Comparative Efficacy of Non-invasive Brain Stimulation for Pain, Anxiety, and Depression: A Systematic Review and Network Meta-analysis

Abstract

Introduction

Although non-invasive brain stimulation (NIBS) has been used to manage chronic musculoskeletal pain (CMP) and related emotional symptoms, the comparative efficacy of different NIBS techniques remains unclear. Therefore, this network meta-analysis (NMA) compared the effects of NIBS interventions on pain intensity, anxiety, and depressive symptoms in CMP.

Methods

We searched seven databases from inception to 7 October 2025, including PubMed, the Cochrane Library, Web of Science, Embase, MEDLINE, PsycINFO, and CINAHL, and included only English-language randomized controlled trials evaluating NIBS for CMP that reported pain, anxiety, or depression outcomes. Effect sizes were synthesized using a frequentist NMA and expressed as standardized mean differences (SMDs) or mean differences (MDs) with 95% confidence intervals (CIs). Risk of bias was assessed with the Cochrane Risk of Bias 2.0 tool.

Results

A total of 50 randomized controlled trials were included, involving 2,203 participants with an average age ranging from 31 to 75 years; approximately 82% of the participants were women. The NMA showed that: (1) in the overall CMP population, low-frequency repetitive transcranial magnetic stimulation (LF-rTMS) targeting the dorsolateral prefrontal cortex (DLPFC) (SMD = − 1.19, 95% CI − 1.86 to − 0.52), transcranial random noise stimulation (tRNS) targeting the primary motor cortex (M1) (SMD = − 1.66, 95% CI − 3.05 to − 0.28), and intermittent theta-burst stimulation (iTBS) targeting the DLPFC (SMD = − 0.78, 95% CI − 1.63 to 0.08) ranked first for pain, anxiety, and depression, respectively. Sensitivity analyses suggested that LF-rTMS targeting the DLPFC may be the most robust option for CMP with comorbid emotional symptoms; (2) in fibromyalgia, LF-rTMS targeting the DLPFC ranked first for pain (SMD = − 1.19, 95% CI − 1.99 to − 0.39) and depression (SMD = − 0.63, 95% CI − 0.98 to − 0.27), while tRNS targeting M1 (SMD = − 1.66, 95% CI − 2.99 to − 0.34) ranked first for anxiety; after excluding sparse-evidence interventions, transcranial direct current stimulation (tDCS) targeting M1 became the preferred option for anxiety; (3)in non-fibromyalgia populations, cranial electrotherapy stimulation (CES) (SMD = − 1.39, 95% CI − 2.41 to − 0.37) ranked first for pain, while iTBS targeting the DLPFC (MD = − 0.79, 95% CI − 1.48 to − 0.10) showed potential benefit for depression, although supporting evidence was limited.

Conclusions

Different non-invasive brain stimulation modalities may confer outcome-specific benefits in chronic musculoskeletal pain. Overall, low-frequency repetitive transcranial magnetic stimulation targeting the dorsolateral prefrontal cortex showed relatively robust effects across outcomes in chronic musculoskeletal pain and on pain and depression in fibromyalgia; however, further high-quality clinical trials are needed.

PROSPERO registration

CRD420251131206.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s40122-026-00860-9.

Key Summary Points

Why carry out this study?

Chronic pain frequently coexists with anxiety and depression, but the comparative effectiveness of non-invasive brain stimulation interventions remains unclear.

A systematic review and network meta-analysis were performed to evaluate and compare available non-invasive brain stimulation approaches.

What was learned from this study?

A network meta-analysis of 50 studies suggested that LF-rTMS targeting the DLPFC may provide the greatest overall benefit for CMP, especially in fibromyalgia.

LF-rTMS targeting the DLPFC may represent the most robust option for improving pain and emotional symptoms in patients with CMP, particularly pain and depressive symptoms in those with fibromyalgia.

tDCS targeting M1 may be a more appropriate option for anxiety in fibromyalgia after exclusion of sparsely supported studies, whereas CES showed favorable effects on pain relief and iTBS targeting the DLPFC for depressive symptoms, although the evidence remains limited.

tRNS targeting M1 may be more effective for reducing anxiety in fibromyalgia, whereas CES may offer greater pain relief in non-fibromyalgia patients.

These results provide comparative evidence that may inform future trials and the clinical application of non-invasive brain stimulation for chronic musculoskeletal pain management.

Introduction

Chronic musculoskeletal pain (CMP) refers to persistent or recurring pain resulting from pathological processes that directly affect bones, joints, muscles, or soft tissues [1]. Epidemiological studies estimate that CMP affects approximately 20–30% of the global population and is the second leading cause of nonfatal disability worldwide [2–4]. CMP substantially reduces quality of life and increases medical and socioeconomic burden. It is frequently comorbid with anxiety and depression, which worsen pain-related functional impairment and further diminish quality of life [5]. Current clinical practice recommends pharmacological, non-pharmacological, and interdisciplinary multimodal approaches for CMP management [2, 6]. Pharmacological therapies, including acetaminophen and nonsteroidal anti-inflammatory drugs, remain the cornerstone of management [7]; however, treatment selection should be guided by pain type, pain severity, and patient comorbidities [8]. At the same time, long-term use of medications, particularly opioids, may lead to adverse effects such as drug dependence and gastrointestinal injury [9, 10]. Conversely, non-pharmacological interventions such as exercise therapy and cognitive behavioral therapy are widely recommended due to their excellent safety profile. Exercise therapy reduces pain intensity and improves physical function, potentially by modulating inflammatory responses and central pain-processing pathways [11, 12]. Evidence from systematic reviews and meta-analyses further indicates that cognitive behavioral therapy produces small to moderate improvements in pain, function, and psychological distress, with benefits for pain-related disability and mood lasting up to 12 months after treatment [13, 14]. Although non-pharmacological interventions are widely recommended, the evidence supporting their efficacy remains limited, and some interventions produce suboptimal outcomes [10, 15, 16]. Therefore, identifying safer and more effective adjunctive therapies with well-defined mechanisms of action has become a central objective in contemporary pain management.

In this context, non-invasive brain stimulation (NIBS) techniques, such as repetitive transcranial magnetic stimulation (rTMS) and transcranial direct current stimulation (tDCS), have emerged as promising treatments for chronic pain because of their non-invasive nature, favorable safety profile, and therapeutic potential [17, 18]. These approaches may benefit CMP by modulating cortical excitability, reorganizing pain-regulatory networks, and alleviating central sensitization [19, 20]. Although multiple randomized controlled trials have shown that NIBS can effectively reduce pain intensity in conditions such as low back pain and fibromyalgia and may also improve depressive symptoms [21–23], the current evidence remains quite limited and fragmented. First, most trials have small sample sizes, lack head-to-head comparisons, and show substantial heterogeneity in stimulation protocols, including treatment targets, frequency, and total dose, producing a weak evidence base that limits clinical decision-making without high-quality support [24–26]. Second, most reviews have focused on single interventions or specific diseases; few have specifically examined CMP as a distinct subgroup to compare the relative benefits and limitations of different NIBS modalities [27, 28]. Third, no network meta-analysis has synthesized recent randomized controlled trials to rank NIBS modalities for pain, anxiety, and depression in CMP. Therefore, this study used a network meta-analysis (NMA) to systematically evaluate the relative effects of NIBS on pain, anxiety, and depression in patients with CMP. The NMA allows for the integration of direct and indirect evidence in the absence of direct comparisons, providing a relative ranking of different interventions. This study may help inform evidence-based clinical decision-making, identify key gaps and limitations in the current evidence base to guide future research and highlight the most promising NIBS modalities for specific symptom domains for patients with CMP.

Methods

This study was conducted in accordance with the PRISMA extension for systematic reviews incorporating NMA of healthcare interventions (PRISMA-NMA) [29]. This systematic review and NMA were previously registered on PROSPERO (CRD420251131206; available at: https://www.crd.york.ac.uk/PROSPERO/view/CRD420251131206). The full review protocol is publicly available via the link above. This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.

Inclusion and Exclusion Criteria

The inclusion and exclusion criteria were defined according to the Population, Intervention, Comparison, Outcome, and Study Design (PICOS) framework. Studies were included if they met the following criteria: (1) participants were adults aged 18 years or older with CMP lasting at least 3 months who met multidimensional diagnostic criteria or the ICD-11 diagnostic criteria for CMP, including conditions such as fibromyalgia, knee osteoarthritis, low back pain, myofascial pain, and neck pain;(2) interventions involved any NIBS technique, such as rTMS, tDCS, cranial electrotherapy stimulation (CES) or transcranial random noise stimulation (tRNS); (3) comparators included sham stimulation or an eligible active intervention; (4) reported at least one of the following outcomes: pain intensity measured by the Visual Analogue Scale or Numerical Rating Scale, anxiety assessed with the Hamilton Anxiety Rating Scale, Beck Anxiety Inventory, or Hospital Anxiety and Depression Scale (HADS); or depression assessed using the Hamilton Depression Rating Scale, Beck Depression Inventory, or HADS. The psychometric properties of the outcome measures included in this study have been thoroughly validated in the existing literature and have been shown to possess good reliability and validity [30–33]; and (5) the study design was a randomized controlled trial published in English and conducted in human participants; there are no restrictions on study duration or setting.

Studies were excluded if they met any of the following criteria: (1) they involved ineligible populations, such as healthy volunteers, patients with acute pain, or patients with other pain conditions, including primary headache disorders, neuropathic pain, visceral pain, orofacial pain, or cancer pain, or they included mixed chronic pain populations without separate reporting of relevant results; (2) they used irrelevant interventions, such as invasive procedures or peripheral stimulation techniques, including deep brain stimulation or transcutaneous electrical nerve stimulation; (3) they did not report the outcomes of interest, or the reported data were unavailable, could not be transformed using standard statistical methods, or contained major errors; or (4) they were nonrandomized controlled trials, reviews, secondary analyses, observational studies, quasi-experimental studies, case reports, animal studies, letters, or conference abstracts. In cases of duplicate publications, the report with the larger sample size and more detailed information was included.

Search Strategy

We systematically searched PubMed, the Cochrane Library, the World Health Organization International Clinical Trials Registry Platform, Web of Science, Embase, MEDLINE, PsycINFO, and CINAHL from database inception to 7 October 2025. Searches of the title and abstract fields combined Medical Subject Headings (MeSH), Emtree terms, and relevant free-text keywords using appropriate truncation and Boolean operators. The subject search terms included, but were not limited to, “Chronic Pain [MeSH],” “Musculoskeletal Pain [MeSH],” “Musculoskeletal Diseases [MeSH],” “Transcranial Magnetic Stimulation [MeSH],” “Transcranial Direct Current Stimulation [MeSH],” “Electric Stimulation Therapy [MeSH],” “Low Back Pain [MeSH],” “Osteoarthritis [MeSH],” “Neck Pain [MeSH],” “Shoulder Pain [MeSH],” “Patellofemoral Pain Syndrome [MeSH],” “Myofascial Pain Syndromes [MeSH]” “Fibromyalgia [MeSH],” and “Randomized Controlled Trial [Publication Type]”. The complete search strategy, with PubMed as an example, is provided in Appendix S1 of the Supplementary Materials. The search strategy was adapted to the syntax and indexing characteristics of each database.

Literature Screening and Data Extraction

All records were imported into EndNote, and duplicate records were removed using the software’s built-in duplicate detection function. Two researchers independently screened titles and abstracts against the predefined inclusion and exclusion criteria to identify studies eligible for full-text assessment. Full texts were then reviewed independently by the same two researchers. Data extraction was conducted independently using standardized forms, and the extracted data were cross-checked. Additionally, Cohen’s kappa coefficient was used to assess inter-rater agreement, where agreement was classified as poor (0–0.20), fair (0.21–0.40), moderate (0.41–0.60), good (0.61–0.80), or very good (0.81–1.00) [4]. The extracted information included: (1) general study characteristics, such as article title, first author, publication year, country, disease type, sample size, sex distribution, and mean age; (2) characteristics of the intervention and comparator groups, including stimulation technique, target site, treatment frequency, and treatment duration; and (3) outcome data included the sample size of each group before and after treatment, as well as the values or changes in pain, anxiety, and depression scores of baseline and post-treatment. When the standard deviation (SD) of the change score was not reported, it was calculated using the following formula: ${\mathit{SD}}_{\mathit{change}}=\sqrt{{\mathit{SD}}_{\mathit{baseline}}^2+{\mathit{SD}}_{\mathit{final}}^2-2\times Corr\times {\mathit{SD}}_{\mathit{baseline}}\times {\mathit{SD}}_{\mathit{final}}}$ with the correlation coefficient set at 0.50, as commonly applied in previous studies. Any disagreements regarding study selection or data extraction were resolved through discussion with a third researcher. For studies reporting results only in graphical form, data were extracted using GetData Graph Digitizer.

Risk of Bias Assessment and Certainty of Evidence

Two researchers independently assessed the risk of bias using the Cochrane Risk of Bias 2 (RoB 2) tool, which evaluates bias arising from the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selection of the reported result. Studies were classified as having low risk, some concerns, or high risk of bias according to the RoB 2 algorithm [34]. Disagreements were resolved through discussion with a third researcher.

The certainty of evidence for pairwise meta-analyses was assessed using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. The certainty of evidence for NMA was assessed using the Confidence in Network Meta-Analysis (CINeMA) framework for both direct and indirect comparisons [35, 36].

Statistical Analysis

Statistical analyses were performed using Stata version 17.0. Because all outcomes were continuous, effect estimates were expressed as standardized mean differences (SMDs) with 95% confidence intervals (CIs) when similar outcomes were measured using different scales. and as mean differences (MDs) with 95% CIs when the same outcome was measured using the same scale. Effect sizes were interpreted as trivial (< 0.20), small (0.20–0.49), moderate (0.50–0.79), or large (≥ 0.80) [37].

For each outcome, the geometry of the evidence network was illustrated using a network plot, in which node size was proportional to the number of participants and edge width to the number of studies directly comparing the two interventions. To assess the transitivity assumption, we compared the distribution of potential effect modifiers, including age, sample size, duration of a single stimulation session, total number of stimulation sessions, and overall treatment duration, across treatment comparisons. When closed loops were present, inconsistency between direct and indirect evidence was assessed globally using the design-by-treatment interaction model and locally using node-splitting and loop-specific methods.

Interventions were ranked using the surface under the cumulative ranking curve (SUCRA), which reflects the relative probability that each intervention is among the most effective treatments. Higher SUCRA values indicate more favorable rankings [38]. For interventions informed by fewer than three studies, SUCRA estimates were considered potentially unstable. Therefore, sensitivity analyses were performed by excluding these interventions and by recalculating SDs under different correlation coefficient assumptions to assess the robustness of the main findings. Potential publication bias and small-study effects were evaluated using funnel plots. We assumed that small-sample studies were more likely to favor active treatments over sham controls, and ranked the treatments in order from sham to active (sham = 1) [39]. Post hoc meta-regression was conducted to examine the effects of demographic variables (age and sample size) and clinical covariates (duration of a single stimulation session, total number of stimulation sessions, and overall treatment duration) on the outcome measures. In addition, subgroup analyses were conducted according to disease type. When fewer than ten studies were available, pairwise random-effects meta-analyses were performed using restricted maximum likelihood estimation.

Protocol Amendments

Before the study began, we registered the protocol for this systematic review in PROSPERO (CRD420251131206). During the review process, several modifications were made to the originally registered protocol. The initial protocol prespecified six outcomes. However, only five included studies reported sleep quality, and none reported quality of life, making a meaningful NMA infeasible. Substantial heterogeneity in the instruments used to assess functional outcomes made data pooling inappropriate. Consequently, sleep, function, and quality of life were not included in the final NMA. Importantly, all outcome modifications were made before data extraction and statistical analysis and were based solely on methodological considerations. No changes were made on the basis of the direction, magnitude, or statistical significance of the results. In addition, to explore potential sources of heterogeneity, we conducted random-effects meta-regression analyses based on the primary analysis. Each covariate was examined in a separate univariate meta-regression model. Because these analyses were not prespecified in the original PROSPERO protocol, they are consistently described as post hoc throughout this paper. Given their exploratory nature, the findings should be interpreted with caution.

Result

Research Selection

A total of 9257 records were initially retrieved from the databases, including PubMed (n = 1398), Web of Science (n = 1317), Embase (n = 2113), Cochrane (n = 3014), Medline (n = 883), CINAHL (n = 329), and PsyCINFO (n = 203). After removing 3732 duplicates, 5525 records were screened by title and abstract. Based on the predefined inclusion and exclusion criteria, 231 full-text articles were assessed for eligibility, and 49 studies were ultimately included. One additional eligible study was identified through the reference lists of the included studies and relevant reviews and was incorporated into the NMA. Therefore, a total of 50 studies were ultimately included. Among the included studies, 21 evaluated rTMS [40–60], 23 evaluated tDCS [50, 61–82], two evaluated CES [83, 84], two evaluated transcranial pulsed current stimulation (tPCS) [85, 86], one evaluated intermittent theta-burst stimulation (iTBS) [87], one evaluated transcranial alternating current stimulation (tACS) [88], and one evaluated tRNS [89]; one study evaluated both rTMS and tDCS [50]. The study selection process is shown in Fig. 1. The inter-rater reliability for abstracts (k = 0.83) and full-text articles (k = 0.75) was good to very good.

Fig. 1.

Flow chart of the study identification, screening, eligibility assessment, and inclusion processes

Research Characteristics

Table 1 summarizes the characteristics of the 50 included studies, comprising 2203 participants with mean ages ranging from 31 to 75 years. Women accounted for approximately 82% of the total sample. Of the included studies, 34 focused on fibromyalgia, eight on low back pain, five on knee osteoarthritis, one on myofascial pain, one on neck pain, and two on unspecified CMP. One study included both neck pain and low back pain.

Table 1.

Characteristics of the included studies in the systematic review and network meta-analysis

Study	Country	Type of disease	Intervention measures	Sample size (n)	Gender (Male/Female)	Average age mean ± SD/mean (Age range)	Stimulation parameters	Treatment sessions	Outcomes measures
Ackermann L 2025	Germany	FM	HF-rTMS (DLPFC)	14	0/14	52.93 ± 7.30	10 Hz 1600 pulses (20 min)	10 sessions, 5 sessions/week for 2 weeks	NRS
			Sham	13	0/13	53.62 ± 6.91
Ahn H 2020	United States	KOA	CES	15	5/10	59.77 ± 6.88	0.50 Hz 0.1 mA 60 min	10 sessions, 5 sessions/week for 2 weeks	NRS
			Sham	15	5/10	59.13 ± 5.83
Altas EU 2019	Turkey	FM	HF-rTMS (M1)	10	0/10	46.30 ± 9.01	10 Hz (90% RMT) 1200 pulses (30 min)	15 sessions, 5 sessions/week for 3 weeks	VAS BDI
			HF-rTMS (DLPFC)	10	0/10	47.90 ± 7.89
			Sham	10	0/10	48.20 ± 9.38
Ambriz TM 2016	Mexico	LBP	HF-rTMS (M1)	41	27/14	52.57 ± 10.50	20 Hz (95% RMT) 2000 pulses	5 sessions, 5 sessions/week for 1 week	VAS
			Sham	12	8/4	51.50 ± 12.30
Argaman Y 2022	Israel	FM	HF-rTMS (M1)	27	0/27	37.90 ± 10.00	10 Hz (80% RMT) 2000 pulses (20 min)	10 sessions, 5 sessions/week for 2 weeks	VAS
			Sham	27	0/27	37.90 ± 10.00
Azizi S 2021	Iran	KOA	tDCS (M1)	27	21/6	61.30 ± 13.50	2 mA 20 min	5 sessions, 5 sessions/week for 1 week	VAS
			Sham	27	9/18	56.40 ± 11.70
Badr MY 2024	Egypt	FM	LF-rTMS (DLPFC)	20	9/11	31.90 ± 7.49	1 Hz (120% RMT) 1200 pulses (30 min)	20 sessions, 5 sessions/week for 4 weeks	HDRS HARS
			Sham	20	8/12	34.55 ± 8.32
Bilir I 2021	United States	FM	HF-rTMS (DLPFC)	10	0/10	46.70 ± 9.06	10 Hz (90% RMT) 1500 pulses (15 min)	Induction phase: 10 sessions, 5 sessions/week for 2 weeks Maintenance phase: 4 sessions, 1 session/week for 4 weeks	VAS HADS (anxiety; depression)
			Sham	10	0/10	43.80 ± 9.37
Boyer L 2014	Italy	FM	HF-rTMS (M1)	19	0/19	49.10 ± 10.60	10 Hz (90% RMT) 2000 pulses	Induction phase: 10 sessions, 5 sessions/week for 2 weeks Maintenance phase: 4 sessions, 1 session/2 weeks for 8 weeks	VAS BDI HADS (anxiety)
			Sham	19	1/18	47.70 ± 10.40
Brietzke AP 2020	Brazil	FM	tDCS (DLPFC)	10	0/10	/	2 mA 30 min	60 sessions, 5 sessions/week for 12 weeks	VAS BDI
			Sham	10	0/10	/
Carretero B 2009	Spain	FM	LF-rTMS (DLPFC)	14	0/14	47.50 ± 5.70	1 Hz (110% RMT) 1200 pulses (30 min)	20 sessions, 5 sessions/week for 4 weeks	HDRS
			Sham	12	2/10	54.90 ± 4.90
Caumo W 2022	Brazil	FM	tDCS (DLPFC)	32	0/32	49.60 ± 9.00	2 mA 20 min	20 sessions, 5 sessions/week for 4 weeks	BDI
			Sham	16	0/16	48.00 ± 6.72
Caumo W 2024	Brazil	FM	tDCS (M1)	34	0/34	48.98 ± 10.08	2 mA 20 min	20 sessions, 5 sessions/week for 4 weeks	NRS
			Sham	17	0/17	49.70 ± 8.97
			tDCS (DLPFC)	34	0/34	48.56 ± 8.93
			Sham	17	0/17	40.61 ± 9.70
Corti EJ 2024	Australia	LBP	tDCS (DLPFC)	15	/	52.13 ± 15.26	1.5 mA 20 min	8 sessions, 2 sessions/week for 4 weeks	VAS
			Sham	15	/	62.80 ± 11.95
Curatolo M 2017	Italy	FM	tRNS (M1)	10	0/10	41.40 ± 10.25	1.5 mA 10 min	10 sessions, 5 sessions/week for 2 weeks	VAS HADS (anxiety; depression)
			Sham	10	0/10	44.20 ± 9.81
Dall’Agnol L 2014	Brazil	MPS	HF-rTMS (M1)	12	0/12	45.83 ± 9.63	10 Hz (80% RMT) 1600 pulses	10 sessions, 5 sessions/week for 2 weeks	VAS
			Sham	12	0/12	44.83 ± 14.09
de Melo GA 2020	Brazil	FM	tDCS (M1)	11	0/11	/	2 mA 20 min	5 sessions, 5 sessions/week for 1 week	VAS
			tDCS (M1)	9	0/9	/
			Sham	11	0/11	/
Fagerlund AJ 2015	Norway	FM	tDCS (M1)	24	0/24	49.04 ± 8.63	2 mA 20 min	5 sessions, 5 sessions/week for 1 week	NRS HADS (anxiety; depression)
			Sham	24	3/21	48.17 ± 10.56
Forogh B 2021	Iran	FM	HF-rTMS (DLPFC)	15	0/15	45.73 ± 9.32	10 Hz (100% RMT) 1000 pulses 2 mA 20 min	3 sessions, 3 sessions/week for 1 week	VAS
			tDCS (DLPFC)	15	0/15	46.07 ± 11.73
Fregni F 2006	Brazil	FM	tDCS (M1)	11	0/11	54.80 ± 9.30	2 mA 20 min	5 sessions, 5 sessions/week for 1 week	VAS BDI
			tDCS (DLPFC)	11	0/11	54.20 ± 7.40
			Sham	10	0/10	50.80 ± 10.20
Gabis L 2009	Israel	CMP(NP; LBP)	tPCS	19	8/11	53.90 (36–78)	77 Hz (Up to 4 mA, adjusted by the patient to the peak self-tolerable level) 30 min	8 consecutive sessions	VAS
			Sham	23	11/6	52.10 (28–77)
			tPCS	17	10/13	51.20 (22–77)	77 Hz (Up to 4 mA, adjusted by the patient to the peak self-tolerable level) 30 min	8 consecutive sessions	VAS
			Sham	16	6/10	53.60 (27–82)
Gabis L 2003	Israel	CMP	tPCS	10	6/4	45.80 (20–77)	77 Hz (Up to 4 mA, adjusted by the patient to the peak self-tolerable level) 30 min	8 consecutive sessions	VAS
			Sham	10	3/7	46.70 (27–69)
Harvey MP 2022	Canada	CMP	tDCS (M1)	12	2/10	69.00 ± 7.00	2 mA 20 min	5 sessions, 5 sessions/week for 1 week	NRS BDI BAI
			Sham	12	2/10	68.00 ± 8.00
Hodkinson DJ 2025	United Kingdom	KOA	iTBS (DLPFC)	31	17/14	68.40 ± 7.20	5-Hz repetition rate: 3 pulses at 50 Hz every 200 ms (80% RMT) 9000 pulses	4 consecutive sessions	NRS HADS (anxiety; depression)
			Sham	12	5/7	69.20 ± 9.50
Is EE 2024	Turkey	LBP	tDCS (M1)	16	5/11	65.30 ± 11.80	2 mA 20 min	10 sessions, 5 sessions/week for 2 weeks	VAS
			Sham	14	3/11	69.90 ± 8.90
Jiang NF 2020	China	LBP	tDCS (M1)	26	14/12	39.90 ± 14.20	2 mA 20 min	1 session	NRS
			Sham	25	13/12	44.10 ± 13.00
Kankane AK 2024	India	FM	LF-rTMS (DLPFC)	30	0/30	38.13 ± 7.33	1 Hz (80% RMT) 27 min 10 Hz (80% RMT) 20 min	10 sessions, 5 sessions/week for 2 weeks	NRS HAMA HDRS
			HF-rTMS (DLPFC)	30	0/30	43.23 ± 14.71
			Sham	30	0/30	42.77 ± 10.43
Khedr EM 2017	Egypt	FM	tDCS (M1)	18	1/17	31.30 ± 10.99	2 mA 20 min	10 sessions, 5 sessions/week for 2 weeks	VAS HAMD HAMA
			Sham	18	1/17	33.89 ± 11.18
Lacroix A 2022	France	FM	HF-rTMS (M1)	41	2/39	47.90 ± 7.70	10 Hz (90% RMT) 2000 pulses (20 min)	15 sessions, 5 sessions/week for 3 weeks	BDI
			Sham	37	3/34	47.20 ± 8.50
Lin AP 2022	China	FM	tACS (M1)	19	6/13	48.30 ± 13.60	50 Hz 1 mA 20 min	10 sessions, 5 sessions/week for 2 weeks	NRS BDI BAI
			Sham	19	2/17	48.90 ± 12.30
Loreti EH 2023	Brazil	FM	tDCS (M1)	17	0/17	42.82 ± 9.02	2 mA 13:20:13 min	10 sessions, 5 sessions/week for 2 weeks	VAS HAMA HAMD
			Sham	18	0/18	41.17 ± 11.31
Luedtke K 2015	Germany	LBP	tDCS (M1)	67	34/33	45.00 ± 9.00	2 mA 20 min	5 sessions, 5 sessions/week for 1 week	VAS HADS (anxiety; depression)
			Sham	68	38/30	44.00 ± 10.00
Mariano TY 2019	United States	LBP	tDCS (DLPFC)	10	9/1	65.70 ± 8.80	2 mA 20 min	10 sessions, 5 sessions/week for 2 weeks	NRS
			Sham	11	9/2	60.70 ± 11.80
Martorella G 2022	United States	KOA	tDCS (M1)	60	20/40	65.32 ± 8.41	2 mA 20 min	15 sessions, 5 sessions/week for 3 weeks	NRS
			Sham	60	18/42	66.60 ± 8.43
McPhee ME 2021	Denmark	LBP	tDCS (Other)	12	3/9	28.60 ± 5.90	2 mA 20 min	3 consecutive sessions	VAS
			Sham	12	3/9	28.60 ± 5.90
Mhalla A 2011	France	FM	HF-rTMS (M1)	20	0/20	51.80 ± 11.60	10 Hz (80% RMT) 1500 pulses	Induction Phase: 5 sessions, 5 sessions/week for 1 week Maintenance phase: 3 sessions, 1 session/week for 3 weeks; 3 sessions; 3 sessions, 1 session/month for 3 months	BDI HADS (anxiety; depression)
			Sham	20	0/20	49.60 ± 10.00
Passard A 2007	France	FM	HF-rTMS (M1)	15	/	52.60 ± 7.90	10 Hz (80% RMT) 2000 pulses	10 sessions, 5 sessions/week for 2 weeks	VAS HDRS HADS (anxiety; depression)
			Sham	15	/	55.30 ± 8.90
Short BE 2011	United States	FM	HF-rTMS (DLPFC)	10	1/9	54.20 ± 8.28	10 Hz (120% RMT) 4000 pulses (20 min)	10 sessions, 5 sessions/week for 2 weeks	NRS HDRS
			Sham	10	2/8	51.67 ± 18.19
Silva VA 2025	Multicenter initiated by Brazil	FM	HF-rTMS (M1)	52	0/52	49.00 (25–79)	10 Hz (80% RMT) 3000 pulses (15 min)	Induction Phase: 10 sessions, 5 sessions/week for 2 weeks Maintenance phase: 6 sessions, 1 session/week for 6 weeks Extended maintenance phase: 4 sessions, 1 sessions/2 week for 8 weeks	NRS
			Sham	49	0/49	47.10 (31–83)
Taylor AG 2013	United States	FM	CES	17	1/16	51.90 ± 10.60	0.5 Hz (100 μA) 60 min	Daily sessions for 8 weeks	NRS
			Sham	14	1/13	51.50 ± 10.90
Tekin 2014	Turkey	FM	HF-rTMS (M1)	27	3/24	42.40 ± 7.63	10 Hz (100% RMT) 1500 pulses	10 consecutive sessions	VAS
			Sham	24	1/23	46.50 ± 8.36
Tiwari VK 2024	India	FM	LF-rTMS (DLPFC)	17	0/17	40.70 ± 9.60	1 Hz (90% RMT) 1200 pulses	20 sessions, 5 sessions/week for 4 weeks	VAS
			Sham	17	0/17	40.50 ± 8.10
To WT 2017	Belgium	FM	tDCS (Other)	15	3/12	47.13 ± 10.01	1.5 mA 20 min	8 sessions, 2 sessions/week for 4 weeks	NRS
			tDCS (DLPFC)	11	1/10	47.81 ± 10.17
			假tDCS	16	2/14	46.19 ± 10.00
Villamar MF 2013	United States	FM	tDCS (M1)	18	3/15	50.30 ± 8.50	2 mA 20 min	1 session	VAS
			Sham	18	3/15	50.30 ± 8.50
			tDCS (M1)	18	3/15	50.30 ± 8.50
YaĞCi İ 2014	Turkey	FM	LF-rTMS (M1)	12	0/12	45.25 ± 9.33	1 Hz (90% RMT) 1200 pulses	10 sessions, 5 sessions/week for 2 weeks	VAS BDI
			Sham	13	0/13	43.00 ± 7.63
Yoo HB 2018	United States	FM	tDCS (Other)	21	1/20	47.81 ± 8.23	1.5 mA 20 min	8 sessions, 2 sessions/week for 4 weeks	NRS BDI
			Sham	16	1/15	47.19 ± 8.14
Jales J 2015	Brazil	FM	tDCS (M1)	10	0/10	/	1.0 mA 20 min	10 sessions, 1 session/week for 10 weeks	VAS
			Sham	10	0/10	/
Tanwar S 2020	India	FM	LF-rTMS (DLPFC)	45	0/45	41.54 ± 8.58	1 Hz (90% RMT) 1200 pulses (27 min)	20 sessions, 5 sessions/week for 4 weeks	NRS HARS HDRS
			Sham	41	0/41	39.05 ± 7.12
Tavares DRB 2021	Brazil	KOA	tDCS (M1)	51	9/42	74.78 ± 7.44	2.0 mA 20 min	15 sessions, 5 sessions/week for 3 weeks	VAS
			Sham	53	7/46	73.13 ± 8.51
Fitzgibbon BM 2018	Australia	FM	HF-rTMS (DLPFC)	14	1/13	45.07 ± 11.02	10 Hz (120% RMT) 3000 pulses	20 sessions, 5 sessions/week for 4 weeks	NRS BAI BDI
			Sham	12	1/11	46.25 ± 15.04

FM fibromyalgia, KOA knee osteoarthritis, LBP lower back pain, NP neck pain, MPS myofascial pain syndrome, CMP chronic musculoskeletal pain, HF-rTMS high-frequency repetitive transcranial magnetic stimulation, LF-rTMS low-frequency repetitive transcranial magnetic stimulation, tDCS transcranial direct current stimulation, iTBS intermittent theta-burst stimulation, CES cranial electrotherapy stimulation, tRNS transcranial random noise stimulation, tACS transcranial alternating current stimulation, tPCS transcranial pulsed current stimulation, DLPFC dorsolateral prefrontal cortex, M1 primary motor cortex, VAS Visual Analogue Scale, NRS Numeric Rating Scale, BDI Beck Depression Inventory, HDRS (HAMD) Hamilton Depression Rating Scale, BAI Beck Anxiety Inventory, HARS (HAMA) Hamilton Anxiety Rating Scale, HADS Hospital Anxiety and Depression Scale, SD standard deviation

For tDCS, sample sizes ranged from 20 to 135 participants, current intensity from 1 to 2 mA, number of sessions from 1 to 60, and stimulation duration from 20 to 30 min. For rTMS, sample sizes ranged from 20 to 101 participants, stimulation frequency from 1 to 20 Hz, number of sessions from 3 to 20, and pulse counts from 1000 To 4000. For tPCS, sample sizes ranged from 20 to 42 participants; the maximum current intensity was 4 mA, with eight sessions and a stimulation duration of 30 min. For CES, two studies enrolled 30 and 31 participants, respectively; both used 0.1 mA stimulation for 60 min and included 10 and 40 sessions, respectively. For tACS and tRNS, sample sizes were 38 and 20 participants, stimulation intensities were 1 and 1.5 mA, both used ten sessions, and stimulation durations were 20 and 10 min, respectively. The single iTBS study included 43 participants and used four sessions, a stimulation duration of approximately 10 min, and 9000 pulses.

Transversality Assessment

The studies included in this review shared key characteristics (see Table 1). Although heterogeneity was present in stimulation targets and intervention methods, we addressed this issue by classifying the studies into different intervention nodes. The distribution of potential effect modifiers across studies appeared balanced, supporting the transitivity assumption.

Risk of Bias Assessment

Figures 2 and 3 summarize the risk of bias of the 50 included randomized controlled trials. Of these, 12 were rated as high risk, 23 as having some concerns, and 15 as low risk. The most common sources of bias were insufficient reporting of randomization, inappropriate statistical analyses, and inadequate handling of missing data.

Fig. 2.

Risk of bias summary of included studies

Fig. 3.

Proportion of risk levels of bias in each domain

NMA Results

Primary Outcome

Network Structure and Geometry

Figure 4 shows the network structure used to compare interventions for pain, anxiety, and depression. The pain network had 13 nodes and 17 direct comparisons; the anxiety network included eight nodes and eight direct comparisons; and the depression network consisted of 11 nodes and 13 direct comparisons. Sham treatment was the most connected node.

Fig. 4.

Network geometry for pain, anxiety, and depression in chronic musculoskeletal pain. a Pain; b Anxiety; c Depression. Nodes denote treatments, and edges connecting nodes represent direct comparisons. Node size is weighted by the total sample size, and edge thickness is weighted by the standard error. HF-rTMS high-frequency repetitive transcranial magnetic stimulation, LF-rTMS low-frequency repetitive transcranial magnetic stimulation, tDCS transcranial direct current stimulation, iTBS intermittent theta-burst stimulation, CES cranial electrotherapy stimulation, tRNS transcranial random noise stimulation, tACS transcranial alternating current stimulation, tPCS transcranial pulsed current stimulation, DLPFC dorsolateral prefrontal cortex, M1 primary motor cortex

Inconsistency Test

The global inconsistency test showed no significant inconsistency between direct and indirect evidence for pain (p = 0.10), anxiety (p = 0.26), or depression (p = 0.20), indicating that the consistency assumption was satisfied for all models. These results support the validity of the network models. In the local inconsistency assessment using the node-splitting method, inconsistency was identified in two comparisons for the pain outcome. For all other comparisons, p values were not statistically significant, indicating a low level of local inconsistency. Loop-specific analyses showed no inconsistency in any closed loop.

Network Estimates

Figure 5 presents the standard mean differences of all pairwise comparisons and their 95% confidence intervals, with statistically significant results highlighted in red. Appendix S2 in the electronic supplementary material presents forest plots for primary analyses of pain, anxiety, and depression, with effect estimates and corresponding 95% CIs shown in figures.

Fig. 5.

League table of pain, anxiety, and depression outcomes in chronic musculoskeletal pain: standardized mean differences and 95% credible intervals. a Pain; b Anxiety; c Depression. HF-rTMS high-frequency repetitive transcranial magnetic stimulation, LF-rTMS low-frequency repetitive transcranial magnetic stimulation, tDCS transcranial direct current stimulation, iTBS intermittent theta-burst stimulation, CES cranial electrotherapy stimulation, tRNS transcranial random noise stimulation, tACS transcranial alternating current stimulation, tPCS transcranial pulsed current stimulation, DLPFC dorsolateral prefrontal cortex, M1 primary motor cortex

Pain

A total of 45 studies assessed 13 interventions for pain. Compared with sham stimulation, high-frequency repetitive transcranial magnetic stimulation (HF-rTMS) over the primary motor cortex (M1) (SMD = − 0.74, 95% CI − 1.18 to − 0.30; eight studies, moderate certainty), HF-rTMS over the dorsolateral prefrontal cortex (DLPFC) (SMD = − 0.51, 95% CI − 1.00 to − 0.02; six studies, low certainty), low-frequency repetitive transcranial magnetic stimulation (LF-rTMS) over the DLPFC (SMD = − 1.19, 95% CI − 1.86 to − 0.52; three studies, very low certainty), tDCS over M1 (SMD = − 0.74, 95% CI − 1.06 to − 0.42; 15 studies, moderate certainty), and CES (SMD = − 1.04, 95% CI − 1.97 to − 0.11; two studies, very low certainty) were associated with significant pain relief in CMP. No statistically significant difference was found in direct comparisons among these five active interventions.

Anxiety

For anxiety, 16 studies examined eight interventions. LF-rTMS over the DLPFC (SMD = – 0.66, 95% CI − 1.26 to − 0.07; three studies, low certainty), tDCS over M1 (SMD = − 0.66, 95% CI − 1.16 to − 0.16; five studies, very low certainty), and tRNS over M1 (SMD = − 1.66, 95% CI − 3.05 to − 0.28; one study, low certainty) performed better than sham stimulation in reducing anxiety symptoms. Direct comparisons among these three interventions did not show significant differences.

Depression

A total of 25 studies evaluated 11 interventions for depression. Compared with sham stimulation, HF-rTMS over the DLPFC (SMD = − 0.49, 95% CI − 0.88 to − 0.10; five studies, moderate certainty), LF-rTMS over the DLPFC (SMD = − 0.63, 95% CI − 1.00 to − 0.26; four studies, moderate certainty), tDCS over M1 (SMD = − 0.38, 95% CI − 0.71 to − 0.05; six studies, moderate certainty), and tDCS over the DLPFC (SMD = − 0.58, 95% CI − 1.09 to − 0.07; three studies, moderate certainty) were more effective in reducing depressive symptoms. Pairwise comparisons among these active interventions showed no significant differences.

Ranking Probabilities

According to the SUCRA rankings, LF-rTMS over the DLPFC had the highest probability of being ranked first for pain relief, followed by tRNS over M1 and CES. For anxiety, tRNS over M1 was most likely to achieve the top ranking, with tDCS over M1 and LF-rTMS over the DLPFC next. Regarding depression, iTBS over the DLPFC appeared most likely to rank first, followed by LF-rTMS over the DLPFC, and then tDCS over the DLPFC (Fig. 6).

Fig. 6.

The surface under the cumulative ranking curve of pain, anxiety, and depression in chronic musculoskeletal pain. a Pain; b Anxiety; c Depression. HF-rTMS high-frequency repetitive transcranial magnetic stimulation, LF-rTMS low-frequency repetitive transcranial magnetic stimulation, tDCS transcranial Direct Current Stimulation, iTBS intermittent theta-burst stimulation, CES cranial electrotherapy stimulation, tRNS transcranial random noise stimulation, tACS transcranial alternating current stimulation, tPCS transcranial pulsed current stimulation, DLPFC dorsolateral prefrontal cortex, M1 primary motor cortex

Sensitivity Analysis

Because only a limited number of studies were available for several interventions, including LF-rTMS targeting M1, iTBS targeting the DLPFC, CES, tACS targeting M1, and tRNS targeting M1, a sensitivity analysis was performed after excluding interventions supported by fewer than three studies, and inconsistency was reassessed. The results showed no significant inconsistency for pain (p = 0.10), anxiety (p = 0.26), or depression (p = 0.20), supporting the continued use of the consistency model. The sensitivity analysis further indicated that LF-rTMS targeting the DLPFC had the highest probability of ranking first for pain, anxiety, and depression. These findings suggest that the effects of interventions supported by sparse evidence should be interpreted with caution (see Appendix S3 in the electronic supplementary material). Furthermore, sensitivity analyses using SD estimates derived under different correlation coefficient assumptions for each outcome measure showed no evidence of significant inconsistency, further supporting the use of the consistency model. When the correlation coefficient was set to 0.70, LF-rTMS targeting the DLPFC ranked first among treatments for depression. This shift may reflect the limited evidence available for interventions such as iTBS targeting the DLPFC and tRNS targeting M1, which were supported by only a small number of studies in the network and were therefore more susceptible to ranking fluctuations under different correlation assumptions. The top-ranked interventions for all other outcomes remained unchanged across scenarios, indicating that these findings were relatively robust to this additional assumption (Appendix S4).

Subgroup Analysis

Fibromyalgia Group

Network Structure and Geometry

Figure 7 shows the network structure of interventions for fibromyalgia. The pain network included 11 intervention nodes and 15 direct comparisons, the anxiety network included seven nodes and seven direct comparisons, and the depression network included ten nodes and 12 direct comparisons.

Fig. 7.

Network geometry for pain, anxiety, and depression in fibromyalgia syndrome subgroups. a Pain; b Anxiety; c Depression. Nodes denote treatments, and edges connecting nodes represent direct comparisons. Node size is weighted by the total sample size, and edge thickness is weighted by the standard error. HF-rTMS high-frequency repetitive transcranial magnetic stimulation, LF-rTMS low-frequency repetitive transcranial magnetic stimulation, tDCS transcranial direct current stimulation, CES cranial electrotherapy stimulation, tRNS transcranial random noise stimulation, tACS transcranial alternating current stimulation, DLPFC dorsolateral prefrontal cortex, M1 primary motor cortex

Inconsistency Test

The global inconsistency test showed no significant inconsistency between direct and indirect evidence for pain (p = 0.28), anxiety (p = 0.21), or depression (p = 0.08), indicating that the consistency assumption was satisfied across all models. These findings support the adequacy of the network models. In addition, both loop-specific and local inconsistency assessments yielded non-significant p values across all comparisons, indicating no evidence of local inconsistency. Therefore, the consistency model was considered appropriate for the NMA.

Network Estimates

Appendix S5 in the electronic supplementary material presents forest plots of pain, anxiety, and depression, with effect estimates and corresponding 95% CIs shown in the figures. The forest plots and league table (Fig. 8) showed that, in patients with fibromyalgia, HF-rTMS targeting M1 (SMD = − 0.73, 95% CI − 1.33 to − 0.13; six studies, low certainty), LF-rTMS targeting the DLPFC (SMD = − 1.19, 95% CI − 1.99 to − 0.39; three studies, low certainty), tDCS targeting M1 (SMD = − 0.95, 95% CI − 1.48 to − 0.43; eight studies, low certainty) were more effective than sham stimulation in reducing pain. For anxiety, LF-rTMS targeting the DLPFC (SMD = − 0.66, 95% CI − 1.21 to − 0.10; three studies, moderate certainty), tDCS targeting M1 (SMD = − 0.97, 95% CI − 1.59 to – 0.36; three studies, low certainty), and tRNS targeting M1 (SMD = − 1.66, 95% CI − 2.99 to − 0.34; one study, low certainty) were superior to sham stimulation. For depression, HF-rTMS targeting the DLPFC (SMD = − 0.49, 95% CI − 0.87 to − 0.12; five studies, moderate certainty), LF-rTMS targeting the DLPFC (SMD = − 0.63, 95% CI − 0.98 to − 0.27; four studies, moderate certainty), tDCS targeting M1 (SMD = − 0.51, 95% CI − 0.92 to − 0.11; four studies, moderate certainty), and tDCS targeting the DLPFC (SMD = − 0.60, 95% CI − 1.10 to − 0.10; three studies, moderate certainty) all showed significant benefits compared with sham stimulation.

Fig. 8.

League table of pain, anxiety, and depression outcomes in fibromyalgia syndrome subgroups: standardized mean differences and 95% credible intervals. a Pain; b Anxiety; c Depression. HF-rTMS high-frequency repetitive transcranial magnetic stimulation, LF-rTMS low-frequency repetitive transcranial magnetic stimulation, tDCS transcranial direct current stimulation, CES cranial electrotherapy stimulation, tRNS transcranial random noise stimulation, tACS transcranial alternating current stimulation, DLPFC dorsolateral prefrontal cortex, M1 primary motor cortex

Ranking Probabilities

According to the SUCRA results, LF-rTMS targeting the DLPFC has the highest probability of ranking first for pain relief in fibromyalgia, followed by tRNS targeting M1 and tDCS targeting M1. For anxiety relief, tRNS targeting M1, tDCS targeting M1, and LF-rTMS targeting the DLPFC are in the top three. For depression relief, LF-rTMS targeting the DLPFC has the highest probability of ranking first, followed by tDCS targeting the DLPFC and tRNS targeting M1 (Fig. 9).

Fig. 9.

The surface under the cumulative ranking curve of pain, anxiety, and depression in fibromyalgia syndrome subgroups. a Pain; b Anxiety; c Depression. HF-rTMS high-frequency repetitive transcranial magnetic stimulation, LF-rTMS low-frequency repetitive transcranial magnetic stimulation, tDCS transcranial direct current stimulation, CES cranial electrotherapy stimulation, tRNS transcranial random noise stimulation, tACS transcranial alternating current stimulation, DLPFC dorsolateral prefrontal cortex, M1 primary motor cortex

Sensitivity Analysis

After interventions supported by fewer than three studies were excluded, the inconsistency was re-evaluated. No significant inconsistency was detected for pain (p = 0.16), anxiety (p = 0.21), or depression (p = 0.08), supporting the continued use of the consistency model. The sensitivity analysis indicated that LF-rTMS targeting the DLPFC, tDCS targeting M1, and LF-rTMS targeting the DLPFC were the interventions most likely to rank first for pain, anxiety, and depression, respectively. Given the limited evidence for tRNS targeting M1, its apparent benefit for anxiety should be interpreted cautiously, and further studies are needed to confirm this finding (Appendix S6). Furthermore, sensitivity analysis of the SD estimates for each outcome measure, based on different correlation coefficient assumptions, also showed no evidence of significant inconsistency, further supporting the use of the consistency model. When the correlation coefficient was set to 0.70, tRNS targeting M1 and iTBS targeting the DLPFC ranked second and third, respectively, for the depression outcome. However, high-frequency repetitive transcranial magnetic stimulation (HF-rTMS) targeting the DLPFC remained ranked first in terms of the area under the curve, indicating that the optimal intervention was unchanged. All other results were consistent across assumptions, suggesting that the findings were robust to this additional assumption (Appendix S7).

Non-Fibromyalgia Group

Pain

Appendix S8 presents the network diagram for CMP in patients without fibromyalgia, comprising eight intervention nodes. Studies in this subgroup shared key characteristics, including a randomized controlled design and comparisons between NIBS and sham stimulation. Although the included patients had different chronic musculoskeletal conditions, their pain was primarily associated with peripheral tissue injury or inflammation, in contrast to the predominantly central mechanisms underlying fibromyalgia, suggesting a degree of clinical similarity across these conditions. However, because no closed loop was formed in the network, inconsistency could not be assessed; therefore, the findings should be interpreted with caution.

Appendix S9-S10 in the electronic supplementary material presents forest plots and the league table of pain, anxiety, and depression, with effect estimates and corresponding 95% CIs shown in the figures. The forest and league table analyses showed that, compared with sham stimulation, HF-rTMS targeting M1 (SMD = − 0.79, 95% CI − 1.48 to − 0.11; two studies, low certainty), tDCS targeting M1 (SMD = − 0.53, 95% CI − 0.84 to − 0.23; seven studies, low certainty), and CES (SMD = − 1.39, 95% CI − 2.41 to − 0.37; one study, low certainty) were more effective for pain relief in chronic musculoskeletal conditions other than fibromyalgia.

According to the SUCRA results, CES had the highest probability of ranking first for pain relief, followed by HF-rTMS targeting M1 and tDCS targeting M1(Appendix S11).

Anxiety and Depression

Because of the limited number of studies in the non-fibromyalgia group, pairwise meta-analyses were performed only for anxiety and depression outcomes. For anxiety, the effects of tDCS targeting M1 (SMD = − 0.18, 95% CI − 0.71 to 0.35, two studies, very low certainty) and iTBS targeting the DLPFC (MD = 0.37, 95% CI − 0.30 to 1.04, one study, low certainty) compared with sham stimulation were uncertain. For depression, iTBS targeting the DLPFC (MD = − 0.79, 95% CI − 1.48 to – 0.10; one study, low certainty) may reduce depressive symptoms in patients without fibromyalgia; however, this finding should be interpreted cautiously because it was based on a single study. In contrast, the evidence for the effect of tDCS targeting M1 on depression was very uncertain (SMD = − 0.20, 95% CI − 0.78 to 0.37, two studies, very low certainty) (Appendix S12-S13).

Sensitivity Analysis

Because few studies were available for pain outcomes, interventions supported by fewer than three studies were retained in the sensitivity analysis; therefore, the results should be interpreted with caution. Analysis of standard error estimates under different correlation coefficient assumptions showed no changes in the direction of effect sizes or in treatment rankings, suggesting that the results were robust to this additional assumption (Appendix S14-S16).

Publication Bias and Confidence in the Network Estimates

Publication bias was assessed using funnel plots; the relevant figures are provided in Supplementary Materials Appendix S17-S19. Overall, the plots were generally symmetrical, suggesting no strong evidence of small-study effects or publication bias among the included studies. The certainty of evidence, assessed using the GRADE approach for pairwise meta-analyses and the CINeMA framework for NMA, is presented in Supplementary Materials Appendix S20-S23.

Meta-Regression Results

A post hoc meta-regression was conducted to examine potential moderators. Univariate meta-regression showed that, for anxiety outcomes, the mean age of the study population was significantly associated with effect size (coefficient = 0.03, 95% CI [0.00, 0.06], p = 0.04), accounting for 28.8% of the between-study variance. However, substantial residual heterogeneity remained (I²res = 61.12%), suggesting that other unmeasured moderators may be present. Because meta-regression is an observational, study-level analysis, this association should not be interpreted as causal at the individual level and requires further validation using individual participant data from the original studies. None of the other covariates significantly affected the results (Appendix S24).

Discussion

To our knowledge, this is the first NMA to quantify the effects of NIBS on pain, anxiety, and depression in patients with CMP. The primary analysis provided very low-certainty evidence that LF-rTMS targeting the DLPFC was the most effective intervention for pain relief, low certainty evidence indicates that tRNS targeting M1 and iTBS targeting the DLPFC showed the greatest benefit for anxiety and depression, respectively. However, sensitivity analyses showed that after exclusion of interventions supported by limited evidence, LF-rTMS targeting the DLPFC remained consistently superior, suggesting that it may be a robust treatment option for CMP and comorbid mood symptoms. Furthermore, when the correlation coefficient was set at 0.7, LF-rTMS targeting the DLPFC emerged as the most effective intervention for reducing depressive symptoms, surpassing tRNS targeting M1. This finding suggests that the apparent advantage of tRNS targeting M1 in the primary analysis was largely driven by a small number of studies. Therefore, the apparent benefits of tRNS targeting M1 and iTBS targeting the DLPFC should be interpreted cautiously, and further high-quality head-to-head trials are needed to confirm these findings. Subgroup analyses further showed that disease heterogeneity substantially influenced treatment effects. In the fibromyalgia subgroup, LF-rTMS targeting the DLPFC showed the greatest efficacy for both pain (low certainty) and depression (Moderate certainty), consistent with the pathophysiology of fibromyalgia, which is characterized by central sensitization and frequent emotional comorbidity. For anxiety, low certainty evidence suggests that tRNS targeting M1 ranked highest; however, this finding was based on a single study. Sensitivity analyses excluding studies with sparse evidence indicate that tDCS targeting M1 may be more suitable for treating anxiety in fibromyalgia. In contrast, the non-fibromyalgia subgroup, including conditions such as osteoarthritis and chronic low back pain, showed a different efficacy profile, with CES (low certainty) based on a single study ranking first for pain relief. Therefore, the result should be interpreted cautiously and warrant confirmation by future high‑quality studies. Evidence for emotional outcomes in the non-fibromyalgia subgroup was also extremely limited, allowing only pairwise meta-analyses. The potential benefit of iTBS targeting the DLPFC (low certainty) for depression in this subgroup should therefore be interpreted cautiously, whereas tDCS targeting M1 and iTBS targeting the DLPFC did not show clear efficacy for anxiety.

From a neurobiological perspective, the DLPFC is a key hub involved in cognitive control, emotional regulation, and descending pain modulation [90–92]. In chronic pain, the DLPFC exhibits reduced volume, abnormal functional activity, and altered connectivity with the limbic system, somatosensory networks, and the default mode network; these changes are closely related to pain and negative affect [93, 94]. LF-rTMS targeting the DLPFC may decrease local cortical excitability via long-term depression mechanisms, thereby enhancing descending pain inhibitory pathways involving the periaqueductal gray and rostral ventromedial medulla. Strengthening this endogenous analgesic pathway can reduce pain intensity [95–97]. In addition, the DLPFC exerts top-down control over cognitive reappraisal and pain attribution; in patients with comorbid emotional disorders, LF-rTMS targeting the DLPFC may thus reduce pain and functional impairment both directly and indirectly by improving emotional and cognitive processing [98–100]. This is consistent with the findings of the sensitivity analysis and the fibromyalgia subgroup in the present study. As a novel rTMS protocol, iTBS induces long-term potentiation-like effects and transiently increases cortical excitability by mimicking physiological theta rhythms[101]. iTBS targeting the DLPFC may increase cortical excitability, normalize aberrant coupling between the posterior default mode network and the salience network, and enhance prefrontal inhibitory control over limbic structures, including the amygdala and anterior cingulate cortex, thereby alleviating depressive symptoms and reducing negative pain appraisal [102–105]. The U.S. Food and Drug Administration has approved 10-Hz high-frequency rTMS targeting the left DLPFC for the treatment of depression [106]. As an optimized protocol, iTBS substantially shortens stimulation duration while maintaining similar intensity and shows comparable antidepressant efficacy [107]. This is consistent with the present findings. By contrast, the mechanism by which tRNS may alleviate anxiety remains unclear. One possible explanation is that tRNS enhances neural responses through stochastic resonance, thereby improving the signal-to-noise ratio of weak inputs, promoting synaptic plasticity, and transiently enhancing cortical responsiveness [108, 109]. Although M1 is not a classical emotional center, it is functionally connected to emotion-related regions such as the insula, which is involved in emotional regulation [110, 111]. Thus, tRNS targeting M1 may indirectly modulate anxiety and emotional regulation by altering functional connectivity between M1 and these regions. However, the number of tRNS and iTBS studies included in this review was very small, and the available evidence remains limited. Thus, although the primary analysis suggested potential benefits for anxiety and depression, these findings require confirmation in larger randomized controlled trials.

The subgroup findings also support a mechanism-based interpretation of treatment response. Fibromyalgia is widely regarded as a central sensitization syndrome and is often accompanied by substantial emotional distress [112]. Neuroimaging studies have shown disrupted functional connectivity within the default mode network in chronic pain, including abnormalities in the prefrontal and anterior cingulate cortices [113]. Systematic reviews in fibromyalgia have further identified structural and functional abnormalities in these regions, supporting their central role in the pathophysiology of central sensitization [114]. In addition, a recent meta-analysis showed that rTMS targeting the DLPFC significantly improved pain and depressive symptoms in fibromyalgia, with effects lasting up to 1 month after treatment [115]. These findings provide a neurobiological basis for the dual benefits of LF-rTMS targeting the DLPFC observed in our fibromyalgia subgroup analysis. In the non-fibromyalgia subgroup, which was characterized primarily by peripheral structural damage, CES showed the most favorable pain-relieving profile among the included interventions. A review of CES in chronic pain management reported benefits for spinal pain, headache, toothache, and pain-related muscle spasm [116]. The mechanism underlying the analgesic effects of CES remains incompletely understood. Available evidence suggests that CES may modulate autonomic function, increase the production of neurotransmitters such as dopamine, raise plasma β-endorphin levels, and lower plasma cortisol levels. These physiological changes attenuate pain facilitation and stress-related hyperalgesia and could account for the greater analgesic response observed in the CMP subgroup characterized predominantly by peripheral structural pathology [117]. This may provide a biological rationale for its analgesic effects. However, only one CES study, conducted in patients with knee osteoarthritis, was included in this review. Therefore, the generalizability of this finding is limited and may not extend to other CMP conditions. In addition, the network had a star-shaped structure, and the number of studies for each intervention was small; accordingly, these results should be interpreted with caution. For emotional outcomes in the non-fibromyalgia subgroup, the evidence was extremely limited, underscoring the need for future high-quality randomized controlled trials of NIBS in patients with peripheral CMP.

Limitations and Future Directions

This study has several limitations. First, fibromyalgia accounted for a large proportion of the included CMP conditions, whereas disorders primarily characterized by peripheral inflammation, such as osteoarthritis, were underrepresented. At the same time, sleep, function, and quality of life were excluded because of sparse data and heterogeneity in measurement tools. Consequently, this review provides limited evidence on these outcomes, which may have introduced reporting bias. In addition, some interventions were supported by only a small number of studies, resulting in substantial uncertainty in the effect estimates. Therefore, these findings should be interpreted cautiously. Future studies should expand evidence on underrepresented interventions and disease categories through more studies and larger sample sizes. Direct head-to-head trials and standardized assessment tools for multidimensional outcomes, such as sleep and quality of life, are also needed to support future evidence synthesis and make up for the lack of existing evidence. Second, the analysis was based only on stimulation modality and target site, although stimulation parameters such as frequency, intensity, and treatment duration varied across studies. This heterogeneity may have affected the pooled effect estimates. Future research should standardize these parameters and evaluate their influence on treatment outcomes to identify the optimal NIBS protocol. Third, the included studies used a variety of outcome assessment scales. Although data were pooled using SMDs, this approach limits direct clinical interpretation of effect sizes on specific scales, and measurement-related bias may still remain. Therefore, caution is warranted when interpreting the magnitude of SMDs. Future studies should adopt standardized assessment guidelines and use validated, reliable, and consensus-based instruments. Fourth, this study mainly evaluated efficacy immediately after the final treatment session. The long-term effects of NIBS remain uncertain and require further investigation. Future studies should include longer follow-up periods to assess durability and determine the optimal maintenance regimen.

Conclusions

In conclusion, LF-rTMS targeting the DLPFC may represent the most robust option for improving pain and emotional symptoms in patients with CMP, particularly pain and depressive symptoms in those with fibromyalgia. tDCS targeting M1 may be a more appropriate option for anxiety in fibromyalgia after exclusion of sparsely supported studies. Among patients without fibromyalgia, CES showed favorable effects on pain relief and iTBS targeting the DLPFC for depressive symptoms, although the evidence remains limited. Therefore, these findings are exploratory and require confirmation in future high-quality randomized controlled trials.

Supplementary Information

Below is the link to the electronic supplementary material.

Author Contributions

Yan Li: Conceptualization, Methodology, Software, Study screening and quality assessment, Data extraction, Statistical analysis, Data interpretation, Writing—original draft, Writing—review and editing. Xiangyin Ye: Software, Methodology, Study screening and quality assessment, Statistical analysis, Writing—original draft. Shufang Li: Software, Methodology, Study screening and quality assessment, Statistical analysis, Writing—original draft. Jing Xu: Literature search, Data extraction, Statistical analysis. Hongyuan Wang: Conceptualization, Data curation, Writing—review and editing. Lina Pang: Conceptualization, Data curation, Writing—review and editing. Yuanlin Li: Conceptualization, Data curation, Writing—review and editing. Song Jin: Conceptualization, Supervision, Writing—original draft, Writing—review and editing.

Funding

The author(s) declared that financial support was received for this work and/or its publication. This study was financially supported by the Sichuan Province Rehabilitation Medicine Association Project, No. SCKFKY20250103, the Sichuan Provincial Health Science Project, No. ZH2025-501, the Sichuan Provincial Administration of Traditional Chinese Medicine Project, No. 25MSZX267, and the Deyang Municipal Health Commission, No. 2025SYQYYL048. The Rapid Service Fee for this article was funded by the Sichuan Province Rehabilitation Medicine Association Project (No. SCKFKY20250103).

Data Availability

All data generated or analyzed during this study are included in this published article/as supplementary information files.

Declarations

Conflict of Interest

All authors (Yan Li, Xiangyin Ye, Shufang Li, Jing Xu, Hongyuan Wang, Lina Pang, Yuanlin Li, Song Jin) declare that there are no conflicts of interest in this study.

Ethical Approval

This article is based on previously conducted studies and does not contain any new studies with human participants or animals performed by any of the authors.