Sixteen RCTs (n=2,116) were included in the review. Study quality with regard to risk of bias was variable: 10 trials reported methods of randomisation, 11 reported on blinding of assessors and intention to treat was reported in six trials. The quality of treatment implementation of the included studies was good: treatment manual (13 studies), trained therapists (15 studies) and assessment of integrity of treatment (12 studies). The studies were mostly undertaken in North America or Europe.
Psychotherapy had a small but significant effect (d=0.23, 95% CI 0.06 to 0.41; eight studies) on depression when compared to control groups, with no significant heterogeneity and a number needed to treat of 7.7. Psychotherapy was significantly less effective than pharmacotherapy in direct comparisons (d=-0.31, 95% CI -0.53 to -0.09, NNT=5.8; 10 studies). Heterogeneity for this comparison was moderate to high.
In subgroup analyses, the benefit of pharmacotherapy appeared to be limited to selective serotonin releasing inhibitors (SSRIs) rather than tricyclic antidepressants (TCA) (d=-0.47, 95% CI -0.75, -0.18); this finding was limited to participants with dysthymia. When pharmacotherapy alone was compared with combined treatment (pharmacotherapy plus psychotherapy) there was a strong trend in favour of combined treatment (d=0.23, 95% CI -0.01 to -0.47, NNT=7.7; nine studies), with moderate heterogeneity. Combined treatment was significantly more effective than psychotherapy alone (d=0.45, 95% CI 0.2 to 0.7, NNT=4; four studies), with moderate heterogeneity. No significant differences were found in the drop out rates between psychotherapy and other interventions. Metaregression that explored the association between effect size and number of sessions of psychotherapy indicated that each extra psychotherapy session increased the effect size by 0.04 points. There was no evidence of publication bias.