Comparative efficacy of non-pharmacological interventions for persistent postural-perceptual dizziness: a systematic review and network meta-analysis of balance, anxiety, and depression outcomes Authors: Jiawei Shen, Xinyu Zhao, Yan Huo, Shaoli Chen, Wenhao Deng, Wei Zheng, Jianning Zhang Published: 2026 ABSTRACT AbstractIntroductionPersistent postural-perceptual dizziness (PPPD) is a chronic functional vestibular disorder characterized by persistent dizziness, postural instability, and visual hypersensitivity. Although several non-pharmacological interventions have been proposed, their comparative efficacy remains unclear.MethodsPubMed, Embase, and the Cochrane Library were searched from inception to January 2026 for randomized controlled trials involving adults with PPPD or equivalent conditions. Interventions included CBT, vestibular rehabilitation therapy (VRT), neuromodulation, galvanic vestibular stimulation (GVS), and combined approaches. Random-effects network meta-analysis was performed using standardized mean differences (SMDs) and 95% confidence intervals (CIs).ResultsEleven RCTs comprising 518 patients were included. For balance outcomes (518 participants), neuromodulation significantly outperformed control (SMD −1.22, 95% CI −1.90 to −0.54). It also demonstrated superiority over CBT (SMD −1.03, 95% CI −1.96 to −0.09) and VRT (SMD −1.33, 95% CI −2.49 to −0.18). For anxiety outcomes (518 participants), neuromodulation demonstrated significant superiority over control (SMD −0.87, 95% CI −1.40 to −0.33), CBT (SMD −0.88, 95% CI −1.62 to −0.14), and Advanced VRT (SMD −1.36, 95% CI −2.31 to −0.41). No significant effects were observed for any intervention on depression outcomes; moderate-to-substantial heterogeneity was detected (I2: 47.8%–68.9%), while no significant inconsistency was found. Most studies (81.8%) were at high risk of bias, with overall evidence quality rated low to moderate.ConclusionsNeuromodulation may provide relative benefits for balance and anxiety in PPPD, whereas evidence for alleviation of depression remains inconclusive. Given the limited and low-quality evidence base, further high-quality randomized controlled trials are needed.RegistrationPROSPERO (CRD420261285141).Supplementary InformationThe online version contains supplementary material available at 10.1007/s00415-026-13891-1.Keywords: Persistent postural-perceptual dizziness, Network meta-analysis, Vestibular rehabilitation, Anxiety, Balance control, Depression Introduction Dizziness and balance disorders impose a substantial global health burden, contributing to functional impairment, reduced quality of life, and increased healthcare utilization [1]. Among chronic vestibular syndromes, persistent postural-perceptual dizziness (PPPD) is a common and highly disabling condition. Formally defined by the Bárány Society in 2017, PPPD is characterized by persistent dizziness, unsteadiness, or non-spinning vertigo present on most days for 3 months or longer, exacerbated by upright posture, active or passive motion, and exposure to complex or moving visual stimuli [2]. This diagnosis unified previously described entities—including chronic subjective dizziness (CSD), phobic postural vertigo (PPV), and visual vertigo—under a cohesive framework [2, 3]. In the International Classification of Diseases, 11th Revision (ICD-11), PPPD is classified under code AB32.0 as a functional vestibular disorder, reflecting its pathophysiological basis in maladaptive sensory integration rather than structural vestibular deficits. Epidemiological studies indicate that PPPD is the second or third most common diagnosis in tertiary dizziness clinics, accounting for approximately 15–20% of patients with chronic dizziness, underscoring its clinical significance [2, 4]. The impact of PPPD on patients’ lives is profound. Persistent symptoms often lead to avoidance behaviors, functional disability, increased fall risk, and significant psychological distress [5, 6]. The prevalence of anxiety and depression in patients with vestibular disorders—including PPPD—is markedly higher than that in the general population. A recent narrative review highlighted that anxiety and depression are common psychiatric comorbidities in PPPD patients [7]. These comorbidities exacerbate core vestibular symptoms, increase healthcare costs, prolong disability, and diminish health-related quality of life [5, 6]. The underlying pathophysiology involves maladaptive neuroplasticity in multisensory integration networks combined with heightened anxiety-mediated postural control, creating a vicious cycle in which vestibular dysfunction triggers anxiety, which in turn perpetuates vestibular symptoms [2, 8]. Consequently, interventions targeting both vestibular and psychological dimensions have been proposed. Mounting evidence supports non-invasive interventions for PPPD. Vestibular rehabilitation therapy (VRT)—comprising gaze stabilization, balance retraining, and habituation exercises—improves postural control and reduces dizziness-related disability [9, 10]. Advanced protocols incorporating virtual reality (VR) or home-based version have shown enhanced outcomes in recent randomized controlled trials (RCTs) [11, 12]. Galvanic vestibular stimulation (GVS), which modulates vestibular afferent activity through cutaneous electrical stimulation, has emerged as a promising adjunctive option [13, 14]. Neuromodulation techniques, including repetitive transcranial magnetic stimulation (rTMS) targeting the dorsolateral prefrontal cortex has also demonstrated efficacy in alleviating both dizziness and anxiety in PPPD patients [15, 16]. Non-invasive vagus nerve stimulation (nVNS), which activates the auricular branch of the vagus nerve and modulates vestibular–autonomic integration through central projections to the nucleus tractus solitarius, has shown significant therapeutic potential in PPPD [17]. Additionally, cognitive-behavioral therapy (CBT) addresses maladaptive illness perceptions, catastrophic thinking, and avoidance behaviors, with sustained benefits in reducing dizziness handicap and comorbid anxiety [18, 19]. However, the optimal treatment strategy remains unclear due to a fragmented evidence base. RCTs have typically compared individual interventions against placebo or usual care, with few direct head-to-head comparisons between active treatments. Moreover, studies predating the 2017 diagnostic criteria used heterogeneous diagnostic labels, complicating synthesis. Conventional pairwise meta-analyses cannot simultaneously compare multiple competing treatments or integrate evidence from direct and indirect comparisons [20]. Network meta-analysis (NMA) overcomes these limitations by synthesizing evidence through a network of common comparators, enabling simultaneous ranking of all interventions and incorporation of indirect treatment effects [20–22]. This approach is particularly valuable for PPPD, where the absence of comprehensive head-to-head trials across non-invasive modalities has hindered evidence-based treatment sequencing. This systematic review and network meta-analysis therefore aims to evaluate the comparative efficacy and acceptability of non-invasive interventions, including VRT, neuromodulation (nVNS, rTMS), GVS, and psychotherapy (CBT) in reducing symptom severity, anxiety, and depression in patients with PPPD, and to generate probability rankings to guide evidence-based treatment selection. Methods General guidelines of the study The current NMA followed an a priori defined unpublished protocol (available upon reasonable request to the corresponding author) and was performed according to the latest preferred reporting items for systematic reviews and meta-analyses (PRISMA) 2020 guideline [23] and AMSTAR2 (assessing the methodological quality of systematic reviews) guidelines [24]. The current study had been registered in PROSPERO (CRD420261285141) with the first submission date of January 14, 2026 and final approval of registration date by January 15, 2026. Search strategy To enhance the reliability and clinical relevance of the present NMA, candidate non- pharmacological interventions were pre-specified based on recommendations from previous systematic reviews, evidence-based clinical guidelines, and expert consensus statements, thereby defining the intervention categories and comparison nodes to be included. Subsequently, a systematic literature search was conducted in PubMed, Ovid embase, and the Cochrane Library, from database inception to the final search date (26/01/2026), to identify RCTs meeting the eligibility criteria (see Supplementary document). No restrictions other than language were applied during the search process, and only RCTs were considered. In addition, manual searches were performed by screening the reference lists of all included studies, relevant systematic reviews, clinical practice guidelines, and previous pairwise meta-analyses to identify potentially eligible studies that may have been missed by the electronic search. In this study, the definition of PPPD primarily followed the diagnostic criteria proposed by the Bárány Society in 2017 and subsequently incorporated into the ICD-11. For studies published prior to the formalization of the PPPD construct that used the diagnostic terms CSD or PPV, inclusion was permitted if the reported clinical phenotype and symptom duration met the predefined criteria for diagnostic equivalence with PPPD. Inclusion and exclusion criteria This study included published RCTs enrolling adult patients with PPPD. Eligible studies were required to adopt a waitlist control, usual care control, sham intervention control, or another active non-pharmacological comparator, and to compare the effectiveness of different non-pharmacological intervention strategies in individuals with PPPD. The target population at first comprised patients with PPPD and its historical diagnostic equivalents, including CSD and PPV, and the scope of comparisons encompassed a broad range of non-invasive interventions. During the subsequent systematic search and study selection process, to reduce potential clinical heterogeneity and to improve comparability across interventions, the eligibility criteria were further refined. Specifically, only published RCTs evaluating the effectiveness of the following core non-pharmacological intervention pathways in patients with PPPD or diagnostically equivalent conditions (functional vestibular disorder characterized by chronic non-vertiginous dizziness, postural instability, and hypersensitivity to visual motion) were included: vestibular rehabilitation, psychological interventions (e.g., CBT or acceptance and commitment therapy), neuromodulation, GVS and their combinations. Accordingly, non-pharmacological interventions were included in the present NMA. Studies in which the primary diagnosis was another vestibular or neurological disorder and from which PPPD-specific data could not be separately extracted (e.g., benign paroxysmal positional vertigo (BPPV), Ménière’s disease, vestibular migraine, or clearly defined central vertigo) were also excluded. A detailed description of the inclusion and exclusion criteria is presented in Supplementary document. Exclusion criteria comprised: (1) non-clinical or observational studies; (2) non-randomized trials; (3) studies involving children or adolescents; (4) failure to report the prespecified primary outcome(s) (defined below); (5) interventions not belonging to the predefined non-pharmacological intervention categories; and (6) studies in which pharmacological, invasive or surgical treatments constituted the primary intervention. Where multiple publications were derived from the same study population, only the report with the largest sample size and the most complete data was included. NMA relies on two key methodological assumptions—transitivity and similarity—for which adequate randomization is a fundamental prerequisite. Inclusion of trials with poor randomization, resulting in significant between-group differences in baseline dizziness severity (e.g., baseline Dizziness Handicap Inventory (DHI) scores), may violate these assumptions. To enhance the credibility of the transitivity and similarity assumptions, RCTs demonstrating significant imbalances in baseline severity between comparison groups were therefore excluded. Data extraction Two authors (JW Shen and XY Zhao) independently screened the studies and extracted the relevant information from the manuscripts. In cases of discrepancy, the corresponding author (JN Zhang) mediated. If data of outcomes of interest were missing from published reports, the corresponding authors or co-authors were approached to obtain additional data. If the data of outcomes of interest still were unavailable, we will only count the other available outcomes of interest from this RCT in the NMA of other outcomes. Node setting Interventions were categorized into nodes based on similarities in mechanism and clinical application. VRT and advanced VRT were treated as separate nodes. Neuromodulation included rTMS and nVNS. GVS was retained as a separate node due to its distinct mechanism of action, directly modulating peripheral vestibular input, in contrast to centrally acting neuromodulation techniques. Control conditions (waitlist, usual care, and sham) were combined to ensure network connectivity. Statistical analysis Network meta-analysis (NMA) was performed using R software (version 4.4.2) and Stata 18.0. For continuous outcomes, effect sizes were expressed as standardized mean differences (SMDs) with corresponding 95% confidence intervals (95% CIs). Because different studies used various instruments to assess balance, anxiety, and depression outcomes, SMDs were adopted to standardize the pooled estimates across scales. All analyses were conducted using random-effects models to account for potential clinical and methodological heterogeneity among studies. Statistical heterogeneity within the network was assessed using Cochran’s Q test and the I2 statistic, with higher I2 values indicating greater heterogeneity. Global inconsistency was evaluated using the design-by-treatment interaction model, while local inconsistency was assessed using node-splitting analysis when closed loops were available within the network. In networks without closed loops, local inconsistency assessment was not feasible. Potential publication bias and small-study effects were explored visually using comparison-adjusted funnel plots. Treatment ranking probabilities were estimated using P-scores (approximately equivalent to SUCRA values), with higher P-scores indicating a greater likelihood of superior treatment efficacy for the corresponding outcome. All statistical tests were two-sided, and a p-value < 0.05 was considered statistically significant. Result Literature screening process and results The study selection process is presented in Fig. 1. A total of 5261 records were identified through database searching, including PubMed (n = 890), Cochrane Library (n = 1061), and Ovid Embase (n = 3310) (supplementary document). After removing 401 duplicate records, 4860 records remained for title and abstract screening. During the screening phase, 3423 records were excluded as they were reviews, letters, editorials, errata, commentaries, or books. Subsequently, 1437 reports were sought for retrieval, of which 237 could not be obtained. The remaining 1200 articles were assessed for eligibility through full-text review. After full-text evaluation, studies were excluded for the following reasons: non-English publications (n = 331), not relevant interventions (n = 7), not the population of interest (n = 846), and no relevant outcomes reported (n = 5). Finally, 11 studies were included in the systematic review and NMA. Schematic diagram of the literature screening process and results Basic characteristics and risk of bias analysis of included studies A total of 11 RCTs [11, 12, 15–17, 19, 25–29] were included in this systematic review and NMA, comprising 518 patients with PPPD or diagnostically equivalent conditions. The included studies evaluated a range of non-pharmacological therapeutic interventions (Table 1). Overall, the study populations, diagnostic frameworks, and outcome measures were considered sufficiently comparable across trials, supporting the transitivity assumption underlying the network meta-analysis. Table 1General information in the included literatureAuthorYearCountryParticipantArmsFollow-upSample sizeAverage age (years) Mean (SD)Male sex ratio (%)Intervention measureOutcome indexExperimental groupControl groupExperimental groupControl groupExperimental groupControl groupExperimental groupControl groupBalance outcomeAnxiety outcomeDepression outcomeEdelman2012AustraliaCSD combined with psychological symptomstwo-arm4 weeks202146.78 (12.65)27%PsychotherapyControlDHI/DSIDASS-21DASS-21Eren2018GermanyPPPDtwo-arm4 weeks/8 weeks10938.8 (9.8)43.2 (14.1)40%44.4%NeuromodulationControlNASHADS-AHADS-DChoi2021KoreaPPPDtwo-arm4 weeks151371.5 (65,78)75 (67,78) 46.67%38.46%Advanced VRTVRTDHI/ADL/VVASBAI/Yamaguchi2022JapanPPPDtwo-arm1 weeks121458 (17.1)63.5 (15.9)25%21.43%Advanced VRTVRTNPQ score/DHIHADS-AHADS-DTeh2023MalaysiaPPPDtwo-arm4 weeks/12 weeks151544.00 (10.61)48.52 (8.72)40%33.3%Advanced VRTVRTDHIDASS-21DASS-21Al-Omari2025JordanPPPDtwo-arm6 weeks202039.60 (8.75)40.90 (13.02)50%65%Advanced VRTVRTNPQ/DHI/FES-I/DGIHADS-AHADS-DAzeez Alsaad2025IranPPPDthree-arm6 weeks9932.44 (10.82)38.00 (13.13)22.2%22.2%VRTGVSDHIHADS-AHADS-DAzeez Alsaad2025IranPPPDthree-arm6 weeks9932.44 (10.82)33.22 (9.96)22.2%33.3%VRTVRT+GVSDHIHADS-AHADS-DAzeez Alsaad2025IranPPPDthree-arm6 weeks9938.00 (13.13)33.22 (9.96)22.2%33.3%GVSVRT+GVSDHIHADS-AHADS-DHerdman2022United KingdomPPPDtwo-arm4 months202044.60 (16.96)44.30 (17.44)20%20%psychotherapyVRTDHI//VVAS/%TSI/GAD-7PHQ-9Jia2024ChinaPPPDtwo-arm2 weeks/4 weeks222051.0 (15.3)52.6 (13.6)22.7%30.0%NeuromodulationControlDHIHAMAHAMDLi2025ChinaPPPDtwo-arm2 weeks/1 month/3 month333354.1 (15.8)56.2 (12.1)39.4%33.3%NeuromodulationControlDHI/VASHAMAHAMDLimburg2021Germanyfunctional vertigo or dizzinesstwo-arm3 month/12 month817853.7 (15.4)53.5 (15.1)39.5%37.2%psychotherapyControlVHQ/VSSBAIBDI-II General information in the included literature Risk of bias was assessed using the Cochrane Risk of Bias 2.0 tool. Overall, most included studies were judged to be at high risk of bias (81.8%), while the remaining showed some concerns (18.2%). In the domain of randomization (D1), 45.5% of studies were rated as low risk, 45.5% as having some concerns, and 9.1% as high risk. For deviations from intended interventions (D2), 27.3% were low risk, 36.4% showed some concerns, and 36.4% were high risk. Missing outcome data (D3) were largely at low risk (81.8%), whereas outcome measurement (D4) showed greater uncertainty, with 18.2% low risk, 45.5% some concerns, and 36.4% high risk. Notably, selective reporting (D5) demonstrated the highest risk, with 81.8% of studies rated as high risk. Blinding of participants and personnel was adequately implemented in only 27.3% of studies, indicating a considerable risk of performance bias (Fig. 2). The results of GRADE evaluation have been listed in the appendix. In brief, the overall quality of evidence of the overall NMA was low to medium (supplementary document). Results of the risk of bias evaluation of the included literature Net meta-analysis Across the three outcomes (BALANCE, ANXIETY, DEPRESSION), the network comprised seven interventions (control, CBT, VRT, advanced VRT, neuromodulation, GVS, VRT+GVS) with consistent topology across analyses. Results were interpreted following PRISMA-NMA and Cochrane recommendations, integrating relative effects, ranking (P-score, representing the frequentist analog of the surface under the cumulative ranking curve (SUCRA)), heterogeneity, inconsistency, and small-study effects. Figure 3 depicts the entire geometric distribution of the treatment arms. To be specific, in Fig. 3, the lines between nodes represent direct comparisons in various trials, and the size of each circle is proportional to the size of the population involved in each specific treatment. The thickness of the lines is proportional to the number of trials connected to the network. Network diagram of interventions for a balance; b anxiety; c depression. The systematic review included 518 unique participants. Due to the inclusion of multi-arm studies, cumulative node sample sizes may exceed the total number of unique participants Balance outcomes For balance outcomes, the NMA included seven interventions across 518 participants and nine direct comparisons. The network was relatively sparse, with most comparisons informed by one to three trials. Control and VRT were the most connected nodes, whereas GVS and VRT + GVS were supported by smaller sample sizes (Fig. 3a). Compared with control, neuromodulation demonstrated a statistically significant improvement in balance (SMD −1.22, 95% CI −1.90 to −0.54). No other interventions showed statistically significant differences. In league table. Neuromodulation was also superior to CBT (SMD −1.03, 95% CI −1.96 to −0.09) and VRT (SMD −1.33, 95% CI −2.49 to −0.18). Other comparisons were not statistically significant and were generally associated with wide confidence intervals (Fig. 4a and Fig. 5a). Forest plot of SMDs for a balance; b anxiety; c depression League table of pairwise difference estimates in balance outcomes (lower triangle) and the probabilities that one treatment is better than another (upper triangle). a Balance; b anxiety; c depression. The league table presents pairwise differences in balance handicap outcomes with 95% confidence intervals (CI) derived from the NMA. The lower triangle shows standardized mean differences (SMDs) between interventions, while the upper triangle presents the probabilities that the treatment listed in the column is more effective than the treatment listed in the row. Results that exclude the null value are highlighted in bold. For each comparison, the treatment listed in the lower row or right-most column serves as the reference treatment Ranking analysis indicated that neuromodulation had the highest probability of being the most effective intervention, followed by control and CBT, whereas GVS, VRT + GVS, and advanced VRT occupied intermediate positions, and VRT ranked lowest (Fig. 7a). SUCRA diagram for a balance; b anxiety; c depression Moderate to substantial heterogeneity was observed in balance outcomes (I2 = 64.5%, τ2 = 0.200; Q = 16.88, df = 6, p = 0.0097), primarily arising from within-design variability (Q = 15.04, p = 0.0102), while between-design inconsistency was not statistically significant (Q = 1.84, p = 0.175). The control versus neuromodulation comparison contributed the largest proportion of heterogeneity (Q = 11.40, p = 0.0034). No evidence of global inconsistency was detected, and node-splitting analysis showed no significant differences between direct and indirect estimates (all p = 0.463). The comparison-adjusted funnel plot did not indicate clear asymmetry (Fig. 6a). Comparison-adjusted funnel plot for a balance; b anxiety; c depression Anxiety outcomes For anxiety outcomes, the NMA included seven interventions involving 518 participants. Control (n = 166) and VRT (n = 86) were the most connected nodes, contributing the majority of direct comparisons, whereas GVS (n = 18) and VRT + GVS (n = 18) were supported by relatively small sample sizes (Fig. 3b). In the random-effects model, neuromodulation was the only intervention associated with a statistically significant reduction in anxiety compared with control (SMD −0.87, 95% CI −1.40 to −0.33). Although several other interventions showed effect estimates favoring treatment, none reached statistical significance and the corresponding confidence intervals were wide (Fig. 4b). Network estimates from the league table were broadly consistent with the direct comparisons. Neuromodulation showed a tendency toward greater improvement in anxiety compared with Control (SMD −0.87, 95% CI −1.40 to −0.33), CBT (SMD −0.88, 95% CI −1.62 to −0.14), and advanced VRT (SMD −1.36, 95% CI −2.31 to −0.41). However, most other pairwise comparisons were not statistically significant and were associated with considerable uncertainty, as reflected by wide confidence intervals (Fig. 5b). Ranking analysis indicated that neuromodulation had the highest probability of being the most effective intervention, followed by GVS and control. CBT, VRT + GVS and VRT occupied intermediate positions, whereas advanced VRT ranked lowest (Fig. 7b). For anxiety outcomes, the network demonstrated moderate heterogeneity (I2 = 47.8%, τ2 = 0.098), with a non-significant Cochran’s Q test (Q = 11.49, df = 6, p = 0.074), suggesting limited between-study variability. No significant global inconsistency was observed (Q = 2.26, df = 1, p = 0.133). Decomposition of the Q statistics indicated that neither within-design heterogeneity (Q = 9.23, p = 0.100) nor between-design inconsistency (Q = 2.26, p = 0.133) reached statistical significance. Design-specific analysis suggested that heterogeneity was primarily driven by the control versus neuromodulation comparison (Q = 8.55, p = 0.014). Node-splitting analysis did not reveal any significant inconsistency between direct and indirect evidence (all p > 0.05). The comparison-adjusted funnel plot did not suggest substantial small-study effects (Fig. 6b). Depression outcomes For depression outcomes, the network included seven interventions and formed a connected but relatively sparse structure. Control was the largest and most central node, followed by CBT and VRT. Most direct comparisons were centered on control, resulting in a predominantly star-shaped configuration. Direct comparisons between active interventions were limited and were mostly supported by single studies, such as those involving GVS and VRT + GVS. VRT connected multiple interventions and acted as a secondary hub within the network. No closed loops were observed (Fig. 3c). The forest plot based on a random-effects model showed that none of the interventions demonstrated statistically significant differences in depression outcomes compared with control, as all confidence intervals crossed zero (Fig. 4c). The league table analysis was consistent with the forest plot findings, showing that none of the pairwise comparisons between interventions reached statistical significance, as all confidence intervals crossed zero. Some comparisons, such as neuromodulation versus control (SMD −0.58, 95% CI −1.24 to 0.09), approached statistical significance; however, the confidence intervals remained wide. Overall, most estimates were imprecise and associated with substantial uncertainty (Fig. 5c). Ranking analysis based on cumulative probability curves indicated that neuromodulation had the highest probability of being ranked among the most effective interventions. Control and CBT also showed relatively high cumulative probabilities across upper ranking positions. In contrast, VRT consistently showed the lowest cumulative probability of being ranked highly (Fig. 7c). Heterogeneity analysis indicated moderate to substantial variability within the network (I2 = 68.9%, p = 0.012). No inconsistency could be assessed, as no comparisons contained both direct and indirect evidence due to the absence of closed loops in the network. The comparison-adjusted funnel plot did not indicate evident small-study effects (Fig. 6c). Discussion This NMA systematically evaluated the comparative efficacy of multiple non-invasive interventions across different symptom domains in patients with PPPD. The findings reveal a nuanced landscape of therapeutic efficacy, with treatment effects varying substantially across outcome domains and intervention types. However, the findings should be interpreted within the context of the limited evidence base for non-invasive interventions in PPPD. Although numerous potentially relevant studies were identified, only 11 randomized controlled trials met the inclusion criteria, resulting in relatively small overall and treatment-node sample sizes. Therefore, the present findings should be interpreted as relative estimates rather than definitive evidence of treatment superiority. In addition, variability in outcome measures across studies may further complicate direct comparisons between interventions. Different instruments were used to assess similar symptom domains across studies. Although standardized mean differences were applied to improve comparability across different instruments, the findings should still be interpreted with caution. Balance outcome For balance outcomes, neuromodulation was the only intervention that demonstrated a statistically significant improvement compared with control (SMD = −1.22, 95% CI −1.90 to −0.54). Although other interventions, including VRT, advanced VRT, and combined approaches, did not reach statistical significance, their effect estimates generally favored improvement [30]. This finding is novel and significant, as prior systematic reviews have not identified a clear "gold standard" treatment for PPPD, often concluding that the evidence base is sparse and of low quality. From a mechanistic perspective, the significant therapeutic effects observed for neuromodulation are supported by a well-established neurobiological rationale. PPPD is increasingly conceptualized not as a primary peripheral vestibular disorder, but as a central functional condition characterized by maladaptive sensory integration and predictive processing, involving dysfunctional interactions among vestibular, visual, and limbic systems [31]. Clinically, patients often exhibit increased visual dependence [32], reduced utilization of vestibular inputs, and rigid postural control strategies [33], which can be interpreted as manifestations of abnormal sensory reweighting [34] and a high-threat postural control mode. Within this framework, neuromodulatory techniques, such as rTMS and GVS, may exert their therapeutic effects by directly targeting the underlying dysfunctional neural networks, thereby conferring potential mechanistic advantages. rTMS, through modulation of cortical excitability in regions such as the dorsolateral prefrontal cortex and temporoparietal junction [35], may influence key functional networks involved in multisensory integration, spatial orientation, and emotional regulation [15]. In contrast, GVS delivers electrical stimulation to vestibular afferents and can modulate activity within brainstem vestibular nuclei and their ascending pathways, thereby influencing vestibulo-cortical integration [36]. Existing evidence suggests that GVS can alter postural control strategies and facilitate sensory reweighting processes [37], potentially reducing excessive reliance on visual and somatosensory inputs. This mechanism is broadly consistent with current pathophysiological models of PPPD, which emphasize disturbed sensory integration and maladaptive postural control. Anxiety outcome The finding that neuromodulation was the only intervention to demonstrate a statistically significant reduction in anxiety symptoms (SMD = −0.87, 95% CI −1.40 to −0.33) is notable and somewhat unexpected, particularly in light of the non-significant effect observed for CBT. The relationship between PPPD and anxiety is well established and forms a core component of its pathophysiology [38]. PPPD evolved conceptually from earlier diagnostic constructs such as PPV, and its current diagnostic criteria, as defined by the Bárány Society [1], explicitly recognize the role of anxiety and threat perception in symptom exacerbation [2]. Earlier work on CSD described a psychophysiological framework in which anxiety-related hypervigilance and vestibular symptoms interact bidirectionally, contributing to persistent dizziness and maladaptive avoidance behaviors [39, 40]. Within this theoretical framework, CBT is considered a well-founded intervention, primarily aiming to disrupt the maladaptive cycle between anxiety and vestibular symptoms by modifying catastrophic cognitions, reducing avoidance behaviors, and regulating autonomic responses [41]. Previous studies, including meta-analyses examining CBT as an adjunct to other interventions, have suggested that CBT may provide benefits in improving dizziness-related functional impairment and psychological distress [18]. However, it should be noted that the current evidence regarding the efficacy of CBT in PPPD remains inconsistent, with a lack of agreement across studies. The finding that CBT did not reach statistical significance in the present study does not necessarily indicate that it is ineffective for alleviating anxiety symptoms in PPPD. Rather, it is more likely to reflect the context-dependent nature of its effects and the influence of multiple methodological factors. First, the number of high-quality randomized controlled trials evaluating CBT in PPPD remains limited [2], particularly those using standardized anxiety scales as primary outcomes, thereby restricting statistical power. Second, substantial heterogeneity exists across CBT protocols, including variations in intervention duration, treatment frequency, delivery format (individual versus group), and therapist expertise, which may reduce the stability of pooled effect estimates. Furthermore, it is important to recognize that network meta-analysis reflects relative treatment effects. Although CBT may demonstrate superiority over waitlist or sham controls in pairwise comparisons, its relative effect may be attenuated within a network that includes multiple interventions—particularly those with stronger effect sizes, such as neuromodulation—thereby appearing statistically non-significant [22]. From a mechanistic perspective, the effects of CBT in PPPD are primarily mediated through top-down cognitive–emotional regulatory pathways. Similar to the dual-pathway model proposed in anxiety disorders, patients with PPPD often exhibit heightened activity in limbic regions (e.g., the amygdala and insula), alongside insufficient regulatory control from the prefrontal cortex [42]. This imbalance may lead to exaggerated threat appraisal of bodily sensations and postural instability. CBT, through strategies such as cognitive restructuring and exposure, is thought to enhance regulatory interactions among the amygdala, insula, and prefrontal cortex, thereby reducing hypervigilance and catastrophic interpretations of symptoms [42]. Furthermore, a growing body of evidence suggests that CBT may modulate broader large-scale brain networks, including the default mode network [43] as well as regions involved in self-referential processing, such as the precuneus [44]. Regulation of these networks may help reduce excessive symptom-focused attention and abnormal internal monitoring, although such effects are likely to vary across individuals. Overall, CBT in PPPD appears to function predominantly as an indirect, cognitively mediated intervention, primarily targeting the interpretation of emotional and perceptual experiences, rather than directly addressing the core sensory integration deficits underlying the disorder. In contrast, the significant efficacy of neuromodulation in reducing anxiety symptoms has important clinical implications. This finding is consistent with a growing body of evidence supporting the use of neuromodulatory techniques, such as rTMS, in the treatment of anxiety disorders [45]. By targeting key brain regions involved in emotion regulation and fear processing, particularly the prefrontal cortex [46], neuromodulation may exert a direct top-down regulatory effect on hypervigilance and autonomic arousal, both of which are core features of anxiety and PPPD. This suggests that these conditions may share partially overlapping neurobiological substrates, and that neuromodulation, by directly modulating these pathological circuits, may offer a mechanistic advantage over CBT, which primarily operates through cognitive and behavioral pathways. Interestingly, although GVS-based interventions did not demonstrate statistically significant effects on anxiety outcomes, they were ranked relatively high in the SUCRA analysis. This apparent discrepancy may reflect the probabilistic nature of ranking metrics, which can be influenced by the relative distribution of effect estimates despite substantial uncertainty. The wide confidence intervals observed for GVS suggest that its estimated effects remain imprecise, and therefore its ranking should be interpreted with caution. From a mechanistic perspective, GVS primarily modulates peripheral vestibular input [47] and may influence vestibulo-autonomic pathways [48], which have been implicated in arousal and emotional processing. However, such bottom-up modulation may be insufficient to meaningfully engage higher-order cortical networks involved in anxiety regulation, potentially limiting its clinical effects in PPPD. Depression outcome The present study found no statistically significant effects of any included interventions on depressive symptoms. However, this finding should not be interpreted as evidence of treatment ineffectiveness, but rather as reflecting substantial uncertainty within the current evidence base. From a statistical perspective, the number of studies contributing to depression outcomes was relatively limited, and the corresponding effect estimates were characterized by wide confidence intervals crossing the line of no effect, indicating considerable imprecision and insufficient statistical power. In addition, variability in depression assessment instruments across studies (e.g., HADS-D, PHQ-9, HAMD) may have further increased heterogeneity, thereby reducing the ability to detect significant effects [22]. This finding aligns with the current understanding of PPPD, where psychiatric comorbidities—particularly anxiety and depression—are well recognized, yet intervention-specific evidence targeting depressive symptoms remains sparse and methodologically limited [49]. Existing literature has predominantly focused on symptom characterization rather than treatment efficacy, resulting in underdeveloped evidence base for depression-specific outcomes in PPPD. From a clinical and mechanistic perspective, depressive symptoms in PPPD are more likely to represent secondary or reactive manifestations arising from chronic dizziness, persistent discomfort, functional limitations, and social withdrawal [2, 38]. Consequently, improvements in depressive symptoms may lag behind those observed in core vestibular symptoms and anxiety, and may require longer treatment durations or more targeted interventions. The majority of randomized controlled trials included in this analysis had relatively short follow-up periods (typically 8–12 weeks), which may be insufficient to capture delayed changes in mood. Measurement-related factors may also have contributed to the observed null findings, as commonly used depression scales include somatic components (e.g., fatigue, impaired concentration) that overlap substantially with PPPD-related symptoms, potentially reducing measurement specificity and complicating interpretation. From a treatment perspective, although the non-invasive interventions evaluated in this study may confer benefits for balance or anxiety outcomes, their mechanisms of action may not adequately target the neurobiological or psychosocial processes that maintain depressive symptoms. In addition, the absence of pharmacological interventions within the network represents an important limitation. Pharmacotherapies such as selective serotonin reuptake inhibitors (SSRIs) and serotonin–norepinephrine reuptake inhibitors (SNRIs) [2], which are commonly used in clinical practice, may exert more direct effects on depressive symptoms, and their exclusion may have influenced the overall assessment of treatment effects. Taken together, these findings highlight a critical evidence gap in the management of depressive symptoms in PPPD. While reinforcing the substantial uncertainty surrounding current non-invasive interventions, the present analysis also underscores the need for future research with larger sample sizes, standardized outcome measures, and longer follow-up durations. In addition, the identification of potential advantages of neuromodulation in other domains, alongside the lack of expected efficacy for established interventions such as CBT, further emphasizes the importance of conducting more rigorous, well-designed trials to clarify the relative effectiveness of available treatments and to identify patient subgroups most likely to benefit. Strength This study has several notable strengths that enhance the robustness and interpretability of the findings. First, by employing an NMA framework, we were able to simultaneously compare multiple non-invasive interventions for PPPD, including both direct and indirect evidence, thereby providing a more comprehensive synthesis than conventional pairwise meta-analysis. Second, we included studies using historical diagnostic labels (CSD and PPV) alongside modern PPPD criteria, enhancing comprehensiveness while ensuring clinical consistency with contemporary diagnostic standards; the substantial clinical and pathophysiological overlap between these entities provides a valid rationale for their inclusion. Thirdly, we adopted a rigorous approach to intervention node classification, balancing network connectivity with mechanistic homogeneity. VRT and advanced VRT were distinguished as separate nodes due to differences in their implementation and neural targets—advanced approaches incorporating virtual reality or optokinetic stimulation may engage higher-order cortical networks involved in visual–vestibular integration. The neuromodulation node included both rTMS and nVNS, which may converge on shared neural circuits related to emotional regulation and multisensory integration [35, 50]. GVS was maintained as an independent node due to its distinct mechanism of directly modulating peripheral vestibular input [47]. Although this classification introduced within-node heterogeneity, it was necessary to ensure network connectivity while preserving clinically meaningful mechanistic distinctions. Also, this study adopted a domain-specific analytical approach, evaluating treatment effects separately across balance, anxiety, and depression outcomes. Given the multidimensional nature of PPPD, this approach allows for a more nuanced understanding of how different interventions may differentially impact distinct symptom domains, rather than assuming uniform treatment effects across heterogeneous outcomes. We conducted systematic assessments of heterogeneity and inconsistency, which strengthens the internal validity of the analysis and enhances confidence in the robustness of the network estimates. In addition, the use of SMD enabled the integration of diverse outcome measures across studies, improving comparability while retaining statistical rigor. Finally, this study explicitly addressed key methodological considerations specific to NMA, including diagnostic mapping and intervention node classification. By acknowledging and critically evaluating these factors, the present analysis provides a transparent and methodologically grounded interpretation of the available evidence. Limitation Several limitations of this study should be acknowledged. First, many included studies (81.8%) were classified as high risk of bias, particularly in domains of selective reporting (81.8% high risk) and blinding (72.7% some concern or high risk). High risk of bias in primary studies typically leads to overestimation of treatment effects in meta-analysis [22]. Additionally, incomplete reporting of study design features, insufficient detail regarding randomization procedures, and lack of clarity regarding outcome assessor blinding were prevalent across the evidence base. Second, the limited number of included studies and small sample sizes within certain intervention nodes—particularly the GVS and VRT+GVS nodes with only 18 participants each—resulted in a sparse network structure that constrains statistical power and precision. This sparsity affects interpretation in two important ways. First, wide confidence intervals, particularly for depression outcomes, reflect substantial uncertainty in effect estimates. Second, comparison-adjusted funnel plots, while showing no clear asymmetry, have limited power to detect small-study effects in sparse networks; therefore, the absence of observed publication bias should be interpreted cautiously, especially for interventions supported by single studies or small sample sizes. Third, the aggregation of heterogeneous control conditions (wait-list, usual care, and sham) into a single reference node, though necessary for network connectivity, represents a critical methodological limitation. Sham interventions—commonly used in neuromodulation trials—may induce stronger placebo responses than wait-list or usual care conditions, potentially biasing effect estimates. This concern is particularly relevant for subjective outcomes such as dizziness, anxiety, and depression. Future studies should employ more homogeneous control conditions or conduct sensitivity analyses stratified by control type. Fourth, the sparse network structure—a consequence of limited available evidence—constrained the analysis in multiple ways. The absence of closed loops in the depression network prevented formal inconsistency assessment, though global testing showed no significant violations. However, the statistical power to detect inconsistency was correspondingly limited by the network structure, warranting cautious interpretation of depression findings. Additionally, the sparse evidence base meant that pharmacological interventions—commonly used in clinical practice for PPPD—were not adequately represented, limiting the comprehensiveness of the analysis. Furthermore, most included studies had relatively short follow-up durations, which may be particularly important given that PPPD is a chronic and fluctuating disorder in which patients may experience symptom relapse even after an apparent initial response to treatment. Consequently, these limited follow-up periods may restrict the ability to assess the long-term efficacy and durability of treatment effects, potentially underestimate interventions requiring longer periods to achieve clinically meaningful benefits, and overestimate the long-term value of interventions demonstrating favorable early responses. Taken together, these limitations highlight the need for cautious interpretation of the findings and underscore the importance of future high-quality, large-scale randomized controlled trials. Conclusion In summary, the present network meta-analysis suggests that neuromodulation may offer relative advantages for balance and anxiety outcomes in PPPD, whereas evidence for depression remains inconclusive. However, given the limitations of the current evidence base, these findings should be interpreted cautiously and viewed as hypothesis generating rather than practice changing. Supplementary Information Below is the link to the electronic supplementary material. Supplementary file1 (DOCX 36 KB) Acknowledgements We sincerely thank the researchers from the Department of Otolaryngology, Yueyang Hospital of Integrated Traditional Chinese and Western Medicine. Author contributions Conceptualization: JW Shen, XY Zhao. Methodology: JW Shen, XY Zhao. Formal analysis and investigation: JW Shen, XY Zhao, Y Huo, SL Chen. Resources: JW Shen, XY Zhao. Data curation: JW Shen, XY Zhao, Y Huo, SL Chen. Writing—original draft preparation: JW Shen. Writing—review and editing: all authors. Supervision: JN Zhang, W Zheng. Administrative support: JN Zhang, W Zheng. Funding acquisition: JN Zhang. All authors have read and agreed to the published version of the manuscript. Funding This work was supported by the National Natural Science Foundation of China (grant No. 82474263) and the Key Research and Development Program sponsored by the Ministry of Science and Technology (2023YFC2508402). Data availability Most of the data supporting the findings of this study are available in the Supplementary Materials. Additional detailed data, including all relevant raw datasets, will be made freely available to qualified researchers for non-commercial use, provided that participant confidentiality is not compromised. De-identified data can be requested directly from the authors. Declarations Conflicts of interest The authors have no competing interests to declare that are relevant to the content of this article. Ethical approval and consent to participate Inapplicable. Footnotes Xinyu Zhao and Jiawei Shen have contributed equally. Contributor Information Wei Zheng, Email: 17621760727@163.com. Jianning Zhang, Email: eternityz@sina.com. References Associated Data Supplementary Materials Supplementary file1 (DOCX 36 KB) Data Availability Statement Most of the data supporting the findings of this study are available in the Supplementary Materials. Additional detailed data, including all relevant raw datasets, will be made freely available to qualified researchers for non-commercial use, provided that participant confidentiality is not compromised. De-identified data can be requested directly from the authors.