Depressive disorder is a prevalent condition affecting over 5% of adults worldwide. 1 While medication and psychotherapy are first-line treatments, more than 30% of patients fail to achieve a response. Reference Voineskos, Daskalakis and Blumberger2 Besides, pharmacotherapy can cause significant side-effects Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson and Ogawa3 and psychotherapy is not always readily available and can be costly. Reference Maljanen, Knekt, Lindfors, Virtala, Tillman and Härkänen4 Given these challenges, alternative treatments that are both effective and scalable are needed to support individuals suffering from depression.
Over the past decades, non-invasive brain stimulation (NIBS) techniques, including repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS) have been increasingly studied as potential treatments for depression. Reference da Silva, Vanderhasselt, Pilloni, Charvet, Padberg and Bikson5,Reference Brunoni, Sampaio-Junior, Moffa, Aparicio, Gordon and Klein6 rTMS uses magnetic pulses to induce electric currents in targeted cortical areas, leading to either excitatory or inhibitory effects depending on stimulation frequency. Reference Downar, Siddiqi, Mitra, Williams and Liston7 It has been used in clinical settings for over a decade and is approved for clinical use in many countries, particularly for patients with major depressive disorder (MDD) who have not responded adequately to first-line treatments such as antidepressant medications or psychotherapy. Reference Lam, Kennedy, Adams, Bahji, Beaulieu and Bhat8 In contrast, tDCS is a newer NIBS approach that delivers low-intensity direct electrical current to the brain through scalp electrodes. Reference Nitsche and Paulus9 Unlike rTMS, tDCS does not elicit action potentials; rather, it modulates resting membrane potentials, thereby influencing cortical excitability and promoting neuroplasticity in brain regions implicated in depression, such as the dorsolateral prefrontal cortex. Reference Lefaucheur and Wendling10 Compared to first-line treatments and rTMS approaches, tDCS offers Reference Charvet, George, Charlson, Lustberg, Vogel-Eyny and Eilam-Stock11 several advantages, including ease of use, safety, tolerability, lower cost and the potential for home-based application, making it a particularly attractive option for patients with limited access to specialised clinical settings or for long-term maintenance treatment. Reference Brunoni, Sampaio-Junior, Moffa, Aparicio, Gordon and Klein6,Reference Downar, Siddiqi, Mitra, Williams and Liston7,Reference Guarienti, Razza, Carvalho, Fregni and Brunoni12
Despite being a promising treatment, some recent large-scale randomised controlled trials (RCTs) of tDCS in depression have shown null results, Reference Burkhardt, Kumpf, Crispin, Goerigk, Andre and Plewnia13–Reference Aust, Brakemeier, Spies, Herrera-Melendez, Kaiser and Fallgatter15 raising questions about its overall efficacy. These inconsistencies have highlighted the substantial heterogeneity in study designs and clinical populations across trials. Variations in patient characteristics, Reference Kambeitz, Goerigk, Gattaz, Falkai, Benseñor and Lotufo16–Reference Bulubas, Padberg, Bueno, Duran, Busatto and Amaro18 stimulation parameters Reference Albizu, Indahlastari, Suen, Huang, Waner and Stolte19,Reference Razza, Wischnewski, Suen, De Smet, da Silva and Catoira20 and trial methodologies may contribute to divergent findings. However, the specific impact of these methodological and clinical factors on treatment outcomes remains unclear. A systematic investigation of these variables is therefore essential to comprehensively assess the efficacy of tDCS in depression and to advance the field towards more targeted and effective intervention.
To systematically investigate these issues, a one-stage individual patient data meta-analysis (IPD-MA) was conducted to evaluate the efficacy of active tDCS compared to sham for depressive symptoms in patients from RCTs and to identify potential moderators of treatment improvement using both individual- and study-level data. Unlike traditional meta-analyses that rely on aggregated study-level data, one-stage IPD-MA enables a more precise estimation of treatment effects and a detailed examination of moderators. This approach enhances statistical power, reduces bias and helps the identification of factors influencing treatment efficacy. Reference Clarke, Stewart, Tierney, Williamson and Clarke21 While a previous IPD-MA on this topic was published, Reference Moffa, Martin, Alonzo, Bennabi, Blumberger and Benseñor22 the current analysis expands upon that work by incorporating several large, recently completed clinical trials. In total, data from 18 data-sets comprising 1246 individual patient records were included, doubling the sample size of the earlier IPD-MA and establishing this as the most comprehensive data-set in the field of tDCS to date.
Method
This study was registered in PROSPERO (CRD420250652413), and it was guided according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses of Individual Patient Data (PRISMA) (Supplementary Material – Appendix 1 available at https://doi.org/10.1192/bjp.2025.10511) reporting guidelines for IPD-MA. Reference Stewart, Clarke, Rovers, Riley, Simmonds and Stewart23
Search strategy
An updated systematic review was conducted, building on a previously published meta-analysis by our group, Reference Razza, Palumbo, Moffa, Carvalho, Solmi and Loo24 which included studies from inception through January 2020. The search strategy here covered the period from the 1 January 2020 to the 1 February 2025 and was performed using the same search strategy as a previous published meta-analysis Reference Razza, Palumbo, Moffa, Carvalho, Solmi and Loo24 conducted in MEDLINE/PubMED, Embase, Web of Science and Cochrane Library databases, including keywords related to ‘tDCS’, ‘depression’ and ‘clinical trials’ (the detailed search strings are provided in the Supplementary Material – Appendix 2). The literature search and initial eligibility screening were conducted independently by the first and fourth authors by using Rayyan (Rayyan Systems Inc., Doha, Qatar), a web-based tool for systematic reviews (accessed via Windows; https://www.rayyan.ai/). Reference Ouzzani, Hammady, Fedorowicz and Elmagarmid25 Any disagreements were resolved with the last two authors (A.R.B. and M.-A.V.).
Newly identified RCTs were combined with those from the previously published meta-analyses. Reference Razza, Palumbo, Moffa, Carvalho, Solmi and Loo24 The authors of all included studies were contacted to request individual patient data and clarify methodological details. Data were first extracted on 15 February 2025.
Eligibility criteria
Only randomised, sham-controlled trials involving adult participants in an acute depressive episode that was associated with a diagnosis of major depressive disorder, bipolar disorder or secondary depression were considered eligible. Trials were included if they compared active prefrontal tDCS with sham tDCS in an acute treatment regimen, whether tDCS was used as monotherapy or in combination with other treatments, with at least five tDCS sessions.
Requested and analysed data
To obtain IPD, the corresponding authors of the included studies were contacted via email. Each author was provided with a standardised data request outlining the required variables, including demographic characteristics (e.g. gender, age), psychiatric comorbidities and clinical severity markers (e.g. duration of the current depressive episode, treatment-resistant depression status). Additionally, information on treatment history, including concomitant medications and depression subtype (e.g. unipolar, bipolar depression), was requested. If no response was received within a few weeks, two additional follow-up emails were sent. If no reply was obtained after three email attempts, the study was not included in the IPD analysis.
Data from active and sham tDCS groups were required. However, some studies also included additional interventional arms. Reference Borrione, Cavendish, Aparicio, Luethi, Goerigk and Ramos14,Reference Brunoni, Valiengo, Baccaro, Zanão, de Oliveira and Goulart26,Reference Brunoni, Moffa, Sampaio-Junior, Borrione, Moreno and Fernandes27 In such cases, if authors provided data from all arms, only data from the active and sham tDCS arms were included in the IPD-MA.
The IPD were checked by one author (A.P.D.), including variable harmonisation, missing data checks and unit consistency. Authors were contacted as needed. Cleaned data were then merged into the main analysis database.
Outcome measures
The main goals of this study were to investigate (a) depression improvement from baseline to end-point as defined per study (Δ score: end-point – baseline scores); (b) clinical response (defined as ≥50% improvement in the depression scale) and remission (defined as per study). When a study used more than one clinical depression scale, the main scale was used. Two studies did not report remission criteria. Therefore, standard thresholds were applied: for Palm et al (2012), Reference Palm, Schiller, Fintescu, Obermeier, Keeser and Reisinger28 using the Hamilton Depression Scale-24 (HAMD-24), remission was defined as a score ≤10; for Lee et al, Reference Lee, Lee, Jang, You, Park and Ji29 using the HAMD-17, remission was defined as a score ≤7; (c) moderators of clinical improvement:
-
(a) clinical variables: treatment-resistant depression (TRD) status (yes/no, defined as failure to respond to at least two antidepressant), depression subtype (unipolar, bipolar or secondary depression), baseline symptom severity (depression score at baseline) and medication use (due to data-set heterogeneity, individual-level medication data could not be analysed; instead, a study-level variable indicating whether concurrent medication was allowed (yes/no) was used);
-
(b) tDCS protocol parameters: total number of stimulation sessions, session duration, total treatment duration in weeks, cathode electrode placement (F4 versus non-F4) and initial combination with other intervention;
-
(c) study-level metadata: publication year, sample size;
-
(d) demographic data: age and gender.
The Brunoni et al study was divided into two data-sets: one without sertraline (including pure tDCS and pure sham conditions) and another with sertraline (active and sham tDCS combined with sertraline). Reference Brunoni, Valiengo, Baccaro, Zanão, de Oliveira and Goulart26 Moreover, for the cross-over trial by Palm et al, only data from the initial sham-controlled phase were included in the analysis. Reference Palm, Schiller, Fintescu, Obermeier, Keeser and Reisinger28
For the Aust et al and Burkhardt et al studies, only the baseline and post-intervention 6-week time points were analysed, as the follow-ups at weeks 18 and 30 were passive and did not involve any intervention sessions. Reference Burkhardt, Kumpf, Crispin, Goerigk, Andre and Plewnia13,Reference Aust, Brakemeier, Spies, Herrera-Melendez, Kaiser and Fallgatter15
Quality assessment
The methodological quality of each study was evaluated per the Cochrane Risk of Bias tool, version 2, Reference Higgins, Altman, Gotzsche, Juni, Moher and Oxman30 that assesses the risks of selection, performance, detection, attrition, reporting and other biases according to standardised criteria. Risk of bias was evaluated by two independent authors (L.B.R. and C.C.) that classified studies in an overall risk of bias category (Supplementary Material – Appendix 3).
Statistical analysis
Statistical analyses were conducted using R software (version 4.4.0 for Windows; R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org/). A one-stage IPD-MA approach was applied to examine depression improvement (delta scores) and response and remission rates, implemented through (generalised) linear mixed-effects models (package lmer), having group (active/sham tDCS) as the fixed factor and ‘study’ as the random effect. Model assumption was confirmed with inspection of a quantile–quantile (Q–Q) plot of the residuals. Missing values were imputed using the last observation carried forward (LOCF) method, except for participants with only baseline measurements.
The main effect size was standard mean difference (Hedges’ g) and odds ratio, for depression improvement and response/remission rates, respectively. Acceptability was also measured using number of drop-outs per group (i.e. attrition rate). Heterogeneity between studies was assessed using the I 2 statistics (>50% for substantial heterogeneity). Small study effects were checked via funnel plots (and Egger’s test in a two-stage IPD-MA). Sensitivity analysis was performed using a leave-one-out approach. Forest plots were done using the package forestplot. A two-stage IPD-MA (aggregate data) was performed to confirm the findings.
Potential moderators of depression improvement (delta scores) were tested using clinical, demographic and methodological variables that presented relatively low variability across the provided trial data. All continuous variables were standardised (mean-centred and scaled to a standard deviation of 1 to improve model convergence). First, univariable analyses were conducted, testing one moderator at a time. Linear mixed models were used with delta as the outcome, group*moderator as fixed effects and study as a random intercept. This was followed by a multiple regression including moderators that showed a p-value ≤ 0.05 in the main or interaction effects in the univariable analyses. These variables were then entered together as interaction terms with the treatment group to assess moderation.
Significant results were defined as those with an α level ≤0.05.
Results
The literature search yielded 730 studies that were combined with the 23 RCTs from a previous meta-analysis. Reference Razza, Palumbo, Moffa, Carvalho, Solmi and Loo24 A total of 29 RCTs met inclusion criteria, of which 17 (18 data-sets) shared individual patient data Reference Burkhardt, Kumpf, Crispin, Goerigk, Andre and Plewnia13–Reference Aust, Brakemeier, Spies, Herrera-Melendez, Kaiser and Fallgatter15,Reference Brunoni, Valiengo, Baccaro, Zanão, de Oliveira and Goulart26–Reference Lee, Lee, Jang, You, Park and Ji29,Reference Brunoni, Boggio, De Raedt, Benseñor, Lotufo and Namur31–Reference Loo, Sachdev, Martin, Pigot, Alonzo and Malhi40 (Fig. 1). The shared data included 1409 patients (mean age 42.9 years, s.d. = 13.5; 65.5% female), of whom 1246 were included in the IPD-MA (active tDCS: n = 651, mean age 44, s.d. = 13.6, 63.1% female; sham tDCS: n = 595, mean age 42.5, s.d. = 13.9, 63.7% female) (Table 1) and 163 were excluded due to allocation to other treatment arms (see Requested and Analysed Data section). Most studies (90%) presented a low risk of bias (Supplementary Material – Appendix 3).
Preferred reporting items for systematic reviews and meta-analyses of individual patient data flow chart.
IPD, independent patient data.
Summary of the included studies
tDCS, transcranial direct current stimulation; CBT, cognitive behavioural therapy; MDD, major depressive disorder; MADRS, Montgomery–Asberg Depression Rating Scale; HDRS, Hamilton Depression rating Scale; iBT, internet-based therapy; CCT, cognitive control training; NA, not available; SOR, supraorbital right.
Data of the end-point score were imputed for 37 patients in the sham group, and for 50 patients in the active group, representing 6.2 and 7.7% of the total sample, respectively.
Main outcomes
For depression improvement, delta scores were the primary outcome. The active tDCS group showed a small, yet statistically significant, improvement compared with the sham group (estimate 1.68, p < 0.001, Hedges’ d = 0.24, 95% CI 0.11–0.36) (Fig. 2). Heterogeneity was low to moderate (I 2 = 32%), suggesting consistent effects across studies.
Forest plot continuous outcome (Hedges’ g). s.m.d., standardised mean difference.
Similarly, for response, the active tDCS group presented a modest but statistically significant higher response rate (32.6%) compared with the sham group (27.1%) (estimate 0.29, p = 0.013, odds ratio 1.33, 95% CI 1.04–1.73) (Fig. 3(a)). Heterogeneity was also low to moderate for this measure (I 2 = 40%). Remission rate only showed a statistical trend to significance between the active (24.6%) and sham (20.5%) groups (estimate 0.26, p = 0.07, odds ratio 1.30, 95% CI 0.98–1.74) (Fig. 3(b)).
Forestplot for (a) response and (b) remission rates (odds ratio). Palm et al Reference Palm, Schiller, Fintescu, Obermeier, Keeser and Reisinger28 reported no responders or remitters in either group, while Blumberger et al Reference Blumberger, Tran, Fitzgerald, Hoy and Daskalakis39 reported no remitters. Therefore, a continuity correction of 0.5 was applied in these cases to allow the forest plot to be constructed.
A complementary two-stage IPD-MA yielded similar results, supporting the observed effects (Supplementary Material – Appendix 4). This was followed by an aggregate meta-analysis including all published studies in the field (Supplementary Material – Appendix 5), which showed a higher effect size (Supplementary Material – Appendix 6).
Funnel plot and sensitivity analysis
Visual inspection of the funnel plot suggested potential publication bias, which was further supported by Egger’s test in the two-stage IPD-MA (p = 0.04). Therefore, a sensitivity analysis was conducted by excluding the smallest and most asymmetrical study Reference Segrave, Arnold, Hoy and Fitzgerald38 (Supplementary Material – Appendix 7). After this exclusion, Egger’s test in the two-stage IPD-MA no longer indicated significant publication bias (p = 0.10) and results remained consistent for depression improvement (Hedges’ g = 0.23, 95% CI 0.10–0.34).
Leave-one-out sensitivity analysis was performed to systematically exclude one of the 18 data-sets at a time to assess the robustness of the findings and confirmed that the overall effect size remained stable, indicating that no study disproportionately influenced the results (Supplementary Material – Appendix 8).
Uni- and multivariable analyses
Univariable analyses were conducted using all variables listed in the Outcome Measures section. Moderators with a p-value ≤ 0.05 for either the main effect or the interaction with group were included in the multivariable analysis. TRD status, baseline symptom severity and the total number of tDCS sessions showed significant main effects, while year of publication and sample size showed significant interactions with group (Table 2).
Univariable and multivariable analyses
TRD, treatment-resistant depression; MDD, major depressive disorder; tDCS, transcranial direct current stimulation.
Bold text indicates statistically significant results (p ≤ 0.05).
In the multivariable model, baseline severity (estimate −1.89, s.e. = 0.35, p < 0.001), TRD status (estimate 1.95, s.e. = 0.77, p = 0.01) and number of tDCS sessions (estimate −2.1, s.e. = 0.61, p < 0.001) remained significant predictors of treatment outcomes. These results suggest that greater symptom improvement was observed among patients with more severe baseline depression and those who received more tDCS sessions, while individuals with TRD showed less improvement. However, no significant interactions were found between these variables and the treatment group (active versus sham).
By contrast, the previously significant interaction between year of publication and group lost significance in the multivariate model (estimate 0.60, s.e. = 0.57, p = 0.29), whereas sample size retained a significant interaction with group (estimate 1.27, s.e. = 0.58, p = 0.03), showing that larger RCTs presented smaller differences between active and sham tDCS groups (Table 2).
Acceptability
A total of 83 out of 651 (12.7%) patients in the active group and 67 out of 595 (11.3%) patients in the sham group did not complete the trials. The difference was not statistically significant between groups (odds ratio 1.13 (95% CI 0.80–1.61)).
Discussion
This study aimed to evaluate the efficacy of tDCS for depressive symptoms and explore potential moderators of treatment response through an IPD-MA. Comprising data from 1246 patients (651 active tDCS, 595 sham tDCS) across 18 data-sets, it represents the largest IPD-MA conducted to date in brain stimulation, doubling the sample size of previous analyses (n = 572). The results showed that active tDCS was associated with statistically significant, though small, improvements in depressive symptoms and response rates compared with sham. Remission rates showed a non-significant trend favouring tDCS. Drop-out rates did not differ between groups, suggesting similar acceptability. Multivariable analyses of moderators of depression improvement identified TRD, baseline symptom severity and the number of sessions showing a significant main effect in modulating depression improvement. Patients with TRD benefited less, while those with more severe symptoms and those who received more sessions showed greater improvements. However, only the sample size of the study significantly interacted with the treatment group, indicating that the difference between active and placebo groups diminished in more powered studies.
By incorporating individual patient data from both earlier studies and more recent large-scale trials, Reference Burkhardt, Kumpf, Crispin, Goerigk, Andre and Plewnia13–Reference Aust, Brakemeier, Spies, Herrera-Melendez, Kaiser and Fallgatter15 this updated analysis offers a more nuanced assessment of tDCS efficacy for depression symptoms. Although overall effects were attenuated compared with the earlier IPD-MA, Reference Moffa, Martin, Alonzo, Bennabi, Blumberger and Benseñor22 the results indicate that active tDCS was more effective than placebo. Notably, remission was no longer statistically significant in this updated analysis. It is also important to note that this meta-analysis used more relaxed criteria compared with previous ones, allowing for the inclusion of individuals with secondary depression.
Although the effect size observed was small for overall depression improvement (Hedges’ g = 0.24, 95% CI 0.11–0.35), this finding aligns closely with large-scale meta-analyses evaluating the efficacy of first-line treatments for depression improvement, including medication and psychotherapy. For instance, a comprehensive network meta-analysis reported a s.m.d. of 0.30 (95% CI 0.26–0.34) across multiple classes of antidepressants, Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson and Ogawa3 while another reported a more conservative s.m.d. of 0.24. Reference Stone, Yaseen, Miller, Richardville, Kalaria and Kirsch41 Moreover, one of the largest meta-analyses to date on psychotherapy efficacy reported an effect size of 0.31 (with a 95% CI of 0.24–0.38) compared with the placebo Reference Cuijpers, Noma, Karyotaki, Cipriani and Furukawa42 when corrected for publication bias and considering high-quality studies. These findings highlight that small effect sizes are typical across therapeutic modalities for depression. In general, overall improvements using medication and psychotherapy tend to be small in mental health disorders. Reference Leichsenring, Steinert, Rabung and Ioannidis43
In turn, response and remission rates appear to be higher for both pharmacological and psychotherapeutic treatments. For example, Cipriani et al Reference Cipriani, Furukawa, Salanti, Chaimani, Atkinson and Ogawa3 found average response rates of 50–60% and remission rates around 35–40% for antidepressants in adults with MDD in the active arms, while Cuijpers et al Reference Cuijpers, Karyotaki, Weitz, Andersson, Hollon and van Straten44 reported psychotherapy response rates of approximately 54% and remission rates around 38% in the verum psychotherapy group. This is also seen for rTMS. Rates of response are approximately 40% for rTMS modalities in the active arm. Reference Vida, Sághy, Bella, Kovács, Erdősi and Józwiak-Hagymásy45 Although rTMS appears to yield higher response rates than tDCS, a direct head-to-head comparison has not been conducted, and network meta-analyses have not established rTMS as superior to tDCS. Reference Mutz, Vipulananthan, Carter, Hurlemann, Fu and Young46
In the moderator analyses, a significant interaction between sample size and treatment group was observed, suggesting that while active tDCS generally shows greater efficacy than sham, this difference tended to diminish in studies with larger samples. Although the interaction between year of publication and treatment group did not remain significant in the multivariable model, it initially pointed in the same direction – toward a reduction in treatment effect over time. This pattern may reflect the increasing sample sizes in more recent trials, which typically come with improved methodological rigour.
Interestingly, contrary to previous findings in the tDCS field, depression severity and the level of TRD were associated with symptom improvement but did not significantly influence the overall efficacy of tDCS. This suggests that while these clinical variables may broadly affect depression improvement, their effect may be more related to general non-specific responsiveness (i.e. placebo effect, regression to the mean, etc) rather than the specific effects of tDCS. Indeed, such factors are known to influence the trajectory of clinical improvement across various depression treatments. Reference Cusin and Colloca47,Reference Fournier, DeRubeis, Hollon, Dimidjian, Amsterdam and Shelton48 Notably, although previous IPD-MAs have suggested that TRD is a moderator of tDCS efficacy, they did not test interactions between moderators and treatment groups. Reference Moffa, Martin, Alonzo, Bennabi, Blumberger and Benseñor22,Reference Brunoni, Moffa, Fregni, Palm, Padberg and Blumberger49 As a result, they identified predictors of overall depression improvement across both active and sham conditions, making it difficult to determine whether TRD specifically modulates the efficacy of tDCS or simply reflects a general moderator of clinical improvement.
Similarly, a higher number of tDCS sessions were associated with greater improvement in depressive symptoms. However, this effect did not significantly differ between groups, suggesting that while more sessions were linked to better outcomes, the benefit was comparable across both conditions. One possible explanation could be increased patient engagement, which could also increase non-specific effects. Additionally, two out of the four studies that included a higher number of sessions (>20) were home-based trials, Reference Borrione, Cavendish, Aparicio, Luethi, Goerigk and Ramos14,Reference Lee, Lee, Jang, You, Park and Ji29 both of which reported null results across groups. These studies may have diluted any potential dose–response relationship by introducing greater variability in factors such as adherence, device usage and therapeutic environment, such as social interaction. Unfortunately, the individual patient data of Woodham and colleagues Reference Woodham, Selvaraj, Lajmi, Hobday, Sheehan and Ghazi-Noori50 – a home-based tDCS trial that reported a positive session-response relationship – was not provided, preventing a more comprehensive analysis of this variability. In fact, although home-based tDCS is a promising technique to reduce the need for daily travel to clinical centres, there are still uncertainties behind its optimal use, particularly regarding the challenge of properly managing treatment parameters without professional supervision, which can result in improper device use (commentary in Reference Borrione, Cavendish, Aparicio, Luethi, Goerigk and Ramos14 ).
Overall, the findings indicate that non-specific effects may substantially contribute to the observed outcomes. To accurately assess the specific effects of tDCS, future research should be designed to control for these influences. For instance, implementing a pre-treatment observation period to monitor spontaneous improvements (i.e. naturalistic observation period) could help identify and exclude individuals who improve before receiving tDCS, thereby minimising potential confounding effects.
Finally, an important insight from this IPD-MA is the observation of a lower effect size compared with the aggregate meta-analysis encompassing all published RCTs in the field of tDCS. A crucial point to highlight is the difference in methodological quality: the studies that provided IPD for this analysis were predominantly rated as having a low risk of bias, whereas nearly 85% of the studies that did not provide IPD (see Supplementary Material – Appendix 5) were judged to have suboptimal quality – 66.7% with a high risk of bias and 16.7% with some concerns. This indicates that the aggregated meta-analyses may present increased effect size with the great proportion of studies with higher bias risk. However, studies with very small sample sizes in the aggregate meta-analysis reported highly significant positive effects, which may have biased the overall results. Reference Sharafi, Taghva, Arbabi, Dadarkhah and Ghaderi51–Reference Fregni, Boggio, Nitsche, Marcolin, Rigonatti and Pascual-Leone53 The findings of this IPD-MA, based on higher-quality studies, therefore provide a more conservative and potentially more accurate estimate of tDCS efficacy.
To conclude, this IPD-MA, the largest to date in the field of tDCS for depression, shows that tDCS yields a statistically significant but small effect on depressive symptoms, with an acceptable profile compared with placebo. The current findings indicate that TRD and baseline depression are general predictors of clinical improvement, regardless of treatment allocation. Furthermore, findings revealed that larger samples reported smaller effect sizes, a trend likely associated with improved methodological rigour in more recent trials. Future research should not only clarify the specific mechanisms of tDCS and ensure stimulation protocols effectively engage these mechanisms, but also account for the role of non-specific therapeutic effects in depression.
Limitations
This study employed an IPD meta-analytic approach, which, while offering more granular and statistically powerful analyses than traditional meta-analyses, is inherently limited by the availability of shared data. Not all eligible studies in the tDCS literature were included, either due to lack of author response, data-sharing restrictions or data unavailability. This may have introduced selection bias, thereby potentially affecting the representativeness of the pooled sample. Second, although IPD allows exploration of individual-level moderators, the data-sets differed considerably in the range of variables provided. Key clinical information such as duration of the depressive episode, comorbidities, medication status or psychosocial factors was inconsistently reported or absent in several data-sets. This heterogeneity limited the ability to investigate further moderators of treatment improvement. Similarly, only a limited number of studies shared specific reasons for discontinuation, which limited a more detailed analysis of drop-out causes. As a result, only an analysis of all-cause drop-out was performed. Third, included studies varied in parameters such as electrode montage, number of sessions, session duration and treatment setting (e.g. supervised versus home-based), all of which could have had an impact on clinical outcomes. This protocol heterogeneity likely diluted potential treatment-specific effects and reduced the statistical power to detect consistent moderators of tDCS efficacy. As such, these findings further highlight the need for increased standardisation in future tDCS trials – including harmonisation of stimulation protocols and consistent reporting of clinical variables – to enable more precise and reproducible insights into treatment mechanisms.
Supplementary material
The supplementary material is available online at https://doi.org/10.1192/bjp.2025.10511
Data availability
The data used in this study consist of individual participant data pooled from multiple independent trials. Each data-set is held by the respective study investigator and is subject to institutional and ethical regulations. As such, the data are not publicly available. However, data may be available from the corresponding authors of each study upon reasonable request and subject to appropriate approvals and data-sharing agreements.
Author contributions
L.B.R.: conceptualisation, data curation, formal analysis, methodology, visualisation, writing – original draft. M.S.L.: formal analysis, writing – review and editing; A.M.: supervision, data provision, writing – review and editing. S.A.: data provision, writing – review and editing. D.M.B.: data provision, writing – review and editing. J.B.: data provision, writing – review and editing. G.B.: data provision, writing – review and editing. C.C.: data curation, writing – review and editing. C.K.L.: data provision, writing – review and editing. R.M.: data provision, writing – review and editing. W.M.: data provision, writing – review and editing. C.L.N.: data provision, writing – review and editing. U.P.: data provision, writing – review and editing. R.S.: data provision, writing – review and editing. L.V.: data provision, writing – review and editing. F.P.: data provision, writing – review and editing. A.P.D.: data curation, writing – review and editing. A.R.B.: upervision, project oversight, writing – review and editing. M.-A.V.: supervision, conceptualisation, methodology, writing – review and editing.
Funding
L.B.R. is supported by a Research Foundation – Flanders (FWO) grant (12AZO24N). M.-A.V. received funding from FWO for research projects for fundamental research (G044222N; G044019N). M.-A.V. is supported by a grant for a Concerted Research Action of the Special Research Fund of Ghent University (01G00623). D.M.B. receives research support from Canadian Institutes of Health Research, National Institute of Mental Health, the Wellcome Trust, Brain Canada and the Temerty Family through the CAMH Foundation and the Campbell Family Research Institute. C.K.L. is supported by a National Health and Medical Research Council Investigator Grant (1195651). G.B. is supported by the Medical Faculty of LMU Munich (1127; CS063). A.R.B. is supported by the Sao Paulo Research State Foundation (2019/06009-6 and 2022/11557-5).
Declaration of interest
D.M.B. has received research grant support and in-kind equipment support for an investigator-initiated study from Brainsway Ltd; was the site principal investigator for three sponsor-initiated studies for Brainsway Ltd; has received in-kind equipment support from Magventure for investigator-initiated studies; has received medication supplies for an investigator-initiated trial from Indivior; is a scientific advisor for Sooma Medical; and is a member of the Clinical Standards Committee of the Clinical TMS Society (unpaid). C.K.L. is on the Medical Advisory Board of Sigmastim and Douglas Pharmaceuticals. A.R.B. holds equity in Flow Neuroscience. No other disclosures were reported. The other authors have no competing interests to declare.
eLetters
No eLetters have been published for this article.