The databases, including Embase, PubMed, the Cochrane Library, the Web of Science, and China National Knowledge Infrastructure (CNKI), were searched to identify studies on the effect of rPMS on post-stroke upper limb motor dysfunction up to 20 February 2025. The English keywords for the database searches included “stroke,” “cerebrovascular accident,” “upper extremity,” “motor function,” “motor performance,” “repetitive peripheral magnetic stimulation,” “peripheral magnetic stimulation,” “magnetic field,” “rPMS,” and “PMS.” The reference lists of the identified articles were checked for potential studies. The detailed search strategies for each database are provided in Supplementary Table S1.

Two reviewers independently conducted the literature screening. Disagreements were recorded and resolved through discussions with a third reviewer. The inclusion criteria were as follows: (1) randomized controlled trial (RCT) studies; (2) studies involving participants who experienced a first-time stroke with upper limb motor dysfunction, confirmed by magnetic resonance imaging or computed tomography; and (3) studies in which the experimental group received rPMS treatment in addition to the control group’s intervention, which could be a placebo, sham, or routine rehabilitation. The exclusion criteria were as follows: (1) animal experiments or studies including healthy volunteers; (2) studies without the target outcome measures; (3) studies for which the full text was not available; and (4) studies lacking complete outcome data. For studies with overlapping data, those with larger or more complete datasets were prioritized. Both published and unpublished studies were considered, and authors were contacted if additional details not reported in the articles were needed.

Two reviewers independently evaluated the methodological quality of all included studies. A third reviewer recorded and resolved any disagreements. The Cochrane Collaboration Tool was utilized to assess the risk of bias for each RCT, including adequacy of sequence generation, concealment of allocation, blinding of participants and personnel, blinding of result evaluators, incomplete result data, and selective reporting (11, 12). The Grading of Recommendations, Assessment, Development and Evaluation (GRADE) guidelines for systematic reviews were followed to assess the quality of outcomes (13).

All included studies were conducted by two independent reviewers. If there was any disagreement, a third reviewer made the final decision. The following data were extracted from the included studies: basic information (study authors, year of publication), participant characteristics (age, time post-stroke, and sample size), rPMS parameters (site, frequency, intensity, and regimen of stimulation), outcome indicators of upper limb motor function, and activities of daily living (ADL).

The primary outcome included the Fugl-Meyer Assessment of the Upper Extremity (FMA-UE). Secondary outcome indicators for efficacy included the Modified Ashworth Scale (MAS), Functional Independence Measure (FIM), and Modified Barthel Index (MBI), as well as dropout rate and adverse effects.

As the first quantitative tool developed to assess the recovery of sensory and motor function after stroke, the FMA has been extensively tested in clinical settings and proven to be both feasible and effective in stroke. The scale is divided into five domains, namely motor function, sensory function, balance, joint range of motion, and joint pain. The FMA-UE evaluates the movement, coordination, and reflexes of the upper limbs (the shoulders, elbows, forearms, wrists, and hands), with its score ranging from 0 (hemiplegia) to 66 (normal motor performance) (14). The MAS is the most commonly used clinical tool for assessing muscle tension and spasticity. It is graded from 0 to 4 (0, 1, 1+, 2, 3, 4), where 0 indicates no resistance and 4 indicates limb stiffness during flexion or extension (15). The FIM, as a scale for evaluating functional independence, assesses 18 kinds of activities of daily living on a 7-point scale, ranging from 1 (completely dependent) to 7 (unassisted independent) (16). The MBI is a five-point scale used to measure activities of daily living, and Tomoko Ohura et al. affirmed its reliability and validity after stroke (17). The outcomes post-intervention from the follow-up phase were selected for meta-analysis if they were reported at multiple time points.

All statistical analyses were performed using RevMan 5.4 software (The Nordic Cochrane Centre, Cochrane Collaboration, Copenhagen, Denmark). Heterogeneity among studies was assessed using the chi-squared (Chi²) test and the I² statistic. A fixed effects model was applied when heterogeneity was low (I² <50%), whereas a random effects model was used when heterogeneity was substantial (I² ≥ 50%). Dichotomous data were presented as risk ratios (RRs) with 95% confidence intervals (CIs), and continuous data were expressed as mean differences (MDs) or standardized mean differences (SMDs) with 95% CIs. Subgroup meta-analyses of the primary outcome were conducted based on predefined variables, such as disease stage, stimulation frequency, coil type, stimulation duration, and stimulation intensity. Combined effect sizes were calculated within each subgroup, and differences between the subgroups were compared.

The selection process of this study is shown in Figure 1. A total of 1,268 potentially relevant studies were screened from four English databases and CNKI using relevant search strategies. Then, 58 duplicates were removed, and 72 studies were excluded because they were too old to obtain the full text. An additional 1,028 studies were removed after screening titles and abstracts. Finally, after reviewing the full texts of the remaining 110 articles, a total of 12 studies were finally included.

A total of 12 studies with 484 participants were included in this study. The characteristics of the included studies are presented in Table 1. The site, frequency, treatment intensity, number of pulses, on–off ratio, treatment duration, and coil type of rPMS stimulation differed between these studies.

A total of eight studies (18–24) stimulated more than two groups of upper limb muscles, with one study (6) stimulating only the triceps brachii muscle and another (25) stimulating the axilla of the affected arm. Furthermore, two studies targeted nerves: one at the cervical nerve root (9) and the other at the radial nerve above the elbow joint (26). The frequency of stimulation was ≤20 Hz in six studies (6, 9, 18, 22, 25, 27) and >20 Hz in six studies (19–21, 23, 24, 26). In addition, six studies (19, 21, 23, 24, 26, 27) determined the intensity of treatment based on the percentage of the maximum output value of the treatment apparatus, and four studies (6, 9, 20, 25) determined it based on the intensity of movement occurring in the wrist at rest. Moreover, one study (22) individualized treatment for patients, while one study (18) did not mention details. In total, five studies had a daily stimulation time of >20 min (18, 19, 23, 25, 27), six studies had a daily stimulation time of ≤20 min (6, 9, 20, 21, 24, 26), and one study (22) did not report the duration. There were seven studies (6, 18, 22–26) with a treatment duration of ≤2 weeks, four studies (9, 19, 21, 27) with a duration of >2 weeks, and one study (20) determined the duration based on the transfer time to the hospital. Six studies (18, 19, 22–25) used sham stimulation as the control, four studies (9, 21, 26, 27) used conventional rehabilitation, one study (20) used standard care, and one study (6) had no treatment in the control group.

As shown in Figures 2, 3, the 12 included studies showed low risk of bias in terms of blinding of outcome assessment (detection bias) and selective reporting (reporting bias), indicating that these studies were relatively standardized in their design and execution. However, a small number of high-risk biases were identified in random sequence generation (selection bias), blinding of participants and personnel (performance bias), and incomplete outcome data (attrition bias), indicating potential statistical errors in these areas. In addition, five studies (6, 9, 19, 23, 24) were assessed as low risk and high quality in terms of methodological quality. There were four studies (18, 21, 26, 27) with high-risk indicators. Among them, Chen et al.’s study (18) showed a high risk of bias in random allocation and allocation concealment, which likely led to selection bias because of non-random allocation. In total, three studies (21, 26, 27) had a high risk of loss of follow-up bias in terms of data integrity. In addition, three studies (20, 22, 25) had unknown risks on a small number of indicators and showed low levels of bias. Overall, the included studies showed high methodological quality and a high evidence level.

Furthermore, 11 studies (6, 9, 18–22, 24–27), involving a total of 442 participants, evaluated the effects of rPMS on the FMA-UE. Subgroup analyses were conducted based on disease stage, stimulation frequency, coil type, stimulation duration, and stimulation intensity. The meta-analysis conducted using a random effects model showed the following pooled SMDs: disease stage, SMD = 0.69 (95% CI 0.20–1.17, p = 0.006, I² = 78%) (Figure 4); stimulation frequency, SMD = 0.58 (95% CI 0.19–0.98, p = 0.004, I² = 74%) (Figure 5); coil type, SMD = 0.82 (95% CI 0.31–1.32, p = 0.001, I² = 75%) (Figure 6); stimulation duration, SMD = 0.62 (95% CI 0.19–1.05, p = 0.004, I² = 76%) (Figure 7); and stimulation intensity, SMD = 0.79 (95% CI 0.29–1.29, p = 0.002, I² = 78%) (Figure 8).

In the disease stage subgroup, rPMS significantly improved upper limb motor function in patients during the subacute phase (SMD = 0.74, 95% CI 0.08–1.40, p = 0.03), while no significant effects were observed in acute-phase (SMD = 1.24, 95% CI −0.52–3.00, p = 0.17) or chronic-phase patients (SMD = 0.12, 95% CI −0.28–0.52, p = 0.56) (Figure 4). Regarding stimulation frequency, the ≤20 Hz subgroup showed significant improvement (SMD = 0.86, 95% CI 0.19–1.52, p = 0.01), whereas the >20 Hz subgroup did not demonstrate significant effects (SMD = 0.27, 95% CI −0.09–0.63, p = 0.14) (Figure 5). In terms of coil type, the figure-eight coil subgroup significantly enhanced FMA-UE scores (SMD = 0.92, 95% CI 0.15–1.68, p = 0.02), while the circular coil subgroup did not reach statistical significance (SMD = 0.73, 95% CI −0.03–1.49, p = 0.06) (Figure 6). Stimulation durations of 15–20 min showed significant effects (SMD = 0.62, 95% CI 0.05–1.20, p = 0.03, I² = 78%), whereas durations over 20 min did not show significant differences (SMD = 0.64, 95% CI −0.12–1.39, p = 0.10) (Figure 7). For stimulation intensity, the high-intensity subgroup demonstrated significant effects (SMD = 1.02, 95% CI 0.33–1.71, p = 0.004), while the low-intensity subgroup did not reach statistical significance (SMD = 0.59, 95% CI −0.10–1.28, p = 0.09) (Figure 8).

A total of three studies (18, 21, 23), involving 114 participants, discussed the influence of rPMS on the MAS. The fixed effects analysis showed no significant improvement in upper limb spasticity compared to the control group (MD = 0.25, 95% CI −0.13−0.63, p = 0.20 > 0.05) (Figure 9).

A total of four studies (9, 19, 22, 26), involving 178 participants, discussed the FIM (19, 22) and MBI (9, 26). The fixed effects analysis showed that rPMS significantly improved patients’ ability to perform ADLs compared to the control group (SMD = 0.66, 95% CI 0.35–0.96, p < 0.0001) (Figure 10).

A total of seven studies (18, 21, 23–27), involving 492 participants, evaluated the effects of rPMS on the dropout rate. Heterogeneity among the included studies was low (I² = 25%, p = 0.24 > 0.05), and therefore a fixed effects model was used for meta-analysis. The results showed no significant difference between the rPMS group and the control group (RR = 0.99, 95% CI 0.65–1.50, p = 0.95) (Figure 11). Four participants in the control group dropped out due to temporary pain, and no adverse events were reported in the rPMS group in the study by Xie et al. (26).

According to the GRADE assessment, the overall level of evidence for the effect of rPMS was “High” for the FMA-UE, “Low” for the MAS, “Moderate” for the FIM and MBI, and “Low” for the dropout rate. (Table 2).