Following a request from the European Commission, the Panel on Food Additives, Flavourings, Processing Aids and Food Contact Materials (AFC) was asked to assess the results of a recent study on the effect of mixtures of additives on children’s behaviour and provide an opinion on the findings, taking into account, if possible, other available scientific literature in the related area.
A recent study by McCann et al. (2007) has concluded that exposure to two mixtures of 4 synthetic colours plus a sodium benzoate preservative in the diet result in increased hyperactivity in 3-year old and 8- to 9-year old children in the general population. In an earlier study by the same research team there was some evidence for adverse behavioural effects of a mixture of 4 synthetic colours and sodium benzoate in 3-year old children on the Isle of Wight (Bateman et al,. 2004).
In this recent study the effects of two combinations of Tartrazine (E102), Quinoline Yellow (E104), Sunset Yellow FCF (E110), Ponceau 4R (E124), Allura Red AC (E129), Carmoisine (E122) and sodium benzoate (E211) on children’s behaviour were studied. Five of the six food colours belong to the class of synthetic azo dyes and one, Quinoline Yellow (E104), is a quinophthalone. Sodium benzoate is used as a preservative.
The study involved one hundred and fifty three 3-year old and one hundred and forty four 8- to 9-year old children, selected to represent a broad range of behaviour in the general population including children with normal to high level behavioural activity. Children who were medicated for ADHD were not included. A global hyperactivity aggregate (GHA) score was the main outcome of the study, and this parameter was based on aggregated z-scores of observed behaviours and ratings by teachers, class room observers and parents, plus, for 8- to 9- year old children, a computerised test of attention.
Mix A containing Tartrazine (E102), Ponceau 4R (E124), Sunset Yellow FCF (E110), Carmoisine (E122) and sodium benzoate significantly increased GHA scores for all 3-year old children compared to the placebo control GHA scores (effect size 0.20 [CI 0.01 to 0.39], p<0.05).
Mix B containing Sunset Yellow FCF (E110), Carmoisine (E122), Quinoline Yellow (E104), Allura Red AC (E129) and sodium benzoate had no effect on GHA scores in 3-year old children as compared to the placebo control GHA scores (effect size 0.17 [CI -0.03 to 0.36]).
This result persisted when analysis was restricted to 3-year old children who consumed more than 85% of juice and had no missing data (complete case group); in this analysis the effect of Mix A in the 3-year old children was still significantly increased compared to placebo control (effect size 0.32 [CI 0.05 to 0.60, p<0.05) but for Mix B no significant effect on GHA scores was observed (effect size 0.21 [CI -0.06 to 0.48]).
For the 8- to 9- year old children a significant effect of Mix A (effect size 0.12 [CI 0.02 to 0.23], p<0.05) or Mix B (effect size 0.17 [CI 0.07 to 0.28], p<0.01) was seen when analysis was restricted to those children consuming at least 85% of drinks with no missing data (complete case group). When all 8- to 9- year old children that completed the study were taken into account, Mix A had no effect on the GHA scores compared to the placebo control (effect size 0.08 [CI -0.02 to 0.17]) and Mix B had a significant effect on GHA scores (effect size 0.12 [CI 0.03 to 0.22] p<0.05).
The authors concluded that exposure to synthetic colours or a sodium benzoate preservative (or both) in the diet result in increased hyperactivity in 3-year old and 8- to 9-year old children in the general population.
Based on surveys conducted from 2002 to 2005, the target colours are more frequently used in sweets but also occur commonly in soft drinks and benzoate is frequently present in soft drinks. Children consuming brightly coloured sweets may be exposed to levels comparable to those considered in the protocol of the McCann et al. study for one or more of the food colours studied. Comparable levels may also be reached in those children who consume brightly coloured soft drinks. The level of exposure to sodium benzoate is also likely to occur.
The Panel considers that the steps taken for score normalisation and aggregation are mathematical transformations that might affect the assumptions of normality and independence of the data which are essential for the whole statistical analysis. Therefore, the authors’ primary analysis was repeated using a more justifiable and conventional statistical model, and this was supplemented by a set of additional analyses with the aim of aiding the interpretation of the results.
The Panel considers the re-analysis undertaken by EFSA, in which all single variables (minus the individual baseline value for that variable) were considered without normalisation, so that each subject served as its own reference, as the most adequate. This re-analysis was undertaken both at the level of the individual parameters as well as on the aggregated scores.
Based on the results obtained it was concluded that the analysis with the recalculated GHA score led to broadly similar conclusions to that in the original paper by McCann et al, except for the following:
(1) The Mix A versus placebo comparison was not statistically significant for the 3-year olds when all subjects were included (entire sample), while the significance for the > 85% consumption and complete case groups was increased slightly;
(2) For the 8- to 9- year age group, the Mix A versus placebo comparison was no longer statistically significant in any of the three consumption groups.
In addition the data were analysed on the basis of a modified GHA score in which the parental scores were not included. The results from this analysis no longer revealed any statistically significant effects of Mix A or Mix B versus placebo, except for Mix B versus placebo in 8- to 9-year old completers.
A further analysis was carried out on the whole data set, comprising analysis of the single variables of parental scores, teacher scores and observer scores, and, in the case of 8- to 9-year old children, computer-based scores. There is a suggestion from these analyses that the statistically significant effects seen in the 3-year olds (Mix A versus placebo) and in the 8-to 9- year olds (Mix B versus placebo) are largely driven in the data by the parental scores and, in the older males in both comparisons, by the computer score.
The Panel notes that some, but not all, earlier studies have also reported effects of food colours on child behaviour, the majority of these studies being conducted on children described as hyperactive or with a clinical diagnosis of ADHD.
The Panel concludes that the McCann et al. study provides limited evidence that the two different mixtures of synthetic colours and sodium benzoate tested had a small and statistically significant effect on activity and attention in some children selected from the general population, although the effects were not observed for all children in all age groups and were not consistent for the two mixtures. The findings may thus be relevant for specific individuals within the population, showing sensitivity to food additives in general or to food colours in particular.
However, it is not possible to assess the overall prevalence of such sensitivity in the general population and reliable data on sensitivity to individual additives are not available.
The clinical significance of the observed effects also remains unclear, since it is not known whether these small alterations in attention and activity would interfere with schoolwork and other intellectual functioning. The clinical significance could possibly be clarified by assessments that used scales for functional impairment and diagnostic interviews, especially if a high proportion of children with high symptom scores were to be included in such a study.
There are thus a number of uncertainties that are apparent from this new research, some of which are echoed in earlier research. These include:
- the limited consistency of the results with respect to age and gender of the children, the effects of the two mixtures of additives tested and the type of observer (parent, teacher or independent observer);
- the unknown clinical relevance of the novel metric, i.e. the GHA score;
- the unknown relevance of the small effect size (as was also seen in the meta analysis of earlier studies by Schab and Trinh, (2004));
- the fact that the study has not been designed to identify the effects of individual additives;
- a lack of information on dose-response;
- the lack of a biologically plausible mechanism for induction of behavioural effects from consumption of food additives.
The Panel concludes that the McCann et al. study provides limited evidence that the two different mixtures of synthetic colours and sodium benzoate tested had a small and statistically significant effect on activity and attention in children selected from the general population excluding children medicated for ADHD, although the effects were not statistically significant for the two mixtures in both age groups.
Since mixtures and not individual additives were tested in the study by McCann et al., it is not possible to ascribe the observed effects to any of the individual compounds.
The clinical significance of the observed effects also remains unclear.
In the context of the overall weight of evidence and in view of the considerable uncertainties, such as the lack of consistency and relative weakness of the effect and the absence of information on the clinical significance of the behavioural changes observed, the Panel concludes that the findings of the study cannot be used as a basis for altering the ADI of the respective food colours or sodium benzoate.