The study was designed as an open-label randomized controlled trial with two stages. The first stage (pre-post) comprised three conditions (miCBT, RLX, WLC), and the second stage (4- and 12-month follow-up) comprised two conditions (miCBT, RLX) since the pre-post completers of the WLC were randomly assigned to the miCBT- or RLX-condition. The study was conducted at the psychotherapy outpatient clinic of the Department of Psychology (University of Mainz, Germany). The study protocol was approved by the Ethics Committee of the State Chamber of Medicine in Rhineland-Palatinate, Germany, reference number 837.291.16 (10610).

Participants were recruited via flyers, social media, several local newspaper articles, and one TV clip. Inclusion criteria were (i) meeting the ICHD-3 beta criteria (24) of either migraine without aura, migraine with aura, or chronic migraine for at least one year; (ii) a minimum of four headache days per month, and a pattern of migraine symptoms stable over the last six months; (iii) psychological factors, such as dysfunctional thoughts (e.g., overambitious achievement orientation), emotions (e.g., attack-related fear), and behavior (e.g., excessive avoidance of triggers), or the experience of emotional distress, were associated with migraine [meeting the DSM-5 criteria (25) of either “somatic symptom disorder” or “psychological factors affecting other medical conditions”]; (iv) fluency in German, Internet access; (v) age of at least 18 years. Exclusion criteria were (i) diagnosis of medication-overuse headache; (ii) currently taking a headache prophylactic medication (3-month wash-out) or therapy with botulinum toxin or neuromodulation during the trial period; (iii) previous completed or current psychotherapy; (iv) a severe mental disorder or medical comorbidity (which was likely to interfere with the ability to participate in group therapy, e.g., an acute psychosis, a major depressive episode, or an advanced Parkinson's disease); (v) suicidal tendency; and (vi) pregnancy or lactating.

Pre-screening was performed by phone followed by a face-to-face screening session. The face-to-face session included a structured interview to validate the migraine- and DSM-5-diagnosis. When checking possible DSM-5 diagnoses, the individual criteria of the disorder were queried in each case if the answer to the corresponding screening question was positive. All participants had to give written informed consent. Since all screenings were conducted by Master's level clinical psychologists, the participants had to provide a medical certificate stating the migraine diagnosis.

Eligible participants were randomly assigned with a 1:1:1 ratio to miCBT, RLX, or the WLC using block-randomization. In a second stage, the completers of the WLC were randomly assigned to miCBT or RLX (1:1 ratio). Randomization was performed by an independent statistician, using a computer-generated random sequence. Participants were then informed by the principal investigator via email about their allocation (i.e., whether they were receiving a behavioral intervention or had been placed on the WLC). The exact type of intervention (miCBT or RLX) was disclosed in the first treatment session since the treatment programs provided that the participants were given an agenda, and since participants were assumed to realize the type of intervention in the course of the therapy anyway. Since all outcomes were self-reported data, blinding of outcome assessment was not provided.

Both treatments (miCBT, RLX) comprised seven, weekly group sessions, each lasting 90 min. The treatments were conducted by five Master's degree clinical psychologists. To avoid bias, each therapist provided a similar amount of both miCBT- and RLX-treatments. Therapists were trained in a 1-day workshop before conducting the treatment and received regular supervision at least two times per treatment cycle by the first author. Both interventions are described in detail in a treatment manual. Since a participation rate of at least five sessions was aimed for in both treatments, missed group sessions were made up in extra group or one-to-one sessions.

The miCBT included a mixture of several behavioral approaches (Figure 1) (23). In session 1 (Education about Migraine) participants were informed about the disease (e.g., symptoms, different phases of a migraine attack) and the different factors, which are assumed to contribute to the emergence and maintenance of migraine or migraine attacks. A biopsychosocial etiology model of migraine, integrating pathophysiological findings (e.g., central neuronal hyperexcitability as an underlying vulnerability), psychological factors (e.g., dealing with triggers, especially stressors), and a threshold for migraine attacks was provided. Participants were encouraged to reflect on their own life situation and own coping styles regarding the etiology model. In session 2 (Lifestyle Counseling), basic recommendations to practice a healthy, migraine-compatible way of life (e.g., relaxation, recreation, and physical activity on a regular basis) were presented. The participants were motivated to formulate their own goals with regard to an even more balanced lifestyle. Session 3 (Coping with Fear of Migraine Attacks) contained behavioral techniques to analyze and modify fear-related cognitions. Thus, the participants were informed about the vicious circle of anxiety and how to cope with attack-related fear. In Session 4 (Coping with an Ongoing Migraine Attack) various behavioral options in managing an acute attack were discussed, with the assumption, that improved coping skills lead to decreased fear of attacks. In doing so, the four different issues (i) medication intake behavior, (ii) activity level, (iii) communication of attack-related complaints to other persons, and (iv) automatic thoughts were addressed. Above all, the participants were encouraged to reduce their activity level during an ongoing attack as far as possible. Subsequently, the participants were asked to reflect on their own typical behavior patterns in an ongoing attack and to formulate appropriate change goals in this regard. Session 5 (Trigger Management) provided five different strategies to manage triggers: (i) experiment, (ii) avoid, (iii) coping with stress, (iv) exposure, and (v) acceptance. The participants were asked to select suitable strategies for their own five most relevant triggers and to apply them in the subsequent. Session 6 (Stress Management) was focused on cognitive-behavioral techniques to improve coping with stress since emotional distress is regarded as the most important trigger in migraine attacks (6). Thus, the model of behavioral analysis was applied for the assessment of an individual, prototypical stressful situation (26). Subsequently, each participant was encouraged to consider a suitable coping strategy for this situation in the future. In Session 7 (Closing) the participants were encouraged to reflect on what they have learned and to formulate long-term behavioral goals for coping with their migraine. Each session (except for the first session) started with a reflection on the previous session (including feedback about successful or failed behavioral changes). Subsequently (except for the last session), the new issue (e.g., stress management) was introduced (including a referred behavioral analysis) and an individual formulation of behavioral change goals was encouraged. At the end of each session, the participants were taught a different, brief relaxation method.

The relaxation training (RLX) was aimed at teaching progressive muscle relaxation (PMR) and transferring it to everyday life. PMR is a systematic relaxation technique with Grade A in migraine prophylaxis (9). The RLX was developed based on the manual by Bernstein and Borkovec (27). In session 1, an education about migraine and PMR was provided. Further, the long PMR-version (16 muscle groups) was conducted. In session 2, the long PMR-version was conducted again, and a rationale for regular, independent practice was given. It was recommended to perform one relaxation exercise (lasting about 20 min) per day (15). In the following sessions 3–7, variations of PMR (version with 7 or 4 muscle groups, envision exercise, conditioned and differential relaxation) were taught. The participants were repeatedly encouraged to regularly perform PMR on their own. To support transfer into everyday life, participants were given logs to document their weekly exercise practice. Keeping the logs was not mandatory, but was strongly recommended. From the third session onward, the everyday life exercise practice was reported by each participant at the beginning of the session. Difficulties in implementation were addressed and appropriate ways to establish a regular exercise practice were developed.

All outcomes were assessed at baseline (pre-treatment), post-treatment, 4-month follow-up, and 12-month follow-up. Each assessment included a 4-week online diary and a one-time online survey (about 60 min). The online diaries were completed by recording hourly ratings of headache intensity using a 4-point rating scale from 0 (no headache) to 3 (severe). Additionally, the amount of daily pain medication had to be entered. Participants were instructed to enter data at least once per day. The frequency of data entry was monitored using specific software. If participants did not enter data for more than three days, they were reminded via a personal email or phone call. The online survey was conducted via SoSci-Survey (28) and contained a set of questionnaires on headache-related factors (e.g., disability). After the successful completion of the pre-measurement, the treatment (miCBT or RLX) or the waiting period (WLC) started. Immediately after the last treatment session (miCBT or RLX) or after seven weeks (WLC), the post-assessment was carried out. The WLC-participants were randomized to one of the treatments after the successful completion of the post-assessment. Follow-up assessments were conducted 4 and 12 months after the successful completion of the treatment. At the 12-month follow-up, participants were additionally asked, whether relaxation had continued to be practiced and to what extent.

After each treatment session, participants filled out the Group Therapy Session Evaluation by Patients (GTS-P) (29) as a paper-pencil questionnaire. The GTS-P comprises 8 items regarding the therapeutic process, including one item addressing treatment expectancy. Since there are no standardized methods for assessing the adverse effects of behavioral therapy given to adults with migraine, a free response field was added to each paper-pencil questionnaire. In doing so, the participants were given the opportunity to report adverse events as well as to make suggestions for changes to the sessions.

All therapy sessions (in total 112, excluding catch-up sessions) were videotaped. In each cycle, 2 sessions were randomly selected for evaluation (in total 32 sessions). The rating was done each by two Master's level clinical psychologists. Raters received training provided by the first author on how to apply the evaluation forms or scales. Therapists' adherence was rated on a 10-item evaluation form with a 3-point rating scale, with the analogous rating possibilities 0 = not adherent, 1 = partly adherent, and 2 = adherent. The evaluation forms were largely similar for both conditions. Both evaluation forms were developed following the Cognitive-Behavioral Maintenance Adherence Scale (30). Therapists' competence was assessed each with the German Version of the Cognitive Therapy Scale (adapted for group therapy, CTS-D-G) (31). The CTS-D-G comprises 18 items with each a 7-point rating scale from analogous 0 = poor to 6 = excellent.

Four primary outcomes were defined: (i) headache days/28 days, (ii) headache-related disability, measured with the Headache Disability Inventory (HDI), (32) (iii) emotional distress, measured with the Depression Anxiety Stress Scales (DASS), (33) and (iv) headache-specific self-efficacy, measured with the German short form of the Headache Management Self-efficacy Scale (HMSE-G-SF) (34). The HDI assesses the impact of headaches on daily living. It includes 25 items and has a very good internal consistency (Cronbach's α = 0.90) (32). The DASS comprises three scales (Depression, DASS-D, Anxiety, DASS-A, and Stress, DASS-S). Each scale includes seven items (short version). The three scales have satisfying to good internal consistency (Cronbach's α = 0.78 to 0.92) (33). The HMSE-G-SF provides an assessment of headache-specific self-efficacy beliefs. It includes six items and has a satisfying internal consistency (Cronbach's α = 0.72) (34).

Secondary outcomes were (v) headache index/28 days, (vi) medication days, defined as the number of days of using headache medication for a 28-day period, (vii) sensitivity to and avoidance of headache triggers, measured with the Headache Triggers Sensitivity and Avoidance Questionnaire (HTSAQ), (35) (viii) headache-related disability, measured with the Pain Disability Index (PDI), (36) (ix) headache impact, measured with the Headache Impact Test (HIT-6), (37) and (x) acceptance of chronic pain, measured with the Chronic Pain Acceptance Questionnaire (CPAQ) (38). The HTSAQ requires an assessment of 26 potential triggers (24 common triggers and 2 triggers not listed before) using four scales: (i) Triggers, which refers to whether the named trigger is a trigger for the respondent's headaches, (ii) Sensitivity to triggers compared with others, S (O), which refers to how sensitive the respondent is to the trigger compared with other persons, (iii) Sensitivity to triggers compared with time of least sensitivity, S (T), which refers to how sensitive the respondent is to the trigger compared with the time of least sensitivity, and (iv) Avoidance, which refers to how hard the respondent tries to avoid the trigger. All four scales have a good internal consistency (Cronbach's α = 0.83 to 0.86) (39). The PDI includes seven items, each referring to disability in a specific area of life (e.g., social activity). Since the PDI assesses pain-related disability in general, the instruction was adapted toward headache-related disability. Its internal consistency is good (Cronbach's α = 0.85 to 0.90) (36). The HIT-6 includes six items and has a high internal consistency (Cronbach's α = 0.83) (40). The CPAQ includes 20 items and showed good internal consistency in a sample of persons with headache (Cronbach's α = 0.84) (39). All questionnaires were applied in a German Version.

A headache day was defined as “a day with headache pain that lasts ≥ 4 h with a peak severity of moderate or severe intensity, or of any severity or duration if the subject takes and responds to a triptan or ergot” (p. 490) or another headache medication (41). Since clinical experience shows that persons with migraine cannot always clearly assign existing headaches to migraine headache, it was decided to record headache days instead of migraine days. The headache index was determined by averaging the headache intensity ratings across the total day (24 h) and calculating a mean over the 28 days. Since the frequency of headache days per 28 days is more comprehensible than the headache index, this measure was defined as the primary outcome. To be analyzed, participants had to complete at least 80% of the diary entries (Figure 2) (42).

An a priori power analysis was conducted using G^*Power (43). Based on previous evidence a moderate effect of miCBT vs. WLC was assumed (9). Thus, with Cohen's f = 0.25 at a power of 1 – β = 0.95 with an α = 0.05, a necessary sample size of at least 90 participants was calculated. Considering a dropout rate of 25%, the target sample size was increased to 120 participants.

Analyses were based on the intention-to-treat (ITT) sample, including all participants that were allocated to one of the three conditions and provided baseline data. To compare the change in outcomes from pre- to post-assessment between groups, mixed-model analyses for repeated measures (MMRMs) were conducted. Analyses included group (miCBT, RLX, or WLC), time (pre- and post-assessment), and the group × time interaction as fixed factors, as well as a random intercept (subjects) to model interindividual differences, with diagonal covariance matrices and using restricted maximum likelihood estimation. To analyze the second stage of this trial (long-term effects of miCBT vs. RLX), MMRMs were conducted with all participants that were allocated to one of the two treatments and provided baseline data (Figure 2). MMRMs included group (miCBT or RLX), time (pre, 4-month follow-up, 12-month follow-up), and the group × time interaction as fixed factors, as well as a random intercept to model interindividual differences, based on an auto-regressive structure with heterogenous variances (ARH1). Additionally, a sensitivity analysis using MMRMs was performed with complete cases (participants that completed ≥5 treatment sessions or WLC and provided data both at baseline and post-assessment). Since the analysis of covariance (ANCOVA) can be considered as robust method with high statistical power, (44) a further sensitivity analysis using this approach was performed. In view of the tightly calculated sample size, assuming medium-sized effects, a more permissive composite analysis was carried out post-hoc. The participants, which had received a behavioral treatment (either miCBT or RLX) between pre- and post-assessment were combined into one condition (“behavioral therapy”, BT) and compared with the participants of the WLC. Accordingly, an ANCOVA was performed with the outcomes' pre-values as a covariate and the outcomes' post-values as the dependent variable.

Within-group effect sizes were calculated as Cohen's d_av (45) (pre to 12-month follow-up) based on observed values. Responder analyses were performed for both follow-up assessments (4-month and 12-month follow-up). For the symptom-related outcomes (headache days, headache index, medication days), a reduction of ≥30% was defined as a response. In contrast to a more conservative ≥50% response operationalization, this threshold seems realistic to detect minor changes that can represent a veritable improvement for patients as well and can be considered more suitable for non-pharmacological interventions (46). For the psychometric questionnaires (e.g., HDI), the Reliable Change Index (RCI) (47) was calculated with a CI of 95% for observed values. For the 12-month follow-up, differences between the two treatment groups in terms of frequency and extent of relaxation practice were calculated using Fisher's exact test or a t-test. All analyses were performed with a two-tailed α of 0.05 in SPSS statistics version 23. A two-sided α was chosen to cover potential inferiority of the miCBT. This corresponds to the conservative approach commonly used in the scientific community (48, 49). We did not adjust the level of significance for multiple testing of primary outcomes to be able to control the type II error rate, to account for the pilot-character of this study in testing a newly developed miCBT, and since we compared distinct treatments with a WL control (50).

Inclusion occurred between 9 January 2017 and 9 April 2019. In total, 243 persons were pre-screened by phone, which led to a face-to-face screening in 129 potential participants (Figure 2). Of these, 121 participants were randomly assigned to miCBT (40), RLX (41), or WLC (40), and 106 participants entered the intervention phase with a pre-measure (ITT; Table 1). Ninety-seven participants (91.5%) completed the post-measurement, and dropout rates were comparable between groups. Twenty-nine participants of the WLC thereafter were randomized to either miCBT (15) or RLX (14).

The treatment sessions were conducted from 5 May 2017 to 18 July 2019. Each of the two programs (miCBT and RLX) was carried out eight times. Treatment sessions were conducted in groups of four to nine participants. The average session completion rate of the 91 treatment-completers (Figure 2) was high (miCBT: 93%, M = 6.53; SD = 0.59; range 5 to 7 sessions; RLX: 91%, M = 6.37; SD = 0.71; range 5–7 sessions).

Therapists' adherence was very high in both treatment conditions (miCBT: M = 1.80; SD = 0.52; RLX: M = 1.83; SD = 0.48). The inter-rater reliability was very high for each of the two evaluation forms, ICC (2,2) = 0.957, p < 0.001 (miCBT), ICC (2,2) = 0.980, p < 0.001 (RLX). Therapists' competence was also rated very high in both treatment conditions (miCBT: M = 5.24; SD = 0.98; RLX: M = 5.19; SD = 1.02). The inter-rater reliability was moderate for the CTS-D-G, ICC (2,2) = 0.571, p < 0.001. Thus, the treatment integrity was given in that a very high therapists' adherence as well as a very high therapists' competence (both assessments done with a statistically significant inter-rater reliability) were observed.

ITT-analyses of the primary outcomes yielded significantly stronger pre-post improvements in self-efficacy (assessed by the HMSE-G-SF) in both treatments (miCBT: p = 0.027; RLX: p = 0.032) compared to the WLC, whereas no significant between-group differences were found for headache days, disability, and emotional distress (Table 2). The between-group comparisons for the secondary outcomes did not yield any statistically significant differences (Table 2, all p-values > 0.05). Sensitivity analyses with complete cases (N = 97) showed a similar pattern, except for a significant difference at post-assessment between RLX and WLC for the secondary outcome headache index in the ANCOVA (Supplementary Tables 1, 2; observed values in Supplementary Table 3). The post-hoc composite analysis by ANCOVA with complete cases yielded a significantly stronger pre-post improvement in self-efficacy as well as a significantly higher reduction in the headache index in the treatment condition (BT) compared to the WLC (Supplementary Table 4).

Analyses of long-term effects (miCBT vs. RLX; Figure 3; Supplementary Tables 5, 6) did not reveal any significant group differences between the two treatments for changes from baseline to the 4- and 12-month follow-up for primary and secondary outcomes (all p-values > 0.05). However, the within-group change from baseline to 12-month follow-up showed significant improvements in all primary and most of the secondary outcomes. Here, the frequency of headache days decreased by 1.75 (miCBT) or 2.60 (RLX) days from baseline to 12-month follow-up (each based on a 28-day period, difference based on estimated marginal means, Figure 3, Supplementary Table 5). The within-group effect sizes (Cohen's d_av) of the pre to 12-month follow-up changes ranged from 0.04 (HTSAQ, Scale B) to 0.80 (CPAQ-D; Supplementary Table 7).

After 12-months, a large proportion of the participants reported to continue practicing some form of relaxation (81% miCBT; 78% RLX; p = 1), whereas the participants of the RLX-group reported a higher extent of weekly relaxation praxis in minutes, which however did not differ significantly (CBT: M = 40.4; SD = 43.6; RLX: M = 59.9; SD = 69.4; p = 0.151).

The responder analyses yielded a similar pattern with no statistically significant difference between the two treatments in the frequency of reliable change (both at 4-month and 12-month follow-up; Supplementary Table 8). For the symptom-related outcomes (headache days, headache index, and medication days), the response rate was higher at 12-month follow-up compared to 4-month follow-up (in both treatments). The miCBT, as well as the RLX, showed a ≥30% reduction of headache days for about 44% of the participants at 12-month follow-up.

Nine participants (miCBT: four; RLX: five) reported a total of 13 adverse events on the session evaluation form. All reported events were classified as temporary and not serious. Five participants (miCBT: four; RLX: one) discontinued the treatment. None of them reported adverse events on the evaluation form. Four of them stated, that the treatment did not meet their expectations. In both treatments, the session evaluation by the GTS-P showed consistently positive assessments (all means >3; Table 3). However, the RLX treatment was rated significantly better in some items and in total (Table 3).