The effect of repetitive transcranial magnetic stimulation on immediate and long-term cognitive functions in Alzheimer’s dementia and mild cognitive impairment: a meta-analysis Authors: Bojun Wang, Ya Wang, Fenglei Yang, Fang Han, Kuncheng Li, Kai Sun, Peipeng Liang Published: 2025 ABSTRACT AbstractBackgroundAlzheimer’s dementia (AD) and mild cognitive impairment (MCI) are characterized by progressive cognitive decline. Repetitive transcranial magnetic stimulation (rTMS), a non-invasive brain stimulation technology, has been widely used to improve cognition in AD and MCI. However, the impact of rTMS on immediate and long-term cognitive functions in AD and MCI, as well as the optimal stimulating parameters, need further clarification.MethodThirty-one randomized controlled trials were included to examine the effects of rTMS on immediate cognition (post-treatment cognition), while nine trials were specifically used to assess its impact on long-term cognition (follow-up cognition). All participants were tested on at least one of the neuropsychological scales of the Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA) and Alzheimer’s disease Assessment Scale-Cognitive Section (ADAS-Cog) to evaluate global cognitive outcomes of rTMS. The study followed the guidelines for meta-analysis of intervention studies and assessed the risk of bias of the included studies. Moderator analysis was used to examine factors that might affect treatment effect. Sensitivity analysis was used to evaluate stability of results. Meta-regression analysis was used to assess the interpretability of heterogeneity. Begg’s and Egger’s test and funnel plot were applied to assess publication bias.ResultsThe results demonstrated that rTMS significantly improved immediate cognition (SMD = 0.93, 95% CI = 0.64 to 1.22, p < 0.001) and long-term cognition (SMD = 0.42, 95% CI = 0.13 to 0.70, p = 0.004) across all patients (including AD and MCI). Moderator analysis revealed that targeting dorsolateral prefrontal cortex (DLPFC), using intensities of 80% or less of the motor threshold, total time of 800 min or more and stimulation sessions greater than 20 significantly enhanced immediate cognitive performance than their counterparts. In contrast, stimulation frequency, duration/session and patient type showed no significant impact on immediate cognition. For long-term cognition, differences in stimulation protocols had no influence on treatment effect for all patients.ConclusionrTMS had a positive effect on immediate and long-term cognition in AD and MCI, stimulation region, intensity, total time and number of sessions significantly improve immediate cognition. These findings suggest that rTMS has potential as a promising adjunctive treatment for AD and MCI.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12984-025-01777-8.Keywords: Repetitive transcranial magnetic stimulation, Alzheimer’s dementia, Mild cognitive impairment, Cognitive function, Meta-analysis Introduction Alzheimer’s dementia (AD) is a progressive degenerative disease that affects the central nervous system, usually characterized by progressive memory loss followed by loss of more extensive cognitive functions [1]. Mild cognitive impairment (MCI) is the prodromal stage of AD, and 15%-20% of MCI patients may develop AD within one year [2]. Current drug treatments for AD demonstrate limited efficacy, which only temporarily alleviates symptoms of cognitive decline without reversing its progression [3]. In addition, drug treatments for AD may also face the risk of side effects [4, 5]. Current treatments for AD only slow disease progression, developing new interventions that target cognitive function remains an urgent priority in order to slow disease progression. Repetitive transcranial magnetic stimulation (rTMS) is a non-invasive neuromodulation technology designed to modulate neural activity to achieve therapeutic effects. This technique employs magnetic pulses to modulate neural activity in targeted brain regions, exerting excitatory or inhibitory effects that influence emotion and cognition [6]. It has also shown therapeutic potential in neurodegenerative disorders, including AD and MCI. After penetrating the skull, the pulsed magnetic field of rTMS induces eddy currents in the brain tissue. The distribution and density of these currents are modulated by the heterogeneity of biological tissues. Therefore, both stimulation parameters and target region selection influence the characteristics of the induced currents, ultimately affecting the efficiency of neuronal depolarization and plasticity changes [7]. Specific stimulating protocols may have profound effects on neural functions [8]. For example, high-frequency rTMS stimulation of the left dorsolateral prefrontal cortex (DLPFC) in patients with major depression may increase cortical excitability by counteracting pathological hypoactivity [9, 10]. Another stimulation method is low-frequency rTMS to the right DLPFC, which appears to reduce cortical activity [11]. The variability of rTMS effects with protocol adjustments has been consistently observed in neurodegenerative diseases [12]. Thus, identifying optimal rTMS parameters is critical for maximizing therapeutic efficacy in neurological disorders. Several recent meta-analyses have examined the efficacy of rTMS or the combination of rTMS and other methods in the treatment of cognitive functions or psychiatric symptoms in AD patients [13–15]. However, few meta-analyses compared rTMS stimulation protocols [16, 17], there is still a lack of direct comparisons between different stimulation parameters and the optimal parameters to improve cognitive functions in AD and MCI remains unknown. This meta-analysis of randomized controlled trials evaluates the effects of rTMS on immediate and long-term cognition in AD and MCI. Given the heterogeneity in rTMS parameter selection across prior studies, this meta-analysis further aimed to examine the effect of different stimulation parameters of rTMS on the cognitive intervention effects in patients with AD and MCI, to help identify optimal treatment parameters. Method literature search This meta-analysis was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines [18]. We searched Web of Science, PubMed, Cochrane Library and Embase databases from their inception to June 30, 2025, using customized syntax. Databases were searched using combinations of the following keywords: [Alzheimer disease or Alzheimer’s disease or Alzheimer-type dementia or cognition(s) or cognitive function(s) or mild cognitive impairment or amnestic mild cognitive impairment or amnestic-type mild cognitive impairment] and [repetitive transcranial magnetic stimulation or rTMS]. Additionally, we also searched WANFANG and CNKI databases from their inception to June 30, 2025, using the same keywords but in Chinese. This study has been registered with the International Prospective Register of Systematic Reviews (PROSPERO) website (Registration Number: CRD42024548872). Inclusion and exclusion criteria Study inclusion criteria included the following: (1) subjects with AD were diagnosed according to recognized clinical criteria, such as the Chinese Classification of Mental Disorders-3 (CCMD-3), International Classification of Diseases-10 (ICD-10), Diagnostic and Statistical Manual of Mental Disorders IV/5 (DSM-IV/5), National Institute on Aging-Alzheimer’s Association (NIA-AA) guidelines, National Institute of Neurological and Communicative Disorders and Stroke-Alzheimer’s disease and Related Disorders Association (NINCDS-ADRDA) criteria, etc., subjects with AD-related mild cognitive impairment based on Petersen’s criteria or 2018 Guidelines for the Diagnosis and Treatment of Dementia and Cognitive Impairment in China. (2) the intervention of the experimental group was rTMS, and the control group received sham stimulation. (3) randomized control study based on patients, double-blind placebo-controlled studies excluding cross-over studies. (4) Cognitive function was assessed with the Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), or Alzheimer’s disease Assessment Scale-Cognitive Subscale (ADAS-Cog). Study exclusion criteria included the following: (1) Studies in languages other than English or Chinese. (2) Non-RCT study. (3) Patients comorbid with other neurodegenerative disease such as Parkinson’s disease. (4) Cognitive impairment caused by other disease. (5) Duplicated studies. (6) Studies lacking clear diagnostic or therapeutic criteria were excluded. (7) Studies with other interventions, such as theta-burst stimulation. (8) The studies with incomplete or unreported data, including basic information or the detailed intervention parameters. In cases of missing data, we contacted the authors of the original studies to request additional data. If no additional data was obtained, the study would be excluded. (9) Non-human studies. (10) Reviews, meta-analyses, letters, case report/series, conference abstracts, editorials, posters and commentaries. Study selection The literature was managed and reviewed using EndNote 19. The entire process of literature screening is presented in Fig. 1. The duplicated studies were first excluded, and then the titles and abstracts were read to exclude irrelevant or unqualified literature according to inclusion and exclusion criteria. Finally, the full texts of the remaining literature were read and screened according to inclusion and exclusion criteria independently by BJW and FLY. Discordance was resolved by discussion with the senior author (PPL). The screening of literature Data extraction A thirteen-item, standardized and pre-selected data extraction form was used to record data for analysis under the following headings: (1) First author, (2) Year of publication, (3) Type of patients, (4) Sample size, (5) Stimulation site, (6) Stimulation intensity, (7) Stimulation frequency, (8) Stimulation duration per session, (9) Stimulation total time, (10) Stimulation sessions, (11) Type of scales, (12) Mean value, (13) Standard deviation. The efficacy of treatment is presented in Supplementary materials for Table S1. Risk of bias of studies The risk of bias was assessed using the Newcastle-Ottawa Scale (NOS) for all studies included in the meta-analysis. Specifically, studies were assessed using a quality score in three domains as follows: selection of the study groups (maximum 4 points), comparability of the groups (maximum 2 points), and measurements of the outcomes (maximum 3 points). A higher score indicates better quality. An NOS score of 9 represents an excellent-quality (low risk of bias) study, a score of 8 and 7 indicates a fair-quality (intermediate risk of bias) study, and a score of ≤ 6 indicates a low-quality (high risk of bias) study [19]. This meta-analysis included 7 studies with high risk of bias, 15 studies with intermediate risk of bias, and 4 studies with low risk of bias. Data analyses We used Comprehensive Meta Analysis V3 software (https://zh.meta-analysis.com/) to compute the effect sizes (standard mean difference, SMD) and the 95% confidence intervals (CI). If a study had different cognitive measures from the same patient group, we obtained an averaged effect size across multiple effect sizes within that study. If two different stimulation parameters were adopted within a single study, then the data were included as two independent units in the meta-analysis. We used the Cochran Q test and I2 method to assess the heterogeneity between included studies. With high heterogeneity (I2 ≥ 50%, p value of Q test < 0.05), we applied the random effects model; with low heterogeneity (I2 < 50%, p value of Q test ≥ 0.05), we used the fixed effects model. In addition, moderator (subgroup) analysis was conducted to examine the source of heterogeneity. Our pre-specified categories for subgroup analyses included type of patient (AD vs. MCI), stimulation site (DLPFC vs. brain regions), stimulation intensity (> 90% motor threshold (MT) vs. ≤90%MT), stimulation frequency (< 20 Hz vs. ≥20 Hz), stimulation duration/session (< 30 min vs. ≥30 min), stimulation total time (> 800 min vs. ≤800 min), stimulation sessions (< 20 vs. ≥20), and type of scales (MMSE vs. MoCA vs. ADAS-Cog). Sensitivity analysis was used to evaluate stability. Short-term cognition/immediate cognition was defined as post-treatment cognition, that is, the cognitive state of the patient after the end of the treatment cycle. Long-term cognition is defined as follow-up cognition, which refers to the cognitive status after a certain period of time following the end of the periodic treatment course. To explore sources of heterogeneity, a meta-regression analysis was performed to evaluate the explanatory power of moderator variables. Moreover, Begg’s test, Egger’s test and funnel plot were applied to assess publication bias. Results A summary of the basic information of each study and the detailed intervention parameters are presented in Table 1. The risk of bias for each study is presented in Table 2. Table 1The basic information and the detailed intervention parameters of studiesFirst authorPublished yearType of patientsSample size (experiment/control)Stimulation siteStimulation intensityStimulation frequencyStimulation duration/sessionStimulation sessionsStimulation total timeAn [20]2014MCI20/18DLPFC80%MT10 Hz30 min20600 minDu [21]2022AD35/35DLPFC100%MT20 Hz25 min30750 minLi [22]2022MCI17/12DLPFC90%MT10 Hz14 min32 s10145.3 minLiu [23]2017AD12/10DLPFC100%MT10 Hz15 min10150 minRan [24]2021MCI17/12DLPFC90%MT10 Hz14 min36 s10146 minWang [25]2021MCI19/20DLPFC90%MT10 Hz14 min32 s10145.3 minYang [26]2014MCI18/15DLPFC80%MT20 Hz30 min401200 minZhou [27]2019MCI17/12DLPFC100%MT10 Hz14 min36 s10146 minLiu [28]2017 AAD60/60DLPFC80%MT5 Hz30 min601800 minLiu [28]2017BAD60/60DLPFC80%MT15 Hz30 min601800 minZhang [29]2021MCI20/20DLPFC80%MT10 Hz30 min401200 minCao [30]2019MCI13/9DLPFC80%MT10 Hz200 s1033.3 minLi [31]2014AD25/25DLPFC100%MT20 Hz30 min401200 minLi [32]2018MCI8/9DLPFC110%MT1 Hz30 min30900 minLong [33]2015MCI15/15DLPFC100%MT15 Hz21 min30 s10215 minWen [34]2018MCI23/22DLPFC80%MT10 Hz200 s2066.6 minZhu [35]2020MCI13/12DLPFC80%MT10 Hz200 s2066.6 minHan [36]2012MCI22/18DLPFC80%MT20 Hz30 min401200 minWu [37]2015AD27/27DLPFC80%MT20 Hz30 min201200 minJia [38]2021AD35/34Left parietal region100–110%MT10 Hz20 min10200 minLee [39]2016AD18/8DLPFC + Parietal somatosensory associated cortices + Broca’s and Wernicke’s areas90–110%MT10 Hz60 min301800 minHu [40]2022AD21/21AG90%MT40 Hz30 min12360 minYao [41]2022AD15/12CB90%MT5 Hz20 min20400 minWei [42]2022AD29/27Left parietal region100–110%MT10 Hz20 min10200 minLi [43]2021AD37/38DLPFC100%MT20 Hz20 min30600 minCotelli [44]2011AD5/5DLPFC100%MT20 Hz25 min20500 minLv [45]2024MCI39/21AG100%MT20 Hz20 min20400 minChen [46]2025MCI17/10AG100%MT20 Hz20 min20400 minJung [47]2024AD18/12Left parietal region100%MT20 Hz20 min20400 minZhang [48]2024 AAD48/51DLPFC90–100%MT1 Hz30 min10300 minZhang [48]2024BAD50/51DLPFC90–100%MT10 Hz30 min10300 minYan [49]2024 AAD18/18DLPFC110%MT5 Hz30 min10300 minYan [49]2024BAD20/18DLPFC110%MT10 Hz35 min10350 minMCI, Mild Cognitive Impairment; AD, Alzheimer’s dementia; DLPFC, Dorsolateral Prefrontal Cortex; MT, Motor Threshold; AG, Angular Gyrus; CB, Cerebellum The basic information and the detailed intervention parameters of studies MCI, Mild Cognitive Impairment; AD, Alzheimer’s dementia; DLPFC, Dorsolateral Prefrontal Cortex; MT, Motor Threshold; AG, Angular Gyrus; CB, Cerebellum Table 2Risk of bias assessment for all studies using the Newcastle-Ottawa scale (NOS)Study nameCriterion 1Criterion 2Criterion 3Criterion 4Criterion 5Criterion 6Criterion 7Criterion 8Criterion 9TotalAn et al., 2014 + + + + - - + + + 7 Du et al., 2022 + + + + + - + + + 8 Li et al., 2022 + + + + - - + + - 6 Liu et al., 2017 + + + + + + + + + 9 Ran et al., 2021 + + + + - - + + + 7 Wang et al., 2021 + + + + - - + + - 6 Yang et al., 2014 + + + + - - + + - 6 Zhou et al., 2019 + + + + - - + + - 6 Liu et al., 2017 A + + + + + + + + - 8 Liu et al., 2017B + + + + + + + + - 8 Zhang et al., 2021 + + + + + - + + + 8 Cao et al., 2019 + + + + + + + + + 9 Li et al., 2014 + + + + - - + + + 7 Li et al., 2018 + + + + + + + + + 9 Long et al., 2015 + + + + + + + + + 9 Wen et al., 2018 + + + + + + + + - 8 Zhu et al., 2020 + + + + + + + + + 9 Han et al., 2012 + + + + + + + + - 8 Wu et al., 2015 + + + + + + + + + 9 Jia et al., 2021 + + + + + + + + - 8 Lee et al., 2016 + + + + + + + + - 8 Hu et al., 2022 + + + + + + + + + 9 Yao et al., 2022 + + + + + + + + - 8 Wei et al., 2022 + + + + + + + + - 8 Li et al., 2021 + + + + + + + + - 8 Cotelli et al., 2011 + + + + + - + + + 8 Lv et al., 2024 + + + + + + - + - 7 Chen et al., 2025 + + + + + + + + - 8 Jung et al., 2024 + + + + + + + + + 9 Zhang et al., 2024 + + + + + - - + - 7 Yan et al., 2024 + + + + + + - + - 7 If the study meets a criterion, it will receive a “+”, and if it does not, it will receive a “-“. Criterion (1) The case definition was adequate with independent validation; Criterion (2) The cases were representative; Criterion (3) The control group was recruited from cases; Criterion (4) The control group had no history of endpoint; Criterion (5) Age was matched between cases and controls; Criterion (6) Sex or education was matched between cases and controls; Criterion (7) Status of cases and controls was blinded to research personnel. Criterion (8) Same method of ascertainment for cases and controls; and Criterion (9) Same non-response rate for both groups Risk of bias assessment for all studies using the Newcastle-Ottawa scale (NOS) If the study meets a criterion, it will receive a “+”, and if it does not, it will receive a “-“. Criterion (1) The case definition was adequate with independent validation; Criterion (2) The cases were representative; Criterion (3) The control group was recruited from cases; Criterion (4) The control group had no history of endpoint; Criterion (5) Age was matched between cases and controls; Criterion (6) Sex or education was matched between cases and controls; Criterion (7) Status of cases and controls was blinded to research personnel. Criterion (8) Same method of ascertainment for cases and controls; and Criterion (9) Same non-response rate for both groups Immediate effect of rTMS on general cognitive function in AD and MCI A total of 31 published studies were finally included in the meta-analysis, including 532 AD and 278 MCI patients treated with rTMS and 444 AD and 225 MCI patients treated with sham rTMS, to examine the immediate effect (post-treatment cognition) of rTMS on general cognitive functions in AD and MCI. Increased scores on MMSE and MoCA scales and decreased scores on the ADAS-Cog scale indicate improvements in general cognitive function. Results revealed that general cognitive functions at post-training assessment (immediate effect) in AD and MCI in the rTMS group were significantly higher than those in the sham rTMS group (SMD = 0.930, 95% CI = 0.638 to 1.222, I2 = 90.00%, p < 0.001, see Fig. 2). Meta-regression analysis results indicate that stimulation total time accounted for 25% of the observed heterogeneity, while type of scales, stimulation sessions, stimulation duration per sessions, stimulation site, and stimulation intensity explained 5%, 14%, 5%, 5%, and 16% of the heterogeneity, respectively. In contrast, stimulation frequency and type of patients did not significantly contribute to the heterogeneity. Forest plot regarding the immediate effect of rTMS on general cognitive function in AD and MCI. Each subgroup represents the type of scale used to assess cognitive functions in different studies, among which the “combined” subgroup represents a study that used two or more scales to measure patients’ cognitive functions. MMSE, Mini-Mental State Examination; MoCA, Montreal Cognitive Assessment; ADAS, Alzheimer’s disease Assessment Scale-cognitive subscale Subgroup analysis of stimulation parameters in immediate cognitive function Thirty-five studies stimulated the DLPFC, and fourteen studies stimulated other brain regions (Table 3). Our subgroup analysis showed that the effect sizes of rTMS estimated from both DLPFC (SMD = 1.28, p < 0.001) and other brain regions (SMD = 0.52, p = 0.003) were significant (Table 3). And there was significant difference in effect size between the two subgroups (p = 0.004). Table 3The immediate intervention effect on rTMS parametersrTMS parameterskSMDSEZ p 95%CIQ p Stimulation siteDLPFC351.2750.1976.480< 0.001(0.889, 1.661)Other140.5210.1752.9700.003(0.177, 0.864)Total between8.1890.004Stimulation intensity> 90%MT240.6030.1484.071< 0.001(0.312, 0.893)≤ 90%MT251.5070.2525.981< 0.001(1.013, 2.001)Total between9.5760.00280%MT142.0770.4274.864< 0.001(1.240, 2.913)90%MT110.4500.2032.2140.027(0.052, 0.848)Total between11.8370.001Stimulation frequency< 20 Hz331.1510.2065.602< 0.001(0.748, 1.554)≥ 20 Hz160.8740.2243.904< 0.001(0.435, 1.313)Total between0.8310.362Stimulation duration/session< 30 min280.8080.1724.687< 0.001(0.470, 1.146)≥ 30 min211.3750.2615.277< 0.001(0.864, 1.886)Total between3.2970.06914–15 min90.4790.2002.3980.01620–25 min160.6810.2163.1600.002Total between0.4720.492Stimulation total time> 800 min131.8840.3774.991< 0.001(1.144, 2.624)≤ 800 min360.7450.1285.801< 0.001(0.493, 0.997)Total between8.1560.004Stimulation sessions> 20161.7150.3544.840< 0.001(1.020, 2.409)≤ 20330.7170.1205.986< 0.001(0.482, 0.952)Total between7.1170.008Type of patientAD280.9820.2124.637< 0.001(0.567, 1.397)MCI211.1590.2255.153< 0.001(0.718, 1.600)Total between0.3290.566Type of scalesMMSE230.9700.2254.301< 0.001(0.528, 1.412)MoCA191.4050.2864.918< 0.001(0.845, 1.965)ADAS-Cog70.4350.1273.4130.001(0.185, 0.684)Total between11.8010.003k, number of studies; SMD, Standardized Mean Difference; SE, Standard error; 95%CI, 95% confidence interval (Lower CI, Upper CI); rTMS, repetitive transcranial magnetic stimulation; MCI, mild cognitive impairment; AD, Alzheimer’s dementia; DLPFC, granular dorsolateral prefrontal cortex; MT, motor threshold; MMSE, Minimum Mental State Examination; MoCA, Montreal Cognitive Assessment; ADAS-Cog, Alzheimer’s disease assessment scale-cognitive section The immediate intervention effect on rTMS parameters k, number of studies; SMD, Standardized Mean Difference; SE, Standard error; 95%CI, 95% confidence interval (Lower CI, Upper CI); rTMS, repetitive transcranial magnetic stimulation; MCI, mild cognitive impairment; AD, Alzheimer’s dementia; DLPFC, granular dorsolateral prefrontal cortex; MT, motor threshold; MMSE, Minimum Mental State Examination; MoCA, Montreal Cognitive Assessment; ADAS-Cog, Alzheimer’s disease assessment scale-cognitive section Twenty-four studies reported the effectiveness of rTMS stimulation intensity greater than 90%MT, twenty-five studies presented the effectiveness of rTMS stimulation intensity less than or equal to 90%MT (Table 3). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation intensity greater than 90%MT (SMD = 0.60, p < 0.001) and less than or equal to 90%MT (SMD = 1.51, p < 0.001) were significant (Table 3). And there was a significant difference in effect size between the two subgroups (p = 0.002). To parse the effectiveness of rTMS stimulation intensity on immediate cognitive function in the group with rTMS stimulation intensity less than or equal to 90%MT, we ran subgroup analyses (80%MT & 90%MT) across the included studies. The results showed that there was a significant difference in the effect size between the two groups (p = 0.001). The effect size of the 80%MT group (SMD = 2.08, p < 0.001) was greater than that of the 90%MT group (SMD = 0.45, p = 0.027). Thirty-three studies reported the effectiveness of rTMS stimulation frequency lesser than 20 Hz, sixteen studies presented the effectiveness of rTMS stimulation frequency greater than or equal to 20 Hz (Table 3). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation frequency lesser than 20 Hz (SMD = 1.15, p < 0.001) and greater than or equal to 20 Hz (SMD = 0.87, p < 0.001) were both significant (Table 3). There was no significant difference in effect size between the two subgroups (p = 0.362). Stimulation duration of single session in each study was different, ranging from 200s to 60 min. Twenty-eight studies reported the effectiveness of rTMS stimulation duration lesser than 30 min, twenty-one studies presented the effectiveness of rTMS stimulation duration greater than or equal to 30 min (Table 3). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation duration lesser than 30 min (SMD = 0.81, p < 0.001) and greater than or equal to 30 min (SMD = 1.38, p < 0.001) were both significant (Table 3). There was no significant difference in effect size between the two subgroups (p = 0.069). To parse the effectiveness of rTMS stimulation duration on immediate cognitive function, we ran further subgroup analyses in the group with rTMS stimulation duration lesser than 30 min. In this group, there were twelve studies with rTMS stimulation duration of 20 min, one study of 21.5 min, three studies of 25 min, four studies of 14.6 min, four studies of 14min32s, one study of 15 min and three studies of 200s. We divided the subgroup into 14–15 min and 20–25 min groups. We did not include the three literatures with a stimulation duration of 200s into the consideration, because the number of articles was too few to be a separate subgroup. The results showed that there was no significant difference in the effect size between the two groups (p = 0.492). The effect size of the 14–15 min group was 0.48 (p = 0.016), and effect size of the 20–25 min group was 0.68 (p = 0.002). On the other hand, we did not perform subgroup analyses in the group with rTMS stimulation duration greater than or equal to 30 min, because of insufficient literature on independent subgroup. Stimulation total time in each study was ranging from 33.3 min to 1200 min. Thirteen studies reported the effectiveness of rTMS stimulation total time greater than 800 min, thirty-six studies presented the effectiveness of rTMS stimulation total time less than or equal to 800 min (Table 3). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation total time greater than 800 min (SMD = 1.88, p < 0.001) and less than or equal to 800 min (SMD = 0.75, p < 0.001) were both significant (Table 3). There was significant difference in effect size between the two subgroups (p = 0.004), the effect size of the subgroup with stimulation total time greater than 800 min was significantly greater than that of the other subgroup. Sixteen studies reported the effectiveness of rTMS stimulation sessions greater than 20, thirty-three studies presented the effectiveness of rTMS stimulation total duration less than or equal to 20 (Table 3). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation sessions greater than 20 (SMD = 1.72, p < 0.001) and less than or equal to 20 (SMD = 0.72, p < 0.001) were both significant (Table 3). There was significant difference in effect size between the two subgroups (p = 0.008). Twenty-eight studies reported rTMS effects in AD, twenty-one studies presented rTMS efficacy in MCI (Table 3). Our subgroup analysis revealed that the effect sizes of rTMS estimated from both MCI (SMD = 1.16, p < 0.001) and AD (SMD = 0.98, p < 0.001) were significant (Table 3). However, there was no significant difference in effect size between the two subgroups (p = 0.566). Twenty-three studies measured immediate cognitive function with MMSE, nineteen studies measured immediate cognitive function with MoCA and seven studies measured immediate cognitive function with the ADAS-Cog (Table 3). Our subgroup analysis showed that immediate cognitive function estimated from MMSE (SMD = 0.97, p < 0.001), MoCA (SMD = 1.41, p < 0.001) and ADAS-Cog (SMD = 0.44, p = 0.001) were all significant (Table 3). There was significant difference in effect size between the three subgroups (p = 0.003). We further made pairwise comparisons among these three scales. The comparison results between MoCA and MMSE showed that immediate cognitive function estimated from MMSE (SMD = 0.97, p < 0.001), MoCA (SMD = 1.41, p < 0.001) were both significant. There was no significant difference in effect size between the subgroups (p = 0.232). The comparison results between ADAS-Cog and MMSE showed that immediate cognitive function estimated from MMSE (SMD = 0.97, p < 0.001), ADAS-Cog (SMD = -0.29, p = 0.001) were both significant. There was significant difference in effect size between the subgroups (p < 0.001). And there was significant difference in effect size between the two subgroups of MoCA (SMD = 1.41, p < 0.001) and ADAS-Cog (SMD = -0.29, p = 0.050) (p < 0.001). Sensitivity analyses and risks of publication bias Sensitivity analyses showed that the direction of effect showed no change when any one study was excluded (Supplementary materials for figure S1). Sensitivity analysis of the AD and MCI subgroups separately also showed that the direction of effect showed no change when any one study was excluded (Supplementary materials for figure S4 and S5). Additionally, Begg’s test, Egger’s test and funnel plot showed no significant risks of publication bias (Begg’s test: p = 0.17; Egger’s test: p = 0.25). We further used the trim-and-fill method [50] to impute the potential missing studies (Fig. 3). The adjusted ES after the use of the trim-and-fill method indicated a large effect of 1.06 (95% CI = 0.75, 1.36) favoring the rTMS intervention as compared to sham. Orwin’s fail-safe N test showed that 5555 studies with an ES of 0 would be needed to overthrow the conclusion that we were able to draw from 49 studies, indicating that the result may not be attributed to publication bias. Publication bias funnel plot (trim and fill) with standard errors and ES of the included rTMS studies Long-term effect of rTMS on general cognitive function in AD and MCI According to before mentioned search strategy, 9 published studies were identified from the selected database, including 200 AD and 55 MCI patients treated with rTMS and 186 AD and 52 MCI patients treated with sham rTMS. These studies explored the long-term effect of rTMS on general cognitive function in AD and MCI. However, studies on long-term cognitive effects (n = 9) were fewer than those examining immediate effects (n = 31), suggesting a research gap in longitudinal cognitive assessment. Meta-analysis showed significant difference in long-term effect of rTMS on general cognitive function between rTMS group and sham rTMS group with a random effects model (SMD = 0.42, 95% CI = 0.13 to 0.70, I2 = 72.0%, p = 0.004, Fig. 4). In contrast, for long-term cognition, type of scales accounted for 49% of the observed heterogeneity, while stimulation site explained 2%. The remaining moderator variables did not demonstrate significant explanatory power. Forest plots regarding the long-term effect of rTMS on general cognitive function in AD and MCI Subgroup analysis of rTMS parameters in long-term cognition Seven studies reported rTMS effects in DLPFC, eleven studies presented rTMS efficacy in other brain regions (Table 4). Our subgroup analysis showed that the effect sizes of rTMS estimated from both DLPFC (SMD = 0.68, p = 0.029) were significant, but other brain regions (SMD = 0.23, p = 0.097) were non-significant (Table 4). And there was no significant difference in effect size between the two subgroups (p = 0.184). Table 4The long-term intervention effect on rTMS parametersrTMS parameterskSMDSEZ p 95%CIQ p Stimulation siteDLPFC70.6820.3122.1860.029(0.071, 1.294)Other110.2290.1381.6600.097(-0.041, 0.499)Total between1.7650.184Stimulation intensity> 90%MT100.4170.2171.9180.055(-0.009, 0.842)≤ 90%MT80.4290.2551.6840.092(-0.070, 0.929)Total between0.0010.970Stimulation frequency> 10 Hz90.4810.2262.1240.034(0.037, 0.924)≤ 10 Hz90.3580.2411.4880.137(-0.114, 0.829)Total between0.1380.710Type of patientAD150.3580.1692.1120.035(0.026, 0.689)MCI30.7490.5121.4630.143(-0.254, 1.752)Total between0.5270.468Stimulation duration/session< 30 min120.5590.2252.4870.013(0.118, 0.999)≥ 30 min60.1510.1411.0740.283(-0.125, 0.427)Total between2.3660.124Stimulation total time> 500 min80.4080.2571.5880.112(-0.096, 0.911)≤ 500 min100.4320.2122.0340.042(0.016, 0.848)Total between0.0050.942Stimulation sessions> 2060.4620.3381.3690.171(-0.199, 1.124)≤ 20120.3940.1742.2660.023(0.053, 0.735)Total between0.0320.858Type of scalesMMSE8-0.0300.106-0.2790.780(-0.279, 0.780)MoCA40.8550.3752.2790.023(0.120, 1.591)ADAS-Cog60.7730.2802.7660.006(0.225, 1.321)Total between11.2880.004k, number of studies; SMD, Standardized Mean Difference; SE, Standard error; 95%CI, 95% confidence interval (Lower CI, Upper CI); rTMS, repetitive transcranial magnetic stimulation; MCI, mild cognitive impairment; AD, Alzheimer’s dementia; DLPFC, granular dorsolateral prefrontal cortex; MT, motor threshold; MMSE, Minimum Mental State Examination; MoCA, Montreal Cognitive Assessment; ADAS-Cog, Alzheimer’s disease assessment scale-cognitive section The long-term intervention effect on rTMS parameters k, number of studies; SMD, Standardized Mean Difference; SE, Standard error; 95%CI, 95% confidence interval (Lower CI, Upper CI); rTMS, repetitive transcranial magnetic stimulation; MCI, mild cognitive impairment; AD, Alzheimer’s dementia; DLPFC, granular dorsolateral prefrontal cortex; MT, motor threshold; MMSE, Minimum Mental State Examination; MoCA, Montreal Cognitive Assessment; ADAS-Cog, Alzheimer’s disease assessment scale-cognitive section Ten studies reported the effectiveness of rTMS stimulation intensity greater than 90%MT, eight studies presented the effectiveness of rTMS stimulation intensity less than or equal to 90%MT (Table 4). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulating intensity greater than 90%MT (SMD = 0.42, p = 0.055) and less than or equal to 90%MT (SMD = 0.43, p = 0.092) were non-significant (Table 4). And there was no significant difference in effect size between the two subgroups (p = 0.970). Nine studies reported the effectiveness of rTMS stimulation frequency greater than 10 Hz, nine studies presented the effectiveness of rTMS stimulation frequency less than or equal to 10 Hz (Table 4). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation frequency greater than 10 Hz (SMD = 0.48, p = 0.034) was significant and greater than or equal to 10 Hz (SMD = 0.36, p = 0.137) were non-significant (Table 4). There was no significant difference in effect size between the two subgroups (p = 0.710). Fifteen studies reported rTMS effects in AD, three studies presented rTMS efficacy in MCI (Table 4). Our subgroup analysis revealed that the effect sizes of rTMS estimated from AD (SMD = 0.36, p = 0.035) was significant and MCI (SMD = 0.75, p = 0.143) were non-significant (Table 4). And there was no significant difference in effect size between the two subgroups (p = 0.441). Twelve studies reported the effectiveness of rTMS stimulation duration/session lesser than 30 min, six studies presented the effectiveness of rTMS stimulation duration/session greater than or equal to 30 min (Table 4). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation duration/session lesser than 30 min (SMD = 0.56, p = 0.013) was significant and greater than or equal to 30 min (SMD = 0.15, p = 0.283) was non-significant (Table 4). There was no significant difference in effect size between the two subgroups (p = 0.124). Eight studies reported the effectiveness of rTMS stimulation total time greater than 500 min, ten studies presented the effectiveness of rTMS stimulation total time less than or equal to 500 min (Table 4). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation total time greater than 500 min (SMD = 0.41, p = 0.112) was non-significant and less than or equal to 500 min (SMD = 0.43, p = 0.042) was significant (Table 4). There was no significant difference in effect size between the two subgroups (p = 0.942). Six studies reported the effectiveness of rTMS stimulation sessions greater than 20, twelve studies presented the effectiveness of rTMS stimulation sessions less than or equal to 20 (Table 4). Our subgroup analysis showed that effect sizes of rTMS estimated from the groups with rTMS stimulation session greater than 20 (SMD = 0.46, p = 0.171) was non-significant and less than or equal to 20 (SMD = 0.39, p = 0.023) was significant (Table 4). There was no significant difference in effect size between the two subgroups (p = 0.858). Eight studies measured long-term cognitive function with MMSE, four studies measured long-term cognitive function with MoCA and six studies measured long-term cognitive function with the ADAS-Cog (Table 4). Our subgroup analysis showed that long-term cognitive function estimated from MMSE (SMD = -0.03, p = 0.790) was non-significant, but ADAS-Cog (SMD = 0.77, p = 0.006), MoCA (SMD = 0.86, p = 0.023) were both significant (Table 4). There was significant difference in effect size between the three subgroups (p = 0.004). We further made pairwise comparisons between the groups of MMSE and ADAS-Cog, due to the amounts of literature, we did not include the group of MoCA into our consideration. The comparison results between the groups of ADAS-Cog and MMSE showed that long-term cognitive function estimated from MMSE (SMD = -0.03, p = 0.780) was non-significant, and ADAS-Cog (SMD = 0.77, p = 0.006) was significant. There was significant difference in effect size between the subgroups (p = 0.007). Sensitivity analyses and risks of publication bias Sensitivity analyses showed that the direction of effect showed no change when any one study was excluded (Supplementary materials for figure S2). Sensitivity analysis of the AD and MCI subgroups separately also showed that the direction of effect showed no change when any one study was excluded (Supplementary materials for figure S6 and S7). Additionally, Begg’s test, Egger’s test and funnel plot showed no significant risks of publication bias (Begg’s test: p = 0.36; Egger’s test: p = 0.74). We further used the trim-and-fill method [50] to impute the potential missing studies (Fig. 5). The adjusted ES after the use of the trim-and-fill method indicated a medium effect of 0.42 (95% CI = 0.11, 0.74) favoring the rTMS intervention as compared to sham. Orwin’s fail-safe N test showed that 121 studies with an ES of 0 would be needed to overthrow the conclusion that we were able to draw from 18 studies. This indicates that the results of this study regarding the delayed cognitive efficacy of rTMS may have certain publication bias. Publication bias funnel plot (trim and fill) with standard errors and ES of the included rTMS studies Discussion Our meta-analysis of 31 randomized controlled studies found that rTMS had a significant improvement effect on immediate and long-term cognition in patients with AD and MCI compared with sham stimulation group. Results of this study further revealed that in the rTMS stimulation protocol, when the DLPFC site, stimulation intensity of 80% MT, total stimulation time greater than 800 min and sessions greater than 20 were selected as stimulation parameters, rTMS showed better efficacy in improving immediate cognitive function of AD and MCI patients. However, in terms of long-term cognition, there was no significant difference in the rTMS efficacy of rTMS parameters in AD and MCI patients. Immediate effect of rTMS on general cognitive function in AD and MCI Our meta-analysis showed that rTMS had a significant therapeutic effect on the immediate cognition in AD and MCI patients. Moderator analysis shows that when the stimulation site is DLPFC, the stimulation effect is larger than stimulating other sites, including parietal lobe, angular gyrus, cerebellum and multi-site combined stimulation area (Broca area, Wernicke area, DLPFC and parietal lobe) (1.28 [0.89, 1.66] vs. 0.52 [0.18, 0.86], p = 0.004). Previous meta-analyses of rTMS efficacy in patients with AD or MCI generally explored the effectiveness of DLPFC as the stimulation site to improve cognitive function in patients, mainly compared the differences in efficacy when stimulating unilateral DLPFC and bilateral DLPFC [51], as well as the changes in efficacy of rTMS stimulation when stimulating the left side [52–54]. However, there is a lack of direct comparison between the efficacy of stimulation at different sites [55]. The cerebellum [56], parietal-temporal region [51, 57] and precuneus [58] were all effective stimulation sites for TMS treatment of AD. In order to further explore the value of DLPFC in rTMS stimulation scheme, this study compared the magnitude of the immediate cognitive improvement of DLPFC and non-DLPFC stimulation sites in patients with AD and MCI, suggesting that DLPFC stimulation may be more effective for enhancing rTMS treatment outcomes. This may be attributed to rTMS inducing local neuronal depolarization and repeatedly delivering high-intensity magnetic fields through the scalp in the target area of ​​the brain to change cortical activity [59]. When rTMS stimulates the DLPFC, it may activate the connections of distal structures, which in turn improves cognition [60]. The pathological feature of AD is dysregulation of the cholinergic system. Cholinergic pathways can regulate synaptic plasticity and cognitive function and are crucial for memory formation and consolidation [61]. The converging evidence from multiple studies demonstrated that short latency afferent inhibition (SAI) is significantly weakened in AD, suggesting that the function of central cholinergic circuits mediating intercortical communication is impaired [62, 63], which may be due to the increase in Aβ content. Studies have shown that the individual amount of cholinergic activity assessed by SAI in AD patients is negatively correlated with the level of Aβ42 in CSF [64]. Notably, therapeutic intervention with TMS has been shown to partially restore SAI magnitude in AD cohorts, concomitant with measurable improvements in domain-specific cognitive performance metrics, suggesting a potential mechanistic link between cholinergic circuit modulation and cognitive remediation. Compared with normal aging individuals, AD and MCI patients exhibit more pronounced atrophy in the basal forebrain [65], this structural degeneration may lead to reduced cholinergic input [66], which is directly associated with global cognitive decline in patients. Furthermore, healthy subjects revealed a significant increase in SAI during memory retrieval, suggesting that normal memory processes are closely linked to cholinergic neurotransmission [67]. Therefore, rTMS therapy targeting the DLPFC may improve cognitive function in AD or MCI patients by enhancing cholinergic innervation from the basal forebrain. Under rTMS stimulation, cholinergic regulation is facilitated through indirect pathways, such as enhancing GABAergic signaling [68], promoting the microglial mitochondrial transporter (TSPO) in its role of clearing Aβ [69], and upregulating brain-derived neurotrophic factor (BDNF). These mechanisms work synergistically to enhance GABAergic neuron activity and neuroprotection [70]. Another key mechanism may involve long-term potentiation (LTP)-like plasticity, a sustained increase in synaptic strength following specific stimulation. AD is increasingly recognized as a synaptic disorder [71], with synaptic degeneration occurring early in its pathophysiology and a general reduction in synaptic marker density in the neocortex [72]. Moreover, elevated t-Tau levels are linked to enhanced long-term depression (LTD)-like plasticity and impaired LTP mechanisms [73, 74], indicating disrupted cortical plasticity in AD. The DLPFC exhibits LTP-like plasticity deficits that correlate with cognitive dysfunction in AD [43, 75]. Stimulating this region can directly activate neurons, reduce synaptic activation thresholds, and enhance synaptic connectivity and plasticity, thereby improving cognitive performance [76, 77]. High-frequency rTMS promotes postsynaptic depolarization, boosts neurotransmitter release, and induces hippocampal LTP via NMDA receptor activation and Ca²⁺ influx [78]. Thus, rTMS targeting the DLPFC may increase cation permeability at the molecular level, facilitating depolarization and driving cortical LTP-like plasticity. Additionally, rTMS-induced enhancement of LTP-like plasticity in the DLPFC is closely associated with BDNF. In animal models, BDNF regulates NMDAR-dependent LTP [79] and modulates synaptic plasticity through both presynaptic and postsynaptic mechanisms [80, 81]. Clinical studies have shown that rTMS can elevate serum BDNF levels in humans [82]. Increased BDNF may help maintain cortical volume by boosting presynaptic neurotransmitter release [83] and promoting postsynaptic structural maturation [84, 85], potentially explaining the observed stabilization of cortical volume in post-rTMS AD patients [86]. Traditional views emphasize the local effects of rTMS, but emerging neuroimaging evidence suggests that its therapeutic mechanism involves not just local modulation, but multiscale neural circuit reorganization [87, 88]. While the bioelectromagnetic effects of rTMS are initially confined to target regions such as the DLPFC, the induced neural activity dynamically propagates across the brain via functional connectivity networks. The default mode network (DMN) exhibits functional connectivity disruptions in AD and MCI patients, which correlate positively with cognitive decline severity [89]. Neuroimaging studies have confirmed a dynamic antagonistic relationship between the DLPFC and DMN, mediated by distinct neural pathways. Precisely targeted accelerated intermittent theta burst stimulation (iTBS) can significantly enhance the negative functional connectivity between the DLPFC and key DMN regions [90]. In addition, the central executive network (CEN) is another core neurocognitive system essential for higher-order cognition [91, 92]. As a key CEN hub, DLPFC stimulation induces targeted neuromodulation by inhibiting functional connectivity (FC) between the anterior cingulate cortex and inferior parietal lobule within the CEN, while also weakening cross-network interactions between the CEN and DMN. Notably, after DLPFC stimulation, FC between the left hippocampus and inferior frontal gyrus increases—a pathway closely linked to working memory consolidation and attention allocation. These dual effects—network rebalancing and enhancement of hippocampal-cortical pathways—contribute to improvements in episodic memory and executive function [88]. Therefore, future research should prioritize systematic studies to delineate the neuroanatomical specificity of neuromodulation and clarify whether cognitive gains result from local plasticity at the stimulation site or from distributed reorganization across interconnected cognitive control circuits. In addition, this study examined the effect of stimulation frequency of rTMS through subgroup analysis. The results showed that in patients with AD and MCI, there was no significant difference in the effect size of rTMS stimulation between frequency < 20 Hz and ≥ 20 Hz on immediate recognition. However, regarding the choice of rTMS stimulation frequency, conflicting results have been shown between different meta-analysis studies. One study by Wei et al. [93], 10 Hz rTMS showed a better therapeutic effect on cognitive function in AD patients compared to 20 Hz. Another meta-analysis study by Wang et al. [94] showed the opposite result, indicating that 20 Hz rTMS was more effective in improving cognitive ability in AD patients than 10 Hz–1 Hz rTMS. Another meta-analysis study by Yan et al. showed that the medium-frequency stimulation (5 Hz and 10 Hz) may have better therapeutic effects on general cognition in AD patients compared with low-frequency (1 Hz) and high-frequency (20 Hz) stimulation [56]. These results suggest that the relationship between stimulation frequency and effect on cognitive improvement might be non-linear, the heterogeneity of methodology or clinical application of rTMS among different studies might be another possibility [95]. Therefore, future high-quality large-sample RCT studies are needed to draw a conclusion. In terms of stimulation intensity, our subgroup analysis showed that in patients with AD and MCI, the therapeutic effect of rTMS on immediate cognition when the stimulation intensity was ≤ 90% MT was better than when the stimulation intensity was >90% MT (1.51 [1.01, 2.00] vs. 0.60 [0.31, 0.89], p = 0.002), further analysis showed that the therapeutic effect of rTMS was significantly larger when the stimulation intensity was 80% MT than when the stimulation intensity was 90% MT (2.08 [1.24, 2.91] vs. 0.45 [0.05, 0.85], p = 0.001). A recent meta-analysis showed similar results, in AD patients, rTMS stimulation at 90% RMT significantly improved cognitive function compared with greater stimulation intensities of 100% and 90%-100% [96]. One potential mechanism may be related to the oxyhemoglobin content in the cerebral cortex. Thomson et al. [97] found that when 120% RMT was used for stimulation, DLPFC blood oxygen saturation was reduced for a prolonged period of time compared with 80% RMT, which possibly because 120% RMT more strongly engages inhibitory mechanisms, leading to reduced neural activity. However, similar studies on rTMS stimulation intensity are limited, and our conclusions still need to be treated with caution. On the whole, the studies included in the subgroup with stimulation intensity of 80%, the corresponding single stimulation duration and total stimulation time were generally on the high side, and this longer stimulation time may have a stronger and more lasting effect on AD patients [12, 98, 99], therefore the effect size of the 80%MT subgroup in this study was significantly larger than that of the 90%MT subgroup. On the other hand, different combinations of different parameters may work through different mechanisms. Previous studies have shown that when the frequency is consistent, stimulating the left motor cortex of healthy people at subthreshold (90% RTM) or suprathreshold (110% RTM) intensities can enhance the activation of cortical and subcortical areas of the motor (and auditory) system [100]. When the TMS stimulation intensity is constant at 120% RMT, 10 and 20 Hz rTMS are equally effective in treating affective symptoms in patients with major depressive disorder [101]. Different combinations of parameters may produce different therapeutic effects. In the future, statistical analysis should be conducted on different combinations of rTMS stimulation parameters to discuss the effectiveness of the stimulation scheme as a whole. We obtained the stimulation total time by multiplying the duration of a single stimulation session by the number of the stimulation sessions. Subgroup analysis of the total stimulation time showed that when the total stimulation time was >800 min, rTMS had a larger effect on immediate cognition in patients with AD and MCI than when it was ≤ 800 min (1.88 [1.14, 2.62] vs. 0.75 [0.49, 1.00], p < 0.001). Through subgroup analysis of single stimulation duration and stimulation sessions, we found that rTMS stimulation sessions >20 times were significantly more effective than stimulation sessions ≤ 20 times (1.72 [1.02, 2.41] vs. 0.72 [0.48, 0.95], p = 0.008). However, the effect of rTMS stimulation duration < 30 min per session and ≥ 30 min per session on immediate cognition in AD and MCI patients was not significantly different (p = 0.125). Previous studies have also shown that longer stimulation time can lead to greater cognitive improvement [102]. Another work by Wang et al. [94] also highlighted that participants with higher education receiving rTMS treatment for more than 10 sessions were more effective in improving their cognitive functions. These studies have shown that longer durations were associated with stronger effects. Evidence has shown that repeateda total of 26 randomized controlled trials were included and long-term stimulation can produce stronger and longer lasting effects in patients with neurodegenerative diseases such as AD [12, 98, 99]. In addition, although the effect size of the subgroup of stimulation duration ≥ 30 min per session was not significantly larger than that of the subgroup of stimulation duration < 30 min per session (SMD = 1.38 vs. SMD = 0.81), it still showed a larger effect size, indicating that the longer the stimulation duration, the greater the stimulation effect. Thus, taken together with previous studies, our results provide some evidence that longer treatment is more beneficial. On the other hand, we conducted corresponding subgroup analysis on types of patients, and the results showed that rTMS was effective on immediate cognition in both AD (0.98 [0.57, 1.40]) (p < 0.001) and MCI patients (1.16 [0.72, 1.60]) (p < 0.001), but there was no significant difference in the therapeutic effect between the two types of patients (p = 0.566). Previous studies have suggested that AD patients may not benefit most from these training techniques, as they may be more effective in prodromal stages of the disease, such as MCI [52]. The differences in the study results may be related to the heterogeneity of the included patients. Previous studies have confirmed that the effectiveness of rTMS may be related to the age of the participants. Teselink et al.‘s [52] meta-regression revealed a negative association between age and the effect of non-invasive brain stimulation (NIBS) on general cognition. NIBS has a greater beneficial effect in young people, regardless of diagnosis, and a younger age does not necessarily correspond to the prodromal stage of the disease. The AD and MCI patients included in this study were all over 60 years old, which may limit the effectiveness of rTMS to some extent. Age serves as a primary non-modifiable risk factor for neurodegenerative disorders such as AD, which progressively impair cognitive functions. Notably, all AD and MCI patients included in this meta-analysis exceeded 60 years of age. At the regional level, older adults (aged 60 years and above) exhibit a general reduction in gray matter volume, particularly in the frontal, insular, and cingulate cortical regions, compared to younger individuals [103, 104]. As MCI progresses to AD, cortical atrophy becomes more widespread. Longitudinal voxel-based morphometry (VBM) analyses reveal increasing gray matter loss in the temporal neocortex, parahippocampal cortex, and cingulate regions, with additional involvement of the frontal lobe, insula, and cerebellum [105]. Age-related cortical atrophy, varying in severity across individuals, may partially compromise the therapeutic potential of rTMS. Mechanistically, cortical atrophy increases the scalp-to-cortex distance (SCD), thereby attenuating the electromagnetic field intensity of TMS pulses reaching cortical targets [106]. Advanced neurocomputational studies employing MRI-derived finite element head models have demonstrated that cortical atrophy not only reduces peak current density in targeted cortical regions but also alters current vector orientation patterns. Furthermore, the widening of cerebral sulci associated with atrophy amplifies localized current density variations through geometrical distortion effects, collectively diminishing both the efficacy and spatial precision of TMS interventions [107]. These biophysical findings substantiate cortical atrophy as a critical moderator influencing rTMS-mediated cognitive improvement in AD/MCI populations. Future investigations should incorporate age-stratified subgroup analyses to systematically evaluate differential treatment responses across aging trajectories. For elderly individuals, rTMS stimulation parameters should be individualized by incorporating structural MRI data to calculate the electric field distribution. Stimulation intensity should be determined based on the actual cortical current density rather than a fixed percentage of the RMT, and gradually adjusted according to patient tolerance. Additionally, neuronavigation systems should be employed to accurately target compensatory brain regions with preserved functional connectivity, or alternatively, multiple stimulation sites may be selected to engage several network nodes simultaneously and enhance the effects of network reorganization. Finally, we found that the ADAS-Cog, MoCA and MMSE scores of AD and MCI patients were significantly improved after rTMS; among them, the ADAS-Cog score was not improved as much as MMSE (-0.29 [-0.57, − 0.001] vs. 0.97 [0.53, 1.41], p < 0.001) or MoCA (-0.29 [-0.57, − 0.001] vs. 1.41 [0.85, 1.97], p < 0.001). To date, no studies have conducted a meta-analysis on the selection and validity of different scales. Previous studies have shown that different scales focus on measuring different populations and cognitive domains. MMSE is the most widely used cognitive screening scale in China and abroad. Its content covers orientation, memory, attention, calculation, language ability and visual-spatial ability. It has good value for screening AD, but its role in identifying normal elderly people and MCI and distinguishing MCI from AD is limited [108, 109]. MoCA covers cognitive domains such as attention, executive function, memory, language, visual-spatial structural skills, abstract thinking, calculation and orientation, and is designed to screen patients with MCI [110]. ADAS-Cog is commonly used to evaluate the efficacy of mild to moderate AD, and is more focused on measuring memory and language functions [111]. The scales used in the studies included in this meta-analysis, as well as the different types of rTMS stimulation protocols used for different subjects, are highly heterogeneous, so our results still need to be treated with caution. In addition, only six studies used the ADAS-Cog scale to assess global cognitive ability were included in the subgroup analysis. The small number of randomized controlled trials and small sample size inherently reduced the statistical power of the analysis, limiting our ability to deal with heterogeneity and the generalizability of the results [112, 113]. Long-term effect of rTMS on general cognitive function in AD and MCI rTMS had a significant long-term cognitive effect in patients with AD and MCI. However, in the subgroup analysis of different parameters such as stimulation site, frequency, intensity, duration, and patient type, no superior choice was found. The only difference was in the choice of cognitive scales, where the moderator analysis showed that the ADAS-Cog was more clinically meaningful than the MMSE [114]. However, due to the limitation of sample size, more evidence is needed to support our conclusion. Limitations There are some inherent limitations in this meta-analysis. First, the sample size of the research on long-term cognition in this study was small, with only nine papers included, which limits the in-depth analysis. The absence of a significant moderating effect in the long-term analyses may be attributable to insufficient statistical power, rather than indicating a lack of substantive impact of the moderator variable. The sample size should be expanded for further analysis in the future. Second, we combined AD and MCI as our study participants, which may improve the statistical power and generalizability of our findings. However, AD and MCI are different clinical stages of neurocognitive disorders, suggesting that these diseases may respond differently to treatment. Future work would benefit from stratifying analyses more precisely by diagnostic category (AD vs. MCI) and by follow-up duration. Conclusion In conclusion, the findings from our meta-analytic review suggest that rTMS significantly ameliorates both immediate and long-term cognition in individuals with AD and MCI. Stimulation DLPFC sites, 80%MT and long-term treatment are more effective in improving immediate cognitive performance. In addition, MMSE and MoCA can better reflect the improvement of immediate cognition, and ADAS-Cog can better reflect the improvement of long-term cognition. In the future, larger sample and finer groupings are needed to explore appropriate stimulation parameters to improve general cognition in AD and MCI. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements We would like to thank Xinrui Cui for the assistance in compiling the data. Author contributions Conception and design: K.S., K.C.L. and P.P.L.; Data collation: B.J.W., F.L.Y. and F.H; Statistical Analysis: B.J.W., Y.W. and F.H; Article writing: B.J.W.; Article revision: K.C.L., K.S. and P.P.L. The authors read and approved the final manuscript. Final approval of the manuscript: All authors. Funding This study was supported by the Key Project of Medical and health technology of Longgang District, Shenzhen (grant number LGKCYLWS2022034). Data availability No datasets were generated or analysed during the current study. Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Kai Sun, Email: Henrysk@163.com. Peipeng Liang, Email: ppliang@cnu.edu.cn. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement No datasets were generated or analysed during the current study.