Latent class models and latent transition models for dietary pattern analysis Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 21, 2019
Creator
  • Sotres-Alvarez, Daniela Taryn
    • Affiliation: Gillings School of Global Public Health, Department of Biostatistics
Abstract
  • Dietary patterns (DP) are used to study the effects of overall diet on health outcomes as opposed to the effects of individual nutrients or foods. DP are empirically derived mostly using factor and cluster analysis. Latent class models (LCM) have been shown empirically to be more appropriate to derive DP than cluster analysis, but they have not been compared yet to those derived by factor analysis. We derive DP using LCM and factor analysis on food-items, test how well the resulting classes are characterized by the factor scores, and compare subjects' direct classification from LCM versus two a posteriori classifications from factor scores: one possible classification using tertiles and a two-step classification using LCM on previously derived factor scores. In order to study changes in dietary patterns over time, we propose using latent transition models to study change as characterized by the movement between discrete dietary patterns. Latent transition models directly classify subjects into mutually exclusive DP at each time point and allow predictors for class membership and for probabilities of changing classes over time. There are several challenges particular to DP analysis: a large ([greater than or equal to]80) number of food-items, non-standard mixture distributions (continuous with a mass point at zero for non-consumption), and typical assumptions (conditional independence given the class and time point, time-invariant conditional responses, and invariant transition probabilities) may not be realistic. We compare performance, capabilities and flexibility between two software packages (Mplus and a user's derived procedure in SAS) that allow fitting latent transition models. A key decision involved when deriving DP is whether or not to collapse the primary dietary data into a smaller number of items called food groups. Advantages for collapsing include dimension reduction and decreasing the number of non-consumers to reduce the mass-point at zero. However, not collapsing helps our understanding of which combinations of specific foods are consumed. Further, food-grouping may have an impact on the association between DP and health outcomes. We explore via a Monte Carlo simulation study whether food-grouping makes a difference when deriving DP using LCM. Methods are illustrated using data from the Pregnancy, Infection and Nutrition (PIN) Study.
Date of publication
DOI
Resource type
Rights statement
  • In Copyright
Note
  • "... in partial fulfillment of the requirements for the degree of Doctor of Public Health in the Department of Biostatistics, Gillings School of Global Public Health."
Advisor
  • Herring, Amy
Degree granting institution
  • University of North Carolina at Chapel Hill
Language
Publisher
Place of publication
  • Chapel Hill, NC
Access
  • Open access
Parents:

This work has no parents.

Items