Advanced Dietary Patterns Analysis Using Sparse Latent Factor Models in Young Adults.

Area of Science:

• Nutritional Epidemiology
• Statistical Modeling

Background:

• Principal Component Analysis (PCA) is a common method for dietary pattern analysis but has limitations.
• PCA requires arbitrary decisions for food variable selection and does not easily incorporate covariates.
• Sparse latent factor models offer an alternative approach to address these limitations.

Purpose of the Study:

• To compare Bayesian sparse latent factor models with PCA for identifying dietary patterns in young adults.
• To evaluate the interpretability and covariate incorporation of sparse latent factor models versus PCA.

Main Methods:

• Habitual food intake data from 2730 young adults (TIGER Study) were analyzed.
• A food-frequency questionnaire assessed intake of 102 food items.
• Bayesian sparse latent factor models and PCA were applied to derive dietary patterns, with covariates included in the sparse model.

Main Results:

• Both methods identified seven dietary patterns.
• Sparse latent factor analysis incorporated covariates and provided probabilistic criteria for food relevance.
• Sparse latent factor analysis yielded more interpretable patterns with fewer excluded food items compared to PCA.

Conclusions:

• Sparse latent factor models are a valuable tool for future dietary pattern studies.
• These models reduce arbitrariness in food variable selection and facilitate covariate incorporation.
• Sparse latent factor models enhance the assessment and interpretation of dietary patterns.