|
Abstract:
|
In many studies, dimension reduction methods are used to profile participant characteristics. For example, nutrition epidemiologists might use clustering or factor analysis to illustrate dietary consumption patterns of food groups or nutrients. One challenge using these approaches is being able to consider the generalizability and reproducibility within different comprising subgroups of the study population. One method, robust profile clustering, has been effectively applied to food frequency questionnaire data. Even so, more flexible approaches are necessary to accommodate the increased granularity and sparsity reported in a 24-hour dietary recall. In this paper, we discuss clustering techniques that are able to integrate the properties of local partitioning and sparse finite mixture modelling to identify dietary intake patterns within distinct subgroups, as well as, ascertain shared patterns across subgroups. These new methods are applied to population-based multi-site studies to empirically derive dietary patterns in large, diverse populations, such as the United States, in order to measure risk and determine dietetic influences on cardiovascular disease.
|
