Survival of Campylobacter on frozen broiler carcasses as a function of time

In the Norwegian Action Plan against Campylobacter in broilers, carcasses from flocks identified as positive before slaughter are either heat treated or frozen for 5 weeks to reduce the number of Campylobacter. The objective of this study was to estimate the effect of freezing time and predict the number of Campylobacter on naturally infected or contaminated broiler carcasses following freezing for 2, 4, 6, 8, 10, 13, 21, 35, and 120 days by nonparametric and parametric linear statistical models. From each of the five flocks, 27 carcasses were sampled. Each carcass was cut in two pieces along the chest bone. Half was put into the freezer (-20 degrees C), whereas the other was deskinned and quantitative culturing was conducted from a 10-g sample of the skin. Fifteen frozen halves were selected at random at each time point following freezing from 2 to 120 days, and skin samples from these were cultured quantitatively and qualitatively. In regard to the log reduction of Campylobacter, almost similar results were obtained using three statistical methods; median regression on the change in Campylobacter counts, zero-inflated negative binomial regression, and a Bayesian Markov chain Monte Carlo (decay) model on original counts. Overall, a 2-log reduction of Campylobacter was obtained after 3 weeks of freezing. Only a marginal extra effect was observed when extending the freezing to 5 weeks. Although freezing appears to be an efficient way to reduce the level of Campylobacter on broiler carcasses, in 80% of the carcasses Campylobacter could still be detected using quantitative culturing following 120 days of freezing. Based on the high number of zeros, these data should be modeled by a zero-inflated model. The best statistical fit in regard to goodness-of-fit measures was the zero-inflated negative binomial log link model, closely followed by the Poisson model. Thus, in our continued search for a better way to describe the data, we used the Poisson distribution in the mixed Bayesian decay models.