Relaxation Training for Anxiety: A Ten-Years Systematic Review With Meta-Analysis

Gian Mauro Manzoni; Francesco Pagnini; Gianluca Castelnuovo; Enrico Molinari


BMC Psychiatry 

In This Article


We located 19 studies (see table 1 ) with a random allocation of subjects into a relaxation training treatment or in a control/comparison group. The pooled sample was composed by 1005 subjects, whose 568 were allocated in the experimental training groups, while 437 were included in control/comparison groups. The mean age is 33,27 years, mostly women (62,75%). In 8 studies (42%), the sample is composed by people with physical diseases, in 6 (31,6%) by volunteers or students, in 5 (26,3%) by psychological or psychosomatic patients. Progressive relaxation was used in 10 works (41,7%), autogenic training and meditation in 2 (8,3%). Only one study used applied relaxation. The other researches evaluated the effects of multi-methods training (3 studies, 12,5%) or other techniques (6, 25%). Half the papers were North American publications (9, 47,4%), 5 were Asian (26,3%), 3 European (15,8%) and 2 Oceanian (10,5%). The most used instrument was the state form of the STAI (14 studies, 73,7%). Only one study used the trait Scale, so as the Beck Anxiety Inventory. Two works assessed the level of anxiety with the Hospital Anxiety and Depression Scale. The context of training was equally divided between individual (52,6%) and group sessions (47,4%). The most part (68,4%) of the trainings required (or, at least, recommended) implementing some activities at home or outside the clinical setting.

Overall Efficacy of Relaxation Training. The average effect size, weighted by the pooled variance, is .5136 (95% CI: .46-.634). This result indicates a medium-high efficacy, according to Cohen's convention. The range of effect sizes is considerable (from .03 to 1.389), which contributes to a significant test of heterogeneity Q, χ (18) = 28,93, p < .05. This significant heterogeneity of effect sizes suggests that the overall efficacy of relaxation training must be handled with caution because of the differences among the relaxation approaches considered, the kind of subjects and the questionnaire used.

The fail-safe n (for k = 19 interventions and the overall mean d of .5136) tells us that we would need an additional 79 studies with non-significant findings in order to reduce the mean d to a small effect size (.1).

Effect Sizes by Relaxation Training Types. Applied relaxation shows an higher effect size in comparison with all other treatments (p < .01), but not with meditation. However, this result is not reliable because applied relaxation was used just by one study. Meditation proved to be very effective in the reduction of anxiety, statistically superior to the other techniques (p < .01 against progressive relaxation, autogenic training, multi-methods, other techniques). All the other techniques show good efficacy, even if statistically lower than meditation ( table 2 ).

Effect Sizes by Kind of Subjects. Varying the type of subjects, the effects relaxation training has on anxiety change significantly. Volunteers and students show a reduction greater than other types (p < .001 in both cases). There were no differences between medical and psychological patients ( table 2 ).

Moderator Variables Analysis. At study level there is a negative correlation between the average age of subjects and the effect sizes, indicating that young people gain more benefits. There is also a negative correlation between the percentages of women and effect sizes. However, women's presence is higher in studies with psychological and psychosomatic patients.

The context of implementation doesn't seem to influence significantly the efficacy of treatment, even if group sessions have an higher average score than the individual ones. At study level there is a positive correlation (p < .05) between the length of treatment and its effect size. Homework increases effect size in comparison to the therapist's sessions alone (p < .001). There are also differences of effect size among the instruments used for psychometric assessment. Studies that used the Hospital Anxiety and Depression Scale show lower results (p < .001) in comparison to the other questionnaires, which don't differ significantly from each other ( table 3 ).

This analysis is based on 25 studies (see table 1 ), with a total sample of 748 participants.

The mean age of the sample is 32,65 years, with a higher percentage of women (59,5%). In 10 studies (40%) the sample was composed by patients with psychological or psychosomatic diseases, 9 (36%) by patients with somatic troubles and 6 by volunteers or students.

Progressive relaxation is the most studied training among the papers included in this meta-analysis of observational studies (33,3%). Autogenic training, meditation and applied relaxation were implemented in 3 studies each (11,1%). In 4 papers (14,8%) a multi-methods training was implemented, while in others 5 (18,5%) other techniques.

The state scale of the STAI is the most administered questionnaire (15 studies, 60%). The trait scale was found in 3 papers (12%), as like the Hospital Anxiety and Depression Scale, while 2 papers (8%) used the Beck Anxiety Inventory. The context is mainly individual (14 studies, 56%) and homework is suggested in two third (66,7%) of the papers. A great part of the works come from USA or Canada (44%), a third from Europe (32%), 4 (16%) from Asia and 2 from Oceania (8%).

Overall Efficacy of Relaxation Training. The average effect size is .57 (95% CI: .52-.68) and, according to Cohen's categories, is a medium-high score. The range of the results is quite wide (from -.061 to 1,49). Effect sizes are not homogeneous, χ (30) = 55.469, p < .01. This significant heterogeneity suggests that the overall effectiveness of relaxation training must be handled with caution because of the differences among the relaxation approaches considered, the kind of subjects and the questionnaire used, as it's for the between group analysis.

Failsafe N calculation indicates that an additional 118 studies with an effect size value of zero would be needed to reduce the effect size toward the value of 0.1.

Effect Sizes by Relaxation Training Types. Progressive relaxation, applied relaxation, autogenic training and meditation show great efficacy in decreasing anxiety against the combination of more than one methods and the other techniques. The "others techniques" treatment type shows the lowest score ( table 4 ).

Effect Sizes by Kind of Subjects. Comparing values before and after the treatment ( table 4 ), the category of subjects with psychological and psychosomatic diseases had higher decrease of anxiety level in comparison with volunteers (p < .01) and with participants with medical problems (p < .001). Subjects with medical problems show a less decrease of anxiety also in comparison to volunteers and students (p < .01).

Moderator Variables Analysis. At study level, the average age of the samples correlates negatively with the effect size, indicating that older people has a smaller reduction of anxiety in comparison with younger. A positive correlation emerges also between the percentage of women and the effect size. However, also in this analysis there is a higher presence of women in the studies with psychological and psychosomatic patients. The context of treatment doesn't seem to moderate the treatment effect. In fact, there are no significant differences between group and individual sessions. Larger effect size corresponds to a higher number of days on treatment (p < .001). The suggestion to apply relaxation techniques at home, together with relaxation sessions conducted by a therapist, increases the effect size of the treatment (p < .001). Effect sizes are really influenced also by the chosen assessment instrument. The questionnaire associated with the higher effect size is the BAI in comparison with the other scales (p < .01). State scale of the STAI-Y shows an higher effect size than the Trait Scale (p < .05). With the exception of the BAI, there wasn't any statistical difference between HADS and other instruments ( table 5 ).


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.