Bayesian estimation of flock-level sensitivity of detection of Salmonella spp., Enteritidis and Typhimurium according to the sampling procedure in French laying-hen houses.

A study was carried out to estimate the prevalence of flocks infected by Salmonella spp., S. Enteritidis and S. Typhimurium in 521 French laying-hen farms from October 1st 2004 to September 30th 2005 as part of a European Union-wide baseline study to define targets for Salmonella reduction in member states. The sampling scheme prescribed and financed by the European Commission to detect Salmonella in laying-hen flocks was based on 2 dust-samples and 5 faeces-samples per farm. A latent-class Bayesian approach for correlated tests was used to estimate the sensitivity of detection of reduced sampling schemes corresponding to the 16 combinations of 2 dust- and 5 faeces-samples. For each model the full sampling scheme (7 samples) and the reduced protocol were considered as two correlated tests, the biological principle being identical and the reduced protocol being a subset of the full sampling scheme. As the observed apparent prevalence in cage flocks was higher than in other systems (barns, outdoor, or organic) these two sub-populations were considered separately. Bayesian estimation of posterior medians with 95% probability intervals for true prevalence in cage flocks were 0.34 (0.29; 0.39) and 0.13 (0.10; 0.18) for Salmonella spp. and Salmonella Enteritidis+Typhimurium respectively. In alternative flocks posterior medians with 95% probability intervals for true prevalence were 0.09 (0.06; 0.13) and 0.05 (0.03; 0.08) for Salmonella spp. and Salmonella Enteritidis+Typhimurium, respectively. In cage flocks Bayesian estimation of posterior distributions for sensitivity indicated that at least 5 samples, including 2 dust samples were necessary to attain comparable sensitivity levels to the full sampling scheme. In alternative flocks and for Salmonella spp. 6 samples were required to ensure a comparable sensitivity level to the full sampling scheme. Detection sensitivity was improved by increasing the number of dust samples in cage farms and by increasing the total number of samples whatever their type in alternative farms.