With the approach described in this paper, we developed a user-friendly instrument for the design of risk-based sampling programmes, providing veterinary authorities with a promising tool for future, cost-effective sampling strategies. Taking the gain in information of testing high-risk strata into account, we are able to considerably reduce the sample size. Especially for IBR, a reduction by almost half of the samples was achieved. We explain this fact by the larger number of relevant RF identified and the higher values for their respective RR compared to EBL. Another influential factor in this context is the larger number of farms available in the highest risk strata for IBR compared to EBL. The analysis of cost-effectiveness clearly revealed a financial benefit of cTS&bsRS, when compared to exclusive sRS for both IBR and EBL.
In the case of the theoretical scenario involving solely TS in the highest risk stratum, the reduction in sample size compared to sRS would be major. However, this scenario is hypothetical and based on the assumption that we have a large number of farms available in the highest risk stratum, which in reality is not the case. Nevertheless, it would be possible to conduct solely TS by considering all available farms in the different risk strata with the highest information gain and consequently to further decrease the necessary sample size. But the eventual geographical clustering of an entirely targeted sample due to uneven spread of risk would be a disadvantage in terms of representativeness and coverage of a survey in many regions or countries. The proposed approach of cTS&bsRS assures the representativeness of the survey, while at the same time taking into account the advantages of TS.
The stochastic scenario tree model to calculate CSe or n of the TS component depends on RF selected through literature review, parameterised with estimates based on expert opinion and is therefore subject to some degree of uncertainty. However, the distributions used for the RR and in consequence, consideration of conservative results on the 5%-percentile of the distribution for the CSe of TS provides a certain counterbalance for this issue. A survey based on cTS&bsRS guarantees an OSe of at least over CSebsRS in case the estimations for the RF and RR should have been completely inadequate. Furthermore, the percentage of CSebsRS on the OSe and therefore the degree of uncertainty can be varied and defined according to requirements. It has to be noted that correlation or dependence between RF was not considered in this study. The participants of the expert opinion survey were left free to assign any value to the RR of the evaluated RF. Although it is possible that experts intuitively considered some degree of correlation or dependence between RF, this issue was not addressed in the survey design.
Because a classical validation of a model with reliable field data is nearly impossible for rare diseases, we chose to verify the accuracy of our RF for IBR with past, well documented cases of the disease [46, 47]. All of the three Swiss IBR outbreak farms from the canton of Jura in 2009 had at least one RF applying to them. One farm even had four RF. Consequently, those farms would have a high probability of being selected for a survey based on cTS&bsRS. For EBL however, even this attempt of validation was difficult to achieve, as only very few, poorly documented cases of leucosis actually occurred in Switzerland since the eradication of the disease.
Further surveillance components for IBR and EBL in Switzerland, such as passive clinical surveillance, slaughterhouse inspection and abortion examination, were not taken into account in this project as we aimed at analysing the legally prescribed annual serological survey only.
Additionally, we simulated and analysed the effect of varying input parameters on the SSC and directly explored the effects of several exchangeable parameters on the OSe. We did this using different values for Pr
and checking if the scenario tree model produced logical results.
The problem of testing the same farms year by year can be reduced by a yearly updating of the risk factors per farm. This is a recommended procedure anyway, as RF for the cattle farms can change over time. More importantly, the bsRS has a certain compensational function also in this respect. Furthermore, if a large number of farms are available in a selected high risk stratum, the farms can be selected randomly within this risk stratum, and not all farms of a certain risk stratum would have to be tested. A verification of the accuracy of and, in consequence, updating of the RF and the RR in regular time intervals (i.e. every 5 years) is also a strategy to consider.
The different approaches described in this paper are all based on whole herd testing which corresponds to the sampling framework of IBR and EBL in Switzerland to demonstrate absence of disease on the farm level. However, the model described in this paper can also be modified for diseases with increased within-herd prevalence. For such diseases, the within-herd prevalence has to be included as an additional infection node in the scenario tree model.