We searched PubMed, MEDLINE, Embase, Cochrane Library, Web of Science, CINAHL, CNKI, Wanfang database, VIP and CBM, and the search period was from the establishment of the database to June 2025. The search strategy used MeSH terms and keywords related to “virtual reality” and “peripheral vestibular dysfunction,” and the references of the included literature were manually searched. The following presents an example of our search process within the PubMed database (Table 1), and the complete search strategy is detailed in Supplementary materials S1.

Inclusion criteria:

Exclusion criteria:

Literature screening was performed in EndNote. Two reviewers (XYL and SY) independently screened each paper according to the inclusion and exclusion criteria, first excluding duplicate studies, then independently screening the full text based on title and abstract, and finally cross-checking. In case of disagreement, a third reviewer (YRW) was consulted.

Two reviewers utilized an Excel sheet to extracted data after reading the full text. The data extracted included the following variables: first author, year of publication, sample size, age, intervention and control group, intervention period, single intervention duration and intervention frequency, level of immersion, type of VR (equipment, VR environment and software), and outcome indicators. Use a separate form (Supplementary material S2) to record the raw outcome data reported for each study. The quality of RCTs was assessed using the Cochrane Risk of Bias tool (30), which includes method of randomization, allocation concealment, blinding of study subjects and investigators, blinding of assessors, completeness of outcome indicators, selective reporting status, and other biases. Each item was categorized as “low risk,” “unclear,” or “high risk.” In case of disagreement regarding the results, this was done by mutual agreement or by consulting a third reviewer.

Statistical analysis was performed using RevMan 5.4, and the included outcome indicators were continuous variables, so mean difference (MD) or standardized mean difference (SMD) was used as the effect indicator, and 95% CI was the effect analysis statistic. Heterogeneity was analyzed using the χ² test and the I² statistic, if there was no heterogeneity among the study results (p > 0.1, I² < 50%), the fixed-effects model was used for the analysis; if there was heterogeneity (p ≤ 0.1, I² ≥ 50%), the random-effects model was used, and sensitivity analysis was performed. The stability of the results was tested by excluding literature on a case-by-case basis to identify sources of heterogeneity. We performed subgroup analysis according to different immersion levels. And further subgroup analyses of the optimal immersion methods were performed to further identify the sources of heterogeneity. Egger’s test was performed using Stata18 software. p < 0.05 indicates a statistically significant difference. Some studies did not provide efficacy data regarding the improvement of dizziness symptoms; therefore, we utilized the data transformation formula recommended by Chi (31) and Luo (32) to compute the mean and standard deviation, which were obtained by XYL under the supervision of another reviewer (YRW).

An initial search identified a total of 2,664 academic articles, including 1,963 in English and 701 in Chinese. Utilizing EndNote X9.1, 950 redundant entries were eliminated, resulting in a total of 1,714 articles. Following a review of the titles and abstracts, 1,677 articles were excluded for several reasons: incorrect target populations, lack of intervention of interest, absence of appropriate controls, wrong outcomes, etc. After a thorough review of the literature, we excluded 25 studies for the same reasons (see Figure 1 for further details). Ultimately, a total of 12 RCTs (33–44) were finally included.

Table 2 outlines the detailed characteristics of each study included in our analysis. Among the twelve studies examined, the majority were conducted in Turkey (40–42) (three studies), followed by Brazil (33, 39), Italy (34, 36), and China (38, 43) (two studies), with the UK (35), Poland (37), and Jordan (44) (one study). The publication dates spanned from 2011 to 2025; eight of them after 2020 (37–44). A total of 600 participants were included in the meta-analysis, comprising 304 in the intervention group and 296 in the control group. The average age of participants ranged from 39.6 ± 8.75 to 70 ± 6 years. Regarding PVD types, four studies focused on unilateral vestibular hypofunction (34, 36, 37, 41), three examined benign paroxysmal positional vertigo (BPPV) (38, 42, 43)[with one specifically addressing residual symptoms of BPPV (43)], and two investigated elderly patients with dizziness (39, 40). Additionally, one study each explored persistent postural-perceptual dizziness (PPPD) (44), adults with dizziness (35), and Ménière’s disease (33).

Ten studies (33, 34, 36–40, 42–44) compared head-mounted devices to conventional rehabilitation and two studies (35, 41) compared non-immersive Nintendo Wii to conventional VRT.

Regarding the number of VR interventions in the intervention group, we observed an average of 13.3 sessions (13.3 ± 5.25); Ozdil et al. (42) applied the most sessions, a total of 24. The studies by Yan et al. (43) and Micarelli et al. (34) were used the least number of times with a total of 8 times. The range of intervention durations varied from 4 to 16 weeks. The average duration range of each intervention was 26 to 35 min. The average number of interventions per week was 6 (13.3 ± 4.8). On the other hand, in the control group, the mean number of regular VRT sessions was 11.7 (11.7 ± 6.4). The mean range of time per exercise was 34.7 min.

Post-intervention vertigo symptoms were assessed using various scales across all participants. These instruments have demonstrated reliability as measures of vertigo symptoms. Ten studies (33–36, 38–41, 43, 44) employed the Dizziness Handicap Inventory (DHI), of which Hasimova et al. (41) and Kanyılmaz et al. (40) studies also used Visual Analog Scale (VAS), Vertigo Symptom Scale (VSS) respectively. Ozdil et al. (42) and Stankiewicz et al. (37) study used the VAS scale to measure vertigo. However, the utilization of various assessment tools across different studies has introduced complexity to our analysis. To address this challenge, we adhered to the Cochrane Handbook and standardized the data by calculating statistical measures, such as SMD (45), thereby converting the scores from diverse studies into comparable effect sizes. This approach alleviated the influence of measurement tool discrepancies, facilitating comparisons and analyses of data from various studies within a cohesive framework. Furthermore, 12 studies (33–44) reported enhanced vertigo outcomes for continuous variables, with one study (39) expressing outcomes using 95% CI and one (40) using interquartile range. Subsequently, we applied the data transformation formula recommended by Chi and Luo to calculate the mean and standard deviation.

The results of the risk of bias assessment for the RCTs are shown in Figure 2. 12 studies (33–44) discussed random sequence generation methods, including random number table methods and computerized random sequence methods. The studies of Viziano et al. (36), Lima Rebêlo et al. (39), and Ozdil et al. (42) reported detailed methods of assigning concealment, whereas the other studies were considered as unclear risks. It is worth noting that only the studies by Viziano et al. (36) and Lima Rebêlo et al. (39) were described as double-blind, and the studies by Kanyılmaz et al. (40) and AL-Omari (44) et al. as single-blind. The studies by Garcia et al. (33), Ozdil et al. (42), and Yan et al. (43) explicitly stated that they did not blind patients, and one of them, Ozdil et al. (42), did not blind the evaluators either, and therefore assessed the risk of bias as “high risk,” while the remaining studies were assessed as “unclear risk of implementation bias.” Ten studies (33, 34, 36–39, 41–44) had complete outcome data, and the studies by Philips et al. (35) and Kanyılmaz et al. (40) were rated as “low risk” with reference to the number of reasons and excluded participants.

In total, 12 studies (33–44) examined the effectiveness of VR (immersive or non-immersive) in the rehabilitation of dizziness for patients with PVD. Two studies (35, 41) compared non-immersive VR with conventional VRT, whereas 10 studies (33, 34, 36–40, 42–44) compared immersive VR with conventional VRT. The study by Hasimova et al. (41) used the DHI, VAS scale, and since the study explicitly stated that it was using VAS scale to evaluate vertigo severity, the VAS scale was chosen for variable inclusion. Although the study by Kanyılmaz et al. (40) also used DHI and VSS scales, the VSS scale was mainly used to evaluate the severity of vertigo, so the VSS scale was chosen for inclusion as a variable. The studies yielded continuous variables, assessed by different scales. Thus, SMD was employed as the effect size indicator.

Subsequently, we categorized the studies into two groups based on immersion levels, specifically the types of devices used—immersive and non-immersive VR—and conducted a subgroup analysis focusing on the primary variable of this review: improvement in dizziness.

In subgroup analyses examining immersion level, both non-immersive VR and immersive VR significantly improved vertigo symptoms in patients with PVD. In the specific study focusing on immersive VR, we concluded that vertigo symptoms significantly improved (SMD = −2.08, 95% CI = −3.13 to −1.04, p < 0.001; Figure 3).

Given the observed effectiveness of immersive VR interventions in the subgroup analysis, along with the significant heterogeneity (I² = 95%, p < 0.001), this may be attributed to small sample sizes or inherent differences in intervention protocols and study designs among the included studies. We further investigated three intervention parameters as potential sources of heterogeneity: intervention period, single intervention duration, and intervention frequency. These analyses aim to systematically evaluate how VR intervention plans affect the effectiveness of immersive VR.

Regarding the intervention period, a duration of ≤7 weeks significantly improved vertigo symptoms in patients with PVD compared to a duration of >7 weeks (SMD = −2.73; 95% CI = −4.17 to −1.28, p < 0.001; Figure 4). Conversely, the longer intervention period >7 weeks showed no statistically significant improvement in vertigo symptoms (SMD = −1.02; 95% CI = −2.56 to 0.53, p = 0.20). In terms of single intervention duration, the synthesized effect size of the subgroup with <30 min/times (SMD = −2.80, 95% CI = −4.89 to −0.70, p = 0.009; Figure 5) exhibited a slightly larger effect than the subgroup with ≥30 min/times (SMD = −1.03, 95% CI = −1.72 to −0.35, p = 0.003). In terms of intervention frequency, intervention frequency ≥5 times/week was effective in improving vertigo symptoms in patients with PVD compared with intervention frequency <5 times/week (SMD = −2.64; 95% CI = −4.91 to −0.38, p = 0.02; Figure 6), and there was no statistical significance for intervention frequency <5 times/week (SMD = −1.98, 95% CI = −1.98 to 0.11, p = 0.08).

According to Cochrane’s recommendations (30), the use of funnel plots to test for publication bias is more reliable when the number of studies included in the Meta-analysis reaches at least 10. Therefore, the present study maps funnels for immersion levels, as well as for studies involving Intervention period in the immersive VR group. The results showed that the funnel plots presented a roughly left–right symmetrical distribution, as shown in Figures 7, 8. The Egger’s test showed that there was no publication bias against the immersion level (p > 0.05). For the studies involving Intervention period in the immersion VR group, there was publication bias (p < 0.05). So the cut-and-patch method was adopted to deal with bias, no literature was added, and the data did not change after the cut-and-patch, thus it can be determined that the impact of bias on the results is small. Furthermore, we performed sensitivity analyses on the outcome indicators for the two groups independently. The results indicated that the effect sizes across the studies exhibited minimal change, suggesting that the findings of the meta-analysis are relatively stable.

The main objective of this study was to investigate the effectiveness of different immersion levels of VR in improving vertigo symptoms in patients with PVD. The systematic review and meta-analysis herein comprehensively analyzed data from 12 RCTs to draw research conclusions. And further review identifies a diverse, generalized range of ideal options for conducting vestibular rehabilitation training. Overall, the use of both immersive and non-immersive VR appeared to help with vestibular rehabilitation, but immersive VR was more effective. Subgroup analyses also found that immersive VR programs with an intervention period of ≤7 weeks, a single intervention duration of <30 min, and an intervention frequency of ≥5 times/week were more effective in improving vertigo.

Our meta-analysis showed that immersive VR showed statistically significant superiority over non-immersive VR and traditional VRT, although the results of this were limited by significant statistical heterogeneity. Patients with PVD utilized VR devices for balance and gait training, enhancing the stability of vestibulo-ocular reflex (46), which in turn stimulated central nervous system compensatory mechanisms and helped in alleviating or eliminating vertigo symptoms (47, 48). This constitutes one of the potential explanations for the systematic review finding that immersive VR is superior to both non-immersive and conventional VRT in mitigating vertigo symptoms, particularly through multisensory stimulation that enhances proprioceptive function (49). Moreover, immersive VR may reduce anxiety and spatial fear during the rehabilitation process, which increasing patients’ confidence in treatment and their motivation to recover (50, 51). Hazzaa’s review corroborates our findings, indicating that immersive VR is a promising approach for effectively improving vertigo symptoms (20). Conversely, Lima Rebêlo’s study (39) reported no significant differences between immersive VR and conventional VRT. This may be due to the fact that the patients included in that study were older, less receptive to VR equipment, and had difficulty concentrating during the training session, ultimately hindering their ability to complete the sessions. Furthermore, the subgroup conducted in this investigation were based on a small number of studies. Importantly, the results of this study are derived from fewer than 1,000 patients and are based on post hoc analyses; thus, a large volume of methodologically rigorous studies is necessary to revalidate these results before endorsing immersive VR for widespread clinical application.

The design and specific technical parameters of immersive VR training scenarios are critical for their clinical applications and subsequent research. Kanyılmaz employed two types of immersive 3D VR scenarios: (1) a 15-min outdoor plaza video, captured with a Samsung Gear 360 camera, featuring pedestrians, vehicle traffic, and environmental sounds, in which patients engaged in seated and standing training within a clinical environment; (2) a 1.5-min video of a supermarket shelf filled with colorful products, where patients walked on a treadmill while wearing Samsung Gear VR goggles connected to a Galaxy S7 phone (40). This setup allowed patients to alter their perspective to view products of varying shapes and colors while hearing authentic environmental sounds, thus helping their brains to mitigate dizziness induced by sensory information misalignment through the integration of inputs from multiple senses. Micarelli utilized VR equipment embedded in a ‘Revelation’ 3D VR headset (dimensions: 140 × 105 × 64 mm; weight: 165 g), integrated with a Windows Phone (5.2-inch display, Lumia 930, Windows 10 Mobile, Microsoft Corporation) (34). Patients were required to adjust the eye-tracking device’s settings according to their pupillary distance and focal length prior to each use. Similarly, Lee employed a VR device comprising an Android smartphone (5.8-inch display, Samsung Galaxy S9) placed inside a ‘BOBO VR Z4’ headset (dimensions: 194 × 117 × 127 mm; weight: 410 g), where patients were instructed to focus on a blue ball set against backgrounds of varying difficulty levels while maintaining a consistent speed during 15 head movements (52). However, the studies included in the review did not report essential parameters such as resolution and refresh rate. Subsequently, our review of relevant research revealed that certain high-end VR devices have achieved a resolution of 4 K per eye, with refresh rates reaching 120 Hz or higher and field of view angles extending up to 200° (53, 54). It is worth noting that the weight of VR devices also serves as a critical parameter (55). Given that patients with peripheral vestibular vertigo are predominantly middle-aged and elderly, the prolonged use of heavier VR devices may compress the cervical spine and diminish comfort (56). However, existing studies have reported limited relevant technical parameter indicators. It is expected that high-quality RCTs will be conducted in the future to establish standardized technical parameter indicators. Additionally, by integrating game levels and immersive life scenarios, VR environments enhance the enjoyment and motivation for vestibular training. It is recommended that rehabilitation programs for middle-aged and elderly individuals incorporate personalized setting options within the system, allowing patients to adjust device parameters (e.g., movement speed, visual intensity) according to their individual physical conditions.

Subgroup analyses found that intervention period of ≤7 weeks and single interventions lasting <30 min significantly improved vertigo symptoms in patients with PVD, which is consistent with the vestibular training program recommended by the guidelines (11). This phenomenon may be attributed to the fact that a shorter intervention period not only reduces vestibular adaptation fatigue but also alleviates vertigo symptoms, thereby enhancing patients’ confidence in treatment adherence and the overall effectiveness of vestibular rehabilitation (37, 38). Furthermore, we found that a frequency of ≥5 times/week was more effective on improving vertigo symptoms (37, 40). It may be due to the fact that frequent interventions in a short period of time help to accelerate and strengthen the adaptive and compensatory mechanisms of the vestibular system and improve patient compliance, ultimately improving vertigo symptoms. It is important to note that fully immersive VR also presents certain challenges. During the treatment process, patients with PVD are required to wear head-mounted displays, which can lead to an overload of visual information. Consequently, some participants may experience motion sickness, resulting in dizziness, nausea, and vomiting, thereby diminishing the effectiveness of the intervention (57). This underscores the need for healthcare professionals to assess vestibular function continuously during rehabilitation, allow for flexible adjustments to the training program, and gradually increase training difficulty to achieve optimal intervention outcomes. Furthermore, the variability in the intervention effects among PVD patients, attributable to individual differences such as age and vertigo symptoms, suggests that comprehensive vestibular function assessments should be performed before designing a training program for these patients. This implies that before developing training programs for patients with PVD, a thorough assessment of vestibular function should be conducted, and individualized rehabilitation programs should be formulated, taking into account age, medical history, and cognitive level.

The study presents several limitations. First, the sample size included in the study is relatively small, and the nature of VR devices means that not all studies employ blinded methods. This significantly contributes to the Hawthorne effect and potential biases introduced by researchers, which may lead to an overestimation of the efficacy of VR in alleviating dizziness. Second, the study’s results demonstrate substantial heterogeneity. Although sensitivity and subgroup analyses suggest that the meta-analysis results are relatively stable, the precise sources of heterogeneity remain undetermined. This may stem from the lack of uniformity in the tools used for measuring dizziness. Different studies utilize various tools, resulting in discrepancies in dimensions, assessments, and sensitivity. Such variations introduce data collection bias and exacerbate the heterogeneity of the results. Furthermore, the complexity of patients with PVD is another crucial factor. The variability in their causes, severity, and disease course may lead to clinical heterogeneity between studies, thereby challenging the stability of the results. Third, the outcome measures included in this research are subjective, which may introduce additional heterogeneity. It is recommended that future studies incorporate objective measures for assessing dizziness symptoms, such as the Mini-Balance Evaluation Systems Test (Mini-BES Test) and the Modified Clinical Test of Sensory Interaction for Balance (mCTSIB). Finally, most of the studies are single-center designs, and the technical parameters and performance of the VR devices differ, which may impose limitations that could affect treatment outcomes.