An Analysis of Four Proposed Measures for Estimating Distributions of Total Usual Vitamin D Intake Among Adults Using National Health and Nutrition Examination Survey Data

Abstract

Abstract Objectives To estimate distributions of total usual vitamin D intakes from foods and supplements among U.S. adults, using four different methods of assessment. Usual intake is defined as long-term average daily intake. Methods Distributions of total usual vitamin D intakes were estimated from data provided by 9954 adults (≥19 y) from the 2011–2014 NHANES. Dietary and dietary supplement (DS) data were collected using two 24-hour recalls (24HR), and DS data were also collected via an in-home inventory that ascertained participants’ DS use in the past 30 days. The four methods used included: combined “add then shrink” (Method 1); combined “shrink then add” (Method 2); stratified “shrink then add” (Method 3); three-part (Method 4). Briefly, Method 1 analyzes DS users and nonusers together and includes DS use as a covariate in the model. Nutrient intake from foods and DS are “added” prior to applying usual intake procedures to “shrink” the total intake distribution. Method 2 combines DS users and nonusers identically, but “shrinks” the nutrient intake distribution from foods prior to incorporating nutrient intake from DS to produce total usual intakes. Method 3 also follows the “shrink then add” framework but stratifies DS users and nonusers separately in the model. Method 4 also stratifies DS users and nonusers but computes total usual intakes in three parts: 1) usual intake from foods for DS nonusers; 2) usual intake from foods for DS users; and 3) usual intake from DS for DS users. Results All methods following the “shrink then add” framework (i.e., methods 2- 4) resulted in similar estimated distributions of total usual vitamin D intakes; but Method 1, the “add then shrink” method, resulted in distributions that were wider as a result of inflated within-person variation. Among the “shrink then add” methods, the main differences existed between methods 2 and 3 vs. method 4, partly due to differences in modeling of DS intake information reported on 24HRs vs. the in-home inventory. Conclusions Currently, no consensus exists on the best method for incorporating DS nutrients into usual intake models designed to estimate intake from foods and beverages. These findings support the notion that the “shrink then add” methods are more amenable to modeling when compared to “add then shrink.” Funding Sources This work was supported by a grant from the National Institutes of Health.