Entrepôts, Représentation et Ingénierie des Connaissances
Séminaires d'ERIC
Pour toute question concernant les séminaires, vous pouvez contacter Nadia Kabachi (Nadia.Kabachi@univ-lyon1.fr) et Jairo Cugliari (Jairo.Cugliari@univ-lyon2.fr)

Recherche approfondie

par Année
par Type
par Thème
feed

(1) Séminaire(s)

-

Lun. 24/04/2017 13:00 K71, Bâtiment K, RdC

ROBERT Valérie (Univ Paris Saclay)
Le modèle des blocs latents multiple. Application à des données de pharmacovigilance

Sommaire:

Adverse drug events are most often discovered after the marketing authori-
sation of these drugs. The pharmacovigilance system therefore aims at detecting as soon as
possible potential associations between some drugs and adverse effects. From this stand-
point, several statistical methods of automatic signal generation (IC, Bate and al., 1998 ;
GPS, Dumouchel, 1999 ; multiple comparisons, Ahmed, 2009...) have been developed for
over twenty years. Nevertheless, these explanatory methods suffer from limitations as they
are based on aggregated data (contingency table), which suppose some homogeneity in the
individuals. But it is reasonable to believe that the studied population is heterogenous.
The aim of this work is to propose an alternative to these methods by dealing with the
heterogeneous dimension of the problem thanks to the study of the individual data which
produce sparse matrices. Within this framework, a new approach is proposed in pharmaco-
vigilance by developing a model adapted from the latent block model (Govaert and Nadif,
2008). It enables to co-cluster rows and columns of two binary tables by imposing the
same row ranking. It also allows to hightlight subgroups of individuals sharing the same
drug profile and subgroups of adverse effects and drugs with links. In this presentation,
the model will be introduced first and sufficient conditions for its identifiability will be
given. Then, the algorithms used for estimating the model will be developed and some
experimentations will be presented.
Keywords. Pharmacovigilance – Bayesian Methods – Mixture Models – Co-clustering
– EM – Variational Approximation – Gibbs Sampler


Pour plus d'informations, merci de contacter Cugliari J.