NMKL71:5, the international standard method EN ISO 6579:2002 and the Draft Annex D of EN ISO 6579:2002, using the MSRV as a selective enrichment medium, were compared, in relation to their accuracy, specificity and sensitivity for different feed ingredients and feed materials.
Four feed ingredients (wheat grain, soybean meal, rape seed meal, palm kernel meal), finished feed (pellets of pig feed) and scrapings from elevator were collected from a Swedish feed mill and stored at 4–8°C until used. A representative sample of all feed materials were analysed with the NMKL71 method before the experiments started. For each feed material the pH value was measured in the buffered peptone water (BPW) (Oxoid CM 0509, Basingstoke, England) after the pre-enrichment. The water activity (aw) of the feed and feed ingredients was determined at ambient temperature before and after addition of bacterial culture. AquaLab Series 3 analyzer (ADAB Analytical Devices AB, Stockholm) was used according to the manufacturer's instructions.
Salmonella enterica ssp. enterica serotype Typhimurium ST115506 (S. Typhimurium), S. Cubana ST58403 and S. Yoruba ST45506 (from the culture collection at the National Veterinary Institute, Sweden), all isolated from animal feed, were used in the experiments. The strains were stored at -70°C and were cultured on brom-cresol-purpure-lactose agar plates (blue-agar) (Oxoid). On day 1, one colony from the plate was inoculated in 5 ml serum broth (National Veterinary Institute, Sweden) and incubated at 37°C for 18–24 h. Ten-fold dilution series in peptone saline water were made on day 2. The number of cfu in the peptone saline water (PSW) was determined by plating out 0.1 ml from selected dilutions on blue-agar plates, which were incubated at 37°C overnight. In order to simulate more natural conditions, where Salmonella might be stressed, the bacterial cells were kept at 4–8°C in PSW for two days and then re-counted before the experiments were carried out.
Sample preparation and pre-enrichment
The NMKL71, MSRV and the EN ISO 6579:2002 methods were run in parallel and in each experiment one Salmonella serotype, three feed materials and a non-spiked sample of each feed were included. Twenty-five grams of each feed material was weighed into a plastic jar and spiked with 0–1, 1–10, 10-102 or 102–103 cfu, except for the palm kernel meal spiked with S. Typhimurium, which received ten times higher levels (1–10, 10-102, 102–103 or 103–104 cfu/25 g). The volumes used for spiking were approx. 370 μl/25 g. The samples were left in room temperature for 4 h before 225 g of BPW was added, followed by incubation at 37° ± 1°C for 18 h.
Selective enrichment and confirmation
Three drops (equivalent to approximately 0.1 ml) of the BPW were placed separately and equally spaced on the surface of Modified Semisolid Rappaport Vassiliadis agar plates (MSRV) (Oxoid CM 0910) supplemented with 1.0% Novobiocin (Sigma-Aldrich N1628) and then incubated in an upright position due to its semi-solid composition at 41.5° ± 0.5°C for 24 ± 3 h. In parallel, three drops of BPW were inoculated in 10 ml Rappaport-Vassiliadis broth (RVS) (Oxoid CM 0866) and 1 ml to 9 ml Müller-Kauffmann tetrathionate-novobiocin broth (MKTTn) (Oxoid CM 0343). The RVS broth was incubated in a water bath at 41.5° ± 0.5°C for 24 ± 3 h, and the MKTTn at 37° ± 1°C for 24 ± 3 h. Five replicates of each level were used.
After incubation the MSRV plates were examined for suspect Salmonella growth and a sample was plated on Xylose Lysine Deoxycholat agar (XLD) (Lab M lab 32, Axel Johnson Lab System Inc. Solna, Sweden) (with 1.5% Novobiocin) and Brilliant Green agar (BGA) (Oxoid CM 0329). If no migration was noted the plates were incubated for additional 24 h at 41.5° ± 0.5°C and the same procedure was applied. A loop (0.1 ml) of each MKTTn and RVS- broth was also plated on XLD and BGA and incubated at 37° ± 1°C for 24 ± 3 h and then screened for Salmonella colonies.
Five colonies, for each XLD/BGA plate from the lowest spiking level, were plated on blue-agar plates and incubated at 37° ± 1°C for 24 ± 3 h. Confirmation was done by serological agglutination with monovalent anti-O sera, O-13 and O-23 to detect S. Cubana, O-4 and O-5 for S. Typhimurium and O-16 for S. Yoruba. Physiological NaCl was used in order to check for auto-agglutination.
Statistical analysis of performance criteria
Since the spiking levels for palm kernel meal with S. Typhimurium were higher than for the other feed materials, those data were excluded from the calculation of overall accuracy, sensitivity and specificity. The detection level of each method for the different animal feed materials spiked with the respective Salmonella serotype was compared. The accuracy (AC), sensitivity (SE) and specificity (SP) were calculated for each method and statistical analysis was done according to the principals used by European Commission Directorate General Joint Research Centre . This approach was recently used by Eriksson and Aspán  in order to evaluate isolation methods for Salmonella in faeces. The assumption is that all non-spiked samples are negative for Salmonella and only those samples spiked with Salmonella are true positives. Samples being positive on at least one selective agar plate, that is XLD, BGA or both, are considered positive. Based on this, the accuracy, sensitivity and specificity rates were obtained by using the following definitions and equations :
TP True Positives
A sample was defined as true positive when Salmonella was detected in a sample where Salmonella had been added
TN True Negatives
A sample was defined as true negative when Salmonella was not detected in a sample where Salmonella had not been added
FP False Positives
A sample was defined as false positive when Salmonella was detected in a sample where Salmonella had not been added
FN False Negatives
A sample was defined as false negative when Salmonella was not detected in a sample where Salmonella had been added
Accuracy (AC) = (TP+TN)/(TP+TN+FP+FN) (1)
Accuracy (AC) is a measure for the ability of a method to correctly classify samples containing Salmonella as positive for Salmonella, and samples not containing Salmonella as negative for Salmonella.
Sensitivity (SE) = TP/(TP+FN) (2)
Sensitivity (SE) is a measure for the ability of a method to classify a sample containing Salmonella as positive for Salmonella.
Specificity (SP) = TN/(TN+FP) (3)
Specificity (SP) is a measure for the ability of a method to classify a sample not containing Salmonella as negative for Salmonella.
In the present study the relative difference in accuracy, sensitivity and specificity between the different methods were investigated.
When accuracy and specificity are investigated at different spiking levels and in different feed materials the sample number are small and the estimated performance measures are associated with a great deal of uncertainty. To account for this the respective accuracy, sensitivity and specificity under these conditions are reported as Bayesian confidence intervals , rather than point estimates.
The accuracy and specificity of the tests are reported as shortest confidence intervals , under the assumption that all values are equally probable. The calculations were performed using an online calculator on (http://www.causascientia.org/math_stat/ProportionCI.html, Dec 16, 2008). The values reported defines the boundaries of an interval that, with 95% certainty, contains the true value of the accuracy, sensitivity or specificity.