This systematic review and meta - analysis followed the PRISMA guidelines (19). The protocol was registered with PROSPERO (CRD420201008459).

We searched six databases (PubMed, Embase, Cochrane Library, Web of Science, EBSCOhost, and CNKI) for RCTs on the effects of virtual reality training (VRT) on balance and functional gait in Parkinson’s disease (PD) patients, from their inception to March 15th, 2025. Three sets of keywords were used, combined with “AND” in the databases.

Chinese translations were used for Chinese databases. We manually checked references of included studies and related systematic reviews to identify more relevant studies. The full search strategy for PubMed is as follows:

(((“Parkinson Disease”[Title/Abstract] OR “Parkinson”[All Fields]) AND “parkinson*”[Title/Abstract]) OR “PD”[Title/Abstract]) AND (“virtual reality exposure therapy”[Title/Abstract] OR “VR”[Title/Abstract] OR “virtual reality”[Title/Abstract] OR “virtual”[Title/Abstract] OR “illusion”[Title/Abstract] OR “immersive”[Title/Abstract] OR “reality system”[Title/Abstract] OR “game”[Title/Abstract] OR “simulation”[Title/Abstract] OR “exergame”[Title/Abstract]) AND (“equilibrium”[Title/Abstract] OR “balance”[Title/Abstract] OR “functional reach”[Title/Abstract] OR “postur”[Title/Abstract] OR “dynamic postural control”[Title/Abstract] OR “gait”[Title/Abstract] OR “locomotion”[Title/Abstract] OR “walking”[Title/Abstract] OR “mobility”[Title/Abstract] OR “treadmill gait”[Title/Abstract] OR “ambulation”[Title/Abstract] OR “Stride Length”[Title/Abstract] OR “Stride Cadence”[Title/Abstract] OR “Gait Speed”[Title/Abstract] OR “Gait velocity”[Title/Abstract] OR “walking speed”[Title/Abstract])

The PICOS framework (Population, Intervention, Comparator, Outcome, Study design) was used to assess study eligibility (20). Studies are included in the review if they meet all of the following criteria:

Studies were excluded if they:

After reviewing the full texts, descriptive data were extracted, categorized into three types: literature characteristics (first author, publication year, country, language), participant characteristics (diagnostic criteria, sample size, gender ratio, age), and intervention plan details. To determine the dose - response relationship of VRT on PD patients’ balance, the training protocol was coded to include training group (experimental and control groups), single - session duration, frequency, total sessions, weekly training time, total training duration, and VRT platform (24, 25). Data on balance and gait (e.g., BBS, 6MWT scores) pre - and post - intervention were extracted. If multiple control groups existed, only data from the active intervention group were used. For missing data, the corresponding authors were contacted via email three times within three weeks. Two reviewers independently extracted data, with a third reviewer checking and adjudicating. Consensus was reached through discussion for any discrepancies.

In this meta - analysis, the intervention effect was assessed using the change in mean (Mean Difference, MD) and standard deviation (SD). If the original study did not directly report the SD, it was estimated from the standard error, 95% confidence interval (CI), p - value, or t - statistic (26). For calculating the SD of the difference between pre - and post - intervention, a correlation coefficient of 0.5 was assumed, reflecting moderate measurement consistency and balancing potential variability to ensure robust and reliable results (26).

The Cochrane Risk - of - Bias Tool (version 2.0) was used to assess bias risk, covering random sequence generation, allocation concealment, blinding, missing outcome data, and selective reporting (27). Overall bias risk was categorized as follows:

Two independent reviewers assessed bias risk, resolving discrepancies through discussion.

Evidence quality was evaluated using the GRADE method via the GRADEpro GDT online tool (www.gradepro.org). Quality was assessed across five dimensions: risk of bias, inconsistency, indirectness, imprecision, and publication bias. Evidence quality was graded as “high,” “moderate,” “low,” or “very low” based on the credibility of effect estimates (28). All assessments were done by independent reviewers, with discrepancies resolved through discussion.

Meta-analysis was performed when ≥2 relatively homogeneous studies existed for an outcome (28). Balance ability (BBS) and functional gait (6MWT) intervention effects were assessed using weighted mean differences (WMD) and 95% confidence intervals (95% CI), calculated based on changes in the intervention group relative to the control group. A random-effects model was used to account for potential heterogeneity across studies, incorporating differences in populations, interventions, and measurement methods (29). Heterogeneity was assessed using the I² statistic, interpreted as follows: <25% (low), 25–75% (moderate), >75% (high) (30). Publication bias was assessed using Egger’s test due to the subjectivity of funnel plots (31). If publication bias was suspected, the trim - and - fill method was used to adjust the overall effect and estimate the impact of potentially missing studies (32). For highly heterogeneous outcomes, sensitivity analysis was conducted by sequentially excluding studies to evaluate result robustness and identify heterogeneity sources. All analyses were performed using Stata (version 17.0; StataCorp, College Station, TX, USA), with forest plots used to visually display pooled effect sizes and CIs. Statistical significance was set at p < 0.05.

Effect size reflected the impact of VRT on PD patients’ balance and gait. For the BBS in Parkinson’s disease, a score change ≥3 points is generally considered to represent the minimal clinically important difference (MCID), indicating a significant improvement in balance (33), For the 6MWT, an improvement of 14.0 to 30.5 meters is considered clinically meaningful (34). We further analyzed the effects of VRT using meta-regression models based on intervention duration, frequency, weekly training time, total number of sessions, and intervention time. However, this meta-regression part was not analyzed in this study due to the lack of quantification of treatment intensity. For studies not providing the above data, the authors were contacted via email to obtain the data and improve the study quality.

This study systematically searched six databases: Web of Science (357), PubMed (475), Cochrane (361), Embase (718), EBSCOhost (1,162), and CNKI (39), yielding 3,116 initial records. After deduplication in EndNote X9 (Bld 12,062), 1,780 records were retained. Independent dual - screeners excluded 1,462 records via title/abstract review for not meeting predefined inclusion criteria. Following full - text assessment, 172 more were excluded, leaving 32 studies (with 547 participants) for meta - analysis (see Figure 1 for the screening process). These studies, published between 2004 and 2024, covered diverse interventions like VRT and conventional rehabilitation, exhibiting high methodological heterogeneity. Among them, 29 (90.6%) used the Berg Balance Scale (BBS) as the primary outcome measure, and 8 (25.0%) reported data on the 6 - Minute Walk Test (6MWT). Table 1 summarizes the study characteristics and quality assessment results.

The Cochrane Risk - of - Bias Tool was used to assess methodological quality (Figure 2). Results showed 62.5% (20/32) of studies were low - risk, 31.3% (10/32) unclear - risk, and 6.3% (2/32) high - risk. In key risk domains, 34.4% (11/32) clearly reported random sequence generation, but only 59.4% (19/32) mentioned allocation concealment. Regarding intervention adherence, only 3.1% (1/32) adopted a blinded design, with 84.4% (27/32) having unclear methodology. Notably, 28.1% (9/32) had selective reporting bias, while all met low - risk standards in outcome measurement. No other potential bias sources were identified.

Evidence quality was assessed using the GRADE system (Table 2), with all outcome indicators rated as moderate - quality. This was mainly due to methodological flaws in some studies (37.6% high/unclear - risk) and significant heterogeneity (I²>75%), suggesting cautious interpretation of results.

Funnel plots were first used to visually assess publication bias in the included studies. The BBS funnel plot (Figure 5A) showed a relatively symmetrical distribution, while the 6MWT funnel plot (Figure 5B) exhibited some asymmetry, making it difficult to accurately judge the presence of publication bias based solely on funnel plots. Therefore, Egger’s test was further used for quantitative analysis. The results showed no significant publication bias for both BBS (p = 0.62 > 0.05) and 6MWT (p = 0.66 > 0.05), indicating a low risk of publication bias in the included studies.

None of the included RCTs reported adverse events related to the interventions. Thus, this study could not extract information on adverse events from the available literature.

This study’s results show that virtual reality training (VRT) can significantly enhance balance function in Parkinson’s disease (PD) patients. PD patients often experience insufficient interaction between the vestibular and proprioceptive systems, altering human biomechanics and affecting balance ability (35). Most included RCTs support this conclusion, mainly due to the following reasons: VRT provides visual and auditory feedback for PD patients, compensating for motor information loss caused by reduced dopamine (36). Additionally, real - time feedback from virtual reality platforms facilitates motor processes (37), and their visual and auditory inputs integrate effectively with vestibular feedback, improving receptor and proprioceptor function, thus enhancing balance ability (38). Furthermore, studies indicate that VRT stimulates PD - related cognitive functions, such as attention and executive function integration, and activates brain reward mechanisms (39). Compared to traditional physical rehabilitation, VRT is more engaging, motivating patients to participate more actively and improving training quality. Moreover, VRT’s reaction to attention, sensory integration, and stimuli in virtual environments offers advantages over traditional rehabilitation (40).

However, the results also show that VRT’s improvement on the 6 - Minute Walk Test (6MWT) distance for PD patients is not statistically significant (p = 0.13), despite reaching the minimal clinically important difference (MCID, WMD = 17.64 m > 14 m), possibly due to high heterogeneity (I² = 99.5). The included studies vary in terms of VRT platforms and methods, so clinical interpretation should be cautious. While motor symptoms are classic PD features, cognitive impairments also pose potential risks (41), Some research indicates that targeted therapy can improve PD patients’ cognitive functions, including overall cognition, attention, and verbal memory (42). Cognitive training for PD shows specific benefits in target treatment areas, including overall cognition, attention, and verbal memory (43). Lau also speculated that improvements might stem from neuroplasticity changes due to training intensity and movement patterns, as well as increased sensory cues from treadmill use combined with VRT (44), because the intervention method used in the study was treadmill combined with VRT.

When analyzing the BBS (Berg Balance Scale), this study conducted subgroup analyses on intervention methods, platforms, and population regions. The results show that intervention method, platform, and region all have statistically significant differences, but the WMD (weighted mean difference) does not consistently meet the MCID (minimal clinically important difference).

Specifically, our analysis indicates that the Minimal Clinically Important Difference (MCID) is achieved when Virtual Reality Training (VRT) is integrated with Physical Therapy (PT). In contrast, the MCID is not consistently met when VRT is employed as a standalone intervention. This observation is consistent with the findings of Wen - Chieh Yang et al. (45), who reported comparable improvements in Berg Balance Scale (BBS) scores between VRT - only and traditional rehabilitation groups. We speculate that this may be due to the fact that both approaches share similar underlying design principles, which in turn lead to similar training effects (46). The combination of VRT with PT may create a more enriched environment for motor learning and adaptation. PT can provide hands - on guidance, manual facilitation, and personalized progression that complements the external feedback and engagement offered by VRT. This integration may enhance the overall effectiveness in improving balance function. Yang also suggested that measurement tool insensitivity might limit the detection of subtle differences between VRT and traditional balance training. Notably, Yang used a self - designed VR balance training system, not a commercial product. Kashif (47) study also focused on this area, showing that VR combined with PT had significantly better training effects than other groups. Kashif believes that combining virtual reality technology with conventional rehabilitation helps patients learn and master new skills based on repetitive movements. Patients recall and consolidate actions from traditional physical therapy using external feedback from VR, enhancing motor memory (48). This method also targets attention span and executive function, activates alternative neural pathways, and supports neuroplasticity (49).

Regarding intervention platforms, both commercial and rehabilitation - specific virtual reality platforms reached the MCID. This contradicts Wu (35) review, which found no significant effect of commercial gaming VR platforms on PD patients’ balance ability, suggesting similar impacts between VR training and traditional rehabilitationThis discrepancy may arise from the inclusion of recent studies or the rapid development of commercial VR platforms (50). Pullia M. noted that commercial VR devices like C - Mill offer better safety, boosting patient confidence and training effectiveness (51).

In terms of regional differences, Asian patients reached the MCID, while non - Asian patients did not. Analysis revealed that Chinese PD patients predominantly used game - combined VR training (52–54),which is more engaging. In contrast, non - Asian regions mainly used conventional motor therapy combined with VR, which is less interesting, potentially reducing patient engagement and training effectiveness, especially during solo sessions (43, 55).

This study shows that virtual reality training (VRT) significantly improves balance and functional gait in Parkinson’s disease (PD) patients, with effects not dependent on specific training methods. Meta - regression and subgroup analyses explored how different dose parameters affect outcomes.

For balance ability (BBS) optimization, the recommended dose is ≤20 min per session, 4–7 times weekly for 4–7 weeks (total >40 sessions, weekly 201–300 min). This aligns with Wu’s (35) review conclusions. By incorporating more literature (15 additional studies), increasing sample size, and integrating the latest data, this study strengthens the credibility of the dose recommendations.

For functional gait improvement (6MWT), the recommended dose is ≤20 min per session, 4–7 times weekly for 4–7 weeks (total 21–40 sessions, weekly 100–300 min).

Notably, short - term training is more effective than long - term training. This may be because PD patients with lower baseline BBS scores have greater room for balance rehabilitation improvement than those with higher scores (56). Also, the optimal dose for 6MWT is similar to that for BBS, so the best dose - response for VRT on PD patients’ balance and gait can be summarized as “high - frequency short - term intervention.” This mode is more suitable for PD patients’ physical tolerance and rehabilitation needs, and helps improve training compliance and effectiveness.