Joint Bayesian Modeling of Irregularly Measured Multivariate Longitudinal Nutrient Consumption and Growth Data

• Activity Number: 162
• Type: Topic Contributed
• Date/Time: Monday, August 1, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Sherry Lin*+ and Robert E. Weiss
• Companies: University of California at Los Angeles and University of California at Los Angeles
• Keywords: nutrient patterns ; missing data ; forward-backward recursion ; MCMC ; dynamic model

Abstract

We have irregularly measured longitudinal nutrient intake data and asynchronously measured longitudinal growth data. The nutrient consumption data is high dimensional and exhibits multivariate correlation across different nutrients and serial correlation across time. A subject's recent dietary intake, along with his or her intake history, can both be associated with growth. We propose a joint model which uses a small number of dynamic latent factors to model the observed nutritional intake covariates and to predict the growth outcome. The latent factors are interpreted as underlying dietary intake patterns which evolve according to a continuous-time autoregressive process. The model simultaneously estimates the correlation structures of the growth and nutrient data. We formulate the model under a Bayesian framework, using a forward-backward algorithm to estimate the latent factors. The model is applied to nutrition data from a Kenya school intervention study used to analyze muscle development of children under different nutritional snack regimens.