This meta-analysis was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) statement for the meta-analyses of RCTs. A number of databases were searched, including PubMed, Medline, OVID, Cochrane Library, EMBASE, CINAHL plus, and PsycINFO. References from selected articles were also reviewed. The inclusion criteria included: (1) RCT, (2) patients aged 18 years or above, (3) presence of comparison group, (4) availability of at least one relevant outcome such as changes in the relapse rate, depressive symptoms, mania severity, and psychosocial functioning, such as HRSD, YMRS, GAF. The exclusion criteria were (1) no relevant data were available for further meta-analysis, and (2) non-RCT. Studies using psychological therapies based on CBT or CBT-modified programs were also included. [6]

The data was extracted by two independent reviewers, and consensus meeting with a third researcher was held to resolve disputes. Quality-control process was done by another researcher according to Cochrane Collaboration's tool for assessing the risk of bias in randomized trials.

Statistical outcomes included odds ratio for relapse, and effect size in terms of Hedges’ g for continuous outcomes. Random effects model was applied. Subgroup analysis, sensitivity analysis and publication bias was also assessed.

19 studies were included in the meta-analysis. Results are summarized as follows:

(i) Depression: 13 RCTs included. Hedges's g = -0.494; 95% CI = -0.963 to -0.026; P = 0.039, with a moderate effect size. Large heterogeneity was observed.

(ii) Mania: 11 RCTs included. Hedges's g = -0.581; 95% CI = -1.127 to -0.035; P = 0.037, with a moderate effect size. Large heterogeneity was noted.

(iii) Relapse rate: 10 RCTs were included. Pooled Odds Ratio = 0.506; 95% CI = 0.278± 0.921; P = 0.026. Large heterogeneity was observed.

(iv) Psychosocial functioning: 7 RCTs included. Hedges's g = 0.457; 95% CI = 0.106±0.809; P = 0.011, with a moderate effect size. Large heterogeneity was noted.

Sensitivity analysis was performed using leave-one-out approach and showed that the findings were robust.

Publication bias was performed using funnel plot and Egger's test, and showed that publication bias was not significant.

Subgroup analysis was performed and revealed that studies assessing subjects with BAD I only reported greater reduction in relapse rate than those assessing subjects of both BAD I and II. For the recovery of depressive symptoms and reduction in mania severity the effect size of CBT was also significantly larger for CBT treatment durations of 90 minutes per session or above.

The 19 RCTs that was included received a total research quality score of >6 (a score of 6-10 is acceptable). The findings of this meta-analysis were similar to those of Jan [7] and Lam [8].

Traditionally CBT for depression deals with distorted cognitions which might lead to negative mood states. However, CBT for bipolar disorder also deals with distorted cognitions during manic states –the “hyperpositive thinking”, which was not a conventional treatment target for CBT but is worth further studies. [6]

The subgroup analysis showed that CBT had greater effectiveness in reducing relapse rate in patients with BAD I compared with patients with both BAD I and II, which might be due to the higher heterogeneity with BAD II. The finding that CBT treatment durations of 90 minutes or more were much more effective implies that treatment duration was a potential moderator for treatment efficacy. It may be due to the fact that CBT as a psychotherapy relies on a strong collaborative therapeutic relationship, which is strengthened by a more thorough process and longer treatment duration. [6]

There were several limitations for this study. First, some comparisons were limited by sample size, as only four studies had more than 100 patients, and other RCTs involved small samples. Second, moderate to high heterogeneity was observed, indicating the possibility that some confounders (such as age, gender, CBT style) may affect the results. Third, it was possible that non-significant findings might not have been published thus biasing the results towards favouring CBT. However, the paper calculated the number of RCTs with zero effect size that was needed to reduce the present effect size to zero, and concluded that this possibility was unlikely. [6]

For future research, the paper suggested that more RCTs with larger sample sizes were warranted, and that the optimized and systematic approaches of CBT should be further investigated [6].