Title: A Bayesian flexible joint model of multivariate longitudinal and survival data
Authors: Kai Kang - The Chinese University of Hong Kong (Hong Kong) [presenting]
Xinyuan Song - Chinese University of Hong Kong (Hong Kong)
Abstract: Joint models for analyzing longitudinal and survival data are widely used to investigate the relationship between a failure time process and time-variant predictors. A common assumption in conventional joint models in the survival analysis literature is that all predictors are observable. However, this assumption may not always be supported because unobservable traits, namely, latent variables, which are indirectly observable and should be measured through multiple observed variables, are commonly encountered in the medical, behavioral, and financial research settings. We propose a novel joint modeling approach to deal with this feature. The proposed model comprises three parts. The first part is a dynamic factor analysis model for characterizing latent variables through multiple observed indicators over time. The second part is a random coefficient trajectory model for describing the individual trajectories of latent variables. The third part is a proportional hazard model for examining the effects of time-invariant predictors and the longitudinal trajectories of time-variant latent risk factors on hazards of interest. We develop a Bayesian approach coupled with a Markov chain Monte Carlo algorithm to perform statistical inference. We conduct simulation studies to assess the empirical performance of the developed methodology. An application of the proposed joint model to a study on the Alzheimer's Disease Neuroimaging Initiative is presented.