Chapter 2 Using Two-Day Food Consumption Survey Data for Longitudinal Dietary Exposure Analyses

摘要

There is no entirely suitable substitute for a robust data base with daily food consumption diaries for thousands of individuals maintained over many weeks, months, and years. In fact, we have little more than two-day food intake records for about 20,000 individuals from the CSFII. But this short-term food consumption data can be used to simulate longer-term consumption trends for many commonly eaten foods. The two simulation methods examined in this paper are complimentary, and neither alone has the ability to predict the entire distribution of food consumption frequencies. At the high end of the frequency distribution (i.e., generally above the 95th percentile), the two-day record repeat method better simulates the directly measured consumption frequencies reported in the NHANES survey. The cohort record sharing method, performed poorly at the high end of the food consumption distribution. However, in the mid-range of the food consumption distribution (generally between the 40th and 70th percentiles) the cohort record sharing method provided more reliable estimates of food consumption frequencies relative to the two-day repeat method. Between the 70th and 95th percentiles, the actual frequencies of food consumption, and by extension the most realistic expectation for exposure values, are most likely bounded by the two estimates provided by these two simulation methods. Therefore, both of these approaches are useful. The risk assessor must select the most appropriate method for the intended purpose.