Using maths to target Salmonella risk
The team from the Finnish Food Safety Authority (Evira) and the Swedish Zoonosis Center used modelling to estimate the occurrence of Salmonella from samples of egg-laying flocks.
How often and the time samples are collected influences the prevalence estimates, according to the research.
Salmonella control
Salmonella control programmes provide data from microbial testing results which can inform about the prevalence in food production.
A positive test result means the pathogen is detected, while a negative result means no detection.
Under low prevalence the control samples are often all negative, showing no observable variation of results. However, the underlying true prevalence may not be zero, because all detection methods are imperfect.
Jukka Ranta, docent and senior researcher from Evira's risk assessment research unit, said the models can be used to update annual estimates.
“This particular model estimates the proportion of production under possible but yet undetected infections in flocks,” he told FoodQualityNews.com.
“The required data represents sampling results from Salmonella control programmes: the number of samples and the corresponding sampling times relative to age of flocks.
“Based on these, and different scenarios related to the detection probabilities, it is possible to estimate the underlying unobserved true prevalence while accounting for some of the inherent uncertainties that were thought to be important by international poultry experts.
“This can have an impact on the planning of control programmes when assessing and comparing residual risks”
Industry-wide potential
The outcome of the flock-level module is needed in the other modules describing prevalence in eggs and the food chain, to get final estimates of exposure.
A continuous time two-state hidden Markov process model was used to describe prevalence of Salmonella infected flocks over laying phase in egg production
Ranta said it is only a part of the models that are used and combined to assess Salmonella risks over food sources, and towards source attribution modelling.
“The combination of the models gives a tool for making annual source attribution estimates for salmonella, based on annual data,” he said.
“These are not used in industry but mainly for government research for evaluating national control programmes. However, an earlier version of this particular model was modified and applied in an EFSA 2010 assessment on Salmonella in laying hens.”
The mathematical models used were Bayesian hierarchical models, including a ‘process model’ and ‘observation model’, to describe the infection state and the probability of detection in flocks of laying hens.
Implementation of computations was done with OpenBUGS software.
Source: Journal de la Société Française de Statistique Volume 154 Number 3 (2013)
Authors: Jukka Ranta, Antti Mikkelä, Pirkko Tuominen and Helene Wahlström
