Poster Topical Area: Nutritional Epidemiology
Location: Hall D
Poster Board Number: 770
Objectives: Household consumption and expenditure surveys are frequently conducted around the world and usually include data on household food consumption. The applicability of these data to nutrition research is limited partly by their collection at the household- rather than individual- (dietary-) level. Using a combination of household food consumption and individual dietary intake data from Mongolia, this study's objective was to evaluated the validity of four approaches for estimating diet from household surveys.
Methods: The following four approaches were evaluated: (1) direct inference from per-capita household consumption; disaggregation of household consumption using (2) a statistical method based on a regression approach and (3) the "adult male equivalent" (AME) method based on relative caloric requirements; and (4) direct prediction of dietary intake given the availability of different household- and individual-level variables with which to build a model.
Results: Per-capita household consumption overestimated dietary energy in single- and multi-person households by factors of 2.63 and 1.89, respectively (correlation: 0.09 and 0.29). Performance of disaggregation methods was variable in terms of mean bias (range: +302 to +1088 kcal/day and -918 to +163 kcal/day for AME and statistical methods, respectively, across two household surveys analyzed), while the statistical method exhibited less bias than the AME method in estimating intake densities (per 100 kcal) of most dietary components in both surveys. Increasingly complex prediction models explained 54% to 72% of in-sample variation in dietary energy (mean absolute error: 229 to 178 kcal/day), with consistent marginal benefits to model fit incurred by additional inclusion of basic dietary measurements and eating behaviors.
Conclusions: In Mongolia and elsewhere, differences in how household and dietary measurements are recorded make their comparison challenging. Validity of disaggregation methods depends on household survey characteristics and the dietary components considered. Relatively precise prediction models of dietary intake can be achieved by integrating basic dietary assessment into household surveys, which should be considered for nutrition surveillance in developing countries.