The effect of treatment on insight in psychotic disorders - A systematic review and meta-analysis

Background: For people with a psychotic disorder lack of insight can be detrimental on their condition and recovery. For this reason, insight has been considered as a target for therapy. We conducted a systematic review of the literature on pharmacological, psychological and other treatments to test the hypothesis that these interventions could improve insight. Methods: We performed a literature search (1970 – 2020) across the following databases: PubMed, EMBASE, PsychINFO, Medline and Web of Science. Within each database the following search terms and the associated Boolean operatives were used: “ Insight AND (treatment OR therapy) AND (psychosis OR schizophrenia) AND (awareness or denial) ” . Further filters were applied to identify peer reviewed controlled trials on adults. Following assessment for bias and inclusion criteria, we calculated the effect size (Cohen's d) for each study and overall, using a random effects model with 95% confidence intervals. Results: Of 94 articles found in the initial literature search, 30 studies that examined the treatment of insight in psychosis met the initial selection criteria and were assessed for bias. A total of 21 studies were included in the final meta-analysis. The overall calculated mean effect size for all interventions was 0.441 (95% CI, 0.23 – 0.66), representing a medium effect size. The effect of psychoeducation studies alone was medium (0.613, 95% CI, (cid:0) 0.35 – 2.06), but not significant. The effect of CBT studies was small (0.235, 95% CI, 0.01 – 0.46), and significant. The effect of combined antipsychotic medication and psychosocial intervention was of medium size and significant (0.683, 95% CI = 0.54 – 0.83). Finally, tDCS over the left fronto-temporal cortex, produced a very large and significant improvement of insight 1.153 (95% CI = 0.61 – 1.70), which was present for at least a month after the intervention. Conclusions: Despite the variation and small number and size of trials into possible interventions, the hypothesis that insight could be improved was confirmed. Whilst most research focuses on psychotherapies, there is scope and potential for pharmacological, as well as other interventions (e.g. physical exercise, self-video observation, Direct Current Stimulation) to improve insight over and above treatment as usual. Given the association of insight with illness severity and treatment adherence, it is important to direct efforts in therapies that target insight improvement in psychosis.


Introduction
A lack of insight or awareness is a common feature amongst patients suffering from schizophrenia or other psychotic disorders (Lincoln et al., 2007). The association between insight and psychosis has been studied since the early 20th century (Kraepelin, 1919), however the presence of insight remained a rather binary concept for the following six decades, i. e. patients either had insight or they didn't (Beck et al., 2004). It wasn't until the late 1980s that insight started being studied as a degree of awareness of illness (David, 1990), and a number of systems were devised to measure and quantify it.
Insight is further subdivided into clinical and cognitive. Clinical insight refers to the "lack of awareness of a mental illness requiring treatment" (Beck et al., 2004), and is generally an observation and assessment made by a clinician. Cognitive insight refers to the patient's own assessment of any delusions and misinterpretations, making it a more subjective assessment. The Insight and Treatment and Attitudes Questionnaire (ITAQ) (McEvoy et al., 1989), Schedule for the Assessment of Insight Extended (SAI-E) (David, 1990), Birchwood Insight Scale (BIS) (Birchwood et al., 1994), Positive And Negative Syndrome Scale (PANSS) Item G12 (Kay et al., 1987), and Scale of Unawareness of Mental Disorder (SUMD) (Amador et al., 1993) all use a variety of questionnaires and patient self-reports to assess the level of clinical insight, whilst the Beck Cognitive Insight Scale (BCIS) (Beck et al., 2004) is used to measure levels of cognitive insight.
Furthermore, lack of insight has been associated with severity of illness, particularly the positive symptoms of schizophrenia (Sevy et al., 2004). Whilst precisely how insight relates to the pathogenesis of disease remains unclear, e.g. whether a lack of insight causes and/or correlates with an increase in psychosis or vice versa, there is also evidence to suggest that it often has a negative effect on treatment adherence (Novick et al., 2015). The associations with disease severity and reduced adherence make insight a suitable target for intervention. Despite this there is a relative paucity of research into the efficacy of targeted therapies. One of the most recent and comprehensive reviews and metaanalyses relating to changes in insight (Pijnenborg et al., 2013) reported that whilst some of the interventions had a moderate effect, they did not yield a statistically significant improvement of insight. Crucially, however, the review also concluded that there was promising potential for further trials and development of therapies targeting insight. This potential for further intervention development, along with the time that has elapsed since that review's publication, therefore warranted further investigation into the treatment of insight. The objective of this systematic review and meta-analysis was to establish the efficacy of treatments on insight for patients diagnosed with schizophrenia or other psychotic disorders. Our hypothesis was that interventions would improve insight.

Inclusion criteria
The criteria for including studies in this review were: randomised or quasi controlled trials that examined interventions associated with the treatment of psychosis in patients diagnosed with schizophrenia or another psychotic disorder. The rationale for including studies of this type was that it would enable a comparison of effect sizes between studies, as well as between different types of intervention and control/ comparison groups.
All types of intervention were eligible and included, but were not limited to: psychological, pharmacological, social, video observation, exercise and psychoeducational therapies. Similarly, there was no restriction placed upon the timing of interventions, mode of delivery or time of follow up. All studies had to be published in the English language in international peer reviewed journals. No grey literature was included.
Insight was a required outcome measure and had to be evaluated by a recognised means of assessment, such as the Insight and Treatment and Attitudes Questionnaire (ITAQ) (McEvoy et al., 1989), Schedule for the Assessment of Insight Extended (SAI-E) (David, 1990), Birchwood Insight Scale (BIS) (Birchwood et al., 1994), Beck Cognitive Insight Scale (BCIS) (Beck et al., 2004), Positive And Negative Syndrome Scale (PANSS) Item G12 (Kay et al., 1987), and Scale of Unawareness of Mental Disorder (SUMD) (Amador et al., 1993).
Study participants were required to have a diagnosis of schizophrenia or other psychotic disorder according to the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV or V). Finally, we only included studies where all participants were adults (ages 18-70 years) and had no underlying neurological conditions or a diagnosis of substance abuse.
These criteria form part of a protocol which has been registered on the National Institute of Health Studies PROSPERO database and is available at: https://www.crd.york.ac.uk/prospero/display_record.php ?ID=CRD42020223080.
The vast majority of the patients were reported to be on medication, to have suffered symptoms for a long period of time (usually several years), and to be in a stable condition for the duration of the study (Supplemental table).

Search methods for identification of studies
We performed the literature search across the following databases: PubMed, EMBASE, PsychINFO, Medline and Web of Science for the time period between January 1970 and December 2020. Papers were required to be written in English and peer reviewed. Within each database the following search terms and the associated Boolean operatives were used: "Insight AND (treatment OR therapy) AND (psychosis OR schizophrenia) AND (awareness or denial)".
We applied additional filters within each database in order to refine the search according to the inclusion criteria. The type and number of filters differed between databases, and the details are listed in Table 1.

Selection of studies
After the database search was complete, the details of the articles were added to an electronic bibliography management software package (Zotero) to aid organisation and synthesis of information. We then independently checked each article to determine whether it met the inclusion criteria.

Assessment of risk of bias
The eligible articles were assessed for risk of bias using the Standard Quality Assessment Criteria for Evaluating Primary Research Papers from a Variety of Fields (Kmet et al., 2004), also known as the Qualsyst Tool. We independently assessed each paper, and calculated a bias risk score. We then compared these scores to evaluate inter-operator agreement, and we used an inclusion threshold of 0.55.

Data extraction and management
We used a standardised proforma to ensure a systematic and uniform extraction of the data from the various studies.
The primary outcome measure for this study was the change in patient insight following an intervention. Therefore, we extracted the mean insight score and standard deviation at baseline and endpoint for the intervention and control groups (Pijnenborg et al., 2013). Where confidence intervals were reported instead, we calculated the standard deviation from those. If additional follow up points of measurement were reported, we only used the measurements taken at the end of the intervention.

Requests for data
A request was made to the authors of one paper (Senormanci et al., 2021) for data access, including the means and standard deviations of the scores from the intervention and control group SAI questionnaire.

Measures of treatment effect on insight
There are several different scoring systems used to measure insight and consequently the system used varied between studies, namely: ITAQ, BCIS, SUMD, PANSS G12, MIC and SAI and SAI-E.
One study used both the PANSS G12 and SAI-E scoring systems to measure insight (Pijnenborg et al., 2019). In order to generate a single effect size for data synthesis it was necessary to select one of these scoring systems. We chose the PANSS G12 over the SAI-E scores, because the baseline measures of insight for the intervention and control groups were more closely matched when measured with the former scale.

Data synthesis and statistical analysis
We extracted the means and standard deviations of insight values for intervention and control groups, both at baseline and following the intervention. We used the Excel-based calculation formulae developed by Wilson (2001) for calculating the correlation, r, for the control and interventions groups in each study. We then used the software Comprehensive Meta-Analysis, Version 3 (Borenstein, M., Hedges, L., Higgins, J., & Rothstein, H., Biostat, Englewood, NJ 2013), which assessed the heterogeneity of the studies, and calculated the effect size for each study, and overall, using a random effects model, and produced the forest plot with 95% confidence intervals.
The heterogeneity of the studies (the outcome variability due to clinical and methodological differences), was measured with Higgins's I 2 statistic. Higgins's I 2 represents the percentage of variation between the sample estimates that is due to heterogeneity rather than to sampling error (Muka et al., 2019).

Search results
The initial literature search resulted in 129 potential articles. Despite the additional filters applied to each of the databases, after reading the papers, 74 were excluded for not meeting the inclusion criteria, see Fig. 1. The reasons for the articles not meeting the inclusion criteria were most frequently due to the relationship of the intervention with insight not being examined, the trial not being controlled, or even absence of any intervention.
Twenty-nine articles met the inclusion criteria and were included in the bias analysis. Due to the relatively small number of included articles, we applied a threshold of bias of 0.55. Seven articles fell below the inclusion threshold, and 1 did not contain data in a format that could be used for meta-analysis.
Overall 21 studies were included in the meta-analysis (N = 3035). The demographic characteristics of the patients in the studies are presented in Table 2, and methodological information about the included studies is presented in Table 3 and the Supplemental table.

Overall effect of treatment
The effect sizes for the included studies favoured the intervention over control conditions for the treatment of insight in psychosis. The calculated effect sizes and their associated 95% confidence intervals are plotted in Fig. 2.
The mean effect size for all the interventions was 0.441 (95% CI, 0.23-0.66), which corresponds to a medium effect size, and it was statistically significant (z = 4.03, p = 0.000).
The heterogeneity of the studies was high, with a calculated Higgins's I 2 of 80% and Tau squared of 0.18.

Psycho-education
Given the heterogeneity of the included studies, we grouped 5 studies that used psychoeducational interventions together. In  Medalia et al. (2012) and Chien and Thompson (2014), produces a medium, but non significant effect size of 0.608. Despite the fact that all 5 studies use a psychoeducational approach, they show a high degree of heterogeneity, Higgins's I 2 of 90% and Tau squared of 0.468, so these results should be interpreted with caution.
Pijnenborg et al. (2019)  multicentre RCT, conducted across 7 mental health care centres in the Netherlands. Although there was no specific advantage of the psychoeducation programme they used (REFLEX) against a control group receiving cognitive remediation therapy, they showed that patients in both groups had improved insight scores following treatment. It is  possible that the full effect of insight improvement is masked by the improvement also observed in the active control group. Medalia et al. (2012) investigated whether a single 45 minute session of interactive verbal and visual exercises along with psychoeducation affected insight, compared to a control group who watched sitcom reruns. The participants were outpatients at various clinics in New York City. There was no significant effect of that intervention on insight. Chan et al. (2007) trialled the psycho-education programme Transforming Relapse and Instilling Prosperity (TRIP), consisting of 10 wardbased sessions over 2 weeks, and conducted at Pamela Youde Nethersole Eastern Hospital (PYNEH) in Hong Kong. They reported a small but not significant improvement of insight in inpatients with schizophrenia, when compared to TAU. A later study by a different group in Hong Kong, Chan et al. (2009) involved 10 weekly sessions of psychoeducation of patients and their caregivers over 3 months, which produced a large and significant improvement of insight that was sustained at the 12 month follow up.
Finally Chien and Thompson (2014) conducted a study with outpatients from 3 different clinics in Hong Kong. They showed a very large and significant improvement of insight in the intervention group, who received a mindfulness-based psychoeducation programme for 6 months, compared to TAU, and an advantage over patients who received a conventional psychoeducation programme.

Cognitive behavioural therapy
Next we combined six studies that used a CBT or Metacognitive therapy approach and calculated their overall effect. In Fig. 4  The intervention group received 12 weekly CBT sessions for auditory hallucinations, whilst the active control group received 12 weekly enhanced support therapy, focusing on emotional support and counselling on problems that are not related to psychosis. Insight was a secondary outcome of the study, and although insight scores for the CBT group were higher at the end of treatment and at a one year follow up compared to the active control, that was not statistically significant. In a study conducted across 6 secondary care services in the UK, Turkington et al. (2002) compared an intervention group who received 3 to 6 hourly CBT sessions over 2 to 3 months to TAU. The effect size of insight improvement was small and not significant. Lam et al. (2015) compared an intervention of 8 sessions of group metacognitive therapy to a TAU control group in China. They reported a large and significant effect on the self-reflectiveness element of insight. In a later study conducted in Australia, Balzan et al. (2019) compared a similar intervention of 4 2-hourly sessions of individualized metacognitive therapy to a cognitive remediation control condition. Arguably the full effect of insight improvement is masked by the improvement also observed in the active control group, and the medium effect size reported did not reach statistical significance.

Combined treatment vs. medication alone
Guo et al., 2010 conducted a large scale randomised controlled trial to determine the efficacy of combined antipsychotic medication and psychosocial intervention versus antipsychotic medication alone. The psychosocial intervention consisted of psycho education, family intervention, skills training and cognitive-behavioural therapy. The participants of the study were all outpatients across 10 clinical sites in China, and received 12 4-hour monthly group sessions. The combined intervention produced a significant medium effect size of 0.683 (95% CI = 0.54-0.83).

Video self-observation
In two studies patients were filmed during a psychotic episode, which they watched after their symptoms had improved. The study by Davidoff et al. (1998)

was conducted at McLean Hospital in
Massachusetts. It reported a large but non-significant effect, compared to a control group who watched a comedy video instead. This may be partly due to the fact that there were only 9 patients in each group, and a rather poor control condition. In an improved design by David et al. (2012), the experimental group (N = 21) watched themselves experiencing psychotic symptoms, whilst the control group (N = 19) watched a video of someone else (an actor) experiencing a psychotic episode. This study showed a medium size, though not significant, effect on insight improvement.

Compliance therapy
To examine an intervention targeting compliance with medication, Kemp et al., 1998 allocated Maudsley Hospital inpatients randomly to receive 4 to 6 sessions of either individualized compliance therapy or standard supportive therapy. Patient insight was a secondary outcome measure in the trial and showed an improvement with a medium effect size of 0.505 (95% CI = − 0.07-1.08), which although was not significant, showed remarkable stability over an 18-month follow up in the community.

Humorous movies
Gelkopf et al., 2006 examined whether viewing of humorous movies improved various outcome measures of inpatients with schizophrenia in Israel. This was a follow up from earlier work, which demonstrated that humorous movies reduced patient anxiety, depression, aggression (Gelkopf et al., 1993), and improved social support (Gelkopf et al., 1994). The intervention group was compared to a control group who viewed the same number of movies, but with only 15% of them being  humorous. Whilst the intervention group showed significant improvements of other outcomes measures, patient insight remained relatively unaffected.

Physical exercise
A randomised controlled trial by Senormanci et al. (2021), involving patients with schizophrenia in a nursing home setting in Turkey, compared groups randomised to receive either two 60-minute exercise sessions per week, or regular inpatient care alone, TAU. Whilst the trial demonstrated improvements in other outcome measures, the overall effect on insight was small at 0.176 (95% CI = − 0.453-0.805).

Transcranial direct current stimulation (tDCS)
Chang et al., 2018 conducted a novel study in patients with medication-refractory auditory verbal hallucinations at the Tri-Service General Hospital in Taiwan. They compared an experimental group who received twice daily left fronto-temporal stimulation at an intensity of 2 mA for 20 min, on 5 consecutive weekdays to a control group that received sham stimulation (2 mA for 30 s of the 20 minute sham session). Although the auditory hallucinations did not improve, the improvement of insight was significant, of a very large magnitude, and lasted for at least a month after the intervention.

Discussion
The results of this meta-analysis show that specific interventions can have a significant overall effect on the level of insight in patients diagnosed with schizophrenia or other psychotic disorders. This confirmed the hypothesis that insight could be improved with appropriate interventions. There was a substantial variation in the overall effect size, regardless of the type of intervention, therefore the significance of these effect sizes should be interpreted with caution. Whilst the results favoured psychoeducational and CBT interventions, this may be partly due to the higher proportion of these types of studies being included with this analysis. Conversely, other interventions such as humorous movies, physical exercise, or compliance therapy, were only represented by a single study and therefore their overall effect size should also be interpreted with caution. In favour of psychoeducational interventions, a recent review focused exclusively on metacognitive interventions (Lopez-Morinigo et al., 2020), also reports improvements in cognitive insight in patients with schizophrenia spectrum disorders. A further point in favour of CBT, metacognitive therapy and other cognitive remediation interventions, such as Franck et al. (2013), was that the active control groups also received cognitive input that also had a therapeutic effect on anosognosia, thus narrowing the difference between the two groups and masking the full effect size of the intervention.
Despite the fact that some interventions were represented by a single study, the effect size, significance and longevity of insight improvement at follow up is worth noting. Guo et al. (2010) reported benefits of combined antipsychotic medication and psychosocial intervention on insight, alongside benefits such as increased engagement in education or employment. Chan et al. (2009) also reported benefits of their psychoeducation programme for patients and their caregivers, with improvements of insight sustained at the 12 month follow up. Chang et al. (2018) reported a RCT of 5 tDCS sessions that showed a very large effect of insight improvement compared to sham stimulation, and was still present at a 3-month follow up. This was the only study that employed a causal intervention at the level of specific brain areas. It confirmed a previous pilot result on insight improvement as a result of stimulation in schizophrenia (Bose et al., 2014), and is part of a wider research effort to establish non-invasive brain stimulation as an effective treatment for various mental conditions (Hyde et al., 2022).
In over 40% of the studies included within this meta-analysis insight was not the primary outcome measure. This is a feature that had been observed in similar, earlier systematic reviews, e.g. Pijnenborg et al. (2013), and the implication is that our results are likely to be an over representation of the interventions that were designed to improve patient insight. This raises the question as to why there is a lack of studies focusing on improving insight. One possible reason for this could be the conceptual and subjective nature of insight and its assessment (Reddy, 2016). Additionally, there may be some debate as to whether an improvement of insight follows standard pharmacological treatment for psychosis. Therefore, a lack of a mechanistic understanding as to what gives rise to a lack of insight may inhibit clinical and applied research or interventions in this area.
There was a lack of pharmacological studies included in our systematic review. This was the result of the age selection criterion of our protocol, which was between the ages of 18 and 70. Most pharmacological studies included patients up to the age of 85. The age criterion of up to 70 (or earlier) is common in both primary studies and reviews on this topic, in order to reduce uncontrolled variability introduced by changes in the structure and function of the brain, e.g. Cohen et al. (2019), as well as specific concerns about the cognitive function of older adults with schizophrenia, e.g. Murante and Cohen (2017). It would be worthwhile to have a separate more inclusive systematic review, so that the effect of medication, such as long-acting risperidone, on insight can also be assessed, e.g. Gharabawi et al. (2006) and Emsley et al. (2008), including in older adults.
A limitation associated with most of the included studies, and therefore this review, was the relatively small scale of trials. The implications of this are that the studies may lack statistical power and may not be representative of the target patient population. A further concern is that relatively higher cognitive functioning patients are more likely to benefit from psychoeducation, CBT and similar interventions, which creates a selection bias. This should be taken into account when recruiting participants, reporting patient demographics and results in primary studies. For example, some of the studies included here reported exclusion criteria for patients with IQ under 70, and most studies reported years of education as a proxy for level of cognitive function. The small group sample sizes may have also contributed to the variance in some of the studies. An additional factor of variance, although also a strength, is the variety of countries and cultural contexts in which the studies have been conducted. There is a clear international interest in this topic, and careful adaptation of assessment scales, psychoeducational and therapy materials is crucial both for efficacy in treating patients, and for the ability to pool together and compare results.
A final consideration for future studies is the scope of different interventions for insight improvement for different levels of illness severity. The engagement required from interventions such as psychoeducation, CBT and meta-cognitive therapy mean that patients need to be well enough for long periods of time to fully benefit from them. Potential benefits of interventions such as video self-observation (David et al., 2012) and fronto-temporal stimulation (Chang et al., 2018) in more acute phases of the illness need to be further explored.

Conclusion
Despite the small number of studies and variability of interventions assessed, it is possible to improve insight through interventions. Insight therefore remains a suitable target for therapy. The approaches that yielded a significant effect were CBT and metacognitive therapy, tDCS, as well as combined pharmacological and psychosocial, with psychoeducation programmes yielding mixed results. tDCS in particular needs to be further explored as a highly promising approach. There is scope and potential for a variety of interventions to improve insight over and above treatment as usual. It is necessary to continue further large-scale research on this topic, to improve our understanding of insight and its relationship to the pathophysiology of disease.

Role of funding source
This work was funded by the Brighton and Sussex Medical School. The funding source did not have any role in determining the content of the manuscript.

CRediT authorship contribution statement
NS conceived the study, SP and NS carried out the data collection, bias analysis, statistical analyses and interpretation. SP wrote the first draft, and SP and NS revised and approved the final draft.