Validation analysis of probabilistic models of dietary exposure to food additives

Abstract

The validity of a range of simple conceptual models designed specifically for the estimation of food additive intakes using probabilistic analysis was assessed. Modelled intake estimates that fell below traditional conservative point estimates of intake and above 'true' additive intakes (calculated from a reference database at brand level) were considered to be in a valid region. Models were developed for 10 food additives by combining food intake data, the probability of an additive being present in a food group and additive concentration data. Food intake and additive concentration data were entered as raw data or as a lognormal distribution, and the probability of an additive being present was entered based on the per cent brands or the per cent eating occasions within a food group that contained an additive. Since the three model components assumed two possible modes of input, the validity of eight (2(3)) model combinations was assessed. All model inputs were derived from the reference database. An iterative approach was employed in which the validity of individual model components was assessed first, followed by validation of full conceptual models. While the distribution of intake estimates from models fell below conservative intakes, which assume that the additive is present at maximum permitted levels (MPLs) in all foods in which it is permitted, intake estimates were not consistently above 'true' intakes. These analyses indicate the need for more complex models for the estimation of food additive intakes using probabilistic analysis. Such models should incorporate information on market share and/or brand loyalty.

Development of databases for use in validation studies of probabilistic models of dietary exposure to food chemicals and nutrients

“…Procedures useful to calculate the best intake estimate of food additives, pesticide residues and nutrients were developed within the Monte Carlo project in order to validate probabilistic modelling (Arcella et al 2003, Gilsenan et al 2003, Lo´pez et al 2003, Rubingh et al 2003). In the present paper, these best estimates were defined as the 'true' intakes.…”

“…In Italy (Arcella et al 2003) an ad hoc food survey on adolescents called INRAN-RM-2001 was carried out, whereas a re-elaboration of data previously collected was performed in Ireland and the UK (Gilsenan et al 2003).…”

“…The first condition is quite easy to confirm by using one of the conservative methods currently in use. Within the Monte Carlo project different approaches were adopted to obtain conservative estimates, for food additives individual consumption data were combined with MPLs by Arcella et al (2003) and the Step 2 method was used by Gilsenan et al (2003), for pesticide residues Boon et al (2003) used the IESTI method whereas a method which combines the Step 2 and IESTI approaches was used in the validation analyses conducted by Lopez et al (2003).…”

“…To determine the presence of the additives in brands, an existing food ingredient database in Ireland (Gilsenan et al 2003) and a newly constructed food ingredient database in the UK were used. Both databases were based on information obtained from ingredient labels of brand foods.…”

“…The electronic databases of the North South Ireland Food Consumption Survey (NSIFCS) and of the National Diet and Nutrition Survey of Children Aged 1½ to 4½ years (NDNS) in the UK were recoded at brand level (Gilsenan et al 2003). This involved first identifying all eating occasions of foods in the two databases in which the selected food additives are permitted according to the EU food additive Directives.…”