RCTs from PubMed, Embase, MEDLINE (Ovid), Cochrane Library, and APA PsycINFO databases were comprehensively searched. RCTs published in English before April 15, 2021, investigating the impact of VR-based therapies on the cognition and mental health of patients with stroke were included. In addition, we conducted reference tracking on the published trials and meta-analysis reviews in this field to make sure all relevant studies were included. The search strategies, including terms for stroke, VR, cognition function, the QoL, and depression, are presented in Multimedia Appendix 1.

Two reviewers independently assessed studies obtained from the database searches in three phases: title, abstract, and full-paper screening. Data were extracted by the same two reviewers from trials using a Microsoft Excel spreadsheet. The information extracted from each trial included the first author, year, setting, sample size, participant ages, details of intervention and the control, outcomes, and corresponding measures. Disagreements were resolved through a discussion with a senior investigator.

The Cochrane risk-of-bias tool was used to assess the quality of eligible trials, focusing on sequence generation, allocation concealment, blinding, attrition bias, completeness of outcome data, and other sources of bias. Any disagreements were settled by a discussion with a third person.

The standard mean difference (SMD) and weighted mean difference (WMD) with a 95% confidence interval (CI) were determined for continuity data. Statistical heterogeneity was assessed using the Cochran Q test and the Higgins I² statistical test. The results implied low-level heterogeneity when I² ＜ 50%, and we used a fixed-effects model to fix the issue. The results implied moderate or high heterogeneity when I² ≥ 50%, and subgroups were determined based on the different characteristics of the studies to identify the sources of heterogeneity. Sensitivity analysis was performed by comparing the effect sizes and CIs of the remaining RCTs after removing each included RCT at a time. Descriptive analysis was used to deal with nonmergeable data. Potential publication bias was assessed with funnel plots and the Egger test if the group included 10 trials or more [41]. All meta-analyses were performed in RevMan 5.3 (Nordic Cochrane Center, Cochrane Collaboration, Copenhagen, Denmark) for outcomes that were evaluated in at least 2 of the included RCTs.

A total of 7117 studies were identified from the 5 databases searched. A further 2932 studies were identified through manual searching, including searching for systematic reviews, and included studies’ reference lists. After 261 duplicates were removed, 4189 full-text manuscripts were identified by screening their titles and abstracts. After full-text reviews, 23 studies (including 894 participants) from 8 countries satisfied the inclusion criteria and were eventually included in our systematic review (Figure 1).

Multimedia Appendix 2 shows the study characteristics and patient demographic data of the included trials. Sample sizes ranged from 18 to 145, for a total of 894 patients with stroke. In total, 11 (47.8%) of the 23 trials were conducted in Korea [34,38,42-50], 3 (13%) in Portugal [51-53], 2 (8.7%) in Spain [54,55], 2 in China [36,37], 2 in Australia [56,57], and 1 (4.3%) each in Lithuania [58], Brazil [59] and Turkey [60].

Regarding the types of VR-based therapies, 16 (69.6%) trials administered VR as a singular session [36-38,44, 45,47-49,51,52,54-57,59,60], 5 (21.7%) trials were categorized as having used VR-based therapies plus occupational therapy [43,46,50,53,58], and 2 trials as having used VR-based therapies plus computer-assisted cognitive rehabilitation [34,42]. The frequency and duration of VR exposure varied considerably between trials. In our review, the duration of VR-based therapies ranged from 3 to 10 weeks, with the majority being 4 weeks, and the frequency of intervention varied from 2 to 5 times per week. The control group underwent conventional rehabilitation therapy [36-38,44,47,48,51,54,56,57,59] or other support interventions such as occupational therapy [43,45, 46,49,50,55,58], computer-assisted cognitive rehabilitation [34,42], and other rehabilitation therapies [52,53,60].

Of the 23 included trials, outcomes of our interest included global cognitive function (10 trials, 43.5%), executive function (5 trials, 21.7%), memory (5 trials), verbal fluency (2 trials, 8.7%), visuospatial ability (2 trials), attention (6 trials, 26.1%), depression symptom (5 trials), and the QoL (7 trials, 30.4%). Although the tools used for outcome evaluation varied across trials, all were valid scales, and the process of data collection was carried out by experienced staff. Tools used to evaluate global cognitive function included the Montreal Cognitive Assessment (MoCA) [52,53,56,58], the Mini-Mental State Examination (MMSE) [34,45,49,51], and the Loewenstein Occupational Therapy Cognitive Assessment (LOTCA) [38,42]. Five trials evaluated the executive function domain using the CogState Groton Maze Learning Task [56], the Tower of London Test [34], the Stroop test [48], Trail-Making Test-B [51], and the the Digit Span Test [52]. Tools used to evaluate memory of patients with stroke included the visual recognition test [42], the visual span test [34], Wechsler Memory Scale-III [52], the Digit Span Test [54], and the Trail-Making Test [48]. The Digit Span Test [34] and Wechsler Memory Scale-III [52] were used to assess the verbal fluency of patients. Visuospatial ability was evaluated using the Motor-Free Visual Perception Test-3 [43] and the visual span test [34]. Attention was evaluated using the Single Letter Cancelation Test [53], Color of Color Word in Word–Color Test [34], and the Trail-Making Test [48,51,52,54]. The Hamilton Scale [36,46,55], the Beck Depression Inventory [37], and the Neurobehavioral Functioning Inventory [56] were used to assess the level of depression. The Stroke Impact Scale [38,47,50] was primarily used to assess the QoL, followed by the Short-Form 36 Health Survey Questionnaire [59], the Short-Form 8 Health Survey Questionnaire [44], the Nottingham Health Profile [60] and the EuroQoL Five Dimensions Questionnaire [57].

The risk of bias is summarized in Figure 2. In general, all 23 trials included in our review showed an acceptable risk of bias. The randomization sequence was adequately generated in 20 trials (86.9%), and 11 trials (47.8%) adequately concealed allocation. We categorized the risk of performance bias in all studies as low because blinding of participants and personnel was not possible in our systematic review. Approximately half of the trials blinded outcome assessors, and the risk of detection bias of these trials was judged as low. Of the 23 trials, 16 (69.6%) had no dropouts or used the intention-to-treat principle to compensate for dropouts; therefore, their risk of reporting bias was rated as low. All trials were categorized as having a low risk of bias in terms of attrition and other bias.

The outcomes of global cognitive function, domain-specific cognition, and mental health were evaluated using various tools in the included trials. Domain-specific cognition included executive function, memory, verbal fluency, visuospatial ability, and attention in our review. Depression and the QoL were synthesized in our review. The change scores from baseline to final values were used in our final efficacy analysis. The results of our analysis of each outcome are as follows.

Funnel plots for global cognitive function showed symmetry. The Egger test of global cognitive function was not significant (P=.29, Multimedia Appendix 4), indicating no significant publication bias in our review. A sensitivity analysis was conducted, excluding individual trials one by one, to confirm the results of our meta-analysis. We found that except for the outcomes of executive function, the rest of the outcomes showed no substantial modification of the overall effect.

Our meta-analysis, based on 23 RCTs, indicated that VR therapy is an effective method of improving executive function, memory, and visuospatial ability on patients with stroke. However, current evidence fails to support the effects of VR-based therapies in improving global cognitive function, attention, verbal fluency, depression, and the QoL. Unlike prior systematic reviews [8,25] primarily focused more on either one type of VR, such as exercise-based VR, or one aspect of cognition indicators, such as only global cognitive function, our review [40] comprehensively evaluated the effect of VR-based therapies on global and domain-specific cognition and mental health outcomes.

VR-based therapies have demonstrated efficacy in the diagnosis, physical rehabilitation, and cognitive rehabilitation of individuals with neurocognitive disorders [17,19]. A growing body of studies has confirmed that the application of VR is effective in cognitive rehabilitation for patients affected by stroke [34,36,56]. However, our pooled results showed no significant benefits of VR in improving global cognitive function for patients with stroke. Our result was comparable to recently published meta-analysis reviews by Wiley et al [39] and Zhang et al [40], who reported that VR therapy is not superior to control interventions in improving global cognition in individuals with stroke. This result differs from another systematic review [25], which only included 4 studies published in 2018. The review found that there is a small-to-medium effect of VR therapy on cognitive outcomes for people after a stroke. A possible explanation for this conflicting finding is that most VR-based therapies do not always concentrate on the training of cognitive function. It remains to be determined whether more training sessions specifically focused on global cognitive function would affect outcomes.

Furthermore, the samples had relatively heterogeneous characteristics, such as stroke onset, the severity of the impairment, and lesion characteristics, which may affect cognition recovery in poststroke patients.

Cognitive function is a complex concept that includes various domain-specific cognition, such as attention, executive function, memory, and psychomotor speed [61]. In our review, we found improvements in executive function, memory, and visuospatial ability, indicating that VR can be considered an effective therapy for improving domain-specific cognition. These positive findings of VR therapy are in line with previous reviews [19,62,63], which report that VR represents a promising methodological approach, implementing specific cognitive and behavioural functions, such as executive function, attention, spatial cognition, memory, and language. A previous systematic review by Alexander et al [64] also supports these positive findings on domain-specific cognition. Wiley et al [39] employed a systematic review to examine the effectiveness of exercise-based VR therapy for poststroke, with 8 studies involving 196 participants, which did not find improvements in attention, memory, and language poststroke. The added value of VR in domain-specific cognition compared with most of the currently provided conventional therapies may be underlined by the following mechanisms. Flannery et al [65] stated that VR training activates brain metabolism, increases cerebral blood flow, and the release of neurotransmitters. Carrieri et al [66] confirmed that VR can foster the reactivation and improvement of various cortex functions and optimize the efficiency of the sensory cortex and is also effective in improving cognitive function.

Poor mental health after a stroke is common and complex. An individual’s QoL is greatly reduced by re-experiencing the situation with depression, anxiety, agitation, and stress. Many studies and clinical trials have shown the potential of VR in relieving stress, depression, and anxiety in an imagined space, which makes it possible to provide efficient educational and psychological training without causing harm to patients with psychological problems [36,67]. There are several possible explanations for these improvements in mental health. Damsbo [68] explained that the use of VR allows individuals to learn emotion regulation strategies in the context of life-like virtual environments, thereby alleviating their negative emotions during long-term recovery. Choi et al [44] thought VR, through quasi-naturalistic and realistic stimuli in a multisensory fashion, can significantly enhance patients’ awareness of the movement performed, as well as self-identification and self-recognition, and makes patients derive a high level of interest and enjoyment via the use of VR. However, the overall effects of VR on depression and the QoL were not encouraging in our meta-analysis. Some studies agree with the findings of our review [69,70]. The results of Lee [37] showed that some VR-based games are not suitable for patients following a stroke. Furthermore, patients with cognitive impairment may feel frustrated during the therapy, resulting in poor emotional experiences. Hence, the degree to which participants feel motivated and engaged during VR therapy can depend on the individual characteristics and the intervention content. The investigation of the mechanisms underlying how VR can influence mental functions is a critical point in stroke rehabilitation research. A likely cause of disagreement in findings on the QoL is that the QoL may worsen over time after stroke. However, the mean duration since stroke is inconsistent, even with no restrictions on duration among the trials included in our review. It is, therefore, possible that the duration since stroke is an important factor that should be considered when selecting a VR therapy to increase the QoL for stroke patients.

Most subgroup analyses did not reveal significant differences between groups. Subgroup analyses consisted of a small number of comparisons, and the lack of relevant differences in most subgroup analyses might be caused by low statistical power. Only the subgroup analysis on global cognitive function showed that the effect of the usual-care control group was significant. There was also the issue with moderate-to-high heterogeneity in our meta-analysis, and grouping the included trials by measurement tools, duration of intervention, and delivery of control did not eliminate heterogeneity. Given the highly variable interventions of included trials, it is not surprising to find high heterogeneity in this case.

This meta-analysis followed the guidelines for performing rigorous systematic reviews [41]. Our review was only based on RCTs, which reinforces the evidence of our results. We also proposed a rigorous screening and search strategy to identify the most comprehensive literature in five major databases. Consequently, the results of our review are a widespread belief. However, there were still some potential limitations to our review. First, the number of trials included in our meta-analysis was limited. Hence, the power to detect small effect sizes was limited. Second, the included trials were extremely varied in terms of intervention contents and doses, and delivery of the control and measurement instruments, which made it rather complex to determine the optimal intensity of VR programs. Third, the quality, quantity, and sample size of included trials were far from ideal. Concealed treatment allocation was often not guaranteed, and blinding of outcome assessments was often not carried out. For example, only less than half of the included trials reported details of allocation concealment, leading to selection bias or confounding of the pooled results. Concerning detecting bias, almost half of the included trials failed to report information or had a high risk of blinding outcome assessment. Despite the rigorous nature of the included research designs (only included RCTs), our results must be interpreted with caution. Fourth, since we only considered effect post-intervention, the sustainability of VR-based therapy effects was not explored in our review. Indeed, details of long-term effects were unavailable in most trials included. Further research is needed to assess whether the effect observed in our review would persist over time and, if not, to evaluate at what duration and frequency the VR therapy should be repeated in order to sustain its effect. Finally, although a significant amount of work has been done in this area with promising results, the relevant characteristics of VR systems and the quantification of their impact on recovery are not yet clearly understood. As a result, we do not know how the different parameters of the proposed VR scenarios exactly affect recovery or whether they are effective at all. There is also a need to consider individual variability in order to optimize the impact of training.

Notwithstanding potential limitations, the present findings offer some implications for researchers and health practitioners. In this high-tech era, clinicians may have more options and alternatives by providing an interactive and visually stimulating approach for patients with stroke, especially for those who cannot easily access traditional rehabilitation methods. The following issues need to be considered when applying VR-based therapies. First, VR should be adapted to the patient’s needs and characteristics in performing activities, tasks, and tests. Second, safety problems need to be highlighted for older adults with reduced vision or other sensory problems. It is critical to consider a methodology type and an interaction technique that will result in the safe implementation of a VR therapy for patients with sensory and cognitive impairment. Finally, the price of VR equipment should be considered to meet the needs of target populations. In addition, there is a need for further research in this field to promote cost-effective care.

VR is a promising approach and can be used effectively in clinical neurorehabilitation. Although existing studies are limited, this review demonstrated statistically significant effects of VR-based therapies on executive function, memory, and visuospatial function in patients with stroke, but not on global cognitive function, attention, verbal fluency, depression, and the QoL. Larger, multicenter RCTs are warranted to confirm these positive effects. The completion of high‐quality trials will ultimately advance the knowledge about optimal cognitive and psychological rehabilitation strategies for patients with stroke.