Can current dietary exposure models handle aggregated intake from different foods? A simulation study for the case of two foods

Current dietary exposure models provide estimates of long-term intake distributions using short-term food consumption survey data, by statistically modeling the aggregated intakes from different foods consumed on the same day for each participant of the survey. Food consumption behaviour in a population may, however, show all sorts of correlations which are not modelled in these exposure models. We developed a simulation model describing a hypothetical population of consumers, assuming various types of correlation between two foods. Using this simulation model we found that the impact of the correlations in many cases is limited, but in particular circumstances it can be substantial, depending on the properties of the marginal distributions. Further, we found that the usual approach of first aggregating the observed intakes over foods, and then applying the statistical exposure models to the total daily intakes may lead to deviating results, even when all correlations are assumed to be zero. The approach of analyzing the intakes from the separate foods, and then aggregating the results from the statistical model applied to each food performed much better. Our results illustrate that the simulation model can be used for validating dietary exposure models, and for indicating how exposure models may be improved.