Comparing different non-invasive brain stimulation interventions for bipolar depression treatment: A network meta-analysis of randomized controlled trials

Non-invasive brain stimulation (NIBS) is a promising treatment for bipolar depression. We systematically searched for randomized controlled trials on NIBS for treating bipolar depression (INPLASY No: 202340019). Eighteen articles (N = 617) were eligible for network meta-analysis. Effect sizes were reported as standardized mean differences (SMDs) or odds ratios (ORs) with 95% confidence intervals (CIs). Anodal transcranial direct current stimulation over F3 plus cathodal transcranial direct current stimulation over F4 (a-tDCS-F3 + c-tDCS-F4; SMD = (cid:0) 1.18, 95%CIs = (cid:0) 1.66 to (cid:0) 0.69, N = 77), high-definition tDCS over F3 (HD-tDCS-F3; (cid:0) 1.17, (cid:0) 2.00 to (cid:0) 0.35, 25), high frequency deep transcranial magnetic stimulation (HF-dTMS; (cid:0) 0.81, (cid:0) 1.62 to (cid:0) 0.001, 25), and high frequency repetitive TMS over F3 plus low frequency repetitive TMS over F4 (HF-rTMS-F3 + LF-rTMS-F4; (cid:0) 0.77, (cid:0) 1.43 to (cid:0) 0.11, 38) significantly improved depressive symptoms compared to sham controls. Only a-tDCS-F3 + c-tDCS-F4 (OR = 4.53, 95%CIs = 1.51 – 13.65) and HF-rTMS-F3 + LF-rTMS-F4 (4.69, 1.02 – 21.56) showed higher response rates. No active NIBS interventions exhibited significant differences in dropout or side effect rates, compared with sham controls.


Introduction
Bipolar disorder (BD) is a mental disorder with an estimated lifetime prevalence of 1-3% (Merikangas et al., 2011), characterized by recurrent manic or hypomanic episodes that may alternate with depressive episodes (Carvalho et al., 2020).Although manic and hypomanic episodes are hallmark features of BD, depressive episodes often impose the most significant burden on affected individuals, comprising 30-50% of the total illness duration (Judd et al., 2003;Judd et al., 2002).However, research on bipolar depression treatment is limited.Currently, only six drugs are approved by the US Food and Drug Administration (FDA) or recommended as first-line treatments by the Canadian Network for Mood and Anxiety Treatments and International Society for Bipolar Disorders (CANMAT/ISBD) guidelines for managing depressive episodes in patients with BD (Carvalho et al., 2020;Yatham et al., 2018).Therefore, novel therapeutic options to meet clinical needs are urgently required.
Non-invasive brain stimulation (NIBS) interventions, such as repetitive transcranial magnetic stimulation (rTMS), transcranial direct current stimulation (tDCS), and cranial electrical stimulation (CES), may offer promising alternative treatments for several psychiatric conditions (Pallanti, 2021;Pallanti et al., 2021), including bipolar depression.A recent meta-analysis suggested that active rTMS treatment is effective (Nguyen et al., 2021).The 2022 Taiwan treatment consensus and 2020 International College of Neuropsychopharmacology (CINP) guidelines for bipolar depression both advocate for the continuous rTMS integration as a potential therapeutic strategy, assigning it a recommendation grade of level 2 (Cheng et al., 2023;Fountoulakis et al., 2020).However, the US FDA has not yet approved rTMS for bipolar depression treatment, and currently only provides grants for breakthrough device designation for this indication (Camprodon, 2021).Inconsistent results for NIBS treatment for bipolar depression have resulted in discrepancies between guidelines and certifications.For instance, some studies have reported inconsistent findings regarding left-sided excitatory stimulation, with both deep transcranial magnetic stimulation (dTMS) and tDCS showing potential efficacy (Sampaio-Junior et al., 2018;Tavares et al., 2017), while intermittent theta-burst stimulation (iTBS) appears ineffective (McGirr et al., 2021).Moreover, emerging novel stimulation modalities, such as high-definition transcranial direct current stimulation (HD-tDCS) (Zhang et al., 2023), have not yet been directly compared in a previous meta-analysis on tDCS (Dondé et al., 2017).Thus, all current treatment modalities need to be further incorporated and differentiated to help clinicians choose the most effective options.
In recent years, a number of NIBS techniques have been introduced for the treatment of bipolar depression.However, as there are few direct head-to-head comparisons between these NIBS techniques, determining more efficacious and acceptable modalities remains challenging.Network meta-analyses (NMA) provide methodological advantages in assessing comparative outcomes of different NIBS interventions, elucidating the relative advantages of multiple interventions that previous pairwise meta-analyses could not (Chen et al., 2023;Huang et al., 2023).We hypothesized that the efficacies of certain NIBS, such as iTBS, for the treatment of bipolar depression may be different from those for the treatment of unipolar depression (Mutz et al., 2019).To address this research gap, we conducted a systematic review and NMA to investigate the treatment outcomes of different NIBS interventions.

General study guidelines
This NMA followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA, eTable 1) guidelines (Page et al., 2021).The study protocol was registered in the International Platform of Registered Systematic Review and Meta-analysis Protocols (No: INPLASY202340019).The Institutional Review Board of Chang Gung Memorial Hospital reviewed the study protocol and granted an exemption from the requirement for ethical approval (No: 202300497A3).

Search strategy and eligibility criteria
Two authors (CW Hsu and YC Chen) conducted a systematic search of multiple databases, including PubMed, Embase, and Cochrane CEN-TRAL.A literature search was conducted up to April 1, 2023, without language restrictions.Additionally, we searched seven gray literature databases and manually searched for potentially eligible articles cited in review articles and meta-analyses (Hett and Marwaha, 2020;Mutz et al., 2019;Nguyen et al., 2021).We screened the titles and abstracts of all retrieved references and selected potentially eligible studies for full-text review.The details of the search strings used for these databases are provided in eTable 2.
The PICOS criteria for this study were as follows: (1) patient: participants with a diagnosis of bipolar depression; (2) intervention: any NIBS; (3) comparison: sham or active controls; (4) outcome: changes in depressive symptoms, anxiety symptoms, quality of life (QOL), and overall clinical status, and remission, serious adverse events, and dropout rates; and (5) study design: limited to randomized controlled trials (RCTs).
The inclusion criteria were as follows: (1) human participants; (2) patients with a diagnosis of bipolar depression, either bipolar I or II disorder; (3) application of NIBS interventions, including rTMS, tDCS, CES, vagal nerve stimulation, and magnetic field stimulation; (4) preand post-treatment scores or score changes on depression rating scales; and (5) RCTs with either sham-control or active controls.The exclusion criteria were as follows: (1) patients with BD in the euthymic phase (McIntyre et al., 2021;Yang et al., 2019); (2) patients with both unipolar and bipolar depression combined in one study without separate data to analyze bipolar depression (Mutz et al., 2019); (3) not reporting outcomes of interest, such as depression rating scales; (4) not clearly C.-W. Hsu et al. defining stimulation protocols for NIBS interventions (e.g., only describing real stimulation) (Dolberg et al., 2002); (5) case series or reports and non-RCTs (allocation bias); and (6) conference papers, protocols, and non-peer-reviewed articles.If multiple studies used identical datasets, we included the study with the largest sample size and most informative data.If discrepancies occurred regarding inclusion of a particular study, a third investigator (PT Tseng) was consulted.

Data extraction and outcomes
Two authors (CW Hsu and YC Chen) independently extracted data from eligible studies, including the first author's name, publication year, gender (percentage of female participants), age (range or mean), sample size, treatment arm, stimulation protocol, follow-up duration, and research location.When the required data were not reported, we contacted the corresponding author to obtain the original data.
This study had two primary outcomes: (1) efficacy, measured by changes in assessment scores of overall depressive severity after NIBS interventions; and (2) acceptability, defined as the dropout rate (i.e., percentage of patients who discontinued the study for stimulationrelated reasons before completion).We examined several secondary outcomes, including (1) continuous variables: improvement in anxiety severity, clinical global impression-severity (CGI-S), and QOL; and (2) categorical variables: response rate, remission rate, and serious side effect rate.To assess response rates, we calculated the proportion of participants with an adequate response to NIBS, with most studies defining this as a 50% reduction in depression score post-intervention.We assessed remission rates by calculating the proportion of participants with depressive symptom remission after NIBS, with most studies defining this as below the cutoff value for depression scales.Side effect rate was calculated as the proportion of participants who experienced serious adverse events during the study period, including seizures, (hypo)manic episodes, or suicidal ideation.Moreover, we considered the outcomes observed at the first visit after the last NIBS intervention in each eligible RCT.Data from intent-to-treat participants were extracted first; if no data were found, data from per-protocol participants were used.

Risk of bias assessment
Two independent authors (CW Hsu and YC Chen) used the risk of bias tool in the Cochrane handbook to assess the bias risk for each domain (i.e., selection, performance, detection, attrition, reporting, and other biases) in the included trials (Higgins JPT, March 2011)."Other bias" was use to evaluate for any possible biases not covered by the existing domains.The two investigators resolved any disagreements through consensus, and consulted a third investigator (PT Tseng) when needed.

Data synthesis and statistical analysis
In this NMA, we assessed pre-post changes (continuous variables) and incidence rates (categorical variables) of the aforementioned outcomes.We estimated standardized mean differences (SMDs), Cohens d, with 95% confidence intervals (95%CIs) for continuous variables and odds ratios (OR) and 95%CIs for categorical variables.For studies with zero events in either the intervention or control arm, a 0.5 zero-cell correction was applied during the meta-analysis.Pairwise metaanalyses and NMA were conducted using random-effects and frequentist models.We evaluated heterogeneity among the included studies using the tau value, which is the estimated standard deviation of the effects across the included studies.We used STATA (version 17.0, mvmeta command; StataCorp, College Station, TX, USA) to conduct the NMA (White, 2015).All comparisons were performed using a two-tailed test; P-values < .05were considered statistically significant.
We used mixed comparison with generalized linear mixed models to analyze both direct and indirect comparisons (Tu, 2014).Indirect comparisons were conducted using transitivity, whereby the differences between treatments A and B could be calculated from their comparisons with treatment C. To compare multiple treatment arms, the direct and indirect evidence of the included studies were combined (Lu and Ades, 2004).The restricted maximum likelihood method was utilized to assess the variance between the studies (Kontopantelis et al., 2013).To provide further clinical applications, relative ranking probabilities of the effects of all treatments on the target outcomes were calculated.The surface under the cumulative ranking curve (SUCRA) indicated the percentage of the mean rank of each intervention relative to an imaginary intervention that was optimal without uncertainty (Salanti et al., 2011).To evaluate potentially small study effects and publication bias, we used comparison-adjusted funnel plots and Egger's regression.Inconsistencies were evaluated using the design-by-treatment, loop-specific approach, and node splitting model (Higgins et al., 2014).
To justify the transitivity hypothesis, we assessed the effectiveness of different sham interventions by computing the change in depression severity relative to tDCS, TMS, and CES sham interventions using Comprehensive Meta-Analysis (version 4; Biostat, Englewood, NJ, USA).Moreover, considering mixed samples of patients with unipolar and bipolar depression in some previous RCTs (Mutz et al., 2019), we conducted two sensitivity analyses related to depression severity.For RCTs in which bipolar data could be extracted and incorporated into our NMA (Kimbrell et al., 1999;Loo et al., 2018;Speer et al., 2014), we excluded them and reassessed the remaining data to evaluate the robustness of our findings (excluding mixed patients).These RCTs did not consider random allocation for patients with unipolar or bipolar depression, potentially leading to imbalances in group distribution (allocation bias).For RCTs without bipolar data available for extraction, we hypothesized that the overall effect size was equivalent to the effect size observed in patients with BD, then calculated the proportion of those patients in each study and added them to the original NMA (adding mixed patients).This enabled us to estimate the potential influence of excluded studies on our overall findings.Finally, we performed network meta-regression analyses to identify potential moderators (age, gender, and total NIBS sessions) that influence relative efficacy in alleviating depressive symptoms.

Certainty of the evidence
We used the GRADE (Grading of Recommendations Assessment, Development and Evaluation) approach to evaluate certainty in the evidence synthesis results (Puhan et al., 2014).
No NIBS interventions demonstrated a significant difference in dropout rates compared with sham controls (eTable 5B, eTable 6B, and Fig. 3B).

Meta-regression, risk of bias, publication bias, inconsistency, and heterogeneity
In the network meta-regression analyses, moderators such as mean age (p = 0.841), proportion of female participants (p = 0.059), and total number of NIBS sessions (p = 0.447) were not statistically significant as factors affecting changes in depressive symptoms.According to ROB2, 79% (85/108 items) of the included studies had a low risk of bias, 19% (21/108 items) had some risk, and 2% (2/108 items) had a high risk of bias (eTable 7 and eFigure 6).Publication bias analysis using funnel plots and Egger's test indicated no significant publication bias (eFigure 7A-7 H).In this NMA, the outcome for anxiety showed inconsistencies in the design-by-treatment method (p = .003)and loop approach (p = .003);however, other outcomes did not display any significant inconsistencies, including global (design-by-treatment method) and local (loop-specific approach and node-splitting method) inconsistencies.No significant heterogeneity detected by tau values (eTables 8-10).The evidence quality for most comparisons ranged from moderate to low, based on GRADE rating (eTable 11) (Tseng et al., 2022b).
Drawing on previous brain stimulation research on major depression (Mutz et al., 2019), this NMA specifically focused on BD.This study supports the findings that tDCS (HD-tDCS-F3 and a-tDCS-F3 +c-tDCS-F4) and bilateral rTMS (HF-rTMS-F3 +LF-rTMS-F4) can significantly improve depressive symptoms.Neither excitatory stimulation of the left cortex (HF-rTMS-F3 and iTBS-F3) nor inhibitory stimulation of the right cortex (LF-rTMS-F4) showed significant effects.This NMA further demonstrated that HF-dTMS can significantly improve bipolar depression.When considering response or remission rates for depressive episodes, only a-tDCS-F3 +c-tDCS-F4 and HF-rTMS-F3 +LF-rTMS-F4 were significantly better than sham controls for response rates (but not remission rates), while other stimulation modalities might not yield satisfactory results.The study on HD-tDCS-F3 did not report response or remission rates; therefore, this NIBS modality was not included in the comparison (Zhang et al., 2023).Despite the phenotypical similarities of depressive episodes in the context of major depressive disorder or BD, this NMA identified subtle variations in the efficacy of NIBS protocols compared with previous research (Mutz et al., 2019).We hypothesize these discrepancies could exist for a few reasons.Pathophysiological investigations from resting-state functional magnetic resonance imaging have revealed distinct patterns differentiating between two seemingly similar clinical phenotypes (Pastrnak et al., 2021).Compared with unipolar depression, bipolar depression is characterized by more diffuse disruptions in cortico-striatal-limbic-subcortical and somatosensory functional connectivity (Ellard et al., 2021).This more extensive connectivity disruption may require a disorder-specific NIBS protocol (broader or deeper stimulation) with translational therapeutic implications to effectively modify neuronal activity (Camprodon, 2021), such as HD-tDCS-F3 (Jog et al., 2023), HF-dTMS (Feifel and Pappas, 2016;Zibman et al., 2021), a-tDCS-F3 +c-tDCS-F4, and HF-rTMS-F3 +LF-rTMS-F4, the SUCRA analysis identified as the four most effective stimulation modalities.In addition to the four NIBS modalities mentioned above, transcranial alternating current stimulation (tACS) is another NIBS modality not included in this NMA and is emerging as a new paradigm for addressing depressive symptoms (Elyamany et al., 2021).Drawing parallels with HD-tDCS-F3 and a-tDCS-F3 +c-tDCS-F4, tACS may also hold potential in aiding bipolar depression treatment.However, this hypothesis currently has limited clinical evidence, and further study is necessary to substantiate its validity.
This NMA also revealed that specific NIBS modalities may be advantageous for particular comorbidities in individuals with bipolar depression.For instance, HF-dTMS demonstrated significant benefits in reducing anxiety symptoms and improving the CGI-S scale.This finding is in line with the US FDA's approval of HF-dTMS for treating comorbid anxiety in major depressive disorder (510(k) No. K210201) (Pell et al., 2022) and extends the effectiveness of HF-dTMS to include bipolar depression.However, only eight studies evaluated these secondary outcomes (Kazemi et al., 2016;Lee et al., 2022;Mak et al., 2021;Mallik et al., 2023;McClure et al., 2015;McGirr et al., 2021;Sampaio-Junior et al., 2018;Tavares et al., 2017); therefore, other NIBS modalities may also be effective, but their effects remain unknown because of the limited available evidence.Further, dropout and serious side effects rates showed no significant differences between all treatment modalities and sham controls, suggesting that NIBS is well tolerated.Notably, left-sided excitatory stimulation generally showed a trend toward a higher incidence of side effects than right-sided inhibitory stimulation, such as inducing hypomanic symptoms (iTBS-F3: one in active; HF-rTMS-F3: one in active; HD-tDCS-F3: two in active and one in sham) (Hu et al., 2016;McGirr et al., 2021;Zhang et al., 2023).This observation is consistent with previous rationale for the effects of NIBS on the brain (Klomjai et al., 2015), and reminds clinicians to exercise greater caution regarding potential side effects when using excitatory or inhibitory stimulation in patients with depressive episodes of BD.
Certain patient characteristics or stimulation parameters may influence efficacy.Network meta-regression was used to assess the effects of age, gender, and total number of NIBS sessions (Table 1), but results did not reach statistical significance.However, the large amount of missing data for some moderators (disease onset duration) or differences in definitions between tDCS and rTMS (intensity) pose challenges for analysis.Therefore, we cannot conclusively rule out the influence of these moderators.Additionally, based on the discussion surrounding the available evidence in the literature, we would like to provide some recommendations for clinicians considering NIBS for the treatment of bipolar depression: (1) Beyond the commonly employed modalities for treating major depressive disorder, such as HF-rTMS-F3 and LF-rTMS-F4, clinicians may consider using a-tDCS-F3 +c-tDCS-F4, HD-tDCS-F3, HF-dTMS, or HF-rTMS-F3 +LF-rTMS-F4 as potential therapeutic options for bipolar depression.(2) For patients presenting with pronounced anxiety symptoms, HF-dTMS or HF-rTMS-F3 +LF-rTMS-F4 may be particularly beneficial.(3) Current evidence does not advocate the use of TBS (iTBS-F3 or cTBS-F4) for the treatment of bipolar depression.
As the first comprehensive NMA to evaluate NIBS for treating depressive episodes in BD, this study has notable strengths.It encompasses various NIBS techniques, such as rTMS, tDCS, and CES, while evaluating multiple outcomes, including depression and anxiety severity, CGI-S and QOL, response and remission rates, and adverse effect and dropout rates.However, it also has several limitations that should be considered.First, only more recent studies differentiated between bipolar and unipolar depression, while some previous studies did not (eTable 4).Although we added these potential missing data in our sensitivity analysis, it still underscores the possible incompleteness of the data represented in our NMA.Second, the number of studies included in our analysis remains limited, with some treatment arms having only one study, such as HD-tDCS (Zhang et al., 2023), dTMS (Tavares et al., 2017), andCES (McClure et al., 2015).Treatment effect estimates for these active NIBS modalities rely primarily on relative outcomes when compared with the only sham control in the respective trial.Thus, caution is needed regarding the influence of single studies on our overall conclusions.Conversely, some treatment arms are informed by multiple studies.Even though they are classified under the same category in this study, subtle differences in their stimulation parameters are discernible.For instance, the HF-rTMS-F3 and a-tDCS-F3 +c-tDCS-F8 treatment arms in this study exhibited variations in treatment durations (from 2 to 6 weeks), equipment brands, and localization methods (5 cm/5.5 cm rule or 10-20 system).Therefore, the treatment effect size presented in this study might represent a convergence of these stimulation treatments.Nonetheless, we included GRADE ratings to assess the reliability of the available evidence (eTable 11).Third, some of the analyses in our NMA may have been limited by potential heterogeneity between studies, with respect to participant characteristics such as BD severity and subtype, onset duration, co-morbidities, trial duration, concomitant medication, and adjunctive psychotherapy (Ilzarbe and Vieta, 2023).These factors may have also influenced the effectiveness of the NIBS interventions.However, a more refined analysis remains unfeasible due to incomplete reporting of these factors in the included studies.Despite these limitations, our NMA provides valuable insight into the effectiveness of different NIBS modalities for treating bipolar depression, highlighting the need for further research to improve treatment options and potential directions for such research.

Conclusion
This NMA indicates that, compared to the sham control, HD-tDCS-F3, a-tDCS-F3 +c-tDCS-F4, HF-dTMS, and HF-rTMS-F3 +LF-rTMS-F4 can significantly improve depressive symptoms in BD.However, only a-tDCS-F3 +c-tDCS-F4 and HF-rTMS-F3 +LF-rTMS-F4 have evidence of significantly better response rates.These results are not entirely consistent with those found in major depressive disorder, and clinicians should use caution when choosing stimulation modalities for bipolar depression.Moreover, while there are many evidence-based stimulation modalities for major depressive disorder, there is still a need for active development of new stimulation modalities for bipolar depression.

Fig. 2 .
Fig. 2. Network structure of the primary outcome: changes in depression severity and dropout rate.The lines connecting the nodes represent direct comparisons observed in different clinical trials.The size of each circle is proportionate to the number of participants who received a specific treatment.The thickness of the lines is proportional to the number of trials that are interconnected to the network.

Fig. 3 .
Fig. 3. Forest plot of primary outcome: (A) changes in depression severity and (B) dropout rate.(A) When the effect size was less than zero, as presented by the standardized mean difference, the treatment under investigation resulted in a greater reduction in depression severity compared to the sham control.(B) When the effect size was less than one, as presented by the odds ratio, the treatment under investigation demonstrated lower dropout rate compared to the sham control.

Table 1
Characteristics of the included studies.