The controversy about cognitive behavioural therapy for schizophrenia

The effectiveness of cognitive behavioural therapy (CBT) in schizophrenia is currently disputed. For example, the UK National Institute for Health and Care Excellence (NICE)1 recommends this therapy, whereas another influential UK organization, the Cochrane Collaboration, has argued since 2012 that there is no clear evidence that it is effective2, 3, 4.

Of clear relevance here is a network meta‐analysis of psychological interventions in schizophrenia published in this journal5 which found pooled evidence that CBT is effective against positive symptoms. On the contrary, a 2014 meta‐analysis by Jauhar et al6 failed to find clear evidence of effectiveness against this class of symptoms. Since it is important to understand what factors give rise to different results in meta‐analyses7, we, as the authors of those two meta‐analyses, decided to examine why such a discrepancy might have arisen.

Bighelli et al5’s examination of CBT for positive symptoms was based on 27 trials out of a total dataset of 40 that met their inclusion criteria (the remaining studies contained data relevant to one or more of the other outcomes they examined, e.g., overall symptoms, negative symptoms, relapse/rehospitalization, depression, quality of life, functioning and mortality). In these 27 studies, the pooled effect size was at the upper end of the small range, against both treatment as usual (–0.30; 95% CI: –0.45 to −0.14, 18 trials) and inactive control interventions (−0.29; 95% CI: –0.55 to −0.03, 7 trials). A larger effect size was found for CBT compared to supportive therapy (–0.47; 95% CI: –0.91 to –0.03, two trials). Leaving aside the findings for supportive therapy, where the number of trials was small, these findings in themselves are not greatly different from the overall effect size that Jauhar et al6 found for positive symptoms against all controls (–0.25; 95% CI: –0.37 to –0.13, 33 trials).

Where the two meta‐analyses diverged, however, was in relation to the findings in blind trials. Bighelli et al5 continued to find a significant effect against treatment as usual (−0.27; 95% CI: –0.41 to −0.13) in 15 blind trials, but not against inactive control (−0.14; 95% CI: –0.37 to 0.09), although the number of studies here was smaller (n=5). In contrast, Jauhar et al6 found that the pooled effect size for positive symptoms against all controls dropped to very low levels in their sub‐analysis of 20 blind trials (–0.08; 95% CI: –0.18 to 0.03).

The divergent findings in blind studies did not reflect differences in the way in which criteria for blindness were applied to the trials included in the two meta‐analyses. The approach used was similar, and cross‐checking revealed that discrepancies about whether individual studies were rated as “blind” , “non‐blind” or “unclear” were trivial.

The most important difference between the two meta‐analyses was found to concern the inclusion criteria used. While Jauhar et al6 employed a broad strategy similar to those used by NICE1 and the Cochrane Collaboration2, 3, 4, the focus in Bighelli et al5’s meta‐analysis was planned from the outset8 to be on the efficacy of psychological interventions for treating positive symptoms (the indication CBT was initially developed for). Consequently, trials carried out in patients with predominantly negative symptoms and those enrolling stable patients (i.e., relapse prevention studies) were excluded. Bighelli et al8 also decided to exclude studies that were carried out in first‐episode patients; this was on the grounds that such studies have been found to have significantly higher treatment response rates compared with those in chronic patients.

This methodological difference turned out to be consequential. Although the number of studies of CBT included were not greatly different in the two meta‐analyses (27 vs. 33), only 14 of the studies in Bighelli et al5 were also included by Jauhar et al6. This means that Bighelli et al5 had more studies with positive symptoms as explicit inclusion criteria (14 in Jauhar et al6 vs. 27 in Bighelli et al5).

We therefore conclude that the discrepancy concerning the effectiveness of CBT on positive symptoms of schizophrenia (especially in blind studies) found in our two meta‐analyses reflects the substantially differing data sets examined. To reduce confusion in this area, where the study designs are much more variable than those about pharmacological treatments for schizophrenia, we propose that future systematic reviews on psychotherapies for schizophrenia should always document their methods and in particular inclusion criteria in an a priori published protocol.