https://orcid.org/0000-0001-6311-2796
Abstract
Impaired insight into the illness and its consequences is associated with poor outcomes in schizophrenia. While transcranial direct current stimulation (tDCS) may represent a potentially effective treatment strategy to relieve various symptoms of schizophrenia, its impact on insight remains unclear. To investigate whether tDCS would modulate insight in patients with schizophrenia, we undertook a meta-analysis based on results from previous RCTs that investigated the clinical efficacy of tDCS. We hypothesize that repeated sessions of tDCS will be associated with insight improvement among patients.
PubMed and ScienceDirect databases were systematically searched to identify RCTs that delivered at least 10 tDCS sessions in patients with schizophrenia. The primary outcome was the change in insight score, assessed by the Positive and Negative Syndrome Scale (PANSS) item G12 following active tDCS sessions as opposed to sham stimulation. Effect sizes were calculated for all studies and pooled using a random-effects model. Meta-regression and subgroup analyses were conducted.
Thirteen studies (587 patients with schizophrenia) were included. A significant pooled effect size (g) of −0.46 (95% CI [−0.78; −0.14]) in favor of active tDCS was observed. Age and G12 score at baseline were identified as significant moderators, while change in total PANSS score was not significant.
Ten sessions of active tDCS with either frontotemporoparietal or bifrontal montage may improve insight into the illness in patients with schizophrenia. The effect of this treatment could contribute to the beneficial outcomes observed in patients following stimulation.
Introduction
Lack of patient insight into the illness is a key characteristic of many psychiatric disorders, including bipolar disorder,1 obsessive-compulsive disorder,2 and dementias.3 Among patients with schizophrenia, the estimated prevalence of poor insight, corresponding to a general unawareness of illness, can be up to 50%.4,5 Insight deficit includes a lack of awareness of the symptoms of the illness, the need for and willingness to undergo treatment, and the consequences of the illness on the patient’s life.6 Poor insight has been regularly associated with negative attitudes toward medication7 and nonadherence to antipsychotic treatment.8–11 Treatment nonadherence in schizophrenia represents a heavy economic cost, estimated between $13.92 and $18.36 million a year in the United States,12 because it increases the risk of relapse, rehospitalization rates, and suicide attempts.12,13 Poor insight is directly associated with a higher risk of relapse,14 which further emphasizes the importance of fostering better insight into the illness. Unfortunately, available treatments such as antipsychotic medication and psychotherapy provide either little or no improvement in patients’ insight into their illness.15
Transcranial direct current stimulation (tDCS) is a safe, low-cost tool for modulating the activity and connectivity of targeted brain regions and related neural networks.16,17 tDCS consists of applying weak electric currents through two electrodes placed over the scalp, with polarity-dependent effects on cortical excitability: Currents entering the brain at the anode site are thought to increase cortical excitability while currents exiting the brain at the cathode site are thought to decrease cortical excitability.18 For clinical purposes, repeated sessions of tDCS show promising results in alleviating some of the symptoms of schizophrenia and may also improve cognition in these patients.19 The means by which tDCS alleviates symptoms remain unclear, and one can hypothesize that a beneficial effect on metacognitive processes, and especially on insight capacities, could be considered to explain tDCS’s clinical effect because insight impairments have been associated with alteration in the activity of brain regions regularly targeted by tDCS (frontal and temporoparietal regions for example).20 Nevertheless, to date, only 5 articles from 3 different randomized controlled trials (RCTs) have studied the effect of tDCS on insight in patients with schizophrenia using different scales21–25; thus, drawing firm conclusions remains difficult.26 An underutilized source of informative data on the impact of tDCS on insight is a widely used psychiatric assessment scale27: the Positive and Negative Syndrome Scale (PANSS) and its item G12.28 This item assesses the lack of judgment and insight into the illness using a 7-level scale (ranging from 1, no impairment, to 7, extreme lack of insight). Gathering these often-unexamined data would provide a new source of evidence in establishing the potential benefits of treatment with tDCS on insight.
Because insight deficits are a major problem in the management of patients with schizophrenia, we conducted a systematic review and meta-analysis of randomized, sham-controlled clinical trials to evaluate the association between tDCS treatment and improvement in PANSS G12 scores in patients with schizophrenia. We hypothesized that, compared with sham tDCS, active tDCS will be associated with a decrease in G12 scores, which corresponds to an improvement in patient insight.
Methods
The methodology followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) reporting guidelines.29 The protocol was registered on the PROSPERO database (CRD42021243147).
Literature Search and Selection Criteria
Two authors (O.A. and M.M.) searched independently for articles in the PubMed and ScienceDirect databases with no restriction of date for eligible studies using the following keywords: (“tDCS” OR “transcranial direct current stimulation”) AND (“PANSS” OR “positive and negative syndrome scale”) AND “schizophrenia”. The search was conducted between November 2020 and March 2021. In ScienceDirect, research was refined using research articles, book chapters, and other limits. Selected studies met the following inclusion criteria: (1) inclusion of patients with schizophrenia or schizoaffective disorder; (2) comparison between active and sham stimulation; and (3) assessment of symptoms before and after tDCS sessions using the PANSS. The selection procedure was repeated on the full text of eligible studies. We excluded studies that were not RCTs, that combined tDCS with a therapy other than antipsychotic treatments, and that did not offer at least 10 sessions of tDCS. This choice was based on three recent meta-analyses that found a significantly greater clinical effect of tDCS in studies that delivered at least 10 sessions compared to those that delivered fewer than 10 sessions.30–32
Data Extraction and Main Outcome
Age, sex, diagnosis, illness duration, antipsychotic dose, PANSS total scores and G12 score at baseline, stimulation parameters (including the electrode placements), sham, and blinding procedures were extracted from included studies. Authors who reported using the PANSS were contacted to provide the mean change in G12 score, defined as the score assessed immediately after completion of tDCS treatment minus the score assessed before the treatment, that was our main outcome. Additionally, they were asked to check extracted data from articles for accuracy.
Quality and Risk of Bias Assessments
The quality of the included studies was measured independently by one author (OA) and by the contacted authors of each study using the Jadad scale.33 These two scores were averaged to assess overall quality.
Risk of bias was evaluated independently by two authors (OA and MB) using a Traffic Light plot and a Summary plot,34 constructed in accordance with the guidelines from the Agency for Healthcare Research and Quality.35
Statistical Analysis
All statistical analyses were performed with R software, version 4.1.2 (R Core Team, 2021), using the metafor, meta, and dmetar packages. The significant α level was set at 0.05. Hedges’ g effect size (ESg) along with the 95% confidence interval was computed based on mean (SD) scores of G12. The overall ESg was pooled from each study using a random-effects model and was interpreted according to Cohen’s guidelines (0.2 = small; 0.5 = medium; 0.8 = large effect).36 Global heterogeneity significance was calculated from Cochran’s Q-test and quantified using the I² and τ² statistics.37
To explore the causes of the heterogeneity, meta-regression analyses were conducted individually for each moderator, such as electrode montage (studies were grouped into bifrontal montage when both electrodes were placed over prefrontal areas at F3, F4, or FP2, and frontotemporoparietal montage when the anode was placed over prefrontal areas at F3–FP1, F3, or F4 and the cathode over temporoparietal areas at T3–P3), current density (intensity divided by electrode size), total charge per session (intensity by session duration) and per tDCS regimen (intensity by session duration by number of sessions), G12 score at baseline, changes in PANSS total score, PANSS total score at baseline, age of the patient, illness duration, and antipsychotic dose. Age and illness duration can impact insight in patients20,38,39 and antipsychotic dose may influence the response to tDCS.40 In addition, we proceeded to exploratory separate subgroup analyses in cases of significance after FDR correction for multiple tests. Because no clear hypothesis on the relationship between a specific age and response to tDCS or insight could be generated from the literature, we divided studies into one group with a below-average age (< 40 years) and one group with an above-average age (≥40 years), based on the mean age of participants included in the studies combined (mean split of age). Concerning G12 scores, we divided studies into a group with, on average, an absence of, or minimal lack of insight (<3); and a group with, on average, at least a mild lack of insight (≥3). This division was based on the hypothesis that there would have to be an already marked impairment of insight for tDCS to influence it.
Finally, publication bias was assessed by visual examination of the funnel plot and using Egger’s regression test. Potential outlier studies were controlled by visual examination of QQ plots and Baujat plots. Meta-outlier detection and leave-one-out meta-analyses were conducted to investigate the effect of potential outliers on the pooled effect size.
Results
Search Results
The initial search strategy identified 116 records. After the removal of duplicate studies and the exclusion of ineligible studies, 21 full texts were assessed, of which 17 met our inclusion criteria (figure 1). Among them, 3 studies presented the same sample21,24,25 and 2 studies had partially overlapping samples41,42; in these cases, we chose the record with the largest sample size. One eligible study was excluded because the data could not be provided upon request.43 A total of 13 RCTs from 12 independent groups were included in our meta-analysis.21,23,41,44–53
PRISMA 2020 flow diagram of included studies.
Abbreviations: RCTs, randomized controlled trials; tDCS, transcranial direct current stimulation.
Characteristics of Studies
These studies included 587 participants, 297 of whom received active tDCS and 290 of whom received sham tDCS. Details of these studies are provided in tables 1 and 2 and in supplementary appendix 1. All included RCTs showed a high range of quality scores (scores ≥ 3) on the Jadad scale, except for 1 study which had a low-quality score (score = 2).52 Visual inspection of the traffic light plot revealed that most studies showed a low overall risk of bias and that only 2 studies46,47 showed a potential moderate risk of bias (supplementary figure 1). The main risk of bias was the bias arising from the randomization process (supplementary figure 2).
Sociodemographic and clinical characteristics of included studies
| Authors, Year | Jadad Scorea | Diagnosis | Treatment Condition | n | Age (y) | Femae, n (%) | Illness Duration (y) | Antipsychotic Dose (CPZE, mg/day) | PANSS Total Score at Baseline | G12 score at Baseline | Mean Changes in Total PANSS Score After tDCS Treatment | Mean Changes in G12 Score After tDCS Treatment |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | 3 | SZ | Active | 15 | 40.4 (9.9) | 3 (20) | NR | 994 (714) | 76.9 (16.4) | 2.3 (1.4) | −9.2 (8.1) | −0.53 (1.33) |
| Sham | 15 | 35.1 (7.0) | 5 (33) | NR | 1209 (998) | 82.8 (15.4) | 2.2 (1.2) | −2.3 (10.1) | −0.13 (0.83) | |||
| Palm et al., 2016 | 5 | SZ | Active | 10 | 38.4 (12.9) | 5 (50) | 7.1 (6.1) | 559 (304) | 79.5 (20.0) | 2.7 (1.2) | −14.2 (21.1) | −0.333 (1.00) |
| Sham | 10 | 34.1 (10.7) | 0 (0) | 13.8 (12.1) | 481 (226) | 85.6 (6.8) | 2.4 (1.4) | 0.8 (7.1) | −0.125 (0.354) | |||
| Chang et al., 2018 | 4 | SZ - SZAff | Active | 30 | 46.4 (10.3) | 16 (53) | 19.7 (10.4) | 494 (307) | 72.3 (13.2) | 4.2 (1.2) | −2.6 (3.8) | −0.87 (1.14) |
| Sham | 30 | 42.2 (10.3) | 17 (57) | 13.9 (7.5) | 493 (284) | 66.7 (12.5) | 3.7 (0.8) | −0.3 (1.2) | −0.03 (0.18) | |||
| Gomes et al., 2018 | 5 | SZ | Active | 12 | 39.2 (9.3) | 2 (17) | 16.0 (11.6) | NR | 81.6 (16.0) | 3.1 (1.2) | −10.7 (6.1) | −0.42 (0.90) |
| Sham | 12 | 33.7 (12.1) | 5 (42) | 10.0 (7.3) | NR | 71.0 (19.9) | 3.3 (1.4) | −0.8 (6.1) | −0.42 (1.08) | |||
| Jeon et al., 2018 | 5 | SZ | Active | 26 | 40.0 (9.4) | 13 (50) | 12.8 (9.8) | 546 (403) | 85.8 (21.1) | 3.6 (1.3) | −13.8 (27.1) | 0.07 (0.48) |
| Sham | 28 | 39.9 (12.4) | 15 (54) | 14.3 (10.5) | 614 (430) | 79.1 (16.6) | 3.4 (1.3) | −7.0 (19.6) | 0.25 (0.80) | |||
| Koops et al., 2018 | 4 | Mixed sampleb | Active | 28 | 44.0 (11) | 14 (50) | NR | NR | 68.0 (16.0) | 2.7 (1.3) | −5.4 (8.1) | −0.32 (0.94) |
| Sham | 26 | 44.0 (12) | 15 (58) | NR | NR | 64.0 (17.0) | 3.5 (1.3) | −2.2 (7.5) | 0.08 (0.81) | |||
| Mellin et al., 2018 | 4 | SZ - SZAff | Active | 7 | 29.6 (11.0) | 3 (43) | 6.4 | NR | 58.9 (14.7) | 2.9 (2.3) | −4.1 (3.7) | 0.14 (0.38) |
| Sham | 7 | 38.9 (10.0) | 3 (43) | 14.3 | NR | 57.0 (11.6) | 2.6 (1.5) | −1.7 (4.4) | 0.71 (1.11) | |||
| Kantrowitz et al., 2019 | 5 | SZ - SZAff | Active | 47 | 38.2 (9.9) | 15 (32) | NR | 806 (768) | 75.3 (12.9) | 3.3 (1.4) | −3.3 (9.4) | −0.149 (1.23) |
| Sham | 42 | 40.1 (8.6) | 7 (17) | NR | 628 (466) | 73.2 (14.4) | 3.0 (1.3) | −2.1 (9.0) | 0.075 (1.28) | |||
| Lindenmayer et al., 2019 | 4 | SZ - SZAff | Active | 12 | 41.4 (11.3) | 2 (17) | NR | 958 (364) | 81.5 (7.5) | 3.6 (1.1) | −8.2 (7.3) | 0.083 (0.515) |
| Sham | 13 | 39.9 (10.5) | 2 (15) | NR | 896 (275) | 82.8 (9.4) | 3.6 (1.2) | −2.3 (6.6) | 0.077 (0.760) | |||
| Weickert et al., 2019 | 4 | SZ - SZAff | Active | 6 | 45.5 (1.9) | 4 (67) | 22 (0.9) | 555 (249) | 59.5 (NR) | 2.5 (1.6) | −5.0 (5.7) | −1.6 (1.5) |
| Sham | 6 | 31.3 (3.6) | 2 (33) | 8.3 (3.4) | 1009 (261) | 65.3 (NR) | 3.0 (1.3) | −0.5 (15.9) | −0.6 (0.9) | |||
| Chang et al., 2020 | 5 | SZ - SZAff | Active | 30 | 44.7 (10.7) | 11 (37) | 14.7 (9.5) | 532 (348) | 67.4 (13.0) | 4.5 (1.0) | −10.8 (7.1) | −2.00 (1.17) |
| Sham | 30 | 45.0 (10.9) | 19 (63) | 15.8 (10.6) | 523 (332) | 73.3 (10.3) | 4.7 (0.9) | −1.4 (4.4) | −0.23 (0.50) | |||
| Smith et al., 2020 | 2 | SZ | Active | 24 | 43.7 (14.1) | 15 (62) | 17.9 (11.7) | NR | 62.0 (15.6) | 3.6 (1.2) | −3.3 (13.8) | −0.381 (0.921) |
| Sham | 21 | 47.1 (11.9) | 12 (57) | 19.7 (11.1) | NR | 58.7 (9.9) | 3.7 (1.2) | −4.4 (9.9) | 0.0833 (1.06) | |||
| Valiengo et al., 2020 | 5 | SZ | Active | 50 | 34.6 (8.4) | 9 (18) | 14.2 (8.1) | 475 (225) | 73.6 (15.8) | 2.8 (1.4) | −5.3 (10.2) | 0 (1.30) |
| Sham | 50 | 35.9 (10.1) | 11 (22) | 14.1 (8.7) | 500 (400) | 73.9 (13.4) | 3.0 (1.4) | −3.1 (7.9) | −0.24 (1.25) |
| Authors, Year | Jadad Scorea | Diagnosis | Treatment Condition | n | Age (y) | Femae, n (%) | Illness Duration (y) | Antipsychotic Dose (CPZE, mg/day) | PANSS Total Score at Baseline | G12 score at Baseline | Mean Changes in Total PANSS Score After tDCS Treatment | Mean Changes in G12 Score After tDCS Treatment |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | 3 | SZ | Active | 15 | 40.4 (9.9) | 3 (20) | NR | 994 (714) | 76.9 (16.4) | 2.3 (1.4) | −9.2 (8.1) | −0.53 (1.33) |
| Sham | 15 | 35.1 (7.0) | 5 (33) | NR | 1209 (998) | 82.8 (15.4) | 2.2 (1.2) | −2.3 (10.1) | −0.13 (0.83) | |||
| Palm et al., 2016 | 5 | SZ | Active | 10 | 38.4 (12.9) | 5 (50) | 7.1 (6.1) | 559 (304) | 79.5 (20.0) | 2.7 (1.2) | −14.2 (21.1) | −0.333 (1.00) |
| Sham | 10 | 34.1 (10.7) | 0 (0) | 13.8 (12.1) | 481 (226) | 85.6 (6.8) | 2.4 (1.4) | 0.8 (7.1) | −0.125 (0.354) | |||
| Chang et al., 2018 | 4 | SZ - SZAff | Active | 30 | 46.4 (10.3) | 16 (53) | 19.7 (10.4) | 494 (307) | 72.3 (13.2) | 4.2 (1.2) | −2.6 (3.8) | −0.87 (1.14) |
| Sham | 30 | 42.2 (10.3) | 17 (57) | 13.9 (7.5) | 493 (284) | 66.7 (12.5) | 3.7 (0.8) | −0.3 (1.2) | −0.03 (0.18) | |||
| Gomes et al., 2018 | 5 | SZ | Active | 12 | 39.2 (9.3) | 2 (17) | 16.0 (11.6) | NR | 81.6 (16.0) | 3.1 (1.2) | −10.7 (6.1) | −0.42 (0.90) |
| Sham | 12 | 33.7 (12.1) | 5 (42) | 10.0 (7.3) | NR | 71.0 (19.9) | 3.3 (1.4) | −0.8 (6.1) | −0.42 (1.08) | |||
| Jeon et al., 2018 | 5 | SZ | Active | 26 | 40.0 (9.4) | 13 (50) | 12.8 (9.8) | 546 (403) | 85.8 (21.1) | 3.6 (1.3) | −13.8 (27.1) | 0.07 (0.48) |
| Sham | 28 | 39.9 (12.4) | 15 (54) | 14.3 (10.5) | 614 (430) | 79.1 (16.6) | 3.4 (1.3) | −7.0 (19.6) | 0.25 (0.80) | |||
| Koops et al., 2018 | 4 | Mixed sampleb | Active | 28 | 44.0 (11) | 14 (50) | NR | NR | 68.0 (16.0) | 2.7 (1.3) | −5.4 (8.1) | −0.32 (0.94) |
| Sham | 26 | 44.0 (12) | 15 (58) | NR | NR | 64.0 (17.0) | 3.5 (1.3) | −2.2 (7.5) | 0.08 (0.81) | |||
| Mellin et al., 2018 | 4 | SZ - SZAff | Active | 7 | 29.6 (11.0) | 3 (43) | 6.4 | NR | 58.9 (14.7) | 2.9 (2.3) | −4.1 (3.7) | 0.14 (0.38) |
| Sham | 7 | 38.9 (10.0) | 3 (43) | 14.3 | NR | 57.0 (11.6) | 2.6 (1.5) | −1.7 (4.4) | 0.71 (1.11) | |||
| Kantrowitz et al., 2019 | 5 | SZ - SZAff | Active | 47 | 38.2 (9.9) | 15 (32) | NR | 806 (768) | 75.3 (12.9) | 3.3 (1.4) | −3.3 (9.4) | −0.149 (1.23) |
| Sham | 42 | 40.1 (8.6) | 7 (17) | NR | 628 (466) | 73.2 (14.4) | 3.0 (1.3) | −2.1 (9.0) | 0.075 (1.28) | |||
| Lindenmayer et al., 2019 | 4 | SZ - SZAff | Active | 12 | 41.4 (11.3) | 2 (17) | NR | 958 (364) | 81.5 (7.5) | 3.6 (1.1) | −8.2 (7.3) | 0.083 (0.515) |
| Sham | 13 | 39.9 (10.5) | 2 (15) | NR | 896 (275) | 82.8 (9.4) | 3.6 (1.2) | −2.3 (6.6) | 0.077 (0.760) | |||
| Weickert et al., 2019 | 4 | SZ - SZAff | Active | 6 | 45.5 (1.9) | 4 (67) | 22 (0.9) | 555 (249) | 59.5 (NR) | 2.5 (1.6) | −5.0 (5.7) | −1.6 (1.5) |
| Sham | 6 | 31.3 (3.6) | 2 (33) | 8.3 (3.4) | 1009 (261) | 65.3 (NR) | 3.0 (1.3) | −0.5 (15.9) | −0.6 (0.9) | |||
| Chang et al., 2020 | 5 | SZ - SZAff | Active | 30 | 44.7 (10.7) | 11 (37) | 14.7 (9.5) | 532 (348) | 67.4 (13.0) | 4.5 (1.0) | −10.8 (7.1) | −2.00 (1.17) |
| Sham | 30 | 45.0 (10.9) | 19 (63) | 15.8 (10.6) | 523 (332) | 73.3 (10.3) | 4.7 (0.9) | −1.4 (4.4) | −0.23 (0.50) | |||
| Smith et al., 2020 | 2 | SZ | Active | 24 | 43.7 (14.1) | 15 (62) | 17.9 (11.7) | NR | 62.0 (15.6) | 3.6 (1.2) | −3.3 (13.8) | −0.381 (0.921) |
| Sham | 21 | 47.1 (11.9) | 12 (57) | 19.7 (11.1) | NR | 58.7 (9.9) | 3.7 (1.2) | −4.4 (9.9) | 0.0833 (1.06) | |||
| Valiengo et al., 2020 | 5 | SZ | Active | 50 | 34.6 (8.4) | 9 (18) | 14.2 (8.1) | 475 (225) | 73.6 (15.8) | 2.8 (1.4) | −5.3 (10.2) | 0 (1.30) |
| Sham | 50 | 35.9 (10.1) | 11 (22) | 14.1 (8.7) | 500 (400) | 73.9 (13.4) | 3.0 (1.4) | −3.1 (7.9) | −0.24 (1.25) |
Note Otherwise specified, results are given as mean (SD).
Jadad scores correspond to the average of the scores given by the author OA and by the main author of the associated article.
Koops et al. (2018) included 21 patients with SZ, 4 patients with psychosis NOS, 1 patient with schizoaffective disorder, 1 patient with affective disorder, and 1 patient with borderline personality disorder.
Abbreviations: CPZE, chlorpromazine equivalent; NR, not reported; PANSS, positive and negative syndrome scale; SZ, schizophrenia; SZAff, schizoaffective disorder.
Sociodemographic and clinical characteristics of included studies
| Authors, Year | Jadad Scorea | Diagnosis | Treatment Condition | n | Age (y) | Femae, n (%) | Illness Duration (y) | Antipsychotic Dose (CPZE, mg/day) | PANSS Total Score at Baseline | G12 score at Baseline | Mean Changes in Total PANSS Score After tDCS Treatment | Mean Changes in G12 Score After tDCS Treatment |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | 3 | SZ | Active | 15 | 40.4 (9.9) | 3 (20) | NR | 994 (714) | 76.9 (16.4) | 2.3 (1.4) | −9.2 (8.1) | −0.53 (1.33) |
| Sham | 15 | 35.1 (7.0) | 5 (33) | NR | 1209 (998) | 82.8 (15.4) | 2.2 (1.2) | −2.3 (10.1) | −0.13 (0.83) | |||
| Palm et al., 2016 | 5 | SZ | Active | 10 | 38.4 (12.9) | 5 (50) | 7.1 (6.1) | 559 (304) | 79.5 (20.0) | 2.7 (1.2) | −14.2 (21.1) | −0.333 (1.00) |
| Sham | 10 | 34.1 (10.7) | 0 (0) | 13.8 (12.1) | 481 (226) | 85.6 (6.8) | 2.4 (1.4) | 0.8 (7.1) | −0.125 (0.354) | |||
| Chang et al., 2018 | 4 | SZ - SZAff | Active | 30 | 46.4 (10.3) | 16 (53) | 19.7 (10.4) | 494 (307) | 72.3 (13.2) | 4.2 (1.2) | −2.6 (3.8) | −0.87 (1.14) |
| Sham | 30 | 42.2 (10.3) | 17 (57) | 13.9 (7.5) | 493 (284) | 66.7 (12.5) | 3.7 (0.8) | −0.3 (1.2) | −0.03 (0.18) | |||
| Gomes et al., 2018 | 5 | SZ | Active | 12 | 39.2 (9.3) | 2 (17) | 16.0 (11.6) | NR | 81.6 (16.0) | 3.1 (1.2) | −10.7 (6.1) | −0.42 (0.90) |
| Sham | 12 | 33.7 (12.1) | 5 (42) | 10.0 (7.3) | NR | 71.0 (19.9) | 3.3 (1.4) | −0.8 (6.1) | −0.42 (1.08) | |||
| Jeon et al., 2018 | 5 | SZ | Active | 26 | 40.0 (9.4) | 13 (50) | 12.8 (9.8) | 546 (403) | 85.8 (21.1) | 3.6 (1.3) | −13.8 (27.1) | 0.07 (0.48) |
| Sham | 28 | 39.9 (12.4) | 15 (54) | 14.3 (10.5) | 614 (430) | 79.1 (16.6) | 3.4 (1.3) | −7.0 (19.6) | 0.25 (0.80) | |||
| Koops et al., 2018 | 4 | Mixed sampleb | Active | 28 | 44.0 (11) | 14 (50) | NR | NR | 68.0 (16.0) | 2.7 (1.3) | −5.4 (8.1) | −0.32 (0.94) |
| Sham | 26 | 44.0 (12) | 15 (58) | NR | NR | 64.0 (17.0) | 3.5 (1.3) | −2.2 (7.5) | 0.08 (0.81) | |||
| Mellin et al., 2018 | 4 | SZ - SZAff | Active | 7 | 29.6 (11.0) | 3 (43) | 6.4 | NR | 58.9 (14.7) | 2.9 (2.3) | −4.1 (3.7) | 0.14 (0.38) |
| Sham | 7 | 38.9 (10.0) | 3 (43) | 14.3 | NR | 57.0 (11.6) | 2.6 (1.5) | −1.7 (4.4) | 0.71 (1.11) | |||
| Kantrowitz et al., 2019 | 5 | SZ - SZAff | Active | 47 | 38.2 (9.9) | 15 (32) | NR | 806 (768) | 75.3 (12.9) | 3.3 (1.4) | −3.3 (9.4) | −0.149 (1.23) |
| Sham | 42 | 40.1 (8.6) | 7 (17) | NR | 628 (466) | 73.2 (14.4) | 3.0 (1.3) | −2.1 (9.0) | 0.075 (1.28) | |||
| Lindenmayer et al., 2019 | 4 | SZ - SZAff | Active | 12 | 41.4 (11.3) | 2 (17) | NR | 958 (364) | 81.5 (7.5) | 3.6 (1.1) | −8.2 (7.3) | 0.083 (0.515) |
| Sham | 13 | 39.9 (10.5) | 2 (15) | NR | 896 (275) | 82.8 (9.4) | 3.6 (1.2) | −2.3 (6.6) | 0.077 (0.760) | |||
| Weickert et al., 2019 | 4 | SZ - SZAff | Active | 6 | 45.5 (1.9) | 4 (67) | 22 (0.9) | 555 (249) | 59.5 (NR) | 2.5 (1.6) | −5.0 (5.7) | −1.6 (1.5) |
| Sham | 6 | 31.3 (3.6) | 2 (33) | 8.3 (3.4) | 1009 (261) | 65.3 (NR) | 3.0 (1.3) | −0.5 (15.9) | −0.6 (0.9) | |||
| Chang et al., 2020 | 5 | SZ - SZAff | Active | 30 | 44.7 (10.7) | 11 (37) | 14.7 (9.5) | 532 (348) | 67.4 (13.0) | 4.5 (1.0) | −10.8 (7.1) | −2.00 (1.17) |
| Sham | 30 | 45.0 (10.9) | 19 (63) | 15.8 (10.6) | 523 (332) | 73.3 (10.3) | 4.7 (0.9) | −1.4 (4.4) | −0.23 (0.50) | |||
| Smith et al., 2020 | 2 | SZ | Active | 24 | 43.7 (14.1) | 15 (62) | 17.9 (11.7) | NR | 62.0 (15.6) | 3.6 (1.2) | −3.3 (13.8) | −0.381 (0.921) |
| Sham | 21 | 47.1 (11.9) | 12 (57) | 19.7 (11.1) | NR | 58.7 (9.9) | 3.7 (1.2) | −4.4 (9.9) | 0.0833 (1.06) | |||
| Valiengo et al., 2020 | 5 | SZ | Active | 50 | 34.6 (8.4) | 9 (18) | 14.2 (8.1) | 475 (225) | 73.6 (15.8) | 2.8 (1.4) | −5.3 (10.2) | 0 (1.30) |
| Sham | 50 | 35.9 (10.1) | 11 (22) | 14.1 (8.7) | 500 (400) | 73.9 (13.4) | 3.0 (1.4) | −3.1 (7.9) | −0.24 (1.25) |
| Authors, Year | Jadad Scorea | Diagnosis | Treatment Condition | n | Age (y) | Femae, n (%) | Illness Duration (y) | Antipsychotic Dose (CPZE, mg/day) | PANSS Total Score at Baseline | G12 score at Baseline | Mean Changes in Total PANSS Score After tDCS Treatment | Mean Changes in G12 Score After tDCS Treatment |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | 3 | SZ | Active | 15 | 40.4 (9.9) | 3 (20) | NR | 994 (714) | 76.9 (16.4) | 2.3 (1.4) | −9.2 (8.1) | −0.53 (1.33) |
| Sham | 15 | 35.1 (7.0) | 5 (33) | NR | 1209 (998) | 82.8 (15.4) | 2.2 (1.2) | −2.3 (10.1) | −0.13 (0.83) | |||
| Palm et al., 2016 | 5 | SZ | Active | 10 | 38.4 (12.9) | 5 (50) | 7.1 (6.1) | 559 (304) | 79.5 (20.0) | 2.7 (1.2) | −14.2 (21.1) | −0.333 (1.00) |
| Sham | 10 | 34.1 (10.7) | 0 (0) | 13.8 (12.1) | 481 (226) | 85.6 (6.8) | 2.4 (1.4) | 0.8 (7.1) | −0.125 (0.354) | |||
| Chang et al., 2018 | 4 | SZ - SZAff | Active | 30 | 46.4 (10.3) | 16 (53) | 19.7 (10.4) | 494 (307) | 72.3 (13.2) | 4.2 (1.2) | −2.6 (3.8) | −0.87 (1.14) |
| Sham | 30 | 42.2 (10.3) | 17 (57) | 13.9 (7.5) | 493 (284) | 66.7 (12.5) | 3.7 (0.8) | −0.3 (1.2) | −0.03 (0.18) | |||
| Gomes et al., 2018 | 5 | SZ | Active | 12 | 39.2 (9.3) | 2 (17) | 16.0 (11.6) | NR | 81.6 (16.0) | 3.1 (1.2) | −10.7 (6.1) | −0.42 (0.90) |
| Sham | 12 | 33.7 (12.1) | 5 (42) | 10.0 (7.3) | NR | 71.0 (19.9) | 3.3 (1.4) | −0.8 (6.1) | −0.42 (1.08) | |||
| Jeon et al., 2018 | 5 | SZ | Active | 26 | 40.0 (9.4) | 13 (50) | 12.8 (9.8) | 546 (403) | 85.8 (21.1) | 3.6 (1.3) | −13.8 (27.1) | 0.07 (0.48) |
| Sham | 28 | 39.9 (12.4) | 15 (54) | 14.3 (10.5) | 614 (430) | 79.1 (16.6) | 3.4 (1.3) | −7.0 (19.6) | 0.25 (0.80) | |||
| Koops et al., 2018 | 4 | Mixed sampleb | Active | 28 | 44.0 (11) | 14 (50) | NR | NR | 68.0 (16.0) | 2.7 (1.3) | −5.4 (8.1) | −0.32 (0.94) |
| Sham | 26 | 44.0 (12) | 15 (58) | NR | NR | 64.0 (17.0) | 3.5 (1.3) | −2.2 (7.5) | 0.08 (0.81) | |||
| Mellin et al., 2018 | 4 | SZ - SZAff | Active | 7 | 29.6 (11.0) | 3 (43) | 6.4 | NR | 58.9 (14.7) | 2.9 (2.3) | −4.1 (3.7) | 0.14 (0.38) |
| Sham | 7 | 38.9 (10.0) | 3 (43) | 14.3 | NR | 57.0 (11.6) | 2.6 (1.5) | −1.7 (4.4) | 0.71 (1.11) | |||
| Kantrowitz et al., 2019 | 5 | SZ - SZAff | Active | 47 | 38.2 (9.9) | 15 (32) | NR | 806 (768) | 75.3 (12.9) | 3.3 (1.4) | −3.3 (9.4) | −0.149 (1.23) |
| Sham | 42 | 40.1 (8.6) | 7 (17) | NR | 628 (466) | 73.2 (14.4) | 3.0 (1.3) | −2.1 (9.0) | 0.075 (1.28) | |||
| Lindenmayer et al., 2019 | 4 | SZ - SZAff | Active | 12 | 41.4 (11.3) | 2 (17) | NR | 958 (364) | 81.5 (7.5) | 3.6 (1.1) | −8.2 (7.3) | 0.083 (0.515) |
| Sham | 13 | 39.9 (10.5) | 2 (15) | NR | 896 (275) | 82.8 (9.4) | 3.6 (1.2) | −2.3 (6.6) | 0.077 (0.760) | |||
| Weickert et al., 2019 | 4 | SZ - SZAff | Active | 6 | 45.5 (1.9) | 4 (67) | 22 (0.9) | 555 (249) | 59.5 (NR) | 2.5 (1.6) | −5.0 (5.7) | −1.6 (1.5) |
| Sham | 6 | 31.3 (3.6) | 2 (33) | 8.3 (3.4) | 1009 (261) | 65.3 (NR) | 3.0 (1.3) | −0.5 (15.9) | −0.6 (0.9) | |||
| Chang et al., 2020 | 5 | SZ - SZAff | Active | 30 | 44.7 (10.7) | 11 (37) | 14.7 (9.5) | 532 (348) | 67.4 (13.0) | 4.5 (1.0) | −10.8 (7.1) | −2.00 (1.17) |
| Sham | 30 | 45.0 (10.9) | 19 (63) | 15.8 (10.6) | 523 (332) | 73.3 (10.3) | 4.7 (0.9) | −1.4 (4.4) | −0.23 (0.50) | |||
| Smith et al., 2020 | 2 | SZ | Active | 24 | 43.7 (14.1) | 15 (62) | 17.9 (11.7) | NR | 62.0 (15.6) | 3.6 (1.2) | −3.3 (13.8) | −0.381 (0.921) |
| Sham | 21 | 47.1 (11.9) | 12 (57) | 19.7 (11.1) | NR | 58.7 (9.9) | 3.7 (1.2) | −4.4 (9.9) | 0.0833 (1.06) | |||
| Valiengo et al., 2020 | 5 | SZ | Active | 50 | 34.6 (8.4) | 9 (18) | 14.2 (8.1) | 475 (225) | 73.6 (15.8) | 2.8 (1.4) | −5.3 (10.2) | 0 (1.30) |
| Sham | 50 | 35.9 (10.1) | 11 (22) | 14.1 (8.7) | 500 (400) | 73.9 (13.4) | 3.0 (1.4) | −3.1 (7.9) | −0.24 (1.25) |
Note Otherwise specified, results are given as mean (SD).
Jadad scores correspond to the average of the scores given by the author OA and by the main author of the associated article.
Koops et al. (2018) included 21 patients with SZ, 4 patients with psychosis NOS, 1 patient with schizoaffective disorder, 1 patient with affective disorder, and 1 patient with borderline personality disorder.
Abbreviations: CPZE, chlorpromazine equivalent; NR, not reported; PANSS, positive and negative syndrome scale; SZ, schizophrenia; SZAff, schizoaffective disorder.
Characteristics of included studies
| Authors, Year | Electrode Montage (Anode/Cathode) | Electrode Size (cm²) | Intensity (mA) | Session Duration (min) | Number of tDCS Session | Current Density (mA/cm2) | Total Charge Per Sessionb (C/cm2) | Total Charge Per Regimenc (C/cm2) |
|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Palm et al., 2016 | F3/FP2 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Chang et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Gomes et al., 2018 | F3/F4 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Jeon et al., 2018 | F3/F4 | 25 | 2 | 30 | 10 | 0.08 | 0.144 | 1.44 |
| Koops et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Mellin et al., 2018 | F3–FP1/T3–P3 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Kantrowitz et al., 2019 | F3–FP1/T3–P3 | 38.81 | 2 | 20 | 10 | 0.05 | 0.062 | 0.62 |
| Lindenmayer et al., 2019 | F3–FP1/T3–P3 | 35 | 2 | 20 | 40 | 0.06 | 0.068 | 2.74 |
| Weickert et al., 2019 | F4/T3–P3 | 35 | 2 | 20 | 20 | 0.06 | 0.068 | 1.37 |
| Chang et al., 2020 | F3–FP1 + F4–FP2/forearmsa | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Smith et al., 2020 | F3/FP2 | 5.08 | 2 | 20 | 10 | 0.39 | 0.472 | 4.72 |
| Valiengo et al., 2020 | F3/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Authors, Year | Electrode Montage (Anode/Cathode) | Electrode Size (cm²) | Intensity (mA) | Session Duration (min) | Number of tDCS Session | Current Density (mA/cm2) | Total Charge Per Sessionb (C/cm2) | Total Charge Per Regimenc (C/cm2) |
|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Palm et al., 2016 | F3/FP2 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Chang et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Gomes et al., 2018 | F3/F4 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Jeon et al., 2018 | F3/F4 | 25 | 2 | 30 | 10 | 0.08 | 0.144 | 1.44 |
| Koops et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Mellin et al., 2018 | F3–FP1/T3–P3 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Kantrowitz et al., 2019 | F3–FP1/T3–P3 | 38.81 | 2 | 20 | 10 | 0.05 | 0.062 | 0.62 |
| Lindenmayer et al., 2019 | F3–FP1/T3–P3 | 35 | 2 | 20 | 40 | 0.06 | 0.068 | 2.74 |
| Weickert et al., 2019 | F4/T3–P3 | 35 | 2 | 20 | 20 | 0.06 | 0.068 | 1.37 |
| Chang et al., 2020 | F3–FP1 + F4–FP2/forearmsa | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Smith et al., 2020 | F3/FP2 | 5.08 | 2 | 20 | 10 | 0.39 | 0.472 | 4.72 |
| Valiengo et al., 2020 | F3/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
Chang et al. (2020) used a bi-anodal montage targeting the left and right dorsolateral prefrontal cortex with extracephalic cathodes.
Total charge per session(C/cm2) = stimulation intensity (A)/electrode size (cm2) × duration (s).
Total charge per tDCS regimen (C/cm2) = stimulation intensity (A)/electrode size (cm2) × duration (s) × number of sessions.
Characteristics of included studies
| Authors, Year | Electrode Montage (Anode/Cathode) | Electrode Size (cm²) | Intensity (mA) | Session Duration (min) | Number of tDCS Session | Current Density (mA/cm2) | Total Charge Per Sessionb (C/cm2) | Total Charge Per Regimenc (C/cm2) |
|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Palm et al., 2016 | F3/FP2 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Chang et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Gomes et al., 2018 | F3/F4 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Jeon et al., 2018 | F3/F4 | 25 | 2 | 30 | 10 | 0.08 | 0.144 | 1.44 |
| Koops et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Mellin et al., 2018 | F3–FP1/T3–P3 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Kantrowitz et al., 2019 | F3–FP1/T3–P3 | 38.81 | 2 | 20 | 10 | 0.05 | 0.062 | 0.62 |
| Lindenmayer et al., 2019 | F3–FP1/T3–P3 | 35 | 2 | 20 | 40 | 0.06 | 0.068 | 2.74 |
| Weickert et al., 2019 | F4/T3–P3 | 35 | 2 | 20 | 20 | 0.06 | 0.068 | 1.37 |
| Chang et al., 2020 | F3–FP1 + F4–FP2/forearmsa | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Smith et al., 2020 | F3/FP2 | 5.08 | 2 | 20 | 10 | 0.39 | 0.472 | 4.72 |
| Valiengo et al., 2020 | F3/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Authors, Year | Electrode Montage (Anode/Cathode) | Electrode Size (cm²) | Intensity (mA) | Session Duration (min) | Number of tDCS Session | Current Density (mA/cm2) | Total Charge Per Sessionb (C/cm2) | Total Charge Per Regimenc (C/cm2) |
|---|---|---|---|---|---|---|---|---|
| Brunelin et al., 2012 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Palm et al., 2016 | F3/FP2 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Chang et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Gomes et al., 2018 | F3/F4 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Jeon et al., 2018 | F3/F4 | 25 | 2 | 30 | 10 | 0.08 | 0.144 | 1.44 |
| Koops et al., 2018 | F3–FP1/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Mellin et al., 2018 | F3–FP1/T3–P3 | 25 | 2 | 20 | 10 | 0.08 | 0.096 | 0.96 |
| Kantrowitz et al., 2019 | F3–FP1/T3–P3 | 38.81 | 2 | 20 | 10 | 0.05 | 0.062 | 0.62 |
| Lindenmayer et al., 2019 | F3–FP1/T3–P3 | 35 | 2 | 20 | 40 | 0.06 | 0.068 | 2.74 |
| Weickert et al., 2019 | F4/T3–P3 | 35 | 2 | 20 | 20 | 0.06 | 0.068 | 1.37 |
| Chang et al., 2020 | F3–FP1 + F4–FP2/forearmsa | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
| Smith et al., 2020 | F3/FP2 | 5.08 | 2 | 20 | 10 | 0.39 | 0.472 | 4.72 |
| Valiengo et al., 2020 | F3/T3–P3 | 35 | 2 | 20 | 10 | 0.06 | 0.068 | 0.68 |
Chang et al. (2020) used a bi-anodal montage targeting the left and right dorsolateral prefrontal cortex with extracephalic cathodes.
Total charge per session(C/cm2) = stimulation intensity (A)/electrode size (cm2) × duration (s).
Total charge per tDCS regimen (C/cm2) = stimulation intensity (A)/electrode size (cm2) × duration (s) × number of sessions.
Insight and tDCS Treatment
A forest plot of the distribution of effect sizes is presented in figure 2. The overall ESg was −0.46 (medium effect, 95% CI [−0.78; −0.14]) and reached statistical significance (P = .005), indicating a greater decrease in G12 scores among patients who received active tDCS than among those who received sham tDCS. The individual ES ranged from −1.94 to 0.19. We found significant heterogeneity (Q = 41.9, P < .001), which was moderate-to-high (I² = 71.4%, τ² = 0.23).
Forest plot showing comparison of insight, as measured by the G12 score, among patients with schizophrenia who received either active or sham tDCS.
Abbreviations: ES, effect size; CI, confidence interval; RE, random effect.
Moderator and Subgroup Analyses
Post hoc analyses were conducted to explore potential sources of global heterogeneity (individual results of meta-regression shown in supplementary table 1). Only the age of participants and baseline G12 score model reached significance, therefore, influencing ESg.
Next, we conducted exploratory subgroup analyses to further investigate the influence of age of participants and baseline G12 score on the association between tDCS and insight improvement (supplementary table 2). Regarding age, studies including patients with an average age over 40 years showed a significantly larger ESg compared with those including patients with an average age below 40 years (Q = 4.11, P = .04). No significant difference was found between studies with on average at least a mild lack of insight (average ≥ 3) and those with on average an absence of or minimal lack of insight (average < 3, Q = 0.62, P = .43).
Sensitivity Analyses
No outlier study could be identified by the QQ plot (supplementary figure 3) but the study of Chang et al.23 substantially influenced the overall ESg, as indicated by the Baujat plot (supplementary figure 4). In that respect, meta-outlier detection led to exclusion of the Chang and colleagues’ study. Leave-one-out meta-analysis showed that the estimated ES, as well as the between-study heterogeneity, were stable by excluding a single study with the exception of Chang et al., 2020: Range of SMDs −0.40 to −0.53 compared to −0.31 (supplementary table 3). When excluding the Chang et al.’ study, the ESg remained significant at −0.31 (small effect, 95% CI [−0.54; −0.09], P = .006) and no significant heterogeneity was observed (Q = 16.92, P = .11). Additional sensitivity exploratory analyses without the study of Chang and colleagues were conducted (supplementary table 4). Only meta-regression for baseline G12 score changed dramatically when excluding Chang et al. 2020 (P = .02 vs P = .84).
Publication bias was unlikely since the funnel plot and Egger’s regression test did not indicate any significant asymmetry (P = .38, supplementary figure 5).
Discussion
Our meta-analysis indicated that tDCS treatment was significantly associated with improvement in insight among patients with schizophrenia. Indeed, we observed a significant, moderate effect size (ESg = −0.46) in favor of a decrease in G12 scores after repeated sessions of tDCS (at least ten), associated with moderate-to-high heterogeneity (I² = 71.4%). Our results are consistent with those reported by 5 articles investigating the effect of tDCS using other insight measures that showed significant improvement in insight after active tDCS compared with sham tDCS.21–25 Although our findings highlighted a statistically significant effect on insight, it remains to be determined whether this effect is clinically meaningful and translates into a functional improvement. To do so, several recent studies have recommended testing the clinical meaning of treatment-induced changes against clinicians’ subjective perception of clinical improvement, as measured with the scores at the Clinical Global Impression Scales.54–56 We recommend future tDCS studies to systematically evaluate and report such measures.
Improvements in insight should be considered a mechanism underlying the clinical efficacy of tDCS by leading to better medication adherence38 and compliance with treatment.57 Improvements in insight could also benefit patients’ quality of life, as insight improvements correlate with reductions in symptoms57 and with the enhancement of social functioning.4 Finally, awareness of illness and symptoms could enhance prognosis by improving the involvement of patients in their own recovery.58 Impaired insight is a transdiagnostic construct, since it is found in many psychiatric disorders,1–3 and it is underpinned by several commonalities among them.59 It is therefore important to focus on insight because it represents a therapeutic target that can affect many patients.
In view of the tDCS parameters, we note that all the electrode setups used targeted the dorsolateral prefrontal cortex (DLPFC). Several neuroimaging studies support the role of the DLPFC as a key region for insight into the illness. For instance, insight has been inversely correlated with gray matter volume deficits in the right DLPFC, bilateral cerebellum, and posterior part of the right temporal inferior gyrus in schizophrenia.60 Lack of insight has also been associated with white matter deficits in the frontal, temporal and parietal regions,61 and structural hemispheric asymmetry in the DLPFC.20 Additionally, a functional magnetic resonance imaging study in patients with schizophrenia highlighted the implication of the DLPFC in both unawareness of symptoms and misattribution of symptoms.62 The combination of structural and functional abnormalities of frontal, parietal, and temporal areas of the brain and dysfunction in their connectivity may lead to the impairment of insight.59 Stimulating the DLPFC using either a bifrontal tDCS montage with the anode placed over the left DLPFC or a frontotemporoparietal montage with the anode over the left or right DLPFC may promote its excitability18 and reestablish the proper functioning of such networks, since tDCS can modulate the activity of networks associated with the targeted regions.17 Given that targeting the DLPFC with transcranial alternating current stimulation has led to improvements in retrospective self-evaluations and metacognition among healthy subjects,63 one could hypothesize that in patients with schizophrenia, stimulating the DLPFC with the anode will also enhance these cognitive functions, thereby improving insight into their illness. Our moderator analyses did not reveal the superiority of any particular electrode arrangement between bifrontal and frontotemporoparietal montages. However, Chang et al. study (2020) which was identified as a significant outlier because of its large ESg was the only one using a bi-anodal montage, ie, with two anodes targeting the left and right DLPFC and two extracephalic cathodes. These findings nevertheless suggest a potential influence of the electrode arrangement on the observed effect.
In addition to the DLPFC stimulation, it seems also probably that the improvement of insight into illness results from cathodal tDCS-induced inhibition18 of left temporoparietal/posterior parietal areas. Indeed, almost half of the included studies used a frontotemporoparietal montage with the cathode placed at T3–P3, which can partially cover the temporoparietal areas and adjacent posterior parietal areas. These regions, namely the left angular gyrus and posterior parietal areas, have been reported to be hyperactive in patients with schizophrenia who present impairment in insight.64,65 Furthermore, an alteration in interhemispheric balance, characterized particularly by a left-hemispheric dominance of the temporoparietal areas and prefrontal regions, has been associated with poor insight65 and could be related to alterations in the white matter tracts of the posterior part of the corpus callosum.66 The association between the stimulation of the left temporoparietal/posterior parietal areas and improved insight is also supported by two studies that delivered cathodal stimulation to these regions. A single biparietal tDCS session with the cathode over the left posterior parietal area has been reported to reduce the interhemispheric imbalance in patients with impaired insight and this mechanism has been proposed to contribute to improved insight.22 In addition, a study investigating the impact of high-definition tDCS targeting the left temporoparietal area with the cathode described a significant improvement in insight in patients with schizophrenia.67
Exploratory moderator analyses showed a significant influence of participants’ age and G12 score at baseline on the improvement of insight after tDCS treatment. First, it appears that the higher the G12 score at baseline, the greater the improvement in insight following tDCS. However, these findings are mainly driven by Chang et al., 2020’s study, in which patients had the highest mean G12 score at baseline. Indeed, sensitivity analysis revealed that the influence of G12 score at baseline did not remain significant after the exclusion of this study. Because a significant overall ESg is still observed even after excluding Chang et al., 2020’s study, G12 score at baseline does not appear to be a determinant factor of the improvement in insight induced by tDCS. Second, the older the patient the greater the improvement. These results are surprising since recent meta-analyses showed a negative association between age and the tDCS-induced reduction of hallucinations and negative symptoms in patients with schizophrenia.30,31 In addition, age has been associated with an alteration of the efficiency of tDCS to induce brain plasticity.68 The opposite association found here between age and tDCS-induced improvement in insight could be related to the complex interaction between age and insight: the course of insight impairment follows a U-shaped trajectory where insight impairment increases with the first episode of psychosis, decreases over midlife, and increases again in late life, with fluctuations depending on the other episodes occurring throughout the patient’s life.69 The complex interaction between age, insight, and tDCS-induced neural plasticity deserves further investigation.
We identified a moderate risk of bias associated with the randomization process in about half of the studies. This risk seems to be mainly due to a lack of reporting sufficient random sequence generation descriptions in the published articles. It is noteworthy that the lack of reporting sufficient details does not necessarily reflect a bias in the randomization process itself but may be related to a lack of awareness of the importance of reporting such information, or a lack of space related to the inherent space limitation of scientific articles. The second questionnaire on the study quality33 we used helps to mitigate this bias since the authors were given the opportunity to rate their studies themselves, so as not to penalize the study for aspects that were merely not mentioned in the published article. Nevertheless, it is critical to report details about the randomization process since this process prevents the influence of potentially confounding factors, whether known or not.70 The potential risk found in some of the included studies is therefore difficult to interpret but could be responsible for the existence of unidentified biases that were not subsequently controlled for.
Finally, one could question the specificity of tDCS effects on insight compared to other symptoms. However, adding the change in PANSS total score as a moderator in our main analysis revealed no significant influence of this parameter on tDCS-induced insight changes. These findings suggest that, although tDCS was associated with improvements in insight (current study) and other symptoms of schizophrenia (eg,30–32), these effects may occur independently and result from nonrelated mechanisms. A large prospective RCT or a meta-analysis based on individual participant data is needed to properly address this question of the specificity of the effect of tDCS on insight vs. other symptoms, and to investigate the causal relationship and temporality between these improvements.
Limitations
Some methodological limitations must be emphasized. First, most of the included studies had small sample sizes, which could be responsible for inflated individual ES estimation. Second, using a single item to assess modifications in insight into the illness could limit the interpretations of current results and their clinical significance. Nevertheless, a high correlation of G12 scores with detailed scales of insight has repeatedly been observed,61,71 suggesting that G12 scores can be considered a good indicator of changes in insight. It should be noted that despite its correlation with the total scores of detailed scales of insight, G12 score does not systematically correlate with the subscales of those insight scales and therefore does not present the same subtlety in the assessment of insight. However, a recent article highlighted a high variability in insight dimensions between those detailed scales, questioning their respective interpretation.72 Third, exploratory post hoc analyses identified some variables that significantly influenced our primary outcome. These results enable us to establish new working hypotheses that should be investigated in further studies, but they do not allow us to draw firm conclusions about the effect of moderators on the association between improvement in insight and tDCS. Fourth, PANSS interrater reliability measures were not systematically reported (supplementary table 5), thus this potential bias could be not controlled. Fifth, we were interested in the short-term effect of tDCS. Just as the improvement of positive or negative symptoms is maintained over time after repeated tDCS sessions, it is reasonable to assume that improvement in insight, as measured by a decrease in G12 scores, will be observed over a few months. Furthermore, studies should nevertheless investigate this question. Finally, controversies exist regarding the effects of tDCS, mostly arising from meta-analyses that find no reliable effects of tDCS on behavior (eg,73,74 for a review of the key points of controversy). The most notable criticisms raised are the lack of reproducibility of the effects of tDCS and the large inter-individual variability, restricting the generalizability of the results. Several factors have been pointed out as plausible reasons for the variability of the findings including differences in stimulation settings and parameters such as stimulation intensity and duration, size of electrodes, location and orientation of electrodes (table 2), as well as differences in experimental design including sample sizes, choices of control conditions and blinding procedures (ie, procedures implemented to ensure that participants/tDCS operators/outcome assessors are not aware of the type of stimulation they are receiving/delivering). Namely, control conditions and blinding procedures are of paramount importance in clinical tDCS studies since they could bias the adequate estimation of tDCS effects through the occurrence of a “placebo effect.” In the reviewed studies, control conditions all consisted of sham stimulations supposed to only mimic transient effects of active tDCS on the scalp but not to induce physiological effects (supplementary table 6). Not only are these sham conditions questionable when successful blinding is not achieved, but they could also mask a differential effect of active stimulation by causing cerebral changes and therefore be a hidden source of variability.75 In addition, the effectiveness of the chosen blinding strategy was not systematically assessed or reported by the authors (supplementary table 7). When done, a post-stimulation questionnaire was generally used in which participants and providers had to guess whether they had received active or sham stimulation, and the validity of such method for measuring blinding effectiveness has recently been questioned.76 Efforts should be made to improve blinding and sham procedures and their evaluation through the development of new methods.
Conclusion
The results of the current systematic review and meta-analysis revealed a significant association, with a moderate effect size, between tDCS treatment and improvement in the PANSS G12 score for insight into the illness among patients with schizophrenia. Participants who received at least 10 sessions of active tDCS displayed greater improvements in insight than patients who received sham tDCS. It is important to note that although the effect is statistically significant, it is unclear whether this effect would transfer into clinical meaningful improvement in real life. Furthermore, randomized controlled studies are needed to investigate the role of different variables of tDCS-induced changes in insight.
Funding
None.
Acknowledgments
We thank Dr Fröhlich for providing data, Dr Fitzgerald, Dr Shiozawa and Dr Trevizol for responding to our requests.
U.P. received speaker honoraria from NeuroCare Group, Munich, Germany. A.R.B. reports grants from Sao Paulo Research State Foundation (2018/10861-7, 2019/06009-6), Brazilian National Council of Scientific Development productivity support (PQ-1B), the UK Academy of Medical Sciences (NAFR 12/1010-2), and University of Sao Paulo Medical School productivity support (PIPA-A). L.V. reports grants from Narsad (2020/28755) and the SIRS Research Fund Award 2020–2021. The other authors do not report any potential conflicts of interest.
Author Contributions
M.M. and O.A. had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. M.M. and J.B. conceived and designed the study and shared last authorship. O.A., M.B., J.B., and M.M. contributed to the acquisition, analysis, or interpretation of data. O.A. drafted the manuscript. All authors contributed to the critical revision of the manuscript for important intellectual content. O.A., J.B., and M.M. contributed to the statistical analysis.
References
Author notes
Shared last authorship.
