A meta-analytical estimation of the detection limits of methods for Salmonella in food

The validation of microbial detection methods for foods does not typically state a limit of detection value. Therefore performances of rapid and reference methods for Salmonella detection in foods were compared retrospectively using data from 49 published studies. A total of 576 values for the limit of detection (LOD50) were calculated. The major scientific variables in these independent validation studies were food matrices, Salmonella serovars, and method types (reference, cultural and immunological and nucleic acid based). The basic design feature of the original studies was comparison of the performances of new methods with those of cultural reference methods. A major experimental design variable was the number of laboratories per study. There were 29 single laboratory studies with 20 replicates per level of Salmonella spiked into the food matrix. The 20 multi-laboratory (N ≥ 10) studies had 5–6 replicates per level of Salmonella. The LOD50 values of the dataset had a mean value of 0.021 MPN g− 1 (standard deviation range (0.013–0.036) and the distribution of their logarithms was symmetrical about the logarithm of the mean if the outlier values were discounted. Eighty-nine percent of the values ranged from 0.01 to 0.04 MPN g− 1. On this basis about 11% of the values were deemed outliers. About three-quarters of them were > 0.04 MPN g− 1. The distribution derived in this study can be used as a bench mark against which to evaluate LOD50 data generated in future studies of detection methods for Salmonella in foods and potentially offers a possible way to streamline method validation study experimental design.