A total of nine databases were searched for studies published in Chinese or English from their inception until July 31, 2024. These databases included the Cochrane Library, PubMed, Embase, Web of Science, Scopus, China National Knowledge Infrastructure (CNKI), and Wanfang Data. Within these databases, we will conduct a thorough search for gray literature and trial registries to ensure comprehensive coverage of all relevant studies. The retrieval strategy, which combined MeSH terms and entry terms, was finalized after three rounds of pre-retrieval. During these rounds, the search terms were adjusted based on their relevance, specificity, and comprehensiveness, ensuring they captured all relevant studies. We consulted with an information specialist to refine the search strategy and ensure its alignment with best practices.

The search terms used for both English and Chinese studies included repetitive transcranial magnetic stimulation (重复经颅磁刺激), stroke (脑卒中), cerebrovascular accident (脑血管意外), sleep (睡眠), sleep disorder (睡眠障碍), dyssomnias (睡眠异常), and insomnia (失眠). The Boolean search structure used across databases was as follows: (stroke OR cerebrovascular accident OR apoplexy) AND (sleep OR insomnia OR dyssomnias) AND (transcranial magnetic stimulation OR repetitive transcranial magnetic stimulation OR rTMS).

Taking Pubmed as an example, our literature search formula is as follows:

(((cerebral hemorrhage[Title/Abstract]) OR (((((stroke[Title/Abstract]) OR (cerebrovascular accident[Title/Abstract])) OR (apoplexy[Title/Abstract])) OR (cerebral in farction[Title/Abstract]))) AND (((sleep[Title/Abstract]) OR (insomnia[Title/Abstract])) OR (dyssomnias[Title/Abstract]))) AND (((transcranial magnetic stimulation[Title/Abstract]) OR (repetitive transcranial magnetic stimulation[Title/Abstract])) OR (rTMS[Title/Abstract]))).

This research has been registered with the PROSPERO International Systematic Review Registration Platform, and no deviations from the registered protocol occurred during the study. Full details of the protocol are available under Registration No. CRD42024589437, which can be accessed at https://www.crd.york.ac.uk/PROSPERO. The review adheres to the PRISMA 2020 guidelines (24), and the risk of bias was evaluated in accordance with the PRISMA extensions for network meta-analysis.

The Pittsburgh sleep quality index (PSQI) is a self-reported questionnaire used to measure sleep quality and disturbances over a one-month period. It has been validated in post-stroke populations, confirming its reliability and suitability for use in this cohort (25). The Pittsburgh Sleep Quality Index (PSQI) is composed of multiple components, with a total of 19 items distributed across seven domains, providing a global score to evaluate overall sleep quality. Each domain reflects a distinct aspect of sleep quality, including subjective sleep quality, sleep latency, sleep duration, sleep efficiency, sleep disturbances, use of sleep medication, and daytime fatigue, performance, or alertness. Scoring: Each component is scored on a scale from 0 to 3, with a global score ranging from 0 to 21. Higher scores indicate poorer sleep quality, with a global score above 5 typically suggesting significant sleep problems; however, this cut-off may vary depending on the clinical context and population norms (26).

While the PSQI provides valuable subjective insights into sleep quality, the inclusion of sleep efficiency, an objective measure derived from polysomnography, allows for a more comprehensive assessment of sleep quality, capturing both subjective experiences and objective sleep metrics. Sleep efficiency refers to the percentage of time spent asleep relative to the total time spent in bed (27). Polysomnography provides an accurate assessment of sleep efficiency by measuring the time spent asleep in comparison to the total time in bed. It is calculated as the total sleep time divided by the time in bed, multiplied by 100. Higher sleep efficiency indicates better sleep quality, as more time in bed is spent sleeping rather than awake (28). However, it is important to note that sleep efficiency derived from polysomnography can differ significantly from the self-reported estimates obtained from the PSQI. Polysomnography provides an objective, direct measure of sleep duration and disturbances, while the PSQI relies on subjective reports from the patient, which may lead to discrepancies between the two measures due to individual perceptions of sleep quality. Therefore, PSQI and sleep efficiency have been included in the meta-analysis to represent the impact on sleep quality following rTMS, providing a comprehensive assessment of both subjective and objective measures.

The Hamilton Depression Scale (HAMD-17) is a clinician-administered scale used to assess the severity of depressive symptoms. The 17-item version is preferred due to its wide use in clinical settings, particularly for stroke populations (29). It contains 17 items that measure various aspects of depression, including mood, guilt, insomnia, anxiety, and somatic symptoms. Each item is scored on a scale of 0 to 4 for symptoms such as depressive mood, guilt, suicidal tendencies, work and interest, slow thinking and speech, agitation, mental anxiety, somatic anxiety, and hypochondriasis. For symptoms like insomnia, shallow sleep, early awakening, gastrointestinal symptoms, general symptoms, sexual symptoms, weight loss, and insight, the scoring scale ranges from 0 to 2. The total possible score ranges from 0 to 52. Higher scores indicate more severe depression. A HAMD-17 score of 8 or above indicates the presence of depression, with severity categorized as follows: 0–7 (normal), 8–16 (mild), 17–23 (moderate), and ≥24 (severe) (30). Therefore, the HAMD-17 assesses the severity of depressive symptoms and serves as a tool for monitoring mood changes in post-stroke patients.

We selected randomized controlled trials of the application of TMS in post-stroke sleep disorders patients. To be eligible for inclusion, studies needed to satisfy the following conditions: (1) The study subjects were patients with sleep disorders after stroke, all were aged>18 years, and all had stable vital signs.; (2) Diagnostic criteria for sleep disorders post-stroke were based on the International Classification of Sleep Disorders (ICSD-3) or The Diagnostic and Statistical Manual of Mental Disorders Fourth Edition (DSM-V), depending on the study.; (3) All participants had an MMSE score greater than 27, ensuring appropriate cognitive function.; (4)The experimental group received rTMS treatment with detailed standardized reporting, including coil type, stimulation protocol, stimulation intensity expressed as a percentage of the resting motor threshold, and other relevant parameters.; (5) In the control group, participants received conventional therapy (defined as a combination of pharmacological, behavioral, and/or physiotherapy interventions depending on the study), sham rTMS (no stimulation), conventional therapy combined with sham rTMS, or other treatments such as medication or acupuncture.; (6) outcomes comprising valid tests and measures of Pittsburgh Sleep Quality Index (PSQI), Sleep Efficiency, and Hamilton Depression Rating Scale-17 (HAMD-17); and (7) reporting the number of participants and all essential data for effect size calculations. Studies were excluded if they reported results from a previously published source (duplicated data). Our study protocol adhered to the Quality of Reporting of Meta-analyses (QUOROM) guidelines (31) and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statements (24). All included studies were analyzed exclusively using the data reported in the published articles. The study adhered to the PRISMA checklist, ensuring transparency in the search strategy and the assessment of risk of bias, with the full checklist available in the Supplementary materials. The exclusion criteria are detailed in the figure.

We extracted information on participants and trial characteristics using a structured form (Table 1) to ensure consistency, enable cross-study comparison, and reduce extraction bias. Specifically, this included participants’ mean age, sex, and baseline symptom severity, as well as trial characteristics such as first author, publication date, type of intervention, type of control, outcome assessment, and intervention duration.

Two researchers (CF and WZ) assessed the risk of bias in the included studies using the risk of bias assessment tool for RCTs as outlined in the Cochrane Handbook (24). The identity and role of researchers are clearly stated, and independent assessments and cross-validation were performed to ensure high inter-rater reliability. The evaluation criteria included the following domains: random sequence generation; allocation concealment; blinding of participants and personnel; blinding of outcome assessors; completeness of outcome data; selective reporting; and other potential sources of bias. Each domain was rated as ‘low risk’, ‘high risk’, or ‘unclear risk’ of bias based on the guidelines of the Cochrane Handbook (24), which provide specific criteria for judgment. To ensure reliability, two reviewers independently performed the risk of bias assessment, and any disagreements were resolved through discussion or consultation with a third reviewer.

Publication bias was evaluated using a comparison-adjusted funnel plot and Egger’s test. In the comparison-adjusted funnel plot, study-specific effect sizes were plotted against their standard errors while adjusting for different comparisons within the treatment network, under the assumption of transitivity and consistency across the network (23). Egger’s test was applied to assess funnel plot asymmetry, with a p-value <0.05 indicating potential small-study effects or publication bias (32). This approach accounts for the multi-arm structure of the treatment network and incorporates heterogeneity across studies.

A frequentist network meta-analysis (NMA) using a random-effects model was employed to synthesize both direct and indirect evidence concurrently. All analyses were conducted using Stata statistical software version 16.0 and Review Manager software version 5.3 (provided in the Supplementary materials) (23). The effect size for continuous variables was assessed using mean difference (MD) with a 95% confidence interval (CI). CIs were computed using normal approximations from random-effects models, with between-study variance (τ²) estimated via restricted maximum likelihood (REML), under the assumption of approximately normal sampling distributions of study-level effect estimates. In cases where a closed loop existed, statistical inconsistency between direct and indirect evidence was evaluated through local (node-splitting method) and global (design-by-treatment interaction model) approaches (32). If the p-value exceeded 0.05, it indicated no statistically significant inconsistency, and the consistency model was applied for further analysis. A network plot was generated to illustrate the relationships among the different interventions, where the size of the nodes represented the sample size of each intervention, and the line thickness indicated the number of randomized controlled trials (RCTs) with direct comparisons. The SUCRA probabilities were calculated to assess the efficacy of various treatment strategies, with higher SUCRA values indicating a more favorable outcome (33). Cumulative rank-probability curves were derived by summing rank probabilities across categories, with surface under the cumulative ranking curve (SUCRA) corresponding to the normalized area under these cumulative curves.

Seven databases were searched utilizing the designated search strategy, yielding 389 papers during the initial screening. Among them, seven papers were retrieved from Cochrane Library, 22 from PubMed, 20 from Scopus, 199 from Embase, 63 from China National Knowledge Infrastructure, and 78 from the Wanfang Data databases. After removing duplicates, reviewing the titles, abstracts, and full texts, and conducting a quality assessment, 15 papers (34–48) were ultimately included. The detailed screening process, including the reasons for excluding the papers, is presented in Figure 1. Furthermore, the main characteristics of the included studies are summarized in Tables 2, 1.

A total of 15 randomized controlled trials (RCTs) involving 1,113 patients with PSSD were selected, with the patients having an average age range of 43–74.34 years. The frequency, target location, and intensity of rTMS differed across the included studies. Eleven studies employed low-frequency stimulation (≤1 Hz) (34–38, 40, 43–47), while four utilized high-frequency stimulation (≥5 Hz) (39, 41, 42, 48). The commonly targeted brain region was the DLPFC, with three localization methods: right, left, and bilateral. In particular, eight studies targeted the r-DLPFC (34–38, 43, 45, 47), six targeted the l-DLPFC (39, 41, 42, 44, 46, 48), and only one targeted the b-DLPFC (40). The stimulation intensity of rTMS is determined by the motor threshold of each patient. During the assessment of the motor threshold, patients are seated or placed in a supine position, and a single-pulse stimulation is applied to the motor cortex area controlling the thumb of the dominant hand. The motor threshold is established as the intensity at which thumb abduction is induced in 5 out of 10 stimulations (19). The motor threshold percentages varied across different studies. Five studies applied 80% of the motor threshold (34, 40, 41, 45, 47), four employed 90% (39, 42, 46, 48), three utilized an 80–120% range (35, 37, 43), and one used 100% (38). Only two studies did not specify the motor threshold (36, 44). The rTMS treatments were administered within a range of 2 to 8 weeks, with a frequency of once per day, 5–7 times per week, and a session duration ranging from 10 to 30 min (Table 1).

Among the 15 included studies, 14 explicitly described the methods used for random sequence generation, such as computer-based randomization or random number tables; thus, they were assessed as having a low risk of bias for random sequence generation. However, the remaining study merely mentioned using randomization without providing details, leading to an unclear risk of bias for the randomization process. Furthermore, none of the studies provided information on allocation concealment or blinding of experimental staff. Consequently, all studies were considered to have an unclear risk of bias for allocation concealment and blinding of experimental staff. Moreover, all studies were evaluated as having a low risk of reporting bias. Patient dropouts were reported in five studies, indicating a low risk of bias for data completeness in those studies. Conversely, five other studies did not provide details on patient dropout, resulting in an unclear risk of bias for data completeness in those studies. Lastly, all studies were determined to have an unclear risk of bias for selective reporting of results and other sources of bias. The overall risk of bias for the included studies is illustrated in a risk-of-bias graph (Figure 2A) and a risk-of-bias summary (Figure 2B).

PSSD is a prevalent and critical barrier to recovery in stroke survivors, manifesting as abnormalities in the quality, quantity, or timing of sleep. Symptoms of PSSD commonly encompass nighttime awakenings, excessive daytime sleepiness, and disrupted sleep–wake rhythms (50), with stroke severity, age (51), and gender (52) serving as the contributing factors. Insomnia affects nearly 75% of stroke survivors, representing a significantly higher proportion than in the general population (25). Moreover, PSSD can exacerbate cognitive deficits, further impairing rehabilitation outcomes (53). The interplay between sleep disorders and depression after stroke is of major concern. PSSD often leads to fatigue, tension, and anxiety, which can compound the psychological burden, reduce the quality of life, and hinder recovery motivation of the affected patients (54). Poor sleep quality has been linked to greater depressive symptoms, and depression, in turn, is associated with worsened sleep disturbances, suggesting a bidirectional relationship between the two conditions after stroke (55). Along with psychological effects, PSSD exerts detrimental effects on the nervous and cardiovascular systems, thereby increasing the risk of recurrent cerebrovascular events and even mortality (56).

Although the stroke-induced effects on brain tissue are well-documented, the specific mechanisms underlying PSSD remain unclear. Furthermore, stroke-related changes in the upper airway anatomy and function may contribute to sleep-disordered breathing, a primary cause of these sleep disturbances (50). Previous research suggests that the disruption of the cerebellar neural circuits coordinating the upper airway and diaphragm can diminish the function of the upper airway muscles, ultimately contributing to sleep apnea (57). Moreover, damage to the brainstem nuclei involved in sleep–wake regulation may contribute to sleep disorders following a stroke. For example, stroke lesions in the hypothalamus have been associated with sleep disorders primarily characterized by reduced non-rapid eye movement (non-REM) sleep (58). In line with this finding, studies using animal models of middle cerebral artery occlusion have demonstrated a significant reduction in non-REM sleep duration, along with elevated levels of orexin and orexin receptor 1 (59). Orexin neurons in the hypothalamus play a crucial role in sleep inhibition, particularly during wakefulness (60). Apart from lesion location, post-stroke sleep-disordered breathing is also affected by several other factors. Studies have revealed higher rates of sleep-disordered breathing following strokes in the brainstem or bilateral hemispheres (61, 62). Conversely, Stahl (63) have found that the incidence and severity of obstructive sleep apnea syndrome following stroke were not significantly correlated with specific lesion locations. The pathogenesis of sleep disorders can also involve psychological factors. For instance, external environmental changes and psychological stress, such as those arising from stroke, may exacerbate sleep problems. In certain individuals, this worsening of sleep issues can lead to the progression from transient insomnia to persistent sleep disorders, particularly among those who are unable to regulate the stress caused by their condition. Our findings showed that Low-frequency rTMS targeting the right and bilateral DLPFC not only significantly improved sleep function in patients with stroke but may also alleviate depression.

Sleep is vital in the processes of neural repair and regeneration. Experimental studies have demonstrated that sleep deprivation reduces the proliferation of neurons and vascular cells and inhibits axonal growth, ultimately impeding neuroplasticity and recovery. This result highlights that sleep disturbances can significantly affect the restoration of neurological function (64). However, in clinical settings, neuroplasticity outcomes may also be affected by factors such as rehabilitation intensity, comorbidities, and individual variability. Moreover, sleep disorders can affect the brain’s autoregulatory mechanisms of cerebral blood flow, thus increasing the susceptibility of the brain to hypoxic damage. This oxidative stress environment also activates inflammatory pathways, leading to the release of inflammatory cytokines and further exacerbating neuronal injury (65, 66).

rTMS is an innovative, non-invasive neuroregulation technique. This stimulation method involves placing a coil perpendicular to the brain surface and passing an electric current through the coil to generate a pulsed magnetic field. Based on electromagnetic induction principles, this magnetic field induces a reverse current in the cortical tissue, which depolarizes local neurons, modulates cortical excitability, and influences neurotransmitter activity within targeted neural circuits. Such repeated, continuous, and rhythmic stimulations result in the modulation of the neural networks, eventually facilitating neuroplasticity (67). rTMS may treat PSSD through the following mechanisms:

In clinical practice, the effectiveness of rTMS largely depends on the combination of various stimulation parameters, including frequency, intensity, pulse number, duration, and the relative positioning of the coil to the target brain area. The DLPFC is the common target for rTMS, given its crucial role in generating and regulating emotions and its close connections to other emotion-related brain regions, such as the limbic system, amygdala, and cingulate cortex. Consequently, the modulation of DLPFC activity by rTMS not only directly influences DLPFC excitability but also indirectly affects these emotional control areas (77). For example, stimulating the DLPFC was demonstrated to reduce amygdala hyperactivity, thereby alleviating anxiety and negative emotions. Stimulation frequency is defined as the rate at which rTMS pulses are delivered. Low-frequency rTMS (1–5 Hz) is typically utilized to inhibit cortical activity, whereas high-frequency rTMS (10–20 Hz) is employed to activate cortical regions. The interhemispheric competition model, a theoretical model developed to explain rTMS applications in stroke rehabilitation, suggests that the balance of mutual inhibition between the two hemispheres via the corpus callosum is disrupted by stroke (78). Moreover, activity in the l-DLPFC is usually associated with positive emotions, while the r-DLPFC is more closely linked to negative emotions. Previous studies have indicated that depressive states are often associated with reduced l-DLPFC activity and a relative elevation in r-DLPFC activity (79). Hence, rTMS is frequently employed to activate the l-DLPFC (80) or inhibit the r-DLPFC (81), aiming to rebalance the activity in these regions and alleviate conditions such as insomnia. Lastly, research has also suggested that stimulating the r-DLPFC may result in comparatively fewer side effects, underscoring its better suitability as a target region in patients with lower tolerance to stimulation (81).

Our findings suggest that low-frequency rTMS targeting the r-DLPFC may be more effective than high-frequency rTMS of the l-DLPFC in improving sleep function and depression in patients with PSSD. Earlier studies have shown that different rTMS protocols are more suitable for specific symptoms of sleep disorders. For instance, right-side low-frequency rTMS appears to be more beneficial for patients experiencing difficulty falling asleep (82), while high-frequency stimulation of the l-DLPFC is more efficient in addressing the symptoms associated with early awakening and vivid dreaming (83). Additionally, the differences in rTMS efficacy may be related to individual variations in neural network structures, physiological characteristics, and symptom heterogeneity. In light of this aspect, some studies have recommended using neuro-navigation systems to select personalized stimulation targets, including a study that showed that 19 out of 27 patients with major depressive and generalized anxiety disorders had target sites in the L8Av region of the DLPFC (84). However, the clinical feasibility of implementing such personalized approaches in standard stroke rehabilitation settings presents several challenges. A nationwide survey of 1,129 physiatrists in South Korea revealed that while 86.1% expressed interest in utilizing neuro-navigation systems for rTMS therapy, only 7.4% reported having access to such systems within their institutions (85). The primary barriers identified included high device costs, lack of reimbursement coverage, and the need for specialized training. In contrast, the three localization methods mentioned in this study, based on standard brain localization, are more cost-effective, allowing for broader application in rTMS therapy.

Moreover, the process of systematically optimizing target selection and stimulation parameters requires further exploration. Although some patients may experience a sleep disorders relapse following rTMS treatment, scarce systematic research and supporting data are available on strategies for extending the therapeutic effects through long-term maintenance or personalized adjustment to stimulation frequencies. All these considerations underline the need for future studies to refine the rTMS therapeutic approach for PSSD, focusing on individualized treatment plans and sustained efficacy.