Comparative efficacy of different repetitive transcranial magnetic stimulation protocols for lower extremity motor function in stroke patients: a network meta-analysis

Background Lower extremity motor dysfunction is one of the most severe consequences after stroke, restricting functional mobility and impairing daily activities. Growing evidence suggests that repetitive transcranial magnetic stimulation (rTMS) can improve stroke patients’ lower extremity motor function. However, there is still controversy about the optimal rTMS protocol. Therefore, we compared and analyzed the effects of different rTMS protocols on lower extremity motor function in stroke patients using network meta-analysis (NMA). Methods We systematically searched CNKI, WanFang, VIP, CBM, PubMed, Embase, Web of Science, and Cochrane Library databases (from origin to 31 December 2023). Randomized controlled trials (RCTs) or crossover RCTs on rTMS improving lower extremity motor function in stroke patients were included. Two authors independently completed article screening, data extraction, and quality assessment. RevMan (version 5.4) and Stata (version 17.0) were used to analyze the data. Results A total of 38 studies with 2,022 patients were eligible for the NMA. The interventions included HFrTMS-M1, LFrTMS-M1, iTBS-Cerebellum, iTBS-M1, dTMS-M1, and Placebo. The results of NMA showed that LFrTMS-M1 ranked first in FMA-LE and speed, and HFrTMS-M1 ranked first in BBS, TUGT, and MEP amplitude. The subgroup analysis of FMA-LE showed that HFrTMS-M1 was the best stimulation protocol for post-stroke time > 1 month, and LFrTMS-M1 was the best stimulation protocol for post-stroke time ≤ 1 month. Conclusion Considering the impact of the stroke phase on the lower extremity motor function, the current research evidence shows that HFrTMS-M1 may be the preferred stimulation protocol to improve the lower extremity motor function of patients for post-stroke time > 1 month, and LFrTMS-M1 for post-stroke time ≤ 1 month. However, the above conclusion needs further analysis and validation by more high-quality RCTs. Systematic Review Registration:www.crd.york.ac.uk/prospero/, identifier (CRD42023474215).


Introduction
Stroke is a local brain dysfunction caused by acute cerebrovascular disease (Park et al., 2011).By 2019, the proportion of stroke patients had increased to 2.58% among residents aged ≥40 years in China (Wang et al., 2022).Stroke is the leading cause of adult death and disability in China, with the five characteristics of high incidence, high disability rate, high mortality rate, high recurrence rate, and high economic burden, which has seriously endangered the health of Chinese people.With the development of medical technology, the mortality rate of stroke has decreased year by year, but 72% of the survivors still have lower extremity dysfunction (Ng and Hui-Chan, 2010), which affects the walking function of patients.Walking dysfunction is one of the most severe consequences of stroke.Nearly 30% of stroke patients are unable to walk even in the chronic recovery stage (Park et al., 2011), which significantly affects the patients' social interactions and can lead to lifelong disability in severe cases (Park and An, 2016).Therefore, improving the motor function of the lower extremity, restoring independent walking as soon as possible, and improving activities of daily living (ADL) are the problems that many stroke patients are eager to solve urgently.However, pharmacological therapy (Mead et al., 2013) and traditional rehabilitation therapies [e.g., neurodevelopmental therapy (Langhammer and Stanghelle, 2011), proprioceptive neuromuscular facilitation (Eng and Tang, 2007), and electromyography biofeedback (Woodford and Price, 2007)] seem to have limited effects on improving motor function of the lower extremity after stroke.Therefore, a more effective treatment is needed.
Repetitive transcranial magnetic stimulation (rTMS), one of the brain stimulation techniques without any trauma, can induce neuroplastic changes and promote brain function restoration (Nathou et al., 2018).At present, the interhemispheric competition (IHC) model is the primary theoretical basis for applying rTMS in stroke rehabilitation.This model suggests that stroke destroys the balance of mutual inhibition of the bilateral cerebral hemispheres through the corpus callosum, resulting in the decreased inhibition of the ipsilateral hemisphere to the contralateral hemisphere and the increased inhibition of the contralateral hemisphere to the ipsilateral hemisphere (Di Pino et al., 2014).Therefore, in clinical practice, there are two main ways to use rTMS to promote functional recovery after stroke.One is to reduce the excitability of the contralateral hemisphere through low-frequency (≤1 Hz) rTMS to reduce the inhibitory effect of the contralateral hemisphere on the ipsilateral hemisphere.The other is to restore the balance of competitive inhibition between the bilateral cerebral hemispheres by stimulating the ipsilateral hemisphere with high-frequency (≥5 Hz) rTMS to increase its excitability (George and Aston-Jones, 2010).Both stimulation modes have been used to treat motor/non-motor dysfunction after stroke.
Theta burst stimulation (TBS), a novel mode of rTMS, saves time in the rehabilitation of motor function after stroke (Huang et al., 2005).There are two types of TBS: intermittent TBS (iTBS) and continuous TBS (cTBS), which generate excitatory and inhibitory effects, respectively (Larson et al., 1986;Huang et al., 2011).Compared with conventional rTMS protocols, TBS provides significant advantages due to its reduced stimulation time (Chung et al., 2015) and long-lasting effects with lower-intensity stimulation (Cárdenas-Morales et al., 2010).Deep transcranial magnetic stimulation (dTMS) is a new non-invasive neuromodulation technique based on rTMS technology, which uses a different coil type (Hesed coil).dTMS has the advantages of deeper and wider stimulation, more precise localization, and less damage to the superficial cortex than conventional TMS (Roth et al., 2014).The primary motor cortex (M1) is typically the target of rTMS.However, some studies have shown that the cerebellum is one of the alternative targets of M1, and rTMS targeting the cerebellum can also improve motor function in stroke patients (Wessel and Hummel, 2018).In conclusion, rTMS can regulate the asymmetry of excitability between the bilateral cerebral hemispheres by changing the stimulation mode, stimulation target, stimulation frequency, and coil type to promote the recovery of lower extremity motor dysfunction after stroke (Kesikburun, 2022).
The effect of rTMS on lower extremity motor function in stroke patients has been demonstrated in previous meta-analysis (Li et al., 2018;Tung et al., 2019).However, an important drawback is that conventional meta-analysis can only compare two interventions simultaneously.At the same time, in these studies, the intervention protocols of the experimental group were roughly classified, and the effects of different stimulation frequencies, stimulation targets, stimulation modes, and post-stroke times on treatment effects were not comprehensively considered.Although Fan et al. (2021) conducted a detailed systematic review of rTMS to improve lower extremity motor function in stroke patients, this study did not consider the new stimulation mode-iTBS, the new stimulation target-cerebellum, and the effect of stroke phase on the efficacy of rTMS.
Network meta-analysis (NMA) is developed from conventional meta-analysis, and its primary function is to comprehensively evaluate and rank multiple interventions simultaneously (Rouse et al., 2017).Therefore, we took the stimulation mode, stimulation frequency, and stimulation target of rTMS into account and summarized the following five different rTMS protocols: high-frequency rTMS-M1 (HFrTMS-M1), low-frequency rTMS-M1 (LFrTMS-M1), iTBS-Cerebellum, iTBS-M1, and dTMS-M1.Then, we compared and analyzed the effects of different protocols on lower extremity motor dysfunction in stroke patients by NMA.In addition, considering the effects of the stroke phase on the efficacy of rTMS at different protocols, we also carried out a subgroup analysis for FMA-LE according to the phase of the stroke to provide sufficient evidence for future clinical practice.

Study enrollment and reporting
Our NMA was conducted using the Cochrane Handbook for Systematic Reviews of Interventions, and the findings were reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyzes (PRISMA) statement (Page et al., 2021).This NMA was prospectively registered in PROSPERO (registration ID: CRD42023474215).

Search strategy
Two authors separately searched for randomized controlled trials (RCTs) and crossover RCTs about rTMS improving lower extremity motor function in stroke patients from China National Knowledge Infrastructure (CNKI), Wanfang database, VIP database, China Biomedical Literature Database (CBM), PubMed, Embase, Web of Science, and Cochrane Library.The search time limit was from the establishment of the database to 31 December 2023.By combining medical subject headings (MeSH) with free words using Boolean logic operators, we integrated the following terms for a comprehensive search: "Stroke, " "cerebrovascular accident, " "CVA, " "Brain Vascular Accident, " "hemiplegia, " "apoplexy, " "hemiparesis, " "repetitive transcranial magnetic stimulation, " "Transcranial Magnetic Stimulation, " "TMS, " "rTMS, " "Theta burst stimulation, " "θ-brust stimulation, " "random, " "randomized controlled trial, " "RCT." In addition, we also reviewed meta-analysis, reviews, and references of the included studies to supplement the search.PubMed was used as an example, and the specific search strategy was provided in Supplementary Table 1.

Eligibility criteria
Eligibility criteria were defined in accordance with the PICOS framework (Hutton et al., 2015).

Study selection
First, two authors (CSW and QZ) used EndnoteX9 software to eliminate duplicate articles.Then, they screened out articles that did not meet the criteria by reading their titles and abstracts.Finally, they browsed the full text to select articles that met the criteria.In case of any disagreement during the review process, the decision was made by consultation between the two authors or by joint decision with the third author (LLZ).

Data extraction
Two authors (DYZ and YNX) independently reviewed all articles and extracted data.The extracted data included basic published information: first author's name, year of publication, country of origin, participant characteristics (age and sample size), intervention characteristics (intervention protocol, coil type, rTMS target, rTMS frequency, rTMS intensity, No. of pulses, and duration of intervention), and outcome indicators (FMA-LE, BBS, TUGT, MEP amplitude, and speed) at baseline and at last observation to obtain their change scores.The collected data were put into an Excel spreadsheet and cross-checked by two authors (DYZ and YNX).In case of disagreement during data extraction, the third author (ZJL) participated in discussion and decision-making.

Quality assessment
Two authors (CLW and SZW) independently assessed the risk of bias for the included articles through the Cochrane Risk of Bias Tool (Savović et al., 2014) (VII) Other bias.Assessment indicators were rated "low risk, " "unclear, " or "high risk" based on the available information.If there was any dispute during the evaluation, the third author (ZJL) would participate in the discussion and make decisions together.

Statistical analysis
Odds ratio (OR) for binary variables and mean difference (MD) for continuous variables were used as the effect indicators, and the Frontiers in Neuroscience 04 frontiersin.org95% confidence interval (CI) was provided for each effect size.If a particular study used different methods or scales to measure the same outcome, standardized MD (SMD) was calculated instead of MD.We calculated the difference before and after treatment for continuous variable indicators and the standard deviation (SD) according to the method provided in 16.1.3.2 of Cochrane Handbook 5.0.2 and then performed the statistical analysis.We used RevMan (version 5.4) for pairwise meta-analysis.The p-value of the chi-square test and the I 2 index from the heterogeneity test were used to express the level of statistical heterogeneity.Different effect models were selected according to the test data's heterogeneity level.When the level of heterogeneity was low (p ≥ 0.1, I 2 ≤ 50%), we selected the fixed effect model for analysis.Otherwise, a random effect model (p < 0.1, I 2 > 50%) was used (Higgins et al., 2003;Tufanaru et al., 2015).We used Stata (version 17.0) to perform the NMA and produce various charts, such as network meta-analysis diagrams of eligible comparisons, surface under the cumulative ranking area (SUCRA), funnel plot of publication bias, etc (Shim et al., 2017).When there were closed loops between interventions, we first needed to assess global inconsistency.If p > 0.05, the inconsistent model was not significant, and the consistent model was selected (White et al., 2012).We used a node-splitting approach to assess local inconsistency (Dias et al., 2010).At the same time, it was also necessary to evaluate the loop inconsistency and calculate the inconsistency factors (IF) and 95% CI for each closed loop.If the lower limit of 95% CI included or was close to 0, the consistency between the direct and indirect comparison results was good; otherwise, the closed loop was considered to have apparent inconsistency.If no closed loops existed between interventions, the consistency model was used directly for analysis.We used the SUCRA to rank interventions.The closer SUCRA was to 100%, the better the effect of the intervention.Finally, the publication bias of the included articles was evaluated using the funnel plot of publication bias and Egger's test.Asymmetry in the funnel plot of publication bias and p < 0.05 in Egger's test indicated publication bias in the included articles (Fleiss, 1993).

Search results
We strictly searched the above 8 databases according to the inclusion and exclusion criteria and preliminarily obtained 12,814 articles.After the duplicate articles were removed, 10,166 articles remained in the database.By reading the titles and abstracts of the articles, we excluded articles that did not meet the inclusion criteria, leaving 176 remaining in the database.By reading the full text, we again excluded 138 articles.Ultimately, 38 articles met our study requirements.Figure 1 shows the article search process and results.

Quality evaluation
25 studies (65.8%) had a low risk of bias concerning random sequence generation.9 studies (23.7%) had a low risk of bias concerning allocation concealment.24 studies (63.2%) had a low risk of bias concerning blinding of participants and personnel.27 studies (71.1%) had a low risk of bias concerning blinding of outcome assessment.37 studies (97.4%) had a low risk of bias concerning incomplete outcome data.38 studies (100.0%) had a low risk of bias concerning selective reporting.Other biases were not known.Details of the evaluation of bias results for the included articles are shown in Figure 2.

Pairwise meta-analysis
A pairwise meta-analysis was used to compare the two interventions comprehensively.We carried out 6 pairwise metaanalysis to compare FMA-LE, 6 to compare FMA-LE (post-stroke time > 1 month), 3 to compare FMA-LE (post-stroke time ≤ 1 month), 5 to compare BBS, 4 to compare TUGT, 4 to compare MEP amplitude, and 2 to compare speed, respectively, which can be summarily seen in Table 2.The detailed results of pairwise meta-analysis are shown in Supplementary Figures 1-7.

FMA-LE
A total of 29 included studies evaluated FMA-LE, involving 6 intervention protocols: HFrTMS-M1, LFrTMS-M1, iTBS-Cerebellum, iTBS-M1, dTMS-M1, and Placebo.A total of 1,685 patients were included.The inconsistency model evaluated global inconsistency, which showed p = 0.0772 (>0.05;Supplementary Figure 8A).The inconsistency test was not significant, so we used the consistency model.The node-splitting approach was used to evaluate local inconsistency.The test of local inconsistency from the node-splitting model showed a small percentage of inconsistency (1 of 6 comparisons), as detailed in Supplementary closed loop.The results showed that the 95% CI included 0, and IF was close to 0, indicating that our NMA was highly credible (Supplementary Figure 9A).
The NMA results showed that FMA-LE generated a total of 15 pairwise comparisons.Compared with Placebo, LFrTMS-M1 (MD = 2.83, 95% CI: 1.96 to 3.70) and HFrTMS-M1 (MD = 2.74, 95% CI: 1.60 to 3.87) significantly improved FMA-LE in stroke patients.There was no statistically significant difference between the other two interventions (p > 0.05; Figure 4).Figure 5A and Table 3 show the SUCRA rankings for all interventions.According to the analysis, LFrTMS-M1 (SUCRA, 84.6%) may be the most effective intervention to improve FMA-LE in stroke patients.
Different phases of stroke may lead to different therapeutic effects of rTMS.Therefore, we performed a subgroup analysis for FMA-LE according to the phase of stroke, including post-stroke time > 1 month and post-stroke time ≤ 1 month (Brunelli et al., 2019).The global consistency model of FMA-LE (post-stroke time > 1 month) and FMA-LE (post-stroke time ≤ 1 month) showed that the inconsistency test was not significant (Supplementary Figures 8B,C).There was a small percentage of inconsistency (1 of 3 comparisons) only for the local inconsistency test of FMA-LE (post-stroke time ≤ 1 month), details of which are provided in Supplementary Tables 3, 4.

BBS
A total of 11 included studies evaluated BBS, involving 5 intervention protocols: HFrTMS-M1, LFrTMS-M1, iTBS-Cerebellum, iTBS-M1, and Placebo.A total of 820 patients were included.The inconsistency model evaluated global inconsistency, which showed p = 0.2800 (>0.05;Supplementary Figure 8D).The inconsistency test was not significant, so we used the consistency model.The node-splitting approach was used to evaluate local inconsistency.The measured p-values were all greater than 0.05, indicating good local consistency (Supplementary Table 5). 1 closed loop was formed for the 3 interventions, so we assessed the inconsistency of the closed loop.The results showed that the 95% CI included 0, and IF was close to 0, indicating that our NMA was highly credible (Supplementary Figure 9D).

TUGT
A total of 8 included studies evaluated TUGT, involving 5 intervention protocols: HFrTMS-M1, LFrTMS-M1, iTBS-Cerebellum, iTBS-M1, and Placebo.A total of 382 patients were included.There was no closed loop, so we did not need to perform a consistency check.The NMA results showed that TUGT generated a total of 10 pairwise comparisons.

MEP amplitude
A total of 8 included studies evaluated MEP amplitude, involving 4 intervention protocols: HFrTMS-M1, LFrTMS-M1, iTBS-Cerebellum, and Placebo.A total of 246 patients were included.The inconsistency model evaluated global inconsistency, which showed p = 0.5656 (>0.05;Supplementary Figure 8E).The inconsistency test was not significant, so we used the consistency model.The nodesplitting approach was used to evaluate local inconsistency.The measured p-values were all greater than 0.05, indicating good local consistency (Supplementary Table 3). 1 closed loop was formed for the 3 interventions, so we assessed the inconsistency of the closed loop.The results showed that the 95% CI included 0, and IF was close to 0, indicating that our NMA was highly credible (Supplementary Figure 9E).
The NMA results showed that MEP amplitude generated a total of 6 pairwise comparisons.Compared with Placebo, HFrTMS-M1 (SMD = 0.99, 95% CI: 0.17 to 1.82) significantly improved MEP amplitude in stroke patients.There was no statistically significant    4). Figure 5F and Table 3 show the SUCRA rankings for all interventions.
According to the analysis, HFrTMS-M1 (SUCRA, 93.7%) may be the most effective intervention to improve MEP amplitude in stroke patients.

Speed
A total of 13 included studies evaluated speed, involving 3 intervention protocols: HFrTMS-M1, LFrTMS-M1, and Placebo.A total of 667 patients were included.There was no closed loop, so Red and bold numbers are statistically significant.HFrTMS, high-frequency repetitive transcranial magnetic stimulation; LFrTMS, low-frequency repetitive transcranial magnetic stimulation; iTBS, intermittent theta-burst stimulation; dTMS, deep transcranial magnetic stimulation; M1, primary motor cortex; NR, not reported.(Supplementary Table 7).The adverse events reported were mild, such as headache, dizziness, nausea, and vomiting.

Discussion
Stroke is a common disease worldwide and causes severe disabilities for patients.More than two-thirds of stroke survivors have post-stroke sequelae, including impairment in motor function, balance, gait, and ADL (Paul et al., 2007).Improving lower extremity motor function and balance ability can significantly impact gait function, ADL, and quality of life in stroke patients (Smith et al., 2017).Although the use of rTMS for stroke has attracted considerable attention, there is still a lack of consensus on the optimal protocol for rTMS to improve lower extremity motor function in stroke patients.To the best of our knowledge, this is the first NMA to compare the effects of different rTMS protocols on lower extremity motor function in stroke patients by taking stimulation frequency, stimulation target, and stimulation mode of rTMS and post-stroke time into account simultaneously.
The FMA-LE can predict lower extremity motor recovery in individuals with stroke (Balasubramanian et al., 2016).This assessment exhibits good internal consistency and reliability, discriminative validity, and responsiveness to interventions Network meta-analysis of head-to-head comparisons.Red and bold numbers are statistically significant.HFrTMS, high-frequency repetitive transcranial magnetic stimulation; LFrTMS, low-frequency repetitive transcranial magnetic stimulation; iTBS, intermittent theta-burst stimulation; dTMS, deep transcranial magnetic stimulation; M1, primary motor cortex. 10.3389/fnins.2024.1352212 Frontiers in Neuroscience 13 frontiersin.org(Hsieh et al., 2009).This study found that compared with Placebo, LFrTMS-M1 and HFrTMS-M1 significantly improved FMA-LE in stroke patients, and LFrTMS in the contralateral hemisphere was more effective than HFrTMS in the ipsilateral hemisphere.Xie et al. (2021a) suggested that LFrTMS in the contralateral hemisphere was more effective than HFrTMS in the ipsilateral hemisphere, which was consistent with our findings.However, a novel finding from this NMA is that the subgroup analyses of FMA-LE showed that at ≤1 month after stroke, HFrTMS-M1 was the optimal stimulation protocol for improving stroke patients' lower extremity motor function.At HFrTMS, high-frequency repetitive transcranial magnetic stimulation; LFrTMS, low-frequency repetitive transcranial magnetic stimulation; iTBS, intermittent theta-burst stimulation; dTMS, deep transcranial magnetic stimulation; M1, primary motor cortex.≤1 month after stroke, LFrTMS-M1 was the best protocol, which was superior to HFrTMS-M1.The difference between these two studies was that when grouping rTMS protocols, we considered that the same rTMS mode applied to different stimulation targets would produce different therapeutic effects.Therefore, we divided stimulation targets into M1 and Cerebellum rather than simply grouping very different stimulation targets together, which allowed the intervention protocol to be more refined.Second, we included more articles that met our research objectives to expand the sample size.Finally, we also carried out a subgroup analysis of the patients on the efficacy of rTMS at different protocols to improve the accuracy of the outcome evidence.
In addition, we also found that 10 Hz and 1 Hz were the most commonly used stimulation frequencies for HFrTM and LFrTMS, respectively, regarding rTMS prescription settings, which was the same conclusion as Fan et al. (2021).Meanwhile, rTMS sessions of 15 or 20 min each and lasting 3 or 4 weeks were the most common.In clinical practice, clinicians or rehabilitation therapists can flexibly formulate the best stimulation prescription according to the specific situation of patients and the recommended protocols mentioned above.
In fact, early hyperexcitability and increased interhemispheric inhibition of the contralesional motor cortex have been demonstrated using TMS after unilateral stroke.Therefore, LFrTMS can effectively improve the motor function of stroke patients by reducing the excitability of the motor cortex of the contralateral hemisphere to restore the balance of competitive inhibition between the two hemispheres in the acute phase of stroke.However, in the post-stroke convalescent phase, the interhemispheric competition is less pronounced than in the acute phase, as it is commonly observed that the transcallosal asymmetry decreases with time (Swayne et al., 2008).LFrTMS may reduce the compensatory effect of the contralateral hemisphere by inhibiting its excitability, thereby hindering functional recovery after stroke.Therefore, HFrTMS-M1 may be more effective than LFrTMS-M1 in the convalescent phase of stroke.Xia et al. (2022) also recommended the application of HFrTMS in patients with stroke patients during the convalescent phase.
The BBS is the most widely used clinical scale for assessing balance performance in individuals with neurological conditions, including static and dynamic balance (Neuls et al., 2011).The sum of the scores for the 14 items (each item was rated from 0 to 4) yielded a balance score ranging from 0 to 56 (Neuls et al., 2011).TUGT is a rapid and quantitative assessment of dynamic balance and functional walking ability and is closely related to other measures of gait and balance in stroke patients (Flansbjer et al., 2005).TUGT assesses the time taken to complete a series of actions, including standing up from a chair, walking forward three meters, turning, and returning to the chair.According to the ranking probability of our NMA, HFrTMS-M1 was more advantageous in improving BBS and TUGT.Therefore, we recommend HFrTMS-M1 as a complementary rehabilitation therapy to improve balance function in stroke patients in clinical practice.
Walking speed can reflect the recovery of lower extremity function and walking quality in stroke patients (Patterson et al., 2008).Our NMA results showed that compared with Placebo, LFrTMS-M1 and HFrTMS-M1 significantly improved the speed in stroke patients.LFrTMS-M1 had more advantages in improving the speed of stroke patients.However, Tung et al. (2019) found that HFrTMS was superior to LFrTMS in improving speed in stroke patients.Further understanding of the relationship between different rTMS protocols and walking speed is needed in the future.Transcranial magnetic stimulation (TMS) is a non-invasive neuromodulation technique that produces pulsed magnetic fields that form induced currents in the motor cortex of the brain.After the induced current stimulates one side of the motor cortex, the conduction nerve impulses are transmitted downward, which will cause the target muscle on the opposite side of the subject to produce action potentials, called motor-evoked potentials (MEPs).MEPs is a quantitative evaluation index of central motor conduction function, which can objectively reflect the excitability of the motor cortex.In this study, we used MEP amplitude to assess the functional status of motor conduction pathways, and the results showed that HFrTMS-M1 was the most effective in improving MEP amplitude.However, MEP amplitude included only a few studies, meaning this ranking result should be treated critically.In addition, to ensure the objectivity of the study results, more high-quality RCTs with large sample sizes are needed for further verification.
Like TMS, functional magnetic resonance imaging (fMRI), electroencephalography (EEG), and other neuroimaging techniques are also of great significance in the assessment of motor function in stroke.fMRI can evaluate the neurovascular response induced by rTMS, and the activity changes of brain nerves can be observed through intuitive and visual images (Bergmann et al., 2016).In addition, fMRI can also be used to study the excitability and functional connectivity of the cerebral cortex and subcortex in stroke patients under rTMS intervention.Guo et al. (2021) used fMRI to find that both LFrTMS stimulation in the unaffected hemisphere and HFrTMS stimulation in the affected hemisphere could promote the reorganization of the motor network, and the changes in functional connectivity between the contralateral PMA and the ipsilateral M1, and between the bilateral M1 induced by rTMS were related to motor recovery.Transcranial magnetic stimulation combined with electroencephalography (TMS-EEG) is a new evaluation method in recent years.TMS-EEG can reflect the direct relationship between brain regions and motor function in stroke patients and predict the recovery of motor function after stroke.Koch et al. (2019) performed motor assessment on 34 stroke subjects who received iTBS-Cerebellum or sham iTBS treatment by Fugl-Meyer scale, gait analysis, and so on, and recorded cerebral cortical activity through TMS-EEG to achieve the combination of treatment and assessment feedback.At present, there are relatively few studies using fMRI and EEG to explore the treatment of rTMS to promote the recovery of dysfunction after stroke, and most of the trials are small in sample size and scale.In the future, large-sample, multi-center, and high-quality RCTs should be carried out, and the results of imaging and electrophysiology of stroke patients should be comprehensively analyzed to improve the accuracy and scientificity of rehabilitation efficacy evaluation.
The M1 is the most essential part of the motor cortex in the human cerebral cortex, located in the precentral gyrus.M1 is the most frequently stimulated target in noninvasive brain stimulation studies for post-stroke gait and balance recovery (Parikh et al., 2023).Among the 38 studies included in this article, 33 studies selected M1 as the stimulation target.Recent electrophysiological and imaging evidence underlined that a large motor network includes other key brain areas during the process of post-stroke functional recovery (Koch and Hummel, 2017).The cerebellum is a crucial structure involved in balance and motor control, and it is essential in motor adaptation and learning processes.Therefore, the cerebellum has been proposed as one of the alternative targets of M1.A total of 5 studies in our NMA used the cerebellum as the stimulation target and selected the iTBS mode for intervention.iTBS is a new rTMS mode that lasts only about 5 min, with the characteristics of short time-consuming, low intensity, and strong effect.iTBS can induce long-term potentiation, which helps promote neural plasticity and produce a safer and lasting intervention effect for stroke patients.Our findings showed that iTBS-Cerebellum can significantly improve balance function in stroke patients.Liao et al. (2024) compared the efficacy and safety of iTBS to the cerebellum or M1 on balance and motor recovery in stroke patients.They found that both iTBS-M1 and iTBS-Cerebellum could improve balance function and that iTBS-Cerebellum, but not iTBS-M1, had a more significant effect on motor recovery.Like our findings, Manto et al. (2012) suggested that iTBS-Cerebellum could be a potential therapeutic approach to improve balance and gait function in stroke patients.Thus, iTBS-Cerebellum may be a valuable new therapeutic option in stroke rehabilitation programs.At the same time, our NMA also included a study on the iTBS-M1 protocol, but the limited number of studies and participants may have led to inaccurate results.
It is worth noting that among the included studies, Chieffo et al. (2014) and Chieffo et al. (2021) used H-coil.The H-coil differs from conventional figure-of-8 coil and circular coil in that it can stimulate deeper cortical areas and neural networks (Roth et al., 2014).Roth et al. (2007) found that H-coil, figure-of-8 coil, and double-cone coil could generate the maximum induced electric field in the surface region of the saline head model at 0.9% concentration.With the increased distance from the simulated skull to the brain tissue, the induced electric field generated by the figure-of-8 coil and the double-cone coil decreased to less than 10% of the maximum induced electric field at 6 cm.In comparison, the electric field intensity of the H-coil was more than 63% of the maximum induced electric field.H-coil can theoretically stimulate deeper leg-related cortical motor areas within the intercerebral fissure approximately 3 to 4 cm below the skull.However, the results of Chieffo et al. (2014) and Chieffo et al. (2021) did not show a favorable advantage of dTMS.We suspect this may be due to the small number of current studies.In the future, more RCTs are needed to confirm the applicability and safety of dTMS in stroke patients.

Limitations
The study also has some limitations, including: (1) Coil type, stimulation intensity, total number of pulses, and duration of intervention were not exactly the same among the included studies, resulting in potential heterogeneity.(2) Despite including the full stimulation protocol in this analysis, the iTBS-M1 and dTMS-M1 groups accounted for 2.6 and 5.3% of the total data, respectively.This affects, to some extent, the quality of the conclusions of this study.
(3) The age and disease severity of the included patients were slightly different, and some data indicators will be affected.Further subgroup analysis according to age and disease severity is needed in the future.
(4) Adverse events may not be strictly reported in the included studies, so the safety of each intervention protocol needs to be further studied.

Conclusion
In this NMA, we found differences in the therapeutic effects between different rTMS protocols.Considering the impact of the stroke phase on the lower extremity motor function, the current research evidence shows that HFrTMS-M1 may be the preferred stimulation protocol to improve the lower extremity motor function of patients for post-stroke time > 1 month, and LFrTMS-M1 for poststroke time ≤ 1 month.In the future, high-quality, large-sample, multicenter, and long-term follow-up RCTs are needed to verify the conclusions of this study.

Figure 3
Figure 3 shows the network meta-analysis diagrams of eligible comparisons, where the blue circles represent the different interventions.The circle size represents the sample size.The straight line between the two circles represents a direct comparison between the two different interventions.The thicker the solid line, the greater the number of studies in that pairwise comparison.

FIGURE 1
FIGURE 1Flow diagram of the eligible studies selection process.CNKI, China National Knowledge Infrastructure; WanFang, WanFang Knowledge Service Platform; VIP, Chinese Scientific Journals Database; CBM, Chinese Biomedical Literature Service System; n, number of publications.

FIGURE 2
FIGURE 2Quality assessment of selected studies by the Cochrane Risk Of Bias Tool.(A) Risk of bias graph: review authors' judgments about each risk of bias item presents as percentages across all included studies.(B) Risk of bias summary: review authors' judgments about each risk of bias item for each included study.
Table 2. 1 closed loop was formed for the 3 interventions, so we assessed the inconsistency of the

TABLE 1
Characteristics of included studies.

TABLE 1 (
Continued) difference between the other two interventions (p > 0.05; Figure

TABLE 3
SUCRA values of different interventions for outcomes.