| Food Composition Data: A User's Perspective (1987) |
|Consideration of food composition variability: What is the variance of the estimate of one-day intakes? Implications for setting priorities|
Some implications for data analyses
In assessment of the distribution of observed intakes, as in estimating the prevalence of inadequate or excessive intakes, the focus of attention should be on the estimation of the distribution of usual intakes. This is achieved by obtaining replicate one-day intakes. In the past these have been averaged for the individual to improve the estimate of his or her intake (eliminate the effect of day-to-day variation). A better approach involves application of an analysis of variance to estimate the distribution of usual intakes (the inter-individual variation component) . Either way, a part of the random variation in food composition or in intake estimation will be factored out along with the removal of day-to-day variation. Thus, it can be demonstrated that the presence of variance affects the estimation of prevalence of inadequate intakes or of excessive intakes. However, the effect is surprisingly small! Improvement of the food composition data base by increasing replications will improve the confidence of the prevalence estimate, but the cost-effectiveness for this purpose needs to be examined very carefully.
Another application of computed nutrient intakes is in connection with epidemiologic studiesas in regression or correlation analyses of the relationship between observed intake and some biological outcome. It has been demonstrated by several authors that an error term in the estimation of usual intake can attenuate correlations and will bias regression slopes toward 0 if intake is used as the independent variable.
Table 4. Potential error associated with estimated one-day intakes
|Mean||SD||CV (%)||Mean||SD||CV (%)|
|Vitamin A||3,798.4||281.24||7.40||5,142.0||603 61||11.74|
Table 5. Impact of the number of food items in a record on the error term of computed nutrient intake for an individuala
CV of nutrient content of individual food serving
a. Table assumes that all foods make an equal contribution to nutrient intake.
A major concern in this regard is the day-to-day variation in intake [2, 3, 4, 5, 8], which can have a CV in the order of 25 to 35 per cent or even higher for some nutrients [2,3]. It is clear that food composition variation also contributes to the true total variability (it would be in addition to estimated intra-individual or within-person variation based on published food composition data). Clearly, improvement of the food composition data would improve correlation or regression analyses; it might not be very cost-effective. If we must continue to estimate food intakes on several days, and then average these to estimate "used intake," we again find that the random component of food composition variation will be reduced. The phenomenon is illustrated in table 9.
Table 6. Estimate of error term in one-day intakes associated with combined variability of food composition and error of the intake estimate (assumed measurement error CV = 10 per cent of reported intake, normally distributed)
|Mean||SD||CV (%)||Mean||SD||CV ( %)|
a For composition variability, see table 5
Table 7. Impact of random error in intake and food composition data on the error of calculated nutrient content of an individual serving of fooda
a. All values expressed as CV = 100 x SD/Mean It is not important which variable is I or 2.