A hierarchical Bayesian mixture model for repeated dietary records.

Area of Science:

• Nutritional epidemiology
• Statistical modeling

Background:

• Dietary consumption data often exhibits high skewness and numerous zeros.
• Standard transformations (e.g., logarithmic) can fail to accurately model these distributions.
• Accurate modeling is crucial for estimating population consumption and identifying those exceeding recommended limits.

Purpose of the Study:

• To introduce and apply finite mixture models for analyzing complex dietary consumption data.
• To extend finite mixture models to handle hierarchical data structures from repeated dietary records.
• To compare the performance of mixture models against traditional methods for estimating consumption distributions.

Main Methods:

• Utilized finite mixture models, a flexible statistical approach for skewed and multi-modal data.
• Extended mixture models to accommodate hierarchical data from repeated measurements over time.
• Applied a Bayesian approach to a finite mixture model extension.
• Incorporated covariates like sex and age into the mixture models.

Main Results:

• Finite mixture models provided superior estimates of probability distributions for daily dietary consumption.
• The approach yielded improved estimates for maximum consumption over a 7-day period.
• The models successfully handled skewed and zero-inflated dietary data, outperforming alternative methods.

Conclusions:

• Finite mixture models offer a robust and flexible alternative for analyzing complex dietary consumption patterns.
• This methodology enhances the accuracy of estimating population-level dietary intake and identifying at-risk individuals.
• The Bayesian extension of finite mixture models is effective for repeated dietary consumption data, accounting for individual variability and covariates.