Abstract

Alzheimer’s disease is the most frequent form of dementia in the elderly, and age is its most powerful risk factor. One idea is to model the probability of being diagnosed with dementia at different ages in order to construct trajectories for different categories of people. Mixed membership models constitute the most promising method for this problem. We develop a few ideas of Manrique-Vallier (2010) to extend the basic TGoM model. In particular, we propose a parametric dependence between the distribution of the membership vectors and a few time-invariant covariates that allows us to interpret their effect on the individual trajectories.