We searched nine commonly used electronic databases: PubMed, Cochrane Library, Embase, Web of Science, SinoMed, OVID, the China National Knowledge Infrastructure (CNKI), China Science and Technology Journal Database (VIP), and Wanfang Data for RCTs of rTMS for stroke patients with NFA. Furthermore, relevant systematic evaluations and reference lists of included studies were searched manually to ensure the comprehensiveness of included studies. Keywords were determined after preretrieval: repetitive transcranial magnetic stimulation, rTMS, TBS, stroke, cerebrovascular accident, aphasia, non-fluent aphasia, and post-stroke aphasia. The final literature search was conducted on January 8, 2024, using neither language nor publication date restrictions.

Inclusion criteria (1) study design: RCTs; (2) study population: individuals with NFA following a clinical diagnosis of stroke; (3) interventions: In addition to the interventions applied to the control group, the experimental group underwent rTMS. Alternatively, the experimental group received rTMS, whereas the control group received sham-rTMS. For studies encompassing more than two groups, the groups fulfilling the inclusion criteria were also included; (4) the outcome indicators ought to incorporate at least one metric of aphasia assessment, such as Western Aphasia Battery (WAB), Aphasia Battery of Chinese (ABC), China Rehabilitation Research Center Standard Aphasia Examination (CRRCAE), and Boston Naming Test (BNT).

Exclusion criteria: (1) duplicate studies or data cannot be extracted; (2) case studies, animal experimental studies, or reviews; (3) the unavailability of full text even after contacting the author via email.

Endnote X9 was utilized for document organization and deduplication. Two reviewers independently conducted literature screening, data extraction, and cross-verification according to predetermined inclusion and exclusion criteria. Any discrepancies between the reviewers were resolved by achieving a consensus with an unbiased third-party researcher. The process of extracting data entailed gathering details regarding the title, first author, publication year, the number of patients, diagnostic criteria, intervention and control protocols, rTMS parameters (e.g., stimulation site, frequency, intensity, and duration), outcome measures, any reported adverse events and follow-up duration.

Meta-analysis was conducted utilizing Revman 5.4 software. The mean differences (MD) along with the 95% confidence interval (CI) were employed for statistical analysis, and the standardized mean difference (SMD) was used when using different measurement methods or units. Statistically significant differences were indicated by p < 0.05, and the magnitude of heterogeneity was quantified using I². I² shows the proportion of heterogeneity in the total variation of effect size based on the Student–Newman–Keuls test, ranging from 0 to 100%. I² ≤ 50% was deemed as an indication of low heterogeneity, employing the fixed-effects model for Meta-analysis; I² > 50% was considered a clear indication of significant heterogeneity, utilizing the random-effects model.

The initial search returned 1,244 articles. Following the removal of duplicate articles, 607 studies were left, and 59 studies remained upon reviewing the titles and abstracts. Finally, 47 studies were included after a rigorous evaluation of the full-text articles (Supplementary Figure S1).

Out of the 48 studies (4, 23, 35–80), 37 studies (35–37, 39–49, 51, 52, 54–66, 68, 69, 72–74, 76, 78, 79) had reported diagnostic criteria for stroke and 35 (4, 35–39, 41–53, 55, 57–60, 62–65, 67–70, 74, 76, 77) for aphasia. In 43 studies (4, 23, 35–52, 54–65, 67–76, 78, 80), the age of the patient was reported as mean ± standard deviation, while two studies (53, 66) reported specific age information for each patient, and the age range was reported in two studies (77, 79). Except for two studies (50, 59) that failed to disclose the post-stroke time, the mean course of disease for patients in the other 44 studies (4, 23, 35–49, 51, 52, 54–58, 60–65, 67–80) ranged from 6.9 days to 4.46 years, and two studies (53, 66) reported specific course of disease for each patient. Of the total studies, 39 studies (4, 35–43, 45–48, 51–55, 58–70, 72–74, 77–80) specifically reported that the patients were right-handed, whereas the remaining 9 studies (23, 44, 49, 50, 56, 57, 71, 75, 76) failed to furnish this information. Duration of the intervention in 45 studies (4, 23, 35, 36, 38–40, 42–46, 48–80) spanned from 2 to 4 weeks, except for 2 studies (37, 41) with an intervention duration of 30 days and one (47) with 8 weeks. Regarding the content of the intervention, five studies (49, 50, 59, 71, 80) used rTMS or sTMS alone, and the remaining studies also had aphasia treatment components including SLT, acupuncture, and electroacupuncture (EA). Patients were followed up after treatment in 15 studies (4, 39, 40, 53, 57, 66, 67, 70–74, 77, 78, 80), ranging from 30 days to 12 months.

Adverse events were mentioned in 10 studies (4, 40, 42, 45, 51, 57, 60, 61, 73, 77). Six studies (4, 40, 42, 45, 60, 73) reported headache; three (40, 45, 51) reported dizziness; adverse effects of epilepsy seizures were seen in two studies (61, 77). One study (57) documented the occurrence of adverse events such as disorientation, injuries resulting from falls, and aspiration caused by dysphagia among both control and experimental participants, the ratio of aspiration was the highest in the experimental group (3 cases), and the control group (11 cases), which revealed that the combination of rTMS and speech training could reduce the complications and improve the therapeutic efficiency.

Regarding the NFA type, 16 studies (23, 37, 41, 42, 44, 45, 47–49, 54, 56, 57, 61, 74, 78, 79) included patients with Broca aphasia. Four studies (36, 53, 63, 64) included patients with global aphasia, and 5 studies (39, 43, 72, 73, 76) reported on the number of patients with different types of aphasia. The aphasia types of patients in the three studies (43, 72, 73) were global, Broca’s, and transcortical motor aphasia, and Wang’s (73) study included patients with mixed transcortical aphasia. In addition to the intervention received by the control group, the experimental group in one study (48) received high-frequency TMS (HF-rTMS) on the IFG-L. One study (67) performed bilateral cerebral pulse stimulation, with HF-rTMS (> 1 Hz) stimulating the left cerebral Broca area and low-frequency TMS (≤1 Hz) (LF-rTMS) stimulating the right Broca area; three studies (38, 46, 56) stimulated the right inferior frontal gyrus (IFG-R) with 0.5 Hz. Of the 40 studies with a frequency of 1 Hz, except for one study (53) in which the stimulation site was the right superior temporal gyrus, the stimulation site was the IFG-L. Three of these studies (32, 66, 80) further stimulated Brodmann 45 (pars triangularis) at the site of the IFG-R partition. Three studies (43, 58, 77) proceeded theta burst stimulation (TBS) on patients, one (77) conducted passive cTBS on the cerebellum in patients of the experimental group, and the other two (43, 58) conducted iTBS on the IFG-L. The critical characteristics of the included studies have been summarized in Table 1.

The outcomes of the risk-of-bias assessments conducted for each included study are shown in Supplementary Figures S2A,B.

Potential publication bias across the included studies was evaluated based on Egger’s tests in Stata MP 17, with a significance level set at p < 0.1 (Supplementary Table S1).

A total of 12 studies were excluded from this meta-analysis due to the following reasons. Specifically, the outcomes of two studies (71, 73) were shown as bar graphs, and two (66, 74) were presented as line charts rather than specific values. Individual outcome indicators in four studies (56, 58, 67, 74) could not be integrated with other studies. Zheng’s (77) date was unacquirable. Data in three studies (62–64) of Zhu was described by Median and quartile which could not be counted. Finally, a total of 35 studies were included.

Fifteen studies (23, 36–38, 41, 42, 44, 46, 51, 52, 57, 59, 61, 65, 75) involving 960 patients assessed aphasia quotient (AQ) in patients after rTMS stimulation. The random-effects model was utilized due to significant heterogeneity across the studies (I² = 78%). Global language ability of the rTMS-treated group exhibited a substantial improvement in comparison to the control group [p < 0.00001] (Figure 1). Subgroup analysis was conducted according to the stroke stages of patients after removing a study (50) with unidentified stroke duration and another study (75) that patients were in the sequelae stage. The random-effects model was employed for the following studies: three studies (37, 41, 44) with patients in the acute phase [SMD = 1.41, 95% CI (0.91, 1.91), p < 0.00001, I² = 78%] and 11 studies (23, 36, 38, 42, 46, 51, 52, 57, 59, 61, 65) with patients in the recovery stage [p < 0.00001, I² = 77%] (Figure 2). Upon conducting sensitivity analysis, the removal of two studies (36, 57) from the subgroup of recovery, the following outcomes were observed: 9 studies (23, 38, 42, 46, 51, 52, 59, 61, 65) of recovery subgroup [p < 0.00001, I² = 0%] (Figure 3). The findings suggest that rTMS has the potential to enhance the overall language abilities of stroke patients during both acute and recovery phases.

Twenty-two studies (23, 36–39, 41–44, 46, 48–52, 55, 59, 61, 65, 68, 69, 75) involving 1,228 patients assessed repetition ability in patients after rTMS treatment. The random-effects model was utilized due to significant heterogeneity across the studies (I² = 85%). Repetition ability of rTMS-treated group exhibited a substantial improvement in comparison to the control group [p < 0.00001] (Figure 4). Subgroup analysis was conducted according to the stroke stages of patients after removing a study (50) with unidentified stroke duration. The random-effects model was employed for the following studies: four studies (37, 41, 44, 59) with patients in the acute stage [p < 0.00001, I² = 84%]. The fixed-effects model was employed for the following studies: 14 studies (23, 36, 38, 39, 42, 43, 46, 48, 49, 51, 52, 55, 61, 65) with patients in the recovery stage [p < 0.00001, I² = 41%] and three studies (68, 69, 75) with patients in the sequelae stage [p = 0.005, I² = 0%] (Figure 5). These findings suggest that rTMS enhance repetition abilities in patients across various stages of stroke recovery. Notably, beyond immediate benefits, rTMS appears to exert medium- to long-term effects on language improvement. For individuals with NFA post-stroke, the positive effects of rTMS on speech enhancement persisted for up to 12 months (66).

Twenty-two (23, 36–38, 41–44, 46, 48–52, 55, 59–61, 65, 68, 69, 75) studies involving 1,229 patients evaluated naming ability in patients after rTMS. The random-effects model was utilized due to significant heterogeneity across the studies (I² = 86%). Naming ability of the rTMS-treated group exhibited a substantial improvement in comparison to the control group [p < 0.00001] (Figure 6). Subgroup analysis was conducted according to the stroke stages of patients after removing a study (50) with unidentified stroke duration. The random-effects model was employed for the following studies: four studies (37, 41, 44, 59) with patients in the acute stage [p < 0.00001, I² = 83%]. The fixed-effects model was employed for the following studies: 14 studies (23, 36, 38, 42, 43, 46, 48, 49, 51, 52, 55, 59, 61, 65) with patients in the recovery stage of [p < 0.00001, I² = 1%] and three studies (68, 69, 75) with patients in the sequelae stage [p = 0.0003, I² = 0%] (Figure 7). Naming is among the most challenging functions for stroke patients with NFA to regain. The restoration of naming capabilities necessitates the involvement of extensive neural networks, and rTMS has been shown to facilitate the recovery of these abilities across various stages of stroke, primarily by enhancing the connectivity among pertinent brain areas (65).

Seventeen studies (4, 23, 36–38, 41, 42, 44, 46, 48, 50–52, 55, 59, 61, 65) involving 1,046 patients assessed spontaneous language ability in patients after rTMS stimulation. The random-effects model was utilized due to significant heterogeneity across the studies (I² = 67%). Spontaneous language ability of the rTMS-treated group exhibited a substantial improvement in comparison to the control group [p < 0.00001] (Figure 8). Subgroup analysis was conducted according to the stroke stages of patients after removing a study (50) with unidentified stroke duration and another study (75) that patients were in the sequelae stage. The random-effects model was employed for the following studies: four studies (37, 41, 44, 59) with patients in the acute stage [p < 0.00001, I² = 64%]. The fixed-effects model was utilized for the following studies: 11 studies (23, 36, 38, 42, 46, 48, 51, 52, 55, 61, 65) with patients in the recovery stage [p < 0.00001, I² = 0%] (Figure 9). These results suggested that rTMS ameliorated spontaneous language capability in patients in acute and recovery stages. Acupuncture therapy constitutes a significant component of complementary and alternative medicine, which has potential therapeutic effects in the treatment of PSA (81). The combined therapeutic application of acupuncture and rTMS on NFA has demonstrated a markedly superior efficacy compared to monotherapy. Specifically, the overall effectiveness of combining 1 Hz rTMS with acupuncture for NFA reached 96.66%, strongly associated with enhanced blood flow velocity and perfusion in the left middle cerebral artery (MCA) (44). Interaction between various acupuncture techniques and rTMS frequencies in producing distinct clinical outcomes warrants further investigation.

Two studies (43, 65) involving 103 patients assessed the concentration of brain-derived neurotrophic factor (BDNF) within patients. The fixed-effects model was utilized due to low heterogeneity across the studies (I² = 0%). The rTMS-treated group exhibited a substantial enhancement in serum BDNF concentration in comparison to the control group [p < 0.00001] (Figure 10). The findings indicated that rTMS has the potential to elevate the concentration of BDNF in the serum of stroke patients.

Two studies (48, 49) involving 84 patients used the Hamilton Depression Scale (HAMD) to evaluate the effects of rTMS on mood in stroke patients. The fixed-effects model was utilized due to low heterogeneity across the studies (I² = 0%). The HAMD score of rTMS group was notably lower in comparison to the control group [p = 0.0005] (Figure 11). The findings indicated that rTMS had potential to ameliorate the depression of these patients.

Recognition of numerous constraints is essential in this systematic review. Firstly, the high proportion of Chinese literature may cause some publication bias. Secondly, clinical and methodological heterogeneity among the included literature may influence overall results, including the age of patients and varied rTMS treatment regimens. Thirdly, owing to restrictions on article length, only some meta-analysis findings were reported. In the future, we can discuss the therapeutic effects of different dosages of rTMS and different rTMS approaches, and design accurate rTMS parameters according to aphasia types, combined with functional imaging technology to further explore the related mechanisms.