19 December 2002

Dear Editor,

I write concerning the reporting of a controlled trial of cumulative behavioural effects of a common bread preservative by Dengate and Ruben in the August 2003 issue of the Journal1.

My organization is quoted in the Introduction as stating: According to the Australian Food and Grocery Council, ‘conventional scientific wisdom worldwide has not established a link between behavioural problems in children and calcium propionate used as a food preservative’.

It is our contention that this study does not alter that conclusion. The authors do not present sufficient data to permit a complete analysis of their results.

While accepting the difficulties inherent in this type of study, the fact that there was a ‘placebo’ effect on both treatment (some improved) and placebo (some worsened) suggests that their reported effect is an artefact of the statistics used and lack of power in the study design. The title of the paper should more properly have been ‘Lack of effect of a common bread preservative on behaviour in children responding positively to an elimination diet’. Our detailed critique follows.

The authors acknowledge the lack of power in their study when they state that ‘due to four placebo responders, there was no significant difference by anova of weighted placebo and challenge behaviours’. Yet they then go on to state that there was a ‘statistically significant difference between the proportion of children who worsened on challenge (52%) compared with those that improved on challenge (19%) relative to placebo’.

Data are not provided on the responses to placebo, apart from mention of the four who worsened on placebo. This information is necessary to determine, in a cross-over design of this nature in which the authors acknowledge that the effect is ‘cumulative’, if those on intervention first, remained ‘worse’ on placebo; and second, compared with those on placebo first. With random assignment, presumably 13/14 started on placebo and 14/13 on intervention. It is essential to have individual scores reported by order of treatment to determine any ‘order’ effect before amalgamating into an intervention minus placebo grouping.

Design issues: Data were collected via the Rowe Behaviour Rating Inventory from both teachers and parents on the Friday covering the three previous school days ‘including Thursday night’. The data from teachers were not analysed due to ‘incomplete collection’. This means behaviour was rated on the school days of the study for the school-aged children, only for the period preschool and after school. I do not believe that this limited parent view is sufficient to get a reliable 3-day behaviour score.

The parental rating for that portion of the study population who did not attend school, given the age range was 4−12 years, would be expected to be quite different.

Three of the children were also medicated. Were their responses different from that of the unmedicated group?

Was a washout period established between one arm and the next? The paper is silent on this. Failure to have a washout period for a ‘cumulative effect’ agent would compound the errors in the study.

We would welcome the authors’ comments on these matters.


  1. Top of page
  2. Reference