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

1. INTRODUCTION
Clinical mastitis in the dairy cow
(CMAST) is the archetype of a multifacto-
rial disease [8, 31]. Udder inflammations-
infections are determined through numer-
ous factors (breeding practices, including
housing hygiene and the quality of milking,
feeding characteristics, weather conditions,
genetic parameters) as well as aetiological
determinants (major, minor and facultative
pathogens, macro and micro traumas) [4, 6,
28, 81]. Moreover, since such factors can be
implicated in different study levels (herd,
cow, teat) which are not independent,
CMAST is at the same time a predilection
field for modelling and a restricting meth-
odological framework.
All statistical models used by the epide-
miologist to study CMAST risk factors or
predict CMAST occurrence require suita-
bility between the hypotheses for statistical
modelling and the samples analysed. What-
ever analysis method used to identify risk
factors, the hypothesis of independence
between the observed events is especially
necessary and remains a major restricting
problem concerning this field of study.
CMAST modelling involves different lev-
els of dependence, which are potential
sources for the overdispersion of the
observed data at the different study levels:
between different herds of the same region,
within the same herd surveyed over a long
time period, between different animals
within the same herd, between successive
lactations/cases for the same animal, or
between udder quarters (Fig. 1). Moreover,
important potential sources for overdisper-
sion are factors which are not taken into
account or, worse, are unidentifiable. Iden-
tifying and taking into account an overdis-
persion in the analysis process is essential
for the validity of the statistical results, the
estimation of the parameters and the level
of significance of the statistical tests [21,
61].
Faced with such difficulties, the authors
tried to take these different constraints into
consideration. The aim of the present arti-
cle was to review and discuss the main
methodological issues in CMAST model-
ling (except through specific genetic mod-
els) and the solutions which are proposed
by the epidemiologists.