A three-part, mixed-effects model to estimate the habitual total vitamin D intake distribution from food and dietary supplements in Dutch young children.

Statistical modeling of habitual micronutrient intake from food and dietary supplements using short-term measurements is hampered by heterogeneous variances and multimodality. Summing short-term intakes from food and dietary supplements prior to simple correction for within-person variation (first add then shrink) may produce estimates of habitual total micronutrient intake so badly biased as to be smaller than estimates of habitual intake from food sources only. A 3-part model using a first shrink then add approach is proposed to estimate the habitual micronutrient intake from food among nonsupplement users, food among supplement users, and supplements. The population distribution of habitual total micronutrient intake is estimated by combining these 3 habitual intake distributions, accounting for possible interdependence between Eq. 2 and 3. The new model is an extension of a model developed by the USA National Cancer Institute. Habitual total vitamin D intake among young children was estimated using the proposed model and data from the Dutch food consumption survey (n = 1279). The model always produced habitual total intakes similar to or higher than habitual intakes from food sources only and also preserved the multimodal shape of the observed total vitamin D intake distribution. This proposed method incorporates several sources of covariate information that should provide more precise estimates of the habitual total intake distribution and the proportion of the population with intakes below/above cutpoint values. The proposed methodology could be useful for other complex situations, e.g. where high concentrations of micronutrients appear in episodically consumed foods.