Content uploaded by Alberto Vidal-Diez
Author content
All content in this area was uploaded by Alberto Vidal-Diez
Content may be subject to copyright.
Evaluation And Optimization Of Scrapie Surveillance In Great Britain: A
Bayesian Framework
Vidal Diez; A1 Arnold; ME1 Del Río Vilas; V2
1 Centre for Epidemiology and Risk Analysis, Veterinary Laboratories Agency, United Kingdom
2 Department for Environment, Food and Rural Affairs, London, United Kingdom
ABSTRACT
There are two main active surveillance sources in Great Britain: the fallen stock (FS) and the abattoir survey
(AS). The FS is a volunteer scheme where farmers report dead sheep on farm. In constrast the AS is a random
survey at the slaughterhouse. Del Río Vilas et al.(2008) estimated using capture-recapture methods the holding
prevalence as 0.45%. Ortiz Pelaez et al. (2009) reported a 0.65% within-holding prevalence. There is evidence
that scrapie prevalence has been reduced in Great Britain in the last years, however we consider it necessary to
evaluate the power of detection of the current surveillance.
Objective: Calculate the sensitivity of the surveillance system and determine an optimal sampling strategy
combining both sources.
We applied a Hierarchical Bayesian Mixture Model based on previous studies (Branscum et al, 2006).
We combined both surveillance sources to study the factors involved in the reduction of the sensitivity at country,
county and holding level. We modelled scenarios where we compared a targeted surveillance (stratification by
county) and a more practical surveillance approach (stratification by holding size).
We proved that the sensitivity of the surveillance could be improved if we sample more sheep within holdings
rather than testing more holdings and a small number of sheep in each holding. We also reported the impact of
changing several parameters in the model. Stratification by holding size was shown to be preferable to stratifying
by county. The sensitivity of the current sampling strategy was established as less than 4.85%.
INTRODUCTION
The prevalence of classical scrapie in Great Britain has decreased since an EU-wide surveillance and Scrapie
control programme was created in 2002. The active surveillance system is based on two main sources: the
abattoir survey (AS) and the fallen stock (FS). DEFRA has sampled different number of sheep through each
stream every year. The FS used to detect more classical scrapie than the AS because infected sheep have high
probability of dying on farm. However atypical scrapie is mainly detected through the AS.
The evaluation of the sensitivity of active surveillance for any disease is essential in order to modify the system
to reflect the disease evolution. It becomes more important for diseases like TSE where testing is expensive and
the prevalence is very low.
Both surveys were designed to produce an estimate of the prevalence of scrapie in the sheep population [further
details of these two surveys are reported elsewhere (Elliott et al., 2005; Del Rio Vilas et al., 2005)]
Sample sizes are calculated based on the country’s adult sheep population and are sufficient to detect a
prevalence of 0.5% and 0.05% in the FS and AS respectively. The geographical distribution of the samples
follows a basic proportional allocation at the two points of sampling: the Animal Health Divisional Office (AHDO)
for the FS and a selection of slaughterhouses for the AS.. This narrow approach, however, seems wasteful
particularly when the application of control measures is done at the holding level (Ortiz-Pelaez and Del Rio Vilas,
2009).
Ebel et al. (2008) discussed several approaches which assess the effectiveness of surveillance systems
assuming a design prevalence below which a population can be considered disease free. Branscum et al. (2006)
used Bayesian Hierarchical Models (BHM) where he included non-infected models by using mixture models.
This approach provides great flexibility to analyse multiple levels and parameters within the surveillance systems
in order to find ways of improvement.
We aimed to study the efficacy of the sampling strategy and the accuracy of the current holding prevalence
estimates. Three aggregation levels were introduced into the model and other sampling strategies were also
investigated.
METHODOLOGY
Data
In order to represent the current surveillance system in Great Britain, we collected the following information:
• The population and location of adult sheep in Great Britain from the GB census database of 2004.
• Surveillance results from 2005 at holding (Del Rio Vilas et al, 2008) and within holding level (Ortiz
Pelaez et al, 2009)
• European standard estimates of within-holding prevalence.
• Number of collected adult sheep from the National Fallen Stock Scheme and Company (NFSCo).
Model
Three main stages make up the structure of the solution process:
On the first stage we generated the population and its level of infection at different aggregation levels. Then we
emulated the sampling strategy of both surveillance sources. Our samples were simulated with similar levels of
infection and sample sizes found in previous years. We saved the sample results.
The second stage consisted of implementing the BHM. The prior information was represented by distributions
which informed the level of infection in the population. The data came from the previously simulated sample
results. Two outputs were obtained from the model: the 10th percentiles of the distributions of the proportion of
truly infected holdings detected within the sample and the proportion of truly infected holdings within the
population.
The last step involved the repetition of stage one and two generating different populations. The final output will
be the 10th percentile of the distributions of the results across each population.
RESULTS
We investigated two main sampling strategies: 1) a simulation of a targeted surveillance scenario implemented
by a multi-stage sampling design with a stratification by county. 2) A simulation modeling a more practical
surveillance where we stratified by holding-size. In general the latter strategy produced better sensitivity.
A significant decrease in the sensitivity was found when we tested small numbers of sheep within a holding but
testing many holdings in comparison to testing fewer holdings but sampling more sheep within each holding.
The influence of the within-holding prevalence was presented comparing scenarios with the estimates of Ortiz-
Pelaez and Del Río Vilas (2009) and the suggested estimates by the European Union. We reported the impact
of the parameter which represents the probability that an infected sheep will die on farm.
The simulation of a scenario similar to the current situation produced a sensitivity estimate of 3.01% for the fallen
stock stream. The abattoir survey did not even reach 1% sensitivity.
Some scenarios studied the possibility of using the National Fallen Stock Scheme and Company as a
surveillance source.
CONCLUSIONS
We conclude that BHM models are very flexible and powerful for investigating issues in surveillance systems.
Optimal sampling designs can be found using these models.
We highlighted that the sensitivity was significantly affected by the small sample size within the holdings. A
sampling design based on stratification by holding size was found more efficient than a target surveillance
strategy. In general terms the sensitivity of the system is poor but the voluntary nature of the FS and the small
probability of finding classical scrapie through the AS make changes difficult. We propose to investigate other
possible sources like the National Fallen Stock Scheme and Company.
REFERENCES
Branscum A.J., Johnson W.O, Gardner,I.A.(2006) Sample size calculations for disease freedom and prevalence
estimation surveys. Statistics in Medicine 25;2678-2674
Del Río Vilas, V.J., Sayers, R., Sivam, K., Pfeiffer, D., Guitian, J., Wilesmith, J.W. (2005) A case study of
capture–recapture methodology using scrapie surveillance data in Great Britain. Prev Vet Med 67(4) 303-317
Del Rio Vilas, V.J., Bohning, D. (2008). Application of one-list capture-recapture models to scrapie surveillance
data in Great Britain. Prev Vet Med 85; 253-266.
Ebel E.D, Williams, M.S., Tomlinson, S.M (2008) Estimating herd prevalence of bovine brucellosis in 46 U.S.A.
states using slaughter surveillance. Prev Vet Med 85; 295-316
Elliott, H., Gubbins, S., Ryan, J., Ryder, S., Tongue, S.,Watkins, G.,Wilesmith, J., (2005). Prevalence of scrapie
in sheep in Great Britain estimated from abattoir surveys during 2002 and 2003. Vet. Rec. 157, 418–419.
Ortiz Pelaez, A., Del Rio Vilas, V.J. (2009). Within-holding prevalence of sheep classical scrapie in Great Britain.
BMC Vet Res 5:1
