To be considered for the review, studies were required to be prospective, controlled trials of behavioral or physical treatments aimed at the prevention of attacks of tension-type or cervicogenic headache or the relief of symptoms of individual episodes of headache in patients with tension-type or cervicogenic headache. The behavioral interventions considered included the broad categories of relaxation, biofeedback, cognitive-behavioral (or stress-management) therapy, and hypnosis. Physical interventions considered for this report included acupuncture; cervical spinal manipulation; low-force techniques, such as cranial sacral therapy; massage (including trigger point release); mobilization; stretching; heat therapy; ultrasound; transcutaneous electrical nerve stimulation (TENS); surgery; and exercise (including postural exercises). Acceptable control treatments included wait-list/no intervention, sham interventions (placebo), other behavioral or physical treatments, and preventive or acute drug therapies.

Although the use of a specific set of diagnostic criteria (e.g., those developed by the Ad Hoc Committee on the Classification of Headache or the Headache Classification Committee of the IHS) was not required, diagnoses had to be based on at least some of the distinctive features of tension-type headache or cervicogenic headache and had to exclude features characteristic of migraine. Both episodic and chronic tension-type headache were included. Trials involving patients with "mixed" migraine and tension-type headache, chronic daily headache, and post-traumatic headache were considered on a case-by-case basis and were included only if they met reasonable criteria for tension-type or cervicogenic headache.

Studies were included only if allocation to treatment groups was randomized or quasi-randomized (based on some nonrandom process unrelated to the treatment selection or expected response). Concurrent cohort comparisons and other non-experimental designs were excluded.

Relevant controlled trials were identified by searching MEDLINE (January 1966 through September 1999) using the MeSH term "headache" (exploded) and a published strategy for identifying randomized controlled trials. Additional search strategies included computerized bibliographical searching of the PsycINFO, MANTIS, and CINAHL databases and the Cochrane Controlled Trials Register; hand-searching of the Chiropractic Research Archives/Abstracts Collection (CRAC) (conducted by members of a research team headed by Drs. Gert Bronfort and Niels Nilsson); hand-searching of the non-MEDLINE-indexed journal, Headache Quarterly: Current Treatment and Research; searching the references of relevant review articles, meta-analyses, and included trials; and consulting with experts in the field of headache.

Studies identified by the literature search were screened for further review based on criteria focusing on patient population, intervention, study design, and type of outcome data reported.

Included studies were evaluated for methodological quality with respect to three domains: randomization, blinding, and description of dropouts. In addition, the behavioral and physical interventions tested were assessed for clinical appropriateness by experienced clinicians using a scale previously developed for trials of physical treatments for headache.

Information on patients, methods, interventions, outcomes, and complications/adverse effects were abstracted from the original reports directly into specially designed, computerized tables similar in format to the final evidence tables envisaged for the report. We collected trial data on symptomatic outcomes related to head pain and did not consider physiological or other measures not directly relevant to the patients’ symptomatic experience.

We preferred that outcome data be based on daily recording of headache symptoms by patients, rather than on global or retrospective assessments performed by patients or investigators. Outcomes were recorded for all time points reported for which the dropout rate was £ 20%.

For preventive trials, we recorded results for headache frequency, headache index, headache duration, and headache intensity. In the relatively few cases in which a behavioral or physical intervention was aimed at the relief of symptoms of an individual attack of headache, we recorded results for headache relief and headache intensity.

For dichotomous outcomes (e.g., success/failure), we required that the threshold for distinguishing between success and failure be clinically significant; for example, we interpreted a 50% or more decrease in headache frequency as meeting this criterion. Dichotomous outcomes meeting our definition of a clinically significant threshold were reported as proportions (or response rates for each treatment) which may be directly compared (difference in proportions). We also used these proportions to calculate odds ratios and numbers-needed-to-treat.

When outcome data were reported on an ordinal scale, we selected a threshold based on the definition of clinically significant improvement described immediately above and converted these data into dichotomous form. If categorical data could not be split into dichotomous outcomes meeting the a priori definition, they were not included in the analysis.

When outcomes were reported on a continuous scale (e.g., mean headache index or mean headache frequency) and variance estimates were also available, we re-scaled and standardized the continuous outcome data for each treatment condition in each study using a published method. In the case of the behavioral trials, we then used the resulting standardized outcome measures to calculate summary effect sizes for each type of treatment, using a multi-variable, random-effects model, controlling for study. For the purposes of this meta-analysis, the behavioral interventions were grouped into categories based in part on statistical considerations and in part on clinical considerations.

Because some of the behavioral trials that reported continuous data did not permit effect size calculation, the sample of studies included in the meta-analysis may be subject to bias. To investigate this potential bias, we calculated another measure of effectiveness, the percentage of improvement (in headache index or frequency) from pre- to post-treatment. Because large differences between the percentage improvement scores from studies included in the meta-analysis and those from studies excluded from the meta-analysis would suggest bias, we compared the mean percentage improvement scores (weighted for sample size) of the two groups.

We also used the standardized outcome measures described above to calculate individual effect sizes for pair-wise comparisons of active behavioral treatments with control treatments for every trial with a control arm, and to calculate effect sizes for all pair-wise comparisons in those few trials of physical treatments for which effect sizes could be calculated.

Summary of Findings

      Behavioral Treatments

Thirty-five trials of behavioral treatments were included in the report; 23 of these reported continuous outcome and variance data and were included in a meta-analysis. The principal findings of the analysis were:

• Behavioral treatments for tension-type headache have a consistent body of research indicating efficacy. The effect size data suggest that each of the interventions examined (relaxation training, cognitive-behavioral therapy with or without relaxation training, EMG biofeedback combined with relaxation training, and EMG biofeedback alone) is effective for reducing tension-type headache symptoms when compared to wait-list control.
  

• The collection of trials and the results of the meta-analysis provide little guidance for choosing among the treatments considered. The summary effect size estimates for the various categories of behavioral therapy are statistically indistinguishable.
  

• Clinically, behavioral treatments are often used in combination. Five of the trials we reviewed were designed to test the incremental benefit of adding EMG biofeedback to relaxation training, and seven trials allowed estimating the incremental benefit of adding cognitive-behavioral therapy to relaxation training. Finally, three trials examined the effect of adding relaxation to EMG biofeedback. None of these studies found a statistically significant incremental benefit to the added component; however, all the studies were too small to detect small, but potentially clinically significant differences.
  

• The question of combining drug and behavioral therapy has been examined in a single study which suggested that amitriptyline with cognitive-behavioral therapy and relaxation training leads to better headache outcomes than the behavioral component alone. Longer-term 6-month results no longer showed significant differences, perhaps because the behavioral therapy resulted in slower onset of improvement.
  

• A large number of studies could not be included in the meta-analysis because they did not report variance data to allow calculation of effect size scores, even though they met all other inclusion criteria. Comparison of percentage improvement scores from trials included in, and excluded from, the meta-analysis did not substantially change our interpretation of the analysis.