NR AXOY
AU Ingravalle,F.; Maurella,C.; D'Angelo,A.; Bona,M.C.; Possidente,R.; Corci,C.; Caramelli,M.; Ru,G.
TI Learning to Diagnose Scrapie by Means of a Probabilistic Approach
QU International Conference - Prion 2007 (26.-28.9.2007) Edinburgh International Conference Centre, Edinburgh, Scotland, UK - Book of Abstracts: Epidemiology, Risk Assessment and Transmission P04.07
IA http://www.prion2007.com/pdf/Prion Book of Abstracts.pdf
PT Konferenz-Poster
AB
So far from a clinical point of view scrapie is still a puzzling disease. In particular in Italy the passive surveillance does not work so well yet. Nevertheless it shows some quite peculiar clinical features.
Aim of this study is to provide a list of clinical signs (features) that allows the classification of a sheep as scrapie or non-scrapie affected minimising the probability of error (Pe). This is the first step in the construction of bayesian classifiers to estimate the likelihood of being scrapie affected or not in presence or absence of certain features.
282 sheep coming from outbreaks have been visited during a four-years period. On the basis of a standardised clinical examination, the presence or absence of 29 features and the disease were recorded. The work is based on the assumption of accuracy of the features and their non total independence. A bayesian approach was adopted, then we calculated the Pe in classifying a sheep as scrapie or non-scrapie affected given that feature; the same operation was repeated for grouped features.
In order to define a stair of relevance to the disease, single or grouped features were ranked by their Pe using 2 methods: single best method (SB) and sequential forward selection method (SFS).
The results show that the bigger the information (i.e. larger is the subset of feature), the smaller the Pe; a subset of 4 features (applying SFS: nibble, depression, abnormal gait from clinical exam, ataxia; applying SB: nibble, ataxia, depression, pruritus) allows the reduction of Pe lower than 10%, while if we consider 12 feature then Pe decreases around 5%. Then SFS and SB methods may create subsets with same size but different features.
The mandatory report of scrapie suspects is a very important step in the struggle for its eradication. For this reason supplying a brief list of features to the practitioners in order to identify a scrapie affected sheep with low probability of error is very useful.
Two aspects of the present study are relevant: first, data come straight from the on field practice and refer to individual clinical signs; second, Bayesian approach allows to attach a Pe to a subset of a fixed number of features.
AD F. Ingravalle, C. Maurella, M.C. Bona, R. Possidente, C. Corci, M. Caramelli, G. Ru, Istituto Zooprofilattico Sperimentale di Piemonte, Liguria e Valle d'Aosta, Italy; A. D'Angelo, Faculty of Veterinary Medicine, Animal Pathology, Italy
SP englisch
PO Schottland