^{1}

^{*}

^{2}

^{1}

^{1}

^{2}

^{3}

^{1}

^{4}

^{1}

^{4}

^{1}

^{4}

^{1}

^{2}

^{3}

^{4}

Edited by: Francisco Ruiz-Fons, Consejo Superior de Investigaciones Científicas (CSIC), Spain

Reviewed by: Cristina Lanzas, North Carolina State University, United States; Andrea Isabel Moreno Switt, Pontificia Universidad Católica de Chile, Chile

This article was submitted to Veterinary Epidemiology and Economics, a section of the journal Frontiers in Veterinary Science

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

Poultry is considered a major source of foodborne salmonellosis globally (

Still, due to the absence of a national control/monitoring program for

Geographic information systems and some spatial statistical analyses have been applied to epidemiological research on

A BYM model contains two spatial random effects, often called spatially structured and unstructured components. The structured component has an intrinsic conditional autoregressive (CAR) prior that takes the geographical contiguity into account (correlated heterogeneity). The geographical contiguity is described by the neighborhood relationships between each pair of areas and a full spatial dependency. The unstructured one is a random effect for non-spatial heterogeneity at the same area level as the structured component. Riebler et al. (

In the current study, we aimed to evaluate the spatial distribution and potential spatial trends of

Data on samples collected for monitoring antimicrobial resistance in

Columbia 5% sheep blood agar (bioMérieux) was used for the incubation of colonies of presumptive

The potential of production-related characteristics to explain at least part of the observed spatial patterns of the risk of

Univariable generalized linear regression results for the risk of

Number of farms | 0.08 | 0.03 | (0.03 to 0.13) | 1.010 (1.004 to 1.016)^{1} |

Density of farms (per km^{2} |
0.07 | 0.02 | (0.03 to 0.12) | 1.014 (1.005 to 1.023)^{2} |

Number of fattening pigs | 0.06 | 0.02 | (0.03 to 0.10) | 1.002 (1.001 to 1.003)^{3} |

Number of sows | 0.06 | 0.02 | (0.01 to 0.10) | 1.010 (1.002 to 1.018)^{3} |

Number of piglets | 0.05 | 0.02 | (0.01 to 0.08) | 1.002 (1.000 to 1.004)^{3} |

Number of weaners | 0.06 | 0.02 | (0.03 to 0.09) | 1.003 (1.001 to 1.005)^{3} |

Number of gilts | 0.05 | 0.02 | (0.01 to 0.09) | 1.040 (1.009 to 1.068)^{3} |

Number of boars | 0.04 | 0.04 | (−0.04 to 0.11) | 1.262 (0.787 to 1.953)^{3} |

Total number of pigs | 0.06 | 0.02 | (0.02 to 0.10) | 1.001 (1.000 to 1.001)^{3} |

Proportion of fattening pigs | 0.08 | 0.04 | (0.00 to 0.15) | 1.009 (1.001 to 1.018)^{2} |

Proportion of sows | −0.11 | 0.06 | (−0.24 to 0.00) | 0.989 (0.976 to 1.000)^{2} |

Proportion of piglets | −0.08 | 0.03 | (−0.15 to −0.02) | 0.989 (0.981 to 0.997)^{2} |

Proportion of weaners | 0.09 | 0.03 | (0.02 to 0.15) | 1.016 (1.004 to 1.028)^{2} |

Proportion of gilts | −0.06 | 0.06 | (−0.19 to 0.04) | 0.976 (0.929 to 1.017)^{2} |

Proportion of boars | −0.09 | 0.06 | (−0.22 to 0.03) | 0.799 (0.566 to 1.091)^{2} |

Density of fattening pigs | 0.07 | 0.02 | (0.04 to 0.11) | 1.003 (1.002 to 1.005)^{4} |

Density of sows | 0.05 | 0.02 | (0.01 to 0.09) | 1.011 (1.002 to 1.020)^{4} |

Density of piglets | 0.05 | 0.02 | (0.01 to 0.08) | 1.002 (1.000 to 1.004)^{4} |

Density of weaners | 0.08 | 0.02 | (0.04 to 0.12) | 1.005 (1.002 to 1.008)^{4} |

Density of gilts | 0.06 | 0.02 | (0.01 to 0.10) | 1.050 (1.012 to 1.089)^{4} |

Density of boars | 0.05 | 0.03 | (−0.02 to 0.11) | 1.643 (0.813 to 3.211)^{4} |

Density of pigs | 0.07 | 0.02 | (0.03 to 0.11) | 1.001 (1.000 to 1.002)^{4} |

Data cleaning, manipulation, and analyses were performed in Microsoft Excel 2013 (Microsoft Corp.), and R program version 3.5.2 (

The overall, yearly and province-level proportion of

The presence of global and local spatial autocorrelation in the spatial distribution of

Additionally, the Poisson model of the scan statistic was also applied to detect the presence of provinces with an increased risk of

Bayesian spatial modeling to assess the associations between the risk of

The BYM2 Poisson model including the selected covariates and the spatial components (Equation 1) was then fitted (

Markov chain Monte Carlo diagnostics for the final model were performed with (a) the potential scale reduction statistic (R^) (

τ_{t}:

For the variables potentially associated with the

Up to 3,730 samples collected over the 15 years in which sampling was conducted, representing the same number of farms, were included in the current study, with an average of 249 (range: 163–384) samples per year. The number of abattoirs where the samples were collected each year, except for 2019 when this information was not available, ranged between 7 and 20. A median of 18 samples (interquartile interval: 11–29, range: 1–60) were collected from each abattoir each year during the study period. Abattoirs were located in 11 out of the 18 autonomous communities in Spain, and 977 (29.2%), 670 (20.0%) and 455 (13.3%) samples came from abattoirs in Cataluña, Castilla La Mancha and Murcia, respectively (

A total of 1,409 of the 3,730 samples were positive, yielding an overall percentage of

Annual proportion of

Proportion of

The serotype of 1,360 (96.5%) out of the total 1,409

Changes in the proportion of

Empirical Bayesian smoothed proportions of farms positive to the four most represented

No global (Moran's

Provinces included in the significant, high-risk

According to the univariable models, provinces with a higher number or density of pig farms showed a higher risk of

After predictive projection, only one covariate, the density of weaners, was selected to be included in the final Poisson BYM2 model (

Regression results from the final multivariable modeling for risk of

Intercept | −0.07 | 0.05 | −0.17 to 0.02 | – |

Density of weaners | −0.01 | 0.06 | −0.13 to 0.10 | −0.001 (−0.008 to 0.007) |

Standard deviation of the spatial component | 0.23 | 0.05 | 0.14 to 0.35 | – |

Proportion explained by the structured component | 0.65 | 0.25 | 0.10 to 0.99 | – |

The spatial component in the final BYM2 model suggested a West-East increasing risk of

The total (left) spatial risk of

Markov chain Monte Carlo and model diagnoses are presented in

The density of weaners at the province level did not experience major changes over the study period, with higher values reported consistently for provinces in the Northeast corner (

Human salmonellosis outbreaks have been linked to pork and pork products in the past (

We detected an overall percentage of

Our results indicated that the recent increase in the percentage of

Several factors may affect the determination of the

Our results showed a clear pattern suggesting a higher risk of

The software used here, Stan, is a highly-expressive probabilistic programming language that allows full Bayesian inference using Hamiltonian Monte Carlo samplers (

To our knowledge, the BYM2 model has been little utilized in the field of veterinary science. The BYM2 model has advantages over the original and some of the reparameterized BYM models (

In the current study, we employed a relatively unexplored approach for variable selection in veterinary epidemiology—predictive projection with a Bayesian penalized regression model as the reference model. Shrinkage methods have been recommended when the ratio of the number of observations to the number of variables is ≤ 10 (

Here, a number of different spatial analytic tools were applied, offering different results. While the global and local Moran's

The current study has some limitations. First, as the sample collection was conducted in abattoirs that have high slaughter capacity, the results might not be necessarily representative of the farms that did not (usually) send their pigs to those abattoirs. Second, the current study was conducted using secondary data. Therefore, it may face some common issues of using secondary data, such as out-of-date information, suboptimal sampling procedure for answering specific research questions, insufficient sample size, and lack of information that would be, otherwise, included. For example, the individual or within-farm prevalence of

The current study shows a notable increasing trend in the risks of

The datasets generated for this study are available on request to the corresponding author.

The current work did not involve any live animal nor human data. No ethical approval was required as all isolates analyzed were retrieved through the ongoing Spanish national surveillance program on antimicrobial resistance, performed according to national and EU regulations.

KT and JA contributed conception and design of the study. KT, MM, MU-R, CB, AT, GL, MAM, LD, and JA participated in the generation, collection, and curation of the data. KT performed the statistical analysis. KT, MM, JA interpreted the results. JA provided supervision. MU-R, GL, and JA facilitated project administration. KT wrote the first draft of the manuscript. MU-R and JA wrote sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

The authors thank Vicente Lopez-Chavarrias for assisting data curation, María García-Martín for the logistics, and all the laboratory staff who have performed laboratory work during the 17 years.

The Supplementary Material for this article can be found online at: