Estimation of Salmonella prevalence on individual-level based upon pooled swab samples from swine carcasses

Pooling of samples might be an effective means to increase cost-effectiveness in routine surveillance. The present study assessed the effect on the sensitivity of detection of Salmonella when pooling swab samples from swine carcasses compared to individual analyses. A total of 18,984 samples from nine Danish swine abattoirs were collected during 1 year, covering 2017 slaughter days. At each abattoir, swab samples were taken on a daily basis from 10 carcasses randomly selected. From each carcass, an area of 3 cm x 100 cm was swabbed. Five of these samples were analysed individually and the other five were analysed as one pooled sample. Standard culture methods were used. A logistic regression model was built, where the response was whether a sample was Salmonella positive or not. The explanatory factors were abattoir, type of sampling (individual or pooled sample), and season of year 2000 (four quarters). The odds ratio (OR) of the effect of type of sampling in the logistic model accounting for abattoir and season was interpreted as the conversion factor between pooled and individual sample prevalence. The results of the individually analysed samples showed a low prevalence of Salmonella (1.4%). When Salmonella was isolated, mostly only one positive sample was found among the five individually analysed samples per slaughter day. On a few days >1 positive samples' were found (9 out of 2017 days approximately 0.4%). The pooled sample prevalence was 4.1%. Because the individual prevalence was low, the pooled sample prevalence would have been around five times higher than the individual-level prevalence-if there had been no loss of sensitivity. However, we found that due to loss of sensitivity the pooled prevalence was only three times higher (OR = 2.7; CI 2.0-3.7). Therefore, a conversion factor of 3 instead of 5 should be applied to calculate the individual prevalence from a pooled prevalence. This approach has been used in the national surveillance of Danish pork since 2001. The estimated conversion factor and accept of pooling samples do not necessarily apply to a population with a higher prevalence or to other types of samples (e.g. faeces or lymph nodes) or diagnostic procedures.