Use of a Poisson-gamma model to assess the performance of the EC process hygiene criterion for Enterobacteriaceae on Irish sheep carcasses

Gonzales-Barrón, Úrsula; Lenahan, M.; Sheridan, J.J.; Butler, Francis

doi:10.1016/j.foodcont.2011.10.035

Cited by 11 publications

(8 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Here, it is worthy to mention that such practice of replacing zero counts by the limit of enumeration has been very common for statistical treatment either merely descriptive or inferential. Thus, it is clear that, while logarithmic transformation to induce data normality can be suitable for bacterial counts of high occurrence such as mesophile or total viable counts (Gonzales- Barron & Butler, 2011a), this approach may be inappropriate for bacterial counts of lower occurrence such as coliforms or pathogens.…”

Section: Assessment Of the Coliforms Datamentioning

confidence: 99%

“…On the other hand, the discrete heterogeneous Poisson distributions have been recently shown to closely describe microbial data sets containing zeros. While Jongenburger, Reij, Boer, Zwietering, and Boer (2012) indicated that both the Poissonlognormal and the Poisson-gamma can be equally adequate for representing low microbial counts and a large proportion of zero counts; Gonzales-Barron, Lenahan, Sheridan, and Butler (2012), and Gonzales- Barron, Zwietering, and Butler (2013) pointed out that the Poisson-gamma distribution is more suitable than the Poissonlognormal, as the latter tends to overestimate the mean concentration. In fact, Gonzales- Barron & Butler (2011a) have shown that the concentration of coliforms on beef carcasses is best described by the Poisson-gamma distribution.…”

Section: Assessment Of the Coliforms Datamentioning

confidence: 99%

“…While Jongenburger, Reij, Boer, Zwietering, and Boer (2012) indicated that both the Poissonlognormal and the Poisson-gamma can be equally adequate for representing low microbial counts and a large proportion of zero counts; Gonzales-Barron, Lenahan, Sheridan, and Butler (2012), and Gonzales- Barron, Zwietering, and Butler (2013) pointed out that the Poisson-gamma distribution is more suitable than the Poissonlognormal, as the latter tends to overestimate the mean concentration. In fact, Gonzales- Barron & Butler (2011a) have shown that the concentration of coliforms on beef carcasses is best described by the Poisson-gamma distribution. Thus, to carry out descriptive, inferential and predictive statistics on low microbial counts, the alternative approach proposed by the authors is to conduct count data regression models based on heterogeneous Poisson distributions; in this particular case, based on the Poisson-gamma.…”

Section: Assessment Of the Coliforms Datamentioning

confidence: 99%

“…Microbial mean concentrations are given in CFU/g or cm 2 . This is the arithmetic mean, which is a more suitable measure from the view point of risk-based metrics and sampling plans (Gonzales- Barron et al, 2012).…”

Section: Estimated Mean Valuesmentioning

confidence: 99%

“…On the other hand, Gonzales- Barron & Butler (2011a,b) have demonstrated that the lognormal (and Poisson-lognormal) distributions are only appropriate for the representation of high microbial counts, while the Poisson-gamma distribution performs much better in the characterisation of low microbial counts and even for highly clustered microbial data consisting of a large proportion of zero counts (non-detections). Whereas the lognormal distribution cannot naturally take in zero values, the Poisson-lognormal distribution greatly overestimates the mean microbial concentration when it attempts to fit in a high frequency of zero counts (Gonzales- Barron & Butler, 2011a). As a consequence, parametric statistical analyses, such as ANOVA or regression models, should only be performed on bacterial counts of high occurrence (such as mesophile or total viable counts), for which a logarithmic transformation can mostly bring about an approximate normal distribution.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Conducting inferential statistics for low microbial counts in foods using the Poisson-gamma regression

Gonzales-Barrón¹,

Cadavez²,

Butler³

2014

Food Control

Self Cite

View full text Add to dashboard Cite

Mixed Poisson distributions have been shown to be able to represent low microbial counts more efficiently than the lognormal distribution because of its greater flexibility to model microbial clustering even when data consist of a large proportion of zero counts. The objective of this study was to develop an alternative modelling framework for low microbial counts based on heterogeneous Poisson regressions. As an illustration, Poisson-gamma regression models were used to assess the effect of chilling on the concentration of total coliforms from beef carcasses (n ¼ 600) sampled at eight large Irish abattoirs. Three Poisson-gamma and three zero-modified (hurdle and zero-inflated) models were appraised with a series of random-effects variants in order to extract any variability in microbial mean concentration, dispersion and/or proportion of zero counts. Models were compared and validated in their ability to predict the coliforms counts on carcasses after chilling. In all five test batches, the hurdle Poisson-gamma distributions predicted the observed post-chill counts closer than the Poisson-gamma distributions. This is justified by the better capacity of the hurdle model to represent a higher proportion of zero counts, which were in fact observed in the post-chill batches. Thus, with a coded variable (pre-chill/post-chill) as treatment, and extracting the significant variability of batches nested in abattoirs for the coliforms mean concentration (s 2 u ¼ 2.68), the dispersion measure (s 2 v ¼ 2.39) and the probability of zero counts (s 2 w ¼ 0.89), the validated hurdle Poisson-gamma model confirmed that chilling has a decreasing effect on the viability of coliforms from beef carcasses, and that the concentration is reduced by an average (pre-chill to post-chill) factor of 2.2 (95% CI: 2.15e2.24) at batch level. The model also indicated that chilling increases the odds of producing a zero count from a carcass swab in about 13.5 times, and that the higher the coliforms concentration in a batch, the weaker the effect that chilling has to reduce such contamination on the beef carcasses.

show abstract

Section: Assessment Of the Coliforms Datamentioning

confidence: 99%

Section: Assessment Of the Coliforms Datamentioning

confidence: 99%