(PTFS06-04-23) Multivariate Markov Chain Monte Carlo Method for Evaluating Diet Quality and General Usual Intake Analysis

Objectives: To illustrate the Markov Chain Monte Carlo (MCMC) method for jointly modeling multiple dietary components in the Healthy Eating Index (HEI-2015), and to compare MCMC performance with methods using 1-day intake, average of 2-day intakes, and the other National Cancer Institute (NCI) methods.

Methods: Using the 24-hour dietary recall data from the Special Supplemental Nutrition Program Women, Infants, and Children (WIC) Infant and Toddler Feeding Practices Study-2 (WIC ITFPS-2), methods mentioned were used to estimate selected nutrient intakes for calculating HEI-2015 component scores and total scores. Results were compared in two types of analyses: population distributions of a single nutrient and ratio of intakes, and regression models assessing factors, including WIC participation, associated with HEI-2015 scores.

Results: The MCMC method and the other NCI methods produced comparable distribution estimates (i.e., mean (SE), median (Q1, Q3)) of single nutrient and ratio, overall and by demographic characteristics including WIC participation status. For example, sodium intake (mg/d) estimated from the MCMC method is 2,583 (60), 2,545 (2,201, 2,925), compared to the NCI univariate analysis 2,572 (57), 2,520 (2,135, 2,949). After accounting for children’s total daily calories and sociodemographic characteristics, enrollment in WIC and/or SNAP was not significantly associated with children’s sodium intake at age 5 using the MCMC approach. But the 1-day intake method showed significantly lower sodium intake of children on WIC only, compared to children who were not on WIC or SNAP. Although all methods showed that children who were on WIC consistently for 5 years had higher HEI scores than children who were not, significance of the group differences differed across methods.

Conclusions: The MCMC method provides more accurate estimation of diet quality by appropriately accounting for within-person variation and correlations between diet components—something the other methods considered in this paper either fail to do or do with limitations. In addition to its unique value among methods in fitting person-level regression models with HEI scores, the MCMC also facilitates general univariate and bivariate usual intake analyses that are comparable with the other methods.

Funding Sources: Funded under USDA FNS Contract Number GS-XXX, Order Number XXX