This systematic review incorporating meta-analysis is reported according to the expanded 2020 PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analysis) checklist [33]. This review and protocol were registered under review number CRD420250650110 through PROSPERO International prospective register of systematic reviews on February 24, 2025.

IVR is a form of computer-human interface that allows the user to become immersed in a 3D, computer-generated environment while wearing an HMD [8].

Aerobic exercise is a structured and planned physical activity in which the body’s large muscles move in a rhythmic manner for a sustained period of time [34].

Anaerobic exercise is a structured and planned physical activity consisting of brief high-intensity bursts of exercise (such as strength training or fast boxing), where oxygen demand surpasses oxygen supply [34].

The search strategy used in this systematic review is reported according to the PRISMA-S (PRISMA–Search) statement for reporting literature searches in systematic reviews [35]. Systematic searches were conducted by one reviewer (RCCB) across 4 databases from inception until January 14, 2025, and re-run on January 6, 2026. The databases searched included PubMed, Embase (Elsevier), Web of Science (Clarivate), and CINAHL Complete (EBSCOhost). Study registries were not searched for article identification purposes. Key search terms included “immersive virtual reality,” “immersive virtual reality exercise,” and “physical activity.” Search strategies for each database were cocreated by all members of the review team and are available in Multimedia Appendix 1. Search strategies from previous systematic reviews were not reused. Recursive and forward searching of reference lists for all included studies and other systematic reviews (those published in 2024 through Google Scholar) was completed. Covidence software (Veritas Health Innovation) was used to deduplicate records from the databases and complete screening procedures. Google Translate was used to translate reports from other languages to English. Only randomized controlled trials (RCTs) were included in this review. The non-RCT automation tool within Covidence software was applied to assist with the identification of ineligible study types prior to study screening. No additional information sources or search methods were used. Two reviewers (RCCB and MHR) contributed to study screening and selection. All disagreements in screening were resolved through consensus discussion.

Data describing participant and trial characteristics, eligibility criteria, interventions, trial funding sources, clinical effectiveness outcomes, and feasibility outcomes were first extracted by 2 independent reviewers (RCCB and MHR) for 30% of trials (n=8) using Covidence and Microsoft Excel (Microsoft Corp) and tabulated into descriptive tables. After it was ascertained that there was 100% agreement between reviewers, one independent reviewer extracted data for the remainder of the trials (n=18; RCCB). Where further information was needed (n=3), authors were contacted via email by one reviewer (RCCB). Clinical effectiveness data were obtained for all reported timeframes in each individual trial. For studies presenting multiple assessments for the same clinical effectiveness domain, one was chosen for entry into the meta-analysis. All data in this review that were not reported in the original manuscripts were obtained via author correspondence.

The revised Cochrane Risk of Bias 2 tool (RoB2) [36] was used for quality appraisal for included trials. Study bias was assessed as “high,” “some concerns,” or “low” across 5 domains (randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, selection of the reported result) and facilitated by 2 independent reviewers (RCCB and MHR) for 28% of outcomes (n=15). Journal articles, protocol papers, and trial registrations were used in quality appraisal. Once consensus was achieved, the remaining outcomes were assessed by one reviewer (n=38; RCCB). An overall risk of bias for each outcome was determined using the RoB2 method [36]. Quality appraisal was based on published manuscripts. The certainty of evidence for clinical effectiveness outcomes was evaluated using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) approach [37,38]. This was facilitated using GRADEPro GDT software [39], and assessed risk of bias, indirectness, inconsistency, imprecision, and other factors at the outcome level. The certainty of evidence was categorized as “high,” “moderate,” “low,” or “very low.” Outcome domains were downgraded if ≥50% of included trials presented issues in each criterion. Given the effect of exercise per se on clinical effectiveness outcomes, GRADE was facilitated according to the comparator group activity.

Meta-analyses were completed using Comprehensive Meta-Analysis Software (version 4; Biostat) [40]. Clinical effectiveness domains reported in 3 or more trials with comparable control groups (ie, exercising or nonexercising) progressed to meta-analyses. Random effects meta-analyses with the Hartung-Knapp-Sidik-Jonkman adjustment allow for differences in treatment effect to be present and accounted for throughout the included trials with greater precision [41,42] and were conducted for the effects of the IVR exercise interventions for mobility and functional balance, condition severity (eg, WOMAC), quality of life, pain intensity, composite static and dynamic balance assessments, aerobic physical activity (device-measured), and static balance. Eligibility of outcome inclusion into individual meta-analyses was completed as part of the data extraction process. Double-counting of data for individual trials in meta-analyses did not occur. Where applicable, data were converted from 95% CIs into SDs for inclusion into the meta-analysis. For the meta-analyses, preintervention and postintervention means and SDs, and sample size per group were used. A within-group Cohen d effect size was calculated to estimate the change from baseline for each group. We used a plausible and conservative pre-post correlation of .5 measured within each comparison group [43,44]. For the effect size difference between groups, standardized mean differences (SMDs) were used. This was due to the differing forms of measurement for the above-listed clinical effectiveness domains. Prespecified levels of magnitude for SMD were set at .2 for small, .5 for moderate, and .8 for large [45]. The SMD and 95% CIs were calculated using random-effect meta-analyses with the inverse of variance. Prediction intervals were calculated for meta-analyses with 5 or more included studies. Statistical heterogeneity was assessed via the Q-test and prediction intervals, and between-study variability was calculated using the I² statistic (0%‐25%=low, 26%‐74%=moderate, ≥75%=high) [46]. Prediction intervals were not calculated where statistical heterogeneity was low [47]. Data are presented in tables as per population group due to substantial heterogeneity identified. Small-study effects were assessed using the Egger test and visual inspection of funnel plots [48]. Additionally, the risk of bias due to missing results in meta-analyses was assessed through Egger test, funnel plots, and domains 3‐5 of the RoB2 tool. Analyses were grouped according to comparator group activity (exercising/nonexercising). Sensitivity analyses conducted included (1) removing individual trial results from the models to ascertain the effects on the overall results, and (2) running pre-post correlation at levels of 0.6, 0.7, 0.8, and 0.9 to assess influence on the overall result in line with Cochrane guidelines [43,44]. Subgroup analyses based on population were not completed due to the scarcity of common data.

This systematic review uses data that are publicly available through previous literature and did not require research ethical approval. The review was registered through PROSPERO International prospective register of systematic reviews on February 24, 2025 (CRD420250650110).

Figure 1 displays that 5241 records were identified in the initial search. A total of 26 trials were included, with 23 progressing to meta-analyses.

Trial characteristics are displayed in Table 1. Trials were conducted from 2021 to 2025 in Spain [28,49-51] (n=4), China [29,52,53] (n=3), Germany [54-56] (n=3), India [57-59] (n=3), and other countries [60-72] (n=13). A total of 846 people participated across all trials. Average sample size across trials was 33.5 (SD 18.0) participants, and mean age was 60.7 (SD 15.8) years. Trials included participants with neurological disorders [49-51,59,70,71], older adults [28,52,56,60,61,63,64,67-69], musculoskeletal conditions [29,55,57,58,62,66], cardiopulmonary diseases [53,65], cancer [54], and metabolic conditions [72]. The most common population group was older adults, accounting for 349 (41%) of all participants.

Details of the virtual reality interventions are summarized in Table 2. Aerobic interventions were used in 9 trials [50,52,54,57,60,61,63,65,72], anaerobic interventions in 4 [29,59,68,71], and a mix of aerobic/anaerobic in 13 [28,49,51,53,55,56,58,62,64,66,67,69,70]. Exercise intensity metrics were reported in 9 trials [50,53-56,63,65,70,72]. Dedicated IVR systems were used in 19 trials [28,49-57,60-67,69], 4 used smartphones in combination with a head mountable stand [29,59,68,72], and 3 did not specify [58,70,71]. Physiotherapists were the most common intervention provider (14/26, 54%). Intervention duration ranged from 30 days to 26 weeks, with 6‐8 weeks being the most common (14/26, 54%). Most participants in virtual reality groups were supervised in person (19/26, 73%).

Data on all 30 clinical effectiveness outcome domains identified in this review are presented in Table S1 in Multimedia Appendix 1. Clinical effectiveness outcome domains progressing to meta-analyses included mobility and functional balance, condition severity, quality of life, functional leg strength, pain intensity, composite static and dynamic balance assessments, aerobic physical activity (device-measured), and static balance. Outcome domains included in meta-analyses are summarized in Table 3.

The between-group analyses for IVR versus comparator, grouped by comparator group activity (exercising/nonexercising) are presented in Figure 2 [28,50,51,58,60,64,67-71] for mobility and functional balance (exercising: SMD −0.345, 95% CI −1.095 to 0.406; P=.29; nonexercising: SMD −0.322, 95% CI −0.931 to 0.288; P=.22), Figure 3 [28,50,51,56,60,67,69-71] for functional leg strength (exercising: SMD −0.161, 95% CI −0.573 to 0.250; P=.34; nonexercising: SMD −0.351, 95% CI −1.750 to 1.049; P=.48), Figure 4 [28,29,50,53-55,65,69] for quality of life (exercising: SMD 0.036, 95% CI −0.444 to 0.516; P=.85; nonexercising: SMD −0.053, 95% CI −0.839 to 0.728; P=.80), Figure 5 [29,49,50,55,57,58,62] for condition severity (SMD −0.153, 95% CI −2.227 to 1.921; P=.86), Figure 6 [29,55,57,58,62,66] for pain intensity (SMD −0.783, 95% CI −2.069 to 0.502; P=.18), Figure 7 [49-51,68,69] for composite static and dynamic balance assessments (SMD −0.310, 95% CI −0.870 to 0.249; P=.39), Figure 8 [60,64,67] for static balance (SMD −0.189, 95% CI −2.020 to 1.642; P=.70) and Figure 9 [29,63,66] for aerobic physical activity (device-measured; SMD 0.145, 95% CI −2.427 to 2.717; P=.83).

No significant effects were observed compared with exercising and nonexercising comparators for any outcome domain. Substantial heterogeneity was identified among most exercising (mobility and functional balance: T²=0.399, Tau=0.630, I²=80.3%, Q=25.38, P<.01; condition severity: T²=2.182, Tau=1.477, I²=92.7%, Q=82.06, P<.01; pain intensity: T²=0.646, Tau=0.804, I²=77.9%, Q=22.58, P<.01; composite static and dynamic balance: T²=0.291, Tau=0.540, I²=73.9%, Q=15.31, P<.01) and nonexercising (functional leg strength: T²=0.471, Tau=0.687, I²=75.9%, Q=12.46, P<.01; static balance: T²=0.308, Tau=0.555, I²=73.3%, Q=7.49, P=.02) comparator meta-analyses. Statistical heterogeneity was not identified for functional leg strength (T²=0.00, Tau=0.00, I²=0.0%, Q=2.65, P=.61), quality of life (T²=0.026, Tau=0.162, I²=18.4%, Q=4.91, P=.30) or aerobic physical activity (device-measured; T²=0.501, Tau=0.708, I²=63.8%, Q=5.53, P=.06) exercising meta-analyses, and mobility and functional balance (T²=0.096, Tau=0.031, I²: 42.0%, Q=6.90, P=.22) and quality of life (T²=0.000, Tau=0.000, I²=0.0%, Q=0.69, P=.71) nonexercising meta-analysis. Calculated prediction intervals were often substantially wider than the 95% CI, indicating significant heterogeneity for most outcome domains (Figures 2-9). Visual inspection of funnel plots did not suggest the presence of small-study effects for all outcome domains (Figures S1-S8 in Multimedia Appendix 1). However, the Egger statistic indicated evidence for a small-study effect for mobility and functional balance outcomes (Figure 2; P=.01). Sensitivity analyses identified single study removal of Zak et al [68] (functional balance, composite static and dynamic balance), Nishitha et al [58] (condition severity), Naqvi et al [57] (pain intensity), Stamm et al [55] (pain intensity), and Barsasella et al [60] (functional leg strength) resulted in altered statistical significance in meta-analytical models. Manipulation of pre-post correlation values did not result in altered statistical significance for any meta-analysis.

Feasibility outcomes are reported in Table 4.

Risk of bias assessments for individual trials progressing to meta-analyses are presented alongside forest plots in (Figures 2-9). Risk of bias proportions for separate outcome domains are presented in Table S2 in Multimedia Appendix 1. Many trials exhibited issues with measurement and selection of reported results across all outcome domains (domains 4 and 5 on the Cochrane RoB2 tool). In turn, 75% of outcomes included across all meta-analyses were rated as “some concerns” (n=18) or at “high” risk of overall bias (n=21). GRADE assessments for outcome domains progressing to meta-analysis are presented in Table S2 in Multimedia Appendix 1. All outcome domains were deemed to have “low” or “very low” certainty levels. Risk of bias, indirectness of intervention, and imprecision of effects had the main impacts on the level of certainty.

This systematic review with meta-analyses assessed the clinical effectiveness of IVR interventions using aerobic or anaerobic exercise. The analysis combined 26 RCTs, including 846 participants. Population groups included adults with neurological disorders, older adults, musculoskeletal conditions, cardiopulmonary diseases, metabolic conditions, and cancer. Pooled data indicated no differences compared with the comparator for clinical effectiveness, with high levels of statistical and clinical heterogeneity and low certainty ratings leading to decreased confidence in findings. While scarcely reported among included trials, findings were generally positive regarding the feasibility of IVR exercise. This has important ramifications for both practitioners who are seeking to adopt IVR as a novel alternative for exercise delivery and researchers conducting future trials.

For all clinical effectiveness outcome domains progressing to meta-analysis, certainty in effect was classified as “very low” or “low.” Many included trials exhibited issues with the measurement and selection of reported results, leading to increased risk of bias. Limited comparative data, small sample sizes, substantial heterogeneity, and increased risk of bias across all outcomes decreased certainty in all findings. This outlines a need for robust clinical trials to be conducted to improve confidence in observed findings.

Pooled data seem to suggest that aerobic or anaerobic exercise via IVR may be comparable to exercising controls. This observation seems to be common across supervised and unsupervised comparator groups across populations. Recent evidence suggests that IVR may be effective for mitigating short-term pain intensity during and post exercise (≈30% reduction) [74], although evidence on longer-term pain is unclear [75]. Attentional distraction and altered pain modulation may be potential mechanisms enhanced in IVR [76]. This review identified a nonsignificant effect favoring IVR for pain intensity compared with exercising controls, marred by high statistical and clinical heterogeneity across population groups. The relationship between IVR exercise and pain intensity outcomes should be emphasized in future robust clinical trials.

No significant differences were observed between IVR and nonexercising comparators for all meta-analysis outcome domains. This observation may be in part due to a low number of trials with nonexercising comparators progressing to meta-analyses. This may have led to a potential lack of power to detect any differences in the meta-analytical models. Exercise prescription, supplementary equipment, and additional intervention components received by the IVR groups varied substantially across all trials. While it cannot be expected that all weekly physical activity be completed in IVR, most included trials either did not meet the World Health Organization physical activity [77] or American College of Sports Medicine resistance exercise guidelines [78] (inclusive of other intervention elements) or did not report exercise prescription in sufficient detail. This indicates a potential issue with appropriate intervention dosage and a lack of reporting for activities being completed outside of the trial context. This lack of clarity in dosage or reporting could have contributed to the nonimprovement of relevant outcomes and may help to partly explain meta-analysis findings.

Inclusion criteria for this review mandated the inclusion of aerobic or anaerobic exercise as part of the IVR intervention. As such, it is surprising that common exercise-related outcome measures were not regularly assessed (eg, muscular fitness and cardiorespiratory fitness). A 2024 systematic review identified that IVR improved physical activity enjoyment, intrinsic motivation, and exercise intention compared with traditional and nonimmersive physical activity interventions [20]. Muscular and cardiorespiratory fitness outcomes were not often assessed across trials. Given the clinical importance of these outcomes for overall health [79,80], future research should prioritize their collection. A 2022 RCT in apparently healthy adults used a resistance training-based IVR intervention compared with a self-directed exercise approach [24]. The trial found that IVR exercise improved pertinent cardiometabolic and physical fitness measures (eg, muscular strength, systolic blood pressure) significantly more than control, possibly due to gamification elements improving exercise adherence [24]. Future research should assess whether resistance training-based IVR has a similar effect on cardiometabolic outcomes for clinical populations. Given the broad population inclusion criteria in the current review, there was a lack of literature from certain chronic condition populations where exercise interventions may be used for disease control (eg, type 2 diabetes) [81]. In addition to the populations identified in this review, future work should attempt to assess clinical effectiveness outcomes in these groups.

Determining the feasibility of IVR exercise for service delivery is important for widespread implementation. Safety concerns for IVR exercise may be compounded by technology-related factors (eg, immersiveness decreasing spatial awareness) contributing to potential injury risk [82]. However, regular physical activity and exercise are often critical for condition management [83-85], and IVR presents a unique opportunity to stimulate exercise enjoyment [20]. Previous literature suggests that IVR is safe for physical rehabilitation [86-88], but exercise-specific literature is lacking. While IVR may present an opportunity to deliver exercise in a novel way, individual risk analysis is imperative to address safety concerns. This is particularly true for population groups susceptible to IVR-specific side effects, such as cognitive overload and motion sickness [89]. While only just over half of the trials reported safety metrics (14/26, 56%), no trials in the current review reported an increased exercise-related adverse event rate in IVR groups for any population.

A common side effect reported from IVR interventions is motion sickness, commonly derived from eye strain, disorientation, or nausea [90]. Recent literature reports high variability concerning the incidence of motion sickness, with between 25% and 60% of participants experiencing symptoms [91]. Occurrence of motion sickness in IVR is multifaceted and linked to hardware type (eg, display type and mode, time delay), IVR content (eg, duration, controllability, realism), and individual susceptibility [90,91]. This review identified minimal-slight severity symptoms, indicating that motion sickness was not common or severe in controlled environments. However, findings regarding motion sickness were not commonly reported among trials (8/25, 32%). This is surprising given the high incidence of motion sickness symptoms in IVR and the unclear additional effects of exercise. Therefore, for clinicians using IVR exercise, we recommend an individualized approach to combat potential motion sickness symptoms. This includes (1) selecting IVR hardware and content based on patient feedback [90,91], (2) accounting for motion sickness history and IVR-readiness [90,91], and (3) using IVR interventions based on safety guidelines for individual systems [92].

Perceived technical issues have been cited as a barrier to the implementation of IVR among health care professionals [93]. This includes technical malfunctioning of IVR equipment, unstable Internet connectivity, and system usability [93]. Technical issues were poorly reported in the current review, with only 2 trials describing experiences with technical challenges. Given that a perceived lack of experience/confidence and fear of technical issues disrupting clinical sessions may dissuade clinicians from using IVR [93], future trials should prioritize their reporting and troubleshooting techniques. While trials in the current review suggest that attendance to sessions and adherence to exercise prescription were high, they were not reported commonly and lacked standardized presentation. The inability to verify whether participants attended and completed prescribed sessions makes it difficult to determine whether the IVR exercise was associated with the observed findings. This notion is supported by a 2024 systematic review of physiotherapist-led telerehabilitation interventions, which identified a lack of standardized reporting for attendance and adherence [94]. Since these are often linked with clinical effectiveness, more nuanced and comprehensive investigations into attendance and adherence are warranted [94].

Generally positive participant experiences were observed for participants using the IVR exercise in the included trials. Self-report questionnaires identified high levels of enjoyment and motivation, system usability, and overall acceptability among participants. Qualitative descriptions of experience found that perceived personal benefits and novelty of IVR may be the driving factors for positive experiences. These themes are consistent throughout the literature, with high degrees of immersion [32], distraction [32,95], and incorporation of gamification (eg, leaderboards) [32,96] often cited as factors influencing positive experiences with IVR. Sources of negative experiences related to the inconvenience of wearing an HMD for exercise. A 2022 systematic review observed that older adults reported several problems with wearing an HMD [97]. These included that the HMD was too heavy, caused general discomfort, and resulted in negative affect [97]. Many participants were willing to tolerate these discomforts to experience positive IVR features (such as immersive networking) [97]. These findings, combined with the current review, may suggest that the selection of IVR exercise should be individualized and based on participant preference.

Due to the advancement of IVR in recent years, the potential for integration into exercise practice is immense. This review identifies that future trials should investigate the clinical effectiveness of IVR exercise with more rigorous methodology. Future studies should focus on the measurement and selection of the reported results being derived from pre-established protocols to minimize bias, measuring common exercise-related outcome measures, sample size calculations to acquire statistical power, and intention-to-treat analysis to increase the validity of results. Future studies should emphasize the reporting of session attendance and exercise adherence, technical issues, and participant experiences to assess overall feasibility.

This systematic review has several important limitations to consider. Cumulatively, there was a considerable amount of bias among included studies resulting from the measurement and selection of reported clinical effectiveness results. In turn, “low” to “very low” certainty in the findings was identified through GRADE analysis. Both comparative (IVR vs alternate intervention) and additive (IVR + usual care vs usual care) intervention designs were included for exercising comparators, limiting the specificity and interpretability of findings. While SMD allowed for the comparison of multiple outcome measures across different scales, variability in SDs between populations may limit generalizability. This may have resulted in confounding and potentially distorted the observed effects in meta-analyses [98,99]. Additionally, small-sample bias may have been present in meta-analyses [73]. High statistical heterogeneity was observed in most meta-analyses, and sensitivity analyses revealed statistically significant differences in multiple meta-analytical models through the removal of single trials. These could not be accounted for using meta-regression due to the small number of trials included for each outcome and indicate the fragility of meta-analytical models. Additionally, a confidence distribution approach for the calculation of prediction intervals in meta-analyses with small trial numbers [100] could not be implemented through CMA software. Therefore, the reported prediction intervals may underestimate the uncertainty of the generated effects. A large amount of variety in population groups and participants is present in this review, and subgroup analyses by condition domain were not possible due to sparse outcome data. Therefore, overall findings should be observed with caution when applied to specific populations.

The findings of this systematic review incorporating meta-analyses provide initial evidence for the clinical effectiveness of IVR exercise interventions. The initial evidence may suggest that IVR exercise interventions do not seem to statistically differ from nonexercising or exercising comparators for changes in clinical effectiveness outcomes. However, significant heterogeneity, high risk of potential bias, and low certainty ratings decrease confidence in the observed findings. While the results indicate that IVR may be a viable option for the delivery of exercise, future trials with more robust methodology for the monitoring and reporting of clinical effectiveness outcomes are needed to verify findings. Importantly, future trials should also emphasize reporting of session attendance and exercise adherence, safety metrics, technical issues, and participant experiences. Implementing these measures will limit the risk of bias observed and provide more accurate insights into the effectiveness and real-world applicability of IVR exercise interventions.