Development of STOC-FREE MODEL

The aim of WP1, which is led by ONIRIS France, is to develop a method (STOC-FREE MODEL) for the quantitative comparison of the confidence in freedom of disease in different control programmes for non-regulated diseases in the EU.

The method STOC-FREE MODEL will allow the estimation of the confidence of freedom from infection and the associated uncertainty from heterogeneous data inputs available for different epidemiological units such as animal, herd, sector, region or country. The method will be developed and evaluated using BVDV in cattle as an example disease.

WP1: Highlights of year 1

During the first year, WP1 focussed on the development of a conceptual model representing the course and dynamics of infection at different levels and the exploration of possible statistical methods that showed potential to be used in this specific context. From 5 September on, PhD student Mathilde Mercat started to work on the project.

Conceptual model

The conceptual model described the infection process for the STOC free case disease - BVD - at 3 levels. At the animal level, the different infection states and the transitions between states (such as susceptible, infectious or resistant) were evaluated.  At the herd level, the model considered herd demography, contact structure and the transmission pathways. At the territory level, the model represented possible transmission pathways from outside to within the territory.  The conceptual model was developed and mapped the different types of information that existed for a given infectious disease onto the true status regarding infection.

The model connected:

·         The biological system: the true status regarding infection which is of interest for different levels of analysis:  animal, herd and territory.

·         Information that is extremely diverse. Conceptually, two types of information that are different in nature can be distinguished:

o   Information generated and collected to specifically detect the pathogen of interest such as test results from control programmes

o   Information associated with an increased probability of pathogen presence such as risk factors of infection

The conceptual model was delivered in April 2018 and will be used to design the appropriate statistical models that will integrate different pieces of information (data) for the estimation of probabilities of being in each single state of interest (outcome) at different levels.

Statistical model

After evaluating and discussing different statistical approaches, development of a Bayesian network model appeared the most promising method to use in STOC free. Bayesian networks are flexible and allow for heterogeneous input information. Such information can be incorporated by inclusion of

prior distributions for the parameters in the model. The prior distributions can be based on default information on for example country level but can be tailored to each specific situation by entering more specific information. Data to specify the distributions to specific situations can be obtained from databases of control programmes, demographic data and contact structures between herds that will have a heterogeneous nature. In addition, frequency of occurrence and risk estimates for factors that influence either the probability of introduction or delayed detection of the infection in an animal or herd will be included in the model. 

WP1: Highlights of year 2

During the second year, WP1 proceeded with the development of a statistical model based on the chosen method: a special type of Bayesian networks called a hidden Markov model, which allows incorporating infection dynamics in the estimation. The results of the conceptual framework were delivered together with a document containing guidelines for identification and sources of data.

An initial simple version of the model was developed and discussed between the partners. It is a herd-level model in which the probability of becoming infected (τ1) is influenced by the occurrence of risk factors and the probability of clearing an infection (τ2). The latter (τ2), among other things, depends on the CP in place. The first version was discussed and decisions were made about the time steps used in the model (monthly), the risk factors that should be included and the amount of CP information that should be taken into account. Risk factors that are included are herd size, introduction of cattle into the herd and the risk from neighbouring herds (prevalence of disease and/or livestock density). The model includes parameters describing the CP in place (risk mitigation (including vaccination) + test system), the test characteristics and information such as the time since freedom was achieved. Later on the probability of freedom and associated uncertainty will be estimated for specific strata in the population based on risk factors such as herd size (small/medium/large), introduction of cattle (yes/no), test scheme (BTM test, tag test, spot test), and neighbourhood risk (low/high).

In the second year, an initial simple version of the model was developed using French data. In the third year, it is foreseen to expand the model by adding risk factors and more detailed CP information. Thereafter, the model will be tested using case studies to validate and further improve the model.

WP1: Highlights of year 3

During the third year, WP1 proceeded with the development of a statistical model based on the chosen method: a special type of Bayesian networks called a hidden Markov model, which allows incorporating infection dynamics in the estimation.

The initial version of the model that was developed on herd-level was expanded with an animal level module for countries that perform their CPs on animal level. The sensitivity of the different input parameters in the models were tested by evaluating the model results of the simulated data when incorporating a range of pre-defined parameter values. The results were discussed and the model adapted accordingly. After this exercise the model with the computer code and a handbook was provided to the members of the team together with an exercise dataset. Members could opt to start directly with their own data (WP3) or to apply the model on the sample data first. During the annual meeting in October 2019, a half day workshop was organised by the members of WP1. The aim of this workshop was to let the team members get acquainted with the model and to answer all questions so far. Between July 2019 and march 2020, each member was applying the model on the data of their own country and in this process feedback was provided which led to several updates of the model. In April 2020 the model was finalised and delivered to EFSA together with the computer code.

Deliverables of all years can be found here