Noninvasive brain stimulation combined with evidence-based psychotherapy for psychiatric disorders: A meta-analysis of optimal implementation parameters

Abstract

Evidence-based psychotherapies are first-line treatments for psychiatric disorders, yet response rates remain suboptimal. Noninvasive brain stimulation (NIBS) may augment psychotherapy by modulating treatment-engaged circuits. We conducted a systematic review and meta-analysis of randomized controlled trials comparing active NIBS plus evidence-based psychotherapy versus sham NIBS plus psychotherapy. We searched six databases through February 2025, screening 1017 records. Twenty-eight trials (31 treatment arms; 1506 participants) met inclusion criteria. Active NIBS combined with psychotherapy produced significantly greater symptom improvement than sham NIBS with psychotherapy (SMD = − 0.38, 95% CI [−0.68, − 0.08]), with substantial heterogeneity. Moderator analyses revealed critical implementation parameters: repetitive transcranial magnetic stimulation (rTMS) showed significant benefit while transcranial direct current stimulation did not. Non-concurrent delivery showed significant effects; concurrent delivery did not. Cognitive behavioral therapy (CBT) combined with NIBS showed significant benefit; other psychotherapy modalities did not. Human-delivered psychotherapy significantly enhanced outcomes; computerized formats did not. Significant effects were observed only for anxiety disorders (SMD = −0.70, 95% CI [−1.26, −0.14]). While there was a null finding for depression, it likely reflects insufficient statistical power rather than true ineffectiveness. Secondary analyses found no significant effects on executive functioning or quality of life. Treatment integrity was under-reported: only 39.3% of studies used fully manualized protocols and 10.7% documented therapist adherence. Timing and modality are largely confounded in the current evidence base, and priming versus consolidation effects cannot be distinguished. These findings provide an evidence-based framework for optimizing combined treatment protocols and highlight the need for standardized psychotherapy fidelity monitoring.

1. Introduction

Psychiatric disorders impose a large burden of disability and economic loss worldwide (Arias et al., 2022). Evidence-based psychotherapies (EBPs), typically delivered in manualized formats and supported by empirical evidence, are first-line treatments across diagnoses (APA Presidential Task Force on Evidence-Based Practice, 2006; Cook et al., 2017). In posttraumatic stress disorder (PTSD), clinical guidelines from the U.S. Department of Veterans Affairs and Department of Defense strongly recommend manualized EBPs such as prolonged exposure, cognitive processing therapy, and eye movement desensitization and reprocessing over medications when feasible (Schnurr et al., 2024). In obsessive–compulsive disorder (OCD) and body dysmorphic disorder, international guidelines designate exposure and response prevention as the therapy of first choice (National Institute for Health and Care Excellence, 2005). In adult major depressive disorder (MDD), psychotherapy families, including cognitive behavioral therapy (CBT), behavioral activation, interpersonal psychotherapy, and problem-solving therapy, are efficacious and broadly comparable, with benefits that can persist up to 12 months (Cuijpers et al., 2021). Despite this efficacy, short-term response to psychotherapy in MDD is limited to approximately 40% and many do not remit (Cuijpers et al., 2021). Effects vary substantially across trials, and durability is limited: across randomized trials with long-term follow-up, relapse after psychotherapy occurs in roughly 39% of patients (Steinert et al., 2014). Taken together, these data establish EBPs as foundational yet underscore the need for strategies that are more targeted, more durable, and more adaptable to individual profiles, including potentially synergistic combination and/or sequencing with noninvasive brain stimulation (NIBS).

NIBS augments EBPs by directly modulating treatment-engaged circuits with greater focality and typically fewer systemic adverse effects compared with pharmacotherapies (Regenold et al., 2022). Two NIBS modalities are in clinical use: transcranial magnetic stimulation (TMS, including repetitive TMS [rTMS] and theta burst stimulation [TBS]) and transcranial electrical stimulation (tES). TMS generates intracranial electric fields via rapidly changing magnetic fields, which can depolarize neurons and modulate cortical network excitability and plasticity. In the U.S., TMS is cleared for MDD, OCD, smoking cessation, and migraine. For OCD specifically, the FDA-cleared protocol is adjunctive and incorporates brief, individualized symptom provocation immediately before each session (Carmi et al., 2019). tES, which encompasses transcranial direct current stimulation (tDCS), transcranial alternating current stimulation, and transcranial random noise stimulation, perturbs membrane potentials with weak currents and alters cortical excitability (Reed and Cohen Kadosh, 2018). Meta-analytic evidence supports tES superiority over sham for several psychiatric disorders, including MDD, OCD, PTSD, and anxiety disorders (Kalu et al., 2012; Xie et al., 2024). In December 2025, the U.S. FDA approved a home-delivered tDCS device for the treatment of moderate to severe major depressive disorder as monotherapy or adjunctive therapy (Bikson et al., 2026). Because psychotherapies harness learning-dependent plasticity (Craske et al., 2014; Hofmann et al., 2012), temporally aligning stimulation with therapeutic learning episodes provides a biologically grounded rationale for combining these modalities to amplify clinical gains.

The integration of NIBS with psychotherapy is a multimodal approach designed to exploit neural plasticity and the principle of state-dependency to optimize clinical outcomes (He et al., 2022; Saccenti et al., 2024; Sathappan et al., 2019). By time-locking brain stimulation with in-session psychotherapy processes, clinicians utilize “functional targeting” or “cortical paired associative stimulation” to synchronize the modulation of disorder-related neural circuits (e.g., prefrontal cortex), while those circuits are simultaneously engaged by cognitive or emotional interventions (Deng et al., 2020; Luber and Lisanby, 2014; Sathappan et al., 2019). This synergy is grounded in the idea that while psychotherapy promotes behavioral and neural plasticity, NIBS directly targets dysfunctional circuitry to facilitate long-term potentiation or depression, potentially producing faster and more durable therapeutic changes (Saccenti et al., 2024).

The current evidence base suggests potential overall benefit of pairing stimulation with psychotherapy, with larger effects dependent upon specific domains and designs. Across disorders, active rTMS paired with psychological interventions significantly improves overall clinical symptoms compared to sham combinations, yielding a small-to-moderate effect size (SMD = 0.31) (Xu et al., 2023). This multimodal approach appears particularly effective for anxiety, demonstrating a large effect size (Cohen’s d = −0.82) when brain stimulation is combined with mindfulness-based interventions (Demina et al., 2025). For depressive symptoms, the evidence is more nuanced; while some data show a small-to-medium effect (Cohen’s d = −0.24) that fails to reach statistical significance across all populations (Demina et al., 2025), the combination of NIBS and psychosocial therapy shows a significant positive effect specifically for moderate-to-severe depression (SMD = −0.46) (He et al., 2022). The reported magnitude of these benefits varies significantly based on the trial design and control conditions used. Uncontrolled pretest-posttest comparisons of active combinations show a very large therapeutic effect (Hedges’ g = −1.91), whereas controlled comparisons against sham rTMS paired with active non-pharmacological methods result in a medium effect size (Hedges’ g = −0.55) (Giron et al., 2025). Furthermore, when compared directly against NIBS monotherapy, the combined approach demonstrates a large and significant effect in reducing depression severity (SMD = −0.84) (He et al., 2022). Beyond mood regulation, integrating rTMS with cognitive training produces significant enhancements in global cognition (SMD = 0.45) (Xu et al., 2023). The augmentation effect of rTMS on CBT also becomes statistically significant when treatments include ten or more sessions (SMD = 0.21) (Xu et al., 2023). While most findings are small-to-moderate, some pilot research has reported exceptional outcomes, such as tDCS paired with positive psychotherapy reaching a very large effect size (Cohen’s d = 1.99) compared to stimulation alone (Kochanowski et al., 2024). However, active combinations have not yet consistently outperformed active rTMS when compared against protocols using sham psychological interventions (Giron et al., 2025; He et al., 2022).

Despite the therapeutic promise, the integration of NIBS and psychotherapy faces substantial feasibility and technical hurdles. Concurrent administration can be difficult; for example, the percussive noise and somatic sensations of TMS can hinder mindfulness practice, leading to increased anxiety and high dropout rates (Cavallero et al., 2021). Attrition is a significant concern, with some trials reporting up to a 40% loss to follow-up (Demina et al., 2025; Xu et al., 2023). Timing remains a primary uncertainty; optimal temporal alignment between stimulation and psychotherapy is not established, and protocols currently use three strategies: priming (NIBS before psychotherapy to increase plasticity), online/synergistic (NIBS during psychotherapy to exploit state dependence), and consolidation (NIBS after psychotherapy to stabilize learning). Another challenge concerns psychotherapy treatment integrity, which consists of therapist adherence to the prescribed theoretical model and the competence to deliver those techniques skillfully and appropriately (Esposito et al., 2024). Systematic reviews indicate that treatment integrity is an important dimension associated with favorable clinical outcomes (Esposito et al., 2024; Power et al., 2022); however, this factor is not always systematically accounted for or reported in combination studies or meta-analyses. For example, in some combined protocols, the psychological component is reduced to audio-guided or modified versions that lack the substantive clinician–patient relationship necessary to navigate the negative emotional states or psychological discomfort that may arise during stimulation (Cavallero et al., 2021). The efficacy of combined treatments may be further blunted by external factors, such as the concurrent use of benzodiazepines or antipsychotics, which can interfere with brain response to stimulation (Kochanowski et al., 2024). Finally, widespread implementation remains limited by the high cost and time commitment required for multi-session combined protocols. Taken together, these issues argue for larger, standardized randomized controlled trials to determine feasible schedules and optimal timing, and ensure psychotherapy integrity (Saccenti et al., 2024; Tatti et al., 2022; Xu et al., 2023).

The present meta-analysis quantifies the efficacy of NIBS delivered adjunctively with EBPs in psychiatric populations and tests moderators with clinical and mechanistic salience. We restrict inclusion to clinical psychiatric samples and EBPs. We model moderators spanning NIBS factors (modality: rTMS, tES; timing of stimulation relative to therapy) and psychotherapy factors (treatment modality; treatment integrity), as well as diagnostic category and treatment-resistance status. Our central analysis evaluates whether the temporal relationship between stimulation and psychotherapy modulates effect size. By isolating these parameters, we seek to generate practical recommendations for protocol design and a mechanistic account of how NIBS augments psychotherapy to support durable remission. Our goal is an actionable framework for when and how to combine NIBS with EBPs to improve durability and remission in psychiatric care.

2. Methods

This quantitative meta-analysis was performed according to the recommendations of the Cochrane group. Procedures included a comprehensive database search, duplicate removal, dual independent screening at title/abstract and full-text stages according to prespecified inclusion/exclusion criteria, quality and risk-of-bias assessments, data extraction, and quantitative synthesis. The protocol was registered with the International Prospective Register of Systematic Reviews (PROSPERO; registration number CRD42024570287).

2.1. Literature search

On February 5, 2025, one investigator (EW) executed searches in the following databases: Scopus, Web of Science, PubMed, Embase, the Cochrane database, and PsycINFO, without date limits and restricted to English-language records. Our chosen keywords for the database search were: (“psychotherapy” OR “psychological intervention”) AND (“noninvasive brain stimulation” OR “NIBS” OR “repetitive transcranial magnetic stimulation” OR “rTMS” OR “transcranial electric stimulation” OR “TES” OR “transcutaneous auricular vagus nerve stimulation” OR “taVNS” OR “transcranial direct current stimulation” OR “tDCS” OR “transcranial focused ultrasound” OR “tFUS” OR “non-invasive brain stimulation” OR “TMS” OR “transcranial magnetic stimulation” OR “TBS” OR “theta burst stimulation” OR “transcranial bio-photomodulation” OR “tPBM” OR “transcranial alternating current stimulation” OR “tACS” OR “transcranial random noise stimulation” OR “tRNS” OR “temporal interference stimulation” OR “TI”) AND (“randomized controlled trial” OR “RCT”).

2.2. Eligibility criteria

The following selection criteria were applied to papers identified by our search, and based on the Population, Intervention, Comparison, Outcome and Settings framework (PICOS) (Moher et al., 2009). The following criteria were used:

2.2.1. Participants

Males and females aged 15 years or older with psychiatric disorders (including mood disorders, anxiety disorders, psychotic disorders, trauma and stressor related disorders, substance use disorders, eating disorders, impulsive/compulsive disorders, dissociative disorders, somatoform disorders, conduct disorders, and personality disorders); or neurodevelopmental disorders (including autism spectrum disorder and attention-deficit/hyperactivity disorder). Studies in which patients were not formally diagnosed with a DSM–5 disorder were excluded. Ultimately, no studies with neurodevelopmental disorders as the diagnostic category of interest met our inclusion criteria.

2.2.2. Intervention

The included interventions involved NIBS combined with evidence-based psychotherapy. Evidence-based psychotherapy refers to scientifically validated treatments tailored to clients’ individual needs. The National Institute of Mental Health emphasizes that rigorous research has shown that evidence-based therapies reduce psychopathology severity and improve quality of life effectively, and these therapies are generally supported by at least two randomized controlled trials. Studies including cognitive training in tandem with NIBS were not considered to include evidence-based psychotherapy and were excluded.

2.2.3. Comparison

Studies in which the experimental condition was active NIBS with active evidence-based psychotherapy, and the control condition sham NIBS with active evidence-based psychotherapy were included. Studies in which the control condition was sham NIBS alone without psychotherapy, or active NIBS without psychotherapy, or in which no stimulation was considered sham NIBS were excluded.

2.2.4. Outcome

Psychopathology symptoms reported in original studies as the primary or secondary outcome were included. If means and standard deviations were not reported and could not be accessed when contacting the authors, the study was excluded. Studies that did not include psychopathology improvement as an outcome measure (i.e., only included changes in cognitive function or physiological characteristics) were excluded.

2.2.5. Study design

We included randomized controlled trials (including parallel group and crossover designs). Reviews, case report studies, and re-analyses of the same dataset were excluded.

2.3. Title, abstract, and full text screening

After deduplication (EW), two independent screeners (EW, NB) evaluated titles/abstracts against eligibility criteria. Discrepancies were discussed and resolved with the rest of the team, and inter-rater reliability was assessed across the corpus. Next, the included studies underwent a full-text screening, performed independently by the same two investigators. Discrepancies were also discussed and resolved with the rest of the team, and inter-rater reliability was assessed.

2.4. Clinician screening to evaluate fidelity to evidence-based psychotherapy

Three clinicians (AN, EJ, BG) independently screened the included articles based on the following criteria: 1) the presence of a valid psychiatric diagnosis in participants, as defined by the DSM–5 and confirmed through the use of appropriate evidence-based rating scales; 2) whether the psychotherapy was evidence-based, meaning it followed a treatment manual and had at least two randomized controlled trials supporting its efficacy (Chambless and Hollon, 1998; Tolin et al., 2015); and 3) the way in which psychotherapy was administered, which could be computerized, administered by a human, or a combination of both. In the absence of evidence-based treatments, the clinicians rated whether some evidence-based elements from an evidence-based treatment were included in the type of psychotherapy provided. If no elements of evidence-based treatments were administered, the study was excluded. Additionally, the clinicians recorded the time spent on therapeutic tasks and evaluated whether adherence to the therapy protocol was assessed, that is, whether the study explicitly stated that adherence measurement was conducted. Each study was independently rated by exactly two of the three clinicians in a balanced distributed design, and inter-rater reliability was assessed by computing Cohen’s κ for each clinician pair separately and averaging across pairs. After the screening, all clinicians and the research team convened to discuss and resolve discrepancies.

2.5. Quality assessment

The Cochrane risk-of-bias tool 2 (RoB 2) was used to assess the possibility of bias presented by the primary outcome in each of the 28 randomized controlled trials examined (Sterne et al., 2019). The RoB 2 is an adapted version of the original Cochrane risk-of-bias tool that addresses user- and expert-identified limitations, advances its methodology, and includes structured signaling questions and an algorithm to provide evaluators with suggested risk levels (Higgins et al., 2011, 2019; Sterne et al., 2019). RoB 2 signaling questions span five domains: bias due to the randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Each signaling question can be answered in one of five ways: “Yes,” “Probably yes,” “Probably no,” “No,” or “No information.” Each domain’s respective signaling questions are ultimately considered together, allowing the evaluator to classify each domain as having a “low” risk of bias, “high” risk of bias, or “some concerns.” Evaluators then decide on the overall risk of bias. Two investigators (LB, EG) thoroughly reviewed the Cochrane guidelines and each signaling question’s criteria (Higgins et al., 2019; Sterne et al., 2019). Evaluators independently completed the RoB 2 for the primary outcome measure of each study, and discrepancies were resolved through a discussion with three additional research team members.

2.6. Data extraction

Two investigators (EW, NB) independently extracted relevant data. First, the general characteristics of each study were collected, such as sample size, participant characteristics (age, percentage female, concurrent medications) across active and sham groups, as well as participants’ diagnoses.

Information about NIBS protocols (i.e., NIBS modality, stimulation target and targeting approach), stimulation amplitude (in mA for tES vs. percent of motor threshold for rTMS), stimulation frequency (only for rTMS), stimulation duration, type of sham control, and the NIBS administration schedule (i.e., the number of sessions per week); and psychotherapy (i.e., type of psychotherapy and therapy schedule [number of sessions per week]) were extracted. We also extracted information regarding the timing of NIBS relative to psychotherapy (concurrently or non-concurrently), to draw more complete inferences on the role of brain state in the efficacy of combined psychotherapy-NIBS treatment approaches.

Finally, the same investigators extracted the means and standard deviations of active/sham pre-/post-treatment for: primary outcome, quality of life outcome (if available), executive functioning outcome (if available), anxiety and depression outcome (regardless of diagnosis, if available). If the primary outcome measure was not a psychopathology outcome (e.g., physiological or cognitive), we consulted the corresponding author and selected the most relevant secondary clinical outcome listed. Only one of our included studies (Osuch et al., 2009) utilized a crossover design with a 2-week washout period between active and sham rTMS phases. We extracted active versus sham means and standard deviations in the same manner as all parallel-group studies. We note that the linear mixed models applied by the original authors modeled active versus sham differences over time but did not explicitly account for sequence or carryover effects; the potential impact of psychotherapy-specific carryover across phases is therefore acknowledged as a limitation. A sensitivity analysis excluding this study confirmed that primary conclusions were not materially affected (see Section 4.4).

2.7. Categorization of variables

Some variables required categorization to have a large enough sample size per variable for analysis. NIBS modalities used in the studies included tDCS, rTMS, deep TMS (dTMS), and intermittent theta burst stimulation (iTBS). We combined rTMS, dTMS and iTBS into one category called “rTMS.” No screened studies utilizing forms of tES other than tDCS appeared in the eligible literature; tACS, tRNS, and other tES modalities were absent from the pool of studies meeting inclusion criteria. Consequently, the tES category in all analyses reflects tDCS exclusively, and findings should not be generalized to other tES modalities. There were also many distinct psychotherapies utilized across studies, and, via a discussion with the clinicians, three global categories were generated: Cognitive Behavioral Therapy (CBT), mindfulness, and exposure. CBT included any therapies based on cognitive strategies to process thoughts and behaviors. Mindfulness included any therapies described as using mindfulness techniques. Exposure included symptom provocation and exposure and response prevention therapies. Lastly, with the help of the clinicians, we categorized diagnoses included in the studies into the following categories: OCD, depression, anxiety (studies combining the diagnoses of specific fears, acrophobia, social anxiety, and panic disorder), addiction (studies combining nicotine addiction, methamphetamine addiction, and alcohol use disorder), PTSD, transdiagnostic (studies that allowed any DSM–5 disorder), pain (including fibromyalgia, which is listed under somatic symptom disorder in the DSM–5), and personality (including borderline personality disorder [BPD]).

2.8. Quantitative Analysis

All statistical analyses were conducted using R (v.4.5.2), with the “meta” and “metafor” packages, which include a collection of functions for conducting meta-analyses (https://CRAN.R-project.org/package=meta). We used the “metacont” function to compute standardized mean differences (SMD) with 95% confidence intervals under a random-effects model. Heterogeneity was summarized with τ² and I². In cases where standard deviations were unavailable for clinical outcomes (n = 2), we used the pooled standard deviation from all other studies (an average of all available standard deviations). Publication bias was assessed using a funnel plot of the primary outcome measures and Egger’s test (Egger et al., 1997).

2.8.1. Primary analysis

The primary clinical outcome defined by the study was extracted. For studies that had physiological outcomes as their primary measure, the most relevant secondary clinical outcome was used. In the rare event where an increase in clinical outcome score signaled improvement, we inverted the scores to match the direction of change of the rest of the outcomes. Further, subgroup analyses were conducted with the following variables: NIBS modality (tDCS or rTMS), NIBS timing (concurrent or non-concurrent), therapy category (mindfulness, CBT, or exposure), diagnostic category (OCD, depression, anxiety, addiction, PTSD, transdiagnostic, pain, or personality), the degree to which the psychotherapy was evidence-based (i.e., whether it was a fully evidencebased intervention as defined by a manual, or if it only contained partial elements derived from evidence-based practice), and administration method (human, computerized, or mixed). We intended to conduct a subgroup analysis comparing studies where adherence was rated and reported, and studies lacking this information. However, because only one study reported adherence ratings, this analysis could not be conducted. Interactions between variables were also tested: NIBS modality by timing, therapy category by diagnostic category, and therapy category by NIBS modality. Further, if a category, or a combination of categories, did not contain at least three studies, we did not include that category in our subgroup analysis.

2.8.2. Secondary analysis

We performed secondary analyses for the following domains: depression, anxiety (regardless of the diagnosis), executive functioning, and quality of life. Only studies that listed an outcome in each domain were included in each secondary analysis, and, given sample size, we only conducted subgroup analyses within each domain for NIBS modality and NIBS timing.

3. Results

3.1. Search results

Search criteria yielded 1728 citations. After removing duplicates (n = 711), 1017 articles underwent abstract and title screening, with 31 total disagreements between the two evaluators (Cohen’s κ = 0.94). A total of 668 articles were kept, and then the full text screening was done, with 10 total disagreements between the two evaluators (Cohen’s κ = 0.97). 45 papers passed the full text screening and were sent to the clinicians for rating the level of evidence supporting the psychotherapy, after which 31 papers remained eligible, with mean κ = 0.74, indicating substantial agreement. However, data were missing for three of the papers despite contacting the authors, those publications were not included later in the meta-analysis. Thus, 28 studies were included in the final cohort (Fig. 1).

Fig. 1.

Consort diagram of randomized controlled trials (RCTs) included in the meta-analysis.

3.2. Descriptive statistics

The following sections provide a general overview of the NIBS parameters, clinical populations and psychotherapies used in combination with NIBS. The following results are expressed as a percentage out of the 28 included studies, except for the stimulation target. As three of the studies (Fitzsimmons et al., 2025; Hu et al., 2022; Neacsiu et al., 2022b) included two active arms, each with different targets, we present the results from 31 total arms across 28 studies. More details about each study are provided in the Supplementary Material.

3.2.1. NIBS

Of the 28 studies identified by our search, only studies using TMS (64.3%) and tDCS (35.7%) have been combined with various psychotherapy approaches in clinical trials matching our criteria. No studies combining tACS, tRNS, or tFUS with psychotherapy were published.

For TMS studies, most used conventional rTMS protocols (72.2%), while some used intermittent theta-burst stimulation (27.8%). To determine the stimulation site, the predominant targeting method was scalp measurement, with or without the 10–20 EEG grid as a reference (83.3%). A minority of studies used fMRI-based targeting (11.1%) or targeting based on an anatomical MRI scan (5.6%). Of the 31 arms in our analysis, rTMS stimulation targets included the left prefrontal cortex (PFC; 38.7%), medial PFC (25.8%), right PFC (22.6%), motor cortex (9.7%), and bilateral PFC (3.2%). As expected, given the FDA-approved label for treatment, the most frequently used stimulation frequency was 10 Hz (44.4%), followed by 5 Hz (22.2%), 1 Hz (11.1%), 20 Hz (11.1%), 18 Hz (5.6%), and 50 Hz (5.6%). The stimulation amplitudes used were 80% rMT (27.8%), 100% rMT (27.8%), 110% rMT (16.7%), and 120% rMT (27.8%).

All studies using tDCS relied on scalp measurements/EEG grid to define their stimulation target. The most common target was either the left PFC (30%) or the medial PFC (30%), followed by the right PFC (20%) and the motor cortex (20%) (Fig. 2B). Most studies used a stimulation amplitude of 2 mA (80%), with only one study using 1.7 mA (10%) and another using 1.5 mA (10%) for increased comfort during stimulation (see Supplementary Table S2 for more NIBS details).

Fig. 2.

Study landscape and protocol characteristics in trials combining NIBS with psychotherapy. (A) Annual number of included randomized trials by primary diagnostic group (stacked colors: Addiction, Anxiety, BPD, MDD, OCD, Pain, PTSD, Transdiagnostic). (B) NIBS modalities and targeting: TMS accounted for 64% of trials and tDCS for 36%; TMS targeting was predominantly scalp/EEG-based (83.3%), with fewer MRI-guided approaches (fMRI 11.1%, sMRI 5.6%). (C) Psychotherapy categories and integrity reporting: Exposure-based interventions were most common (46%), followed by CBT (36%) and mindfulness (18%). Treatment adherence/competence (“integrity”) was explicitly assessed in only 20% of the CBT trials. (D) Delivery timing and format: Non-concurrent scheduling (stimulation before/after psychotherapy) was more frequent (57%) than concurrent delivery (43%). Psychotherapy was delivered by a human clinician (53.6%), in hybrid formats (28.6%), or computerized (17.9%). Percentages reflect the proportion of studies within the included RCT sample (n = 28). Abbreviations: NIBS, noninvasive brain stimulation; TMS, transcranial magnetic stimulation; tDCS, transcranial direct current stimulation; CBT, cognitive behavioral therapy; BPD, borderline personality disorder; MDD, major depressive disorder; OCD, obsessive–compulsive disorder; PTSD, posttraumatic stress disorder.

3.2.2. Clinical diagnosis, therapy, and combination of therapy with NIBS

Across all studies, the average total sample size was 53.8 ± 33.6 subjects. Clinical populations included patients with anxiety (21.4%), PTSD (21.4%), addiction (17.9%), OCD (10.7%), and pain (10.7%). Only a small percentage of the studies included patients with MDD (7.1%), transdiagnostic emotion dysregulation (7.1%), and BPD (3.6%). Concurrent medication was allowed in most studies (78.6%), with the remaining studies prohibiting medication (10.7%) or not specifying medication status (10.7%). The most common type of therapy used was exposure (46.4%), followed by CBT (35.7%), and mindfulness (17.9%). Across all therapy types, the average total time in therapy was 448.5 min (~7.5 h), with the average session total being 9.6 sessions. Most studies employed a therapist (53.6%), while some relied on a mix of therapist interaction and computerized elements (28.6%), and several used computerized interventions only (17.9%). A minority of studies used a fully evidence-based procedure (39.3%), with most using only elements of an evidence-based protocol (60.7%). This classification was performed via a discussion between clinicians (AN, EJ, BG). Of note, only 3 studies (10.7%) reported adherence measures for the therapeutic protocol used (Fig. 2C).

Finally, regarding the combination of NIBS with psychotherapy, the majority of studies delivered NIBS right before or right after the psychotherapy intervention (57.1%), with the rest concurrent (Fig. 2D). Four studies (14.3%) used only one session of NIBS combined with one therapy session, while most of the studies (85.7%) administered multiple sessions of NIBS combined with psychotherapy (see Supplementary Table S3 for more psychotherapy parameters).

3.3. Risk of bias assessment

Two investigators (EG, LB) independently assessed risk of bias using the RoB 2 tool for the primary outcome of each study. Inter-rater agreement was high (80% agreement), with concordant judgments on 112 of 140 ratings. Discrepancies between the two investigators were resolved through group discussion.

Domain-level assessment revealed the following risk profiles: randomization process (Domain 1): 27 low risk (96.4%), 1 some concerns (3.6%); deviations from intended interventions (Domain 2): 21 low risk (75.0%), 7 some concerns (25.0%); missing outcome data (Domain 3): 26 low risk (92.9%), 2 some concerns (7.1%); measurement of outcome (Domain 4): 27 low risk (96.4%), 1 some concerns (3.6%); selection of reported result (Domain 5): 4 low risk (14.3%), 24 some concerns (85.7%). The predominant concern across studies was lack of pre-registration (Domain 5), which affected the majority of trials.

We deviated from the RoB2 algorithm in one specific instance: when lack of pre-registration (Domain 5) was the only concern, we adjusted the overall rating from “some concerns” to “low risk” for 14 studies (50%). This decision reflects that pre-registration is a relatively recent practice and several included trials preceded widespread adoption of this standard. Without this adjustment, the majority of studies would have been classified as having “some concerns” solely based on publication era, obscuring more substantive methodological risks. We acknowledge that this deviation from the standard RoB2 algorithm may have produced a more favorable overall risk-of-bias profile than a strict application of the tool would yield. We made this decision deliberately, reasoning that pre-registration became standard practice only recently and that classifying the majority of trials as “some concerns” solely on this basis would obscure more substantive methodological risks. Nonetheless, it should be noted that under strict RoB2 classification, the majority of included studies would be classified as “some concerns,” and this should be considered when interpreting the overall risk-of-bias profile.

Overall, 20 studies (71.4%) were rated as low risk of bias, 6 (21.4%) as some concerns, and 2 (7.1%) as high risk (Fig. 3).

Fig. 3.

Summary of the risk of bias across the 28 included studies.

3.4. Meta-analysis

3.4.1. Clinical factors

3.4.1.1. Primary clinical outcomes.

Out of 28 included papers (and 31 treatment arms, given that three papers included two active arms), 26 reported a clinical outcome as their primary outcome measure, while two reported physiological outcomes as their primary. For those two papers, we consulted the author and selected the most relevant secondary clinical outcome listed. Across all studies, the random-effects meta-analysis revealed significantly greater improvement in primary clinical symptoms after active NIBS with psychotherapy compared to sham NIBS with psychotherapy (SMD = −0.38, 95% CI = [−0.68; −0.08]), with substantial heterogeneity (I² = 82.9%). We then conducted subgroup analyses to elucidate differences in clinical outcomes within various categories of interest.

3.4.1.2. Diagnostic category.

The combination of NIBS and psychotherapy was statistically significant in improving clinical symptoms in those diagnosed with anxiety (SMD = −0.70, 95% CI = [−1.26; −0.14]), but not in those diagnosed with addiction (SMD = −0.07, 95% CI = [−0.48; 0.33]), OCD (SMD = −0.81, 95% CI = [−2.10; 0.49]), pain (SMD = −0.32, 95% CI = [−3.95; 3.32]), or PTSD (SMD = 0.07, 95% CI = [−1.30; 1.45]). The diagnostic categories of depression, personality disorders, and transdiagnostic disorders did not meet the minimum study sample size of three studies and therefore were excluded from this subgroup analysis.

3.4.2. Intervention factors

3.4.2.1. NIBS modality

rTMS had a statistically significant benefit in improving clinical outcomes when paired with psychotherapy (SMD = −0.47, 95% CI = [−0.80; −0.13]), while tDCS did not (SMD = −0.20, 95% CI = [−0.88; 0.47]) (Fig. 4).

Fig. 4.

Primary outcome forest plot: Active NIBS + psychotherapy vs. sham NIBS + psychotherapy, stratified by modality. Standardized mean differences (SMD; negative values favor Active) are shown for each trial with 95% CIs, pooled within tDCS and rTMS subgroups, and pooled overall using a random-effects model. Both subgroups show a small benefit of combining stimulation with psychotherapy relative to sham + psychotherapy. Heterogeneity is substantial (see I² and τ²), consistent with variation in protocols, diagnoses, and timing. Column headings report group Ns, means, and SDs used for SMD estimation; studies are weighted by inverse variance. Abbreviations: NIBS, noninvasive brain stimulation; tDCS, transcranial direct current stimulation; rTMS, repetitive transcranial magnetic stimulation; CI, confidence interval.

3.4.2.2. Psychotherapy format.

In the subgroup of psychotherapy category (CBT, exposure, and mindfulness), the combination of CBT and NIBS resulted in statistically significant benefit in improving clinical outcomes (SMD = −0.48, 95% CI = [−0.78; −0.17]), while exposure therapy and mindfulness did not (SMD = −0.35, 95% CI = [−0.94; 0.24], SMD = −0.22, 95% CI = [−1.45; 1.01], respectively).

3.4.3. Methodological factors

3.4.3.1. Level of evidence supporting the psychotherapy.

Studies that included fully evidence-based psychotherapy had a statistically significant improvement in clinical outcomes for active vs. sham NIBS (SMD = −0.33, 95% CI = [−0.59; −0.07]), while studies that only used partial elements of evidence-based therapy did not (SMD = −0.41, 95% CI = [−0.90; 0.08]).

3.4.3.2. Psychotherapy administration method.

Only studies in which psychotherapy is administered exclusively by a human significantly improved clinical outcomes for active vs. sham NIBS (SMD = −0.49, 95% CI = [−0.78; −0.21]). No significant improvement was found with administration solely by a computer (SMD = −0.17, 95% CI = [−1.87; 1.52]) or by a mix of computer and human (SMD = −0.28, 95% CI = [−0.98; 0.43]).

3.4.3.3. NIBS timing; NIBS timing x NIBS modality.

When administered non-concurrently with the psychotherapy (before or after), NIBS had a statistically significant benefit in improving clinical outcomes (SMD = −0.56, 95% CI = [−0.94; −0.18]), while concurrent NIBS did not (SMD = −0.11, 95% CI = [−0.61; 0.39]). Interestingly, we found that only rTMS given non-concurrently had a statistically significant benefit in improving clinical symptoms (SMD = −0.47, 95% CI = [−0.91; −0.02]). No effects were found with concurrent tDCS (SMD = 0.26, 95% CI = [−0.56; 1.07]), non-concurrent tDCS (SMD = −0.92, 95% CI = [−2.04; 0.20]), or concurrent rTMS (SMD = −0.48, 95% CI = [−1.17; 0.21]) (Figs. 5 and 6). Unfortunately, the limited number of studies prevented us from testing for an interaction between timing, NIBS modality and each type of psychotherapy, which could have provided stronger insight on this result.

Fig. 5.

Primary outcome by stimulation–therapy timing. Forest plot for individual trials, pooled within Non-concurrent (stimulation delivered before/after psychotherapy) and Concurrent (stimulation during psychotherapy) subgroups and overall (random-effects). The Non-concurrent subgroup shows a significant small-to-moderate symptom reduction relative to sham + psychotherapy, whereas the Concurrent subgroup’s pooled effect is not statistically significant.

Fig. 6.

Primary outcome by timing × modality. Forest plot of SMD for individual trials, pooled within four subgroups: Concurrent tDCS, Non-concurrent tDCS, Concurrent rTMS, and Non-concurrent rTMS. Under random effects, only Non-concurrent rTMS shows a significant pooled benefit; the other subgroups are not significant.

3.4.3.4. NIBS modality × psychotherapy category.

We additionally explored the combination of NIBS modality with psychotherapy category. When CBT was administered with rTMS, there was a statistically significant benefit in improving clinical symptoms (SMD = −0.48, 95% CI = [−0.88; −0.07]), while CBT with tDCS (SMD = −0.47, 95% CI = [−1.31; 0.36]), exposure therapy with tDCS (SMD = −0.38, 95% CI = [−2.69; 1.93]), exposure therapy with rTMS (SMD = −0.35, 95% CI = [−0.99; 0.29]), and mindfulness with tDCS (SMD = 0.22, 95% CI = [−11.91; 9.79]) did not. There were only two studies that utilized the combination of mindfulness with rTMS, and those were excluded from this subgroup analysis.

3.5. Summary of all results for the primary outcome measures

Given the number of comparisons, we provide in Table 1 a summary of all the results from the primary outcomes.

Table 1.

Standardized mean differences (SMD, active vs. sham) for various NIBS modalities, clinical diagnoses, and combined therapeutic interventions. SMD represent the central effect size, flanked by 95% Confidence Intervals (CI). Values in bold denote statistical significance (where the CI does not cross zero), indicating a superior outcome for active stimulation over sham.

Category	Label	SMD	Lower CI	Upper CI	Significance
NIBS type	tDCS	−0.20	−0.88	0.47	Not Significant
	rTMS	−0.47	−0.80	−0.13	Significant
Diagnosis	Addiction	−0.08	−0.48	0.33	Not Significant
	Anxiety	−0.70	−1.26	−0.14	Significant
	Depression	−0.39	−1.24	0.47	Not Significant
	OCD	−0.81	−2.10	0.49	Not Significant
	Pain	−0.32	−3.95	3.32	Not Significant
	Personality	−0.47	−1.10	0.17	Not Significant
	PTSD	0.08	−1.30	1.45	Not Significant
	Transdiagnostic	−0.76	−0.84	−0.68	Significant †
Therapy	CBT	−0.48	−0.78	−0.17	Significant
	Exposure	−0.35	−0.94	0.24	Not Significant
	Mindfulness	−0.22	−1.45	1.01	Not Significant
Evidence-based	No	−0.41	−0.90	0.08	Not Significant
	Yes	−0.33	−0.59	−0.07	Significant
Therapy delivery	Computerized	−0.18	−1.87	1.52	Not Significant
	Mixed	−0.28	−0.98	0.43	Not Significant
	Human	−0.50	−0.78	−0.21	Significant
Timing	Not concurrent	−0.56	−0.94	−0.18	Significant
	Concurrent	−0.11	−0.61	0.39	Not Significant
Timing × NIBS	Concurrent tDCS	0.26	−0.56	1.07	Not Significant
	Non-concurrent tDCS	−0.92	−2.04	0.20	Not Significant
	Concurrent rTMS	−0.48	−1.17	0.21	Not Significant
	Non-concurrent rTMS	−0.47	−0.91	−0.02	Significant
Therapy × NIBS	CBT with tDCS	−0.48	−1.31	0.36	Not Significant
	CBT with rTMS	−0.48	−0.88	−0.07	Significant
	Exposure with tDCS	−0.38	−2.69	1.93	Not Significant
	Exposure with rTMS	−0.35	−0.99	0.29	Not Significant
	Mindfulness with tDCS	0.22	−0.88	1.31	Not Significant
	Mindfulness with rTMS	−1.06	−11.91	9.79	Not Significant
Therapy × Diagnosis	CBT for Addiction	−0.11	−0.65	0.44	Not Significant
	CBT for Transdiagnostic	−0.76	−0.84	−0.68	Significant †
	Exposure for Anxiety	−0.61	−1.28	0.06	Not Significant
	Exposure for OCD	−0.81	−2.10	0.49	Not Significant
	Exposure for PTSD	0.32	−1.35	1.98	Not Significant
	Mindfulness for Pain	−0.22	−2.09	1.66	Not Significant

^†Significant but aggregated across only two studies

3.6. Publication bias and heterogeneity

Visual inspection of the funnel plot did not suggest asymmetry, and Egger’s test was non-significant (p > 0.05), indicating no evidence for small-study effects (Fig. 7).

Fig. 7.

Funnel plot for small-study effects in the primary analysis. Scatter of individual studies’ SMD against their standard errors; negative SMD indicates symptom improvement favoring Active NIBS + psychotherapy. The vertical dashed line marks the pooled random effects estimate; dotted lines denote the pseudo 95% confidence limits forming the funnel.

3.7. Executive functioning and quality of life outcomes

No significant results were found for active compared to sham NIBS with psychotherapy in improving executive functioning (n = 9 studies, SMD = 0.12, 95% CI = [−0.45; 0.70]) or quality of life (n = 9 studies, SMD = 0.16, 95% CI = [−0.11; 0.42]).

3.8. Other outcomes

We sought to investigate changes in fMRI BOLD activity, as neuroimaging outcomes could provide mechanistic insights into how combined treatment alters neural circuits. However, only four studies collected fMRI at any time-point, precluding meta-analytic synthesis. Two used fMRI for the purpose of individualized NIBS targeting (Cybinski et al., 2024; Neacsiu et al., 2022a), while the other two explored changes in network activation post-NIBS and psychotherapy (Adams et al., 2022; Fitzsimmons et al., 2025). Given the lack of studies and the variability in the tasks employed, we were unable to conduct this analysis.

Finally, we conducted an analysis of studies that included depression and anxiety outcome measures, independently of the diagnosis. We did not find any significant results for active compared to sham NIBS with psychotherapy in improving depression symptoms (SMD = −0.08, 95% CI = [−0.35; 0.19]). However, we did find an overall statistically significant improvement in anxiety scales for active NIBS with psychotherapy compared to sham NIBS with psychotherapy (SMD = −0.82, 95% CI = [−1.40; −0.24]). With subgroup analysis, we found that this improvement in anxiety measures was specific to rTMS (SMD = −0.73, 95% CI = [−1.35; −0.10]) (Fig. 8). We additionally explored the subgroup of NIBS timing and found that, when administered concurrently, the combination of NIBS and psychotherapy was statistically significant in improving anxiety symptoms (SMD = −0.80, 95% CI = [−1.38; −0.23]), while non-concurrent NIBS with psychotherapy was not significant (SMD = −0.83, 95% CI = [−1.83; 0.17]) (Fig. 9).

Fig. 8.

Forest plot for subcategory of NIBS modality on anxiety outcomes.

Fig. 9.

Forest plot for subcategory of NIBS timing on anxiety outcomes.

4. Discussion

This meta-analysis provides a comprehensive quantitative synthesis of randomized controlled trials examining the efficacy of NIBS combined with evidence-based psychotherapy for psychiatric disorders. Our findings demonstrate that active NIBS paired with psychotherapy produces significantly greater improvement in primary clinical symptoms compared to sham NIBS with psychotherapy, with a small-to-moderate overall effect size (SMD = −0.38). Critically, our moderator analyses reveal that the magnitude and significance of this benefit depend on specific implementation parameters: NIBS modality, temporal alignment of stimulation with therapy, diagnostic category, and the psychotherapy delivery method. These findings offer mechanistic insights into how brain state modulation enhances therapeutic learning and provide a framework for optimizing combined treatment protocols.

4.1. Clinical factors

4.1.1. Primary findings and clinical significance

The overall effect size we observed (SMD = −0.38) aligns with prior meta-analytic estimates reporting small-to-moderate benefits of combined approaches (He et al., 2022; Xu et al., 2023). However, our results diverge from these earlier syntheses in critical ways. First, by restricting inclusion to trials that compared active NIBS plus psychotherapy against sham NIBS plus psychotherapy—thereby controlling for nonspecific effects of psychotherapy—we isolate the specific augmentation effect attributable to brain stimulation. Second, by requiring clinician-verified evidence-based psychotherapy, we minimize heterogeneity introduced by non-manualized or poorly validated interventions. This stringent choice reduces confounding but also narrows the evidence base, underscoring the limited number of well-controlled trials in this emerging field.

We found no significant effects of combined NIBS and psychotherapy on executive functioning (n = 9 studies, SMD = 0.12) or quality of life (n = 9 studies, SMD = 0.16). This null finding should be interpreted cautiously given the limited sample and heterogeneity in assessment instruments. Prior meta-analyses have reported inconsistent results: Fu et al. (2025) found no cognitive benefits of rTMS in depression across multiple domains (Fu et al., 2025), while Xu et al. (2023) observed improvement in global cognition but not quality of life (Xu et al., 2023). Future trials should continue assessing these outcomes to clarify whether combined treatment impacts cognitive function and wellbeing.

4.1.2. Diagnostic considerations and transdiagnostic potential

At the diagnostic level, significant improvement was observed only for anxiety disorders (SMD = −0.70), with no significant effects for addiction, OCD, pain, or PTSD. This pattern may reflect disorder-specific neural mechanisms, sample size limitations, or protocol optimization. Anxiety disorders—particularly social anxiety, specific phobias, and panic disorder—are characterized by hyperactivation of threat-detection circuits involving the amygdala, insula, and medial prefrontal cortex (Etkin and Wager, 2007), which are amenable to prefrontal neuromodulation during exposure-based interventions. The integration of NIBS with exposure therapy may reduce anticipatory anxiety, enhance extinction learning (Raij et al., 2018), or modulate dlPFC-mediated anxious arousal (Balderston et al., 2020), mechanisms that align well with the neurobiological targets of prefrontal TMS.

The limited representation of MDD in our sample (only two studies) is striking, given that MDD is the most extensively studied indication for NIBS monotherapy (Lefaucheur et al., 2020). Most combination trials in MDD have compared active NIBS plus psychotherapy against active NIBS alone, rather than against sham NIBS plus psychotherapy, and were therefore excluded by our eligibility criteria. This design choice reflects a pragmatic clinical question (does adding psychotherapy improve outcomes over stimulation alone?) but precludes conclusions about whether NIBS augments psychotherapy specifically. Future research should address this gap by testing whether TMS enhances established psychotherapies like behavioral activation or interpersonal therapy in MDD, particularly for patients with partial response to either modality alone.

The absence of significant effects in PTSD is particularly notable given that six studies included PTSD populations, comparable to the anxiety subgroup. PTSD involves complex disruptions in fear extinction, emotion regulation, and memory reconsolidation (Liberzon and Abelson, 2016; Pitman et al., 2012), processes that theoretically should benefit from NIBS-augmented psychotherapy. However, PTSD trials in our sample exhibited high heterogeneity in trauma types, stimulation parameters, and psychotherapy formats, which may have obscured benefits. Additionally, PTSD is often comorbid with depression, dissociation, and chronic hyperarousal (Brady et al., 2000; Flory and Yehuda, 2015), which may require more intensive or prolonged combined treatment (Schnurr et al., 2024) than the protocols tested to date.

4.2. Intervention factors

4.2.1. NIBS modality

Our subgroup analyses demonstrate that rTMS, but not tDCS, yielded statistically significant improvement when combined with psychotherapy. This divergence may reflect mechanistic differences between modalities. TMS directly induces action potentials and modulates cortical excitability with high spatial precision, whereas tDCS shifts membrane potentials subthreshold, producing weaker and more diffuse neuromodulatory effects. The greater efficacy of rTMS could also be attributed to more established clinical protocols, higher dose consistency across trials, and more robust sham control procedures. However, the tDCS subgroup included fewer studies and exhibited substantial heterogeneity, which may have precluded detection of smaller effects. Emerging home-based tDCS protocols, recently FDA-approved for major depressive disorder (Bikson et al., 2026), may offer unique advantages for scalability and adherence that warrant further investigation in combination paradigms.

4.2.2. Temporal alignment of psychotherapy and NIBS

Perhaps the most theoretically and clinically important finding from this meta-analysis is the moderating effect of NIBS timing. Non-concurrent delivery, where stimulation preceded or followed psychotherapy, produced significant symptom reduction (SMD = −0.56), whereas concurrent delivery did not (SMD = −0.11). This pattern held specifically for rTMS delivered non-concurrently (SMD = −0.47), with no significant effects for concurrent rTMS, concurrent tDCS, or non-concurrent tDCS. These findings appear to contradict prevailing theories that emphasize state-dependent plasticity, which predict maximal benefit when stimulation co-occurs with active cognitive or emotional engagement (Luber and Lisanby, 2014; Sathappan et al., 2019).

Several mechanisms may account for the superiority of non-concurrent protocols. First, priming paradigms—in which stimulation precedes therapy—may enhance synaptic plasticity and metaplasticity, rendering treatment-engaged circuits more responsive to subsequent therapeutic learning (Tatti et al., 2022). Preclinical work demonstrates that TMS can upregulate brain-derived neurotrophic factor (BDNF) and long-term potentiation, creating a neurobiological window of heightened plasticity that extends beyond the stimulation session itself (Gersner et al., 2011). Second, consolidation paradigms—in which stimulation follows therapy—may stabilize newly formed memory traces through reconsolidation mechanisms, strengthening extinction learning or cognitive restructuring achieved during psychotherapy (Saccenti et al., 2024). Third, concurrent delivery may introduce practical barriers that undermine psychotherapy fidelity. The percussive noise, somatic sensations, and attentional demands of concurrent TMS can interfere with mindfulness practice, exposure tolerance, or therapeutic alliance, as evidenced by high dropout rates in some concurrent protocols (Cavallero et al., 2021; Demina et al., 2025).

Our secondary analysis of anxiety outcomes revealed that both concurrent and non-concurrent NIBS produced comparably large point estimates for anxiety symptom reduction (concurrent: SMD = −0.80, 95% CI [−1.38; −0.23]; non-concurrent: SMD = −0.83, 95% CI [−1.83; 0.17]). Although only the concurrent estimate reached statistical significance, the near-identical effect sizes and the substantially wider confidence interval for the non-concurrent subgroup strongly suggest this difference reflects statistical power rather than a true mechanistic distinction. The non-concurrent subgroup contained fewer anxiety studies, and the overlapping confidence intervals preclude any conclusion that concurrent delivery is preferentially efficacious for anxiety outcomes. Nonetheless, the hypothesis that anxiety reduction may be particularly responsive to concurrent neuromodulation warrants prospective investigation. Anxiety scales often capture state-dependent reactivity, that is, transient arousal and avoidance indexed in the moment, rather than the stable symptom severity and functional impairment measured by primary outcomes (Spielberger, 1983). If concurrent NIBS modulates anxious arousal in real time during exposure or emotional processing, it may be especially well-suited to attenuating this reactivity. This mechanistic account is biologically plausible but remains untested; the present data are consistent with it but do not confirm it. Future trials directly comparing timing schedules within anxiety populations, with adequate power and pre-registered hypotheses, are needed to determine whether concurrent delivery holds a genuine advantage for anxiety-specific outcomes.

4.2.3. Spatial alignment of psychotherapy and NIBS

Another important consideration when combining NIBS with psychotherapy is the spatial alignment of neural targets to maximally promote Hebbian plasticity and behavioral activation. This spatial alignment is complex, as the ideal NIBS target should be optimized both for the dysregulated neural circuitry specific to the population’s psychopathology and for the circuits actively engaged by the chosen psychotherapy. CBT has been found to primarily alter cognitive and emotional control regions such as the prefrontal cortex, precuneus, and anterior cingulate cortex (Yuan et al., 2022); mindfulness has been found to act on the dACC, amygdala, and insula to increase central executive network to salience network connectivity and decrease default mode network activity (Sim et al., 2024); exposure therapy has been found to dampen fear-network activity in regions like the amygdala and insula while heightening cognitive control regions in the ventromedial and dorsolateral prefrontal cortex (Hauner et al., 2012). Despite these three psychotherapies having distinct mechanisms of action, all but two of the studies selected left, right, or medial prefrontal cortex targets. The prefrontal cortex is a broad region implicated in many psychiatric disorders and thought to be functionally altered by all three psychotherapy methods, making it a logical target for NIBS. However, most of the included studies (83.3%) used scalp landmarking rather than anatomical or functional neuroimaging to locate their prefrontal cortex targets, even though a more optimal neural target could be found if individualized for patients’ neuroanatomy, their psychopathology, and psychotherapy method being used for treatment. Neuroimaging to improve spatial targeting of TMS specifically has become an area of great interest in the field, particularly to alter deeper brain regions and networks via more superficial targets (Cash et al., 2021). This approach was exemplified by Neacsiu et al. (2022a), who administered an emotion-regulation task during fMRI to identify neural regions activated by cognitive restructuring and selected a TMS target specifically aligned with the therapy. Future studies should prioritize this approach, using task-based or resting-state fMRI to identify individualized targets that are simultaneously optimized for psychopathology-relevant circuitry and psychotherapy-engaged networks, moving the field toward truly circuit-informed combined treatment protocols.

4.2.4. Psychotherapy format

Among psychotherapy modalities, only CBT combined with NIBS produced statistically significant improvement (SMD = −0.48), while exposure therapy and mindfulness did not. This finding likely reflects both the larger sample of CBT trials and the heterogeneity within the exposure- and mindfulness-based therapies. Exposure-based interventions, which constituted nearly half of all studies, included protocols ranging from brief symptom provocation (as in the FDA-cleared OCD protocol) to full multi-session exposure and response prevention. Similarly, mindfulness interventions varied from brief audio-guided exercises to structured mindfulness-based stress reduction programs. The CBT category, by contrast, encompassed manualized protocols with clearer structural consistency across trials.

Our analysis also revealed that only human-delivered psychotherapy—as opposed to computerized or hybrid formats—resulted in significant augmentation by NIBS. This finding underscores the importance of the therapeutic relationship, real-time adaptation to patient responses, and the clinician’s ability to navigate complex emotional states that may arise during treatment. Computerized interventions, while offering standardization and scalability, may lack the interpersonal depth necessary to capitalize on NIBS-induced plasticity. These programs often present abbreviated or simplified versions of evidence-based techniques, which may not fully engage the cognitive and emotional processes targeted by NIBS. Hybrid approaches, which combine human oversight with digital content delivery, showed intermediate but non-significant effects, suggesting a potential middle ground worthy of further refinement.

4.3. Methodological factors

4.3.1. Treatment integrity and level of evidence

Treatment integrity, encompassing therapist adherence to manualized protocols and competent delivery, is a critical but under-reported dimension in this literature. Our clinician screening identified that only 39.3% of studies employed fully evidence-based, manualized psychotherapy, with the remainder using only partial elements of validated treatments. Moreover, only three studies (10.7%) explicitly reported adherence monitoring or competence ratings. This gap is concerning because treatment integrity is consistently associated with superior clinical outcomes in psychotherapy research (Esposito et al., 2024; Power et al., 2022). Notably, studies using fully manualized psychotherapy showed a significant augmentation effect (SMD = −0.33), while those using only partial elements did not (SMD = −0.41, 95% CI [−0.90; 0.08]), though the point estimates were similar and confidence intervals overlapping, suggesting power rather than a true difference may account for this pattern. When psychotherapy is diluted or poorly implemented, the synergistic potential of combined NIBS-psychotherapy protocols may be undermined. Future trials must prioritize treatment integrity by using manualized interventions, training therapists to fidelity, and systematically monitoring adherence and competence throughout the study.

4.4. Limitations and interpretive considerations

Several limitations warrant consideration when interpreting these findings. First, the relatively small number of eligible trials (n = 28) limited statistical power for subgroup and interaction analyses, particularly for underrepresented diagnoses such as MDD, BPD, and transdiagnostic samples. Confidence intervals were wide for several moderators, and null findings may reflect insufficient power rather than true absence of effects. Second, substantial heterogeneity (high I²) indicates variability across trials in stimulation parameters (frequency, intensity, target, number of sessions), psychotherapy protocols (dose, fidelity, delivery format), and patient characteristics (symptom severity, treatment resistance, medication status). While we conducted moderator analyses to parse this heterogeneity, unmeasured confounders likely remain. Third, we faced constraints in categorizing NIBS timing. We originally intended to distinguish between stimulation delivered before, during, or after psychotherapy to enable more specific mechanistic recommendations. However, many studies did not specify exact timing beyond reporting non-concurrent delivery, resulting in insufficient sample sizes for separate before/after categories. Consequently, we collapsed all non-concurrent protocols into a single category. While this approach maintained statistical power, it precluded differentiation between priming effects (pre-session stimulation) and consolidation effects (post-session stimulation), which may operate through distinct neuroplastic mechanisms. Future trials should explicitly report whether NIBS precedes or follows therapy to enable more granular temporal analyses. We also acknowledge that timing is inherently confounded with stimulation modality and protocol architecture in the current evidence base. For example, rTMS may be louder and more disruptive to therapy engagement than tDCS, and perhaps especially so for psychotherapies like mindfulness that require a calm environment. Critically, the significant effect observed for non-concurrent delivery was driven entirely by the rTMS subgroup, meaning the timing finding and the modality finding cannot be cleanly separated with the current data. It is therefore not possible to determine whether the observed advantage of non-concurrent protocols reflects timing per se, modality-specific features, or their interaction. Definitively establishing a causal effect of timing will require a prospective RCT that systematically varies timing within a single modality and psychotherapy type, with adequate power to test the timing by modality interaction directly.

Fourth, one study with a crossover design (Osuch et al., 2009) was included. Review of their statistical methods indicates that the linear mixed models applied by the original authors were used to examine active versus sham differences over time with a compound symmetry covariance structure, but did not explicitly model sequence or carryover effects. Consequently, the extracted means and standard deviations cannot be guaranteed to be free of psychotherapy-specific carryover, as the 2-week washout period in that trial may be insufficient to reverse the long-lasting cognitive and behavioral changes that effective psychotherapy induces. To assess whether this study materially influenced our findings, we conducted a post-hoc sensitivity analysis excluding Osuch et al. from the primary meta-analysis. The pooled effect size remained virtually unchanged (SMD = −0.42, 95% CI [−0.71; −0.13], k = 30), with nearly identical heterogeneity (I² = 82.6%), indicating that our primary conclusions are robust to the inclusion of this study. Future crossover trials in this area should explicitly model period and carryover effects, or consider parallel designs to avoid the inherent limitations of washout periods for skill-based psychotherapeutic interventions.

Fifth, our decision to compare active NIBS plus psychotherapy against sham NIBS plus psychotherapy was motivated by the goal of isolating the specific augmentation effect of neurostimulation. However, this design precludes inference about whether combined treatment is superior to psychotherapy alone, a clinically crucial question that remains unaddressed by the current work and should be a priority for future trial design. Furthermore, the limitations of sham as an inert comparator are not unique to this meta-analysis but reflect a broader methodological challenge in the NIBS field. Sham stimulation may interact with psychotherapy in multiple ways: it may enhance outcomes through expectancy effects when paired with an active therapeutic context, or it may interfere with psychotherapy by introducing sensory distractions such as scalp sensations or auditory clicks that compete for attentional resources during in-session therapeutic work. Critically, this ambiguity bears directly on the interpretation of our moderator finding that non-concurrent delivery was associated with significant effects while concurrent delivery was not. If concurrent sham NIBS actively disrupts psychotherapeutic processes, the apparent superiority of non-concurrent protocols may partly reflect reduced interference in the sham condition rather than, or in addition to, enhanced neuroplasticity priming or consolidation in the active condition. Disentangling these mechanisms will require designs that include a psychotherapy-alone arm alongside both active and sham NIBS augmentations.

Finally, treatment integrity was under-reported. Only 10.7% of studies documented therapist adherence or competence, making it difficult to determine whether psychotherapy was delivered as intended. The majority of studies (60.7%) used only partial elements of evidence-based protocols rather than fully manualized treatments, introducing additional heterogeneity. Without systematic fidelity monitoring, it is impossible to disentangle whether null findings reflect ineffective combinations or poor psychotherapy implementation. Fifth, the majority of studies allowed concurrent psychotropic medications, which may interact with NIBS effects. Benzodiazepines, for instance, can blunt TMS-induced plasticity through GABAergic mechanisms, while antipsychotics may alter cortical excitability. Although medication status was reported in most studies, few conducted sensitivity analyses to evaluate its moderating role. Finally, our analysis focused on short-term outcomes, typically assessed immediately post-treatment or within a few weeks. Evidence for the durability of combined NIBS-psychotherapy effects is limited. Long-term follow-up data—at 6, 12, or 24 months—were reported in only a subset of trials, and we lacked sufficient data to conduct meta-analytic synthesis of maintenance effects. This is a critical gap, as one of the theoretical advantages of combining psychotherapy with NIBS is the potential to enhance durability through consolidation of therapeutic learning. Future trials should prioritize extended follow-up periods to determine whether augmentation by NIBS produces more sustained remission than psychotherapy alone.

4.5. Future directions

Future research should also prioritize mechanistic endpoints. Incorporating neuroimaging (resting-state and task-based fMRI), electrophysiology (EEG, TMS-evoked potentials), and peripheral biomarkers (BDNF, inflammatory cytokines) can illuminate how combined treatment alters neural circuits and whether neuroplastic changes mediate clinical outcomes. Biomarker-stratified trials could identify which patients are most likely to benefit from combined treatment, moving the field toward precision psychiatry. fMRI-individualized TMS targeting may also allow for more precise manipulation of diagnosis-specific neural circuits of interest, a method that was only used in approximately 11% of included studies. Additionally, researchers should explore novel NIBS modalities not yet tested in combination with psychotherapy, including tACS, tFUS, and closed-loop neurostimulation paradigms that adapt in real time to brain state.

Finally, implementation science research is critical to translate these findings into real-world practice. Combined NIBS-psychotherapy protocols are resource-intensive, requiring specialized equipment, trained operators, and extended treatment schedules. Cost-effectiveness analyses, scalability studies, and dissemination trials are needed to determine whether combined approaches can be feasibly integrated into routine clinical care, particularly in underserved settings. The recent FDA approval of home-based tDCS for depression raises intriguing possibilities for remotely supervised combined treatment, pairing home-delivered stimulation with teletherapy. Such models could dramatically expand access while maintaining treatment fidelity.

5. Conclusion

This meta-analysis demonstrates that noninvasive brain stimulation, when combined with evidence-based psychotherapy, produces significant improvement in psychiatric symptoms beyond the effects of psychotherapy with sham stimulation. However, this benefit is conditional on critical implementation parameters. Among the combinations examined, non-concurrent rTMS paired with human-delivered, manualized CBT was the only one to show significant effects across primary moderator analyses, particularly for anxiety disorders. The finding that non-concurrent rather than concurrent delivery was associated with significant effects warrants further investigation, though it should be interpreted cautiously: non-concurrent protocols collapse priming and consolidation mechanisms that may be functionally distinct, and timing is inherently confounded with stimulation modality and protocol architecture in the current evidence base. These findings do not establish non-concurrent delivery as definitively superior, but rather identify it as a priority target for prospective trials designed to isolate timing effects. Treatment integrity, including the use of fully evidence-based psychotherapy and fidelity monitoring, is essential but remains critically underreported in the current literature and represents an important direction for future work.

While these findings provide an evidence-based framework for optimizing combined treatment protocols, significant gaps remain. The field needs larger, mechanistically informed trials with long-term follow-up, standardized reporting of psychotherapy fidelity, individualized NIBS targeting, and biomarker-guided patient selection. By addressing these challenges, the integration of brain stimulation with psychotherapy holds promise as a next-generation approach to enhance the speed, magnitude, and durability of therapeutic change in psychiatric care.

Appendix A. Supporting information

Supplementary data associated with this article can be found in the online version at doi:10.1016/j.neubiorev.2026.106786.