Using Latent Variables with Longitudinal Data to Identify Traits with Common Underlying Disease Processes

• Activity Number: 653
• Type: Contributed
• Date/Time: Thursday, August 13, 2015 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Jesse Raffa*
• Keywords: multivariate longitudinal response ; latent variables ; hidden disease process

Abstract

Longitudinal data is frequently collected on multiple responses in order to capture features of an underlying disease or other process. Recently, attempts have been made to model these responses as generated from the hidden disease process for both univariate and bivariate longitudinal responses using latent variable or hidden Markov mixed models. In the multivariate response setting, inference can be difficult when responses are generated from different underlying diseases processes. We propose a method for identifying longitudinal responses with common hidden disease processes. We illustrate the approach with a dataset of >100 common clinical traits with a large degree of between-subject heterogeneity from a study where subjects were counseled on how to improve their personal health. Such an approach could be used to identify traits with common disease processes or which are easily or inexpensively measured to monitor health status. Examples using real and simulated data will be discussed.