Cognitive behavioural therapy and the psychopathology of schizophrenia: Systematic review and meta-analysis


*Correspondence should be addressed to Dr Giles Newton-Howes, c/- 76 Wellesley Road, Napier 4110, New Zealand (email:


Purpose. To examine whether cognitive behaviour therapy (CBT) reduces psychopathology in patients with schizophrenia more effectively than the use of non-cognitive psychotherapies.

Method. Systematic review and meta-analysis of the literature was performed. All Randomized Controlled Trials meeting the inclusion criteria were analysed using RevMan software. This design was used to maximize power and study efficacy. Medline, PsycINFO, and Embase were searched using free-text keywords to identify potential papers. Nine were included in the final meta-analysis. Change in psychopathology at the end of therapy was the end point investigated. A random effects model was used to assess the standard mean difference between the CBT and supportive control groups.

Results. Meta-analysis of CBT versus supportive therapy did not find significant differences between the therapy groups at the end of treatment in respect of psychopathology. There was no evidence of publication bias. Post hoc power analysis using the Z test ruled out type one error.

Conclusions. Theoretically based CBT therapies, although proving effective, may not out perform more accessible and simpler forms of therapy for patients with schizophrenia in reducing psychopathology. Consideration of supportive therapy should be made for patients with psychotic mental disorder.

Practitioner Points

  • • CBT may not be the psychotherapeutic treatment of choice to alleviate the phenomenology of Schizophrenia.
  • • It may be valuable trialling simple supportive therapies prior to implementing more costly and complex cognitive therapies.
  • • This review, like the Cochrane review and others, does not suggest CBT in psychosis is not effective, simply that it dose not outperform supportive therapy in effecting change in phenomenology.


The burden of schizophrenia is significant on patients, their families, and society at large and requires a biopsychosocial approach to ensure best outcomes (NICE, 2010). Anti-psychotic medication is been the mainstay of treatment, although it does not relieve all psychotic symptoms in all patients. Cognitive symptoms are increasingly recognized as a core symptom and one that is poorly responsive to pharmacotherapy (Saykin et al., 1994). For this reason, psychological interventions have been increasingly studied and implemented in clinical practice. There have now been numerous randomized controlled trials (RCTs) of cognitive behaviour therapy (CBT), specifically designed to reduce the burden of illness associated with schizophrenia (Haddock et al., 1999; Lewis et al., 2002; Tarrier et al., 1998).

There is an appealing logic to the theory that cognitive psychological approaches can improve psychotic symptoms as some psychotic symptoms have been postulated to be cognitive processing errors (Frith, Stevens, Jonhstone, Owens, & Crow, 1983; Phillips & Silverstein, 2003). Increasingly the literature postulates difficulties in prefrontal cortical processing and subsequent working memory as key areas of deficit (Honey, Bullmore, & Sharma, 2002). This can lead to cognitive problems such as externalized attribution errors (Moritz et al., 2003), a breakdown in normal cognitive attribution leading to delusions (Langdon & Coltheart, 2000) and problems with theory of mind (Langdon, Ward, & Coltheart, 2010). The theoretical framework for cognitive therapy in psychosis suggests these errors are, in theory, accessed and addressed by CBT (Garety, Kuipers, Fowler, Freeman, & Bebbington, 2001). There is also biological evidence that CBT may alter the metabolism of the cortical pathways and as such can be considered as an ‘active treatment’ that is substantially different from placebo (Goldapple et al., 2004). The combination of biological studies suggesting CBT alters brain function and a theoretical framework of understanding CBT in schizophrenia aimed at the brain abnormalities in the prefrontal cortex and working memory combine to support the proposition that CBT should be an effective treatment in reducing the psychopathology of schizophrenia.

Despite numerous trials of cognitive behaviour therapy in schizophrenia (Newton-Howes, 2008; Wykes, Steel, Everitt, & Tarrier, 2008), and its recommendation by some authorities (NICE, 2010), it remains somewhat unclear that any improvement in outcomes is related to improvements in psychopathology that the model of CBT in psychosis implies. A recent meta-analysis would suggest no effect of CBT in schizophrenia symptoms or relapse (Lynch et al., 2010), although this study has been questioned by some and is broad in its approach. There is, in addition to this, a long running debate as to which aspects of therapy can be considered active and if, in fact, common factors are the necessary therapeutic elements as opposed to specific factors such as CBT for psychosis postulates. Some authors suggest relational factors may have more than double the impact of specific ‘techniques’ (Lambert, 1992). This debate would suggest that, despite the logic embedded in the theory of CBT for psychosis, it is in fact the non-specific factors common to all therapies that are important. Indeed, much of the empirical evidence for CBT in this area is derived for academic departments where highly motivated and inspiring academic leaders coach and support therapists, effectively enhancing these common therapeutic facets. Added to this is the concern that in some areas of CBT research, publication bias may lead to an overestimate of efficacy of CBT as opposed to simpler controls, clouding the picture further (Cuijpers, Smit, Bohlmeijer, Hollon, & Andersson, 2010). These types of analyses raise the question as to whether simpler non-theoretically based supportive ‘therapy’ may in fact be as effective as CBT for some clients to help ameliorate their symptoms, despite the appealing rationale for CBT.

In order to address this particular question, we decided to undertake a systematic review and meta-analysis of all placebo controlled randomized controlled trails of cognitive behavioural therapy in schizophrenia. This analysis differs from other recent analyses that are broader and assume effectiveness as their starting point, looking instead at sub-populations. The null hypothesis for this analysis was that there was no difference in the treatment effect on psychopathology for psychosis between theoretically based CBT and atheoritical non-cognitively based therapy. RCTs were chosen as the most robust available to show clear differences between active and placebo arms, maximizing the possibility of finding a true difference between groups if present (Mulrow, 1994). An unavoidable weakness of this approach is the inclusion of all types of psychopathology, rather than a particular facet (such as delusions), although the ‘lumping versus splitting’ difficulty is a common one with meta-analytic approaches to combining studies, and in this case, such an approach was thought likely to provide the most clinically useful results. This review differs from other recent reviews in that by and large these reviews assess general improvement and fail to include only truly matched placebo arms (Jones, Cormac, Silveira da Mota Neto, & Campbell, 2004; Tarrier, 2005; Tarrier & Wykes, 2004). This study also differs by only including direct measures of symptoms, as opposed to indirect measures such as readmission that may be multifaceted in their reasons for being found positive. The importance of using scientifically robust methodologies, due to the difficulty in assessing outcome, has been championed by experts in this field (Tarrier, Haddock, Barrowclough, & Wykes, 2002) as important in the ongoing assessment as to which psychological therapies help patients with schizophrenia best.

Material and Methods

Inclusion criteria

These were kept as broad as possible whilst ensuring the methodologically robust papers were included. Papers were considered for inclusion if:

  • 1Patients were adults aged 18–65.
  • 2Schizophrenia was diagnosed using a peer-reviewed instrument.
  • 3A cognitive behavioural therapy was being used in one arm of the trial. This therapy needed to conform to the outline of Pilling (Pilling, Bebbington, & Kuipers, 2002) to ensure its theoretical basis.
  • 4A non-cognitive therapy was used in the placebo arm that was not based in a theory of cognitive change. This ensured a formal placebo was administered to account for non-theoretical factors such as contact, support, and discussion. These general factors for therapy were considered important to account for to ensure the ‘active’ component assessed was the theoretical cognitive aspects of the model used.
  • 5A peer-reviewed instrument was used to measure the change in psychopathology.
  • 6Papers were written in English.
  • 7Ethical approval given for the study.

Papers not meeting the inclusion criteria were excluded.

Search strategy

Medline, PsycINFO, and Embase were searched electronically from 1950, 1906, and 1974, respectively. The Medline platform was used for the search. The search terms 'psychosis' or 'schizophrenia' were combined with 'psych$ therapy', 'psychological therapy', or 'CBT'. All the titles and abstracts of potential papers were read and any the full article of all potential papers accessed and assessed.

Data extraction and checking

Every paper had a data extraction sheet that ensured uniformity of the data gathered. Two authors extracted all data independently. Disagreements were resolved by consensus.


For each study, a standardized mean difference (SMD) was calculated using the means and standard deviations from the papers included, if they were not directly reported. This statistic represents the difference between the means of the CBT and placebo therapy groups. RevMan weights these SMD's by sample size. Odds ratios of difference were calculated for papers where the data did not allow for the calculation of standard mean differences. All the data was analysed using RevMan analysis software (RevMan). Funnel plotting of the standard error of the SMD was constructed to assess the data for evidence of publication bias. Symmetrical distribution of papers through the funnel plot would be expected as this represents a standard statistical scatter expected according to the likelihood of greater confidence intervals in smaller studies. If this is not found, it suggests some studies have not been published. Following this, forest plots were constructed comparing cognitive therapy to placebo therapy to allow for combination of the data. Heterogeneity was assessed using the I2 statistic and random effects’ modelling was used for heterogeneous data, according to the Cochrane criteria. A confidence interval and z-value were calculated to assess the overall effect size of the different interventions.


Results of the search

A total of 1,756 studies were revealed by the initial search for review. A total of 1,705 studies were rejected for being obviously unsuitable. The remaining 51 studies were reviewed in full. Of the 51 studies, nine fulfilled the inclusion criteria including 602 patients as is shown in Figure 1. Papers were excluded for a variety of reasons, primarily because no control group was included in the trial.

Figure 1.

QUORUM (Moher et al., 1999) statement.

Description of the papers

There were five studies from England and one each from Scotland, Italy, Germany, and the Netherlands. All of the studies were RCTs and they all compared CBT with an appropriate non-cognitive therapy. The placebo therapies used varied between studies; four used Supportive Counselling and the others used Recreation and Support, Befriending, Psychoeducation, Supportive Psychotherapy, and Supportive Therapy. Two studies investigated outpatients, three inpatients while the remaining four examined mixed inpatient/outpatient groups. A combination of the positive and negative syndrome scale (PANSS), brief psychiatric rating scale (BPRS), the scale for the assessment of negative symptoms (SANS) and the perceptual aberration scale (PAS) were used to measure the change in severity of psychopathology. The advantage of assessment by SMD's allows for comparison of these non-uniform measures. Table 1 shows the characteristics of the included studies.

Table 1. Study characteristics.
First author (Year)StudyIP/OPSizeCriteriaTheoretical therapyAtheoretical therapyMeasuresTime
Valmaggia et al. (2006) NetherlandsRCTOP58DSM IVCBTSupportive counselling (SC)PANSS general22 weeks
Lewis et al. (2002) EnglandRCTBoth172DSM IVCBTSupportive counselling (SC)PANSS total sum7 weeks
Haddock (1999) EnglandRCTIP20DSM IVCBTSupportive counselling (SC)BPRS4 months
Sensky et al. (2000) EnglandRCTOP90DSM IV + ICD 10CBTBefriendingSANS9 months
Drury et al. (1996) EnglandRCTIP40DSM IVCTRecreation and supportPAS12 weeks
Tarrier et al. (1998) EnglandRCTBoth52DSM III-RCBTSupportive counselling (SC)Severity of symptoms10 weeks
Pinto et al. (1999) ItalyRCTBoth41DSM IVCBTSupportive therapyBPRS6 months
Durham et al. (2003) ScotlandRCTBoth41ICD 10CBTSupportive psychotherapyPANSS total sum8 weeks
Bechdolf et al. (2004) GermanyRCTIP88ICD 10CBTPsychoeducation (PE)PANSS general8 weeks

Meta-analysis of the sample

Initially a funnel plot was constructed to assess for the possibility of publication bias (see Figure 3). If publication bias was evident, many points would lie outside the expected ‘funnel’ comparing the study size with the estimates made. Occasional small negative studies, for example, are not published. This was not indicated by the funnel plot suggesting appropriate publication of studies in this area.

All trials had the number of patients included and the standard mean difference for outcomes calculated. In the study by Sensky et al. (2000), the standard deviations were not given and so they were calculated from the 95% confidence intervals that were given using a formula from the Cochrane Handbook for systematic reviews of interventions. This allowed for combination of the differences by meta-analysis despite the use of various tools to measure change. Table 2 outlines the means and standard deviations of the included papers.

Table 2. Data extracted.
First author (Year)TheoreticalAtheoretical
N Mean SD N Mean SD
Valmaggia et al. (2006) 3530.46.282329.586.16
Lewis et al. (2002) 8461.7319.698859.9616.39
Haddock (1999)  946.88.751138.317.4
Sensky et al. (2000) 46221.614420.71.8
Drury et al. (1996) 202.062.57204.592.72
Tarrier et al. (1998) 282016.5242316.8
Pinto et al. (1999) 2038.19.72145.711
Durham et al. (2003) 2296.217.71995.216.2
Bechdolf et al. (2004) 40289.248256.2

A random effect model was used to combine the data as heterogeneity was significant. The overall standard mean difference between the CBT group and the placebo group was 0.04 (95% CI –0.29 to 0.36). This non-significant result suggested no statistical difference in symptom reduction rates between the groups as is shown in Figure 2.

Figure 2.

SMD analysis of all studies.

Figure 3.

Funnel plot of studies included in the meta-analysis performed in Figure 2.

Post hoc power analysis was performed using G-Power software (Erdfelder, Faul, & Buchner, 1996). The analysis used the Z test in a post hoc setting and found that with an α level of 0.05 and the group sample sizes being 304 and 298, respectively, the critical z was 1.64 and the power (1-β error probability) was 0.80. This suggested the meta-analysis included sufficient patients to detect a true difference in the groups examined if one had actually existed. This supported the statistical accuracy of the analysis and effectively ruled out type one error.


The aim of this study was to assess the impact of cognitive behavioural therapies on psychopathology in schizophrenia. Direct measures of psychopathology were assessed improving the likelihood the finding from this review relates to the cognitive aspects of CBT as they relate to symptoms, not social markers of change. Similarly power analysis indicated this meta-analysis had sufficient numbers to avoid type one error, the individual studies included lack this. Despite a statistically large enough data set in this review, no difference could be found between the CBT and non-cognitive interventions in methodologically robust studies. This suggests the non-specific aspects of CBT are possibly as important in preventing relapse of symptoms as those based in the theory of cognitive therapies for psychosis. Considered in this light, the positive effect of CBT in reducing psychopathology may in fact be due to factors other than the application of formal theoretically based CBT itself in this patient group, a finding at variance with some other recent reviews (Pfammatter, Junghan, & Brenner, 2006; Wykes et al., 2008) and recommendations (NICE, 2010), but in line with the findings from other groups (Jones et al., 2004, Langdon et al., 2010). These reviews tend to take a broader perspective of outcome and positive impacts may be in areas of functioning as opposed to psychopathology. This review focuses more closely on psychopathology than those above, enabling the assessment of CBT's effect on symptoms to be directly addressed, although any potential differential effect of CBT on psychopathology reduction was not explored. For example, possible improvements in delusional symptoms with CBT compared to global psychopathology cannot be assessed with this meta-analytic approach. This is a weakness of this study, however if this differential effect is indeed true, it suggests that other symptom facets perform even more poorly in the CBT arm, as the overall finding is of no difference between the two approaches; if CBT is providing major improvements in a subset of symptoms, this is offset on average by no or negative impacts in other symptom domains. This is important in the conceptualization and formulation of the approach, using CBT, as broader based goals other than simply symptom reduction may improve success rates. What is undisputed is that CBT, as an additional intervention, outperforms treatment as usual (Jones et al., 2004).

The debate in psychotherapy regarding specific versus general factors in delivering therapy remains current. This paper would support the importance of general factors, and one could speculate the ‘expert effect’ of highly motivated clinical leaders involved in the delivery of CBT may enhance the general factors of CBT therapists more than in general or supportive therapists. Although this may remain of clinical utility, it does not support the reasoning behind the application of CBT in this group of patients, rather the need to ensure therapists are motivated, engaged, warm, and caring.

The findings of this paper, like many meta-analyses, are circumscribed by a number of factors. Most of the studies included in this review were of small size, included expert therapists and carried out under trial conditions. This does not reflect actual practice, however this should have maximized the likelihood of a positive outcome in favour of CBT. This was not found to be the case. Considering this, it is difficult to interpret effectiveness type trials (Turkington et al., 2006) as the explanatory trials included in this review are not positive for the treatment. Without proven efficacy, the looseness of statistical criteria for effectiveness trials makes their interpretation more difficult. Secondly, this review looked at the effect of CBT on psychopathology in schizophrenia. Although there is no indication of improved mental functioning as measured by psychopathological rating scales, this does not rule out a useful function for CBT in addressing other areas of morbidity such as social functioning. It does not, however, support the hypothesis that the cognitive processing deficits of schizophrenia are directly amenable to change by CBT. Thirdly, the decision to include only direct outcome measures limited the number of studies to include. This does, however, ensure that psychopathology was directly measured as opposed to social outcomes such as readmission. Fourthly, the measures included potentially identified symptoms that are not specific to schizophrenia, such as the mood component of the BPRS. This does, however, reflect the way outcome is measured in this field and as such is a reflection of what is known about psychopathology in response to CBT to date. Finally, this paper does not take a fine-grained approach to CBT, assuming all CBT is essentially equal. It may be that some of the modifications to CBT designed to improve its effectiveness in schizophrenia would improve outcome; however, accounting for this introduces significant selection problems. It allows the authors to develop highly selective inclusion criteria, either reducing the clinical utility of the paper or potentially biasing its finding. For these reasons, a broader based approach was used. These points limit the breadth, but not the principle finding of this review and are similar to the findings of other reviews that take methodologically different approaches to the questions posed by CBT in schizophrenia (Jones et al., 2004) and supportive therapies (Buckley, Pettit, & Adams, 2007).

There are a number of possible explanations that may account for this somewhat surprising finding. Firstly, it is possible that the ‘active ingredient’ in the included trials is the activity of research itself. It is well know that inclusion in research trials tends to improve outcome, and it may be that this effect, in these psychotherapy trials leads to equivalent outcomes at endpoint. This would explain both the lack of differential effect and the evidence for intervention being better than treatment as usual. It does not, however, explain the differences in this review compared to others in the field.

The equivocal nature of this review may also be a consequence of the ‘allegiance effect’ whereby academic professionals show an allegiance to a form of therapy in particular patient groups on which both academic careers and therapeutic modalities gain prominence. This bias would be expected to influence not only individually run trials but also meta-analyses and systematic recommendations collated by such groups. This effect has been shown to account for a significant proportion of effect size in outcome in psychotherapy trials (Luborsky et al., 1999) and the bias in favour of cognitive behavioural therapies in schizophrenia has already been identified (Paley & Shapiro, 2002). This bias could lead to the construction of meta-analyses that tend to favour cognitive strategies (for example, by predetermined choice of endpoint, patient inclusion group or trial methodology) as the studies in this area would be well know to those developing the search protocols and undertaking the analysis. Finally, this meta-analysis may be limited by its inclusion criteria, failing to show the modest improvements found in other reviews. This meta-analysis specifically looks for changes in psychopathology on the basis that the theory of CBT naturally leads to an expected change in psychopathology in psychosis. This would be expected to be greatest at trial endpoint – hence the use of these outcome measures and endpoints. Other analyses and trails have shown improvements in other aspects of care such as social adjustment or interpersonal connectedness (Kemp & David, 1996; Roder et al., 2002) and may explain the modest improvement in effects size noted in other reviews. As such, this paper does not suggest CBT is not a useful therapeutic tool for patients with schizophrenia, rather its effect on psychopathology is not greater than for structured supportive therapies of similar intensity when investigating the putative causal factors expected to change with CBT.

This meta-analysis suggests the unfolding nature of CBT in schizophrenia continues to require both theoretical development to explain some of the benefits found in other research and empirical study to see how CBT can maximize its effectiveness. At the current time, it is difficult to see how CBT can be recommended on the basis of change in psychopathology alone, but may be appropriate with a more holistic patient perspective.