This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement, which is displayed in Figure 1. Web of Science, Pubmed, and Embase were searched in January 2026 using a combination of the following terms: “transcranial magnetic stimulation” AND dystonia”, OR “focal hand dystonia”, OR “musician’s dystonia”, OR “writer’s cramp.” No publication status or year limiters were applied; however, only studies reported in English were considered. The reference lists of all included articles were searched for studies missed in the initial search. All studies resulting from the searches were uploaded to Covidence software. As this study did not involve any interaction with human subjects, no ethical approval or consent was obtained.

Studies were screened using inclusion and exclusion criteria based on the PICO (participants, intervention, control, outcome) framework. Studies were first selected for qualitative review based on the following criteria: (P) participants who had a clinical diagnosis of FHD, with a study sample size of 3 or more; (I) TMS used as an intervention intended to reduce dystonia symptom severity; (C) no comparison group or randomization necessary; and (O) a behavioral outcome measure that assessed changes in clinical symptoms of dystonia [12]. Studies were selected for quantitative review (i.e., any statistical analysis) if (C) included a comparison group of FHD controls who received sham stimulation (parallel trials), or a design where FHD patients received both sham and real stimulation (crossover trials); (O) as above.

Literature search results were exported to Covidence. Two reviewers (ZBS and PJM) independently screened titles and abstracts obtained from the literature search against the inclusion and exclusion criteria. Full-text articles were then assessed against inclusion criteria, with disagreements resolved through discussion, and where necessary by a third member of the study team (NBP). Following the screening and inclusion of full text articles, data were extracted from individual studies into Covidence spreadsheets and exported into Microsoft Excel spreadsheets, including participant demographics, clinical information, clinical trial design, TMS stimulation site, TMS stimulation parameters and behavioral outcome measures. The primary outcome measures were changes in dystonic behavior post-TMS compared to pre-TMS, or active-TMS compared to sham-TMS. Dystonic behavior included digital behavioral measures, clinician rating scales and patient reported outcomes. In this review, we defined digital behavioral measures as behavior measured using digital tools such as measures of tracking error, and pen pressure generated from a digital writing assay. Clinician rating scales included Burke Fahn Marsden Dystonia Severity scale and patient reported outcomes included Arms Dystonia Disability scale [13, 14]. All three categories of measures were extracted and included in the meta-analyses. In studies where graphs or images were used to report the means and standard deviations (SDs) standard errors (SE), Plot Digitizer (version 3.0; https://plotdigitizer.com) software was used to extract values. Plot Digitizer software has high interrater reliability and is more accurate than traditional methods for extracting data from figures and graphs [15, 16].

Two independent reviewers (ZBS and PJM) assessed the methodological quality of the included studies using Cochrane Collaboration’s Risk of Bias (RoB) checklists [17]. The RoB checklist assesses studies on the domain’s randomization, allocation concealment, blinding of participants and personnel, eligibility criteria, intention to treat, between group comparisons, point estimates and variability, and adequate followups defined as greater than 1 week after intervention [17]. Each domain was judged to be of low, unclear, or high risk of bias, with an overall judgment given for each study, of low (low risk of bias for all domains), unclear (some concerns in at least one domain), or high (high risk of bias in at least one domain) risk [17]. The methodological quality of each study was also assessed using the PEDro scale [18]. A higher PEDro score indicates a higher quality of the trial, with scores of 9 and 10 indicating excellent quality, 6–8 indicating good quality, 4 and 5 indicating fair quality, and less than 4 indicating poor quality [18].

Due to the small sample sizes of the included articles, a Hedges’ g effect size was calculated to correct for potential overestimation of the population standardized mean difference (SMD) [19]. For all studies, a Hedges’ g and SD was first calculated for each outcome measure [19]. For studies with more than one outcome measure, the mean and SD of each measure before and after active-TMS and a mean and SD before and after sham-TMS were extracted and used to calculate a mean difference and SD for active-TMS and a mean difference and SD for sham-TMS. For each measure, the mean difference was then used to calculate a Hedge’s g and SE for active-TMS and a Hedge’s g and SE for sham-TMS. The Hedge’s g and SE were then averaged across measures to generate a Hedge’s g and SE for active-TMS and sham-TMS for each study. The direction of the outcome measure was adjusted so that positive values indicated improved dystonia behavioral outcomes, and negative values indicated worsened dystonia behavioral outcomes. The mean Hedge’s g for each TMS condition in each study was then used to calculate two pooled effect sizes: 1-change in dystonic behavior before and after TMS (pre-TMS vs. post-TMS) and 2-change in dystonia behavior after the two TMS conditions (active-TMS vs. sham-TMS). The pooled effect sizes were calculated using a random-effects model in Comprehensive Meta-Analysis (CMA; version 4) software. Both study level and the overall pooled effect size were considered significant if p < 0.05. All meta-analysis forest plots and funnel plots were prepared in CMA software. Between-study heterogeneity in effect sizes was quantified using the Q-test and I² statistic [20, 21]. The Q-statistic provides a test of the null hypothesis that all studies in the analysis share a common effect size [20]. If all studies shared the same true effect size, the expected value of Q would be equal to the degrees of freedom (the number of studies minus 1). The I² statistic for heterogeneity is measured as low, moderate, or high for I² values of 25%, 50%, and 75%, respectively [21].

The presence of publication bias across studies was assessed using funnel plots where effect sizes for each study were plotted against their standard error [19]. In the absence of publication bias, symmetrical distribution of effect sizes around the overall effect size is observed, indicating randomness due to a sampling error [19]. Egger’s test quantitatively evaluates the significance of any detected asymmetry, with non-significant asymmetry suggesting no publication bias [19]. The statistical significance level was set at 0.1 because the statistical power of the publication bias tests is generally low.

The electronic search strategy yielded 330 papers, including 84 duplicates removed automatically by the Covidence software and 45 removed manually (Figure 1). After screening the titles and abstracts for PICO framework, 201 full-text publications were retrieved and screened for eligibility. Of these, nine articles met the inclusion criteria and were included in the qualitative and quantitative synthesis (Table 1). Although three of the nine clinical trials originated from the same hospital and investigator, the study characteristics of the three articles were different [24, 27, 28]. The three studies, therefore, were deemed to be independent and included in the quantitative analysis [24, 27, 28].

Six of the nine included studies used a crossover (full or partial) study design while the remaining three used a parallel study design (Table 1). The sample sizes of the included studies ranged from 6 to 18 participants. All studies delivered two interventions. For eight of the nine studies, the two interventions were active-TMS and sham-TMS. One study delivered the same active-TMS combined with sensorimotor retraining (SMR) or control therapy.

A total of 111 individuals with FHD who received TMS treatment were included in the meta-analysis comparing pre-TMS and post-TMS (Figure 2) and 102 patients with FHD were included in the meta-analysis comparing active-TMS and sham-TMS (Figure 3). The mean age and sex distribution across the 111 FHD participants was 45.5 [SD 4.98] years with 63% male and 37% female. The duration of FHD symptoms ranged from 5.7 years to 16.4 years (Table 1). Overall, the FHD study cohort across the nine clinical trials represented a diverse patient cohort.

Six of the nine studies targeted TMS to the premotor cortex (PMC) [22–24, 26–28, 30]. Across the nine studies, TMS was also targeted to the motor cortex (MC, two studies), primary somatosensory cortex (S1, two studies), supplementary motor area (SMA, one study), and ipsilateral cerebellum (one study) [29]. Of note, cortical target for TMS was identified using scalp measurements in seven of the nine studies. The remaining two studies used fMRI to identify the cortical target for TMS delivery [25, 30].

TMS stimulation parameters were categorized by frequency of stimulation, intensity, waveform, and duration of each TMS session. Since some studies delivered multiple TMS sessions over five consecutive days, the total pulses was also reported for comparison. The majority of studies (6/9) delivered low-frequency rTMS (1 Hz or 0.2 Hz), while two studies delivered continuous theta-burst stimulation (cTBS) and one study used high-frequency rTMS (10 Hz) (Table 1). The intensity of TMS stimulation was in the range of 80%–90% of resting motor threshold (RMT) in half of the studies (5/9) and 80%–100% of active motor threshold (AMT) in the other half. The waveform was biphasic in most of the studies (7/9) and monophasic in the rest. The duration of each TMS session ranged from 40 s for cTBS to 40 min for 10 Hz rTMS. Of note, the 40-min rTMS session consisted of 5 min of TMS delivery interleaved with 5 min of motor behavior for four repetitions. The motor behavior was used to prime the motor circuitry prior to TMS delivery. Across all nine studies, the 40-min study was the only one that delivered TMS after a motor priming task [30]. The remaining studies delivered TMS during rest state. The total number of pulses delivered per stimulation site ranged from 600 to 9,000. All studies used figure-of-eight coils to deliver TMS.

To assess TMS effect in FHD, all nine clinical trials collected dystonic behavioral measures including digital behavioral measures, clinician rating scales and patient reported outcomes (Table 4). A majority of the digital behavioral measures included kinematic features of writing such as pen pressure, pen velocity and peak accelerations during writing. The durability of the effect size ranged from immediately after TMS delivery to 28 days post intervention (Table 1).

Methodological quality of studies was assessed using two different scaling systems: risk of bias scaling (Table 2) and physiotherapy evidence database (PEDro) scale (Table 3). Using the risk of bias scale, all nine studies showed low risk of bias in randomization, eligibility criteria, between group comparisons, and reporting of point estimates and measures of variability. A majority (5/9) of the studies also reported blinding. In these cases, blinding was reported primarily at the level of study subjects. A limited number of studies (3/9) also reported blinding of study assessors and only one study reported blinding of the TMS technician. A majority of the studies (6/9) were designed and maintained an intention to treat analysis. Adequate follow up (> 1 week after each TMS session) was observed in a majority (6/9) of studies. For the remaining studies (2/9), only one TMS session was delivered and follow up was performed on the same day. The PEDro scale was used to assess the overall study quality (Table 3). All nine studies were at minimum fair quality with a majority (7/9) meeting criteria for good quality.

Across all nine studies, a meta-analysis was performed comparing dystonic behavior before and after TMS (Table 4). A mild-moderate significant effect size favoring post-TMS and an overall reduction in dystonic behavior was observed with the random effects Hedges’ g = 0.36, 95% CI [0.17, 0.56], p < 0.001 (Figure 2). Between study heterogeneity as measured by Cochrane’s Q (X2) was 2.24 with 8 degrees of freedom. Since the Cochrane’s Q was less than the degrees of freedom, the amount of between-study variance in the observed effects was actually less than we would expect to see based on sampling error alone. Therefore, the variance of true effects was estimated as zero, and all indices of heterogeneity (I², tau² and tau) were set to zero.

Since a pre-post meta-analysis is vulnerable to regression to the mean, placebo effects, learning effects, and natural fluctuations in dystonic behavior, a second meta-analysis was performed to compare dystonic behavior after sham-TMS and active-TMS (Figure 3). As one study did not include a sham-TMS group, it was excluded from this analysis and the remaining eight studies were used to calculate a pooled effect size. The second meta-analysis also showed a moderate effect size favoring active-TMS for an overall reduction in dystonia with the random effects Hedges’ g = 0.35, 95% CI [0.12, 0.57], p = 0.002 (Figure 3). Between-study heterogeneity as measured by Cochrane’s Q (X2) was 0.93 with 7 degrees of freedom. Since the Cochrane’s Q was less than the degrees of freedom, the amount of between-study variance in the observed effects was less than we would expect to see based on sampling error alone. Therefore, the variance of true effects was estimated as zero, and all indices of heterogeneity (I², tau², and tau) were set to zero.

Since studies that report a treatment effect are more likely to be published and therefore lead to an overestimate of the true treatment effect, it is important to assess the likely extent of publication bias and its potential impact on the true TMS effect in dystonia. A funnel plot was, therefore, used to measure publication bias. In the absence of publication bias, we would expect the published studies to be distributed symmetrically around the combined effect size with larger study sizes clustering near the top of the funnel plot. A funnel plot analysis of the nine studies comparing pre-TMS vs. post-TMS (Figure 4A) and sham-TMS vs. active-TMS (Figure 4B) demonstrated a symmetrical distribution. The effect size calculated with the imputed studies (black diamond) was the same as the effect size with the observed studies (open diamond) suggesting that the extent of publication bias was low.

An Egger’s test was also performed for both meta-analyses. For pre-TMS vs. post-TMS meta-analysis, the Egger’s test indicated no significant publication bias (2.09, 95% CI [−0.95, 5.13], p = 0.148). Similarly, for sham-TMS vs. active TMS, the Egger’s test also indicated no significant publication bias (−1.05, 95% CI [−4.16, 207], p = 0.443). Overall, using a funnel plot and Egger’s test, we found no evidence of publication bias overestimating the TMS effect size in dystonia.

However, funnel plots and Egger’s test have limited power when fewer than ten studies are included, so the absence of detected publication bias should be interpreted with caution.

This review has several limitations that should be considered. First, the limited number of clinical trials investigating the effects of TMS in FHD precluded a comprehensive meta-analysis including secondary analysis of moderators of TMS benefit. This gap in the literature may be addressed in the future as more rigorous and larger scale clinical trials of TMS in FHD and other dystonia subtypes are funded. Second, the TMS studies in FHD in this meta-analysis also varied in brain targets, TMS parameters, TMS dosing, and brain state during TMS delivery. The finding of very low heterogeneity may reflect low sampling power to detect heterogeneity rather than true homogeneity. Third, the included clinical trials provided limited data on long-term TMS effects, with few studies extending follow-up beyond 1 month. This lack of long-term findings hampered our ability to draw conclusions regarding the prolonged effects of TMS on dystonia subtypes. Lastly, there were no uniform digital behavioral outcome measures used by all clinical trials which likely led to the variability in evaluating TMS effect on dystonia symptoms. Nevertheless, the behavioral measures used across the nine trials were reported in sufficient detail to allow calculation of a pooled effect and a low between study heterogeneity.