Better medical treatments in the acute phase after stroke have increased survival and with that the number of patients needing rehabilitation with an associated increased burden on the health care system.1 Novel technologies have sought to meet this increased rehabilitation demand and to potentially allow patients to continue rehabilitation at home after they leave the hospital.2 Also, technology has the potential to gather massive and detailed data (eg, kinematic and performance data) that might be useful in understanding recovery after stroke better, improving the quality of diagnostic tools and developing more successful treatment approaches.3 Given these promises, several studies and meta-analyses have evaluated the effectiveness of technologies that use virtual reality (VR) in stroke rehabilitation. In a first review, Crosbie et al4 analyzed 6 studies that used VR to provide upper-limb rehabilitation. Although they found a positive effect, they concluded that the evidence was only weak to moderate given the low quality of the research. A later meta-analysis analyzing 5 randomized controlled trials (RCTs) and 7 observational studies suggested a positive effect on a patient’s upper-limb function after training.5 Another meta-analysis of 26 studies by Lohse et al,6 which compared specific VR (SVR) systems with commercial VR games, found a significant benefit for SVR systems as compared with conventional therapy (CT) in both body function and activity but not between the 2 types of systems. This study, however, included a variety of systems that would treat upper-limb, lower-limb, and cognitive deficits. Saywell et al7 analyzed 30 “play-based” interventions, such as VR systems including commercial gaming consoles, rehabilitation tools, and robot-assisted systems. They found a significant effect of play-based versus control interventions in dose-matched studies in the Fugl-Meyer Assessment of the Upper Extremity (FM-UE).7 In contrast, a more recent large-scale analysis of a study with Nintendo Wii–based video games, including 121 patients concluded that recreational activities are as effective as VR.8A later review evaluated 22 randomized and quasi–randomized controlled studies and concluded that there is no evidence that the use of VR and interactive video gaming is more beneficial in improving arm function than CT.9 In all, 31% of the included studies tested nonspecific VR (NSVR) systems (Nintendo Wii, Microsoft Xbox Kinect, Sony PlayStation EyeToy). Hence, although VR-based interventions have been in use for almost 2 decades, their benefit for functional recovery, especially for the upper limb, remains unknown. Possibly, these contradictory results indicate that, at present, studies are too few or too small and/or the recruited participants too variable to be conclusive.10 However, alternative conclusions can be drawn. First, VR is an umbrella term. Studies comparing its impact often include heterogeneous systems or technologies, customized or noncustomized for stroke treatment, addressing a broad range of disabilities. However, effectiveness can only be investigated if similar systems that rehabilitate the same impairment are contrasted. This has been achieved by meta-analyses that investigated VR-based interventions for the lower limb, concluding that VR systems are more effective in improving balance or gait than CT.11Second, a clear understanding of the “active ingredients”3 that should make VR interventions effective in promoting recovery is missing. Therapeutic advantages of VR identified in current meta-analyses are that it might apply principles relevant to neuroplasticity,5,9 such as providing goal-oriented tasks,5,9 increasing repetition and dosage,5,9 providing therapists and patients with additional feedback,5,6,9 and allowing to adjust task difficulty.6 In addition, it has been suggested that the use of VR increases patient motivation,6 enjoyment,8,9 and engagement7; makes intensive task-relevant training more interesting4,7; and offers enriched environments.9 Although motivational aspects are important in the rehabilitation process because they possibly increase adherence,3 their contribution to recovery is difficult to quantify because it relies on patients’ subjective evaluation.7,12–15 Rehabilitation methods, whether VR or not, however, need to be objectively beneficial in increasing the patient’s functional ability. Hence, an enormous effort has been expended to identify principles of neurorehabilitation that enhance motor learning and recovery.16–24 Consequently, an effective VR system should besides be motivating, also augment CT by applying these principles in the design.23 Following this argument, we advance the hypothesis that custom-made VR rehabilitation systems might have incorporated these principles, unlike off-the-shelf VR tools, because they were created for recreational purposes. Combining the effects of both approaches in one analysis might, thus, mask their real impact on recovery. Again, in the rehabilitation of the lower limb, this effect has been observed. Two meta-analyses investigating the effect of using commercial VR systems for gait and balance training did not find a superior effect, which contradicts the conclusions of the other systematic reviews.11 In upper-limb rehabilitation, this question has not been properly addressed until the most recent review by Aminov et al.25 However, there are several flaws in the method applied that could invalidate the results they found. Specifically, studies were included regardless of their quality, and it is not clear which outcome measurements were taken for the analysis according to the World Health Organization’s International Classification of Function, Disability, and Health (ICF-WHO).26 In addition, a specifically designed rehabilitation system (Interactive Rehabilitation Exercise [IREX])27 was misclassified as an off-the-shelf VR tool. Because their search concluded in June 2017, the more recent evidence is missing. We decided to address these issues by conducting a well-controlled meta-analysis that focuses only on RCTs that use VR technologies for the recovery of the upper limb after stroke. We analyze the effect of VR systems specifically built for rehabilitation (ie, SVR systems) and off-the-shelf systems (ie, NSVR commercial systems) against CT according to the ICF-WHO categories. Also, we extracted 11 principles of motor learning and recovery from established literature that could act as “active ingredients” in the protocols of effective VR systems. Through a content analysis, we identified which principles are present in the included studies and compared their presence between SVR and NSVR systems. We hypothesized, first, that SVR systems might be more effective than NSVR systems as compared with CT in the recovery of upper-limb movement and, second, that this superior effect might be a result of the specific principles included in SVR systems.
This meta-analysis was conducted following the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines.28
Identification of RCTs
We define VR as a computer-based technology that provides the user with a sense of presence in a virtual environment,29 which is induced by exposing the user to computer-generated sources of sensory stimulation that satisfy their perceptual predictions and expected sensorimotor contingencies.30 The studies included aimed at training the upper extremity of stroke patients through active participation, without assistive robotic devices (eg, exoskeleton, end-effector devices) or exogenous stimulation. We compared the impact on body function and activity of 2 kinds of VR systems with CT: SVR and NSVR systems. SVR systems were developed exclusively for neurorehabilitation purposes. NSVR systems, on the other hand, are recreational and/or off-the-shelf video games (eg, Nintendo Wii, Microsoft Xbox). As CT, we considered occupational therapy and physical therapy. To identify all RCTs in these 2 categories, we performed a computerized search in the bibliographic databases MEDLINE (OVID), Cochrane Library Plus (including EMBASE), CINAHL, APA PsycNET, DARE, and PEDro for studies that were published in English from inception until August 7, 2018, the day of the conclusion of the search. The search strategy (Supplementary Table 1) included only RCTs that tested the efficacy of SVR or NSVR systems in recovering the upper limbs of stroke patients who were either in the acute (up to 21 days poststroke), subacute (between 3 weeks and 3 months poststroke), or chronic (after 3 months poststroke) stage. We combined the effects of various chronicity bands because the current literature suggests that principles of motor learning interact constantly with the biological processes of recovery,31 and therefore, no differential effect between SVR and NSVR systems resulting from chronicity should be expected. This notion has also been confirmed by the latest meta-analysis.25 In addition, splitting the identified literature into VR type, ICF-WHO category, and chronicity reduces statistical power because of the small number of studies remaining in each band. Two reviewers (BRB and MM) assessed the studies for eligibility. We excluded studies that were not carried out on humans, lacked a control group, included less than 5 participants per experimental condition, did not target upper-extremity rehabilitation, used exoskeletons as interfaces, used exogenous stimulation (such as transcranial stimulation), or did not provide information on standard clinical scales (Figure 1). Exoskeletons and exogenous stimulation protocol where excluded for the passive or active support provided in the rehabilitation process that might lead to different outcomes.