*05. V0605. Gasqui. Individual

clinical mastitis / individual model / dairy cow / survival model / recurrence 
Vet. Res. 31 (2000) 583–602
583
© INRA, EDP Sciences
* Correspondence and reprints
Tel.: [phone]; fax: [phone]; e-mail: [e-mail]

---

P. Gasqui et al.
584
1. INTRODUCTION
Mastitis is still the most frequent health
disorder in dairy cattle [7]. It is also the dis-
ease with the most deleterious consequences
on milk quality, hence on economics [22].
The problem has been taken seriously
enough for the dairy industry to envisage
systematic recording of clinical mastitis
cases on dairy cattle farms [24, 28]. Models
for clinical mastitis occurrences are in this
case valuable tools for herd management,
in that they allow to study and rank the
potential influencing factors [32]. Tools that
simulate clinical mastitis occurrence are
emerging, based on results from various ear-
lier studies. They are intended to assist
breeders in the decision making process
while focussing on the economic conse-
quences of the disease [3, 4, 23, 43]. The
elaboration of models that (1) estimate the
effects of various individual and environ-
mental factors and (2) predict clinical mas-
titis occurrence while considering the effects
of these factors, is confronted with a number
of problems. These problems are now
clearly identified and linked, in particular, to
the relatedness of successive mastitis dur-
ing lactation, to the relatedness of succes-
sive lactation during a cow’s career, to the
relatedness of cows in a herd and to the asso-
ciation of clinical mastitis with other dis-
eases that may occur, in particular during
the peri-calving period. To characterize the
relatedness of successive mastitis at herd
level, some authors have proposed to cal-
culate a “reoccurrence rate” [38] (total num-
ber of clinical mastitis cases per number of
lactations with at least one case), and a “lac-
tational incidence risk” [8] (number of lac-
tations with at least one occurrence of the
disease of interest divided by the total num-
ber of lactations at risk).
With this prospect to estimate and pre-
dict, the classic approach generally consists
in fitting to a Generalized Linear Model
(GLM) [33]. However, adjustments using
these models often elicit overdispersion,
considered by many authors as a measure
of deviation from baseline independence