Seven databases (PubMed, Embase, Cochrane Central Register of Controlled Trials (CENTRAL), Chinese Biomedical Literature Database (CBM), Chinese National Knowledge Infrastructure (CNKI), China Science and Technology Journal Database (VIP), and Wanfang) were searched from inception to August 2021 using explicit search strategies combining MeSH terms and free text. The search strategy for all databases is described in detail in Supplementary Material 2. Inclusion and exclusion criteria are shown in Table 1.

The initial selection was conducted independently by two researchers (DX and PZ) through independent screening and analysis of the titles and abstracts of the imported studies. Another two researchers (JC and XT) read the full text of the potentially eligible papers and conducted a second round of selection. Additionally, the reasons for excluding papers were recorded and a third researcher (SYL) was consulted to resolve disagreements during the screening process. Note Express 3.5.0 software (Beijing Aegean Music Technology, Beijng, China) was used to manage all the imported search records.

Data extraction was conducted by two researchers (PZ and XT) using a standardized form. The following data were extracted: study characteristics, trial methodologies, participant information, the stimulation parameters of rTMS, sham or non-rTMS control, and outcomes. Further details on data extraction are described in Supplementary Material 3.

The Cochrane risk-of-bias tool from the Cochrane Handbook for Systematic Reviews of Interventions was applied using Review Manager to evaluate the risk of bias for each study [30]. The risk-of-bias tool consisted of seven items, each of which was rated on a three-level scale. The risk of bias was evaluated independently by two researchers (PZ and JC), and a separate researcher (SYL) made the final decision in the event of a disagreement. The quality of the data was assessed by the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) [31], which contained four criteria (inconsistency, indirectness, imprecision, and publication bias). Each of these four criteria were rated on a four-level scale: high, moderate, low, or extremely low.

To reduce heterogeneity and improve comparability between rTMS groups across studies, we extracted the preintervention means and standard deviation of each study to conduct baseline difference testing.

Heterogeneity between eligible studies was measured using I${}^{\mathrm{2}}$ statistics and Galbraith plots. I${}^{\mathrm{2}}$ statistics were calculated to determine the magnitude of study heterogeneity and Galbraith plots were used to reflect the extent of heterogeneity. To confirm the robustness of the conclusions and measure publication bias, Egger’s test and Begg’s test were conducted.

Data was analyzed using STATA 17 (StataCorp LLC4905 Lakeway Drive, College Station, TX, USA) to determine whether rTMS can improve memory performance, global cognition, and activities of daily living (ADL). All data analyses were performed using postintervention data from both the rTMS group and control group, and the results are presented as the mean difference (MD) and 95% confidence interval (CI). A random-effect model or a fixed-effect model was applied to determine effect sizes. When I${}^{\mathrm{2}}$ = 0.00%, the fixed-effect model was employed; otherwise, the random-effect model was applied.

We performed subgroup analysis based on stimulation frequency (low frequency or high frequency), age (greater or less than 60), sex (male or female), stimulation intensity, duration of intervention (greater or less than 1 month), and intervention in the control group (sham rTMS or non-rTMS). Sensitivity analysis was performed to reduce the level of heterogeneity. The leave-one-out method was used to increase the reliability and credibility of the results.

The databases initially yielded 1230 studies, 42 of which were potentially eligible for the final selection. Thirty-seven papers were subsequently excluded as a result of participants not being diagnosed with memory dysfunction, data from multiple studies being derived from the same trial, lack of appropriate memory assessment scale, and ineligible article types. The remaining five articles were included in the final analysis. The process of study selection is presented in Fig. 1.

Fig. 1.

Flow diagram of study selection.

The relevant features of eligible papers are presented in Table 2 (Ref. [10, 11, 12, 13, 14]). A total of 192 patients were included in the meta-analysis, including 121 males (63.0%) and 71 females (37.0%). The number of subjects in the eligible studies varied from 24 to 50 per study. The mean age range of the participants was 42.5 to 68.3 in the rTMS group and 47.3 to 69.2 in the control group. Three studies reported the prevalence rates of ischemic and hemorrhagic stroke; the number of ischemic strokes was 75 (58.6%) [10, 11, 14]. Participants’ average number of years of education ranged from 9.48 to 12.8 in the rTMS group and from 9.05 to 11.85 in the control group [11, 12, 13, 14].

The stimulation parameters are shown in Table 2. A figure-eight coil was applied in two studies [11, 14], an O-coil was used in two studies [12, 13], and the coil type was not reported in the fifth study [10]. Two studies employed 1-Hz low-frequency stimulation [11, 14], and the other three employed 5-, 10-, or 20-Hz high-frequency stimulation [10, 12, 13]. The stimulation intensity in all the studies ranged from 60% to 120% of the resting motor threshold (five sessions per week for four consecutive weeks). The stimulation sites for rTMS included the left DLPFC [13], the contralesional DLPFC [14], bilateral DLPFC [12], the right DLPFC [11], and ipsilesional frontal cortex [10]. Regarding the interventions in the control group, three studies employed sham rTMS using three different approaches. These included a sham coil, a real coil placed perpendicular to the scalp that was not active, or stimulation stopped after 15 seconds. [11, 12, 14]. Non-rTMS interventions (drugs and regular cognitive rehabilitations) were used in the other two studies [10, 13].

All papers that were included assessed memory function using the RBMT [10, 11, 12, 13, 14]. One study used the Cambridge Prospective Memory Test Scale (CAMPROMPT) in addition to the RBMT to evaluate memory performance [10]. To assess global cognition, the Montreal Cognitive Assessment (MoCA) was administered in four studies [11, 12, 13, 14], one of which used a combination of the MoCA and the MMSE [14]. Ou et al. [10] did not assess global cognition. The Modified Barthel Index (MBI) was utilized to assess ADL in three studies [12, 13, 14]. Ou et al. [10] and Lu et al. [11] did not assess ADL. No adverse events were reported in these studies, except for Huang [13].

Three papers used a random number method to divide participants into treatment groups [11, 12, 14]. One paper used a computerized randomization method [13]. The method of randomization was not specified in the fifth paper [10]. In regard to allocation concealment, one paper used sealed opaque envelopes [13] and the remaining four papers did not report the details of allocation concealment [10, 11, 12, 14]. Regarding blinding, three papers were considered to have a low risk of bias as both the patients and examiners were blinded [11, 12, 13]. Two papers had potential risk of performance and detection bias; only patients were blinded in the study by Wen et al. [14], and the blinding method was unspecified in the study by Ou et al. [10]. All of the included studies had complete data (Fig. 2).

Fig. 2.

Risk-of-bias graph and summary of the included studies.

According to the GRADE, the level of evidence for the primary (RBMT) and secondary outcomes (MoCA and MBI) was moderate. The quality of this outcome was downgraded because the method of allocation concealment was unclear (Supplementary Material 4).

The results of data synthesis based on the baseline data indicated no significant difference between the rTMS and control groups in scores on the RBMT (MD = –0.76, 95% CI [Confidence Interval] [–1.68, 0.15], p = 0.10, I${}^{\mathrm{2}}$= 0.00%, Fig. 3A), MoCA (MD = –0.65, 95% CI [–2.07, 0.77], p = 0.37, I${}^{\mathrm{2}}$ = 0.00%, Fig. 3B), and MBI (MD = –0.73, 95% CI [–8.27, 6.80], p = 0.85, I${}^{\mathrm{2}}$ = 37.93%, Fig. 3C).

Fig. 3.

Comparison of pre-intervention clinical data between the rTMS and control groups. (A) Comparison of differences in RBMT between the rTMS and control groups. (B) Comparison of differences in MoCA between the rTMS and control groups. (C) Comparison of differences in MBI between the rTMS and control groups. rTMS, repetitive transcranial magnetic stimulation; RBMT, Rivermead Behavioral Memory Test; MoCA, Montreal Cognitive Assessment; MBI, Modified Barthel Index; WMD, Weighted Mean Difference; 95% CI, 95% Confidence Interval.

The Galbraith plot also did not indicate heterogeneity, as all dots were distributed between the two regression lines (Fig. 4A). There was no significant heterogeneity among the studies on memory performance (Q = 3.94, p = 0.41, I${}^{\mathrm{2}}$ = 0.00%, Fig. 5A). Neither Begg’s test (t = 0.73, p = 0.462, Fig. 4B) nor Egger’s test (t = 1.00, p = 0.393, Fig. 4B) revealed evidence of publication bias.

Fig. 4.

Series assessments for heterogeneity and publication bias for the effect of rTMS on memory function. (A) Galbraith plot indicating no heterogeneity. (B) Begg’s test and Egger’s test suggesting no publication bias.

Fig. 5.

Comparison of postintervention clinical data between the rTMS and control groups. (A) Comparison of differences in RBMT between the rTMS and control groups. (B) Comparison of differences in MoCA between the rTMS and control groups. (C) Comparison of differences in MBI between the rTMS and control groups.

As mentioned in the Methods and published protocol section [32], a random-effect model or a fixed-effect model was applied to determine effect sizes. When I${}^{\mathrm{2}}$ = 0.00%, the fixed-effect model was employed; otherwise, the random-effect model was applied.

3.6.1 Primary Outcome

The RBMT contains 12 sub scores consisting of 10 retrospective memory tasks and 2 prospective memory tasks. This scale was used to evaluate memory for details. All included papers used the RBMT to assess the impact of rTMS on memory function. The results based on the fixed-effect model indicated that the RBMT scores of the rTMS group were significantly greater than those of the control group (MD = 1.73, 95% CI [0.85, 2.60], p 0.001, I${}^{\mathrm{2}}$ = 0.00%, Fig. 5A).

As shown in Fig. 6A, subgroup analysis revealed no significant difference between the effects of the LF-rTMS and HF-rTMS on memory performance (Q = 0.52, p = 0.47, I${}^{\mathrm{2}}$ = 0.00%). Regarding low-frequency versus control, subgroup analysis demonstrated that the effect of LF-rTMS on memory function in participants with PSMD was significantly greater than that of control treatment (MD = 2.12, 95% CI [0.90, 3.33], p 0.001, I${}^{\mathrm{2}}$ = 0.00%). Regarding high-frequency versus control treatment, there was no significant difference in memory performance between these two conditions (MD = 1.42, 95% CI [–0.07, 2.90], p = 0.06, I${}^{\mathrm{2}}$ = 18.87%, Fig. 6A).

Fig. 6.

Forest plots of subgroup analysis. (A) Stimulation frequency (low frequency or high frequency). (B) Intervention in the control group (sham rTMS or non-rTMS).

As shown in Fig. 6B, the sham rTMS group and the non-rTMS did not significantly differ according to subgroup analysis (Q = 0.01, p = 0.94, I${}^{\mathrm{2}}$ = 0.00%). Regarding sham rTMS versus rTMS, an analysis with three studies [11, 12, 14] indicated rTMS group was superior to the sham rTMS group in RBMT (MD = 1.96, 95% CI [0.84, 3.07], p 0.001, I${}^{\mathrm{2}}$ = 0.00%). Regarding non-rTMS versus rTMS, the analysis with two studies [10, 13] reported that there was no significant difference between the non-rTMS and rTMS groups (MD = 1.85, 95% CI [–0.80, 4.50], p = 0.17, I${}^{\mathrm{2}}$ = 58.96%, Fig. 6B).

Regarding the duration of intervention, a subgroup analysis could not be conducted because all the interventions examined herein lasted for one month. Regarding sex, all included studies recruited both males and females and did not assess the data separately by sex. Regarding stimulation intensity, two studies had the same stimulation intensity but the intensity was different in the remaining three studies. Regarding age, both younger and older patients were simultaneously included in four studies. Therefore, subgroup analyses could not be performed based on the duration of intervention, age, sex, or stimulation intensity (Supplementary Material 5).

The leave-one-out method revealed that the effect sizes fell within the original confidence interval regardless of which study was left out. This result suggested that the assessment of memory evaluation is stable and reliable (Fig. 7).

Fig. 7.

Sensitivity analysis for the RBMT.

rTMS is a noninvasive technique used to modulate cortical activity. Its effects on memory performance among stroke patients may be associated with its influence on long-term depression (LTD) and long-term potentiation (LTP), processes that are required for memory formation [35]. LF-rTMS decreases cortical excitability by influencing LTD, while HF-rTMS facilitates cortical excitability through LTP [36]. Studies have demonstrated that the effectiveness of TMS is related to the activation of brain regions such as the prefrontal cortex (PFC) [37, 38]. and hippocampus [39]. Additionally, rTMS was shown to promote the recovery of damaged nerves and the production of new neurons through upregulation of brain-derived neurotrophic factors [40].

Our findings that rTMS treatment was effective at improving memory performance in patients with PSMD is consistent with previous studies [8, 21, 23, 41]. Conversely, it differs from the studies of Sedlackova et al. [17] and Rektorova et al. [16], that found rTMS to have no effect on memory performance. There are several tenable reasons for these inconsistent findings. It may be associated with differences in sample size between studies. Their studies included 7 participants [16, 17], whereas our study included 192 patients. Additionally, their studies compared baseline and post-intervention data, while we compared the performance of the two groups after the intervention in the absence of differences at baseline. The latter approach may be more appropriate for demonstrating the efficacy of rTMS. Stimulation parameters also varied across different studies, possibly affecting the clinical efficacy of rTMS.

Subgroup analyses revealed that the LF-rTMS condition was superior to the control condition, whereas HF-rTMS was not different from the control condition. These findings contradict those of other studies [21, 42, 43] that have reported HF-rTMS as being superior to control treatments. This inconsistency may be a result of different stimulation sites, frequency, and intensity. There was also heterogeneity detected, which may affect the final effect size. Therefore, additional studies are required to investigate the efficacy of HF-rTMS on memory performance.

In addition, stimulation frequency indicated similar efficacy between LF-rTMS and HF-rTMS subgroups, results that are consistent with previous studies [42, 44, 45, 46]. For instance, Ying-Hui Chou and coworkers reported that both LF-rTMS and HF-rTMS over the right DLPFC and the left DLPFC, respectively, were effective at enhancing memory performance in patients with mild cognitive impairment (MCI) and Alzheimer’s disease (AD) [42]. Another meta-analysis showed that both low- and high-frequency rTMS could significantly enhance memory performance [44]. Additionally, several studies have shown that low-frequency and high-frequency rTMS have similar effects in the treatment of poststroke motor disorder [45, 46]. Xiang H et al. [46] conducted a meta-analysis to investigate how the adjustable parameters of rTMS influence its efficacy as a treatment for poststroke motor disorder and found that stimulation frequency did not have an effect on the efficacy of rTMS. Du J et al. [45] reported that both 10 Hz HF-rTMS and 1 Hz LH-rTMS were superior to sham rTMS for poststroke motor disorder as measured by the Fugl–Meyer Assessment (FMA), while similar efficacy was observed between HF-rTMS and LH-rTMS.

Furthermore, our findings revealed no significant difference between the effects of the sham and non-rTMS control conditions on memory function. This suggests that rTMS does not have a placebo effect on memory rehabilitation. Accordingly, Jelić MB et al. [47], also did not report a placebo effect of rTMS on motor learning. This supports our conclusion as memory function is essential for motor learning. However, the placebo effect was not confirmed due to the considerable heterogeneity. Thus, further studies are required to explore the placebo effect of sham rTMS.

The comparison of the rTMS group versus the control group revealed that rTMS was effective at enhancing global cognition and ADL in patients with PSMD. These results are consistent with those reported in previous publications [20, 24, 38, 41]. For instance, Mengting Liu and colleagues reported positive effects of rTMS on cognitive function among stroke patients [20]. Another study found that intermittent theta-burst stimulation significantly enhanced cognition and quality of life among patients with PSCI [38]. Additionally, studies have shown that rTMS improves quality of daily life and general cognition among patients with AD [48], Parkinson’s disease [49], and MCI [42]. Nevertheless, our results should be followed up with additional evidence as there was mild heterogeneity in our meta-analysis.

Only one article included in our review reported that there were no adverse effects, while the others did not mention them. Thus, no firm conclusions can be drawn from our meta-analysis regarding possible adverse events. Previously studies have reported that rTMS is safe for treating both poststroke motor disorder and aphasia [50, 51]. However, several articles have reported that patients experienced dizziness [52], headache [45], anxiety [53], and other discomforts after rTMS treatment. These adverse events may be related to the high stimulation intensity of rTMS and individual participant differences. Therefore, it is necessary to conduct additional studies to determine the safety of rTMS for treatment of patients with PSMD.

One limitation of our study was the small number of eligible papers on patients with PSMD. As a result, we did not examine funnel plots to evaluate publication bias. While our findings support the use of rTMS in clinical practice for improving memory performance, MoCA and MBI scores among patients with PSMD, the duration of its efficacy is still unknown as long-term follow-up studies are limited. Future studies should extend the follow-up observation period to explore the time point at which the best long-term outcomes can be obtained and to provide new insights into the long-term benefits of rTMS treatment for PSMD. Additionally, mechanical embolectomy is closely associated with stroke outcomes and future analyses should control for the impact of this factor on results. Future clinical trials should control for confounding factors and stratify patients according to the type of PSMD to improve data quality and determine the best stimulation strategy. Furthermore, future studies need to improve clarity of methodological details regarding trial design by including information about allocation concealment and blinding to decrease the risk of bias.