Patients with PSD were recruited from the Affiliated Hospital of North Sichuan Medical College between March 2023 and October 2025. All participants met the predefined inclusion and exclusion criteria. Before the commencement of the study, written informed consent was obtained from each patient or their legal guardian/immediate family member, ensuring full comprehension of the study’s purpose, procedures, potential risks, and benefits. The trial protocol was approved by the Medical Ethics Committee of the Affiliated Hospital of North Sichuan Medical College (approval numbers: 2023ER031-1) and registered in the Chinese Clinical Trial Registry (registration number: ChiCTR2300068730).

This study was designed as a single-center, single-blind, randomized controlled trial with an integrated multimodal evaluation system to comprehensively examine intervention effects and underlying mechanisms. Assessment modalities included the standardized swallowing assessment (SSA), fiberoptic endoscopic evaluation of swallowing (FEES), and suprahyoid muscle motor evoked potentials (MEP). The study aimed to systematically elucidate the intervention’s pathways and neuromodulatory mechanisms from both behavioral and neurophysiological perspectives, providing a more precise scientific basis for the rehabilitation of patients with PSD.

Block randomization was employed, with an independent statistician not involved in patient recruitment or assessment generating a random sequence using computer software. The sequence allocated patients in a 1:1:1 ratio to the sham group, affected rTMS group and bilateral rTMS group. Allocation concealment was ensured using sequentially numbered, opaque, sealed envelopes containing the group assignment. Eligible patients who provided informed consent were assigned to a group by a study coordinator who opened the corresponding envelope in sequential order, thereby determining the group allocation and treatment protocol. Blinding was applied to both patients and outcome assessors; only the treating operator was aware of the group assignment. All outcome measures were assessed at baseline (before treatment) and immediately post-intervention by the same therapist, who was blinded to group allocation and not involved in the intervention. This assessor received standardized training prior to the trial to ensure consistency in evaluation. Throughout the trial, adverse events (including epilepsy, headache, dizziness, syncope, dyspnea, scalp/neck skin redness, tinnitus, etc.) and participant dropout were recorded in detail to comprehensively evaluate treatment safety and tolerability.

This trial employed a random grouping design, consisting of a total of three groups. The significance level ($\alpha$) was set at 0.05, and the statistical power (1–$\beta$) was set at 0.9. Based on effect sizes, standard deviations, and preliminary experimental data reported in previous literature, PASS 15 software (version 15.0, NCSS, LLC, Kaysville, UT, USA) was used to estimate the sample size, resulting in a requirement of at least 17 cases per group, totaling a sample size of 51 cases [19, 20, 21]. To control for potential biases arising from sample dropouts, an additional 20% increase in sample size was added to the estimate, resulting in a final target of 63 cases, with 21 cases allocated to each group.

(1) Unilateral stroke confirmed by computed tomography or magnetic resonance imaging [22]; (2) First-ever stroke occurring within 2 weeks to 6 months after onset; (3) Age $\ge$18 years; (4) Presence of penetration and/or aspiration confirmed by FEES [4]; (5) Water swallowing test grade $\ge$3; (6) Signed informed consent obtained from the patient or their family members.

(1) Dysphagia attributable to neurological conditions other than stroke, such as Parkinson’s disease, Alzheimer’s disease, traumatic brain injury, cerebellar or brainstem strokes, and head/neck tumors; (2) Pregnancy or lactation; (3) Acutely ill or medically unstable patients (e.g., unstable hemodynamics, active progressive illness); (4) Inability to comply with FEES, MEP, or swallowing function assessments due to cognitive, behavioral, or communication impairments; (5) Contraindications to transcranial magnetic stimulation (e.g., intracranial metal implants, epilepsy) or severe adverse reactions during previous rTMS sessions; (6) Disease progression or recurrent stroke during the trial; (7) Withdrawal of informed consent or unwillingness to continue participation.

During the treatment phase, to simulate the real-world scenario of multimodal integrated rehabilitation, all patients received standardized conventional swallowing rehabilitation therapy alongside their group-specific interventions. This conventional therapy consisted of a 2-week regimen administered once daily. The conventional rehabilitation included the following components: swallowing function training, oral sensory stimulation, acupuncture therapy, and postural compensatory training. Swallowing function training, oral sensory stimulation, and postural compensatory training were conducted by qualified rehabilitation therapists [3, 4, 5, 23]. Acupuncture treatment was performed by a certified acupuncturist. The acupoint selection protocol followed the principles of “local point selection, stage-specific treatment, and integration of pattern differentiation with disease diagnosis”. Based on these principles, differentiated acupoint prescriptions were designed to address the distinct stages of PSD (oral phase, pharyngeal phase, and oral-pharyngeal mixed phase) [23, 24]. Additionally, stimulation was delivered using a figure-of-8 coil connected to a MagNeuro 60 transcranial magnetic stimulator (Nanjing Vishee Medical Technology Co., Ltd., Nanjing, China). The sham rTMS group received bilateral sham stimulation over the swallowing cortex; affected rTMS group received real stimulation on the affected hemisphere combined with sham stimulation on the contralesional side; and bilateral rTMS group received real stimulation to both hemispheres. The rTMS parameters were set as follows: frequency of 10 Hz, intensity at 100% of the resting motor threshold (RMT), train duration of 2 s, inter-train interval of 10 s, and a total of 1200 pulses per session. Treatment was administered once daily, 7 days a week, for 2 weeks. Sham stimulation was performed using the flipped-coil method, where the figure-of-8 coil was rotated 180°, aligning the induced magnetic field tangentially to the scalp. This approach produced acoustic artifacts similar to real stimulation without delivering an effective magnetic field intracranially. Prior validation tests confirmed that participants could not reliably distinguish between real and sham stimulation. Immediately after the first treatment session, a blinding assessment was conducted by asking patients, “Do you believe you received real or sham stimulation in this session?” to evaluate their awareness of group assignment. Based on previous research indicating that the excitatory effects of rTMS can last approximately 30 minutes and remain effective over intermittent periods of 20–30 minutes, the bilateral rTMS group received stimulation to the unaffected hemisphere first, followed by the affected hemisphere, with the aim of achieving an additive effect. During treatment, patients were positioned comfortably in a supine or seated posture. A figure-of-eight coil was placed over the primary motor cortex representation area (swallowing cortical hotspot) of the suprahyoid muscles, as determined by MEP mapping. Therapists administered the corresponding stimulation according to group assignment. Treatment was immediately discontinued if patients experienced intolerable discomfort.

This study systematically collected baseline demographic and clinical data from all enrolled patients prior to the trial commencement. Information included age, gender, stroke location, and stroke type, along with relevant outcome measures, to evaluate the comparability of baseline characteristics across groups. The primary and secondary outcome measures comprised the SSA, yale pharyngeal residue severity rating scale (YPR-SRS), and MEP. All assessments were conducted by a rehabilitation therapist blinded to the intervention at two time points: baseline (T0) and after two weeks of intervention (T1). Throughout the trial period, any adverse events (including seizures, headache, dizziness, syncope, dyspnea, local scalp/neck redness, and tinnitus-as well as participant dropouts) were meticulously documented to comprehensively evaluate treatment safety and tolerability. Prior to the trial, all assessors received standardized training on the relevant outcome measures to ensure consistency and accuracy in data collection.

Statistical analyses were performed using SPSS 27.0 (IBM Corp., Armonk, NY, USA). Normality of continuous variables was assessed using the Shapiro-Wilk test and histograms. Normally distributed data are presented as mean $\pm$ standard deviation ($\overline{x}$ $\pm$ s), while non-normally distributed data are expressed as median (interquartile range) [M (P25, P75)]. Categorical data are reported as number (percentage). The primary outcome measure, SSA score, and continuous exploratory secondary outcomes (e.g., MEP latency and amplitude) were compared between groups using the Kruskal-Wallis test. Within-group comparisons were conducted using the Wilcoxon signed-rank test. For ordinal categorical variables, including PAS and YPR-SRS scores, between-group comparisons were also performed with the Kruskal-Wallis test, and within-group changes were analyzed with the Wilcoxon signed-rank test. If the overall between-group test was significant, post hoc pairwise comparisons were performed using Dunn’s test with Bonferroni correction. A linear mixed-effects model was fitted for SSA scores at T0 and T1, with group and time as fixed effects and subject as a random effect, and Bonferroni correction was applied. Furthermore, the change in SSA score from baseline to the end of treatment ($\mathrm{\Delta }$SSA = SSA_T₀ – SSA_T₁) was calculated to assess the degree of swallowing function improvement. Based on previous literature, a $\mathrm{\Delta }$SSA $\ge$4 points was set as an exploratory threshold to define treatment response [30, 31]. The response rate was determined accordingly, and a chi-square test was used to compare response rates between groups. Effect sizes are expressed as Cohen’s d (0.20, 0.50, and 0.80 corresponding to small, medium, and large effects, respectively) [32, 33]. All statistical tests were two-tailed, and p 0.05 was considered statistically significant.

Between March 2023 and October 2025, a total of 75 patients with PSD were enrolled in this study. During the trial, five patients were lost to follow-up: two in the affected rTMS group withdrew one week after intervention due to conflicts with other treatment schedules, and three in the bilateral rTMS group dropped out before completing one week of intervention due to hospital discharge or personal reasons. Ultimately, 70 patients completed all scale assessments, FEES, and MEP data collection, and were included in the statistical analysis. The detailed trial flow is illustrated in Fig. 1. Analysis of baseline characteristics showed no statistically significant differences among the three groups in terms of age, sex, disease duration, stroke location, or stroke type (p $>$ 0.05). Furthermore, no significant intergroup differences were observed at baseline in SSA scores, PAS scores, pharyngeal residue scores, or MEP parameters (p $>$ 0.05), indicating that the baseline characteristics were well balanced across groups (Table 1).

Fig. 1.

Flowchart of the experiment. rTMS, Repetitive transcranial magnetic stimulation; SSA, Standardized swallowing assessment; PAS, Penetration aspiration scale; YPR-SRS, yale pharyngeal residue severity rating scale; MEP, Motor evoked potential. T0, Baseline before the intervention; T1, 2 weeks after rTMS intervention.

The SSA scores of the three groups at T0 and T1 are presented in Table 2 and Fig. 2A. Overall, the SSA scores improved significantly in all groups after treatment, with T1 scores being significantly lower than those at T0 (p 0.001). Further between-group comparisons revealed that at T1, the SSA score in the bilateral rTMS group was significantly lower than that in both the sham stimulation group and the affected rTMS group (p 0.001).

Fig. 2.

Comparison of SSA scores (A), PAS scores (B), vallecular residue scores (C), and pyriform sinus residue scores (D) before and after treatment in the three groups. ^*p 0.05; ^**p 0.01; ^*⁣**p 0.001.

To systematically quantify the independent and interactive effects of group and time on SSA scores, a linear mixed-effects model was constructed based on data from patients who completed the full intervention. The model results were as follows: a significant main effect of group (F = 54.920, p 0.001), indicating that the overall SSA scores differed significantly among the three groups from baseline to follow-up; a significant main effect of time (F = 12.640, p 0.001) suggesting a substantial overall change in SSA scores over time after intervention; and a significant group $\times$ time interaction (F = 7.230, p = 0.001), demonstrating statistically different trends in SSA score changes before and after treatment among the groups (Table 3).

To further evaluate the magnitude of improvement in swallowing function, the change in SSA score ($\mathrm{\Delta }$SSA) was calculated. The results showed that the $\mathrm{\Delta }$SSA in the bilateral rTMS group was significantly greater than that in both the sham group and affected rTMS group (p 0.001), indicating a more pronounced improvement in swallowing function in the bilateral rTMS group (Fig. 3A). The response rate, defined as the proportion of patients achieving a $\mathrm{\Delta }$SSA of at least 4 points, was 77.27% in the bilateral rTMS group, which was significantly higher than the rates in the sham group (32.00%) and the affected rTMS group (47.83%), with a statistically significant between-group difference ($\chi$² = 9.780, p = 0.008) (Table 2, Fig. 3B).

Fig. 3.

Comparison of the $\mathrm{\Delta }$SSA (A), SSA effective ratio (B), MEP amplitudes in the ipsilesional (C) and contralesional (D) regions among the three groups. ns, not significant; ^*p 0.05; ^**p 0.01; ^*⁣**p 0.001.

Additionally, an intention-to-treat analysis was conducted, applying the baseline observation carried forward (BOCF) method to the five patients who dropped out. The results showed that SSA scores still differed significantly among the three groups after treatment (p 0.001). Further pairwise comparisons confirmed that the SSA score in the bilateral rTMS group remained significantly lower than that in both the sham stimulation group and the affected-side rTMS group (p 0.01). A mixed-effects model constructed on this basis continued to support the above conclusions, yielding the following results: a significant main effect of group (F = 46.67, p 0.001); a significant main effect of time (F = 9.406, p = 0.0001); and a significant group $\times$ time interaction (F = 5.232, p = 0.006). Detailed data are provided in Supplemental Material.

The PAS scores of the three groups of patients at T0 and T1 are shown in Table 2 and Fig. 2B. Statistical analysis revealed that the PAS scores in the affected rTMS group and the bilateral rTMS group at T1 were significantly lower than those at T0 (p 0.001). However, the difference in scores for the sham rTMS group before and after treatment was not statistically significant (p = 0.126). Inter-group comparisons showed a significant overall difference in PAS scores among the three groups at T1 (p = 0.017). Further pairwise comparisons indicated that the PAS score of the bilateral rTMS group was significantly lower than that of the sham rTMS group (p = 0.017).

Furthermore, the proportion of patients achieving safe swallowing (PAS $\le$2, indicating penetration without aspiration) increased significantly: from 12% to 52% in the sham group, from 13% to 69.6% in the affected rTMS group, and from 18.2% to 77.3% in the bilateral rTMS group. The proportion achieving basic safe swallowing (PAS $\le$5, indicating that aspirated material can be cleared) also improved: from 80% to 84% in the sham group, from 73.9% to 87.0% in the affected rTMS group, and from 86.4% to 95.5% in the bilateral rTMS group.

Table 2 and Fig. 2C,D summarize the results of the vallecular residue and pyriform sinus residue scoring for the three groups of patients at T0 and T1. Overall, the residue scores for both the vallecula and the pyriform sinus showed significant improvement over time, with scores at T1 being significantly lower than those at T0 (p 0.05). Inter-group comparisons indicated statistically significant overall differences in vallecular residue scores (p 0.001) and pyriform sinus residue scores (p = 0.004) among the three groups at T1. Further pairwise comparisons revealed that for vallecular residue, the scores for the affected rTMS group and the bilateral rTMS group were both significantly lower than that of the sham rTMS group (p 0.05). Regarding pyriform sinus residue, the score for the bilateral rTMS group was significantly lower than those for the affected rTMS group and the sham rTMS group (p 0.05).

The MEP parameters for the three groups at T0 and T1 are summarized in Table 2 and Fig. 3C,D. Overall, the MEP amplitude significantly increased from T0 to T1 in all groups (p 0.001). Regarding MEP latency, a statistically significant difference between pre-and post-treatment values was observed only on the unaffected side in the bilateral rTMS group (p = 0.046). Inter-group comparisons at T1 revealed significant differences in MEP amplitude among the groups, both on the ipsilesional side (p = 0.023) and the contralesional side (p = 0.007), whereas no significant inter-group differences were found in latency (p $>$ 0.05). Further pairwise comparisons showed that the MEP amplitude on the unaffected side in the bilateral rTMS group was significantly higher than that in the sham group (p = 0.020). Moreover, the MEP amplitude on the affected side in the bilateral rTMS group was not only significantly higher than that in the sham stimulation group (p = 0.026) but also significantly higher than that in the affected-side rTMS group (p = 0.013).

At T1, this study calculated the Cohen’s D values for the changes in functional outcomes to assess the effect sizes of different indicators, with the effect sizes for each group summarized as follows (Table 2):

Sham rTMS Group: SSA score improvement demonstrated a medium effect size (D = 0.592); MEP amplitude showed large effects on both the unaffected side and affected side (D = 2.435 and 1.908, respectively); Changes in MEP latency were classified as a small effect (unaffected side D = 0.219; affected side D = 0.317).

Affected rTMS Group: SSA score improvement showed a large effect size (D = 0.969); MEP amplitude on both the unaffected side and affected side also exhibited large effects (D = 2.295 and 1.391, respectively); Changes in MEP latency remained a small effect (unaffected side D = 0.490; affected side D = 0.318).

Bilateral rTMS Group: SSA score improvement resulted in a large effect size (D = 2.339); MEP amplitude on both the unaffected side and affected side demonstrated large effects (D = 2.944 and 2.622, respectively); Notably, the MEP latency on the affected side achieved a large effect size (D = 0.810), while the change in latency on the unaffected side remained a small effect (D = 0.251).

The blinding assessment conducted after the first treatment session revealed that the proportion of patients who correctly identified their actual group assignment was 68% in the sham group, 60% in the affected rTMS group, and 76% in the bilateral rTMS group. A chi-square test showed no statistically significant difference in the accuracy rates among the three groups ($\chi$² = 1.471, p = 0.479). These results indicate that the blinding method was effective, and patients could not reliably distinguish between real and sham stimulation, confirming the successful implementation of blinding in this study.

All patients successfully completed the treatment, with no cases of dropout due to intolerance, indicating overall good tolerability. During the study, a total of 5 cases of transient scalp discomfort were reported (1 case in the sham rTMS group, 2 cases in the affected rTMS group, and 2 cases in the bilateral rTMS group). Symptoms were alleviated after pausing stimulation and taking a short break. No other serious adverse events, such as headache, epilepsy, tinnitus, or psychological discomfort, were observed.

Swallowing function involves multi-stage coordinated activities from the oral cavity to the esophagus, relying on a complex neural network regulated by bilateral cerebral cortices, subcortical structures, the brainstem, and the cerebellum [31, 33]. Within this network, the bilateral cerebral hemispheres jointly control swallowing via corticofugal projections, typically with one hemisphere serving as the dominant side and the other playing a synergistic and compensatory role [11, 34]. When stroke injures key nodes of this network-such as the primary motor cortex, anterior insula, anterior cingulate cortex, or frontal operculum-or disrupts the white matter pathways connecting them, the integrity and coordination of the network are compromised. The core pathological mechanisms involve desynchronization of cortico-subcortical neural circuits and impaired neuromuscular control of muscles such as the suprahyoid group, ultimately leading to clinical symptoms including pharyngeal residue, aspiration, and delayed swallowing initiation [3, 35, 36]. This understanding of network dysfunction provides a theoretical basis for the use of neuromodulation techniques, such as rTMS, to target and repair neural circuits.

The bilateral rTMS intervention strategy adopted in this study was primarily based on the theoretical frameworks of the interhemispheric competition model and the dual-mode balance recovery model [9, 11, 17, 18]. Following stroke, the excitability of the affected hemisphere decreases, while the unaffected hemisphere may become hyperactive due to loss of interhemispheric inhibition, thereby impeding the recovery of swallowing function [12]. In contrast to conventional unilateral intervention approaches, bilateral high-frequency rTMS (e.g., 10 Hz) simultaneously modulates excitability in both hemispheres: it promotes functional reorganization in the affected hemisphere while suppressing excessive activity in the unaffected hemisphere, ultimately restoring the overall function of the swallowing network. As swallowing is a physiologically bilateral cortical process, its neural network relies on compensatory mechanisms from the unaffected hemisphere after unilateral damage [37, 38, 39]. Thus, bilateral synchronous stimulation may activate potential compensatory pathways, yielding synergistic effects. Although there is no unified standard for optimal rTMS parameters in PSD treatment, high-frequency stimulation ($\ge$5 Hz) applied to bilateral swallowing cortices (e.g., the suprahyoid motor cortex) at an intensity of 80%–120% RMT represents a common strategy [15, 19, 21, 40]. This study employed a 10 Hz bilateral protocol to enhance cortical excitability and long-term plasticity (LTP) plasticity, thereby optimizing neuroplasticity and providing a theoretical basis for multi-target neuromodulation strategies.

The findings of this study clearly demonstrate that bilateral high-frequency rTMS offers significant advantages in improving swallowing function in patients with PSD. Specifically: Comprehensive improvement in swallowing function: The bilateral rTMS group showed significantly greater improvements in the SSA score (reflecting overall swallowing ability), PAS score (reflecting swallowing safety), and pharyngeal residue score (reflecting swallowing efficiency) compared to both the sham stimulation group and the unilateral rTMS group. High clinical response rate: Using a $\mathrm{\Delta }$SSA $\ge$4 points was set as an exploratory threshold to define treatment response, the response rate in the bilateral rTMS group reached 77.27%, significantly higher than the other two groups, with an effect size indicating a “large effect”, underscoring the clinical value of this intervention. Thus, these behavioral results align with previous studies, confirming that the bilateral rTMS strategy holds superior advantages in enhancing swallowing function, improving safety, and clearing efficiency, thereby offering a comprehensive approach to addressing the multifaceted challenges of PSD and providing strong evidence for clinical application [41, 42, 43, 44].

Changes in MEPs provide an objective neurophysiological metric for elucidating the mechanisms of rTMS. The results of this study showed that the MEP amplitude significantly increased from baseline in all intervention groups after treatment, suggesting that conventional rehabilitation combined with rTMS may effectively enhance cortical excitability in swallowing-related areas. Notably, the bilateral rTMS group demonstrated a unique advantage: the improvement in MEP amplitude on the affected side was significantly greater than that in the affected-side rTMS group, indicating that bilateral stimulation may more effectively facilitate the recruitment and synchronized firing capacity of neural pathways in the affected hemisphere, potentially by modulating interhemispheric interactions [13, 14]. Particularly important is that only the bilateral rTMS group showed a significant shortening of MEP latency on the unaffected side. Given that latency reflects the neural conduction velocity from the cortex to the target muscles, improvement in this measure suggests that bilateral rTMS not only increased cortical excitability but may also have enhanced the conduction speed of swallowing-related neural pathways, possibly by optimizing myelination or synaptic transmission efficiency [45, 46, 47]. The underlying mechanisms may involve rTMS-induced modulation of neurotransmitter systems (e.g., GABA and glutamate) and upregulation of brain-derived neurotrophic factor expression, as observed in other studies [47, 48]. Therefore, bilateral rTMS may promote neural remodeling and functional recovery related to swallowing through a dual mechanism of “enhancing excitability” and “optimizing conductivity”.

All patients completed the treatment without any dropouts due to intolerance, indicating good overall tolerability of the intervention protocols across all groups. A total of five cases of transient scalp discomfort were reported among the three groups, with no significant difference in incidence rates between groups. Symptoms resolved after briefly pausing stimulation and allowing a short rest. No serious adverse events such as headache, seizures, tinnitus, or psychological discomfort were observed. These findings are consistent with the majority of reported studies, further supporting the favorable safety profile of rTMS in treating post-stroke dysphagia [49]. In this study, bilateral rTMS did not increase the risk of adverse reactions, demonstrating that both dual-target combined intervention and single intervention modalities possess reliable safety and clinical applicability [44].