Estimating the Distribution of Dietary Consumption Patterns.

Area of Science:

• Nutrition Science
• Biostatistics
• Survey Methodology

Background:

• Dietary intake data in the US often relies on 24-hour recalls, which have significant measurement error for assessing usual long-term intake.
• Accurate estimation of usual dietary intake is crucial for public health, but current methods are limited.
• The Healthy Eating Index-2005 (HEI-2005) is a key measure of diet quality, but its population distribution is difficult to estimate accurately.

Purpose of the Study:

• To estimate the population distribution of the Healthy Eating Index-2005 (HEI-2005) among US children aged 2-8.
• To address the considerable measurement error inherent in dietary intake assessment.
• To incorporate survey weights for a nationally representative analysis.

Main Methods:

• Development of a highly nonlinear, multivariate zero-inflated data model with measurement error.
• Application of a Bayesian approach utilizing Markov Chain Monte Carlo (MCMC) for computational feasibility.
• Utilizing the relationship between Bayesian posterior means and maximum likelihood estimation.
• Employing balanced repeated replication, a survey-sampling technique, for standard error estimation.

Main Results:

• The Bayesian MCMC approach successfully resolved computational challenges associated with the complex data model.
• A realistic population distribution estimate for the HEI-2005 total score among children was achieved.
• The study demonstrated a viable method for analyzing complex dietary quality indices with measurement error.

Conclusions:

• Bayesian methods, particularly MCMC, are effective for estimating complex dietary indices like the HEI-2005 when standard statistical software fails.
• This approach allows for more accurate population-level assessments of diet quality in children.
• The findings provide a more reliable understanding of children's dietary patterns and quality in the US.