|
Abstract:
|
Dietary quality data pose a number of changes including truncation of ratios of correlated observations, measurement errors, semi-continuity, and inherent structural clustering among food variables. Further challenges arise when dietary data are collected longitudinally as predictors of future health outcomes. In this talk we will address the necessity of properly dealing with semicontinuity and inherent clusters concerning dietary quality measures, using a dietary intervention study as an example. Specifically, we propose and investigate a two-part multivariate Bernoulli-Gaussian mixture models approach for clustering multivariate semicontinuous variables, and demonstrate its effectiveness in identifying a dietary pattern with which type 1 diabetic children benefit more from a dietary intervention than those with other dietary pattern.
|
