Are Psychological Interventions Efficacious for Adults With Migraine?

A Systematic Review and Meta-Analysis

Joanne Dudeney PhD; Louise Sharpe PhD; Sarah McDonald PhD; Rachel E. Menzies PhD; Brian McGuire PhD


Headache. 2022;62(4):405-419. 

In This Article



This review extends and updates the 2019 Cochrane Review.[21] We updated our search, meta-analyzed all studies that provided data allowing for computation of an overall effect size (e.g., mean change, mean and confidence intervals, effect size, F-value, etc.), only meta-analyzed outcomes in which k > 4 and n > 100, and to enable sensitivity analyses we included studies combining data from patients with migraine and TTH. The review was otherwise conducted using Cochrane systematic review methods, and according to the Cochrane Review protocol (i.e., search strategy, selection criteria, study outcomes, risk of bias methods remained unchanged).[22]

Search Strategy

The article search for the Cochrane Review was conducted from inception to July 2018. We updated the search, using the databases CENTRAL, MEDLINE, Embase, PsycINFO, and CINAHL from July 2018 to July 31, 2021, and searched trial registries and reference lists of prior reviews and included studies. See supporting information for a search strategy example.

Two authors (J.D., L.S.) independently shortlisted titles and abstracts for potential inclusion. Next, three authors (J.D., R.M., S.M.) independently assessed full-text articles to determine whether the article met eligibility criteria; agreement rates among the three authors were 95%, with consensus through discussion. We contacted authors for clarification regarding inclusion criteria where required. Finally, we linked multiple articles on the same trial for the purposes of data extraction. Figure 1 presents the Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram of the study selection process. A list of included and excluded studies can be seen in supporting information.

Figure 1.

Preferred Reporting Items for Systematic Reviews and Meta-Analyses flow diagram of study search selection process. RCT, randomized controlled trial [Color figure can be viewed at]

Selection Criteria

Studies. We included RCTs with ≥15 participants in the treatment and control arms post-treatment, and that were published (or pre-published) in English or with English translation as full-text articles in peer-reviewed scientific journals.

Interventions. We included RCTs designed to test the efficacy of a psychological treatment on migraine-related outcomes. RCTs were required to have at least one active (psychological treatment) and one comparison (control) arm. We defined psychological treatment as any intervention aimed at reducing the psychological processes that maintain pain, and associated disability and distress. These interventions could include behavioral skills (e.g., biofeedback, relaxation), cognitive strategies (e.g., psychoeducation, eye movement desensitization and reprocessing, hypnosis), or a mixture of both (e.g., cognitive behavioral therapy [CBT], acceptance and commitment therapy, mindfulness-based cognitive therapy). The comparison arm could be active (e.g., placebo, another psychological intervention) or inactive (e.g., waitlist, standard care). We excluded studies of treatment equivalence.

Participants. We included adults with episodic or chronic migraine. If participants had TTH, they were included in the sensitivity analyses only if the outcome data were combined with patients with migraine. Consistent with other migraine reviews,[17,23,24] a verified diagnosis was not required for inclusion. We excluded studies of secondary headache disorders due to acute or progressive neurological conditions, if migraine was not the primary pain complaint, or studies of medication overuse headache.

Study Outcomes

Outcomes were based on recommendations proposed by the International Headache Society Clinical Trials Standing Committee[25] and the guidelines for behavioral treatments of recurrent headache.[26] The primary outcome was a reduction in migraine frequency, operationalized as (1) the number of days with migraine (or number of migraine attacks) from a self-reported headache diary, and (2) the proportion of participants for whom there was a reduction of 50% or greater in migraine frequency after treatment (i.e., treatment responders). Secondary outcomes included pain intensity, mood, pain-related disability, quality of life (QoL), medication usage, and adverse events.

Data Extraction

We extracted data on number of participants, demographics (e.g., age, diagnosis), treatment details (e.g., modality, dose, delivery), control group, therapist, and outcome measures at post-treatment and follow-up. In addition, for this review we extracted data on number of participants with TTH and type of TTH, where relevant. One review author (J.D.) extracted 100% of the data from included studies into Microsoft Excel, with three authors (L.S., R.M., S.M.) independently extracting data from approximately 30% studies each, with 100% agreement.

Risk of Bias

For the studies included in both the Cochrane and current reviews (k = 20), we included the risk of bias outcomes as they were published.[21] For the studies unique to this review (k = 19), three authors (J.D., R.M., S.M.) assessed the risk of a bias in an identical manner using the Cochrane risk of bias tool.[27] Studies were assessed on the following risk of bias items: random sequence generation and allocation concealment, blinding of outcome assessors, attrition, and reporting outcomes consistent with trial registry. We additionally assessed for bias in treatment integrity (clinician training and treatment fidelity) and sample size.

Quality of the Evidence

For each post-treatment outcome included in the analysis, we assessed the quality of the evidence provided, using GRADE criteria, following the guidelines set out by Cochrane.[28] The outcomes were rated on the following criteria: risk of bias, inconsistency, indirectness, imprecision, and publication bias (see the Cochrane guidelines[28] for a comprehensive summary). We used GRADEpro software to conduct the ratings.[29]

Statistical Approach

For all outcomes, apart from treatment responders, we calculated Cohen's d effect size statistics (difference between the means of two groups, divided by the pooled standard deviation). A positive d indicates a beneficial effect of receiving the psychological intervention, compared to receiving the control intervention. The magnitude of d was interpreted as: 0.2 = small, 0.5 = medium, and 0.8 = large.[30] To assess significance, two-tailed tests were conducted, with p < 0.05. For one outcome (pain-related disability), the effect size of one study[31] was more than four times greater (d = 4.69) than the next largest effect (d = 1.02), and was deemed an outlier. We analyzed this outcome twice, with and without the study. With the study removed, the overall effect reduced in magnitude, but remained significant; thus, we opted for the conservative approach and conducted analyses for this outcome with the study removed.

For the proportion of treatment responders, we determined the risk ratio (RR) and number needed to treat for an additional beneficial outcome (NNTB) for ≥50% decrease in migraine frequency, for the treatment versus control group.

Additionally, we conducted a sensitivity analysis that included both migraine-only studies and studies that combined outcomes from participants with either migraine or TTH. We meta-analyzed all outcomes using the above methods, per available data. Of note, we changed the name of the migraine frequency outcome to "headache frequency" to better reflect the mixed sample.

Effect sizes were weighted by sample size and calculated using a random-effects model, which assumes heterogeneity. Cochrane's Q tests for heterogeneity were performed, and I2 was reported. The magnitude of heterogeneity for each effect size was interpreted as follows: I2 = 0% to 40%, not important; I2 = 30% to 60%, moderate heterogeneity; I2 = 50% to 90%, substantial heterogeneity; and I2 = 75% to 100%, considerable heterogeneity.[32] Comprehensive Meta-Analysis version 3 was used for all analyses.[33]