Evaluating Intake Estimation Methods for Young Children's Diets

Objectives: This paper illustrates the use of the National Cancer Institute (NCI) Markov Chain Monte Carlo (MCMC) method for usual intake (UI) analyses of 5-year-old children's diets by comparing results from the MCMC method with results from other estimation methods. Methods: This study involves secondary analysis of data from the Infant and Toddler Feeding Practices Study-2 (ITFPS-2), a nationally representative prospective cohort study that followed children from around birth through age 9. Dietary data analyzed were collected between April 2018 and August 2019. All study participants in the longitudinal cohort (n = 1030) had 1 day of dietary recall data, and 122 participants had 2 days of recall. We compare differences in intake distributions for sodium, added sugars, whole grains, energy, and Healthy Eating Index (HEI) scores using the NCI UI methods, as well as single-day and two-day methods. We use regression analysis to assess associations by intake estimation method. Results: Across the methods examined, means for daily consumed nutrients differed by less than 2 percentage points and mean HEI component scores differed by less than half a point. However, for episodically consumed whole grains, the NCI UI methods yielded mean intake estimates that differed by 37%, with the univariate method indicating higher mean intake than the MCMC method. Conclusions: For the daily consumed nutrients examined, the NCI MCMC method is a useful alternative to the univariate method. However, for episodically consumed whole grains, the NCI UI methods yield notably different mean estimates. For episodically consumed dietary constituents, abandoning the NCI univariate method may exacerbate differences between recommended and estimated population mean intake levels for young children.