Fourteen controlled trials (n=1,484) were included: 13 randomised controlled trials (RCTs; n=1,464) and one non-randomised controlled trial (n=20).
Patients receiving CBT benefited more at post-treatment than patients receiving the control treatment. The ES from the fixed-effect model was 0.35 (95% CI: 0.23, 0.47), based on 17 comparisons. No statistically significant heterogeneity was found (P>0.05). The fail-safe N test indicated that publication bias was unlikely.
For comparisons of blinded versus non-blinded outcome assessment, the ES did not differ statistically (P>0.05) between the two subgroups: ES 0.54 (95% CI: 0.29, 0.79) and 0.29 (95% CI: 0.15, 0.43), respectively.
For comparisons of patients with acute versus chronic disease (13 trials, 1,001 patients), the ES was greater for acute than chronic patients, although this difference was not statistically significant (P>0.05); the ESs were 0.57 (95% CI: 0.31, 0.83) and 0.27 (95% CI: 0.11, 0.42), respectively.
For comparisons of CBT with the three control interventions, the ES was greater for waiting-list controls than for treatment as usual or other (non-specific) treatments, although these differences were not statistically significant (P>0.05); the ESs were 0.55 (95% CI: 0.04, 1.14), 0.30 (95% CI: 0.15, 0.46) and 0.42 (95% CI: 0.21, 0.64), respectively.
Further splitting of the analyses by acute or chronic disease and type of control intervention simultaneously did not reveal any subgroup that benefited substantially more than any other subgroup.
Analyses split by subgroup indicated that the post-treatment ES was maintained in early and later follow-up; the ESs were 0.40 (95% CI: 0.24, 0.57) and 0.33 (95% CI: 0.14, 0.51), respectively.
In none of the above analyses was there any evidence of heterogeneity between the studies (P>0.05).