Virtual reality for stroke rehabilitation: time to rethink? Lessons from a Cochrane review

A recent Cochrane review, by Laver and colleagues, of virtual reality as a rehabilitation treatment for people who had had a stroke, included 190 trials and 7188 people . It concluded that virtual reality “may be beneficial in slightly improving upper limb function and activity”, a surprisingly weak conclusion from such an extensive review. The trials were small. Only 36 of the 190 trials had more than 50 participants, the maximum was 152, but meta‐analysis can overcome small trial numbers so what are the main sources of uncertainty?

Heterogeneity was a major factor. The review included all studies of “… human‐computer interface that allows the user to ‘interact’ with a computer‐generated environment”, a definition of virtual reality published in 2001 . So‐called virtual reality treatments underwent considerable changes between 2000 and 2023; in 2023, games such as Assassin's Creed offered an immersive experience, a marked contrast to the initial games, which bore little resemblance to reality. Many studies had other primary outcomes. Some studies focused on motor control (e.g. balance), while others targeted specific impairments, such as memory. Only studies that measured the outcome of interest were included in the meta‐analysis, which significantly reduced the number of studies included in individual meta‐analyses.

A meta‐analysis of a drug treatment aggregates data on the impact of identical interventions on the same outcome. In contrast, this systematic review collated data from studies of various interventions with different specific target effects. Many outcomes were necessarily studied in the review, as the original studies used different outcomes.

It is unlikely that a systematic review of a vast range of interventions with many specific targeted effects could detect an influence on any outcome. Meta‐analysis assumes all interventions included in an analysis have a similar direct impact on a chosen variable.

Rehabilitation is a complex process that combines several treatments with different targets . For example, many therapies focus on altering impairments or a single activity, such as walking. The goals of rehabilitation encompass complex, high‐level functions necessary for living in society, but many treatments are tailored to an individual, combined with other treatments, and have a limited target effect. This feature challenges all rehabilitation research.

A crucial component of rehabilitation is learning simple or complex behaviours, such as walking or organising a family meal. For effective learning, the patient must practice. This requires motivation, knowledge of how to perform the task, regular practice with feedback, and the ability to respond appropriately to changing demands as they arise. Rehabilitation teaches patients how to perform activities and often provides opportunities for practice with feedback. The environment offers direct feedback through the individual's successes or failures, and natural fluctuations encourage an enhanced ability to adapt to change.

How might using the ‘human‐computer interface’ enhance learning? Computer‐generated environments may help in each aspect of learning.

• Motivation: computer games are designed to be enjoyed, offering challenge and reward.

• Method: using sensors, computers can teach people how to move correctly and, using electromechanical robotic devices, they assist with movement. They can provide hints and reminders for cognitive activities.

• Practice: if the interface requires a functional activity, such as moving an arm to a set point or identifying specific objects, it can facilitate endless practice; motivation is crucial for increasing the frequency of practice.

• Feedback: the interface can give immediate visual or audible feedback. Electromechanical limb devices can guide and correct movements. An accumulated score gives more delayed feedback.

• Variability: very advanced computer‐generated environments can add unpredictable changes for the person to react to.

The educational potential of computer games was recognized in 1970 when Clark Abt coined the term, ‘serious games’, for programmes that “have an explicit and carefully thought‐out educational purpose and are not intended to be played primarily for amusement” . A systematic review of studies that evaluated the effects of serious games in rehabilitation found, surprisingly, that “custom‐made casual games that resort to the first‐person perspective, do not feature a visible player character, are played in single‐player mode, and use non‐immersive virtual reality attain the best results in terms of positive clinical outcomes” .

A 2021 systematic review of virtual reality identified 105 definitions of virtual reality, and the most cited one from 2001 (the year of the definition used in the Cochrane review) was only mentioned four times . The 2021 review's scope included all studies that used “… computer‐based programs designed to simulate real‐life objects and events”.

Many programs in the included studies in the Cochrane review were unrealistic, so Laver and colleagues grouped the studies according to their realism, a concept known as ‘immersion’. Immersion is a cognitive phenomenon characterized by the extent to which a participant devotes their attention to the virtual world, in contrast to their physical surroundings in the real world. It reflects the objective reality of the virtual world. Few studies were immersive.

Virtual reality is no longer an adequate descriptor. To improve the interpretation of any study of virtual reality, a multi‐axial description of computer‐based interventions is needed that covers the following features.

• The reality of the environment

  • Immersive: this cognitive phenomenon manifests as the amount of attention the participant devotes to the virtual world in comparison to their presence in the real world. It reflects the objective reality of the virtual world.

  • Credible, perceived reality: this is a subjective judgment by the participant on how believable the presented virtual world is.

  • Unreal, representational: this is accepted as not real, but is easily understood as representing something in the world.

• The interaction and feedback

  • Immediate feedback on performance: the patient receives feedback on their performance in real time. This might be done indirectly, such as accumulating points with success, or directly when performance is manifest within the computer‐generated environment.

  • Delayed feedback: for example, getting a score at the end of an activity.

  • No feedback.

• The targeted mechanism(s)

  • Practice: improving the performance of an activity by motivating the person to undertake the activity.

  • Learning: improving an activity by providing feedback on the person's performance of the activity.

  • Enjoyment: improving mood and overall engagement with all activities.

The Cochrane review by Laver and colleagues underscores some important learning points. As computers have become embedded in most aspects of rehabilitation and interfaces with almost any other technology, it is time to abandon dated terms such as virtual reality. We should develop a holistic approach to computer‐assisted interventions, using generic terms, such as ‘extended reality technology’, when interested in one facet of computer‐assisted rehabilitation . Moreover, we should use multi‐axial descriptions, covering, for example, the intervention's targeted direct effect (mechanism); intended higher‐level effects (outcomes); feedback mechanism; and other features, such as its reality.

Feedback on this editorial and proposals for future editorials are welcome.

End Notes

Declarations of interest

DTW has a blog post on his website related to the topic of the editorial (www.rehabilitationmatters.com), and he acted as Sign‐off Editor for the Cochrane Update, Virtual reality for stroke rehabilitation .

Provenance and peer review

This editorial was commissioned based on a proposal by the authors and was peer reviewed.