The comprehensive search was conducted in the PubMed, Embase, Cochrane Library, OVID Medline, and Web of Science databases, with the search time set from establishing the databases until November 19, 2024. We combine subject keywords and free terms, structuring the search strategy into three parts: disease (Parkinson’s disease or Alzheimer’s disease), intervention (repetitive transcranial magnetic stimulation), and study type (randomized controlled trial). In addition, we carefully searched for meta-analyses related to the topic of this study and read the references in detail to ensure that all relevant information was included (see Supplementary material).

Two independent evaluators, ZY and XK, employed Endnote 20 to screen and review the literature, while author WS resolved any disputes arising from the literature screening process. The literature was initially screened based on titles and abstracts, after which eligible full texts of the studies were obtained for secondary screening.

We searched only evidence from studies published in peer-reviewed journals, and the PICOS principles were used to determine this study’s inclusion and exclusion criteria; manuscripts published only online were also included in our review. The inclusion criteria are as follows: (1) All RCTs (parallel or crossover design) must involve participants aged over 18 years who meet the diagnostic criteria for Parkinson’s disease or Alzheimer’s disease. (2) The study arms must include multiple number of rTMS pulses or compare the effects of a specific number of rTMS pulses with a placebo on outcomes. (3) The study must provide clear parameters regarding total number rTMS of pulses, stimulation sessions, daily number rTMS of pulses, and other stimulation parameters. (4) The study must evaluate improvements in motor or non-motor symptoms of Parkinson’s disease or Alzheimer’s disease.

The exclusion criteria encompassed duplicate studies, animal research, review articles, conference proceedings, non-English publications, and non-randomized controlled trials. Furthermore, this study specifically excluded Lewy body dementia (LBD) and Parkinson’s disease dementia (PDD). Additionally, intervention groups employing more than two combined therapeutic approaches were not included in the final analysis.

We consider the Unified Parkinson’s Disease Rating Scale III (UPDRS III) and Alzheimer’s Disease Assessment Scale-Cognition (ADAS-Cog) as the primary outcomes for Parkinson’s disease (PD) and Alzheimer’s disease (AD), respectively. Unified Parkinson’s Disease Rating Scale (UPDRS), Timed Up and Go Test (TUG), Freezing of Gait Questionnaire (FOGQ), Beck Depression Inventory (BDI), Montreal Cognitive Assessment (MoCA), Mini-Mental State Examination (MMSE), Hamilton Anxiety Rating Scale (HAMA), Hamilton Depression Rating Scale (HAMD), Clinical Dementia Rating (CDR), and Geriatric Depression Scale (GDS) are assessed as secondary outcomes to evaluate the motor or non-motor symptoms of PD or AD.

Two independent reviewers (ZY and XK) extracted data from eligible articles, including publication year, data source, gender, age, education level, treatment course, stimulation target, stimulation frequency, stimulation intensity, number of pulses, and the mean difference in the motor or non-motor symptoms performance of participants before and after the intervention along with its corresponding SD or SEM. If the study used different data forms, such as quartiles or confidence intervals, the data were converted according to the Cochrane Handbook (Cumpston et al., 2019). For studies that reported effect estimates graphically, a web plot digitizer¹ was used to estimate the effect sizes from the graphs. When a study outcome was evaluated at multiple time points, we selected outcome data immediately after treatment. Data were cross-checked to minimize potential errors, and disagreements were resolved through discussion with the corresponding author (WS).

Two reviewers, ZY and XK, utilized the Cochrane Risk of Bias Assessment Tool (Higgins et al., 2011) to assess the methodological quality of the included studies. The risk tool incorporates seven critical sources of bias, including selective bias, implementation bias, measurement bias, follow-up bias, reporting bias, and other biases. Each article was categorized as “low risk,” “high risk,” or “unclear risk” for each type of bias. We utilized the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach to evaluate the quality of evidence for both primary and secondary outcomes. Following the GRADE handbook(Guyatt et al., 2011), we conducted our quality assessment. The GRADEpro Guideline Development Tool (GDT) was employed to produce the results. In cases of disagreement, the corresponding author arbitrated the issues.

We performed all statistical calculations using R software (version 4.4.2). The standardized mean difference (SMD) and its 95% confidence interval (CI) were used as effect sizes for continuous variables. A dose-response meta-analysis was conducted using the doresmeta package developed by Crippa et al. (2019), fitting a one-stage restricted cubic spline (RCS) to evaluate the dose-response relationship between the total number of rTMS pulses and the improvement in motor and non-motor symptoms of PD and AD. Three fixed percentile knots (5, 50 and 90%) were set according to Orsini et al. (2012) recommendations. Using Crippa’s method (Crippa and Orsini, 2016), we calculated the impact of every additional 5000 total pulses on participants’ motor and non-motor symptoms. Heterogeneity was quantified using I² and p-values. When the number of included studies was ≥ 10, Egger’s test and funnel plots were used to detect publication bias. A leave-one-out method was employed for sensitivity analysis to determine the robustness and reliability of the pooled results. P < 0.05 was assessed as statistically significant.

According to the previously established retrieval strategy, a total of 3,846 articles were retrieved (PubMed = 2,297, Embase = 146, Cochrane = 635, OVID Medline = 78, Web of Science = 690). After removing duplicate studies and strictly applying the inclusion and exclusion criteria, 51 studies meeting the required criteria were included, involving 1,938 subjects. Among these, 32 studies (Aftanas et al., 2018; Barboza et al., 2024; Benninger et al., 2011; Benninger et al., 2012; Boggio et al., 2005; Brys et al., 2016; Cardoso et al., 2008; Chang et al., 2016; Cohen et al., 2018; Grobe-Einsler et al., 2024; Huang et al., 2023; Ji et al., 2021; Jiang et al., 2023; Khedr et al., 2024; Khedr et al., 2003; Khedr et al., 2019; Kim et al., 2015; Lench et al., 2021; Li et al., 2020; Makkos et al., 2016; Maruo et al., 2013; Mi et al., 2019; Mitsui et al., 2022; Pal et al., 2010; Romero et al., 2024; Shimamoto et al., 2001; Shin et al., 2016; Song et al., 2024; Spagnolo et al., 2020; Wang et al., 2024; Wu et al., 2024; Zhuang et al., 2020) (63%) focused on PD, comprising 1,164 subjects, and 19 studies (Bagattini et al., 2020; Chen et al., 2023; Cotelli et al., 2011; Hoy et al., 2023; Jia et al., 2021; Jung et al., 2024; Koch et al., 2022; Lee et al., 2016; Leocani et al., 2020; Li et al., 2021; Padala et al., 2020; Saitoh et al., 2022; Vecchio et al., 2022; Wei et al., 2022; Wu et al., 2022; Wu et al., 2015; Yao et al., 2022; Zhao et al., 2017; Zhou et al., 2022) (37%) addressed AD, including 774 subjects. The detailed screening process and reasons for exclusion are presented in Figure 1 and Supplementary material.

In the included studies, 49 adopted a parallel design (96%), while 2 employed a crossover design (4%). There were 842 female subjects (49%), and 6 studies did not report the gender ratio of subjects (203 subjects, 10%). 28 studies utilized a figure-of-8 coil (55%), 13 studies used other types of coils (25%), and 10 studies did not report the type of coil used (20%). The average total number of pulses was 23,300, with a range from 500 to 80,000. The average number of sessions was 14, with a range of 3 to 32. 28 studies (55%) involved single-target stimulation, while 23 studies (45%) involved multi-target stimulation. 5 studies reported stimulation frequencies of ≤ 1 Hz (10%), 28 studies had frequencies of 1–10 Hz (55%), and 18 studies reported frequencies > 10 Hz (35%). Detailed information on all included studies is reported in Supplementary Table 2.

The included 22 articles provided detailed descriptions of the randomization procedures. In 24 studies, only the randomization methods were reported, resulting in an assessment of “unknown risk.” Twenty studies employed appropriate methods to conceal allocation, such as using opaque envelopes. However, five studies did not clearly specify their allocation methods, which were considered “high risk.” Most studies conducted a sham rTMS procedure primarily using fake coils or adjusting coil positions, which ensured that the sounds heard by participants and the visual appearance of the coils were consistent with actual stimulation. As a result, 25 studies were assessed as having a “low risk” concerning the blinding of participants. 8 studies did not clearly describe the blinding procedures for assessors and were thus deemed to be at “high risk.” Specific sources of bias can be found in Figure 2 and Supplementary Table 1.

A total of 24 studies were summarized on the improvement of UPDRS III scores in PD subjects through rTMS, involving 891 participants. The combined results showed that rTMS significantly reduced UPDRS III scores in PD subjects (SMD: −0.66, 95% CI [−0.91, −0.41], p < 0.01), with high heterogeneity among studies (I² = 66%, p < 0.01) (Table 1 and Figure 3). We conducted a re-analysis based on differences in stimulation frequency, and the results showed: for stimulation frequency ≤ 1 Hz, the combined result was (SMD: −0.85, 95% CI [−1.53, −0.16], p = 0.01), with heterogeneity (I² = 72%, p < 0.01); for 1–10 Hz (SMD: −0.68, 95% CI [−1.10, −0.26], p = 0.02), heterogeneity (I² = 72%, p < 0.01); for > 10 Hz (SMD: −0.58, 95% CI [−0.98, −0.18], p = 0.04), with heterogeneity (I² = 52%, p = 0.07); by stimulation target: Single Target Point (SMD: −0.80, 95% CI [−1.14, −0.47], p < 0.001), with heterogeneity (I² = 58%, p < 0.01); Multiple Target Points (SMD: −0.55, 95% CI [−0.94, −0.15], p = 0.004), with heterogeneity (I² = 72%, p < 0.01). Specific details can be found in Table 1 and Supplementary Figures 1–5. Using the leave-one-out method for sensitivity analysis, the results showed that the effect size and heterogeneity did not change significantly, indicating that the combined results are robust (see Supplementary Figures 22).

rTMS secondary outcomes in PD subjects covered various areas, including motor, emotion, and cognition, and showed significant improvements in multiple scores such as TUG, FOGQ, HAMD, and BDI. TUG (SMD: −0.41, 95% CI [−0.61, −0.21], p < 0.0001), heterogeneity (I² = 36%, p = 0.12); FOGQ (SMD: −0.62, 95% CI [−0.92, −0.31], p < 0.0001), heterogeneity (I² = 0%, p = 0.63); HAMD (SMD: −1.40, 95% CI [−2.71, −0.10], p = 0.04), heterogeneity (I² = 90%, p < 0.01); BDI (SMD: −0.61, 95% CI [−0.88, −0.33], p < 0.0001), heterogeneity (I² = 39%, p = 0.13). The improvement of UPDRS total score, HAMA, MoCA, and MMSE in subjects with PD following rTMS treatment does not seem to be significant. Specific details can be found in Table 1 and Supplementary Figures 6–13. The sensitivity analysis indicated that the effect sizes and heterogeneity for FOGQ, BDI, HAMD, and HAMA showed no significant changes, suggesting that the combined results are relatively robust. However, after excluding the studies by Aftanas LI, Mitsui T, and Makkos A, there were significant fluctuations in the effect sizes or heterogeneity of the combined results for UPDRS total score, TUG, and MoCA (see Supplementary Figures 23–30).

A total of 12 studies were summarized on the improvement of ADAS-Cog scores in AD subjects through rTMS, involving 498 participants. The combined results showed that rTMS significantly reduced ADAS-Cog scores in AD subjects (SMD: −0.20, 95% CI [−0.38, −0.02], p = 0.03), with heterogeneity among studies (I² = 28%, p = 0.17) (Table 1 and Figure 4). We conducted a re-analysis based on differences in stimulation frequency, and the results showed: for stimulation frequency ≤ 10 Hz, the combined result was (SMD: 0, 95% CI [−0.25, 0.25], p = 0.99), with heterogeneity (I² = 0%, p = 0.43); for > 10 Hz (SMD: −0.41, 95% CI [−0.67, −0.15], p = 0.002), with heterogeneity (I² = 9%, p = 0.36); by stimulation target: Single Target Point (SMD: −0.22, 95% CI [−0.67, −0.23], p = 0.35), with heterogeneity (I² = 71%, p < 0.01); Multiple Target Points (SMD: −0.55, 95% CI [−0.46, −0.07], p = 0.16), with heterogeneity (I² = 0%, p = 0.96) (Specific details can be found in Table 1 and Supplementary Figures 14–17). After excluding the study by Zhou X, there were fluctuations in the heterogeneity (see Supplementary Figures 31).

The combined results indicate that rTMS can significantly improve secondary outcome measures such as MMSE and MoCA in subjects with AD. MMSE (SMD: 0.43, 95% CI [0.02, 0.84], p = 0.03), heterogeneity (I² = 80%, p < 0.01); MoCA (SMD: 0.38, 95% CI [0.08, 0.67], p = 0.01), heterogeneity (I² = 0%, p = 0.96); The improvement of CDR and GDS in subjects with AD following rTMS treatment does not seem to be significant (Specific details can be found in Table 1 and Supplementary Figures 18–21). The sensitivity analysis indicated that the effect sizes and heterogeneity for MoCA, CDR, and GDS showed no significant changes, suggesting that the combined results are relatively robust. However, after excluding the studies by Li X, there was significant fluctuations in the heterogeneity of the combined results for MMSE (see Supplementary Figures 32–35).

Based on RCS, a non-linear dose-response meta-analysis was conducted on the total number of pulses and the improvement of motor and non-motor symptoms in subjects with PD and AD. The results showed that the total number of rTMS pulses had significant bell-shaped curves for PD subjects in TUG (χ² = 6.87, df = 2, p = 0.03), FOGQ (χ² = 15.17, df = 2, p = 0.001), BDI (χ² = 14.33, df = 2, p = 0.001), HAMD (χ² = 12.63, df = 2, p = 0.001), and HAMA (χ² = 6.06, df = 2, p = 0.04). Specifically, TUG achieved maximum therapeutic effect after receiving 8731 pulses (SMD: −0.41, 95% CI [−0.71, −0.10]); FOGQ achieved maximum effect after 8,763 pulses (SMD: −0.74, 95% CI [−1.14, −0.35]); BDI reached its peak effect after 11,535 pulses (SMD: −0.95, 95% CI [−1.47, −0.43]); HAMD after 7,705 pulses (SMD: −1.28, 95% CI [−2.83, 0.27]); and HAMA after 6,518 pulses (SMD: −0.78, 95% CI [−1.94, 0.37]). This indicates that an increase in total pulse rTMS correlates with improvements in motor or non-motor symptoms after receiving rTMS in the short term. A significant decreasing curve was observed for the UPDRS total score (χ² = 12.14, df = 2, p = 0.002). Furthermore, for UPDRS III (χ² = 27.58, df = 2, p = 1.02), MMSE (χ² = 2.54, df = 2, p = 0.28), and MoCA (χ² = 2.72, df = 2, p = 0.26), relatively flat bell-shaped curves were observed, indicating that the increase in total pulse numbers had no significant effect on the improvement of these outcomes. For AD subjects, the total number of rTMS pulses showed significant bell-shaped curves for MMSE (χ² = 8.76, df = 2, p = 0.01) and MoCA (χ² = 6.79, df = 2, p = 0.03), meaning that MMSE reached its maximum therapeutic effect after receiving 28793 pulses [SMD: 0.45, 95% CI (0.15, 0.74)] and MoCA after 25201 pulses [SMD: 0.42, 95% CI (0.07, 0.77)]. Additionally, no significant effects were observed for ADAS-Cog (χ² = 1.93, df = 2, p = 0.38), CDR (χ² = 0.28, df = 2, p = 0.87), and GDS (χ² = 0.14, df = 2, p = 0.93), as they also displayed relatively flat bell-shaped curves(Specific details can be found in Figures 5, 6, and Supplementary Figures 36–44). In addition, we also presented the effect size and 95% confidence interval corresponding to the improvement in symptoms for every additional 5,000 total number of pulses (see Table 2).

We conducted Egger’s test and generated a funnel plot for studies with more than 10 entries (see Supplementary Figures 45–48). The results showed: UPDRS III (t = −1.46, df = 22, p = 0.1581), UPDRS total score (t = −1.73, df = 11, p = 0.1111), TUG (t = 2.08, df = 8, p = 0.0713), ADAS-Cog (t = −0.66, df = 10, p = 0.5217), MMSE (t = 0.10, df = 12, p = 0.9222). These results suggest that there may not be significant publication bias.

We conducted an evaluation of the quality of evidence for the primary and secondary outcomes related to AD and PD. The results showed that there is high confidence in the evidence for the UPDRS total score, as well as the MMSE and MoCA scores of AD subjects. For UPDRS III, TUG, FOGQ, HAMD, BDI, HAMA, MoCA, ADAS-Cog, CDR, and GDS, the evidence levels maintain moderate confidence. However, there is limited confidence in the evidence level for the MMSE score results of PD subjects (Specific details can be found in Table 1 and Supplementary Table 3).