An oversimplification of psychosis, its treatment, and its outcomes?

Jauhar et al’s meta-analysis¹ of randomised controlled trials in cognitive-behavioural therapy for psychosis (CBTp) is broadly consistent with previous results:² that is, there is an overall significant but modest impact on psychotic symptoms, with blinded studies showing lower effect sizes than those that are not blinded. However, there are a number of problems with this study and especially with its conclusions.

Jauhar et al conclude that they find the advocacy by government (including NICE) for CBTp ‘puzzling’, bearing in mind the low effect sizes found for psychotic symptoms. However, I find it puzzling that the authors comment on NICE recommendations, since a third of the studies included for their overall symptoms analysis (12/34) were not based on therapies recommended by NICE in the first place (based on what we know is effective from the literature so far): they were either group or brief CBT studies. Three further studies were in Chinese, so their relevance to NICE recommendations is hard to tell.

It is a testament to the far-reaching effects of CBTp that the analyses revealed any effects at all, since the authors looked at outcomes that were not always targeted by the therapy. For instance, only a few of the 34 studies included for negative symptoms actually targeted such symptoms specifically. Furthermore, severity of positive symptoms/hallucinations was used as the outcome for studies that did not hypothesise changes in psychotic symptoms since the target was on compliance with command hallucinations,³ emotional dysfunction,⁴ or social functioning.⁵ By contrast, outcomes on depression, anxiety or distress as a result of psychotic symptoms, and trials targeting self-esteem, post-traumatic symptoms, suicidality, or substance misuse, which are all main and legitimate targets in CBTp, were excluded.

The criteria for studies to be included in the final analyses were idiosyncratic. Perhaps the most surprising was the decision to exclude studies that targeted hallucinations specifically from their positive symptoms analyses. A separate ‘supplementary’ meta-analysis was carried out for those studies, with an effect size of 0.34, which is not reported in the abstract (where only the - lower - 0.25 effect on positive symptoms is reported). Clinicians familiar with clinical presentations of patients with psychosis might be surprised at their rationale for excluding trials because patients had a dual diagnosis, or had medication-resistant psychotic symptoms but no further diagnosis specification. None of the follow-up data available was included, meaning that the Sensky et al ⁶ (non-significant) end-of-study results contribute to the findings, but the (significant) 9-month and 5-year follow-up results do not.⁷

Meta-analyses can be highly informative, but they are highly prone to bias.⁸ Those with a ‘washing machine’ approach, such as this one (i.e. amalgamating different populations - from acute in-patients to chronic out-patients, from young people with a first episode of psychosis to older adults; different therapies - from 3 sessions of acceptance and commitment therapy to 18 months of weekly cognitive therapy; different modalities - groups or individual; different targets - from compliance with command hallucinations to emotional dysfunction), tell us very little about what works for whom. Unsurprisingly, the heterogeneity statistics were highly significant for all analyses, with I ² being at 50% or above (i.e. representing ‘substantial heterogeneity’), suggesting that there was too much heterogeneity to obtain meaningful pooled estimates, and that the necessary criteria for rendering a meta-analysis appropriate were not met.⁹

The field of CBTp has now progressed such that it is no longer appropriate to simply lump together psychosis patients assuming that clinical presentations are the same, that therapy is for the same problem, and that the type of CBT is the same. Other recent meta-analyses, which focus on treatment-resistant patients,¹⁰ or on individually tailored, formulation-based CBT for hallucinations and delusions,¹¹ will be more informative to clinicians and researchers about the specific effects of CBTp.

To conclude, the reported analyses reflect an over-simplification of the complexities of psychosis and psychological interventions. The biggest challenges in psychological therapy trial methodology (and in clinical practice) are the quality of/adherence to the therapy delivered and the competence of the therapists, none of which was taken into account in this study. A more meaningful reading of the existing research is that the next steps are to investigate which patients benefit on which outcomes at which stages with which types of therapy, and how to ensure therapist competence (and availability).