Skip to Main content Skip to Navigation
Theses

Modèles mixtes à structure latente pour données
longitudinales multivariées hétérogènes :
application à l'étude du vieillissement cognitif et
de la démence

Abstract : The aim of this work was to propose statistical models for longitudinal, heterogeneous and multivariate data in order to describe cognitive evolution and its association with dementia in the elderly. As cognition is not directly observed but measured by various quantitative non Gaussian psychometric tests, we proposed a nonlinear model with a latent process in which cognitive evolution represented by a latent process is described through a linear mixed model including covariates and a Brownian motion. Psychometric tests are linked to the latent process using nonlinear estimated transformations which account for different effects of covariates on the tests. In addition to the description of metrological properties of the tests and recommandations about which test can be used according to the population of interest, we distinguished the impact of covariates on cognitive evolution from their differential impact on the psychometric tests. We then extended this methodology to take into account the heterogeneous cognitive evolution associated with dementia by developping two nonlinear latent class models. By studying the probability of dementia at the end of the follow-up jointly to the cogntive evolution, we distinguished two profiles of cognitive declines which represent the normal cognitive aging process and a pathological cognitive process associated with dementia. Then, by studying the risk of dementia according to age jointly to the cognitive evolution, we highlighted 5 shapes of cognitive decline with age associated with different risks of dementia. Using these models, we proposed two tools for detecting a dementia according to any available cognitive assessment. For each of the three models proposed in this work, we performed a simulation study, applied the methodology to data from the PAQUID cohort and developed methods to evaluate the adequation of the model.
Document type :
Theses
Complete list of metadatas

Cited literature [164 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-00192386
Contributor : Cecile Proust-Lima <>
Submitted on : Tuesday, November 27, 2007 - 5:13:04 PM
Last modification on : Friday, October 23, 2020 - 4:51:44 PM
Long-term archiving on: : Monday, April 12, 2010 - 5:19:55 AM

Identifiers

  • HAL Id : tel-00192386, version 1

Collections

Citation

Cecile Proust-Lima. Modèles mixtes à structure latente pour données
longitudinales multivariées hétérogènes :
application à l'étude du vieillissement cognitif et
de la démence. Sciences du Vivant [q-bio]. Université Victor Segalen - Bordeaux II, 2006. Français. ⟨tel-00192386⟩

Share

Metrics

Record views

1042

Files downloads

2254