Does Food-Grouping Make a Difference When Deriving Dietary Patterns Using Latent Class Models? A Monte Carlo Simulation Study

• Activity Number: 257
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Daniela Sotres-Alvarez*+ and Amy H. Herring and Anna Maria Siega-Riz
• Companies: The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill and The University of North Carolina at Chapel Hill
• Keywords: latent class models ; dietary patterns ; Monte Carlo simulation

Abstract

Dietary patterns (DP) are used to study the effect of overall diet. First, dietary data (50 to 250 food items) are collapsed into food groups to reduce dimension; then DP are derived using multivariate methods. Collapsing limits the DP's interpretation and can attenuate their association with health outcomes. We explore with a simulation study the effect of food grouping when deriving DP using latent class models. We compared the effect of sample size on the number of DP (chosen by several information criterion and Lo-Mendell-Rubin likelihood ratio test (LMR-LRT)), performance of the estimates, and subjects' classification. BIC performed better than AIC, but slightly worse than LMR-LRT except when using food-groups and sample size was small. Power for LMR-LRT was high both when using food-items and food-groups, but type I error was large preferring models with at least one more class.