regression model for Campylobacter vaccines (Residual I 2 = 54.5%) indicated that the type of interventions administered was not the major source of heterogeneity (Bonferroni adjusted p value = 1; Unadjusted p value = 0.410). The difference in effects was variable when compared to the reference subunit vaccine group; however, the number of comparators within the inactivated vaccine and passive immunisation group was much lower than the subunit group.
A high level of variability remained in the meta-regression model for Salmonella vaccines (Residual I 2 = 77.4%) with little evidence suggesting the type of interventions could explain the heterogeneity in the synthesised effect (Bonferroni adjusted p value = 1; Unadjusted p value = 0.733). Absolute deviations from the reference subunit vaccine group were less than 0.40 SMD units, except for the combined subunit and live vaccine group containing one comparator with a very wide confidence interval.

Sensitivity Analysis
To examine the effects of within-cluster correlations for studies that had a unit of analysis error, an ICC of 0.1 was used to estimate effective sample sizes. New statistically adjusted SMDs were estimated for 19 Campylobacter comparators and 24 Salmonella comparators.
A random effects meta-analysis synthesising the adjusted Campylobacter SMDs estimated a pooled SMD of −0.86 (95% CI: −1.187 to −0.523; p value < 0.001). Compared to the primary meta-analysis (Table 4), the difference in effects was small (absolute SMD difference = 0.07), with a narrower confidence interval and lower levels of heterogeneity (I 2 = 25.3%; Chi 2 p value = 0.123).
The adjusted (ICC = 0.1) random effects meta-analysis synthesising the effects of Salmonella vaccines estimated an average Salmonella reduction of 0.86 SMD units (SMD = −0.86; 95% CI: −1.176 to −0.539; p value < 0.001). The absolute difference in effects (0.24 SMD units) was slightly larger than that observed for the Campylobacter synthesis. Although a significant level of heterogeneity remained (I 2 = 43.6; Chi 2 p value < 0.001), the analysis still indicated that the use of vaccines resulted in a significantly large average reduction of Salmonella. The impact of unit of analysis errors in the syntheses was small. The magnitude of effects was reduced in both adjusted syntheses; however, a high reduction of Campylobacter and Salmonella remained.

Confidence in the Body of Evidence
The level of confidence in the body of evidence for both syntheses was evaluated using the GRADE approach. The consistent unclear assessment in the outcome risk of bias due to intervention methods led to a high overall risk of bias for both syntheses. This high risk of bias in the pooled effect reduced the confidence in evidence by one level. There was a high level of inconsistency in the syntheses that could not be explained through subgroup analysis. As the heterogeneity of effects could not be explained, the confidence in the findings of both syntheses was reduced by one level. The higher level of inconsistency in the Salmonella synthesis led to a further reduction to a very low level of confidence. The indirectness of the included studies was low as studies fit within the pre-specified range of eligible populations, interventions, comparators and outcomes. The level of imprecision of the pooled effects was also low as the 95% confidence intervals excluded the null (SMD = 0) effect of vaccinations and both upper and lower limits were negative. The confidence in the body of evidence was not reduced for indirectness or imprecision in both syntheses. Publication bias, revealed by a gap of smaller studies in areas of statistical non-significance, was present in the Salmonella synthesis. This selective suppression led to a further reduction of confidence by one level. The overall certainty of gathered evidence was increased by one level for the syntheses due to the large magnitude of effects. This resulted in a moderate level of confidence in the Campylobacter vaccine effect estimate, meaning the true effect is likely close to the estimated effect. For Salmonella, however, there was a very low level of confidence in the effect estimate, meaning the true effect is likely smaller than what was estimated.

Discussion
A systematic review and meta-analyses were performed to determine the effects of vaccines on caecal Campylobacter and Salmonella in meat chickens. The meta-analyses indicated that vaccines were effective in reducing Campylobacter and Salmonella in the caeca of meat chickens. However, the confidence in findings was limited by high levels of heterogeneity and consistent risk of bias in the study design. Publication bias was also found in the Salmonella review, further reducing the confidence in findings. This resulted in a moderate level of confidence in the estimated effects on Campylobacter and a very low level of confidence in the estimated effects on Salmonella.
Adjustment for unit of analysis errors was performed using an ICC of 0.1 to calculate effective sample sizes, and the synthesised adjusted effects were not substantially different to the unadjusted estimates. Due to the reduction of sample sizes, standard errors of the SMDs were inflated, contributing to less heterogeneity in the synthesis. It is possible that the correlation that existed within the unadjusted clustered groups may not only have exaggerated the effect of the treatments but may also have contributed to the heterogeneity between the different studies. As the ICC is likely to vary with study characteristics such as challenge status, ages of birds and interventions used [67], results should be interpreted cautiously as the ICC used may not have been appropriate for all the studies included. Although further research into estimating a variety of appropriate ICCs to analyse cluster correlated data would be useful, the prevention of unit of analysis errors at the experimental design stage is recommended for future research.
Many studies (n = 121) were excluded from the broad review due to unavailable full texts. It is unclear how many of these studies would have satisfied the eligibility criteria for this review. A larger number of studies would also have facilitated an improved subgroup analysis investigation. It is also worth noting the variety of Salmonella serovars investigated in the literature compared to the one Campylobacter species. Although approximately 90% of campylobacteriosis is caused by C. jejuni, other species such as C. coli and C. lari can also cause human illness [68].
Most of the Campylobacter vaccine types were subunit vaccines, whereas majority of those for Salmonella were live vaccines. The use of live vaccines is generally considered to be the most effective immunisation type for pathogen protection; however, it also introduces the risk of colonising the host [69]. This is especially true for Campylobacter, as chickens can very easily become colonised with high numbers of C. jejuni in the intestinal tract [37]. There is limited literature on comparing the efficacy of different Campylobacter and Salmonella vaccine types; however, evidence supports a strong but short-lived response to inactivated vaccines [70]. Key disadvantages of inactivated Salmonella vaccines are that they require the use of adjuvants, are less likely to carry their beneficial effects to progeny and carry a risk of improper inactivation [70,71]. Although subunit vaccines are comparatively newer in development and require the use of an adjuvant, subunit vaccines are composed of defined antigens making them the safest choice of the three [71].
Although the subgroup analyses did not unveil any obvious differences between the vaccine types, there were less than the recommended 10 comparators in the remaining subgroups to conclude robust findings for both syntheses [72,73]. Amongst the vaccine types, the effects of Salmonella subunit vaccines had a comparatively low level of heterogeneity with a promising beneficial effect on caecal Salmonella. While no significant differences between the subgroups were seen in the meta-regression for both Campylobacter and Salmonella, conclusions should be interpreted with caution due to the observational nature of the metaregression as well as the low number of comparators within subgroups [72,73]. There are many factors within each study setting that could have contributed to the different effects. As the intervention types were the most anticipated source of heterogeneity, no other subgroup analysis was conducted to prevent data dredging.
It is possible that many of the unclear risk of bias assessments could have been resolved if the authors were contacted for clarification, however, this review intended to evaluate the findings in the readily available literature. The high risk of bias due to the use of visibly different intervention methods can sometimes be avoided using sham methods, however, in other cases, unavoidable especially when the application method of the treatment is what is being investigated. Fulfilling the reporting guidelines recommended in the REFLECT statement (Reporting guidelines for randomized control trials in livestock and food safety) [74] is not common practice in animal experiments, however, would be incredibly beneficial to improve study risk of bias. Registering the trials in an animal registry potentially could have corrected the issue of publication bias apparent in the Salmonella synthesis [75]. Although this is also not common practice, the use of trial registers can also improve transparency in reporting methods, animal welfare and study reproducibility [76].
Consideration of experimental units of analysis, fulfilment of recommended reporting guidelines and addressing publication bias is necessary to improve the certainty of evidence for the evaluation of vaccinations against Campylobacter and Salmonella in meat chickens.

Conclusions
Meta-analysing the effects of vaccines against caecal Campylobacter and Salmonella revealed a large reduction in pathogen levels, however, with different levels of confidence in the findings. There was a moderate level of confidence in the reduction of Campylobacter levels by 0.93 SMD units (95% CI: −1.275 to −0.585; p value < 0.001) and a very low level of confidence in the reduction of Salmonella by 1.10 SMD units (95% CI: −1.419 to −0.776; p value < 0.001). Higher levels of heterogeneity and publication bias were the main limiting factors for the confidence in Salmonella findings. High levels of heterogeneity in both syntheses contributed to wide prediction intervals encompassing a majority of overall beneficial effects, however, were not able to predict similar individual studies will always reduce Campylobacter or Salmonella. Further research into vaccinations for Campylobacter and Salmonella is essential to source areas of heterogeneity. It is likely further research will also support the beneficial and consistent effects of Salmonella subunit vaccines seen in this review. Consideration of experimental units of analysis, a commitment to reporting guidelines and registration of trials is necessary to improve the levels of confidence in future research.