Assessment of food intake input distributions for use in probabilistic exposure assessments of food additives

Abstract

A key component of a food chemical exposure assessment using probabilistic analysis is the selection of the most appropriate input distribution to represent exposure variables. The study explored the type of parametric distribution that could be used to model variability in food consumption data likely to be included in a probabilistic exposure assessment of food additives. The goodness-of-fit of a range of continuous distributions to observed data of 22 food categories expressed as average daily intakes among consumers from the North–South Ireland Food Consumption Survey was assessed using the BestFit® distribution fitting program. The lognormal distribution was most commonly accepted as a plausible parametric distribution to represent food consumption data when food intakes were expressed as absolute intakes (16/22 foods) and as intakes per kg body weight (18/22 foods). Results from goodness-of-fit tests were accompanied by lognormal probability plots for a number of food categories. The influence on food additive intake of using a lognormal distribution to model food consumption input data was assessed by comparing modelled intake estimates with observed intakes. Results from the present study advise some level of caution about the use of a lognormal distribution as a mode of input for food consumption data in probabilistic food additive exposure assessments and the results highlight the need for further research in this area.

Keywords:

• food consumption
• distribution
• probabilistic analysis
• food additive

Notes

^a If the AD statistic > critical value, distribution is rejected. Critical value = 2.49 (95% confidence level).

^b Mean and SD intakes (g kg⁻¹ bw day⁻¹) among consumers only, calculated from the observed data.

^c LognormBestfit: parameters = mean and SD of the log-normal distribution that best fit the observed data, estimated in BestFit®; Parameters = mean and SD of the natural logarithms of the observed data, estimated in BestFit®.