Background
Schizophrenia is a severe disorder, and a leading cause of disability with a dramatic burden on society [
1]. Psychological interventions for schizophrenia have been developed to address several aspects of the disorder, and in agreement with guidelines from the National Institute for Health and Care Excellence in the UK and the Schizophrenia Patient Outcomes Research Team in the USA, are widely regarded as necessary interventions [
2,
3]. The importance of research advancements in the field of psychological treatments has been also recently pointed out by the constitution of the Lancet Psychiatry Commission on psychological treatments research in tomorrow’s science [
4].
Among psychotherapies for schizophrenia, cognitive behavioural therapy (CBT) is the most studied, and it is currently recommended by guidelines [
2]. In a recent systematic review and network meta-analysis by our group, which considered all psychological interventions for schizophrenia, 41 out of 53 included randomized controlled trials (RCTs) examined CBT [
5]. Results revealed that CBT is efficacious for treating patients with schizophrenia who present positive symptoms, which had a significant benefit from the treatment when compared to patients receiving usual care, supportive therapy and inactive control conditions such as befriending. In particular, effect sizes measured as standardized mean differences of CBT in comparison with usual care were − 0.38 (95% CI -0.56 to − 0.20) for overall symptoms, − 0.30 (95% CI -0.30 to − 0.14) for positive symptoms and − 0.16 (95% CI -0.29 to − 0.03) for negative symptoms [
5].
However, efficacy measured with rating scales is difficult to interpret. The clinical meaning of results is especially unclear when different measures are used in different studies, and a standardized mean difference is employed as effect size. A pragmatic outcome like response to treatment would make the results easier to interpret. Moreover, meta-analyses provide the relative treatment effects in comparison to an alternative intervention, while, from a clinical point of view, it is important to know the absolute treatment effect that can be expected from a certain therapy.
Nonetheless, the number of patients who improve with a treatment is rarely reported in the studies, and very heterogeneous criteria are used to define it. Probably for that reason, not one of the existing pairwise meta-analyses on CBT for schizophrenia presented data on response rates. The only exception is represented by a Cochrane review by Jones et al., in which the authors pooled response rates from seven trials under the label of “reliable change”, pointing out that these trials applied different definitions of response [
6]. They presented a pooled relative effect size that did not inform on the absolute treatment effect of cognitive behavioural treatment.
As a result, the extent to which patients with schizophrenia and positive symptoms may benefit from CBT remains unclear.
A possible strategy to deal with this issue was applied by Zhu and colleagues, who calculated response rates from continuous outcomes in the field of antipsychotic medication for patients with first episode schizophrenia [
7]. Thus, we decided to apply the same methodology to calculate response rates from studies on CBT that were included in the previous review [
5], in order to provide an easy-to-interpret measure of treatment effect.
Goals for present meta-analysis are: i) calculating how well patients with schizophrenia and positive symptoms respond to cognitive behavioural therapy; ii) examining the determinants of response to cognitive behavioural therapy in this population.
Methods
Study design and participants
The protocol of the original review was registered in PROSPERO (number CRD42017067795) and published [
8]. We included studies in adult individuals with a diagnosis of schizophrenia or related disorders (such as schizophreniform or schizoaffective disorders), presenting current positive symptoms, as defined by inclusion criteria of the trial, with no restrictions on setting, gender or ethnicity. We excluded studies on patients with predominant negative symptoms or concomitant medical or psychiatric illness, and patients at different stages of illness (first episode, at risk of psychosis). Studies were included if at least 80% of the patients had schizophrenia or related disorders (such as schizophreniform or schizoaffective disorders). Following the rules of the Cochrane Schizophrenia group we included trials regardless of the diagnostic criteria used [
9], in order to increase representativeness and generalizability.
Intervention, comparator and outcome
For the current analysis, unlike our previous review, we considered only studies on cognitive behavioural therapy, compared with any non-pharmacological intervention or control condition. Among the included studies cognitive behavioural therapy was administered usually in addition to standard care, which typically included pharmacological treatment. Studies were included in the analysis if they provided data for overall symptoms and/or positive symptoms measured with validated rating scales.
Search strategy and inclusion criteria
We searched Embase, MEDLINE, PsycINFO, PubMed, WHO International Clinical Trials Registry Platform (ICTRP),
ClinicalTrials.gov and Cochrane Collaboration Controlled Trials Register for reports published up to January 2018 for randomized controlled trials that compared CBT with other psychological treatments or with a non-pharmacological control condition in patients with schizophrenia currently presenting positive symptoms. We applied no restrictions for language or publication period. Previous reviews on CBT were also inspected to determine if some studies met our inclusion criteria as well.
Screening and data extraction
Two reviewers among IB, CR, SW and FS independently inspected all abstracts identified in the searches based on the inclusion criteria. Disagreements were resolved by discussion, and in case of doubts the full paper was retrieved for further inspection. Full articles were obtained for all eligible papers, and were again independently assessed by two reviewers. Disagreements were resolved by discussion, and in case of need, by contacting study authors for further information.
Two of IB, CR, SW and FS independently extracted data from the selected studies, considering main reports, secondary publications and supplementary materials, entered the relevant information into a Microsoft-Access database created especially for this study and assessed risk of bias using the Cochrane risk of bias tool [
10,
11]. We contacted authors of included studies published in the last 30 years for missing or additional information about their studies.
Definitions of response
Response is defined typically in schizophrenia trials as a minimum percentage reduction of the PANSS/BPRS total score from baseline to endpoint. Different cut-offs have been used in the literature to define response (for example at least 20, 25, 30, 40% or 50% improvement [
12]). According to equipercentile linking studies comparing PANSS/BPRS scores with simultaneous CGI ratings [
13], an improvement of at least 20% corresponds approximately to ‘minimally improved’ as measured with the Clinical Global Impressions of the raters, while 50% reduction from baseline means much improved according to the CGI [
14‐
16].
In studies on psychological interventions, the number of patients reaching “response” is not often reported: only 12 out of 62 trials presented this information in our previous review [
5]. In trials included in the present analysis this information was reported in 10 out of 33 studies. Moreover, when number of responders is provided, they are often defined with very heterogeneous criteria, that would not be comparable. In order to obtain a reliable measure of the response rate that could be comparable across studies, we calculated the rate of responders from the scores on continuous scales, using the imputation method proposed originally by Furukawa et al. [
17] and replicated [
7,
18]. We used this method to estimate number of patients who reached at least 20 and 50% reduction from baseline of rating scales measuring overall symptoms (mainly PANSS and BPRS), based on means and standard deviations at endpoint or change scores from baseline. Our primary outcome was the reduction of at least 20% from baseline in overall symptoms scale, that corresponds to a minimal improvement [
14]. Since the efficacy of CBT had already been established in our previous network meta-analysis, we wanted now to determine how many patients benefited from the treatment, and even a small decrease in symptoms was regarded as relevant. Additionally, given the focus on patients with positive symptoms, we also calculated response rates from positive symptoms scales, again for 20 and 50% cut-offs. In the case where a scale had a possible minimum baseline score different from 0 (for example, PANSS rated as 1–7 for each item), the application of this method would result in an underestimation of response rates [
12,
19]. Therefore, we subtracted the minimum score of the scales (for example 30 in the case of PANSS total) before imputing the number of responders.
Data analysis
Unlike from most meta-analyses focusing on comparisons between interventions, the aim of the current meta-analysis was to examine the response rate in a population of patients with schizophrenia receiving CBT. Accordingly, in this case the index is not a between-group difference, but rather, a single-group summary, that uses in essence the same meta-analytic calculations [
20]. To obtain an average response rate, we performed a single-group summary meta-analysis in R using the metaprop function in the meta package [
21,
22]. Analyses were conducted separately for both outcomes (reduction in overall and positive symptoms), for both cutoffs (at least 20% and at least 50% reduction from baseline), and using the intention-to-treat datasets.
Heterogeneity was assessed using the I-square statistic (values > 50% were considered considerable heterogeneity) [
23].
In order to explore which study characteristics might explain heterogeneity, we performed subgroup (dichotomous variables) and meta-regression analyses (continuous variables) for the primary outcome response rate at 20% reduction in overall symptoms. When the analysis revealed a possible role for a specific moderator, we investigated further on the number of responders calculated with a 50% reduction in overall symptoms threshold. The following moderators were chosen a priori: blinding of outcome assessor, treatment resistant patients, researchers’ allegiance (whether study authors also developed the investigational intervention of the study), expertise of the therapist, number of sessions, treatment duration, baseline severity, mean age, gender ratio and percentage of participants taking antipsychotic medication. We assessed small-study effects by visual examination of funnel plots.
Discussion
To the best of our knowledge, this is the first systematic review that informs on how well patients with schizophrenia and current positive symptoms respond to cognitive behavioural therapy in randomized trials.
Our main findings were that 44.5% of patients who received CBT reached an at least 20% reduction from baseline in overall symptoms, and can be considered at least minimally improved, while 13.2% of patients reached an at least 50% reduction from baseline in overall symptoms, being considered much improved [
14]. A decrease in positive symptoms of at least 20%/50% occurred in 52.9%/24.8% of patients, respectively. The observed improvement in positive symptomatology might be explained with the fact that CBT for psychosis actively addresses the thoughts and cognitions related to delusions and hallucinations.
We also found that the patients’ characteristics of being treatment resistant, the severity at baseline, and the clinician’s factors of researchers’ allegiance and expertise could have a role as determinants of response to cognitive behavioural therapy.
The response rates were lower in treatment-resistant patients, who failed to benefit from a previous treatment. This finding may be explained by the fact that treatment-resistant symptoms are more difficult to treat, and therefore a CBT intervention can bring only a lower improvement compared to that of other patients. A trial on clozapine-resistant patients receiving CBT, published after the date of our search, reported responders’ rates that are slightly higher than the ones that we found (46 and 7% for the 20 and 50% PANSS total reduction of symptoms, respectively) [
24].
We found a borderline significance for higher response rates in more severely ill patients. This is consistent with previous findings, in which more severely ill patients at baseline had a higher response rate with antipsychotics than less severely ill patients [
25,
26].
A reason for the higher response rates in studies conducted by researchers testing the efficacy of their own treatments could be that they might have a vested interest in showing better results for cognitive behavioural therapy. In order to assess the role of this factor, studies should always report information on researchers’ allegiance.
We also found that patients treated by expert therapists had higher response rates, especially when considering the 50% reduction from baseline threshold. The expertise of the therapist might play a more important role when aiming to achieve a greater symptom reduction.
We did not find a role for the other variables that we investigated as possible moderators (blinding type, number of sessions, treatment duration, age and gender).
It must be noted that, on average, patients in the included studies were only moderately ill, with a baseline PANSS total of 70.55, that corresponds to a CGI between 3 and 4 [
14], and is importantly lower than the one of patients enrolled in antipsychotics trials [
27].
Some limitations should be considered in interpreting our results.
First, it has been shown that the imputation method of response data tends to overestimate very low values and to underestimate extremely high values [
18]. In the case of the present analysis, we adopted a conservative approach in our calculations and subtracted the minimum scores only where it was explicitly declared that the 1–7 version of PANNS and BPRS was used. This may have led to a certain degree of imprecision in calculating the response rates. Future studies should always clearly report which version of BPRS / PANSS was employed in order to allow more precise calculations.
Second, patients in the included studies were also receiving standard care, which usually included antipsychotics, so that cognitive behavioural therapy was delivered as add-on to the pharmacological treatment. However, detailed information on antipsychotic medication was usually not provided in the studies. As a result, it is not possible to ascertain the respective role of cognitive behavioural therapy and medication on the outcome, neither to evaluate the adequacy of the pharmacotherapy provided in combination with CBT. Moreover, administration of CBT to patients with schizophrenia without concomitant antipsychotic medication is a debated issue: some studies have been conducted by Morrison et al. in patients receiving CBT without medication [
24,
28,
29], but other authors have claimed this to be unethical [
30]. We argue that the situation in the studies included in the present review resembles real-life clinical practice settings, where patients, in general, receive antipsychotics in addition to CBT, making our results more generalizable to the clinical context. We claim that future trials should provide detailed information on antipsychotic medication, such as number of patients who actually received antipsychotics and dosages, so that the role of medication can be assessed and differentiated from the role of cognitive behavioural therapy.
A further weakness of these results is the high heterogeneity that we found across different studies. However, we found possible explanations in the role of different moderators as possible sources for heterogeneity in response rates.
This study also presents some strengths. First, the study was planned carefully in agreement with PRISMA guidelines, and followed a sound methodology that was a-priori published in the protocol, including a comprehensive search and the evaluation of quality of studies with the Cochrane Risk of Bias tool. Second, results presented as response rates are easy to interpret for clinicians, and can provide, at first glance, information on the patients’ probability of receiving a benefit from CBT. This information, together with the relative effect sizes coming from comparison of CBT with control conditions, can provide a more complete picture to be considered in the decision making process of treatment strategies for patients with schizophrenia.