Bayesian risk assessment for Salmonella in egg laying flocks under zero apparent prevalence and dynamic test sensitivity

Ranta, Jukka; Mikkelä, Antti; Tuominen, Pirkko; Wahlström, Helene

Journal de la société française de statistique, Tome 154 (2013) no. 3, pp. 8-30.

Un modèle à processus de Markov caché à deux états en temps continu a servi à décrire la prévalence de cheptels infectés par les salmonelles durant la phase de ponte dans le cadre de la production d’œufs. L’état infectieux d’un cheptel a été traité comme une variable cachée binaire susceptible d’être détectée comme étant positive aux salmonelles uniquement par des tests microbiologiques imparfaits. La sensibilité du test dépend du type d’échantillonnage et de la méthode d’analyse employés mais aussi de la phase inconnue de l’épidémie parmi les poules du cheptel. Dans un jeu de données issu d’un programme de contrôle national sous prévalence très basse, il est possible que tous les tests à tous les âges génèrent des résultats négatifs. Cependant, une certaine incertitude temporellement variable demeure en regard de la prévalence réelle inconnue, du fait des évolutions temporelles de la sensibilité d’ensemble des tests. En définissant la sensibilité comme une fonction de la durée de l’épidémie au sein du cheptel, un modèle Bayésien a été développé pour l’évaluation quantitative des risques. En employant des hypothèses minimales dérivées de connaissances expertes ou de scénarios plausibles, les effets de la sensibilité des tests changeante au cours du temps ont été pris en compte par intégration sur la durée inconnue de l’infection. Le modèle de sensibilité a été combiné avec le modèle de processus de Markov caché, conditionnellement á la séquence temporelle des résultats de tests. Les calculs ont été effectués avec OpenBUGS.

A continuous time two-state hidden Markov process model was used to describe prevalence of salmonella infected flocks over laying phase in egg production. The infection status of a flock was treated as a binary hidden variable that can be detected as salmonella positive only by imperfect microbiological testing. Sensitivity of the test depends on the sampling type and analysis method used, but also on the unknown phase of epidemic among the hens within the flock. In a data set obtained from a national control programme under very low prevalence, all tests at all ages may show negative results. However, some temporally varying uncertainty remains about the unknown true prevalence, due to temporal changes in overall test sensitivity. By defining the sensitivity as a function of duration of within flock epidemic, Bayesian modeling was developed for quantitative risk assessment. Using minimal assumptions derived from expert knowledge or plausible scenarios, the effect of dynamically changing test sensitivity was accounted for by integration over the unknown time of infection. The sensitivity model was combined with the hidden Markov process model, conditional to temporal sequence of test results. Computations were performed using OpenBUGS.

Keywords: hidden Markov process, sensitivity, detection, identifiability, uncertainty, Bayesian Risk Assessment, QMRA, OpenBUGS, laying hens, salmonella
Mot clés : processus de Markov caché, sensibilité, détection, identifiabilité, incertitude, évaluation bayésienne des risques, QMRA, OpenBUGS, poule pondeuse, salmonella

[1] Albert I., Grenier E., Denis J.-B., Rousseau J. Quantitative Risk Assessment from Farm to Fork and Beyond: A Global Bayesian Approach Concerning Food-Borne Diseases, Risk Analysis, Volume 28 (2008) no. 2, pp. 557-571

[2] Albert I., Espié E., de Valk H., Denis J.-B. A Bayesian Evidence Synthesis for Estimating Campylobacteriosis Prevalence, Risk Analysis, Volume 31 (2011) no. 7, pp. 1141-1155

[3] Arnold M.E., Carrique-Mas J.J., McLaren I., Davies R.H. A comparison of pooled and individual bird sampling for detection of Salmonella in commercial egg laying flocks, Preventive Veterinary Medicine, Volume 99 (2011), pp. 176-184

[4] Bureau A., Shiboski S., Hughes J.P. Applications of continuous time hidden Markov models to the study of misclassified disease outcomes, Statistics in Medicine, Volume 22 (2003), pp. 441-462

[5] Branscum A.J., Gardner I.A., Johnson W.O. Estimation of diagnostic-test sensitivity and specificity through Bayesian modeling, Preventive Veterinary Medicine, Volume 68 (2005), pp. 145-163

[6] Bonde M., Toft N., Thomsen P.T., Sørensen J.T. Evaluation of sensitivity and specificity of routine meat inspection of Danish slaughter pigs using Latent Class Analysis, Preventive Veterinary Medicine, Volume 94 (2010), pp. 165-169

[7] Carrique-Mas J.J., Davies R.H. Salmonella Enteritidis in commercial layer flocks in Europe: Legislative background, onfarm sampling and main challenges, Brazilian Journal of Poultry Science, Volume 10 (Jan-Mar, 2008) no. 1, pp. 1-9

[8] Cook R.J. A Mixed Model for Two-State Markov Processes Under Panel Observation, Biometrics, Volume 55 (1999), pp. 915-920 | Zbl

[9] De Vylder J., Dewulf J., Van Hoorebeke S., Pasmans F., Haesebrouck F., Ducatelle R., Van Immerseel F. Horizontal transmission of Salmonella Enteritidis in groups of experimentally infected laying hens housed in different housing systems, Poultry Science, Volume 90 (2011), pp. 1391-1396

[10] Douc R., Garivier A., Moulines E., Olsson J. Sequential Monte Carlo smoothing for general state space hidden Markov models, The Annals of Applied Probability, Volume 21 (2011) no. 6, pp. 2109-2145 | MR | Zbl

[11] Dewaele I., Van Meirhaeghe H., Rasschaert G., Vanrobaeys M., De Graef E., Herman L., Ducatelle R., Heyndrickx M., De Reu K. Persistent Salmonella Enteritidis environmental contamination on layer farms in the context of an implemented national control program with obligatory vaccination, Poultry Science, Volume 91 (2012), pp. 282-291

[12] EFSA Panel on Biological Hazards (BIOHAZ) Scientific Opinion on a quantitative estimate of the public health impact of setting a new target for the reduction of Salmonella in laying hens, EFSA Journal, Volume 8 (2010) no. 4, 86 pages (Available online: www.efsa.europa.eu) | DOI

[13] European Food Safety Authority Quantitative risk assessment of Salmonella Enteritidis in shell eggs in Europe, EFSA Journal, Volume 8 (2010) no. 4, 43 pages (Available online: www.efsa.europa.eu) | DOI

[14] Fienberg S.E. Bayesian Models and Methods in Public Policy and Government Settings, Statistical Science, Volume 26 (2011) no. 2, pp. 212-226 | MR | Zbl

[15] Gradel K.O., Andersen J., Madsen M. Comparisons of Sampling Procedures and Time of Sampling for the Detection of Salmonella in Danish Infected Chicken Flocks Raised in Floor Systems, Acta vet. scand., Volume 43 (2002), pp. 21-30

[16] García C., Soriano J.M., Benítez V., Catalá-Gregori P. Assessment of Salmonella spp. in feces, cloacal swabs, and eggs (eggshell and content separately) from a laying hen farm, Poultry Science, Volume 90 (2011), pp. 1581-1585

[17] Hernes G. Hot Topic - Cold Comfort. Climate Change and Attitude Change, NordForsk, 2012

[18] Klinkenberg D., Thomas E., Calvo Artavia F.F., Bouma A. Salmonella Enteritidis Surveillance by Egg Immunology: Impact of the Sampling Scheme on the Release of Contaminated Table Eggs, Risk Analysis, Volume 31 (2011) no. 8, pp. 1260-1270

[19] Karlin S., Taylor H.M. A First Course in Stochastic Processes, Academic Press, San Diego, 1975 | MR | Zbl

[20] Lunn D., Best N., Spiegelhalter D., Graham G., Neuenschwander B. Combining MCMC with ’sequential’ PKPD modelling, J Pharmacokinet Pharmacodyn, Volume 36 (2009), pp. 19-38

[21] Maijala R., Ranta J., Seuna E., Pelkonen S., Johansson T. A quantitative risk assessment of the public health impact of the Finnish Salmonella control program for broilers, Int. J. Food Microbiology, Volume 102 (2005) no. 1, pp. 21-35

[22] Miconnet N., Cornu M., Beaufort A., Rosso L., Denis J.-B. Uncertainty distribution associated with estimating a proportion in microbial risk assessment, Risk Analysis, Volume 25 (2005) no. 1, pp. 39-48

[23] Mahé A., Bougeard S., Huneau-Salaün A., Le Bouquin S., Petetin I., Rouxel S., Lalande F., Beloeil P.A., Rose N. Bayesian estimation of flock-level sensitivity of detection of Salmonella spp., Enteritidis and Typhimurium according to the sampling procedure in French laying-hen houses, Preventive Veterinary Medicine, Volume 84 (2008), pp. 11-26

[24] Nagelkerke N.J., Chunge R.N., Kinoti S.N. Estimation of parasitic infection dynamics when detectability is imperfect, Statistics in Medicine, Volume 9 (1990), pp. 1211-1219

[25] Ranta J., Maijala R. A probabilistic transmission model of salmonella in the primary broiler production chain, Risk Analysis, Volume 22 (2002) no. 1, pp. 47-58

[26] Ranta J., Tuominen P., Maijala R. Estimation of True Salmonella Prevalence Jointly in Cattle Herd and Animal Populations Using Bayesian Hierarchical Modeling, Risk Analysis, Volume 25 (2005) no. 1, pp. 23-37

[27] Ranta J., Siekkinen K.M., Nuotio L., Laukkanen R., Hellström S., Korkeala H., Maijala R. Causal hidden variable model of pathogenic contamination from pig to pork, Statistical Modelling, Volume 10 (2010) no. 1, pp. 69-87 | MR | Zbl

[28] Sweeting M.J., De Angelis D., Aalen O.O. Bayesian back-calculation using a multi-state model with application to HIV, Statistics in Medicine, Volume 24 (2005), pp. 3991-4007 | MR

[29] Smid J.H., Verloo D., Barker G.C., Havelaar A.H. Strengths and weaknesses of Monte Carlo simulation models and Bayesian belief networks in microbial risk assessment, International Journal of Food Microbiology, Volume 139 (2010), p. S57-S63

[30] Spiegelhalter D.J., Riesch H. Don’t know, can’t know: embracing deeper uncertainties when analysing risks, Phil. Trans. R. Soc. A., Volume 369 (2011), pp. 4730-4750 | MR | Zbl

[31] Thomas M.E., Klinkenberg D., Ejeta G., Van Knapen F., Bergwerff A.A., Stegeman J.A., Bouma A. Quantification of Horizontal Transmission of Salmonella enterica Serovar Enteritidis Bacteria in Pair-Housed Groups of Laying Hens, Applied and Environmental Microbiology, Volume 75 (2009) no. 19, pp. 6361-6366

[32] van Schaik G., Schukken Y.H., Crainiceany C., Muskens J., VanLeeuwen J.A. Prevalence estimates for paratuberculosis adjusted for test variability using Bayesian analysis, Preventive Veterinary Medicine, Volume 60 (2003), pp. 281-295

[33] Van Hoorebeke S., Van Immerseel F., De Vylder J., Ducatelle R., Haesebrouck F., Pasmans F., de Kruif A., Dewulf J. Faecal Sampling Underestimates the Actual Prevalence of Salmonella in Laying Hen Flocks, Zoonoses and Public Health, Volume 56 (2009), pp. 471-476