Effectiveness of neuromodulation techniques for migraine prophylaxis: a systematic review and network meta-analysis of randomized controlled trials

Abstract

Background

Given limited efficacy and potential complications of pharmacological approaches, neuromodulation techniques are emerging as non-pharmacological alternatives for migraine prevention. The present study aimed to conduct a systematic review and network meta-analysis to compare the effectiveness of neurostimulation interventions for migraine prophylaxis.

Methods

The PubMed/MEDLINE, Cochrane, Web of Science, Embase, Clinicaltrials.gov, China National Knowledge Infrastructure, Chongqing VIP, Wanfang, Chinese Biomedical Literature Database, Chinese Clinical Trial Registry, and International Traditional Chinese Medicine Clinical Trial Registry databases were systematically searched up to February 7th, 2025 for randomized controlled trials (RCTs). Outcomes of interest were changes in monthly migraine days, response rate and changes in pain intensity. Network meta-analyses were based on a Bayesian framework.

Results

Forty RCTs (N = 4341; mean age = 38.7 years; % females = 81.2) were included in the network meta-analysis. Transcranial magnetic stimulation (TMS), transcranial electrical stimulation (tES), percutaneous mastoid electrical stimulation (PMES), supraorbital transcutaneous stimulation (STS), and acupuncture were associated with significant improvements in migraine frequency, response rate and pain severity. Both invasive and non-invasive occipital nerve stimulation (ONS) demonstrated significantly higher response rates relative to sham controls. Among all the investigated interventions, tES (SUCRA = 82%; SMD = 0.9; 95% CrI = 0.32, 1) yielded the greatest reduction in monthly migraine days, transcutaneous ONS (tONS) (SUCRA = 90%; RR = 12; 95% CrI = 1.9, 389) exhibited the highest response rate, and PMES (SUCRA = 96%; SMD = 2.4; 95% CrI = 0.75, 3.4) yielded the most decreased pain intensity after intervention.

Conclusions

The main findings of this study highlight the beneficial effect of tES and tONS in reducing migraine frequency and improving treatment response, respectively. Due to scanty evidence for certain interventions and network model limitations, caution is needed when interpreting the results. Future large-scale and well-conducted RCTs are required to strengthen the reliability and validity of the findings.

Trial registration

The study protocol was registered on the International Prospective Register of Systematic Reviews. Registration Number: CRD42025642688.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12906-026-05270-0.

Introduction

Migraine is a highly prevalent neurovascular disorder characterized by recurrent headache attacks, frequently accompanied by photophobia, phonophobia, nausea, and vomiting. Affecting more than one billion people globally, migraine is estimated to be the second leading cause of disability worldwide and the foremost cause among females. Owing to its high prevalence, recurrence, and substantial impact on quality of life, preventive therapies are essential for long-term migraine management [1–5].

Conventional prophylaxis includes antihypertensive agents, calcium channel blockers, antidepressants, and antiseizure medications. Recently, onabotulinumtoxinA and calcitonin gene-related peptide (CGRP) targeted monoclonal antibodies have been integrated into clinical practice [6]. Notwithstanding pharmacotherapy advancements, poor toleration and certain side effects compromise long-term treatment adherence, potentially leading to acute medication overuse and migraine chronification [7–10]. Furthermore, most pharmacologic studies have focused on episodic migraine, resulting in limited evidence for drug efficacy in chronic migraine prevention [11]. Taken together, the development of non-pharmacological alternatives may be a promising strategy for long-term pain management.

Over recent decades, neuromodulation interventions have been proposed for migraine prophylaxis and shown certain efficacy. Theoretically, neuromodulation techniques exert a modulatory effect on the pathophysiological mechanisms of migraine via central or peripheral pathways. Non-invasive brain stimulation, including transcranial magnetic stimulation (TMS) and transcranial electrical stimulation (tES), play a key role in regulating cortical neuronal activity that is involved in the normalization of pain-processing transmission [12, 13]. In contrast, there are numerous targets for peripheral nerve stimulation with possible modalities involving transcutaneous or invasive occipital nerve stimulation (ONS), non-invasive vagus nerve stimulation (nVNS), percutaneous mastoid electrical stimulation (PMES), supraorbital transcutaneous stimulation (STS), remote electrical neuromodulation (REN), acupuncture, caloric vestibular stimulation (CVS), and kinetic oscillation stimulation (KOS) [14–16]. Stimulation of specific peripheral nerves is targeted at the activation of the trigeminovascular system implicated in migraine pain signaling [17, 18].

Given the wide range of neuromodulation interventions for migraine prevention, evidence-based guidance for clinical decision making is required. Network meta-analysis is able to provide precise estimates for comparative efficacy of multiple interventions by integrating both direct and indirect evidence [19]. Previous network meta-analyses, however, were limited to non-invasive neurostimulation modalities [18, 20]. Thus, we conducted a systematic review and network meta-analysis on randomized controlled trials (RCTs) to rank the efficacy of both invasive and non-invasive neuromodulation interventions for migraine prophylaxis.

Methods

The study protocol has been registered on PROSPERO (registration number: CRD42025642688), and the reporting of the present study adhered to the Preferred Reporting Items for Systematic Reviews and Network Meta-analyses (PRISMA-NMA) guidelines [21].

Literature search

PubMed/MEDLINE, Cochrane, Web of Science, Embase, Clinicaltrials.gov, China National Knowledge Infrastructure (CNKI), Chongqing VIP (CQVIP), Wanfang, Chinese Biomedical Literature Database (CBM), Chinese Clinical Trial Registry (ChiCTR), and International Traditional Chinese Medicine Clinical Trial Registry (ITMCTR) from the inception of each database to February 7th, 2025, were systematically searched. Detailed search strategy is documented in Supplementary Method 2.

Inclusion and exclusion criteria

The inclusion criteria based on the PICOS framework comprised: (1) Participants: Patients aged over 15 years meeting the International Classification of Headache Disorders (ICHD) criteria for episodic or chronic migraine; (2) Interventions: Invasive or non-invasive neuromodulation techniques; (3) Comparators: Sham controls or active neuromodulation interventions; (4) Outcomes: Monthly migraine days (MMD), response rate and migraine pain intensity; (5) Study design: Parallel-group RCTs published as full reports, with at least 10 participants per group in final analyses.

Studies were excluded if they (1) assessed the acute treatment of migraine; (2) were cluster or crossover RCTs; (3) included cluster headache, tension-type headache, or migraine comorbid with other neurological disorders; (4) did not report the target outcomes.

Data extraction

Pairs of reviewers worked independently to screen and extract data from eligible studies and evaluate risk of bias using the Cochrane Risk of Bias (RoB) Tool. Any disagreements were resolved through discussion or consultation with other members of the team. Corresponding authors were contacted to obtain unreported essential data.

Outcome measures

Changes in monthly migraine days (MMD) and response rate were considered as primary outcomes, since the prophylactic therapy is aimed to reduce migraine frequency rather than to cure migraine [1]. Due to the variation in the classification of the headache episodes by patients, studies might report either MMD or monthly headache days (MHD). Therefore, with regard to the data extraction for the primary outcome, we gave priority to extract changes in MMD when both MMD and MHD were reported in the study. For studies reporting only MHD, changes in MHD were extracted [18]. Response rate was defined as the proportion of participants achieving ≥ 50% reduction in migraine/headache days or attack frequency, depending on the definition of each study. The secondary outcome was changes in migraine pain intensity measured by Visual Analogue Scale (VAS) or 11-point Numerical Rating Scale (NRS).

Statistical analysis

We performed network meta-analyses based on a Bayesian hierarchical framework using the GeMTC package in R (version 4.5.0). For continuous outcomes, effect sizes were estimated using standardised mean differences (SMD) with 95% credible intervals (CrI) due to the variability of rating scales [20]. Dichotomous outcomes were expressed as risk ratios (RR) with 95% CrI. Random-effects models were used using Markov chain Monte Carlo sampling, with 4 parallel chains, each comprising 20,000 burn-in iterations and 50,000 actual stimulation iterations, and a thinning interval of 10. Convergence of the algorithm was assessed using the Brooks-Gelman-Rubin-Brooks diagnostic, and the visual inspection of trace plots and density plots [22].

The network evidence diagrams were constructed, with each node representing an intervention or sham control and each line connecting two nodes denoting a direct comparison between two groups. Relative treatment effects were presented through forest plots and league tables. The rankings of the treatment effects were estimated using surface under the cumulative ranking (SUCRA) values, where higher values indicate superior efficacy [23]. And cumulative probability plots were generated to display cumulative ranking of treatment effects.

To evaluate the consistency of the network model, the node-splitting method was used, and the outcome of each comparison was expressed as a Bayes p-value. A p-value greater than 0.05 indicated non-significant inconsistency between direct and indirect evidence. Statistical heterogeneity was quantified via I² statistics (low heterogeneity: I² < 25%; moderate heterogeneity: I² of between 25% and 50%; high heterogeneity: I² > 50%) [24]. Publication bias was evaluated using the Begg’s rank-order correlation test and comparison-adjusted funnel plots. If the p-value of the Begg’s rank-order correlation test was greater than 0.05, we considered that there was no significant evidence of bias in the study. Additional sensitivity analyses were conducted through outlier removal.

Bayesian meta-regression analyses for sample size, publication year, and rating scales were used to assess possible heterogeneity of treatment effects. Subgroup analyses based on sample size and rating scales were further conducted to explore contributors to heterogeneity. For comparisons exhibiting high pairwise heterogeneity, additional pairwise meta-regressions were applied to stimulation parameters, including target region, total dose, treatment duration per session, number of sessions, and total treatment duration. Certainty of evidence was evaluated using the Confidence in Network Meta-Analysis (CINeMA) approach based on the Grading of Recommendations Assessment, Development and Evaluation (GRADE). Due to the complexities of indirect comparisons generated by numerous interventions included in this study, only the results of direct comparison based on the GRADE framework were reported.

Results

Study selection

The literature searches identified a total of 9406 records. Following duplicate removal, 4974 entries underwent screening of titles and abstracts, with 4891 records excluded. Full-text assessments of 83 studies yielded 41 eligible for network meta-analysis (Fig. 1) [15, 25–64]. However, one study [52] was subsequently excluded during data extraction due to significant non-normality of outcome data precluding standard normal-based transformation methods. Ultimately, 40 studies involving 4341 participants (mean age = 38.7 years; % female = 81.2) were included in the final analysis. Details of the study characteristics can be found in Supplementary Table 1.

Fig. 1.

PRISMA flow chart diagram

Network meta-analyses

The assessments of convergence and consistency for random-effects network models are presented in Supplementary Figs. 1–2, 10–11 and 17–18. Point estimates and upper bounds of 1 for the Gelman-Rubin diagnostic, together with no specific patterns displayed in trace plots and density plots, indicated satisfactory convergence. Node-splitting analyses indicated non-significant inconsistencies between direct and indirect comparisons for both primary and secondary outcomes, with all p-values greater than 0.05.

Primary outcome: changes in MMD

The network diagram for changes in MMD is shown in Fig. 2A. Compared to sham controls, TMS (SMD of 0.68 with 95% CrI ranging from 0.32 to 1 and I² of 11.1%), tES (SMD of 0.9 with 95% CrI ranging from 0.49 to 1.3 and I² of 0%), PMES (SMD of 0.67 with 95% CrI ranging from 0.14 to 1.2), STS (SMD of 0.91 with 95% CrI ranging from 0.36 to 1.5), REN (SMD of 0.61 with 95% CrI ranging from 0.055 to 1.1), and acupuncture (SMD of 0.34 with 95% CrI ranging from 0.17 to 0.54 and I² of 70.1%) were associated with significant reductions in migraine frequency. The forest plot is available in Supplementary Fig. 3. According to the cumulative ranking results, tES conferred the greatest reduction in MMD among all the interventions (Table 1 and Fig. 3A).

Fig. 2.

The network evidence plots of (A) changes in MMD and (B) response rate. The size of the circle represents the number of migraine patients participating in the study, with larger circles indicating a greater number of participants. The width of the line represents the number of studies, with thicker lines indicating a greater number of studies

Table 1.

League table with SUCRA values of the changes in MMD

ONS SUCRA = 54%										−0.6 (−0.19, 1.4)
−0.08 (−0.94, 0.79)	TMS SUCRA = 62%									−0.68 (−1, −0.32)
−0.3 (−1.19, 0.59)	−0.22 (−0.77, 0.32)	tES SUCRA = 82%								−0.9 (−1.3, −0.49)
0.33 (−0.53, 1.16)	0.4 (−0.09, 0.86)	0.62 (0.1, 1.13)	nVNS SUCRA = 23%							−0.28 (−0.61, 0.03)
−0.07 (−1.02, 0.88)	0.01 (−0.64, 0.64)	0.23 (−0.44, 0.91)	−0.39 (−1.01, 0.24)	PMES SUCRA = 60%	0.35 (−0.3, 0.99)					−0.8 (−1.5, −0.13)
−0.3 (−1.26, 0.66)	−0.22 (−0.88, 0.43)	0 (−0.68, 0.69)	−0.63 (−1.25, 0.01)	−0.23 (−0.75, 0.3)	STS SUCRA = 81%					−0.77 (−1.5, −0.07)
0 (−0.95, 0.95)	0.08 (−0.58, 0.72)	0.29 (−0.38, 0.98)	−0.33 (−0.94, 0.32)	0.07 (−0.7, 0.84)	0.3 (−0.48, 1.08)	REN SUCRA = 54%				−0.6 (−1.2, −0.05)
0.27 (−0.56, 1.07)	0.34 (−0.09, 0.73)	0.56 (0.1, 1.01)	−0.06 (−0.43, 0.3)	0.34 (−0.25, 0.89)	0.57 (−0.03, 1.13)	0.27 (−0.34, 0.82)	Acupuncture SUCRA = 28%			−0.34 (−0.55, −0.17)
−0.1 (−1.18, 0.99)	−0.02 (−0.85, 0.8)	0.2 (−0.65, 1.05)	−0.42 (−1.21, 0.39)	−0.03 (−0.94, 0.88)	0.2 (−0.72, 1.12)	−0.1 (−1.01, 0.82)	−0.36 (−1.11, 0.41)	CVS SUCRA = 62%		−0.7 (−1.4, 0.05)
0.12 (−0.87, 1.11)	0.19 (−0.5, 0.88)	0.41 (−0.31, 1.14)	−0.21 (−0.86, 0.47)	0.18 (−0.62, 0.98)	0.42 (−0.39, 1.23)	0.12 (−0.68, 0.92)	−0.15 (−0.75, 0.49)	0.21 (−0.74, 1.16)	KOS SUCRA = 44%	−0.49 (−1.1, 0.11)
0.6 (−0.18, 1.39)	0.68 (0.32, 1.03)	0.9 (0.49, 1.32)	0.28 (−0.02, 0.61)	0.67 (0.14, 1.21)	0.91 (0.36, 1.45)	0.61 (0.06, 1.15)	0.34 (0.17, 0.54)	0.7 (−0.04, 1.44)	0.49 (−0.11, 1.08)	Sham SUCRA = 2.1%

The results of network (lower-left portion) and pairwise (upper-right portion) meta-analyses are presented as estimate effect sizes (SMD with 95% CrIs). A SMD greater than 0 indicates greater treatment effect of the intervention in the column than that in the row. The confidence interval excluding 0 indicates statistical significance, with values in bold presented

Fig. 3.

The cumulative probability plots of (A) changes in MMD and (B) response rate

The Begg’s rank test statistic indicated significant publication bias, with the comparison-adjusted funnel plot showing substantial asymmetry (Supplementary Tables 3 and Fig. 4). Sensitivity analysis significantly reduced the risk of publication bias, with the p-value of Begg’s rank test exceeding 0.05 (Supplementary Table 3). Although the cumulative rankings of tES and STS shifted marginally after outlier removal, the SUCRA values of the two interventions (Supplementary Fig. 6) remained comparable, suggesting minor effect of publication bias on the robustness of the results. Further, since the high heterogeneity was largely attributable to the direct comparison between acupuncture and sham controls, trim-and-fill analysis was performed to identify potentially missing studies specifically for acupuncture. And the funnel plot with trim and fill (Supplementary Fig. 5) demonstrated a lack of studies with small effect sizes and sample sizes.

Primary outcome: response rate

The network plot for changes in pain intensity is shown in Fig. 2B. ONS (RR of 9 with 95% CrI ranging from 1.3 to 205), tONS (RR of 12 with 95% CrI ranging from 1.9 to 389), tES (RR of 3.1 with 95% CrI ranging from 1.4 to 7 and I² of 0%), PMES (RR of 4.6 with 95% CrI ranging from 2 to 12), STS (RR of 3.6 with 95% CrI ranging from 1.4 to 9.8, and acupuncture (RR of 1.7 with 95% CrI ranging from 1.1 to 3 and I² of 90%) significantly increased the response rate compared to sham controls. The forest plot is presented in Supplementary Fig. 12. Among all the interventions, tONS exhibited the highest response rate (Table 2 and Fig. 3B). In terms of publication bias, the comparison-adjusted funnel plot (Supplementary Fig. 13) did not reveal any significant skewness or asymmetry. Meanwhile, no evidence of bias was detected using the Begg’s rank test, with the p-value not reaching statistical significance (Supplementary Table 7).

Table 2.

League table with SUCRA values of the response rate

ONS SUCRA = 84%										0.11 (0, 0.74)
0.71 (0.01, 27.75)	tONS SUCRA = 90%									0.08 (0, 0.53)
4.66 (0.59, 117.03)	6.55 (0.81, 217.05)	TMS SUCRA = 39%								0.52 (0.22, 1.15)
2.9 (0.36, 71.11)	4.04 (0.51, 137.39)	0.62 (0.2, 1.92)	tES SUCRA = 62%							0.32 (0.14, 0.69)
5.59 (0.73, 131.72)	7.83 (1.04, 254.74)	1.21 (0.41, 3.28)	1.96 (0.67, 5.24)	nVNS SUCRA = 29%						0.63 (0.3, 1.12)
1.94 (0.23, 49.42)	2.71 (0.32, 91.56)	0.41 (0.12, 1.37)	0.67 (0.2, 2.18)	0.34 (0.12, 1.08)	PMES SUCRA = 78%	0.78 (0.36, 1.67)				0.22 (0.09, 0.51)
2.51 (0.29, 64.71)	3.52 (0.41, 118.62)	0.53 (0.15, 1.87)	0.87 (0.25, 2.95)	0.44 (0.14, 1.46)	1.29 (0.6, 2.81)	STS SUCRA = 66%				0.28 (0.1, 0.7)
5.25 (0.62, 135.25)	7.36 (0.88, 255.2)	1.12 (0.32, 3.93)	1.81 (0.53, 6.25)	0.92 (0.31, 3.11)	2.7 (0.75, 9.97)	2.09 (0.56, 8.03)	REN SUCRA = 33%			0.58 (0.22, 1.48)
5.26 (0.72, 123.44)	7.36 (1.02, 234.04)	1.14 (0.43, 2.88)	1.85 (0.71, 4.56)	0.95 (0.43, 2.12)	2.75 (0.98, 7.45)	2.14 (0.71, 6.07)	1.03 (0.33, 2.79)	Acupuncture SUCRA = 31%		0.6 (0.34, 0.91)
5.14 (0.53, 136.39)	7.19 (0.75, 253.08)	1.09 (0.26, 4.24)	1.76 (0.43, 6.81)	0.9 (0.25, 3.35)	2.63 (0.61, 10.89)	2.03 (0.45, 8.81)	0.97 (0.22, 4.08)	0.95 (0.28, 3.29)	CVS SUCRA = 35%	0.57 (0.17, 1.66)
8.88 (1.34, 204.54)	12.43 (1.9, 388.65)	1.92 (0.87, 4.47)	3.1 (1.45, 7.02)	1.58 (0.89, 3.29)	4.63 (1.97, 11.67)	3.58 (1.43, 9.66)	1.72 (0.68, 4.45)	1.67 (1.1, 2.93)	1.76 (0.6, 5.76)	Sham SUCRA = 4%

The results of network (lower-left portion) and pairwise (upper-right portion) meta-analyses are presented as estimate effect sizes (RR with 95% CrIs). A RR greater than 1 indicates greater treatment effect of the intervention in the column than that in the row. The confidence interval excluding 1 indicates statistical significance, with values in bold presented

Secondary outcome: changes in pain intensity

The main result of the network meta-analysis revealed that TMS (SMD of 0.92 with 95% CrI ranging from 0.55 to 1.3 and I² of 84.7%), tES (SMD of 0.71 with 95% CrI ranging from 0.18 to 1.2 and I² of 33.8%), PMES (SMD of 2.4 with 95% CrI ranging from 1.4 to 3.4), STS (SMD of 2.1 with 95% CrI ranging from 0.75 to 3.4), and acupuncture (SMD of 0.47 with 95% CrI ranging from 0.21 to 0.76 and I² of 65%) significantly decreased the pain severity of migraine patients compared to the sham group (Supplementary Fig. 19). Among all the interventions, PMES yielded the greatest pain intensity reduction (Supplementary Tables 9 and Fig. 21).

Meta-regression and subgroup analyses

High heterogeneity was observed in several interventions for both primary and secondary outcomes. Through sensitivity analyses, the presence of extreme data points from certain studies was considered as the primary contributor to the high heterogeneity, and outlier removal substantially reduced the heterogeneity. However, for changes in MMD and pain intensity, pairwise comparisons between acupuncture and sham controls still exhibited moderate heterogeneity (Supplementary Figs. 7 and 22). Thus, network regression analyses based on sample size, publication year, and rating scales (MMD or MHD for changes in MMD and VAS or NRS for changes in pain intensity) were conducted to further identify the source of heterogeneity. The 95% credible intervals of beta coefficients for all regressors included 0, suggesting no significant effects on the results (Supplementary Tables 5 and 12).

Additionally, subgroup analyses were conducted specifically for sample size and rating scales (Supplementary Figs. 8–9 and 23–24). For both outcomes, studies of acupuncture with small sample size showed a tendency towards greater effect size compared to that with large sample size, suggesting potential overestimation of treatment effects in studies with small sample size. Besides, the effect sizes of acupuncture were significant when measured by MMD and VAS rather than MHD and NRS. Similar results were also found in the direct comparisons between TMS and sham controls in measuring the migraine pain intensity. MMD and VAS exhibited superior sensitivity and specificity to the measurements of migraine frequency and severity compared with MHD and NRS.

Further, pairwise meta-regression analyses were conducted based on stimulation parameters (e.g., target region, number of sessions, and total dose) specifically for the acupuncture versus sham comparisons (Supplementary Tables 6 and 13). For changes in MMD, treatment duration per session was identified as the only modulator, explaining 3.27% of between-study heterogeneity. For changes in pain intensity, total dose emerged as the primary modulator, accounting for 32.6% of the heterogeneity. Taken together, the results of meta-regression and subgroup analyses suggested that variations in study design may contribute to heterogeneity.

RoB and certainty of evidence

Regarding RoB, 60% (24/40) of the included studies were deemed as low risk of bias, and 32.5% (13/40) and 7.5% (3/40) of studies were classified as having some concerns and high risk of bias respectively. The assessments of risk of bias for individual studies are detailed in Supplementary Table 2.

For changes in MMD, the certainty of evidence ranged from moderate to very low. The effects of tES on the reduction of migraine frequency were rated moderate certainty, indicating a reasonable level of confidence. For response rate, the quality of evidence ranged from high to very low. The comparison between tONS and sham controls was rated high certainty, indicating a high degree of confidence in the beneficial therapeutic impact of tONS. Given only one study included in assessing the intervention effect of tONS, caution should be exercised when applying this conclusion in routine clinical practice. Full information about CINeMA assessment for both outcomes are provided in Supplementary Tables 4, 8, and 11.

Discussion

The present study showed that TMS, tES, PMES, STS, and acupuncture significantly outperformed sham controls in improving migraine frequency, response rate and pain intensity after intervention, with varying strengths of evidence. In addition, both invasive and non-invasive ONS were associated with significantly higher response rate compared to sham controls. The benefits of most interventions were supported by moderate certainty of evidence, whereas evidence for acupuncture in decreasing migraine frequency and pain intensity. Among all the investigated interventions, tES conferred the greatest reduction in monthly migraine days, tONS produced the strongest effect on response rate improvement, and PMES yielded the most decreased pain intensity after intervention. In particular, effect sizes for changes in MMD and pain intensity were presented as SMDs, which need to be considered with other clinical parameters for the interpretation of intervention effects in clinical practice [65].

The findings of this study are in line with previous meta-analyses establishing therapeutic benefits of transcranial magnetic and electrical stimulation, and peripheral nerve stimulation [66–68]. Beyond conventional pairwise meta-analyses, this network meta-analysis enabled the integration of both direct and indirect comparisons across active interventions. Further, unlike previous network meta-analyses restricted to non-invasive stimulation techniques, the present study encompassed both invasive and non-invasive neuromodulation interventions for migraine prophylaxis.

There were some controversies over the classification of acupuncture as invasive neuromodulation in previous meta-analyses [14, 18]. Acupuncture requires the insertion of needles with electrical or manual stimulation during the entire treatment process [31, 54, 69, 70]. Thus, acupuncture is defined as invasive peripheral neurostimulation in this study. Despite significant treatment effects for both primary and secondary outcomes, the effect sizes of acupuncture were relatively small compared to other active interventions, and the cumulative ranking of acupuncture was the lowest for each outcome. Meanwhile, the pairwise comparisons between acupuncture and sham stimulation, which were rated as very low for the certainty of evidence for both outcomes, were considered as the secondary contributor to the high heterogeneity. Similarly, although invasive ONS was associated with significantly better response rate relative to sham controls, its cumulative ranking was lower than that of the non-invasive tONS. Combined with surgical risks and potential adverse events inherent to the invasive intervention, current evidence does not support invasive neuromodulation as preferred treatment option for migraine prophylaxis [48, 71].

One of the key findings of the current study was that tES yielded the greatest reduction in MMD after intervention. Though tES and STS exhibited comparable SUCRA values, the league table showed that tES outperformed some interventions. Given limited STS studies for analysis, tES emerged as preferred preventive treatment in reducing migraine frequency. Transcranial direct current stimulation (tDCS) and transcranial alternating current stimulation (tACS), which are common types of tES, have been increasingly applied in the prevention of migraine attacks. tDCS can modify cortical spreading depression (CSD), a key mechanism underlying migraine pathophysiology [67]. The activation of CSD may contribute to the progression of central sensitization, which in turn exacerbates the perception of pain and leads to chronification of migraine headaches [72]. Active stimulation of frontal, temporal, and occipital regions with tACS can not only modulate local cortical excitability but also modify the functional connection between cortical regions and downstream second-order nociceptors [52, 73]. The effects of transcranial electrical stimulation are cumulative and multiple stimulation sessions are beneficial to the general reduction of migraine attack frequency [74].

Another major finding of this network meta-analysis was that tONS was the most efficacious intervention for increasing the response rate. The league table demonstrated significantly greater treatment effect of tONS compared to several other interventions. The therapeutic mechanism of occipital nerve stimulation is thought to involve modulation of the trigeminovascular system—a key pathway implicated in migraine pathogenesis [17]. Anatomically, the occipital nerve is a branch of the C2 spinal nerve, which converges with trigeminal afferent inputs within the trigeminocervical complex (TCC) [27]. Nociceptive signals are relayed from the TCC to thalamocortical regions, where they are processed, ultimately resulting in the perception of migraine pain. By attenuating nociceptive transmission at the level of the TCC, stimulation of the occipital nerve may impede the development of central sensitization and dysregulation of descending pain control pathways [75]. Consequently, inhibition of trigeminovascular activity suppresses pain perception and markedly decreases the incidence of migraine attacks [76].

Limitations

There are some limitations that warrant consideration in this study. First, for some interventions, a limited number of RCTs were included in analysis, necessitating cautious interpretation of the results. Second, most of the included RCTs compared neuromodulation interventions against sham stimulation groups, limiting the power of evidence. Thus, head-to-head comparisons of active interventions are required for more conclusive evidence. Third, high heterogeneity was found in both primary and secondary outcomes. Sensitivity analyses allowed for identification of the leading cause to the high heterogeneity and reduction of heterogeneity to moderate. While network regression analyses suggested no potential contributors to the high heterogeneity, subsequent subgroup analyses demonstrated relative difference of treatment effects between studies using different outcome measurement scales. Further, pairwise meta-regression analyses for the comparison between acupuncture and sham controls identified treatment duration per session and total dose as key modulators associated with between-study variance. Therefore, standardization of stimulation parameters and assessment methodologies in future research may be able to improve the reliability and validity of meta-analysis. Fourth, significant publication bias was observed in the changes in MMD, contributing partially to the high heterogeneity. The funnel plot with trim and fill specific for the comparison between acupuncture and sham stimulation, which was considered as the secondary contributor to the high heterogeneity, indicated a lack of studies with small sample sizes and treatment effects. And this was further supported by subgroup analyses based on sample size. Despite non-significant publication bias after sensitivity analyses, caution is still needed when interpreting the findings.

Conclusions

This study provides a summary of existing evidence from previous studies for the guidance of the choice of preventive neuromodulation interventions for migraine. Our main findings indicated that tES and tONS are associated with greater improvements in MMD and response rate compared to other interventions, respectively. Additionally, PMES, which also demonstrated efficacy in improving migraine frequency and treatment response, emerged as the most effective intervention for pain intensity reduction. Because of the limited studies included in the assessments for certain interventions, as well as network model constraints such as heterogeneity and publication bias, caution is required for the interpretation of results. Future large-scale and high-quality RCTs are necessary to provide more direct evidence to confirm the efficacy of neuromodulation interventions.

Abbreviations

95% CrI

95% credible interval

CINeMA

Confidence in Network Meta-Analysis

CSD

Cortical spreading depression

CVS

Caloric vestibular stimulation

GRADE

Grading of Recommendations Assessment, Development and Evaluation

KOS

Kinetic oscillation stimulation

MHD

Monthly headache days

MMD

Monthly migraine days

NRS

Numerical rating scale

nVNS

Non-invasive vagus nerve stimulation

ONS

Occipital nerve stimulation

PID

Peri-infarction depolarization

PMES

Percutaneous mastoid electrical stimulation

RCT

Randomized controlled trial

REN

Remote electrical neuromodulation

RoB

Risk of bias

SMD

Standardized mean difference

STS

Supraorbital transcutaneous stimulation

SUCRA

Surface under the cumulative ranking curve

tACS

Transcranial alternating current stimulation

tDCS

Transcranial direct current stimulation

tES

Transcutaneous electrical stimulation

tONS

Transcranial occipital nerve stimulation

TMS

Transcranial magnetic stimulation

VAS

Visual analogue scale

Authors’ contributions

DK performed concept formation, database searches, data extraction, statistical analysis and manuscript drafting. MC performed database searches, data extraction and visualization. XD, YL, and HC performed literature review, data curation and statistical analysis. XY and ZW made great contribution to concept formation, study design and manuscript revision. All authors read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (Grant No. 82575226), received by ZW.

Data availability

The data of this study would be available upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.