Stroke is one of the global leading causes of death and may cause long-term disability for many stroke survivors (Mendis, 2013; Andrew et al., 2014). Up to three-quarters of patients with poststroke experienced ongoing cognitive impairment (Pasi et al., 2012; Jokinen et al., 2015; Renjen et al., 2015). Cognitive impairment and functional disability are often associated with the following stroke. Furthermore, the depressive mode worsens the difficulties for patients with stroke to maintain their social and personal relationships. Clinical depression is characterized by behavioral, cognitive, and emotional features (Merriman et al., 2019). Cognitive performance is always associated with symptoms of depression (Nakling et al., 2017), and early cognitive deficits in patients after stroke may predict long-term depressive symptoms (Nys et al., 2006). Furthermore, poststroke cognitive impairment is associated with early and enduring activity limitations and participation restrictions (Stolwyk et al., 2021). These disorders might lead to a poor quality of life (QoL) for individuals with stroke and their families.

In recent years, interventions for poststroke motor and cognitive impairment, depression, and reduced functional independence have become the focus of international stroke rehabilitation research, and novel clinical rehabilitation therapies [e.g., virtual reality (VR), repetitive transcranial magnetic stimulation (rTMS), and robotic assistive therapies] have shown great potential in future practice (Langhorne et al., 2011; Winstein et al., 2016a; Gittler and Davis, 2018). VR-based training is defined by using computer hardware and software-generated user-computer interface for users to interact with virtual environments that relate to the real world to facilitate task-oriented training and provide multimodal feedback to augment functional recovery (Laver et al., 2017; Hao et al., 2021). Basic neuroscience behind VR-based treatment was the finding of mirror neurons (MNs) in the primary motor cortex (M1), dorsal premotor cortex, and supplementary motor area (SMA) from animal studies (Gentilucci et al., 1988; Rizzolatti et al., 1996; Rizzolatti and Sinigaglia, 2016; Mekbib et al., 2020). The evidence from human neuroimaging suggested that the neural mechanisms of VR on neural plasticity and motor reorganization in humans might be to stimulate the internal sensorimotor system through activating MNs in the cortical and subcortical motor control-related areas, particularly M1, SMA, and cerebellum (August et al., 2006(Prochnow et al., 2013; Mekbib et al., 2020, 2021; Hao et al., 2021).

Recently, many clinical studies favored VR-based intervention for motor function, balance, gait, and activities of daily living (ADL) in patients with stroke. Although multiple systematic reviews and meta-analyses have indicated that VR-based training was useful for upper limb motor function, lower limb motor function, balance, gait, and activities of daily living (ADL) in stroke (Henderson et al., 2007; Laver et al., 2011; Saposnik et al., 2011; Lohse et al., 2014; Laver et al., 2015; de Rooij et al., 2016; Li et al., 2016; Silver, 2016; Yates et al., 2016; Laver et al., 2017; Aminov et al., 2018; Al-Whaibi et al., 2021; Fang et al., 2021; Peng et al., 2021; Zhang et al., 2021), two recent articles published in The Lancet Neurology by Saposnik et al. (2016) and Silver (2016) argued that the methodological issues that existed in some of the studies (Broeren et al., 2008; Kwon et al., 2012) were the comparison of VR combined with conventional rehabilitation vs. conventional rehabilitation alone without active control. Such study design (Saposnik et al., 2011; Lohse et al., 2014; Laver et al., 2015) might create an imbalance in the total rehabilitation time, and the effect might be induced by any active intervention and might not be explained by VR (Saposnik et al., 2016; Silver, 2016).

Conventional paper-and-pencil exercises and computer-assisted cognitive training designed to improve specific domains of cognitive deficits are widely used for patients with stroke with cognitive impairment. However, traditional cognitive training is limited by its insufficient personalization and adaptation and suboptimal intensity (Faria et al., 2016; Maier et al., 2020). Preliminary results (Kim et al., 2011; Choi et al., 2014; Faria et al., 2016, 2020; De Luca et al., 2018; Kannan et al., 2019; Maggio et al., 2019; Oh et al., 2019; Maier et al., 2020; Manuli et al., 2020) suggested that VR-based training combined with traditional rehabilitation might be more effective for enhancing cognition, depressive mood, and QoL in stroke than traditional cognitive rehabilitation. However, there is no clear evidence concerning the effectiveness of VR for cognition, depression, and QoL in patients with stroke (Laver et al., 2011, 2015; Silver, 2016). Recently, several systematic reviews (Aminov et al., 2018; Wiley et al., 2020; Zhang et al., 2021) have evaluated the effectiveness of VR for cognitive impairment in patients with stroke. Aminov et al. (2018) included 4 studies that assess VR-based rehabilitation on cognitive outcomes and found that VR could induce significant gains on improvements in cognitive function. Zhang et al. (2021) combined 7 RCTs to evaluate the effectiveness of VR interventions for cognitive outcomes compared with control groups, but no significant difference was found. However, in the two meta-analyses (Aminov et al., 2018; Zhang et al., 2021), only global cognition examined by MMSE or MoCA test for screening cognitive impairment was included, and specific domains of cognition were not investigated. Wiley et al. (2020) performed a systematic review that included five manuscripts to evaluate VR-based intervention combined with rehabilitation exercise on global cognition and specific domains of cognition and concluded that VR therapy was not better than traditional rehabilitation interventions for enhancing cognitive function in stroke survivors. However, due to the limited number of original articles, small sample size, different types of VR devices, different VR intervention durations, and different stages after stroke onset, the results remain controversial.

To date, however, few systematic reviews and meta-analyses have investigated VR-based training for cognitive function in contrast to cognitive exercise or motor exercise on the chronic stage of stroke. Therefore, this study aimed to explore the effect of VR-based training on cognition, motor function, mood, and ADL among individuals at the chronic phase of stroke.

In the present meta-analysis from five high-quality RCTs, the baseline global cognitive scores all reported mild cognitive impairment, and we showed the superiority of using VR-based intervention combined with rehabilitation on overall cognition, attention, and executive function in individuals with chronic stroke compared with control groups. However, VR-based training combined with rehabilitation showed non-significant improvement in global cognition in patients with chronic stroke. The positive result of VR-based training on attention/execution and overall cognition of patients with stroke in the current meta-analysis was in consistence with data from a previous systematic review of mild cognitive impairment or dementia (Zhu et al., 2021). However, the negative result of VR-based training on global cognition of patients with stroke in the current meta-analysis was not in consistent with data from VR-based interventions on global cognitive function measured with MMSE or MoCA test in individuals with stroke (Aminov et al., 2018) and neurocognitive disorders (Moreno et al., 2019). The meta-analyses by Aminov et al. (2018) included 4 RCTs investigating the effect of VR-based rehabilitation in patients with stroke and reported small to medium effect VR intervention for cognitive outcomes. While the participants in 3 of the 4 included RCTs were at the subacute phase, 1 RCT did not report the time since onset of stroke before intervention. In terms of the severity for global cognition at baseline, 2 of the 4 included RCTs reported an average of mild cognitive impairment, and another 2 RCTs included an average of moderate cognitive impairment. The positive result of VR-based training on attention/execution of patients with stroke in our study was not in consistence with data from the previous meta-analysis (Wiley et al., 2020), while the negative result of VR-based training on global cognition of patients with stroke in our study was in consistent with data from previous meta-analyses (Wiley et al., 2020; Zhang et al., 2021). Wiley et al. (2020) performed a meta-analysis that included 5 RCTs to investigate the effectiveness of VR-based rehabilitation on global cognition and specific domains of cognition such as memory, attention, and language. However, the participants in 2 of the 5 included RCTs were at the subacute phase, 2 RCTs were at the chronic phase, and 1 RCT was at the subacute or chronic phase. In terms of the severity for global cognition at baseline, 3 of the 5 included RCTs reported mild cognitive impairment, 1 RCT included reported moderate cognitive impairment, and 1 RCT included reported moderate or mild cognitive impairment. Besides, one study included (Oh et al., 2019) did not exclude individuals with visual impairment. A meta-analysis by Zhang et al. (2021) included 7 RCTs using MMSE for measuring cognitive function, and the result showed no significant difference in cognitive outcomes after the VR interventions compared with control groups. However, for PEDro Scale scores, 5 of the 7 included RCTs in this meta-analysis were of low quality, and 3 of the 7 included RCTs were not present in the reference. In terms of the time since onset of stroke before the intervention, participants in 2 of the available 4 included RCTs were at subacute phase, 1 of the available 4 included RCTs was at chronic stage, and 1 of the available 4 included RCTs was at the subacute or chronic stage.

The results of the limited current review provided evidence that VR could not contribute to motor rehabilitation. In this meta-analysis, 3 of the included 5 RCTs measured motor function (upper extremity motor function), while two included RCTs (Oh et al., 2019; Maier et al., 2020) used VR device-based cognitive rehabilitation and one included RCT (Faria et al., 2018) used VR device-based cognitive-motor task. Due to the small number of participants, different VR devices aiming at improving cognition, and the different scales measuring motor function in the included studies, our analyses failed to find positive results. However, cognition and motor function potentially influence each other in many ways, indicating that motor recovery might positively affect cognition. Previous studies have shown the potential benefits of VR-based motor rehabilitation on cognitive and motor outcomes (Gamito et al., 2017; Manuli et al., 2020). Furthermore, VR-based cognitive-motor intervention systems are encouraging (Faria et al., 2018; Kannan et al., 2019; Manuli et al., 2020).

Improvement in ADLs was included in two of the studies. In one included study (Manuli et al., 2020), the FIM and SF-12 increased significantly in accordance with the improvement in cognitive and behavioral outcomes, indicating better QoL after the VR-based treatment. However, Faria et al. (2018) reported no significant difference in improvement for BI along with cognitive function. This potentially indicates that cognition and participant may influence each other. Poor performance in motor activities and poststroke cognitive impairment can lead to depression, anxiety, and impairment of social functioning (Kim et al., 2019). Two studies in this review reported that either motor-based VR training or cognitive-based VR intervention indicated a positive effect on cognitive abilities and improvement in motor functions with a reduction in a depressive mood. Furthermore, mood symptoms of patients should be examined in future studies for VR intervention.

The controversial findings for VR-based training for cognitive and motor function may be related to the following aspects. First, insufficient VR programs designed for cognitive function training might be used. Second, this might be related to the wide range of training duration. Third, the different assessment tools for cognition after stroke were used. Finally, the time since onset of stroke before intervention largely varied from the acute phase to several years after stroke.

While the VR-based intervention was argued not to be exclusively dependent on a higher dose, frequency, or daily intensity (Muratori et al., 2013; Gamito et al., 2014; Laver et al., 2015; Winstein et al., 2016b), the subgroup analysis of our meta-analyses found more positive effects of a greater dose of VR therapy (more than 20-h intervention), higher frequency (more than four times per week), and higher daily intensity of VR therapy (more than 60 min per day) in the recovery of cognition, motor function, mood, and ADL. These findings differed from the recent meta-analysis of VR studies. Positive results were seen (Palma et al., 2017) for VR interventions using immersive, semi-immersive, and non-immersive environments for patients with subacute and chronic stroke with a mean dose of 17.6 h for upper extremity and 13.2 h for motor function. Aminov et al. (2018) reported a mean daily intensity of 42 min and weekly intensity of 153.9 min, a mean frequency of three sessions a week and a median duration of 18 sessions, and a total of approximately 12 h VR interventions and found that VR showed no advantage for a greater duration, higher doses, or massed training schedules in individuals with stroke compared with the control group. However, in the review by Aminov et al. (2018), the duration of stroke before intervention was not mentioned in one of the 4 included RCTs measuring cognitive outcomes, three of the included RCTs reported inclusion criteria for patients with subacute stroke, only 2 included studies targeted cognitive function alone, and the other 2 studies also aimed to improve motor function. In this review, semi-immersive environments reported in 5 included RCTs were used for patients with chronic stroke. Large variability was found for the VR intervention parameters including RCTs, with studies providing up to 1.5 h intervention daily intensity, up to five times per week frequency, and up to 40 sessions duration.

The multisensory stimulation of head-mounted display (HMD) device-induced immersive VR training is thought to have a better effect on behavioral outcomes in patients with stroke. However, a recent study conducted by Gamito et al. (2014) revealed no significant difference between HMD-induced immersive VR and desktop screen-induced semi-immersive VR for increasing working memory and sustained attention in patients with stroke. The widespread use of HMD displays may have some limitations, and the HMD devices are more expensive than screens and often not portable, difficult to justify for lay users, and may cause visual discomfort. Thus, the non-expensive displays of semi-immersive environments of VR-based cognitive rehabilitation might provide a cost-effective schedule for patients with chronic stroke with cognitive and motor dysfunction.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.

YG, LM, CL, and HF conceptualized and performed the analysis. TW and JG supervised the study. YG, LM, and LLY drafted the manuscript. SZ, CL, and XY analyzed the data. YG and TW participated in the whole process and made final decisions. All authors contributed to the writing of this manuscript and read and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.