The design, execution, and reporting of this meta-analysis adhered to the PRISMA guidelines of the 2020 statement [22]. This study has been registered at PROSPERO, and the registration number is CRD42023445000.

No humans or animals were involved in this meta-analysis, so ethical approval was waived.

An exhaustive search was systematically performed across multiple databases, including PubMed, Web of Science, MEDLINE, EMBASE, and Cochrane Library. In light of the progressive maturation of VR technology subsequent to 2010 and the increased scholarly attention toward exploring VR as a modality for supporting psychological and cognitive interventions during this period [23], the timeframe for our literature search shall be delimited from January 2010 to December 2024. The used search methodology was grounded in the PICOS framework, encompassing the specific elements of Patient or Population, Intervention, Comparison, Outcome, and Study type. No specific restrictions were applied regarding demographic variables such as age, gender, education, etc. However, for the language, the search was conducted exclusively in the English language. The search methodology used multiple strategies, including examination of titles and abstracts, use of relevant keywords, and incorporation of free and medical subject headings terms, etc. The specific retrieval strategies for each database are outlined in Multimedia Appendix 1.

Inclusion criteria for our review encompassed the following criteria: (1) participants were diagnosed with neuropsychiatric diseases; (2) interventions used the VR technology; (3) the control groups were assigned to either receive conventional treatment or were placed on a waitlist for comparison purposes; (4) the study quantitatively assessed the cognitive function. No restrictions were imposed on the specific measurement tools or instruments used to evaluate the indices; (5) the literature was published in English; and (6) the included studies were designed per the principles of RCTs. Exclusion criteria for this study encompassed the following criteria: (1) qualitative study, conference paper, letter, review, commentary, or protocol; or (2) studies where access to full-text articles was not available or where relevant full data were not available.

All identified study titles and abstracts were retrieved and imported into Zotero 7.0 (Corporation for Digital Scholarship). Duplicate papers were subsequently removed from the dataset. Two researchers independently assessed the titles and abstracts of the studies in parallel to ensure compliance with the inclusion criteria. Subsequently, the full texts of each study were reviewed per the predetermined eligibility criteria. Discrepancies were resolved through a consensus-seeking process involving discussion and consultation with a third investigator.

Two researchers independently performed data extraction from the selected studies and documented the relevant information within a pre-established data extraction form, encompassing the following essential components: the first author, the published year, the country or region where the trial was conducted, the disease type, the ages (mean and SD), the years of education (mean and SD); the sample size (experimental or control group), and the gender; the type, content, hardware, software, immersive or nonimmersive approach and detailed session information of the intervention; and the assessment time and the instruments of outcome measures.

Using the Cochrane RoB2 (Risk of Bias 2) tool, 2 investigators carried out a separate evaluation of the methodological rigor and potential bias of the selected studies. RoB2 is a systematic review tool developed by the Cochrane Risk of Bias Working Group aimed at assessing the methodological quality of RCTs [24]. This tool identifies potential sources of bias that may affect the reliability of study outcomes through a series of guiding questions that cover key areas such as the randomization process, deviations from intended interventions, missing outcome data, outcome measurement, and selective reporting. It provides a final judgment on the risk of bias. Third-party consultation was used to resolve disagreements.

We used Review Manager 5.3 (Cochrane) to conduct our statistical analyses. For continuous outcome indexes, the mean difference along with a 95% CI was used when all selected studies reported outcomes using the same unit and magnitude; otherwise, the standard mean difference was used to express the results.

Initially, the chi-square test was used to assess the statistical heterogeneity among the outcomes reported in each individual study, accompanied by the calculation of the corresponding P value; to quantitatively evaluate the magnitude of heterogeneity, the I² test was used [25]. Given the likelihood of functional disparities among studies conducted independently by different researchers, a random-effects meta-analysis was conducted for the outcome to account for potential heterogeneity across the included studies. Additionally, we conducted a subgroup analysis based on different types of interventions. To explore the potential reasons for the differences in the integrated outcomes of various intervention types based on VR technology, we conducted another subgroup analysis for different types of neuropsychiatric disorders and carried out sensitivity analyses by excluding certain studies. We constructed a funnel plot to evaluate potential publication bias systematically across the studies. In the final stage, forest plots were generated to visually present the effect estimates of outcome variables across all included studies. A level of statistical significance was established at P<.05.

According to the 5 electronic databases, the initial search yielded a total of 2181 documents. Following a deduplication process, 552 redundant documents were eliminated. The exclusion of 1568 articles was based on a thorough evaluation of their titles and abstracts. Subsequently, the full texts of the remaining 61 articles underwent rigorous examination. Forty studies were excluded after full-text assessment for the following reasons: 7 were protocol studies, 22 did not employ VR technology, and 11 had outcomes unrelated to cognition or produced cognitive assessment results that could not be quantified. Consequently, 21 studies were included in the meta-analysis. A visual representation of the search and selection processes is provided in Figure 1. The unweighted Cohen κ values for study selection (0.86), data extraction (0.84), and risk of bias assessment (0.81) indicate an excellent level of interrater agreement, as per Cochrane Handbook criteria.

The methodological quality of the 21 selected studies is depicted in Figure 2. Most studies (n=18, 85.7%) were overall rated as low risk. However, the study by Buele et al [18] was assessed as high risk overall due to a lack of specific information on whether all participants adhered strictly to the intended intervention. Additionally, during the study, 5 participants withdrew. The text does not mention how these missing data were handled, leading to a high-risk rating due to deviations from intended intervention and bias arising from missing outcome data. Furthermore, the studies by Oliveira et al [26] and Park et al [38] were also rated as high risk overall because they did not specify if blinding was used and did not clearly state whether all participants adhered strictly to the intended intervention, resulting in a high risk of bias due to deviations from the intended intervention.

To assess publication bias among the studies, a funnel plot is drawn. In this plot, the distribution of points is not entirely symmetrical, particularly on the left side, which could be an indication of publication bias. However, since most of the studies included in the meta-analysis report were smaller sample sizes, they may not accurately reflect publication bias (Figure 7).

The principal aim of this meta-analysis was to quantitatively assess the efficacy of VR-based interventions on cognitive function among individuals with neuropsychiatric disorders. Through subgroup analyses, we sought to elucidate the efficacy of distinct intervention types and to investigate potential reasons for differences in results. Our findings demonstrated that VR-based interventions yield a significant positive effect on cognitive function in patients with neuropsychiatric disorders. Specifically, subgroup analyses revealed that cognitive rehabilitation training, exergame-based training, and telerehabilitation and social functioning training were associated with improvements in cognitive performance. Conversely, immersive cognitive training, music attention training, and vocational and problem-solving skills training did not significantly enhance cognitive functions within this patient population. Further analysis from the standpoint of disease categorization suggested that the unique characteristics of conditions for Parkinson disease and brain injury may underlie the lack of statistical significance observed in the meta-analytic synthesis for immersive cognitive training, music attention training, and vocational and problem-solving skills training, which underscored the importance of considering the specific etiology and pathology of neuropsychiatric disorders when designing VR-based therapeutic interventions.

This study is the first to use quantitative analysis methods to evaluate the impact of VR-based interventions on cognitive functions in patients with neuropsychiatric disorders, providing preliminary evidence of potential efficacy in this therapeutic area. It represents an advancement in personalized medicine and the development of digital health tools. We included data from several relatively high-quality RCTs, revealing the efficacy and potential of VR-based interventions in improving cognitive functions, which are critical for the rehabilitation of neuropsychiatric disorders.

In our study, VR-based interventions featured immersive and highly interactive characteristics that can stimulate neuroplasticity in the brain. This property allows the brain to adapt to new information and skills by forming new synaptic connections or strengthening existing ones [45,46]. In VR environments, the simultaneous provision of visual, auditory, and tactile sensory inputs promotes synchronous activity among different sensory processing areas of the brain. This multisensory integration process is essential for constructing a coherent cognitive representation and can, to some extent, compensate for deficits caused by damage to certain sensory channels [47,48]. Different types of VR experiences can specifically activate particular regions of the brain. For instance, visual-spatial navigation tasks may increase activity in the hippocampus, whereas complex social interaction scenarios could impact the prefrontal cortex and other brain areas associated with higher cognitive functions [49,50]. This targeted stimulation may aid in repairing or optimizing the functions of these brain regions. Furthermore, VR-based interventions can also influence the brain’s reward system. When users complete challenges or achieve goals within a VR environment, they typically receive immediate feedback and reward signals, such as increased scores or virtual prizes. This type of positive reinforcement mechanism can activate the reward centers in the brain, such as the nucleus accumbens, triggering the release of dopamine, which in turn can increase motivation levels and enhance learning outcomes [51,52]. Another crucial point is that the VR-based interventions in this review primarily simulate real-world life scenarios, which is beneficial for improving the transfer effect of cognitive enhancements to activities of daily living. This approach aids in social reintegration and boosts the ability to live independently [53].

Through subgroup analysis of intervention types, we found that immersive cognitive training, music attention training, and vocational and problem-solving skills training based on VR technology did not improve cognitive functions in patients with neuropsychiatric disorders. This is inconsistent with previous research findings [54]. In previous studies, the aforementioned interventions primarily targeted patients with mild cognitive impairment or dementia; in contrast, our meta-analysis also included patients with Parkinson disease and brain injury. Second, subgroup analyses in this study revealed that exergame-based training was associated with cognitive improvement. Emerging evidence suggests that body position moderates the effectiveness of exergame-based training on global cognition, with stepping movement training in a predominantly standing position showing greater benefits compared to weight shifting or exergaming in a sitting position [55]. Consequently, the positive effects of exergame-based training on cognition may have been underestimated in our study. Notably, among the 4 exergame-based training studies included in this analysis, 3 used a seated position, while only 1 involved a standing position [17,39-41]. The limited sample size prevented further detailed and effective analysis. Therefore, future research should investigate the influence of body posture on the efficacy of different types of VR-based training to develop more comprehensive and effective cognitive intervention programs. Subgroup analysis by disease types revealed that patients with Parkinson disease and brain injury did not show significant cognitive improvements from VR-based interventions. After excluding the study on Parkinson disease [34] through sensitivity analysis, immersive cognitive training demonstrated statistically significant improvement effects. Parkinson disease, characterized by resting tremors, muscle rigidity, bradykinesia, and postural instability, can limit a patient’s ability to interact effectively within a VR environment, thereby reducing the efficacy of training [54]. For example, decreased fine motor control may impair the ability to complete tasks requiring hand-eye coordination [56]. Furthermore, abnormalities in cortico-basal ganglia circuits in Parkinson disease may exacerbate executive dysfunction, further hindering cognitive training outcomes [57]. Similarly, different types of brain injuries can lead to either localized or diffuse brain tissue damage. Localized injuries typically affect specific functional areas, whereas diffuse injuries disrupt broader neural networks. When critical neural pathways are severed, simple repetitive training may be insufficient to rebuild these connections, thereby limiting the effectiveness of VR-based cognitive training [58,59]. However, emerging evidence suggests that while VR-based training often presumes baseline executive functioning, such as planning and organization [57], recent trials demonstrate its adaptability to populations with neurological impairments. For example, Dubbeldam et al [60] reported that technology use did not lead to higher dropout rates, and adherence was significantly higher in tailored interventions. This finding aligns with Swinnen et al [61], who conducted a systematic review and found high attendance adherence rates even in institutionalized individuals with major neurocognitive disorder. Thus, the limited efficacy observed in our study may reflect limitations in the intervention design rather than being solely attributable to inherent neuroanatomical constraints. These findings underscore the importance of tailoring VR-based interventions to the specific needs and capabilities of patient populations, as well as optimizing the design to maximize therapeutic outcomes.

The strength of this review lies in its novel use of meta-analysis to quantify the effects of VR-based interventions on cognitive functions in patients with neuropsychiatric disorders, significantly enhancing statistical power and providing a more comprehensive and robust estimate of intervention efficacy [62]. Simultaneously, subgroup analyses were conducted to evaluate the effects of different VR-based intervention types and to examine potential reasons for efficacy variations from the perspective of disease types. These findings help identify potentially effective VR interventions and technical components for improving cognitive functions in patients with neuropsychiatric disorders and suggest differences in responses to various interventions across patient groups. Consequently, this provides a scientific basis for the development of personalized treatment plans and promotes the precise application and advancement of VR technology in the rehabilitation of neuropsychiatric disorders.

This study acknowledges several limitations that merit consideration. First, the studies included in our review exhibited certain heterogeneity due to differences among participants’ baseline characteristics, sample sizes, assessment methods, intervention durations, etc. Second, only original research articles published in English were included, which may lead to language-related selection bias. Additionally, there may be some publication bias in this study. Last, despite using a broad search strategy across 5 major electronic databases, some studies have incomplete or entirely missing information sources. Therefore, our conclusions should be interpreted with caution.

As VR-based interventions advance within personalized medicine paradigms, future research must adopt structured, user-centered frameworks to align technological development with the specific needs, preferences, and capabilities of populations with neuropsychiatric disorders. Frameworks such as the Generative Co-Design Framework [63] and the Multidisciplinary Iterative Design of Exergames [64] emphasize participatory, iterative design involving end users, clinicians, and industry partners. These methodologies, exemplified by projects such as Brain-IT (information technology) help ensure that interventions are grounded in real-world usability and adherence while integrating established guidelines such as the Medical Research Council Framework for Complex Interventions [65,66]. This approach prioritizes multidisciplinary collaboration and continuous feedback to enhance translational impact. Beyond efficacy validation, research must investigate the mechanisms through which VR enhances cognitive function, such as neuroplasticity and attentional engagement, to identify potential commonalities or differential pathways across disorders. Mechanistic clarity will inform the optimization of VR components and protocols to better target therapeutic effects. Thus, by integrating user-centered design, mechanistic investigation, and technological innovation, VR has the potential to evolve into a robust, evidence-based tool for neuropsychiatric rehabilitation, optimizing recovery trajectories and long-term patient outcomes.

This meta-analysis showed that interventions based on VR technology have the potential to improve cognitive functions in patients with neuropsychiatric disorders, particularly in areas such as cognitive rehabilitation training, exergame-based training, and telerehabilitation and social functioning training. The findings provide valuable evidence supporting the application of VR technology in the rehabilitation of neuropsychiatric disorders and highlight possible directions for future intervention strategies.