Thirteen studies were included in the review (n=828 participants): four RCTs, four controlled studies and five open studies. Sample size varied between 20 and 155 participants. General quality of the studies was low to moderate (mean 22.8 out of a possible 46). The mean quality scores of the RCTs were significantly higher than mean scores of uncontrolled and open trials. The quality rating scores did not correlate significantly with the effect sizes of individual studies.
Within-group analyses: Pre-post effect sizes ranged from 0.78 to 1.89 with a weighted statistically non-significant mean of 1.18 (95% CI 0.98 to 1.37) and similar results when grouped by different study methodologies. There was considerable heterogeneity for the whole sample of studies (I2=47.2%, p=0.02), mostly because of marked heterogeneity in the open studies (I2=78.9%, p=0.0008). Sensitivity analysis that excluded two studies with the highest effect sizes reduced heterogeneity to 0% and overall effect size to 1.04 (95% CI 0.89 to 1.18). Further outcome measures were presented.
Between-group analyses: Based on three RCTs, a statistically non-significant mean effect size of 1.12 (95% CI: 0.78 to 1.46) was found with no evidence of heterogeneity. One non-randomised controlled study compared group CBT to group-based relaxation training and found no significant difference between groups. However, differential drop-out was noted with a larger number of relaxation training participants withdrawing from the study. One RCT that compared group CBT to individual CBT showing no statistically significant differences between treatments. Two studies compared group therapy to pharmacological treatment, one of which was randomised and the other had treatment allocated by preference. The pooled non-significant ES was 0.80 (95% CI 0.45 to 1.15) with no significant heterogeneity. Comparison of pooled effect sizes of ERP and CBT showed no statistically significant difference.