The experience of psychosis represents possibly the greatest challenge for psychological interventions. Despite seeming like a relatively new approach, Aaron Beck first used cognitive therapy to help with delusional thinking in people diagnosed with schizophrenia back in 1952; precisely the same year that saw the advent of the first antipsychotic medication.
The past two decades has seen the rise and ‘sciencing’ of psychological interventions to help people with psychosis; largely centering on the application of Cognitive Behavioural Therapy for psychosis (CBTp). Although CBTp is advocated by organisations such as the National Institute for Health and Clinical Excellence (NICE), questions have arisen both about its efficacy (Jauhar et al, 2014) and about the quasi-neuroleptic approach of randomised controlled trials (RCTs) in focusing upon symptom-reduction.
More recently, CBTp has now been joined by so-called third-wave CBTp which promote mindfulness, acceptance and compassion as means of change. As Louise and colleagues say in this meta-analysis (2018), these approaches:
…address one’s relationship with, and responses to, experiences and symptoms, rather than attempting to change them.
Louise et al conducted a meta-analysis of randomised controlled trials (RCTs) assessing the efficacy of so-called third-wave interventions (TWI) for psychosis, i.e. mindfulness-, acceptance-or compassion-based intervention.
Their primary outcome was the overall severity of psychotic symptoms at end-of-trial. Multiple secondary outcomes were also assessed (see below).
A systematic search revealed 10 trials employing TWI approaches, with an aggregated sample of 624 participants, with 284 randomised to a TWI (n=284) and 340 to a control condition.
For the primary outcome:
- They found a small, but significant reduction of total symptoms in 8 (K=8) trials (Hedges g = 0.29 95% CI 0.04 to 0.54).
However, this significant effect only emerged in comparisons with TAU/waitlist controls, and not when compared to ‘active control’ conditions (such as befriending).
Meta-regression analyses revealed that the reduction in total symptoms was not significantly related to study quality (as measured by Clinical Trial Assessment Measure [CTAM; Tarrier and Wykes, 2004]), number of treatment sessions or duration of treatment in hours.
Turning to secondary outcomes, no significant improvement emerged for:
- Positive symptoms (g= -0.10 95% CI -0.38 to 0.18, K=4) indeed, positive symptoms were numerically worse after the intervention
- Negative symptoms (g= 0.09 95% CI -0.22 to 0.40, K=4)
- Hallucination-related distress intensity (g =-0.07 95% CI -0.37 to 0.23, K=3) or amount (g = -0.07 95% CI -0.35 to 0.21, K=3)
- Functioning and disability (g = 0.09 95% CI -0.23 to 0.42, K=5)
- ‘Acceptance’ (g = 0.08 95% CI -0.44 to 0.60, K=3)
Significant benefits did emerge on:
- Measures of ‘mindful awareness’. (g= 0.56 95% CI 0.15 to 0.97, K=4)
- Depression (g = 0.39 95% CI 0.09 to .69, K= 3)
Potential publication bias for the total symptom effect size was assessed using Fail-Safe N (FS-N), which estimates how many studies might be required to overturn a significant effect size, and Louise et al concluded the finding was unlikely to be prone to publication bias.
Louise et al conclude that:
…findings from RCTs indicate third-wave interventions are efficacious for the treatment of psychotic symptoms, and for the treatment of depressive symptoms in the context of psychosis.
Strengths and limitations
Some limitations of the original trials and caveats of the current meta-analysis are acknowledged by the authors, although others exist.
At a broad level, the authors rightly question trials using symptom reduction as the primary outcome “…rather than the more specific therapeutic goals of reduced symptom-related distress or improved functioning”. The quasi-neuroleptic approach has been a perennial issue in the CBT for psychosis literature (e.g. Birchwood & Trower 2006) and we might ask: why do trials of psychological interventions continue to measure symptom reduction as the primary outcome?
Surprisingly perhaps, Louise et al found that study quality was not significantly related to effect size; although they acknowledge that low statistical power may have been an issue. This seems likely given that a simple plot of study quality (CTAM scores) against effect sizes suggests a clear inverse relationship, i.e. larger effects occurring in lower quality trials. Indeed, if we look at the two best quality trials (Shawyer et al., 2012; Shawyer et al., 2016), both show numerically worse outcomes for the intervention compared to befriending. In this context, Louise et al found that TWI fared no better than active controls in reducing symptoms and note that “…much simpler interventions may be equally effective in improving psychosis outcomes.” So, study quality may well be an issue and we should be aware of the possibility that TWIs may even have a negative impact on symptoms.
Effect sizes increase as study quality decreases
While all estimates of publication bias are imperfect, the FS-N technique used in this meta-analysis lacks any statistical criterion and more informative techniques exist (see Becker 2005). If Louise et al had used one better approach (the Trim and Fill technique) it would impute two potentially missing trials and the adjusted effect size for total symptoms would become non-significant, so the choice of bias measure here is crucial.
A final, but important limitation of the meta-analysis concerns the power of the original studies themselves to detect the mean effect size reported by Louise et al. A quick calculation shows that the mean statistical power of the 8 trials is just 0.18; such poorly powered trials will discover fewer than around 1 in 5 true effects. Moreover, low power also increases false positives. The underpowering of psychological intervention trials is a perennial issue and alongside evidence of publication bias, might lead us to feel very uncertain about any positive claims.
- Louise et al report a significant reduction in total symptoms following third wave interventions for psychosis. However, this does not occur in trials using active controls such as befriending, suggesting that any such effect is non-specific to TWIs
- Moreover, alternative analyses of publication bias (e.g. trim and fill) suggest, contrary to Louise et als conclusion, that bias may exist and the adjusted effect size becomes nonsignificant
- It is also notable that when larger effects have occurred, they emerge in lower quality trials and that the better-quality trials reveal little or no benefit (or even possibly worse outcomes for TWIs)
- Finally, existing trials are hugely underpowered to detect the effect size reported by Louise et al; and any future planned RCT would require almost as many participants as occur in all 8 trials combined here.
Conflicts of interest
Beck, A. T. (1952). Successful outpatient psychotherapy of a chronic schizophrenic with a delusion based on borrowed guilt. Psychiatry, 15(3), 305-312.
Becker BJ (2006) Failsafe N or File-Drawer Number in Rothstein, H. R., Sutton, A. J., & Borenstein, M. (Eds.). (2006). Publication bias in meta-analysis: Prevention, assessment and adjustments. John Wiley & Sons.
Birchwood, M., & Trower, P. (2006). The future of cognitive-behavioural therapy for psychosis: not a quasi-neuroleptic. British Journal of Psychiatry, 188, 107-108
Cramer, H., Lauche, R., Haller, H., Langhorst, J., & Dobos, G. (2016). Mindfulness-and acceptance-based interventions for psychosis: a systematic review and meta-analysis. Global advances in health and medicine, 5(1), 30-43.
Jauhar, S., McKenna, P. J., Radua, J., Fung, E., Salvador, R., & Laws, K. R. (2014). Cognitive–behavioural therapy for the symptoms of schizophrenia: systematic review and meta-analysis with examination of potential bias. The British Journal of Psychiatry, 204(1), 20-29.