Review article Statistical modelling for clinical mastitis in the dairy cow: problems and solutions

---

493
Vet. Res. 34 (2003) 493–505
© INRA, EDP Sciences, 2003
DOI: 10.1051/vetres:2003029
Review article
Statistical modelling for clinical mastitis in the dairy 
cow: problems and solutions
Patrick G
ASQUI
*, Jacques B
ARNOUIN
Unité d’Épidémiologie Animale, INRA, 63122 Saint-Genès-Champanelle, France
(Received 9 October 2002, accepted 16 June 2003)
Abstract – Modelling case occurrence and risk factors for clinical mastitis, as a key multifactorial
disease in the dairy cow, requires statistical models. The type of model used depends on the choice
of perception or the study level: herd, lactation, animal, udder and quarter. The validity of the tests
that are performed through these models is especially ensured when hypotheses of independence
between statistical units are respected, and when the model adjustments do not involve
overdispersion faced with the observed data. In the article, the main sources of overdispersion are
identified according to the different levels of perception of mastitis risk. Then, the proposed
solutions to control for overdispersion at each study level are discussed and the difficulty to compare
the study results is highlighted through a variety of methodological choices of the authors. Two
main categories of models are used for modelling clinical mastitis, i.e. generalist exploratory models
and explanatory designed models. The contribution of the explanatory models to improve modelling
accuracy and relevance is documented through the two main published methodological approaches,
the first one being based on a states model, and the second on a survival model. The integration and
optimisation of such explanatory modelling methods should be possible in the future in order to
develop a more global explanatory model including herd risk factors, which could pertinently
predict udder infections (both clinical and subclinical) at the cow, lactation, or even udder and
quarter levels.