The Cochrane Library, PubMed, Web of Science, Embass, CNKI, Wanfang Data databases were searched by two independent researchers using subject terms combined with free words from the time of their creation to December 2023. In addition, supplementary manual searches of references and grey literature included in the studies were conducted to ensure the comprehensiveness of the search. The search terms included: Parkinson’s diseases, transcranial magnetic stimulation, high frequency, low frequency, theta burst stimulation, cognitive impairment, randomized controlled studies.

A randomized controlled trial of the application of TMS in Parkinsonian patients. The study subjects were patients with PD patients with cognitive dysfunction, all of them were > 18 years old, and they were conscious and had stable vital signs. The control group was CRT, CRT mainly consists of conventional drug therapy, occupational therapy and physical therapy commonly used in clinical practice. The experimental group was treated with TMS on the basis of CRT. TMS modes included HF-rTMS, LF-rTMS, and iTBS. The primary outcomes were the Montreal Cognitive Assessment (MoCA) and the Mini-Mental State Examination (MMSE), and the secondary outcomes were the Unified Parkinson Disease Rating Scale part I and II (UPDRS I, II).

Non-English and Chinese literature, duplicate publications, studies with non- randomized controlled trials (RCTs) designs, no baseline conditions, incomplete or unextractable raw data and fruitless attempts to contact the authors, conference papers, master’s and doctoral dissertations, and literature without specified outcome indicators.

The retrieved literature was imported into EndNote software and duplicates were excluded. 2 investigators independently screened and extracted data according to inclusion and exclusion criteria and cross-checked the literature, consulting a third investigator in case of disagreement. The data extraction included basic information, baseline conditions, interventions, and outcome indicators. All outcomes were included in the pre-treatment and post-treatment differences.

The quality of the included literature was assessed by two investigators using the Cochrane Handbook of Systematic Reviews 5.1.0. The assessment included (1) Selective bias: assessed by the appropriateness of the random sequence generation method and the perfection of the allocation scheme concealment; (2) Implementation bias: blinding of subjects or researchers and outcome assessors, assessed by the perfection of the blinding method; (3) Follow-up bias: assessed by whether the outcome data were complete and whether missing data were handled appropriately; (4) Reporting bias: assessed by whether the results were selectively reported; (5) Other bias. The risk of bias was evaluated as high, low, or unclear.

The STATA 17.0 software was used to conduct the reticulated Meta-analysis, and the included outcome indicators were continuous variables, and the effect size was expressed by mean difference (MD), and 95% confidence interval (CI) was calculated as the effect size, while the I² statistic was combined to quantitatively determine the size of heterogeneity. When p ≥ 0.05, I²≦50%, it was considered that there was no heterogeneity among studies, and when p < 0.05, I² > 50%, it was considered that there was significant heterogeneity among studies. For the heterogeneous results, the sources of heterogeneity could be explored by subgroup analysis, sensitivity analysis or by conducting qualitative descriptions. Stata was used to plot the network evidence of each outcome indicator, each dot represented different interventions, the dots represented number of cases for that intervention, the line connected between the two dots is the evidence for the existence of direct comparisons between the two interventions, and the thickness of the solid line is the amount of evidence for direct comparisons. Inconsistency tests were performed using nodal analysis, and if p > 0.05, the difference between direct and indirect comparisons was not statistically significant and the two results were consistent, and the analysis was performed using the consistency model, and vice versa. This study used Stata to detect the inconsistency factor and 95% CI in the closed loop, and the inconsistency factor (IF) was small when the 95% CI contained 0, suggesting that the evidence for direct versus indirect comparisons was very consistent. The ranking status of each outcome indicator was expressed using a cumulative probability plot, with a larger area under the curve of the cumulative probability plot representing better efficacy.

Six databases were searched according to the search formula, and a total of 661 papers were obtained from the preliminary search, including 64 papers from Cochrane, 150 papers from PubMed, 229 papers from Web of Science, 107 papers from Embass, 50 papers from CNKI and 61 papers from Wanfang Data. Twenty-two papers (Su et al., 2012; Tang et al., 2015; Yu et al., 2017; Dong and Yang, 2018; Zhu et al., 2018; Li et al., 2019; Wu et al., 2019; Chi, 2020; Guo, 2020; Jiang L. et al., 2020; Khedr et al., 2020; Lang et al., 2020; Li, 2020; Liu and Lu, 2020; Zhang et al., 2020; Zhuang et al., 2020; He et al., 2021; Liao et al., 2021; Cheng et al., 2022; Gao et al., 2022; Huang et al., 2023; Yang, 2023) were finally included after eliminating duplicates and reading through the titles, abstracts, and full text and quality assessment. The detailed screening process and the reasons for exclusion are shown in Figure. The detailed screening process and reasons for exclusion are shown in Figure 1, and the basic characteristics of the included literature are shown in Tables 1, 2.

The Cochrane Handbook of Systematic Reviews 5.1.0 was used to assess the quality of the included literature. Of the 22 included papers, 22 referred to “randomized groups” and 20 (Tang et al., 2015; Yu et al., 2017; Dong and Yang, 2018; Zhu et al., 2018; Li et al., 2019; Wu et al., 2019; Chi, 2020; Guo, 2020; Khedr et al., 2020; Lang et al., 2020; Li, 2020; Liu and Lu, 2020; Zhang et al., 2020; Zhuang et al., 2020; He et al., 2021; Liao et al., 2021; Cheng et al., 2022; Gao et al., 2022; Huang et al., 2023; Yang, 2023) described the specific randomization method. Five papers (Khedr et al., 2020; Lang et al., 2020; He et al., 2021; Liao et al., 2021; Cheng et al., 2022) had an “opaque envelope” allocation concealment. Thirteen papers (Tang et al., 2015; Dong and Yang, 2018; Zhu et al., 2018; Li et al., 2019; Chi, 2020; Jiang L. et al., 2020; Khedr et al., 2020; Lang et al., 2020; Li, 2020; Liu and Lu, 2020; He et al., 2021; Cheng et al., 2022; Huang et al., 2023) were blinded to study subjects or interventionists. Three papers (Khedr et al., 2020; Lang et al., 2020; He et al., 2021) were blinded to outcome measures. Twenty-two of the literature had complete outcome data, no selective reporting of outcomes, and other risks of bias were unclear, as detailed in Figure 2.

This network meta-analysis systematically evaluated the difference in efficacy between four stimulation modalities and conventional rehabilitation in improving cognitive dysfunction in patients with PD patients by means of two primary outcomes and two secondary outcomes. A total of 22 RCTs with 1,473 patients were included. The results of the primary outcomes were as follows: HF-rTMS significantly improved MoCA and MMSE scores, and LF-rTMS significantly improved MMSE scores. This suggests that HF-rTMS may show better efficacy in improving cognitive function. Finally, we performed a ranking of the strengths and weaknesses of the intervention modalities, and consistent with the results of the network meta-analysis, HF-rTMS was considered to have better efficacy in both primary outcomes. Among the secondary outcomes, LF-rTMS significantly improved UPDRS I scores and HF-rTMS significantly improved UPDRS II scores. In the MoCA subgroup analyses, it was shown that HF-rTMS significantly improved Visuospatial, Language, Abstraction, and Delayed recall scores, and iTBS significantly improved Language and Delayed recall scores. In conclusion, HF-rTMS has more significant advantages in improving cognitive function in PD patients.

Most of the current research in TMS in PD patients is in the area of confirming effectiveness. A recent meta-analysis (Deng et al., 2022) showed that TMS significantly improves cognitive function in patients with PD patients, and due to methodological limitations inherent in traditional meta-analyses, the authors of the review were only able to perform pairwise comparisons without taking into account the existing evidence network. In addition, to avoid multiple testing, the authors had to combine treatment combinations with different TMS stimulation modalities into a single TMS stimulation group and did not include iTBS, thus potentially masking differences between different TMS modalities. More than 90% of the rTMS included in this study were HF-rTMS, which is somewhat similar to our final conclusion.

In the present study, we further explored the duration of the efficacy of TMS in patients with PD patients. The results showed that the effect of TMS treatment in MoCA scores became insignificant in the first month of follow-up. Whereas in the MMSE scores there was still a significant efficacy at 2 months after the follow-up. The reason for the different effects of the two scales may be that the MoCA includes the assessment of visuospatial and executive functioning in addition to general cognitive functioning, which makes it relatively difficult to improve the MoCA. Through sensitivity analysis, Yang (2023) study was found to be more influential, and after exclusion, the effect of MoCA was lost in the first month, whereas the effect of TMS in MMSE lasted for 1 month, suggesting that the improvement of cognitive functioning in patients with PD patients can last at least 1 month, which is important for guiding the course and period of TMS treatment in clinical practice.

In addition to the stimulation mode, stimulation intensity, stimulation location, and number of stimulation pulses are also important parameters for the application of TMS. A meta-analysis of stimulation parameters when TMS was applied to cognitive function in PD patients was conducted by Jiang Y. et al. (2020). The study concluded that high-frequency TMS stimulation of the DLPFC was able to significantly improve Parkinsonian executive function, but the study did not specifically analyze whether there was a difference in stimulation of the different sides of the DLPFC. In the present study, we grouped the other three factors affecting TMS with the primary outcome metrics. Among the stimulation sites, stimulation of the DDLPFC at 100 intensities showed superior results in terms of MoCA and MMSE, similar to the conclusions of Jiang Y. et al. (2020). No more reliable conclusions were obtained in terms of the number of pulses.

Compared with existing meta-analyses of TMS to improve cognitive function in PD patients, this study refined the classification based on the effectiveness of TMS and explored the comparative efficacy of different TMS modalities on Parkinsonian cognitive function. Moreover, the stimulation parameters were analyzed and compared, providing some insights for future research on TMS parameters. In addition, this paper analyzes and compares the intervention time of TMS, and concludes that TMS has a 1-month improvement effect on cognitive function in PD patients, which provides an important reference significance for the use of TMS treatment cycle in the clinic. It is worth mentioning that in this study, the control group was subdivided into a sham-stimulation group and a rehabilitation training group, and it was found that the treatment effect of the sham-stimulation group was better than that of the conventional rehabilitation group, which may be due to the fact that the sham-stimulation can also produce a sizable placebo effect by inducing the release of dopamine from basal ganglia region, which has already been demonstrated in certain imaging, so the effect of the sham-stimulation should not be neglected in clinical practice (Kim et al., 2008). Based on the results of this study, in the future, clinical research can refine the stimulation mode of TMS around the stimulation parameters to find the optimal stimulation parameter pattern, and can be carried out for the 1-month cycle of treatment to prove whether it is scientifically effective to carry out clinical research. Our network meta-analysis cross-sectional comparison of different TMS stimulation modalities provides valuable new insights into the use of TMS in patients with cognitive impairment in PD patients. We suggest that this analysis be considered as a complement to previous systematic reviews on this topic (Goodwill et al., 2017; Lawrence et al., 2017; Jiang Y. et al., 2020; Deng et al., 2022; He et al., 2022).

Current research suggests that altered levels of dopamine (DA) in the midbrain-limbic and midbrain-cortical systems may be associated with non-motor symptoms of PD patients (Kang et al., 2010). The feasibility of TMS for the treatment of PD patients is based on the ability of TMS to induce a rise in intracranial dopamine release, regulate cortical excitability, improve local blood flow in the brain, and regulate the secretion of brain-derived neurotrophic factor (BDNF) in the brain (Helmich et al., 2006; Zhu et al., 2018). Brain-derived neurotrophic factor is a special neurotrophic protein widely expressed in the nervous system. Compared with other parts of the brain, the hippocampus and the cortex contain significantly higher levels of BDNF, and TMS can increase the level of BDNF directly in the brain to improve the cognitive function of the brain. Synapse is a special bridge for nerve cells to connect, TMS can increase intracranial BDNF and change the synaptic plasticity at the same time, and its enhanced connectivity can improve the function of dopaminergic neurons, thus improving the cognitive level of PD patients (Ye and Feng, 2016; Zhao et al., 2019). Acetylcholine (Ach) is an excitatory neurotransmitter that plays an important role in learning and memory, and the occurrence of Parkinson’s is closely related to the reduction of Ach, and the main mechanism lies in the impairment of the cholinergic pathway that leads to the reduction of Ach activity and triggers Parkinson’s (Bonnì et al., 2017; Chi, 2020). Some studies have shown (Zhang et al., 2018) that TMS can increase the activity of acetylcholinesterase and choline acetyltransferase and increase the density of cholinergic neurons, and it is speculated that the mechanism by which TMS improves cognitive and learning functions may be related to the restoration of the activity of the cornu ammonis 1 cholinergic system.

MoCA and MMSE are screening scales for cognitive functions such as attention and memory (Arevalo-Rodriguez et al., 2015; Kang et al., 2018). The frontal lobe is an important regulatory area for cognitive function and is the most common stimulation target for TMS application in patients with cognitive impairment. The results of this study suggest that when the frontal lobe is stimulated, the brain is able to mediate the executive function and attention of the patient, thus improving the patient’s memory, visuospatial processing ability (Burton and Tyson, 2015). HF-rTMS stimulation is able to repeat and regular high-intensity magnetic stimulation of the frontal lobe of the patient’s brain, forming a continuous effect, with specific effects on the frontal cortex, which is more conducive to cognitive improvement than other intervention modes with direct stimulation and increased number of neural connections (Lu et al., 2015). The basal forebrain is an important region of the brain that is rich in cholinergic neurons that release Ach to areas of the brain that are closely related to cognitive processes such as learning, memory, attention, and executive function (Gratwicke et al., 2015). It has been reported (McClelland et al., 2016) that repetitive transcranial magnetic stimulation inhibits activity in other regions of the brain, promotes localized activity, enhances the activity of cholinergic neurons in the basal forebrain, and improves the release of neurotransmitter secretion, such as Ach, thereby promoting the recovery of cognitive function. In addition, stimulation of the DLPFC in the brain can lead to enhanced neural connectivity between the DLPFC and the basal forebrain, prompting changes in the activity patterns of regions such as the basal forebrain, which may include increased activation of cholinergic neurons and enhanced synaptic plasticity of neurons, leading to enhanced release of Ach, and consequently, improved cognitive function (Gratwicke et al., 2015). The frontal lobe was used as the stimulation target in most of the present study, and the high stimulation frequency of HF-rTMS, between 5 and 25, could be an important reason for the more significant effect of HF-rTMS.

UPDRS I and UPDRS II are special assessment scales for evaluating the psychoemotional, behavioral cognitive and daily life ability of PD patients, respectively. TMS treatment can regulate the excitability of the cerebral pallidum and striatal loop, improve the symptoms of cognitive, sleep and motor disorders, enhance the quality of life of the patients, and reduce the negative emotions. Daily living ability is closely related to cognitive function. The ability of daily life includes the participation of several cognitive functions such as executive function, reasoning integration of tasks, etc. TMS can achieve the reconstruction of regional cortical function and the remodeling of brain network system through the cortico-cortical pathway, and promote the effective connection of brain network nerves to enhance the recovery of cognitive function (Hallett et al., 2007). With the improvement of cognitive function, the ability of PD patients in daily life has been improved accordingly.

There are still shortcomings in this study: The majority of the current included literature is a comparison between different intervention models relying on conventional rehabilitation treatment, and lacks two-by-two comparisons between different intervention models, hence the lack of a closed-loop mesh relationship. Due to the limited number of included studies, the subgroup analyses were not categorized by stimulation modality, resulting in potentially biased results. The frequency of high-intensity stimulation spanned from 5 to 25 sessions, making it difficult to determine the frequency of HF-rTMS that is most suitable for patients with PD patients. The duration of intervention varied across studies, and it was not possible to determine the difference between the duration of intervention and efficacy, more high-quality RCTs are still needed in the future to help us increase the sample size and improve the effect estimates.