A0532
Title: Tree-regularized Bayesian latent class analysis: Addressing weak separation in small-sized subpopulations
Authors: Zhenke Wu - University of Michigan at Ann Arbor (United States) [presenting]
Abstract: Latent class models (LCMs) have been used to derive dietary patterns, where class profiles represent a probability vector of exposures to a set of diet components queried on diet assessment tools. However, LCM-derived dietary patterns can exhibit strong similarities, or weak separation, resulting in unstable class profile estimates and less accurate class assignments. This issue is exacerbated in small-sized subpopulations. This issue is addressed with a newly proposed tree-regularized Bayesian LCM that shares statistical strength across dietary patterns. These patterns are guided by an unknown tree learned from the data to produce improved estimates of class profiles and assignments using limited data. This is achieved via a Dirichlet diffusion tree process that specifies a prior distribution for the unknown tree over classes. Dietary patterns that share proximity in the tree are shrunk towards ancestral dietary patterns a priori, with the degree of shrinkage varying across pre-specified food groups. Using dietary intake data from the Hispanic Community Health Study/Study of Latinos, the utility of our model is demonstrated to identify dietary patterns of US adults of South American ethnic backgrounds.
