Instituto de Ciências Biomédicas Abel Salazar – Universidade do Porto (ICBAS-UP), Population Studies Department, Largo Prof. Abel Salazar, Porto, 2, 4099-003, PORTUGAL

Instituto de Saúde Pública da Universidade do Porto (ISPUP), Rua das Taipas, Porto, 135, 4050-600, PORTUGAL

CEMS, University of Exeter, Harrison Building, North Park Road, Exeter, EX4 4QF, UK

UTAD, Veterinary Science Department, Apartado, Vila Real, 1013, 5001-801, PORTUGAL

Abstract

Background

The EU Regulation No 2160/2003 imposes a reduction in the prevalence of

Methods

The data used come from the Baseline Survey for the Prevalence of

Results

The significant associations found (when compared to category “no

Conclusions

The risk factors significantly associated with

Background

Methods

Herd selection

The objectives, the sampling frame, the diagnostic testing methods as well as the collection and reporting of data, and the timelines of the Baseline Survey on the Prevalence of

The sample size was calculated by the PVA and considered the number of swine herds existing in April of 2007, stratified by Region. The sampling frame consisted of 4522 herds, with 204,584 breeding pigs and 1,827,533 pigs in total. The herd inclusion criteria for entering the sampling frame were: to have at least 50 breeding pigs, either for breeding or production purposes. The pig population included in the sampling frame represented 87% of the total registered pig population in Portugal in 2007.The sample size was calculated using the sampling criteria specified in the Commission Decision 2008/55/EC Annex I - expected herd prevalence of 50%, desired confidence level of 95%, accuracy of 7.5% and then apply a finite population correction factor, with an increase of 10% for each group (breeding and production holdings) in case of non-response. A sample of 174 swine herds was randomly selected using probability proportional to the number of herds among the regions in Portugal.

Pen selection

In each herd only the pens with breeding pigs over six months of age were randomly selected. The breeding pigs that have been recently introduced into the herd and were in quarantine were not included in the survey. In each selected herd, faecal samples from 10 pens were taken representing a 95% probability of detecting at least one positive sample if the true prevalence of infected pigs in the population was 10%

Faecal samples collection

The faecal samples were collected and pooled together by the herd veterinary assistant and then sent to laboratory for detection of

At the laboratory, the isolation of

Data collection

A questionnaire was used to collect information about the herd management and potential risk factors for

**Number of pen samples by the categories of the outcome variable**

**Number of pen samples by the categories of the outcome variable**

Legend: 1 (no

**1**

**2**

**3**

**1**

**2**

**3**

**HERD VARIABLES**

Open air

29

0

1

more than 90% external source

Intensive

1242

38

110

>90% home raised

189

10

21

Missing observations

229

8

13

10-90% home raised

476

19

35

Selection and Multiplication Unit

292

8

30

No

1254

42

114

Production Unit

1208

38

94

Yes

Alentejo

229

8

13

Insemination centre – IC

491

11

18

Centre

278

8

34

Own boar + IC

869

23

88

Lisbon and Tagus Valley

Boar from another herd

North

79

3

18

Missing observations

Farrow-to-weaners

Just own herd

585

14

41

Farrow-to-growers

250

7

13

Others sources

906

31

83

Farrow-to-finish

794

26

60

Missing observations

9

1

0

Missing observations

292

8

30

Number of finishers pigs/herd

<100

<3

≥100

1221

42

107

≥3

Missing observations

16

1

3

<170

759

16

55

No

1192

34

94

≥170

741

30

69

Yes

<22

713

20

47

No

1017

38

85

≥22

787

26

77

Yes

483

8

39

<203

759

16

55

Yes

1423

46

121

≥203

741

30

69

No

77

0

3

more than 90% external source

606

24

70

Good

434

7

29

without boars or >90% home raised

735

15

40

Bad

1066

39

95

10-90% external source or home raised

159

7

14

Yes

828

30

72

No

672

16

52

**Number of pen samples by the categories of the outcome variable**

**Number of pen samples by the categories of the outcome variable**

Legend:1 (no

**1**

**2**

**3**

**1**

**2**

**3**

**PEN VARIABLES**

No

1146

30

92

No

Yes

354

16

32

Yes

874

27

80

No

1194

41

98

Dry pellet

229

7

27

Yes

306

5

24

Dry non pellet

Missing observations

Wet

41

3

0

No

1445

45

118

Fully slatted

139

5

10

Yes

Others

1361

41

114

Missing observations

22

0

4

Exclusively own

199

8

8

Only gilts or gilts and others

874

38

73

Bought + Mixture

Without gilts

626

8

51

No

1291

38

111

Only females

1430

44

114

Yes

Males and females

No

1229

45

103

Mating room

210

10

21

Yes

Gestation room

789

26

62

Mixture of room

58

1

14

Compose sample

121

1

11

Farrowing room

Swab

Replacement breeders

Number of pigs in the pen

=10

1284

34

94

>10

216

12

30

Data analysis

From the information gathered in the questionnaires, two new binary variables were created. The first variable groups the questions regarding management of replacing breeding pigs and their source, and was codified as Good if more than 90% of the breeding sows and boars were homebred (also included herds with no boars) and if the semen was not from another herd otherwise it was codified as Bad. The second variable combines the questions about biosecurity measures and was codified as Yes when controls for rodents and birds were implemented, and also if herds had provisions for foot bathing and clothe changing before entering the herd and No otherwise. The variables and their categories were recoded or aggregated to fewer categories as necessary to avoid sparse data problems as shown in Tables

The continuous variables were transformed into categorical using the median values as the cut-off points defining the categories. Their summary statistics are shown in Table

**Variable**

**Mean**

**Standard deviation**

**Minimum**

**Percentile 25**

**Median**

**Percentile 75**

**Maximum**

* 258 pens had more that 10 pigs per pen.

Number of boars

3.9

3.9

0

2

3

4

28

Number of sows

226.5

192.9

8

98

170

300

1077

Number of gilts

34.0

38.3

0

12

22

40

300

Size of the herd (number of breeding pigs)

265.0

216.9

41

109

203

355

1214

Number of pigs per pen*

11.6

8.0

10

10

10

10

130

Because of the low number of cases per serotype (Table

**Serotype**

**Percentage of isolates (n)**

**Percentage of herds that have at least one pen sample positive to the serotype (n)**

**Breeding holdings**

**Production holding**

**All Holdings**

**Breeding holdings**

**Production holding**

**All holdings**

Typhimurium

15.8 (6)

25 (33)

23 (39)

13.6 (3)

25.6 (20)

13.2 (23)

Rissen

18.4 (7)

19.7 (26)

19 (35)

22.7 (5)

19.2 (15)

12.0 (20)

London

21 (8)

13.6 (18)

15 (26)

13.6 (3)

11.5 (9)

7.2 (12)

Derby

15.8 (6)

9.1 (12)

11 (18)

13.6 (3)

8.9 (7)

6.0 (10)

Give

13.1 (5)

5.3 (7)

7 (12)

9.1(2)

5.1 (4)

4.0 (6)

Brandenburg

0 (0)

6.1 (8)

5 (8)

0 (0)

2.6 (2)

1.8 (2)

1,3,19:-:-

2.6 (1)

4.5 (6)

4 (7)

4.5 (1)

6.4 (5)

3.6 (6)

1,4,5,12:i:-

5.3 (2)

3.8 (5)

4 (7)

9.1 (2)

3.8 (3)

3.0 (5)

Bovismorbificans

0 (0)

3 (4)

2 (4)

0 (0)

2.6 (2)

1.2 (2)

Gloucester

0 (0)

2.3 (3)

2 (3)

0 (0)

2.6 (2)

1.2 (2)

Muenchen

2.6 (1)

2.3 (3)

2 (4)

4.5 (1)

3.8 (3)

2.4 (4)

Anatum

0 (0)

1.5 (2)

1 (2)

0 (0)

2.6 (2)

1.2 (2)

Bredeney

0 (0)

0.8 (1)

1 (1)

0 (0)

1.3 (1)

0.6 (1)

Goldcoast

0 (0)

1.5 (2)

1 (2)

0 (0)

1.3 (1)

0.6 (1)

Livingstone

2.6 (1)

0 (0)

1 (1)

4.5 (1)

0 (0)

0.6 (1)

Mbandaka

2.6 (1)

0.8 (1)

1 (2)

4.5 (1)

1.3 (1)

1.2 (2)

Senftenberg

0 (0)

0.8 (1)

1 (1)

0 (0)

1.3 (1)

0.6 (1)

The data have a “natural” multilevel structure: pen faecal samples (first level) nested in herds (second level) and were analysed using a Bayesian hierarchical model with a categorical response variable (three categories). Monte Carlo Markov Chain (MCMC) was used for estimation and this was implemented in the freely available software WinBUGS (BUGS project,

Random effects were included at the herd level to account for the fact that the observations are ‘nested’ in herds. Treating the herd effect as random, also allows for the fact that the number of herds here (167) is a sample of all existing herds. All prior distributions were chosen to be as uninformative as possible. A more detailed description of the model is given in Additional file

**Model framework.**

Click here for file

To decide which variables should be included in this multivariable model, an exploratory analysis was performed by fitting univariable models and considering as candidates for the multivariable model, all variables significant at the 0.15 significance level. Associations between the explanatory variables were tested using a chi-square test and if a significant association (p < 0.05) was found, only the variables with more biological justification were allowed to enter the model.

The final multivariable model was built using a forward selection process until all variables with a significant 95% credible interval were included. The significance level was set at 0.05.

The model ran long enough with sufficient burn-in (5000 iterations) to ensure convergence to the posterior distribution of the parameters. Convergence was assessed by visual inspection of the means in time-series plots but also more formally using the Raftery and Lewis, and the Gelman-Rubin R-hat diagnostics

**WinBUGS code for the categorical multilevel model.**

Click here for file

The presence of confounding was investigated by analysing the correlation matrix of the joint posterior distribution for all model parameters but especially the slope parameters. Correlation values higher than 0.5 where takes to indicate significant correlation.

Posterior predictive simulation was used for model checking as described by Gilks et al.

Results

A total 167 herds (33 breeding and 134 production holdings) responded to the questionnaire and were tested: 76 herds were positive to

Several management practices linked to herd and pen were assessed (Tables
_{1}, 1/τ_{2}) are non-zero. Estimates of 1/τ_{1} and 1/τ_{2} are arbitrarily away from zero (5.8 and 1.4 respectively) and their standard errors are relatively small, indicating that both estimates are different from zero (see Table

**Variable**

**Typhimurium or 1,4,5,12:i:-**

**Other serotypes**

**Coefficient**

**SD**

**OR**

**95% OR CrI**

**Coefficient**

**SD**

**OR**

**95% OR CrI**

Legend: SD – standard deviation, OR – odds ratio, CrI – credible interval, in bold the significant OR for a 95%CrI.

HERD

Region of the herd

Alentejo

0

1.0

0

1.0

Centre

−1.3

1.5

0.28

0.01-4.30

1.5

0.7

**4.57**

**1.33-17.57**

Lisbon and Tagus Valley

−0.5

1.1

0.62

0.07-5.05

0.9

0.6

2.56

0.86-8.36

North

−0.1

1.7

0.88

0.03-24.31

2.6

0.8

**12.9**

**2.97-64.33**

Size of the herd: (number of breeding pigs)

<203

0

1.0

0

1.0

≥203

1.9

0.9

**7.04**

**1.46-60.04**

0.5

0.4

1.65

0.83-3.44

Source of semen

Insemination centre – IC

0

1.0

0

1.0

Own boar + IC

0.4

0.8

1.45

0.24-7.77

1.1

0.4

2.91

1.35-6.83

Boar from another herd

3.7

1.6

**41.22**

**2.46-1392.7**

1.4

0.8

4.18

0.94-19.30

Control of rodents

No

0

1.0

0

1.0

Yes

−2.2

1.8

0.11

0.002- 1.85

−2.0

0.7

**0.13**

**0.03-0.45**

PEN

Number of pigs/pen

=10

0

1.0

0

1.0

>10

1.4

0.7

**4.06**

**1.03-19.73**

0.6

0.4

1.82

0.88-3.79

Age of the breeding sows

Only gilts or gilts and others

0

1.0

0

1.0

Without gilts

−1.8

0.8

**0.17**

**0.03-0.65**

0.2

0.3

1.24

0.68-2.24

Breeding sector room

Mating

0

1.0

0

1.0

Gestation

0.1

0.5

1.11

0.44-3.10

−0.2

0.3

0.81

0.45-1.52

Mixture of animals of different sectors

0.2

1.7

1.17

0.03-24.80

0.8

0.7

2.14

0.54-7.78

Farrowing

−1.0

0.6

0.36

0.10-1.22

−1.0

0.4

**0.38**

**0.17-0.80**

Replacement breeders

−0.9

1.1

0.40

0.04-2.72

0.1

0.7

1.15

0.29-3.88

Source of feed

Exclusively own

0

1.0

0

1.0

Not exclusively own

0.5

1.1

1.63

0.18-17.62

2.0

0.7

**7.29**

**2.25-29.46**

Herd random effect variance

5.8

0.66

1.4

0.24

It can be seen from the analysis of Table

Discussion

This study investigated risk factors for

The outcome variable

The different serotypes of

The model

It was anticipated that the hierarchical structure of the data from our sample could influence the outcome of the analysis. Therefore the statistical approach was chosen to take into consideration the multilevel structure of data from our sample where the pen faecal samples (level 1) are nested in herds (level 2). Some important remarks concerning the statistical approach deserve to be highlighted: the model implemented here showed a good fit, despite the fact there was little information to update the prior distributions. The methodology proposed could offer a general modelling approach to researchers who want to incorporate expert knowledge in the specification of the priors or for those who wish to restrict the priors accordingly to account for lack of information in the response variable which was not the case in this study. Lastly, both WinBUGS and R, are freely available software which is particularly appealing for the purpose of presenting the methodology here as a general modelling tool.

Risk factors for

It can be seen from the analysis of Table

In category “Typhimurium or 1,4,5,12:i:-”

Risk factors for infection with other serotypes of

Concerning the category “other serotypes”, the

Application in control

The results from this work should be taken into account when implementing control and biosecurity programmes to

Conclusion

In Portugal, the prevalence of herds with breeding pigs that had at least one sample positive to serotype Typhimurium or

Competing interests

The authors do not have any competing interest.

Authors’ contributions

CCG was involved in the design and performed the statistical modelling analysis and drafted the manuscript. TE performed part of the statistical analysis and was involved in the revision of the manuscript for intellectual contents. DM was involved in the revision of the manuscript for intellectual contents and statistical analysis. MVP was involved in the revision of the manuscript for intellectual contents. JNR was involved in the design of the statistical analysis, in the drafting and revision of the manuscript for intellectual content. All authors approved the final manuscript.

Acknowledgments

We would like to thank FCT for the PhD scholarship (SFRH/BD/40932/2007) and for the strategic research project Pest-OE/AGR/UIO772/2011. We would like also to thank the Portuguese official veterinary authority (DGAV) for the data.