Part A: Salmonella prevalence estimates. (Question N EFSA-Q A) Adopted by The Task Force on 28 April 2008

Transcription

1 Report of the Task Force on Zoonoses Data Collection on the Analysis of the baseline survey on the prevalence of Salmonella in turkey flocks, in the EU, Part A: Salmonella prevalence estimates (Question N EFSA-Q A) Adopted by The Task Force on 28 April For citation purposes: Report of the Task Force on Zoonoses Data Collection on the Analysis of the baseline survey on the prevalence of Salmonella in turkey flocks, Part A, The EFSA Journal (2008) 134, European Food Safety Authority, 2008

2 Summary Salmonella is an important cause of food-borne illnesses in humans. Farm animals and food of animal origin form an important source of human Salmonella infections. Therefore, in order to reduce the incidence of human salmonellosis in the European Union, the Community legislation foresees setting of Salmonella reduction targets for food-animal populations including turkey flocks. To underpin such a target, a European Union-wide baseline survey was carried out to determine the prevalence of Salmonella in commercial turkey holdings with at least 250 birds for breeding turkeys and with at least 500 birds for fattening turkeys. The survey was the third of several baseline surveys to be conducted in the Community. The sampling of turkey flocks took place between October 2006 and September Five environmental faeces samples were taken from breeding turkey flocks within nine weeks of slaughter and from fattening turkey flocks within three weeks of slaughter. A total of 539 breeding turkey flocks and 3,769 fattening turkey flocks with validated results, from the EU and Norway, were included in the survey analyses. In each Member State, the number of reported holdings was combined with the number of birds annually reared in each holding (as evaluated from this survey) to estimate turkey population size. The geographical distribution of breeding turkeys in the European Union was highly heterogeneous. In fact, France accounted for 56.0% of the breeding population, followed by Italy (11.9%) and The United Kingdom (10.1%). None of the remaining Member States reached 5% of the total breeding population. The distribution of fattening turkeys was less heterogeneous. Still, five Member States accounted for 79.3% of the fattening bird population, namely, France (18.7%), Germany (16.4%), Italy (16.0%), Spain (14.7%), and Poland (13.5%). Six of the 14 Member States isolated Salmonella spp. in their breeding flocks, which resulted in a Community observed prevalence of Salmonella-positive breeding flocks of 13.6%. This means that in the European Union around one in seven breeding turkey flocks raised over the one year period of the baseline survey was Salmonella-positive. The Salmonella prevalence in these flocks varied widely amongst the Member States, from 0% to 82.9%. Three of those six Member States isolated Salmonella Enteritidis and/or Salmonella Typhimurium, the two most common serovars found in Salmonella infection cases in humans. This resulted in an estimated Community observed prevalence of 1.7% for these two serovars, varying from 0% to 8.3% within the Member States. The Community observed prevalence of Salmonella-positive fattening flocks was 30.7%, meaning that approximately one in three fattening turkey flocks raised over the one year period of the baseline survey were Salmonella-positive. The Salmonella prevalence in these flocks also varied widely amongst the Member States, from 0% to 78.5%. Thirteen of the 22 Member States with fattening turkey flocks reported to have isolated S. Enteritidis and/or S. Typhimurium resulting in a Community observed prevalence of 3.8% in the fattening turkey flocks. The Member Statespecific observed flock prevalence of S. Enteritidis and/or S. Typhimurium varied from 0% to 18.4% in fattening turkeys. European Food Safety Authority,

3 In breeding flocks no single Salmonella serovar was isolated in more than three of the 14 reporting Member States. The five most frequently isolated Salmonella serovars from fattening turkey flocks in the European Union, in decreasing order, were: S. Bredeney, S. Hadar, S. Derby, S. Saintpaul and S. Kottbus. Out of these, only S. Hadar and S. Derby are frequent causes of Salmonella infections in humans within the European Union. The serovar distribution varied amongst the Member States, with serovars tending towards specific distribution patterns of their own. The number of positive samples in a Salmonella positive breeding or fattening flock ranged between one and five. Almost all Member States had a major part of their Salmonella-infected flocks of fattening turkeys with all five samples positive. Reducing the number of samples taken per flock would have lead to a substantially lower prevalence estimate of S. Enteritidis and/or S. Typhimurium in fattening turkey flocks. Salmonella positive turkey flocks contribute to a consequent contamination of turkey meat. The risk for human health arises from accidental under-cooking of the meat or cross-contamination to other foods. Thorough cooking and strict kitchen hygiene will prevent or reduce the risk posed by Salmonella contaminated turkey meat. While Community reduction target will most likely be set for a transitional period only for S. Enteritidis and S. Typhimurium, Member States may wish to consider addressing in their national Salmonella control programmes also other serovars when these serovars are of public health importance in their country. European Food Safety Authority,

4 Table of contents Summary Introduction Objectives Materials and methods Data description Data validation and cleaning Statistical analysis Estimate of prevalence of infection Sensitivity analysis: expected results if fewer samples had been taken per flock Correlation between the Salmonella flock prevalence in flocks with breeding turkeys and in flocks with fattening turkeys Results Features of the Community turkey population Flocks with breeding turkeys Flocks with fattening turkeys Observed flock prevalence of Salmonella Flocks with breeding turkeys Flocks with fattening turkeys Number of Salmonella spp. positive samples per flock Flocks with breeding turkeys Flocks with fattening turkeys Sensitivity analysis of the effect of number of samples per flock on S. Enteritidis and/or S. Typhimurium EU prevalence Flocks with breeding turkeys Flocks with fattening turkeys Frequency distribution of Salmonella serovars Flocks with breeding turkeys Flocks with fattening turkeys Correlation between the Salmonella flock prevalence in flocks with breeding turkeys and in flocks with fattening turkeys Overview of the quality of the bacteriological testing Discussion Survey design, and data analysis Observed Salmonella turkey flock prevalence Flocks with breeding turkeys Flocks with fattening turkeys Frequency of isolated Salmonella serovars Relevance of the findings to human health...33 European Food Safety Authority,

5 5.5. The Salmonella reduction targets Conclusions Recommendations...36 Task Force on Zoonoses Data Collection members...37 Acknowledgements...37 Abbreviations...38 List of Tables...39 List of Figures...40 Annexes...42 European Food Safety Authority,

6 1. Introduction The following report describes the results of a baseline survey carried out in the European Union (EU) to estimate the prevalence of Salmonella spp. in commercial flocks of breeding and fattening turkeys. This study was the third in a series of baseline surveys of Salmonella carried out within the EU. The objective of the surveys has been to obtain comparable data for all Member States (MSs) through harmonised sampling schemes. According to Regulation (EC) No 2160/2003 on the control of Salmonella spp. and other zoonotic agents 1, which aims to reduce the incidence of food-borne diseases in the EU, results of such a survey will inform the setting of the Community target for reduction of the prevalence of the infection in turkey flocks. The survey was carried out over a one year period, starting 1 October Tested flocks were selected in holdings with at least 250 birds for breeding turkeys and with at least 500 birds for fattening turkeys. Such holdings were considered as representative of approximately 80% of the EU commercial turkey production that constituted the study s target population. Twenty-two EU MSs participated to the study. Norway participated on a voluntary basis. The objectives, the sampling frame, the diagnostic testing methods, as well as the collection and reporting of data, and the timelines of this baseline survey were specified in the Commission Decisions 2006/662/EC and 2007/208/EC 2, 3. 1 Regulation (EC) No 2160/2003 of the European Parliament and of the Council of 17 November 2003 on the control of Salmonella and other specified food-borne zoonotic agents. OJ L 325, , p Commission Decision of 29 September 2006 concerning a financial contribution from the Community towards a baseline survey on the prevalence of Salmonella in turkeys to be carried out in the Member States. OJ L 272, , p Commission Decision of 30 March 2007 concerning a financial contribution from the Community towards a baseline survey on the prevalence of Salmonella in turkeys to be carried out in Bulgaria and Romania. OJ L 92, , p. 18. European Food Safety Authority,

7 2. Objectives The aim of the survey was to estimate the prevalence of Salmonella-positive flocks amongst commercial holdings (i.e. holdings containing at least 250 birds) of breeding turkeys and amongst commercial holdings (i.e. holdings containing at least 500 birds) of fattening turkeys, at the European Community level as well as for each MS. The specific respective objectives for flocks with breeding turkeys, and for flocks with fattening turkeys, were; to estimate the flock prevalence of Salmonella in commercial holdings at the EU level and for each MS individually, to estimate the flock prevalence of the two serovars, S. Enteritidis and S. Typhimurium which pursuant to article 4 of the Regulation EC No 2160/2003, will be the serovars subject to specific reduction targets in the first instance, to investigate the serovar distribution and determine the most frequently occurring serovars in turkey flocks across the EU, to investigate the effect of potential risk factors, such as number of birds per holding, and time of sampling, which may be associated with the occurrence of Salmonella, to evaluate the sampling design especially with regard to the precision and accuracy of the prevalence estimates. MSs were also invited to submit additional information on S. Enteritidis and S. Typhimurium phage types and antimicrobial susceptibility of Salmonella isolates, but this testing was not a compulsory requirement of the survey. This part A report includes the analyses of the prevalence of Salmonella, the most frequent serovars and the sampling design. The analyses of potential risk factors as well as more in depth analyses of serovar and phage type distribution will be provided in the part B report. The analyses of the antimicrobial susceptibility of Salmonella isolates will be specifically addressed with in a separate report to be published by the European Food Safety Authority (EFSA). European Food Safety Authority,

8 3. Materials and methods A detailed description of the design of the baseline survey, the sample design and size and the bacteriological testing is found in the document of European Commission, Directorate General for Health and Consumer Affairs (DG SANCO): Baseline survey on the prevalence of Salmonella in flocks of turkeys in the EU: Technical specifications. SANCO/2083/ Environmental faecal samples were taken from flocks with breeding turkeys and from flocks with fattening turkeys. Breeding flocks on holdings with at least 250 birds were sampled within nine weeks before leaving the selected holding for slaughter. In most MSs, only one flock per holding was sampled even though two flocks with birds of eligible age could have been reared during the survey period. In the fattening turkey sub-survey, the sampling frame covered primarily holdings representing at least 80% of the total population of turkey meat finishing flocks, which was to be achieved by including holdings with at least 500 birds. The fattening turkey flocks were sampled within three weeks before leaving the selected holding for slaughter. On each selected fattening holding, one flock with turkeys of the appropriate age was to be sampled. However, in MSs where the calculated number of flocks to be sampled was greater than the number of available holdings with at least 500 birds, up to four flocks could have been sampled on the same holding in order to achieve the calculated number of flocks. Where possible, the additional flocks from a single holding were to originate from different turkey houses and samples taken in different seasons. If the number of flocks to be sampled was still not sufficient, progressively smaller holdings were to be selected, focussing preferably on holdings with more than 250 birds. Five pooled environmental faeces samples were taken in every selected flock. Each pooled sample comprised faecal material fixed to a pair of boot swabs (or sock samples which were considered equivalent). This sampling procedure should theoretically have provided 95% confidence of detection of 1% within flock prevalence assuming the analytical method was 100% sensitive. For all production types the same sampling approach was applied. For free-range flocks, samples were to be collected in the area inside the house. The number of flocks to be sampled was stratified according to the flock size and region in the MS, meaning that a representative number of flocks in different size categories of flocks as well as in different regions had to be sampled. Samples were taken by the competent authority in each MS or under its supervision and were tested by the National Reference Laboratory (or an authorised laboratory) using the ISO 6579 Annex D method. 1 European Commission Directorate General for Health and Consumer Affairs (DG SANCO). Baseline survey on the prevalence of Salmonella in flocks of turkeys in the European Union: Technical specifications. SANCO/2083/2006. Working document, 18 July Presented at the meeting of the Standing Committee on the Food Chain and Animal Health on 19 July ( European Food Safety Authority,

9 3.1. Data description Data validation and cleaning The European Food Safety Authority (EFSA) received the final dataset of the survey from the European Commission (COM) on 29 January This dataset contained data from 4,406 turkey flocks in 22 MSs and in Norway. Estonia, Latvia and Luxembourg reported not having commercial turkey flocks, but only backyard flocks. No data were received from Malta and Romania. A set of data exclusion criteria (Annex III) was used to identify non-valid and non-plausible information in the dataset. This resulted in a cleaned, validated dataset comprising 4,308 turkey flocks from 22 MSs and Norway (final dataset), which formed the basis for all subsequent analyses. An overview of the number of excluded flocks per MS is given in Table 1. Altogether, 2.2% of the flocks (95 out of 4,329) were excluded from the final EU dataset. The reasons for exclusion of samples or flocks in accordance with the exclusion criteria are summarized in Annex III. The criterion that caused the highest number of flocks to be excluded was a within-flock sample size that differed from five. The second most common cause of excluding samples was delay in start of bacteriological test of more than seven days after the day of sampling Statistical analysis Estimate of prevalence of infection Data from breeding and fattening turkey flocks were separately analysed and the following three outcomes were considered: Positivity for Salmonella spp., Positivity for S. Enteritidis and/or S. Typhimurium Positivity for serovars other than S. Enteritidis or S. Typhimurium. A flock was considered positive if Salmonella was detected in at least one of the five samples collected. Only the observed prevalence was investigated and no correction was made for test sensitivity or specificity. Only flock-level prevalence over the one year period (the proportion of tested flocks that were positive at any stage over the one year period of the survey) was investigated since it was considered as more relevant than the holding-level prevalence, from an epidemiological, as well as from a risk management perspective. European Food Safety Authority,

10 Table 1. Overview of the data validation at flock-level, Salmonella in turkey flocks baseline survey in the EU 1, Member States Number of turkey flocks structurally validated by COM Number of turkey flocks validated by EFSA Number of turkey flocks excluded by EFSA Percent (%) of flocks excluded by EFSA and sent to EFSA, full dataset final dataset Austria Belgium Bulgaria Cyprus Czech Republic Denmark Finland France Germany Greece Hungary Ireland Italy Lithuania Poland Portugal Slovakia Slovenia Spain Sweden The Netherlands The United Kingdom EU Total 4,329 4, Norway Total 4,406 4, Five MSs did not contribute to this data; Estonia, Latvia and Luxembourg that did not have commercial turkey flocks and Malta and Romania that did not participate to the survey. Prevalence was estimated for each MS and at the EU-level, by Generalized Estimating Equations (GEE) taking into account that outcomes (presence or absence of infection) in flocks from the same holdings are expected to be more alike than in flocks from different holdings (PROC GENMOD, SAS, 1999). Weights, more specifically standardised year-aggregated weights (WY), were included in the GEE models to account for a disproportionate stratified sampling design. In fact, MSs and holdings were considered as strata, and the proportion of sampled holdings was not constant across MSs. Similarly, the proportion of sampled flocks was not constant across holdings. The reciprocal of the sampling proportion for holdings (the number of holdings in a MS divided by the number of sampled holdings in the same MS) was used as the MS-level weight (WY1), whereas the reciprocal of the sampling proportion for flocks (the estimated number of flocks produced in a holding during a year, divided by the number of sampled flocks in the same holding) was used as the holding-level weight (WY2). Only WY2 was used to estimate individual MS prevalence, whereas the product between WY1 and WY2 was used to estimate the EU-level prevalence. To avoid artificial inflation of the sample size, weights were standardised so that their sum was equal to the original sample size, and this was done both at the MS (WY2) and at the EU level (WY1 by WY2). More details on statistical models and weighting are given in Annex I, section III-2. European Food Safety Authority,

11 Sensitivity analysis: expected results if fewer samples had been taken per flock The number of samples per flock was expected to affect the probability of finding at least one positive result. Therefore, an in-silico simulation exercise was set up, based on the received data, separately for flocks with breeding turkeys and for flocks with fattening turkeys, to examine the effect of reducing the number of samples per flock on that probability of detection infection in a flock, focusing on the flock prevalence estimation of S. Enteritidis or S. Typhimurium in the EU and for each MS and Norway separately. A bootstrap statistical technique was used to assess what would have been the estimated S. Enteritidis and/or S. Typhimurium EU flock observed prevalence in the baseline survey if only one, two, three or four samples per flock had been collected. More specifically, for each case investigated, 1000 replicates of the baseline survey were simulated by randomly selecting one to four samples. For each of these 1000 simulated studies, the fitting exercise previously described was performed, and the mean observed prevalence derived, with 95% simulated bootstrap confidence interval (accounting for data randomness). This allowed plotting the curve of EU flock prevalence estimates versus number of samples per flock. A SAS macro was developed for each sample size investigated, using SAS version Correlation between the Salmonella flock prevalence in flocks with breeding turkeys and in flocks with fattening turkeys Correlation between the estimated prevalence of Salmonella in breeding and fattening turkeys in each MS and Norway was studied graphically via scatterplots and in a more formal way using the Spearman rank correlation coefficient. European Food Safety Authority,

12 4. Results 4.1. Features of the Community turkey population A summary of the European Community turkey population is presented in Annex II. For breeding turkeys, the number of holdings in each MS was approximately equal to the number of sampled holdings. For fattening turkeys, a sample of holdings was selected for the survey Flocks with breeding turkeys Of the 22 MSs that participated to the survey, 14 reported data on flocks with breeding turkeys, plus Norway. Breeding turkeys were concentrated in a limited number of MSs, and the greatest population was in France, which accounted for 52% of holdings and 56% of birds in the EU. Conversely, several countries had very small breeding turkey populations (Annex II) Flocks with fattening turkeys France, Germany, Italy, Spain, and Poland accounted for 79.3% of the total EU population of fattening turkeys, and the annual production in each of these MSs exceeded 10 million birds (Annex II) Observed flock prevalence of Salmonella A total of 2.2% of the sampled flocks were excluded from the final EU dataset in the data validation and cleaning process. No flock was excluded from Austria, Cyprus, Germany, Spain, and The United Kingdom, whereas 25% of flocks from Bulgaria were excluded. The numbers of Salmonella-positive, excluded flocks by country are reported in Annex IV. Salmonella-positive flocks were excluded in nine MSs. Poland was the only MS where S. Enteritidis and/or S. Typhimurium-positive flocks were excluded. The criterion that caused the highest number of positive flocks to be excluded (16 out of the total of 17) was a within-flock sample size that differed from five. The second most common cause of excluding positive flocks was delay in start of bacteriological test of more than seven days after the day of sampling Flocks with breeding turkeys The weighted Salmonella prevalences in flocks with breeding turkeys in each MS and at EU level as well as for Norway are presented in Table 2. Although in some MS all breeding flocks that where available at the time of the visit were sampled (census sampling), it was decided to report 95% confidence intervals (CI) of prevalence for MSs were at least one flock was found European Food Safety Authority,

13 Salmonella infected. In this way, inference was attempted to the turkey production beyond the time of the data collection. Conversely, it was decided not to report CIs for countries were no flock was found infected. The raw, unweighted prevalence estimates are reported in Annex I, section IV-1. Salmonella spp. was found in six out of 14 MSs providing data on flocks of breeding turkey (Figure 1). No positive flock was found in Norway. The weighted EU-level prevalence (13.6%) was higher than the raw, unweighted prevalence (7.2%). This difference is due to the large weights that were assigned to flocks from Slovakia and Italy (because of relatively large numbers of holdings and of flocks per holding) combined with relatively high prevalence of infection in these MSs. S. Enteritidis and/or S. Typhimurium were found in breeding turkey flocks from only three MSs. The weighted EU-level prevalence was 1.7%, whereas the raw, unweighted prevalence was 0.9%. At the MS-level, prevalence was highest in Italy and it was low in France and the United Kingdom (Figure 2). Salmonella serovars other than S. Enteritidis or S. Typhimurium were found in the breeding flocks from six MSs (Figure 3). European Food Safety Authority,

14 Table 2. Weighted prevalence of Salmonella 1 in breeding turkey flocks in the EU and Norway, Salmonella spp. S. Enteritidis and/or S. Typhimurium Member State N % prevalence CI % prevalence CI % prevalence CI Bulgaria Czech Republic Finland France Germany Greece Hungary Ireland Italy Poland Slovakia Spain Sweden The United Kingdom EU Norway Serovars other than S. Enteritidis and/or S. Typhimurium N = number of tested flocks; % prevalence = weighted prevalence estimate; CI = 95% confidence interval 1 For some MSs, the sum of the weighted prevalence for Salmonella Enteritidis and / or S. Typhimurium plus the weighted prevalence for Salmonella belonging to other serovars is not equal to the weighted prevalence for Salmonella spp. Such apparent discrepancies can be attributed to the combined effects of co-infection of multiple serovar groups in the same flock, the attribution of weights based upon holding size, and rounding error in the model based estimates. European Food Safety Authority,

15 Figure 1. Weighted prevalence* of Salmonella spp. in breeding turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals**. Slovakia Italy EU Spain The United Kingdom Hungary France Germany Finland Bulgaria Norway Greece Poland Czech Republic Ireland Sweden % prevalence of Salmonella spp. * Breeding turkeys flock prevalence estimate (proportion of the total number of breeding turkey flocks over the one year period that were positive). ** Confidence intervals are not represented for MSs where no flock was found positive (prevalence = 0); these negative MSs are ordered, from top to bottom, by decreasing number of tested flocks. European Food Safety Authority,

16 Figure 2. Weighted prevalence* of Salmonella Enteritidis and/or S. Typhimurium in breeding turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals**. Italy EU France The United Kingdom Germany Slovakia Finland Hungary Spain Bulgaria Norway Greece Poland Czech Republic Ireland Sweden % prevalence of Salmonella Enteritidis - S. Typhymurium * Breeding turkeys flock prevalence estimate (proportion of the total number of breeding turkey flocks over the one year period that were positive). ** Confidence intervals are not represented for MSs where no flock was found positive (prevalence = 0); these negative MSs are ordered, from top to bottom, by decreasing number of tested flocks. European Food Safety Authority,

17 Figure 3. Weighted prevalence* of serovars other than Salmonella Enteritidis or S. Typhimurium in breeding turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals**. Slovakia Italy EU Spain Hungary The United Kingdom France Germany Finland Bulgaria Norway Greece Poland Czech Republic Ireland Sweden % prevalence of Salmonella serovars other than S. Enteritidis or S. Typhymurium * Breeding turkeys flock prevalence estimate (proportion of the total number of breeding turkey flocks over the one year period that were positive). ** Confidence intervals are not represented for MSs where no flock was found positive (prevalence = 0); these negative MSs are ordered, from top to bottom, by decreasing number of tested flocks. European Food Safety Authority,

18 Flocks with fattening turkeys The weighted Salmonella prevalences in flocks with fattening turkeys in each MS and at EU level as well as for Norway are presented in Table 3. The raw, unweighted prevalence estimates are reported in Annex I, section IV-1. Salmonella spp. was found in 19 out of 22 MSs providing data on fattening turkey flocks (Figure 4). No positive flock was found in Norway. The EU-level weighted prevalence of Salmonella spp. (30.7%) was close to the unweighted prevalence (30.4%). In individual MSs, prevalence was highest in Hungary followed by Cyprus and Spain. In general, Salmonella infections were more widespread in fattening turkey flocks within MS, than in breeding turkey flocks. S. Enteritidis and/or S. Typhimurium were found in fattening turkey flocks from 13 MSs. The EUlevel, weighted prevalence estimate (3.8%) was almost equal to the unweighted prevalence (3.9%). MS-level prevalence peaked in the Czech Republic followed by Belgium and Italy (Figure 5). Salmonella serovars other than S. Enteritidis and/or S. Typhimurium was found in 19 MSs and prevalence values were similar to those observed for Salmonella spp. (Figure 6). European Food Safety Authority,

19 Table 3. Weighted prevalence of Salmonella 1 in fattening turkey flocks in the EU and Norway, Salmonella spp. S. Enteritidis and/or S. Typhimurium Member State N % prevalence CI % prevalence CI % prevalence CI Austria Belgium Bulgaria Cyprus Czech Republic Denmark Finland France Germany Greece Hungary Ireland Italy Lithuania Poland Portugal Slovakia Slovenia Spain Sweden The Netherlands The United Kingdom EU 3, Norway Serovars other than S. Enteritidis and/or S. Typhimurium N = number of tested flocks; % prevalence = weighted prevalence estimate; CI = 95% confidence interval 1 For some MSs, the sum of the weighted prevalence for Salmonella Enteritidis and / or S. Typhimurium plus the weighted prevalence for Salmonella belonging to other serovars is not equal to the weighted prevalence for Salmonella spp. Such apparent discrepancies can be attributed to the combined effects of co-infection of multiple serovar groups in the same flock, the attribution of weights based upon holding size, and rounding error in the model based estimates. European Food Safety Authority,

20 Figure 4. Weighted prevalence* of Salmonella spp. in fattening turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals**. Hungary Cyprus Spain Czech Republic Italy The United Kingdom EU Ireland Poland Austria Slovakia Slovenia Belgium Greece The Netherlands France Germany Portugal Lithuania Denmark Finland Norway Bulgaria Sweden % prevalence of Salmonella spp. *: Fattening turkeys flock prevalence estimate (proportion of the total number of fattening turkey flocks over the one year period that are positive). ** Confidence intervals are not represented for MSs where no flock was found positive (prevalence = 0); these negative MSs are ordered, from top to bottom, by decreasing number of tested flocks. European Food Safety Authority,

21 Figure 5. Weighted prevalence* of Salmonella Enteritidis and/or S. Typhimurium in fattening turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals**. Czech Republic Belgium Italy Slovenia The United Kingdom Poland France EU Hungary Spain Germany The Netherlands Lithuania Austria Ireland Finland Portugal Norway Denmark Greece Slovakia Bulgaria Cyprus Sweden % prevalence of Salmonella Enteritidis - S. Typhymurium * Fattening turkeys flock prevalence estimate (proportion of the total number of fattening turkey flocks over the one year period that are positive). ** Confidence intervals are not represented for MSs where no flock was found positive (prevalence = 0); these negative MSs are ordered, from top to bottom, by decreasing number of tested flocks. European Food Safety Authority,

22 Figure 6. Weighted prevalence* of serovars other than Salmonella Enteritidis or S. Typhimurium in fattening turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals**. Hungary Cyprus Spain Italy The United Kingdom Ireland EU Czech Republic Austria Poland Slovakia Greece Slovenia The Netherlands Belgium France Germany Portugal Lithuania Denmark Finland Norway Bulgaria Sweden % prevalence of Salmonella serovars other than S. Enteritidis or S. Typhymurium * Fattening turkeys flock prevalence estimate (proportion of the total number of Fattening turkey flocks over the one year period that are positive). ** Confidence intervals are not represented for MSs where no flock was found positive (prevalence = 0); these negative MSs are ordered, from top to bottom, by decreasing number of tested flocks. European Food Safety Authority,

23 4.3. Number of Salmonella spp. positive samples per flock A total of five samples were taken from each flock sampled and, in positive flocks, one to five samples were positive Flocks with breeding turkeys The overall proportions of Salmonella spp. positive breeding flocks found positive on the basis of one, two, three, four and five positive samples was not evenly distributed but was 17.5%; 12.5%; 20.0%; 15.0% and 35.0%, respectively. The MS-specific distribution of the within-flock number of Salmonella-positive samples in the positive breeding turkey flocks varied between countries and is shown in Figure 7. Figure 7. Distribution of the within-flock number of positive samples in Salmonella spp. positive flocks with breeding turkeys observed in the EU MSs, Number of Salmonella spp. positive flocks Spain Sweden The United Kingdom Italy Norway Poland Slovakia Germany Greece Hungary Ireland Bulgaria Czech Republic Finland France Number of positive samples per flock Flocks with fattening turkeys The overall proportions of Salmonella positive fattening flocks found positive on the basis of one, two, three, four and five positive samples was not evenly distributed, but was 16.9%; 13.7%; 12.2%; 13.2% and 44.0%, respectively. The MS-specific distribution of the within-flock number of Salmonella-positive samples in the positive fattening turkey flocks varied among countries and is shown in Figure 8. However, almost all MSs had a major proportion of their Salmonella infected flocks with all five samples positive. European Food Safety Authority,

24 Figure 8. Distribution of the within-flock number of positive samples in Salmonella spp. positive flocks with fattening turkeys observed in the EU MSs, Sweden The Netherlands The United Kingdom Number of Salmonella spp. positive flocks Poland Portugal Slovakia Slovenia Spain Hungary Ireland Italy Lithuania Norway Denmark Finland France Germany Greece Austria Belgium Bulgaria Cyprus Czech Republic Number of positive samples per flock 4.4. Sensitivity analysis of the effect of number of samples per flock on S. Enteritidis and/or S. Typhimurium EU prevalence Flocks with breeding turkeys S. Enteritidis and/or S. Typhimurium were found in breeding flocks in only three MSs. Moreover less than 1% of the sampled flocks were positive. Consequently, simulated prevalence estimates of S. Enteritidis and/or S. Typhimurium, which were obtained through the EU-level sensitivity analysis, were characterised by extreme uncertainty, as demonstrated by very wide confidence intervals. Therefore, the results of sensitivity analysis for breeding flocks were not considered as useful and are not presented in this report Flocks with fattening turkeys Figure 9 represents the simulated effect of the number of samples per flock on the EU-level prevalence of S. Enteritidis and/or S. Typhimurium in fattening turkey flocks. The sensitivity of sampling and the EU-level prevalence decreased with decreasing sample size, i.e. collecting only two instead of five samples would have lead to a prevalence estimate of 2.8% (95% confidence interval: 2.5%, 3.2%), instead of 3.8%. European Food Safety Authority,

25 Figure 9. Simulated EU prevalence of S. Enteritidis and/or S. Typhimurium -positive flocks with fattening turkeys and 95% confidence intervals for sample sizes less than five per flock. % prevalence of Salmonella Enteritidis Typhimurium Number of samples per flock Similarly, results for each MS are reported in Annex I, section IV-3. At the MS level, the general tendency follows a similar pattern as for the EU level. However, variability across MSs is large Frequency distribution of Salmonella serovars The serotyping of Salmonella isolates was mandatory according to the technical specifications of the survey. At least one isolate from each positive sample was to be typed according to the Kaufmann-White Scheme. Results from any flock where the serovar information was not available for any isolate were excluded from the final dataset Flocks with breeding turkeys In total there were 135 Salmonella-positive samples (5.0% of 2,695 samples) originating from 40 positive breeding turkey flocks. More than one Salmonella serovar was not isolated from any Salmonella-positive sample. The frequency distribution of isolated Salmonella serovars in the EU is listed in Table 4. This table is ranked based on the percentages of specific Salmonella serovar-positive flocks, as flock was the epidemiological unit of interest. MS-specific overviews of the frequency distribution of serovars are shown in Annex V. S. Saintpaul was the most frequently reported serovar from the breeding turkey flocks in EU, found in 42.5% of the Salmonella positive flocks. The two next most frequent serovars were S. European Food Safety Authority,

26 Kottbus and S. Typhimurium (17.5% and 10.0% of the Salmonella positive flock, respectively). S. Kottbus was the serovar most commonly isolated in terms of number of MSs (three MSs). The distribution of the reported serovars varied amongst the MSs. Isolation of S. Saintpaul in breeding turkey flocks was only reported by Slovakia. The fact that it was isolated in 17 flocks in this MS resulted in S. Saintpaul being the most frequently reported serovar, at EU breeding flocklevel. For MSs reporting more than one breeding turkey flock positive for specific serovars, S. Kottbus was the most frequently reported serovar for Hungary (two flocks), and The United Kingdom (four flocks positive). S. Typhimurium was the most frequently reported serovar for Italy (three flocks positive) Flocks with fattening turkeys There were a total of 3,814 Salmonella-positive samples (20.2% of 18,845 samples) originating from 1,084 positive fattening flocks. Two different Salmonella serovars were isolated from 20 Salmonella-positive samples. The frequency distribution of isolated Salmonella serovars in the EU is listed in Table 5. This table is ranked based on the percentages of specific Salmonella serovar-positive flocks, as flock was the epidemiological unit of interest. MS-specific overviews of the frequency distribution of serovars are shown in Annex VI. S. Bredeney was the most frequently reported serovar from the fattening turkey flocks in EU, representing 17.2% of the Salmonella positive flocks. The three next most frequent serovars were S. Hadar, S. Derby and S. Saintpaul (14.0%, 11.3% and 10.4% of the positive flocks, respectively). S. Saintpaul and S. Typhimurium were the serovars most commonly isolated in terms of the number of MSs, in total 12. The distribution of the serovars varied widely amongst the MSs. S. Bredeney was the dominant serovar in only two MSs: Hungary and Italy. In fact, Hungary accounted for 75.8% (141 out of 186), while Italy accounted for 20.4% (38 out of 186) of the S. Bredeney-positive flocks. S. Hadar was the leading serovar in three MSs: Austria, Lithuania and Spain. Spain accounted for 71.1% (108 out of 152) of the S. Hadar -positive flocks. S. Derby was the leading serovar in two MSs: France and Portugal, S. Saintpaul in three MSs: Poland, Slovakia and The Netherlands, and S. Typhimurium in Germany. From comparison of the MS-specific overviews of the frequency distribution of serovars in breeding and in fattening turkeys (Annexes V and VI) it can be generally noticed that the diversity of observed serovars in fattening turkeys was bigger compared to the one in breeding flocks. Also all isolated serovars in breeding flocks are almost always also isolated in the fattening flocks in the same MS, with the exception of one serovar, S. Corvalis in France. European Food Safety Authority,

27 Table 4. Frequency distribution of isolated Salmonella serovars in the breeding turkey flocks baseline survey in the EU and Norway, Samples with serovars (N=135) Holdings with serovars (N=26) Flocks with serovars (N=40) Countries with serovars N % N % N % N S. Saintpaul S. Kottbus S. Typhimurium S. Heidelberg S. Derby S. Blockley S. Senftenberg S. Corvallis S. Bredeney S. Bradford S. Enteritidis S. Thompson Total isolates 135 European Food Safety Authority,

28 Table 5. Frequency distribution of isolated Salmonella serovars in the fattening turkey flocks baseline survey in the EU and Norway, Samples with serovars (N=3,834) Holdings with serovars (N=934) Flocks with serovars (N=1,084) Countries with serovars N % N % N % N S. Bredeney S. Hadar S. Saintpaul S. Derby S. Kottbus S. Typhimurium S. Orion S. Infantis S. Enteritidis S. Agona S. Newport S. Blockley S. Indiana S. London S. Heidelberg S. Kedougou S. Senftenberg S. Montevideo S. Zanzibar S. Virchow Others Salmonella untypeable Total isolates 3,852 European Food Safety Authority,

29 4.6. Correlation between the Salmonella flock prevalence in flocks with breeding turkeys and in flocks with fattening turkeys Correlation between the estimated prevalence of Salmonella in breeding and fattening turkeys in each MS was studied formally using the Spearman rank correlation coefficient, ρ, a nonparametric rank correlation procedure which can be used when few data pairs (15) are available. The estimated correlations are displayed in Table 6. This table also includes p-values from testing the null hypothesis of no association between the prevalence estimates in the two types of flocks. Significant correlation was observed for Salmonella spp. and for S. Enteritidis or Typhimurium (p < 0.05), whereas correlation for serovars other than S. Enteritidis or Typhimurium was near the threshold of statistical significance. Table 6. Spearman correlation coefficients and corresponding p-values for the test of the correlation between the prevalence estimates of breeding and fattening turkeys equal to zero. ρ p-value Salmonella spp S. Enteritidis or S. Typhimurium Serovars other than S. Enteritidis or S. Typhimurium These significant results are based on calculations taking into account the results of MSs that reported no positive outcomes for both breeding and fattening turkeys Overview of the quality of the bacteriological testing In the technical specifications of the baseline survey it was indicated that at least one isolate from each positive sample should be serotyped in the National Reference Laboratory (NRL) for Salmonella, following the Kaufmann-White scheme. For quality assurance of the serotyping, a maximum of 16 non-typable isolates of the one year survey had to be sent to the Community Reference Laboratory (CRL) for Salmonella. The CRL-Salmonella reported on the quality of the serotyping of non-typable Salmonella isolates from the baseline survey in turkey flocks performed by the NRLs. Seven NRLs-Salmonella (of the 23 participating countries) sent in some non-typable isolates to the CRL; 16 NRLs indicated that they had not found any non-typable isolates. Only a very low number of non-typable Salmonella strains were found during this baseline survey by the NRLs-Salmonella. A total of 25 non-typable strains were received by the CRL-Salmonella. Of these strains, CRL-Salmonella was able to further identify 13 strains to serovar names. Although the CRL-Salmonella also followed the Kauffmann-White scheme for serotyping the strains, extra or alternative culture steps were used, which are in most cases not routinely used at European Food Safety Authority,

30 the NRLs-Salmonella, and were not required by the technical specifications of the survey. Because of this special treatment the CRL-Salmonella was able to further identify the strains where the NRL was not able to do so. Still 11 isolates could only be identified to the level of subspecies and two isolates could only biochemically be identified as Salmonella while serotyping was not possible as the strains were rough. European Food Safety Authority,

31 5. Discussion 5.1. Survey design, and data analysis The aim of this survey was to estimate the prevalence of Salmonella in turkey flocks in the EU. A flock is defined by Regulation (EC) No 2160/2003, as "a group of birds constituting a single epidemiological unit and, in the case of housed poultry, all birds sharing the same airspace". Three issues were taken into consideration in the statistical analysis, in order to obtain valid prevalence estimates: 1) the potential correlation between outcomes (presence or absence of infection) for flocks belonging to the same holding; 2) sampling of different proportions of the total numbers of holdings present in the participating MSs; 3) sampling of different proportions of the numbers of flocks present in the sampled holdings. The statistical techniques that were implemented in the analysis (GEE) are specific for correlated observations (issue 1). Moreover, disproportionate sampling at the country (issue 2), and at the holding level (issue 3) were considered through weighting of the results. In this way, an effort was made to apply the most appropriate analysis to such a complex study design. The resulting, weighted prevalence estimates are therefore valid and representative indices of the presence of Salmonella spp. in EU turkey flocks. Therefore, such estimates are most suitable to be used in target setting for the control of Salmonella infection in the EU, and as references for further studies at the EU level and within MSs. The boot-swab/sock method of obtaining pooled faecal samples from the environment of the flock is a reliable and sensitive method of detecting Salmonella spp. in a house of litter-bedded birds, and it was used in previous baseline surveys coordinated by the European Commission. Nevertheless, a sensitivity analysis of the sampling design suggests that, in fattening turkeys, the number of samples taken per flock might affect prevalence estimate for Salmonella Enteritidis and/or Salmonella Typhimurium. Accordingly, collecting less than five samples would have yielded lower prevalence values compared with those reported here. This result could be partially associated with the fact that each sample was taken from a different part of the house where a flock was reared. Because the distribution of S. Enteritidis and/or S. Typhimurium in the house may not be uniform, excluding one or more samples would reduce the probability of detecting the infection. Turkey flocks are not uniformly distributed throughout the EU. This was particularly true for flocks of breeding turkeys which are concentrated in a small number of MSs. The very small numbers of breeding holdings in several MSs tended towards virtually complete coverage sampling (census) of breeder flocks. The dataset analysed was not the complete dataset submitted by MSs, due to exclusion of some samples with implausible data values. In total, 2.2% of the flocks were excluded from the final EU dataset. This proportion of excluded data can be considered to be very small at EU-level. Therefore the exclusion is unlikely to have a significant impact on the results at the Communitylevel. On the other hand, in certain MSs, the proportion of excluded data was relatively high and reached 25% in Bulgaria (however, none of the excluded flocks from this MS was Salmonella European Food Safety Authority,

32 positive). Two MSs, Malta and Romania, did not submit any data. Since Romanian turkey population appears to be large (although entirely constituted by relatively small holdings), the impact of this MS to the EU prevalence of Salmonella could have been substantial, but it remains unknown Observed Salmonella turkey flock prevalence Flocks with breeding turkeys The EU-level, weighted prevalence of Salmonella spp. in breeding turkey flocks was 13.6%. This means that one in seven breeding flocks in the EU was expected to be infected with Salmonella spp. within nine weeks of slaughter. However the variation amongst MSs was considerable. Salmonella spp. was not detected in eight of the 14 MSs providing data on breeding flocks or in Norway, and prevalence was less than 6% in four out of six MSs where the infection was found. Prevalence in the remaining two MSs with Salmonella-positive breeding flocks, Italy and Slovakia, was in excess of 20% and 80% respectively. S. Enteritidis or S. Typhimurium were reported from the breeding flocks in three MSs, and the prevalence exceeded 1% only in Italy. In the MS with the highest prevalence of Salmonella spp. in breeding flocks, Slovakia, none of the isolates were S. Enteritidis or S. Typhimurium. The relevance of Salmonella spp. infection in breeding turkeys is mainly related to the potential for vertical transmission to fattening flocks. The significant correlation between prevalences of Salmonella spp. in breeding and fattening flocks was consistent with the hypothesis of an epidemiological association between these two flock types within the same MS. The finding of a prevalence in breeding flocks at about half that of fattening flocks may be explained by clearing of infection by the older breeding birds, and/or the intensified approach to biosecurity for the breeding stock 1. Overall, the findings of the present survey indicate localised Salmonella spp. problems in the breeding flocks of a small number of MSs with relatively high prevalence Flocks with fattening turkeys The overall EU weighted prevalence of Salmonella spp. in fattening flocks was 30.7%. Variation existed amongst MSs, with three MSs reporting no detected Salmonella spp. Of the 19 MSs that reported Salmonella spp. in fattening flocks, six had prevalences in excess of the EU weighted prevalence, although for two of those MSs the prevalence seemed not markedly different from that EU weighted average. In analogy with the breeding turkeys, the contribution of S. Enteritidis or S. Typhimurium to this prevalence was much less than that of serovars other than S. Enteritidis or S. Typhimurium, 3.8% and 26.3% respectively. Of the 13 MS reporting S. Enteritidis or S. 7Typhimurium, only three MS reported an observed prevalence greater than 5%, one of which was 18.4%. In Hungary, where the highest prevalence of Salmonella spp. was observed, most of the isolates belonged to serovars other than S. Enteritidis or S. Typhimurium. 1 Cox, N.A., N.J. Stern, S.E. Craven, M.E. Berrang, and M.T. Musgrove Prevalence of Campylobacter and Salmonella in the cecal droppings of turkeys during production. J. Appl. Poultry Res. 9: European Food Safety Authority,

33 The reported EU prevalence of 30.7% means that almost one in three fattening flocks in the EU harbours Salmonella within three weeks prior to slaughter. Hygiene of slaughter and subsequent cutting and processing of meat may be optimised to minimise spread of Salmonella in the turkey food-chain. However, the presence of Salmonella in about one third of the flocks before the slaughtering reduces chances of producing turkey meat with low Salmonella contamination level Frequency of isolated Salmonella serovars In breeding flocks of turkeys, no clearly dominant Salmonella serovar emerged in the present survey. The distribution of serovars was particularly heterogeneous, with no indication of EUwide serovar of particular relevance. The most frequently-isolated serovar in the EU breeding flock population was S. Saintpaul accounting for 42.5% of the positive breeding flocks, but it was only present in Slovakia. S. Kottbus was the next most common serovar in breeding flocks at 17.5% of positive flocks; and the most widely distributed, being present in breeding flocks in three MSs. S. Typhimurium was the third most common serovar isolated in breeding flocks, present in two MSs, and the most frequent serovar in breeding flocks in one MS. Whilst human disease has been attributed to each of these serovars, only S. Typhimurium was amongst the ten most frequently isolated serovars in human salmonellosis cases in the EU in 2006 (Community Summary Report on Zoonoses in ). In fattening turkey flocks, a dominant Salmonella serovar was equally unapparent as the serovar distribution greatly varied between the MSs. S. Bredeney was the serovar most frequently isolated from fattening flocks accounting for 17.2% of positive units. However, S. Bredeney demonstrated a relatively narrow geographic distribution; while being found in only six MSs, Hungary and Italy accounted together for more than 95% of the positive flocks. S. Saintpaul and S. Typhimurium were the two serovars most widely distributed in fattening turkey flocks, each reported by 12 MSs. In addition to these two serovars, S. Hadar, S. Derby, and S. Kottbus demonstrated frequent contributions to positive flocks, and wide distribution amongst MSs. Three of those six serovars, S. Typhimurium, S. Hadar and S. Derby, were amongst the ten most frequently isolated serovars in human salmonellosis cases in the EU in 2006 (Community Summary Report on Zoonoses in 2006). Overall, this survey demonstrates a wide variation in the distribution of Salmonella serovars in turkeys and the absence of a dominant serovar in this poultry species. These results contrast with those found in Gallus gallus (fowl), where S. Enteritidis (uncommon in turkeys) predominates in both laying hens and broilers in many MSs. A risk factor analysis, as well as a more in depth analysis of the Salmonella serovars including the phage types will be presented in the Part B report Relevance of the findings to human health This survey investigated the prevalence of Salmonella in the rearing environment of turkeys, and therefore, at the primary production phase of the food-chain. Salmonella serovars of potential 1 The Community Summary Report on Trends and Sources of Zoonoses, Zoonotic Agents, Antimicrobial Resistance and Foodborne Outbreaks in the European Union in 2006, The EFSA Journal (2006), 130. European Food Safety Authority,

34 public health significance were detected both in breeding and fattening flocks, in some MSspecific cases at relatively high prevalence. The translation of any risk from live birds to food may be modified, either adversely or positively, by subsequent food production processes such as transport, slaughter, cutting or processing. In particular, slaughter technology might facilitate horizontal transfer to carcasses from Salmonella-free flocks, while de-skinning operations such as filleting, and other processes such as freezing, might afford useful Salmonella reduction. At a consumer level, heat treatments, such as the thorough cooking normally received by turkeys, eliminates the risk arising from Salmonella contamination. However significant potential for cross contamination of the domestic environment including other foodstuffs, might arise from the manual manipulations common to turkey carcasses e.g. stuffing / washing, as well as the issues arising from physical size of turkey carcasses exceeding the domestic capacity for raw meat storage and handling. Such augmentation or amelioration of risk would not be reflected in a prevalence survey such as this. Additionally, despite a relatively biosecure approach to primary production, Salmonella infection in turkeys represents a potential source of risk for other food producing animal species and zoonotic pathways, possibly through vectors such as wildlife. Thus these data provide a useful overall indication of the potential role of turkeys in the epidemiology of Salmonella species in the global ecosystem. As the serovar distribution varied greatly amongst the MSs, the situations are likely to be MS specific. The most common serovar associated with human disease, S. Enteritidis was only rarely isolated from turkeys. The second most common serovar implicated in human disease, S. Typhimurium was one of the more frequently isolated from turkeys in this survey, and serovars of emerging human significance such as S. Hadar and S. Derby were also amongst the more frequent and more widely distributed turkey isolates. A more in depth analysis of serovar distribution, including animal source attribution for serovars provoking human cases of salmonellosis, will be the subject of a specific EFSA report The Salmonella reduction targets A reduction target for S. Enteritidis or S. Typhimurium, as foreseen in Regulation (EC) No 2160/2003 and as justified from a public health perspective, may well be informed by the finding of this survey. With some MS-specific exceptions, the prevalences of these two serovars, particularly S. Enteritidis, is already quite low at an EU level, lower than those identified for broilers (Gallus gallus), which means that a target prevalence of 1% or less is likely to be achievable. MSs may be informed by the specific results, and investigate risk management approaches to address specific reduction targets for serovars such as S. Saintpaul, S. Kottbus, S. Bredeney, S. Hadar, S. Infantis or S. Derby, in case these serovars are of public health importance in their countries. European Food Safety Authority,

35 6. Conclusions This baseline survey has established a baseline turkey flock observed Salmonella prevalence in EU, which should inform the EU Salmonella reduction target. The baseline prevalence figures may be used later to compare future trends and follow the impact of the control programmes. The other variables studied, such as the proportion of positive samples in flocks, impact of reducing number of samples and the serovar distribution, will also contribute to understanding and managing the Salmonella infections. The survey provides valuable data for risk managers on the prevalence and distribution of Salmonella in EU MSs, and results are suitable to be used for setting targets for the reduction of the frequency of the infection in the EU. In the EU about half of the MSs reported having holdings with breeding turkeys and France accounted for about half of the breeding flocks and breeding birds. With regard to fattening turkeys, almost all MSs have flocks and France accounted for about one sixth of the fattening flocks and fattening birds. An effort was made to achieve a harmonised sampling across MS to favour comparability of results. Moreover, statistical analysis was carried out in order to account for survey design. However, very small numbers of flocks were tested in certain MSs. Therefore, even if the sample comprised all available flocks during the sampling period, prevalence, as an estimate of the risk of infection in turkey production in those MS, was affected by a high degree of uncertainty. The observed flock prevalence of Salmonella spp. varied widely amongst MSs. The weighted prevalence of Salmonella spp. in breeding flocks within the EU was estimated to be 13.6% whereas the weighted flock prevalence of S. Enteritidis and/or S. Typhimurium was 1.7%. S. Saintpaul was the most frequently isolated serovar from breeding flocks (42.5% of positive flocks) but it was only found in Slovakia. Of the two next frequent serovars, S. Kottbus was found in three MSs, and S. Typhimurium in two MSs. For fattening turkey flocks the weighted EU prevalence of Salmonella spp. was estimated to be 30.7% whereas the weighted flock prevalence of S. Enteritidis and/or S. Typhimurium was 3.8%. The six most frequently isolated Salmonella serovars from the fattening flocks were S. Bredeney (17.2% of the positive flocks), S. Hadar (14.0%), S. Derby (11.3%), S. Saintpaul (10.4%), S. Kottbus (8.3%), S. Typhimurium (7.9%) and out of these, S. Hadar, S. Derby, and S. Typhimurium are amongst the 10 most commonly reported serovars in humans. The serovar distribution varied amongst the MSs, many of them having a specific distribution pattern of their own. Often, for a specific Salmonella serovar, a few MSs accounted for the majority of the positive flocks. Reducing the number of samples taken from a flock of fattening turkeys is likely to have a strong impact on S. Enteritidis and/or S. Typhimurium flock prevalence estimation. There was correlation between serovars found in breeding flocks and fattening flocks in many MSs. Salmonella infected turkey flocks contribute to consequent contamination of fresh turkey meat. Salmonella infection in humans may result from undercooking of the meat or crosscontamination to other foods. Thorough cooking of the turkey meat and strict kitchen European Food Safety Authority,

36 hygiene would prevent or reduce the risk posed by Salmonella contaminated turkey meat at the consumer level. 7. Recommendations It is recommended that MSs would address in their Salmonella control programmes also serovars other than S. Enteritidis and S. Typhimurium, when these other serovars are of public health importance in their country. The analysis of potential risk factors for Salmonella spp. at the flock-level, including season and flock size, as well as in depth serovar and phage type analyses, should be the subjects of the report part B, and will help to generate hypotheses on maintenance and transmission of the infection and the importance of findings to public health. Although the level of data exclusions was low overall it affected certain MS disproportionately because of low numbers of flocks sampled. It is recommended that MS pay close attention to the data dictionary and exclusion criteria in future surveys to avoid losing data resulting from valuable sampling visits. The one-in-three positivity of fattening flocks would warrant enhanced hygiene measures and related official controls in order to reduce the Salmonella contamination of turkey meat. European Food Safety Authority,

37 Task Force on Zoonoses Data Collection members José Ignacio Arraz Recio, Andrea Ammon, Alenka Babusek, Marta Bedriova, Birgitte Borck, Karen Camilleri, Georgi Chobanov, Adriana Costache, Kris De Smet, Katica Florjanc, Matthias Hartung, Birgitte Helwigh, Merete Hofshagen, Sarolta Idei, Vaidotas Kiudulas, Elina Lahti, Lesley Larkin, Peter Much, Lisa O Connor, Rob A.A. Van Oosterom, Jacek Osek, José Luis Paramio Lucas, Melanie Picherot, Christodoulos Pipis, Saara Raulo, Antonia Ricci, Tatjana Ribakova, Valentina Rizzi, Petr Šatrán, Joseph Schon, Jelena Sõgel, Petra Szabados, Patricia Tavares Santos, Kilian Unger, Luc Vanholme and Dimitris Vourvidis. Acknowledgements The Task Force on Zoonoses Data Collection wishes to acknowledge the contribution of the Working Group that prepared this report: Birgitte Borck, Vojislava Bole-Hribovšek, Rob Davies, Annemarie Käsbohrer, Nicolas Rose, Micheál O Mahony, Arjen W. van de Giessen, Kris De Smet, Francesca Riolo, Kenneth Mulligan, Pablo Nart, Billy Amzal, Pierre-Alexandre Beloeil, Pia Mäkelä, Alessandro Mannelli and Frank Boelaert. The Task Force on Zoonoses Data Collection also wishes to acknowledge the contribution to statistical analysis of personnel of Hasselt University, Center for Statistics: Marc Aerts, José Cortiñas, Christel Faes, Saskia Litière and Kaatje Bollaerts. The implementation of the baseline survey by the Competent Authorities of the Member States and Norway is gratefully acknowledged. European Food Safety Authority,

38 Abbreviations CI CRL EEA EFSA EU MS(s) NRL Confidence Interval Community Reference Laboratory European Economic Area European Food Safety Authority European Union Member State(s) National Reference Laboratory European Food Safety Authority,

39 List of Tables Main report (pp. 1-41) Table 1. Table 2. Table 3. Table 4. Table 5. Table 6. Overview of the data validation at flock-level, Salmonella in turkey flocks baseline survey in the EU 1, Weighted prevalence of Salmonella in breeding turkey flocks in the EU and Norway, Weighted prevalence of Salmonella in fattening turkey flocks in the EU and Norway, Frequency distribution of isolated Salmonella serovars in the breeding turkey flocks baseline survey in the EU and Norway, Frequency distribution of isolated Salmonella serovars in the fattening turkey flocks baseline survey in the EU and Norway, Spearman correlation coefficients and corresponding p-values for the test of the correlation between the prevalence estimates of breeding and fattening turkeys equal to zero...29 European Food Safety Authority,

40 List of Figures Main report (pp. 1-41) Figure 1. Weighted prevalence* of Salmonella spp. in breeding turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals** Figure 2. Weighted prevalence* of Salmonella Enteritidis and/or S. Typhimurium in breeding turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals** Figure 3. Weighted prevalence* of serovars other than Salmonella Enteritidis or S. Typhimurium in breeding turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals** Figure 4. Weighted prevalence* of Salmonella spp. in fattening turkey flocks in the EU and Figure 5. Norway, Horizontal bars represent 95% confidence intervals** Weighted prevalence* of Salmonella Enteritidis and/or S. Typhimurium in fattening turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals** Figure 6. Weighted prevalence* of serovars other than Salmonella Enteritidis or S. Typhimurium in fattening turkey flocks in the EU and Norway, Horizontal bars represent 95% confidence intervals** Figure 7. Figure 8. Figure 9. Distribution of the within-flock number of positive samples in Salmonella spp. positive flocks with breeding turkeys observed in the EU MSs, Distribution of the within-flock number of positive samples in Salmonella spp. positive flocks with fattening turkeys observed in the EU MSs, Simulated EU prevalence of S. Enteritidis and/or S. Typhimurium -positive flocks with fattening turkeys and 95% confidence intervals for sample sizes less than five per flock European Food Safety Authority,

41 List of Annexes Annex I. Statistical analysis of the baseline survey on the prevalence of Salmonella in turkeys, in the EU, Annex II. General features of the European turkey population...66 Annex III. List of criteria used to identify non-valid and non-plausible information in the Salmonella turkey database...75 Annex IV. Overview of the number of records with non-plausible characteristics in the final dataset received by the European Commission and proportions of excluded Salmonella spp. positive flocks...78 Annex V. Frequency distribution of Salmonella serovars in breeding turkey flocks in EU and Norway, Annex VI. Frequency distribution of Salmonella serovars in fattening turkey flocks in EU and Norway, European Food Safety Authority,

42 Report of the Task Force on Zoonoses Data Collection on the Analysis of the baseline survey on the prevalence of Salmonella in turkey flocks, in the EU, Annexes (Question N EFSA-Q A) Adopted by The Task Force on 28 April For citation purposes: Report of the Task Force on Zoonoses Data Collection on the Analysis of the baseline survey on the prevalence of Salmonella in turkey flocks, Part A, The EFSA Journal (2008) 134, European Food Safety Authority,

43 Table of contents Annex I. Statistical analysis of the baseline survey on the prevalence of Salmonella in turkeys, in the EU, I- Introduction...44 II- Objective...44 III- Material and Methods...45 III-1 Data import and management III-2 Methodology and tools for the flock prevalence estimation III-3 Methodology for the sensitivity analysis IV- Results...51 IV-I Summary of raw results IV-2 Results of the prevalence estimations IV-3 Results of the sensitivity analysis V- Discussion and Conclusions...65 Annex II. General features of the European turkey population...66 Annex III. List of criteria used to identify non-valid and non-plausible information in the Salmonella turkey database Annex IV. Overview of the number of records with non-plausible characteristics in the final dataset received by the European Commission and proportions of excluded Salmonella spp. positive flocks Annex V. Frequency distribution of Salmonella serovars in breeding turkey flocks in EU and Norway, Annex VI. Frequency distribution of Salmonella serovars in fattening turkey flocks in EU and Norway, References...91 European Food Safety Authority,

44 Annex I. Statistical analysis of the baseline survey on the prevalence of Salmonella in turkeys, in the EU, I- Introduction This survey follows from the regulation (EC) No. 2160/2003 on the control of Salmonella which provides the setting of Community targets for reducing the prevalence of Salmonella serovars with public health significance in animal populations. The European Union has carried out large baseline surveys. The technical specifications of the baseline surveys are laid down in the Commission decisions: 2006/662/EC, 2006/668/EC, 2007/208/EC and 2007/219/EC. Reduction specific targets are also going to be considered in turkeys (in particular breeding and fattening animals) and in slaughter pigs. In this report (Part A) the focus is on the baseline surveys on Salmonella in turkey flocks. II- Objective The objective of this report is to obtain valid estimates of the prevalence of Salmonella serovars in turkey flocks, at the Community level and for each Member state, and separately for Norway. Positivity for Salmonella spp., positivity for either S. Enteritidis or S. Typhimurium and `positivity for serovars other than S. Enteritidis S. Typhimurium are the three outcome variables to be analyzed separately for flocks with breeding turkeys and for flocks with fattening turkeys. Statistical analyses include: prevalence estimation of the three outcomes of interest, for both breeding and fattening turkeys, in the European Union (EU) and in each Member State (MS), and Norway separately. Correlation between the estimated prevalence in breeding and fattening turkeys will also be evaluated. a sensitivity analysis of the survey design. The number of samples per flock affects the probability of finding at least one positive result in a group. The effect of varying sample size on the probability of detecting infection in a flock will be investigated. Information on the sampling scheme and testing procedures is provided in the technical specifications on the baseline survey in turkey flocks, mentioned in the introduction. European Food Safety Authority,

45 III- Material and Methods III-1 Data import and management All data management and statistical analysis in this report were performed using the SAS System, whereas graphs were obtained using the open source software R ( The data contained information at the level of the environmental faecal samples taken within turkey flocks. However, since flock-level prevalence of Salmonella needs to be estimated, some data manipulation was required. First, three additional variables were created, indicating whether each sample within a flock was found positive for Salmonella spp. S. Enteritidis or S. Typhimurium Serovars other than S. Enteritidis or S. Typhimurium. A flock is defined positive for the outcome of interest when at least one sample is positive. This results in a new data set with information on flock-level, containing 3 new outcome variables: SalmSpp, which equals 1 when the flock is found positive for Salmonella spp, SalmET, which equals 1 when the flock is found positive for S. Enteritidis or S. Typhimurium, and SalmOth, which equals 1 when the flock is found positive for Salmonella serovars other than S. Enteritidis or S. Typhimurium. III-2 Methodology and tools for the flock prevalence estimation The hierarchical structure in the data can essentially be expressed as follows: samples within a flock, flocks within a holding, and holdings within a country. Interest goes to the flock-level prevalence. Therefore, let be the probability for a flock to be positive, let be the number of flocks in holding j from country i. The starting point for inference on the flock-level prevalence of the different outcome a variable is the binomial distribution for the number of positive flocks in holding j from country i:. (1) In a fully random sample these numbers could be combined in a straightforward way to estimate the prevalence for country i. The main complications here are 1. the assumptions on the binomial distribution are violated 2. the sample is not drawn at random (but essentially stratified) Indeed, violation of independence: outcomes from the same holding are expected to be more alike (correlated) as compared to outcomes from a different holding (hierarchical correlation structure), European Food Safety Authority,

46 violation of constant probability: samples, even from the same holding might have different probabilities to be infected (heterogeneity of probability). Clustering To account for the possibility of samples from the same holding being more alike than from different holdings, there exist, broadly, three approaches: Ignore the correlation. While this typically leaves the consistency of point estimation intact, the same is not true for measures of precision. In case of a positive correlation (i.e., samples within a holding are more alike than between holdings), then ignoring this aspect of the data, just as ignoring overdispersion, overestimates precision and hence underestimates standard errors and widths of confidence intervals. Account for correlation. The existence of correlation is recognized but considered as a nuisance characteristic. A crude way of correcting for clustering is done by computing a so-called design effect. Roughly, the design effect is a factor comparing the precision under simple random sampling with the precision of the actual design. Standard errors, computed as if the design had been simple random sampling, can then be artificially inflated using the design effect. Model correlation. In contrast to the previous view-point, one can have a genuine scientific interest in the correlation itself. The intra-class correlation should be addressed in order to obtain valid statistical inferences, and specialized methods which model the correlation should be used. Obviously the third method is much broader. Hence analysis strategies consistent with an interest in the intra-cluster dependence can be applied. There exist two important families of models which can be used for this purpose: random-effects models and marginal models. Given that the objective of the analysis in this report is to obtain a prevalence estimate of Salmonella in the EU and for each MS and Norway separately, the marginal or populationaveraged approach is the obvious path to follow. Indeed, the marginal model can be used to evaluate the overall prevalence (i.e., averaged over all holdings in the EU and in the MS and Norway separately). We will fit a logistic intercept model, which will provide us with an estimate for the prevalence of Salmonella, while correcting the estimated standard errors for clustering. The association structure is typically captured using a set of association parameters, such as correlations or odds ratios. Often, generalized estimating equations (GEE) (Zeger and Liang, 1986; and Liang and Zeger, 1986) are used to account for the clustering of outcomes. In this approach, instead of specifying the full distribution for the correlated binary response, we make assumptions about the mean, variance and correlation. For example, let represent the response of flock k of holding j in country i. There are a variety of possible working correlation structures. Some of the more popular choices are: Independence: The simplest choice is the independence working model, i.e., European Food Safety Authority,

47 . Exchangeable: When there is no logical ordering for the observations within a cluster, an exchangeable correlation structure may be more appropriate: Autoregressive: When repeated samples are taken at the same holding, an autoregressive correlation structure might be of interest, assuming that the correlation between samples depends on the time lag between samples: Unstructured: A totally unspecified correlation matrix is given by Any of these choices are justified since estimation using the GEE method is robust against misspecification of the working correlation structure. However, misspecification of the correlation structure can come at the cost of efficiency of the parameter estimates (Molenberghs and Verbeke, 2005). As was mentioned before, in this report prevalence estimates for Salmonella in fattening and breeding turkeys, are obtained starting from (1), and considering the logit link function such that. (2) Observe that in this model, represents the odds of success. Using the GEE methodology, we can obtain an estimate for together with a 95% confidence interval. This interval is based on the robust or empirical standard errors from assuming an exchangeable working correlation structure, a plausible choice given that there is no logical ordering of flocks within a holding. Independence is chosen only when problems occur while trying to fit the exchangeable structure. Note that independence in itself would also be a good option; however this assumption could decrease the efficiency of the prevalence estimates when an exchangeable working correlation structure is more appropriate. From (2) now also follows the relation between and, given by, which provides us with an estimate for the prevalence, as well as a corresponding 95% confidence interval. Observe that in this report the models are used only to obtain prevalence estimates. Since no model building is performed in this analysis, no model diagnostic or remedial measures are required to study the goodness-of-fit. European Food Safety Authority,

48 Further note that the application of a model which accounts for clustering is not complicated by data where only one flock per holding was available. For instance, in countries where only one observation per holding is available, the GEE will simplify to a classical logistic regression assuming independence between observations. Weighting Most statistical procedures analyze the data as if they were collected as a simple random sample. As a result, these procedures may underestimate the variability present in the data, when the data actually arise from complex surveys. Assigning weights to the observations is one possible approach to correct for the differences between the complex survey design and simple random sampling. In general, by using weights, we try to reconstruct the total population, in order to avoid that certain strata or subpopulations are over- or underrepresented. In this report, two weighting schemes are considered for the prevalence estimation: No weights: takes into account each observation as it is. This would disregard the disproportionate sampling at the level of the countries and within the holdings. Year-aggregated weights: as represented in Table III.2.1 Table III.21.1 Year-aggregated weights to account for the disproportionate sampling of holdings within country, and flocks within holding. Level Country (MS) Holding Flock Weight WY1 = total # holdings / sampled # of holdings (within country) WY2 = total # flocks / sampled # of flocks (over one year period within holding) Note that WY2 is calculated using the total number of flocks in a holding, over a one year period. Since this count is unknown, it needs to be estimated from the available data as the median number of flocks at time of sampling multiplied by the median number of cycles within that holding. The weights to be used when studying prevalence are therefore, on the level of the European Community: o W_EU = WY1*WY2 on the level of the MS and Norway: o W_MS = WY2 Finally, observe that the sum of these weights gives an indication of the size N of the population of flocks in the EU, or the flock population size within each MS. However, the sum of the weights should be equal to the original sample size, to make sure that the estimated standard errors European Food Safety Authority,

49 of the prevalence estimates correctly reflect the variability 1. If the sum of the weights exceeds the sample size, this would lead to overly optimistic standard errors (being too low as they would be based on an artificially (too) large sample). Therefore, to avoid overemphasizing the importance of the flocks used in the sample, the weights W_EU (W_MS) are standardized to sum to N s ( ), i.e., the sample size rather than the total flock population N ( ). This implies that, for flock k, in holding j, in country i: If then. Therefore, we will use the standardized weights. On the level of each MS separately, a similar standardization procedure is applied to obtain the standardized weights. Note that in countries where only one observation per holding is available, the impact of this sample is artificially enlarged to reflect the population of flocks in that holding. In these cases the implicit assumption is then that the within-holding correlation equals 1. Correlation In addition to the association between outcomes from the same holding, it is also of interest to study the correlation between the estimated prevalence of Salmonella in breeding and fattening turkeys in each MS and Norway. A first approach to study the correlation consists in graphically representing the estimated probability of infection in breeding turkeys and in fattening turkeys for each member state i. The estimated probabilities are obtained from models as described in the previous section. A scatterplot of (for all i) gives a first impression of the correlation between outcomes in breeding and fattening turkeys. A second and more formal way of studying the correlation is to calculate a Spearman rank correlation. This is a non-parametric rank correlation procedure which can be used when the joint distribution of the variables of interest considerably deviates from a bivariate normal, or, as is the case in this survey, when few data pairs are available. This coefficient is constructed as follows. First, all prevalence estimates in breeding and fattening turkeys are expressed in terms of ranks, denoted by and respectively. The Spearman rank correlation is then defined as a Pearson correlation based on the rank data, i.e.,, where is the mean of the ranks, and is the mean of the ranks. Like an ordinary correlation coefficient, r S takes values between -1 and 1 only. The coefficient equals 1 when the ranks of the prevalence estimates in breeding turkeys are identical to those of fattening turkeys, 1 The need for standardization is software, in fact, some software procedures automatically standardize the weights whereas other procedures require the user to adjust the weights. European Food Safety Authority,

50 i.e., perfect association between the ranks of the variables. On the other hand, r S equals -1 when there is an inverse association between the ranks. The Spearman rank correlation can be used to test H 0 of no association between the two rankings. The probability distribution of r S can be calculated using the condition that, for any ranking of, all rankings of are equally likely when no association is present between and (Neter et al., 1996; Sprent and Smeeton, 2001). III-3 Methodology for the sensitivity analysis The number of samples per flock is expected to affect the probability of finding at least one positive result. Therefore, it is of interest to examine the effect of varying sample size on the probability of detection of the infection in a flock, focusing on the flock prevalence estimation of S. Enteritidis or S. Typhimurium in the EU and for each MS and Norway separately. This is done through a bootstrap simulation study. In each simulation run, the following steps are taken: data is simulated for each flock by resampling from the original data of samples (without replacement). This is repeated B = 250 times; and the prevalence of finding infection in the flock is estimated. This results in an estimate (and confidence interval) of the prevalence of infection in case fewer than 5 samples would have been taken. European Food Safety Authority,

51 IV- Results IV-I Summary of raw results The number of sampled flocks of fattening and breeding turkeys, and the percent of positive flocks, for each MS, and Norway, and for the EU is reported in the following tables IV 1.1 and IV.1.2. Table IV.1.1 Raw percent of positive breeding flocks for each group of Salmonella serovars considered in the baseline survey in the Member States and Norway Serovars other than Countries N Salmonella spp. S. Entiriditis and/or S. Typhimurium S. Enteritids and/or S. Typhimurium proportion positive (%) Bulgaria Czech Republic Finland France Germany Greece Hungary Ireland Italy Poland Slovakia Spain Sweden The United Kingdom EU Total Norway N = number of sampled flocks European Food Safety Authority,

52 Table IV.1.2 Raw percent of positive fattening flocks for each group of Salmonella serovars considered in the baseline survey in the Member States and Norway Serovars other than S. Countries N Salmonella spp. S. Entiriditis and/or S. Typhimurium Enteritids and/or S. Typhimurium proportion positive (%) Austria Belgium Bulgaria Cyprus Czech Republic Denmark Finland France Germany Greece Hungary Ireland Italy Lithuania Poland Portugal Slovakia Slovenia Spain Sweden The Netherlands TheUnited Kingdom EU 3, Norway N = number of sampled flocks IV-2 Results of the prevalence estimations In this section, we discuss the results obtained from the prevalence estimations. Recall that these were obtained using GEE with either an independent or an exchangeable working correlation structure. Fattening turkeys In Table IV2.1 we have displayed the prevalence estimates of Salmonella spp., S. Enteritidis and/or S. Typhimurium, and Salmonella serovars other than S. Enteritidis and/or S. Typhimurium. European Food Safety Authority,

53 Table IV2.1 Prevalence estimates and 95% confidence intervals for the fattening turkeys in the EU , by outcome variable and weighting scheme. No weights Weights Estimate LB UB Estimate LB UB Salmonella spp S Enteritidis or S. Typhimurium Serovars other than S. Enteritidis or S. Typhimurium LB = lower bound of 95% confidence interval UB = upper bound of 95% confidence interval In the European Union, prevalence of Salmonella spp. in fattening turkeys is estimated as 30.7%, 3.8% of the sampled flocks are infected by S. Enteritidis or S. Typhimurium, whereas 27.3% of the sampled flocks are infected by other types of Salmonella serovars. Further, the use of weights does not affect much the estimates for the EU, at least compared to the estimates obtained without weights. More substantial differences can be observed in the prevalence estimates obtained for each Member State separately. First, note that when no positive flocks were observed in a Member State, confidence intervals were not estimated and are therefore left blank in the tables. Further, given that the data for France and Italy contain very few holdings with more than one sampled flock, there is not enough information available to estimate the correlation structure. Therefore, the results for these countries, are obtained from a GEE model with independent working correlation structure rather than an exchangeable structure. European Food Safety Authority,

54 Table IV2.2 Prevalence estimates and 95% confidence intervals of Salmonella spp. for the fattening turkeys in the Member States and Norway , by weighting scheme. Salmonella spp No weights Weights Estimate LB UB Estimate LB UB Austria Belgium Cyprus Czech Republic Denmark Finland France * Germany Greece Hungary Ireland Italy * Lithuania Poland Portugal Slovakia Slovenia Spain Sweden The Netherlands The United Kingdom Norway Bulgaria * Results from GEE with independent working correlation structure Observe in Table IV2.2 that the use of weights or not can have quite an impact on the obtained prevalence estimates. Consider for instance the prevalence estimates for Austria. Taking into account the weights which are correcting for the survey design increases the prevalence estimate from 17% (without weights) up to 26% (with weights). For this country, 35 positive flocks were observed in the sample of 202 flocks. This results in a prevalence estimate of Using the weights, the sample is then transformed to provide an accurate representation of the total population of fattening flocks, in which around 97 flocks are positive out of a total population of flocks. Here, the count 97 is obtained as the sum of the unstandardized weights W_MS for observations infected by Salmonella spp., whereas corresponds to the sum of the unstandardized weights of all observations in Austria. This results in a prevalence estimate around That the difference between weighted and unweighted prevalence estimates in Austria seems considerably big, can be explained by observations like the flock sampled from holding 76. In this case 1 flock (positive) was sampled from an estimated number of 18 flocks available in that holding. After standardization, this results in a weight of 9.5, clearly influencing the final prevalence estimates. The impact of the weights is expected to be largest when one flock is sampled in a holding housing many. 1 Observe that this value is not exactly the same as the prevalence estimate of presented in Table IV1.2. However, it should be noted that this value was estimated taking into account the clustering of flocks within a holding. European Food Safety Authority,

55 Table IV2.3 Prevalence estimates and 95% confidence intervals of S. Enteritidis and/or S. Typhimurium for the fattening turkeys in the Member States and Norway , by weighting scheme. S. Enteritidis or S. Typhimurium No weights Weights Estimate LB UB Estimate LB UB Austria Belgium Cyprus Czech Republic Denmark Finland France * Germany Greece Hungary Ireland Italy * Lithuania Poland Portugal Slovakia Slovenia Spain Sweden The Netherlands The United Kingdom Norway Bulgaria * Results from GEE with independent working correlation structure European Food Safety Authority,

56 Table IV2.4 Prevalence estimates and 95% confidence intervals for Salmonella serovars other than S. Enteritidis and/or S. Typhimurium for the fattening turkeys in the Member States and Norway , by weighting scheme. Serovars other than S. Enteritidis or S. Typhimurium No weights Weights Estimate LB UB Estimate LB UB Austria Belgium Cyprus Czech Republic Denmark Finland France * Germany Greece Hungary Ireland Italy * Lithuania Poland Portugal Slovakia Slovenia Spain Sweden The Netherlands The United Kingdom Norway Bulgaria * Results from GEE with independent working correlation structure Still, the impact of the weighting scheme should not be overemphasized either. As it can be seen from Table IV1.2, the prevalence estimated without weights is contained in the confidence intervals of the weighted estimates in all countries, indicating that the difference between the estimates may not be significant at all. The same holds for the prevalence estimates displayed in the Tables IV2.3 and 4. Breeding turkeys In Table IV1.5 we have displayed the prevalence estimates in breeding turkeys of Salmonella spp., S. Enteritidis and/or S. Typhimurium, and Salmonella serovars other than S. Enteritidis and/or S. Typhimurium. It is clear that prevalence of Salmonella spp. is much lower in breeding turkeys than in fattening turkeys. Now, only around 14% of the breeding flocks in the EU are affected. Around 2% of the sampled flocks are infected by S. Enteritidis and/or S. Typhimurium, whereas 12% is infected by serovars other than Enteritidis or S. Typhimurium. European Food Safety Authority,

57 Table IV1.5 Prevalence estimates and 95% confidence intervals for the breeding turkeys in the EU , by outcome variable and weighting scheme. No weights Weights Estimate LB UB Estimate LB UB Salmonella spp S. Enteritidis or S. Typhimurium Serovars other than S. Enteritidis or S. Typhimurium Additionally, in contrast to the results displayed in Table IV1.1, the prevalence estimates in the EU are almost doubled when the weighting scheme is taken into account. Many countries sampled only one flock of breeding turkeys in each holding. Therefore, the observations in these countries, like e.g. Italy, can have a big impact on the prevalence estimates when the weights are used. Consider for instance the extract from the data with the observations for Italy in Table IV1.6. European Food Safety Authority,

58 Table IV1.6 Data extract: weights calculated for breeding flocks in Italy Holding SalmSpp WY1 Total Flocks Sampled Flocks WY2 SWeightEU SWeightMS In Italy, 27 holdings were sampled out of 53 (WY1=1.96). Further, in Holding 2317, one flock was sampled from a total of 16 estimated available flocks (WY2=16). This would imply that the weight on the EU-level would be equal to around 32 (standardized afterwards to around 8 in the column labelled SWeightEU), and the weight on the MS-level would be equal to 16 (standardized afterwards to around 2.5, in the column labelled SWeightMS). Clearly, the weights on the EU-level are inflated compared to the standardized weights on the MS level. As a result also the prevalence estimates on the EU level are affected. This is a consequence of the design of the survey, and the impact of countries where only one flock per holding was sampled. If all flocks had been sampled, the impact of the weights would have been much smaller. We should therefore be careful in the interpretation of these results. For future surveys, designs should be considered or encouraged which provide a sample for a holding which is representative for the population in the holding. Further, observe that the impact of these weights can also be observed in the prevalence estimates for Salmonella spp. for the Member States, which are summarized in Table IV1.7. European Food Safety Authority,

59 Table IV1.7 Prevalence estimates and 95% confidence intervals of Salmonella spp. for the breeding turkeys in the Member States and Norway , by weighting scheme. Salmonella spp No weights Weights Estimate LB UB Estimate LB UB Czech Republic Finland France Germany Greece Hungary Ireland Italy * Poland Slovakia Spain Sweden The United Kingdom Norway Bulgaria * Results from GEE with independent working correlation structure The small MS-level weights which are displayed in Table IV1.6 now decrease the estimated prevalence for Italy. In other countries like Hungary and the UK similar results can be observed. For instance, in Hungary, 2 positive flocks were observed out of a total of 13 sampled flocks. This results in a prevalence estimated by (or taking into account clustering). Using the weights, the 2 positive flocks still represent 2 positive flocks, but now in an estimated population of 33 flocks. Now the value 2 follows as the sum of the unstandardized weights W_MS for observations infected with Salmonella spp., whereas 33 represents the sum of all unstandardized weights jointly. As a result, prevalence is estimated as (or taking into account clustering). Again, it should be noted that the impact of the weighting scheme should not be overemphasized as the prevalence estimated without weights are all contained in the confidence intervals of the corresponding weighted estimates. The difference between the estimates may not be significant at all. From Table IV1.8-9 it follows that breeding flocks in Slovakia are only infected with serovars other than S. Enteritidis and/or S. Typhimurium. On the other hand, in Italy infections are caused by S. Enteritidis and/or S. Typhimurium (8.3% of the flocks), as well as other serovars (17% of the flocks). Further, only a small amount of flocks in France and the UK are infected by S. Enteritidis and/or S. Typhimurium. European Food Safety Authority,

60 Table IV1.8 Prevalence estimates and 95% confidence intervals of S. Enteritidis and/or S. Typhimurium for the breeding turkeys in the Member States and Norway , by weighting scheme. S. Enteritidis or S. Typhimurium No weights Weights Estimate LB UB Estimate LB UB Czech Republic Finland France Germany Greece Hungary Ireland Italy * Poland Slovakia Spain Sweden The United Kingdom Norway Bulgaria * Results from GEE with independent working correlation structure Table IV1.9 Prevalence estimates and 95% confidence intervals of Salmonella serovars other than S. Enteritidis or S. Typhimurium. for the breeding turkeys in the Member States and Norway y, by weighting scheme. Serovars other than S. Enteritidis or S. Typhimurium No weights WS4 Estimate LB UB Estimate LB UB Czech Republic Finland France Germany Greece Hungary Ireland Italy * Poland Slovakia Spain Sweden The United Kingdom Norway Bulgaria * Results from GEE with independent working correlation structure European Food Safety Authority,

61 Correlation between prevalence estimates of fattening and breeding turkeys As discussed in Section III.2, the Spearman correlation is a non-parametric rank correlation procedure, i.e., it is expressed in terms of the ranks of the prevalence estimates of fattening and breeding turkeys. In Table IV1.10 we have reported the ranks of the estimates for each member state in the European Union, by outcome variable and flock production group. Table IV1.10 Prevalence estimates and corresponding ranks for the three different outcomes, for fattening and breeding turkeys in the European community, Breeding turkeys Fattening turkeys Estimate Rank Estimate Rank Salmonella spp Finland Sweden Bulgaria Germany France Greece Slovakia Poland Ireland The United Kingdom Italy Czech Republic Spain Hungary S. Enteritidis or S. Typhimurium Finland Greece Ireland Slovakia Sweden Bulgaria Germany Spain Hungary France Poland The United Kingdom Italy Czech Republic Serovars other than S. Enteritidis or S. Typhimurium Finland Sweden Bulgaria Germany France Greece Slovakia Poland Czech Republic Ireland The United Kingdom Italy Spain Hungary Note that in case of ties, the average of the ranks is used. Observe from this table that many (0,0) combinations occur. The scatter plots in Figure IV1.1 indicate that there seems to be some correlation between the weighted prevalence estimates of Salmonella in fattening and breeding turkeys in the Member States. European Food Safety Authority,

62 (a) (b) (c) (d) (e) (f) Figure IV1.1. Correlation: scatterplot of the weighted prevalence estimates of Salmonella in breeding and fattening turkeys in the Member States, (a) Salmonella spp., (b) S. Enteritidis-S. Typhimurium, and (c) serovars other than S. Enteritidis and/or S. Typhimurium. Scatterplot of the ranks of the prevalence estimates in breeding and fattening turkeys in the Member States. (d) Salmonella spp., (e) S. Enteritidis and/or S. Typhimurium, and (f) serovars other than S. Enteritidis and/or S. Typhimurium. This is corroborated by the Spearman Correlation Coefficient displayed in Table IV1.13. Significant correlation was observed for Salmonella spp. and for S. Enteritidis and/or Typhimurium (p < 0.05), whereas correlation for serovars other than S. Enteritidis and/or S. Typhimurium was near the threshold of statistical significance. Table IV1.13. Spearman Correlation Coefficients and corresponding p-values for the test of the correlation between the weighted prevalence estimates of breeding and fattening turkeys equal to zero. Spearman Correlation Coefficient p-value SalmSpp SalmET SalmOth However, these significant results are based on calculations taking into account countries where the infection was absent from both fattening and breeding turkeys. European Food Safety Authority,

63 IV-3 Results of the sensitivity analysis Figure IV2.1 displays the results from the bootstrap simulation to study the impact of the number of samples taken per flock on the prevalence estimates of S. Enteritidis-S. Typhimurium in flocks with fattening turkeys. (a) (b) Figure IV2.1 Simulated EU prevalence and 95% uncertainty intervals of S. Enteritidis-S. Typhimurium in (a) fattening turkey flocks, for sample sizes less than 5. In fattening turkeys, we can observe a reduction of prevalence following sample sizes < 5. Although in most MS the prevalence estimates displayed in Figure IV2.2 seem to stabilize after taking 3 samples per flock, this is not the case for countries like Czech Republic, France, Hungary and the Netherlands, which still seem to have a steep increase (reflected also on the level of the European Union). European Food Safety Authority,

64 Figure IV2.2. Simulated MS-specific prevalence of S. Enteritidis-S. Typhimurium: positive fattening turkey flocks and 95% uncertainty intervals for sample sizes less than 5. Figure IV2.3. Simulated MS prevalence of S. Enteritidis-Typhimurium: positive breeding turkey flocks and 95% uncertainty intervals for sample sizes less than 5. European Food Safety Authority,

65 V- Discussion and Conclusions Using weights to obtain the prevalence estimates reported in Section III.2 is a way of accounting for the fact that statistical methods are used which assume a simple random sample, when the data were obtained from a complex survey design. If this were to be ignored, these procedures could produce biased results. In this study, the weights seemed to have some significant impact only in the prevalence estimates of Salmonella in the breeding turkeys in the EU. When only one flock per holding was sampled, these single observations are assigned a weight which should allow them to represent all flocks in their holding. Therefore, their outcomes can have a considerable impact on the final prevalence estimates. For future surveys, designs should be considered or encouraged which provide a sample for each holding which is representative for the population in the holding. The sampling of more than one flock per holding has considerably reduced the impact of single observations on the final prevalence estimates. The significant correlation between prevalences of Salmonella spp. in breeding and fattening flocks was consistent with the hypothesis of an epidemiological association between these two flock types within the same MS. A sensitivity analysis of the sampling design suggests that, in fattening turkeys, the number of samples taken per flock might affect prevalence estimate for Salmonella Enteritidis S. Typhimurium. Accordingly, collecting less than five samples would have yielded lower prevalence values compared with those reported here. This result could be partially associated with the fact that each sample was taken from a different part of the house where a flock was reared. Because the distribution of S. Enteritidis and/or S. Typhimurium in the house may not be uniform, excluding one or more samples would reduce the probability of detecting the infection. European Food Safety Authority,

66 Annex II. General features of the European turkey population. An overview of the breeding and fattening turkey populations in Europe is presented in Table 7 and in Table 8. The figures for the total number of holdings were directly reported by each country, whereas the number of flocks of birds was inferred from quantitative sampling data originated from the baseline survey (number of flocks per year equalled the number of holdings multiplied by the median number of flock per holding at the time of sampling multiplied by the median number of cycles per holding). Five MSs did not contribute to this data; Estonia, Latvia and Luxembourg that did not have commercial turkey flocks and Malta and Romania that did not participate to the survey.. The maps (Figures 11 to 16) display the proportion of the number of holdings, flocks and birds, for each country out of the EU total using a grey scale code. Breeding Turkeys A total of 401 breeding turkey holdings were reported in the EU. France had the highest number of holdings (209), followed by Italy with 53, while The United Kingdom, Germany, Slovakia and Poland reported 29, 23, 22 and 19 respectively. Other countries reported ten or less holdings. Ten countries reported not having commercial turkey holdings while The Netherlands reported having two, but did not provide any sampling data. The data presented in the tables represent flocks with at least 250 birds. Of a total of 1,702 breeding flocks, France has also the highest number (836) followed by The United Kingdom (232) and Italy (159) and Sweden had the smallest (2). An estimated 3,784,417 birds were present in the EU at the time of the survey. France had the highest number of breeding turkeys (2,119,260), followed by Italy (451,560) and The United Kingdom (384,076). Fattening Turkeys A total of 7,520 holdings of fattening turkeys was reported in the EU. The highest number of holdings (2,591) was submitted by France, followed by Italy with 848 and Poland with 729. Estonia, Latvia and Luxembourg did not report any holding. The data presented in the tables represent flocks with at least 500 birds. A total of 28,701 flocks were estimated in the EU. The highest number was calculated for France (5,182) followed by Germany (4,020) and Italy (3,392) while Cyprus had the smallest (20). A total of 113,931,230 fattening birds were calculated for the EU, France reported the largest population (20,106,160), followed by Germany, Italy, Spain and Slovakia with 17, 17, 15 and 14 million birds respectively. European Food Safety Authority,

67 Table 1. Overview of the breeding turkeys in the EU 1 and Norway, Member States Breeding turkey flocks with at least 250 birds No of holdings No of flocks No of birds Austria Belgium Bulgaria ,823 Cyprus Czech Republic ,789 Denmark Finland ,688 France ,119,260 Germany ,558 Greece ,750 Hungary ,680 Ireland ,000 Italy ,560 Lithuania Poland ,823 Portugal Slovakia ,600 Slovenia Spain ,500 Sweden 1 2 3,310 The Netherlands 2 no data no data The United Kingdom ,076 EU 401 1,702 3,784,417 Norway ,120 European Food Safety Authority,

68 Figure 1. Distribution of breeding turkey holdings in the EU and Norway, For all countries, the EU total population (therefore, without Norway) was used as the denominator. This approach was adopted also for the other maps. European Food Safety Authority,

69 Figure 2. Distribution of breeding turkey flocks in the EU and Norway, European Food Safety Authority,

70 Figure 3. Distribution of breeding turkeys in the EU and Norway, European Food Safety Authority,

71 Table 2. Overview of the fattening turkeys in the EU 1 and Norway, Member States Fattening turkey flocks with at least 500 birds No of holdings No of flocks No of birds Austria ,380,000 Belgium ,884 Bulgaria ,704 Cyprus ,000 Czech Republic ,200 Denmark ,000 Finland ,139,600 France 2,591 5,182 20,106,160 Germany 670 4,020 17,688,000 Greece ,000 Hungary 467 3,736 8,779,600 Ireland ,472,076 Italy 848 3,392 17,180,480 Lithuania ,696 Poland 729 2,916 14,580,000 Portugal ,257,600 Slovakia ,000 Slovenia ,600 Spain 482 2,892 15,877,080 Sweden ,150 The Netherlands ,980,000 The United Kingdom 460 1,380 1,656,000 EU 7,097 27, ,670,830 Norway ,000 European Food Safety Authority,

72 Figure 4. Distribution of fattening turkey holdings in the EU and Norway, European Food Safety Authority,