Salmonella occurrence can be evaluated using statistical models


<p>&nbsp;The occurrence and risks of salmonella in, for example, egg production can be evaluated with the aid of mathematical models. Advances in both these models and the analyses that utilise them can also be of use internationally. Methods developed under the leadership of the Finnish Food Safety Authority Evira were presented at an international scientific symposium, along with results obtained using these methods.</p>

Globally, the majority of salmonella infections in humans originate from foodstuffs, and from eggs and pork in particular. Salmonella still causes significant health risks all across Europe. Finland, Sweden and Norway have been combating the health hazards posed by salmonella for decades using a variety of means, including national monitoring programmes. Cases of salmonella in Finnish eggs and pork are extremely rare.

The salmonella monitoring programmes targeted at eggs and egg products are quite similar throughout the Nordic countries and seek to prevent risks right from primary production. One key method is the taking of regular samples from egg-producing premises.

Models clarify complex connections

Modelling has been used to estimate the actual occurrence of salmonella on the basis of samples taken from egg-laying flocks.

These models help in the analysis of salmonella monitoring results and the identification of complex connections in cases of egg-related illnesses. On the basis of each country's results, the models can also estimate how large a proportion of a production facility's eggs could have been produced during an outbreak before infection was detected at the facility.

In addition to outbreaks confirmed by laboratory tests, the actual occurrence of salmonella also includes cases that miss detection due to, for example, lack of sensitivity in laboratory tests or the sample taking system.

“When data is processed by modelling, you can identify which types of conclusions may be inferred and what uncertainties are associated with them. That way, you can also ascertain to what extent this uncertainty depends on the observations themselves, their number, and the assumptions made,” says Docent and Senior Researcher Jukka Ranta from Evira's Risk Assessment Research Unit. 

Modelling increases the precision of estimates

Uncertainty factors are caused by the timing of sample taking, the number of samples taken from an egg-laying flock, and the stage of any potential infection when the sample was taken. Modelling can also account for the uncertainty concerning when an outbreak commenced and was detected. This requires familiarity with salmonella results obtained from laboratory studies of flocks of certain ages.

Bayesian methods were used to model the connections between the timing of sample taking, the stage of any potential infection within the flock and its detection.

They enabled, for example, the calculation of a conditional probability distribution for salmonella infection and its duration, on which detection indirectly depends, from the body of data based on the monitoring results. The calculation used software specialised in the common simulation method for Bayesian models.

“In theory, the method can be adapted to other disease-causing bacteria and could be harnessed in similar monitoring programmes,” says Ranta.

Help in comparing different countries' results

The models can be adapted to similar datasets in international studies. This enables the comparison of different countries' results and monitoring programmes, so that the uncertainties they contain can be better accounted for. 

“We can also study how the timing of sample taking and the number of samples taken affect results. Using models, we can also construct other interesting predictions that can enhance monitoring and make it as effective as possible,” says Ranta.

The results and methods were presented at an international statistics symposium and in its follow-up scientific publication, whose contributors include researchers from the Finnish Food Safety Authority Evira's Risk Assessment Research Unit.

Ranta, J., Mikkelä, A., Tuominen, P., Wahlström, H.
  Bayesian risk assessment for Salmonella in egg laying flocks under zero apparent prevalence and dynamic test sensitivity.  Journal of the French Statistical Society 2013: Vol. 154, No. 3, pp. 8–30. 

For more information, contact:
Senior Researcher, Docent Jukka Ranta, Ph.D.,
Risk Assessment Research Department, tel. +358 40 489 3374







Related Categories: