Commentary on selected evidence

With advice from topic experts we selected 3 studies for further commentary.

Psychological therapy

We selected the systematic review and meta-analysis by Velthorst et al. (2015) for a full commentary due to the large number of primary studies included in the analysis and overall participant number. This area was chosen for full commentary following topic expert feedback which indicated that there is continuing uncertainty regarding the effects of cognitive behavioural therapy (CBT) for psychosis and that there have been reports suggesting that there may be little or no effect of CBT when evaluating certain outcomes. This study presents evidence which may contradict current recommendations on offering CBT for the treatment of negative symptoms, and therefore a consideration of the interventions in the primary evidence included and the methods used, will help ascertain the impact this study may have on the guideline.

What the guideline recommends

NICE's guideline on psychosis and schizophrenia in adults recommends that individual CBT should be offered to people at risk of developing schizophrenia or psychosis, during first and subsequent acute episodes and during the recovery period for people with persisting positive and negative symptoms. It is recommended that CBT is delivered on a one-to-one basis over at least 16 sessions. A treatment manual should be followed which aids people to establish links between their thoughts, feelings or actions and their symptoms and functioning as well as allowing a re-evaluation of their perceptions, beliefs or reasoning so that it relates to target symptoms. It is also recommended that at least 1 of the following components is included during CBT: monitoring your own thoughts, feelings or behaviours with respect to symptoms or symptom recurrence; promoting alternative ways of coping with target symptoms; reducing distress and improving functioning.

See recommendations,,, and in NICE guideline CG178 for full details.


This systematic review by Velthorst et al. (2015) includes 35 publications covering 30 randomised trials, in 2,312 people with schizophrenia. A random-effects meta-analysis was performed to provide an estimate of the effect size (calculated as Hedges' g) of the influence of CBT on negative symptom severity.

In order to meet the inclusion criteria for this review, primary studies must have evaluated CBT targeted at improving: psychotic symptomology; negative symptoms; social functioning; self-esteem or cannabis use, and included at least 1 behavioural and 1 cognitive technique as part of the primary intervention. All studies must also have reported a negative symptom score as an outcome. Scales used to report this outcome included the Scale for Assessment of Negative Symptoms (SANS), the Brief Psychiatric Rating Scale (BPRS), the Schizophrenia Change Scale (SCS) and the Positive and Negative Syndrome Scale (PANSS).

A number of effect modifiers were evaluated during analysis to account for differences across studies in the delivery of CBT, the participant characteristics and methodology used. Factors analysed included: the number of behavioural techniques included in an intervention; the number of CBT sessions attended; age; gender; illness duration; illness severity; study quality; primary outcome measure used and publication year.


A total of 30 published randomised controlled trials were included in the systematic review (n=2,312). The population included people with chronic illness or recent-onset illness. Only 2 of the 30 studies reported on the effect on negative symptoms as a primary outcome, with 28 studies reporting this as a secondary outcome.

Primary and secondary outcome measures

When considering the effect of CBT from 28 studies which reported negative symptom severity as a secondary outcome, no statistically significant effect was found through a combined estimate for Hedges' g (0.093; 95% CI −0.028 to 0.214; p=0.130). There was also a high level of heterogeneity amongst the outcomes of these studies (χ2=72.392; I2=62.703). No statistically significant effect of CBT was found when considering the combined effect of the 2 studies which reported negative symptoms as a primary outcome (Hedges' g=0.157; 95% CI −0.010 to 0.409; p=0.225).

Short and long term effects

No statistically significant difference in negative symptom severity following treatment with CBT was found by stratifying studies by length of follow up. Measured at short term (3 to 6 months), an effect estimate for Hedges' g of 0.207 (95% CI −0.049 to 0.463, p=0.113) was reported, which decreased to −0.1 (95% CI −0.020 to 0.182, p=0.922) when long term outcomes were evaluated (9 to 12 months).

Potential effect modifiers

CBT which was given on an individual basis, compared to group CBT, was shown to be significantly more effective for reducing negative symptoms (Hedges' g=0.210, p=0.011 for individual CBT verses −0.174, p=0.204 for group CBT). The number of sessions provided, treatment type (CBT versus CBT 'plus'), study population or strictness of the control condition were not significantly associated with treatment effect.

Strengths and limitations


All studies included in the systematic review were randomised controlled trials, which were independently assessed by 2 authors for study quality using a validated tool. The methodology used to identify relevant studies was appropriate, shown by the report of a clear search strategy which was used to identify relevant evidence from a variety of sources.

Thorough analysis has been performed to assess the differences in effect which may be concluded due to differences in study methodology, intervention delivery and participant characteristics across the included studies.

The population and outcomes evaluated are directly relevant to NICE guideline CG178, as all included participants were people with schizophrenia and the effect on symptoms is an important outcome for this guideline.


The majority of studies included were not targeted at improving the presence of negative symptoms, although this was the outcome assessed during this meta-analysis. As such, the overall population had a range of baseline negative symptom scores, which may have impacted the effect which could be identified following intervention. This may also have been impacted by the use of a range of different outcome measurement tools.

There was a high level of heterogeneity shown which could not be fully explained, including within the results for sensitivity analysis.

It was not possible to distinguish between the effect of antipsychotic medication and CBT in individual studies. Although CBT was the main intervention component in all studies included in the systematic review, an inability to ascertain which studies may have been influenced by changes in medication during the intervention is a limitation of this review.

Four of the 30 included studies are likely to include an intervention focusing on CBT to prevent substance misuse, which is not relevant to NICE guideline CG178. It cannot be determined in which sensitivity analyses this may have had a significant impact.

Impact on guideline

The current recommendations regarding CBT are specific, suggesting one-to-one interaction over 16 sessions. The evidence presented here suggests that CBT delivered individually is significantly more effective than CBT delivered within a group, which is consistent with the current recommendations. While the overall meta-analysis outcome suggests that CBT is not effective for the improvement of negative symptoms, a large range of methods for the delivery of CBT are used across the studies included. The topic expert view, that in some circumstances CBT may not be an effective treatment is supported by this. However, methods such as individual CBT, which were reported to be more effective than group CBT in this systematic review, are specifically recommended in this guideline. This, in combination with several important limitations to the study methodology, means that the evidence reported is not sufficient to suggest that the recommendations made in NICE guideline CG178 would be impacted.

Interventions for people whose illness has not responded adequately to treatment

We selected Samara et al. (2016) for a full commentary because it presents a network meta-analysis (NMA) which may contradict current recommendations for the use of clozapine for people with treatment resistant schizophrenia or psychosis. It was also highlighted by topic experts that there may be issues regarding physical health risks associated with clozapine and therefore a commentary of this study's conclusions are warranted.

What the guideline recommends

Currently, NICE guideline CG178 recommends that people with illness which has not responded adequately to treatment of adequate doses of at least 2 other antipsychotic drugs (with at least 1 of these having been a non-clozapine second-generation antipsychotic), should be offered clozapine (recommendation


This NMA considered 40 randomised controlled trials (n=5,172), evaluating drug treatment for adults with schizophrenia, schizophreniform disorder or schizoaffective disorder that has been resistant to treatment, as defined by the primary studies. To meet the inclusion criteria for the NMA, trials must also have: assessed the intervention as monotherapy; had a minimum duration of 3 weeks and included another antipsychotic or placebo treatment as a comparator.

Multiple relevant databases were searched, as well as reference searches of included studies being conducted, which were then screened and analysed by 2 independent reviewers. Studies which presented with a high risk of allocation or treatment concealment bias, as determined by the Cochrane Collaboration's risk of bias tool, were excluded, as were studies conducted in mainland China. There was also an attempt to alleviate missing data bias by contacting study authors.

The primary outcome measure was the change from baseline to end point as measured on the Positive and Negative Syndrome Scale (PANSS), the Brief Psychiatric Rating Scale (BPRS), or any other validated scale which assesses schizophrenia symptoms. Effect sizes were calculated as Hedges' adjusted g standardised mean differences (SMD).


Nine antipsychotics were considered as part of this study[1]. For improvement of the primary outcome – change in symptom severity in people with treatment resistant illness – the NMA indicated the following:

  • Olanzapine was shown to be significantly more effective than quetiapine fumarate (SMD −0.29, 95% credible interval [CRI] −0.56 to −0.02), haloperidol (SMD −0.29, 95% CRI −0.44 to −0.13) and sertindole (SMD −0.46, 95% CRI −0.80 to −0.06).

  • Clozapine was significantly more effective than haloperidol (SMD −0.22, 95% CRI −0.38 to −0.07) and sertindole (SMD −0.40, 95% CRI −0.74 to −0.04).

  • Risperidone was significantly more effective than sertindole (SMD −0.32, 95% CRI −0.63 to −0.01).

All the other comparisons of antipsychotics made in the NMA did not show any significant difference in effect.

Strengths and limitations


This NMA was conducted against an a priori study protocol, which included only blinded randomised controlled trials and excluded studies which showed high risk of bias. The methods used to conduct the NMA were robust, and included a range of relevant databases being searched and study selection and data extraction being performed by multiple study authors.

Inconsistencies throughout the network were identified, and analysed during sensitivity analysis, indicating robust methodology. The network was also entirely connected, without the need to expand the study protocol in order for connections to be made.


The population included in this NMA is not specific to the population included in the current recommendations on inadequate response to treatment. This is due to the fact that NICE guideline CG178 defines treatment resistance as not responding adequately to treatment despite the sequential use of at least 2 different antipsychotics (with at least 1 of these being a non-clozapine second-generation antipsychotic). Treatment resistance was defined differently by many of the included studies. During a sensitivity analysis, only studies which defined treatment resistance as it is in NICE guideline CG178 were included, with only 11 comparisons remaining in the network. Results from this analysis indicated no statistically significant difference in mean change in overall symptoms between antipsychotics.

A number of the antipsychotics included in this NMA are not currently licenced for use in the UK (sertindole, ziprasidone and fluphenazine as an oral formulation).

The median trial length of included studies was 11 weeks, meaning long term outcomes cannot be reliably concluded.

There were high rates of attrition throughout the included studies, with the mean drop-out rate reported as 30.2%. There was also evidence of selective reporting in 45.0% of included studies.

Impact on guideline

While olanzapine, clozapine and risperidone were shown to be more effective than some other antipsychotics evaluated for treatment resistant schizophrenia, the results presented in this NMA do not clearly indicate a single antipsychotic to be most effective. As well as this, a number of limitations lessen the impact this study has on the guideline. Most notably, the overall population was not specific to that referred to in the relevant recommendations in NICE guideline CG178, and when this population was evaluated individually, no significant difference in antipsychotic efficacy was reported. In addition, a number of the antipsychotics included in the NMA are not currently available in the UK. Therefore, in combination with the evidence considered during guideline development, clozapine is likely to be the most effective treatment for the specific population described.

Promoting recovery and possible future care

We selected this network meta-analysis (NMA) (Zhao et al. 2016) for a full commentary because it provides an overview of the effects of a large range of antipsychotic treatments aimed at reducing relapse in people who are clinically stable, allowing a comparison of many treatment options for this specific population in a single study. Compared to other systematic reviews identified specifically for promotion of long term recovery, this is a relatively recent study with a large sample size.

What the guideline recommends

Currently there are no recommendations for treatment with a specific long term antipsychotic, although it is recommended that an antipsychotic should be offered to people at all stages of illness following a diagnosis of schizophrenia or psychosis. It is recommended that the choice of antipsychotic should be made by the service user and healthcare professional together, taking into account a range of benefits and side effects which accompany individual treatment options. See recommendations,, and for details.


This NMA by Zhao et al. (2016) includes 56 trials with a total of 10,177 participants, evaluating 18 antipsychotics[2]. Literature was identified through searches of PubMed, PubMed/Medline and the Cochrane library, as well as through reference list searches. Studies were included if they were blinded randomised controlled trials evaluating the use of antipsychotic monotherapy for relapse prevention in people with clinically stable schizophrenia. Trials conducted in people with predominantly negative symptoms, known treatment resistance or acute illness were excluded.

The primary outcome reported was relapse as defined in each included study at its longest follow up. Definitions of relapse included clinically assessed or rating scale based exacerbation of psychotic symptoms, hospital admission and the need to change medication. Adverse effects such as weight gain were also reported as secondary outcomes.


Results of the meta-analysis indicated that all antipsychotic treatments were significantly more effective than placebo at preventing relapse, with the exception of trifluoperazine, which did not achieve statistical significance. The NMA indicated that olanzapine was more effective at preventing relapse than chlorpromazine (odds ratio [OR]=0.35, 95% CI 0.14 to 0.88) and haloperidol (OR=0.50, 95% CI 0.30 to 0.82). Fluphenazine long acting injectable was also more effective than chlorpromazine (OR=0.31, 95% CI 0.11 to 0.88). Any other differences in efficacy between antipsychotics were minimal and non-significant.

Fifteen studies included the outcome of weight gain. It was found that olanzapine caused significantly more weight gain than amisulpride (OR=2.31, 95% CI 1.04 to 5.17) and haloperidol (OR=4.15, 95% CI 1.97 to 8.71). Quetiapine was shown to cause significantly less weight gain than olanzapine (OR=0.40, 95% CI 0.20 to 0.78), as was risperidone (OR=0.44, 95% CI 0.20 to 0.94), and ziprasidone (OR=0.13, 95% CI 0.06 to 0.27).

Strengths and limitations


The systematic reviewing methodology of this NMA was thorough, as 2 reviewers selected the studies for inclusion, synthesised and analysed the evidence, and a validated tool was used to evaluate the potential risk of bias of each study.

Methodological strengths specific to NMA included that loop-specific tests were performed, for which 17 out of 21 loops showed no significant inconsistency. Furthermore, no significant inconsistency between direct and indirect evidence was identified within the entire network (p=0.14).


Heterogeneity between studies was not specifically reported, and there are concerns that between study differences may have introduced bias. Only very limited sensitivity analysis was performed in order to account for differences in drug dosages across studies, meaning the assumption of transitivity is in doubt. It is also noted that relapse rates were measured differently across studies. Therefore, there is a potential for bias in favour of antipsychotics evaluated by studies which measured relapse using a more extreme outcome, such as hospitalisation, compared to decline on a symptom rating scale.

Some antipsychotics were better represented than others, with 44% of the agents being represented in only 2 to 3 trials. The smaller overall sample size that is likely to accompany fewer representative trials gives these antipsychotics a disadvantage for achieving statistical significance in a NMA.

The search approach conducted was limited, with backward citation searches of systematic reviews identified through PubMed forming the basis of the search, alongside a limited search of PubMed/Medline and the Cochrane library.

The exclusion of primary studies with a population who predominantly experience negative symptoms means that the population of this study does not wholly represent that included within NICE guideline CG178.

Impact on guideline

The evidence identified showed that across nearly all treatment options, antipsychotics are more effective than placebo at preventing relapse in clinically stable people with schizophrenia, consistent with the recommendations to offer antipsychotic medication as treatment. In the NMA, greater efficacy was shown for olanzapine and fluphenazine long acting injectable compared to a minimal number of other treatment options. However, full consideration of antipsychotic side effects such as weight gain – associated with olanzapine – and extrapyramidal effects – associated with typical antipsychotics such as fluphenazine – need to be considered before prescription. Other antipsychotics showed minor and non-significant differences in efficacy. It should also be considered that while it was reported that ziprasidone has less propensity to cause weight gain than olanzapine, this drug is not currently available in the UK. Overall, the evidence presented in this NMA does not convincingly place any antipsychotic above another after consideration of efficacy and side effects. This was also emphasised when the results of the meta-analysis were considered, which showed consistently overlapping confidence intervals, for all antipsychotics considered. Therefore, the current recommendations that choice of antipsychotic should be based on discussion between the service user and the healthcare professional, after consideration of benefits and side effects, are unlikely to be impacted by this evidence.

[1] chlorpromazine, clozapine, fluphenazine, haloperidol, olanzapine, quetiapine, risperidone, sertindole and ziprasidone.

[2] amisulpride, aripiprazole, chlorpromazine, flupentixol decanoate (flupentixol LAI), fluphenazine decanoate (fluphenazine LAI), haloperidol, haloperidol decanoate (haloperidol LAI), olanzapine, paliperidone, paliperidone palmitate (paliperidone LAI), pipotiazine palmitate (pipotiazine LAI), quetiapine, risperidone, risperidone LAI, sulpiride, trifluoperazine, ziprasidone and zuclopenthixol decanoate (zuclopenthixol LAI).

This page was last updated: 09 November 2017