We conducted a randomized controlled trial with a waitlist control group. Measurements were taken at the beginning of the intervention (t0), at the end of the intervention (t1) and for the intervention group after a 7-month follow-up (t2).
Participants were recruited via local advertisements, local neurologists and the Pain Unit of the Medical Center of the University of Freiburg. Inclusion criteria were as follows: (1) aged 18–65 years, (2) diagnosis of migraine with or without aura by the trial physician in accordance with the diagnostic criteria of the International Headache Society , (3) at least two migraine attacks per month on average, and (4) in case of a medical prophylaxis maintaining a stable dose for at least 3 months prior to inclusion until the end of the trial. Exclusion criteria were as follows: (1) chronic migraine with more than 15 migraine days per month, (2) taking headache analgesics on more than 15 days or migraine-specific triptans on more than 10 days per month, (3) regular practice of meditation (> 1 × per week) or yoga (> 2 × per week), (4) plans to start psychotherapy or any other migraine treatments during the course of the trial, (5) prior participation in a mindfulness training, (6) participation in other clinical studies throughout the study duration, and (7) presence of a life-threatening disease or a mental disorder that might severely hinder interpersonal contacts. Taking acute headache medications like non-steroidal anti-rheumatics (NSAR) and triptans during the trial period was allowed.
The patients were recruited between Nov 2014 and Feb 2015. They underwent initial telephone screening and, if potentially eligible, were invited to the study center at the Medical Center of the University of Freiburg for examination by one of three study physicians and the study psychologist. They were provided with information about the trial and gave written informed consent. Next, the participants were asked to complete a first set of questionnaires and received a headache diary with detailed instructions to be filled in daily for the next 4 weeks (t0). The patients were allocated to one of the two trial arms and were informed in writing about the result of the allocation in the middle of their diary assessment. The patients allocated to the MBCT arm started their course a few weeks later and completed their post-assessment (t1) headache diaries and questionnaires shortly afterwards. Headache diaries and questionnaires were provided in the last MBCT session and were sent back by mail by the participants 4 weeks later. The patients allocated to the waitlist group did not receive any treatment within that period, but were sent the same measurements at the same time (t1) by mail. They received the MBCT intervention after their post-assessment. A 7-month follow-up (t2) completed the assessment for the MBCT group; the participants were sent diaries and questionnaires by mail and were asked to send them back. Based on the waitlist design, the participants could not be blinded with respect to their group allocation. Data entry was performed in a blinded manner. Data analysis was not performed in a blinded manner, as the principal investigator of the study was responsible for patient recruitment, patient guidance, data collection, and data analysis.
The patients were assigned to groups through the design-adaptive allocation approach by Aickin [29,30,31]. This regression-based minimization procedure takes advantage of creating balanced groups of participants with respect to chosen prognostic variables. We used age, gender, and self-reported (baseline) number of migraine days per month as factors. An anonymous list of included patients was sent to Mikel Aickin in Arizona. He performed the group allocation according to this method and generated and returned two balanced lists of participants. Finally, a person not further involved in the trial randomly allocated the group assignment of the two lists.
The Migraine-Adapted MBCT Intervention
The formal structure of our adapted MBCT group intervention closely followed the manual by Day , which in turn closely parallels the MBCT manual for depression relapse from Segal et al. . The intervention consisted of eight weekly 2.5-h sessions. At the start an individual intake interview was held with the MBCT teacher in order to assess personal goals and motivations. Finally, a booster session for refreshment was held after 6 months. Overall, four courses (two courses for the invention group and two delayed courses for the control group) were conducted with an average group size of 12 participants. The courses were held by three experienced and certified MBSR/MBCT teachers from the local mindfulness network. In order to ensure expertise with the disease, the three teachers were provided with a professional education by one of the study physicians. In order to ensure adherence to the MBCT protocol, the principal investigator of the study and the teachers met regularly for discussions during the study interval. In these sessions, the specific adaptations of the manual were discussed in detail.
Regarding program content, the depression-related cognitive-behavioral elements of the original MBCT from Segal et al.  were transformed to headache-specific adaptations by Day and Thorn , which were then further adapted to migraine-specific adaptations by our research group. This encompassed (1) educational elements about the condition of migraine, (2) fostering self-monitoring of the cascades of thoughts, feelings, and bodily reactions, (3) identifying cognitive errors, (4) regulating the level of activity and stress in everyday life, and (5) fostering early recognition and regulation of specific signs of stress and overload. The participants were encouraged to practice at home for 30–45 min a day.
Outcomes were chosen with regard to the recommendations of the IMMPACT (Initiative on Methods, Measurement, and Pain Assessment in Clinical Trials) group . A headache diary assessed headache-related impairment as well as migraine-specific related outcomes. Psychological variables of functioning and coping were assessed via self-report questionnaires. Socio-demographic and medical history features as well as treatment satisfaction, compliance, and degree of goal achievement were assessed with the aid of self-constructed questionnaires.
Primary outcome was the group difference at t1 of the variable “headache-related impairment.” It was assessed via three items asking for impairment in everyday life, at work and during leisure. Items were assessed on an 11-point numeric rating scale (0–10) with the anchors “no interference at all” to the “most severe interference”. These items were taken from the Pain Disability Index [34, 35], and they are also used in the German Pain Questionnaire . They were assessed on a daily basis in a headache diary (see below). Cronbach’s α for the three items in our sample at baseline (t0) was 0.93 (0.91 at t1; 0.93 at t2). Proof of validity has been provided by the authors, including correlations with the degree of behavioral disability in the everyday life, the amount of time spent in bed due to pain or the level of depression. The authors state that the three items can be aggregated to an overall value of pain-related disability. We calculated an overall disability value for every headache day. We then aggregated these values over 1 month to a monthly disability value, which served as our primary outcome.
The participants were provided with headache diaries to be completed daily for 4 weeks at pre- and post-intervention for both groups, and also at 7-month follow-up for the intervention group. It consisted of eight columns assessing headache intensity, headache-related impairment, headache characteristics (duration, pain character, aggravation by exercising, presence of the three attendant symptoms aura, sensitivity to light and noise, nausea and sickness), and medication use. Headache intensity was assessed on a numeric rating scale (0–10) ranging from “no pain at all” to the “most severe pain.” In case of a day with no headache, the patients marked the first column (headache intensity) with an “X” and did not fill in the other columns. In case of a headache day, all eight columns had to be filled in. The patients were instructed to fill in this first column in any case, in order to retrieve the number of headache days per month without any missings or need for imputation. Headache characteristics were assessed with dichotomous items (i.e., a duration more or less than 4 h). For assessing medication use, the exact acute medication should be noted for each day. Medication use is reported as number of days with acute medication per month. Headache frequency reflected the number of days per month with any kind of headache. Both variables were converted to a standard month of 30 days.
The participants filled out a battery of the following questionnaires at all three measurement points. All the standardized questionnaires showed at least satisfactory reliability and validity in clinical trials as well as suitability in pain patient samples.
Hospital Anxiety and Depression Scale
To assess the degree of affective disturbance, the Hospital Anxiety and Depression Scale (HADS-D) was used [37,38,39]. The HADS-D is a well-established, low-threshold screening instrument for assessing anxious and depressive symptoms in patients with somatic complaints. We found in our sample a Cronbach’s α for anxiety of 0.78 (t0), 0.83 (t1), and 0.88 (t2), and for depression of 0.79 (t0), 0.82 (t1), and 0.88 (t2) respectively.
Perceived Stress Questionnaire
The degree of stress in everyday life was assessed with the Perceived Stress Questionnaire (PSQ) [40, 41]. The PSQ measures the amount of subjectively experienced stress, according to inner experiences like having sorrows, feeling tension, experiencing lack of positive emotions, and being confronted with requirements. Cronbach’s α was 0.91 (t0), 0.93 (t1), and 0.96 (t2) in our sample.
Perceived Stress Reactivity Scale
The level of stress reactivity was assessed via the Perceived Stress Reactivity Scale (PSRS) [42, 43]. The PSRS asks for the duration and the extent of a person’s affective reaction in stressful situations as a dispositional trait. We found in our sample a Cronbach’s α of 0.89 (t0), 0.91 (t1), and 0.86 (t2).
Questionnaire of Dysfunctional and Functional Self-Consciousness
The amount of rumination was assessed through the scale “dysfunctional self-attention” of the Questionnaire of Dysfunctional and Functional Self-Consciousness (DFS) . It assesses to what extent people are directing their attention inflexibly towards their inner processes and are persevering in these when confronted with disturbances or obstacles. Cronbach’s α was 0.90 (t0), 0.92 (t1), and 0.92 (t2).
Pain-Related Self Statements Scale
The amount of catastrophic thinking in reaction to pain is assessed with the scale “catastrophizing” of the Pain-Related Self Statements Scale (PRSS) [45, 46]. Statements that indicate higher levels of this mode of thinking reflect higher states of helplessness and hopelessness in the face of pain. Cronbach’s α was 0.84 (t0), 0.87 (t1), and 0.78 (t2) in our sample.
The Self-Compassion Scale (SCS) assesses the degree of self-directed compassion or empathy which expresses a positive attitude of a person towards him/herself even in the face of inadequacies, failures or difficult life circumstances [47, 48]. The overall value of this questionnaire has shown a Cronbach’s α of 0.91 (t0), 0.94 (t1), and 0.95 (t2) in our sample.
Freiburg Mindfulness Inventory
The short-version of the Freiburg Mindfulness Inventory (FMI) targets the amount of self-reported mindfulness by assessing the underlying non-judging, accepting, and non-identifying qualities of the mindfulness construct [49, 50]. We found a Cronbach’s α of 0.78 (t0) 0.88 (t1), and 0.82 (t2) in our sample at baseline.
Socio-demographic Data, Feasibility and Goal Attainment Assessment
Additionally, we administered questionnaires assessing personal data on socio-demographic features and characteristics of the migraine disease at baseline. For the feasibility assessment, we assessed treatment satisfaction and homework adherence via questionnaires including Likert scales at the follow-up-assessment. As a final measure, two personally relevant goals from every patient were assessed in the interview session with the MBCT teacher with the use of a goal attainment assessment . At the end of the course, these sheets were given to the patients again along with the other questionnaires in order to rate their goal attainment on a five-point scale (from −2 “achieved goals much less than expected” to + 2 “achieved goals much more than expected”).
Data Analysis and Statistics
Our study was designed to find a medium to large effect on the primary outcome in a waitlist design. In our pilot study, the variable of migraine-related impairment showed an effect size of d = 0.66. By specifying our intervention to our patient sample, we sought to increase this effect to at least 0.7. To detect an effect of d = 0.7 with a statistical power (1 − β) of 0.80, a sample size of n = 52 would be needed.
For the headache diary, missing data at post- or follow-up assessment were replaced through the last-observation-carried-forward procedure (LOCF). This procedure was done at the level of aggregated mean values for all headache variables (i.e., impairment, intensity, number of headache days, and number of medication days). In the MBCT group, the diaries at baseline from three people were found to provide invalid data due to a large amount of missing data (> 50%). They were not included in the analyses, reducing the data set to n = 24. In the control group, one person provided data with a large portion of missings. Two other patients who left the study did not return their diaries, reducing the diary-based data set to n = 24. For the clinical questionnaires, missing data were first checked for random distribution and were then replaced with the expectation–maximization algorithm . The critical rate of 30% of missing values (i.e., single fields in the diary) for the application of the algorithm was at no time exceeded . As predictor variables for the expectation–maximization algorithm, baseline values as well as the three stratification variables of the design-adaptive allocation were chosen. Here, all 54 patients provided valid data sets and could be included.
All the analyses were conducted with SPSS Statistics 18, and the statistical software R. Differences between groups from baseline to post-measurement were assessed via analysis of covariance (ANCOVA). Baseline scores as well as stratification variables were entered as covariates in a model with group as the independent factor. As partial Eta2 (η2) as a common measure of effect size in analyses of variance has no pre-defined range, Cohen’s d was computed instead. As effect size measure for the ANCOVA, we calculated an adjusted Cohen’s d (dadj) by dividing the difference of the adjusted means through the pooled standard deviation at post-measurement.
Variables with frequency data, i.e., number of headache days and number of medication days, are assumed to be Poisson rather than normally distributed. For them a linear mixed model based on a Poisson distribution with the use of the package lme4 in R was computed. We assessed the time × group interaction. Additionally, the stratification variables age, gender, and self-reported days of migraine at intake were entered into the model as fixed factors, while intercept was entered as a random factor.
Regarding correction for multiple testing of secondary outcomes, we refer to the recommendations of Schulz and Grimes . The authors state that formal corrections like the Bonferroni correction inflate the type-II-error and devaluate information stemming from trials with many endpoints. We followed their recommended procedure to set one primary endpoint that will be prioritized for the assessment of the trial and to only report results that have been set by the study protocol.