Comparison of three modelling approaches to predict the risk of campylobacteriosis in New Zealand

New Zealand has the highest rate of reported campylobacteriosis in the developed world. Due to the large economic and health consequences of campylobacteriosis, prediction models for disease are required to be designed to predict accurately the number of campylobacteriosis cases. The Bayesian approach has gained increased interest in recent years for calculating the outcomes of quantitative microbial risk assessment (QMRA). A classical time series and Monte Carlo (MC) modelling were also explored as appropriate techniques to predict campylobacteriosis. A simplified model representing the entire food chain from the farm to the fork with all the variables of interest was used with the Bayesian method. The Auto Regressive Integrated Moving-Average intervention models (ARIMA additive and multiplicative), Holt-Winters (HW multiplicative) and Bayesian methods were considered the best models for predicting the actual 7333 notified campylobacteriosis cases with 7990, 8442, 8666 and 9250 cases, respectively. It is also noteworthy that the notification rate has more or less stabilised since 2008 until 2017. MC modelling provided the least realistic prediction (846,451 cases). The HW method is simple to use and reliable method for time series predictions. However, the Bayesian method provides a prior assessment of any possible intervention in the food chain and provides satisfactory prediction accuracy in spite of the complexity involved in constructing and assigning probabilities from expert knowledge or prior information, linking the nodes and complex software. This study highlighted the importance of the Bayesian model to assess all the factors which may contribute to the campylobacteriosis risk and confirmed that it can provide better conclusions for QMRA than the MC technique because of its interactive link between the data and the parameter (backward inference).