Validation and sensitivity analysis of probabilistic models of dietary exposure to micronutrients: an example based on vitamin B6

Vitamin D concentrations in fortified foods and dietary supplements intended for infants: Implications for vitamin D intake

Due to potential overages to cover losses during shelf life, the actual vitamin D concentration of fortified foods and dietary supplements may deviate from the label. In this pilot study the vitamin D concentrations of fortified foods (n=29; follow-on formula, baby porridge, curd cheese dessert) and dietary supplements (n=15), both specifically intended for infants, were analytically determined. Compared to the declared values, the vitamin D content ranged from 50% to 153% for fortified foods and from 8% to 177% for supplements. In general, both instant follow-on formula and oil-based supplements had a measured vitamin D content similar to or higher than the labelled value. Ready-to-eat baby porridge was the only category in which all measured vitamin D concentrations were below the declared value (74-81%). The use of label information for fortified foods and dietary supplements may result in invalid estimations of vitamin D intake distributions of infants; both under- and overestimation may occur.

Approaches to assessment of exposure to food- and supplement-derived amino acids

Although the amino acid composition of almost all food proteins is known, estimating the amino acid intake from the diet is extremely difficult because of the lack of available data. A conservative approach would be to determine the population distribution of protein intake, select the 97.5(th) or higher percentile of intake, assume all comes from the target protein, and estimate exposure to some specific amino acid. Any number of dietary survey methodologies could be used to conduct such a conservative approach. However, given the great variety of brands of food supplements, estimates of amino acid intakes from supplements are problematic. Firstly, few studies include supplements in their target nutrient sources because brand-level data would need to be retained and nutritional composition data would need to be recorded. Probabilistic modeling offers some solution provided some basic data are gathered. The percentage of the population regularly taking supplements and the frequency of consumption must be known. Therefore, data on the dietary supplement market would need to be known including the percent of brands containing amino acids and if possible specific amino acids together with concentrations. A probabilistic model as follows would ensue: probability of being a consumer of amino acid supplements; probability distribution function of frequency of use of supplements; probability distribution function of dose per eating occasion; market characteristics; probability distribution function for dietary amino acid intake. Using multiple iterations and perhaps bootstrapping on some elements of the model, fully worst-case model scenarios of exposure could be computed.

Introduction to the Monte Carlo project and the approach to the validation of probabilistic models of dietary exposure to selected food chemicals

The Monte Carlo project was established to allow an international collaborative effort to define conceptual models for food chemical and nutrient exposure, to define and validate the software code to govern these models, to provide new or reconstructed databases for validation studies, and to use the new software code to complete validation modelling. Models were considered valid when they provided exposure estimates (e(a)) that could be shown not to underestimate the true exposure (e(b)), but at the same time are more realistic than the currently used conservative estimates (e(c)). Thus, validation required e(b) </= e(a)<e(c). In the case of pesticides, validation involved the collection of duplicate diets from 500 infants for pesticide analysis. In the case of intense sweeteners, a new consumption dataset was created among prescreened high consumers of intense sweeteners by recording, at brand level, all foods and beverages ingested over 12 days. In the case of nutrients and additives, existing databases were modified to minimize uncertainty over the model parameters. In most instances, it was possible to generate probabilistic models that fulfilled the validation criteria.

Validation and sensitivity analysis of probabilistic models of dietary exposure to micronutrients: an example based on vitamin B6

Probabilistic modelling can be used to get an insight into the variability and uncertainty of the nutrient intake in a population. When a probabilistic model is used, it is important that it is validated. Furthermore, a sensitivity analysis of the model output can give an insight into the most important input variables of the model and can be used as an aid to describe the reliability of the model. In this study, four models to estimate vitamin B(6) intake among males and females were validated using the method of Kaaks et al. This method compares the relationship between three different kind of measurements with the unknown 'true' intake. In each of these four models, only one input variable (concentration or bioavailability) was changed compared with a reference model. A sensitivity analysis was also performed. The results of the validation showed that for males, a model using a fixed bioavailability factor at the food group level was valid, while for females a model using either a fixed value or a distribution for the bioavailability factor was valid. Use of a distribution for the concentration of vitamin B(6) in supplements was not valid. The results of the sensitivity analysis showed that the concentration of vitamin B(6) in food and supplements was the key contributor to variability and uncertainty in the model estimates of vitamin B(6) intake, in both males and females. All results indicated that when taking variability and uncertainty into account by using probabilistic modelling, the effect on the nutrient intake for nutrients that are present in many common eaten foods, is small. For these broadly available nutrients, fixed concentrations and bioavailability factors give a good estimate of the nutrient intake in a population. When using probabilistic modelling, it is very important to collect more actual information about the concentration.