The study was a single blind, two arm randomised controlled trial in which participants were randomised to either a "true" diet or "sham" diet control group. Participants were allocated to one of the two diet sheets based on a randomisation schedule developed using a random computer number generator by an independent data manager. Simple randomisation was used with no restrictions (e.g. blocking and/or stratification) placed on the randomisation. All participants and staff at YorkTest Laboratories were blinded to group assignment for the duration of the study.
Participants were recruited through advertisements and a press release. An advert for the trial was placed in the Migraine Action Association newsletter and on the University of York's webpage. Additionally there was a press release to the local media. Participants who expressed an interest in taking part in the study between March and June 2008 were sent further details about the study, a baseline questionnaire and a consent form. Because the participants were recruited through advertisement and through the completion of a postal questionnaire we could not be certain that all of them had a clinical diagnosis of migraine. Consequently, those recruited to the study did have self reported headaches that were 'migraine like'. Whilst most of the participants are likely to have had migraines it is possible that some did not. All participants aged 18-65, who had a self-reported diagnosis of migraine for at least 12 months, had no evidence of any other significant co-existing pathology and experienced 2 or more migraine like attacks (or 4 or more headache days) in the previous 4 week period were eligible for the screening phase. We wanted to recruit participants who were having regular migraines or migraine like headaches in order to maximise the power of the study.
At screening, potentially eligible participants were sent a pin prick test kit, supplied, at no cost, by YorkTest Laboratories Ltd (York, UK), with only a numerical identifier. Participants were asked to follow the instructions of the pin prick test in order to supply a small blood sample. Participants were asked to return this blood sample directly to YorkTest Laboratories. YorkTest Laboratories carried out an ELISA test to detect the presence of IgG antibodies specific to a panel of 113 different food antigens (see Table ). Participants who had one or more food intolerances identified from the ELISA test were eligible for inclusion in the study. The cut-off used to determine food-specific IgG antibodies are detected or not was 10 AU (arbitrary units) per millilitre (AU/ml) of blood. Most of the 113 foods were tested individually. There were some that were tested as mixers (e.g., melon mix includes watermelon, honeydew and Cantaloupe melon). Participants were asked to remove all the foods in that particular mix (if appropriate). The test is CE marked and test reproducibility meets the requirements of the European IVD Directive. Participants who showed no evidence of food sensitivity in the ELISA test were excluded from randomisation and were informed of their true test results.
Foodscan 113 - the foods tested
Treatment and control group interventions
For each participant two diet sheets were generated by YorkTest Laboratories - a true diet sheet and a sham diet sheet. The sham diet sheet asked participants to eliminate the same number of foods to which they exhibited IgG antibodies but not those particular foods. The diet sheets were matched for the number of foods to be eliminated and also for the difficultly of eliminating foods. YorkTest Laboratories sent the true and sham diet sheets for each participant to the York Trials Unit, again with only a number for identification.
Participants were sent the appropriate diet sheet and were advised to remove foods to which they showed a positive reaction from their diet for a period of 12 weeks. At the end of this phase of the study, both groups were asked to reintroduce, in a stepwise fashion, the foods which they had been asked to eliminate. Participants were asked to reintroduce one food at a time on a weekly basis and continue with the food if no migraine or migraine like headache occurred. Food reintroduction occurred over a 4 week period with all eliminated foods (if no migraine or migraine like headache had occurred) being introduced in the final week. Participants were advised to stop consuming any reintroduced food if a migraine or migraine like headache occurred and to continue to reintroduce one of the other remaining foods. At the end of the follow up period all participants were told which diet sheet they had been given. Participants were sent the appropriate diet sheet to commence diet elimination in September 2008.
The baseline questionnaire recorded information about participants experience with migraine, including: frequency of migraines or migraine like headache in the previous 4 weeks (number of headache days), symptoms experienced, migraine medication use, whether they have previously excluded foods from their diet, and any consultations with healthcare professionals about their migraine or migraine like headache. The questionnaire also included a measure of disability using the MIgraine Disability Assessment Scale (MIDAS) [24
] and impact on daily life using the Headache Impact Test (HIT-6™) [25
]. Baseline questionnaires were sent out to participants at the beginning of July 2008. Reminders were sent 3 weeks later if participants had not returned their original questionnaire. Participants were followed-up at 4 weeks (October 2008) and 12 weeks (December 2008) after starting their allocated diet with a short questionnaire which included the MIDAS, HIT-6 and an additional question asking if the participant had consulted with a GP about their migraine or migraine like headache over the past 4 or 8 weeks depending on the timing.
The MIDAS questionnaire assesses headache-related disability. Respondents are asked to answer five questions, scoring the number of days, in the past 3 months, that they have had activity limitations due to headaches. Two additional items (A and B) are also included which ask about the number of headaches over the past 3 months and how painful the headaches were. The Headache Impact Test (HIT) is a tool to measure the impact headaches have on a person's ability to function on the job, at home, at school and in social situations. Respondents are asked to answer six questions covering aspects of functioning mostly impacted by headache: pain, role functioning (the ability to carry out usual activities), social functioning, energy or fatigue, cognition, and emotional distress. The responses on each question are described on a 5-point Likert scale including: never, rarely, sometimes, very often and always.
Participants in both groups completed a daily diary for the 12 weeks they were in the food elimination phase of the trial. The diary recorded whether or not they had a migraine or migraine like headache, the duration of the migraine or migraine like headache, the treatments they used and a migraine/headache severity score (0 to 10 where 0 is no pain and 10 is pain as bad as it can be). Participants in the "true" diet group were also asked to record their adherence to the diet each day in the diary. Participants were asked the following question: "Have you followed the study-related dietary advice today?". If participants answered "yes" to the question 70% of the time they were categorised as "adhering to the diet most of the time."
In a population survey of migraine sufferers it was found that on average patients had 13 migraines or headache days over a 12 week period (our follow-up period) with a standard deviation of 11 [24
]. We sought a reduction of 5 headache days. Assuming a standard deviation of 11 this results in a standardised effect size (difference in means/standard deviation) of 0.45. This reduction of a 0.45 of a standard deviation was similar to that observed in a recent acupuncture trial for migraine prevention [2
]. To detect this effect size, assuming 80% power, an independent samples t-test and a 2-sided 5% significance level we required 78 participants per group, 156 in total. Allowing for 10% loss to follow up we required a total of 174 participants (87 per group).
The primary outcome measure was the number of headache days over a 12 week period (item A on the MIDAS questionnaire) which was calculated by summing responses on the 4 and 12 week questionnaires. A headache day was defined as any 24 hour period in which the patient reported that they had had a migraine attack which lasted more than 4 hours. Secondary outcome measures were the total MIDAS score and total HIT-6 score. The total MIDAS score was calculated by summing the total number of days from questions 1 to 5 (ignoring A and B). The HIT-6 scores were calculated by assigning a value of 6 to a response of "Never," 8 to "Rarely," 10 to "Sometimes," 11 to "Very Often," and 13 to "Always;" these values were then summed. Scores can range from 36 (lowest possible score) to 78 (highest possible score).
All analyses were conducted on an intention to treat bases, including all randomised patients in the groups to which they were randomised. Analyses were conducted in SAS (version 9.1) and SPSS (version 15). We compared the number of migraines in the intervention and control groups using negative binomial regression models with adjustments for baseline scores. These models are used to estimate the number of occurrences of an event when the event has Poisson variation with over-dispersion. We also used a negative binomial regression model to compare the total MIDAS scores in the intervention and control groups with similar adjustments for baseline scores. A linear regression model was used to compare the total HIT6 scores between the intervention and control groups after adjustment for baseline scores. Stricter levels of significance were used (p = 0.01) for secondary outcomes to compensate for multiple testing. Multiple imputation was used to account for the missing data. Five imputations were created using a set of appropriate imputation models constructed using variables that were predictive of the missing data (e.g. age, baseline number of headache days) and the group allocation. Multiple imputation was performed using the mi procedure in SAS with the assumption that the data were missing at random.