Castle et al's study ^{Reference Castle, White, Chamberlain, Berk, Berk and Lauder1} provides further evidence that psychosocial treatments can supplement pharmacological treatments and improve the course and outcome of bipolar disorder. It is a well-conceived and well-implemented study. However, I would like to raise three pertinent issues.

First, the authors reported a significant baseline difference between the two groups in respect of comorbidity with anxiety and eating disorders, and reported a P value of 0.04. The P value denotes the possibility of finding the difference by chance. In a randomised controlled trial, any difference between the groups is by chance. The CONSORT guideline advises against carrying out significance tests for baseline imbalances in randomised controlled trials. ^{Reference Altman, Schulz, Moher, Egger, Davidoff and Elbourne2} Hence, we should stop doing significance tests for baseline differences in randomised controlled trials.

Second, the primary outcomes were three parameters of relapse of any mood episode (mania, depression, hypomania, mixed and other). The authors reported a significant difference between the two groups, favouring the treatment, for the combined outcome. On further analysis, they found significant differences with regard to depressive, and manic and mixed relapses. The rationales for combining manic with mixed episodes and excluding hypomanic episodes are not clear. More patients in the treatment group suffered from hypomanic episodes than in the control group (9 v. 5). Had the authors combined mania with hypomania, or mania, hypomania and mixed episodes, they would have reached a different conclusion. Therefore the authors' conclusion that the intervention would reduce the number and duration of relapses of any type is not entirely supported by the data. In a review of psychosocial interventions in bipolar disorder, Miklowitz & Scott emphasised the need to report adverse effects along with the beneficial effects. However, the authors have primarily focused on positive and beneficial effects of the intervention. ^{Reference Miklowitz and Scott3}

Lastly, both groups had high relapse rates. In the treatment group, 28.1% of participants (9/32) had experienced one or more relapses. In the control group, the comparative figure was 55% (22/40). The difference between the treatment and control groups was most obvious for depression (4 v. 15). However, at face-to-face interviews (both at 3 months and 12 months), no difference was found on the Montgomery–Åsberg Depression Rating Scale. Apart from the authors' explanation, other reasons could be the method of assessment and ascertainment bias because participants and assessors were not masked to treatment. The telephone assessments might have identified more patients with borderline depression as depressed in the control group than in the treatment group.