In this work, we propose a semi-latent class random changepoint mixed model that allows the estimation of the time of differentiation between cognitive decline of future demented and normal subjects from a nested-casecontrol study. Cases are assumed to have a random changepoint trajectory while controls can have either a linear trajectory or a random changepoint trajectory where the class membership follows a logistic model. The log-likelihood of the model is derived and can be optimized using a Levenberg-Marquardt algorithm with Gaussian quadrature for numerical integration. The model is estimated on the Paquid cohort of elderly with very long follow-up (25 years) to estimate the delay between the beginning of the decline of a test of verbal uency and the onset of dementia.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados