Estimating the diagnostic accuracy of three culture-dependent methods for the Listeria monocytogenes detection from a Bayesian perspective

Abstract

The objective of this study was to estimate the test accuracy measures (classification probabilities [CPs], predictive values [PVs], likelihood ratios [LRs] and area under receiving operating characteristic curve [AUC]) of three different culture-dependent methods, commonly used during routine analysis for the detection of the foodborne pathogen Listeria monocytogenes, from a Bayesian perspective. Data from a previous study by Andritsos et al. (2010) were used to define measures of accuracy for the diagnostic tests. Samples of minced pork meat obtained from local markets were tested for L. monocytogenes presence by parallel testing using selective media (PALCAM, ALOA and RAPID'L.mono). Dirichlet distribution, which is the multivariate expression of a Beta distribution, was used to analyze the data. Bayesian analysis determines characteristics of the posterior distribution from available prior information. Results showed that all methods were best at ruling in L. monocytogenes presence than ruling it out. PALCAM seemed to have better performance based on positive PV, positive LR and AUC, but it was not so sensitive as RAPID'L.mono was. Results also showed that none of the media were perfect in detecting L. monocytogenes, i.e. sensitivity and specificity equal to one. Besides, the problem of observing zero counts may be overcome by applying Bayesian analysis, making the determination of a test performance feasible.