Is a mirror necessary for mirror therapy?

Since it was first described as a method to treat phantom limb pain, mirror therapy (MT) has been applied in many areas of rehabilitation. Its seminal application with the aim of restoring motor function following stroke is considered to be that of Altschuler et al. (1999). The key features of MT in this context are that the stroke survivor performs movements of the less-impaired limb while looking in a mirror, that is positioned such that the reflected image gives rise to the impression that the more-impaired limb is also moving. In most discussions of the means through which MT exerts therapeutic effects, emphasis is placed upon an instrumental role of the visual feedback provided by the mirror. Although it is known that such feedback increases the excitability of corticospinal projections to the quiescent limb, there is scant evidence that this in itself gives rise to subsequent increases in functional movement capacity (Carson et al., 2016).

An alternative conception of mirror therapy: Necessarily, in all applications of MT, movements are performed repeatedly (and often intensively) by the less-impaired limb. That this aspect of the MT protocol is accorded minimal attention is perhaps because intentional engagement of the less-impaired limb seems to run counter to the tenets of constraint-induced therapies. It has been known since the nineteenth century that functional capabilities can be enhanced through training performed by the opposite limb alone. There have been many further demonstrations of this “cross education” (CE) phenomenon (Calvert and Carson, 2022), in a wide variety of upper and lower limb tasks. In recognition of the therapeutic potential of this effect, it has been proposed that it may serve as a basis for the restoration of motor function in stroke survivors. Several clinical trials have been undertaken recently with this possibility in mind.

Given that interventions based on CE share with MT the active engagement of the less-impaired limb, one might ask: “what further benefit arises through the use of a mirror”? While the answer to this question has the potential to shed light on the neural mechanisms that mediate the therapeutic response to MT, it also has practical significance. Should it prove to be the case that the functional gains realized by CE are comparable to those of MT, interventions can be simplified dramatically. In short, there would be no requirement for a mirror.

In principle, the answer to this question can be determined empirically by comparing the outcomes of trials in which MT was administered, to those in which therapeutic training was performed using only the less-impaired limb. Ideally, meta-analyses would be used to summarise the impact of interventions based on CE, and to draw comparisons with the effectiveness of MT. As will become apparent, however, meta-analyses sometimes only tell part of the story.

A meta-analytic approach to the question: As spontaneous neurobiological processes (as opposed to interventions) exert the most significant causal influence on the recovery of motor function in the initial weeks following stroke, we focussed on the chronic (> 6 months) period, during which additional gains can still be achieved. Two comprehensive search strategies were implemented using Scopus, PubMed, and EMBASE. The first combined two major themes: stroke and cross education and their synonyms (e.g., “stroke” OR “hemiparesis” OR “post stroke” AND “cross-education” OR “cross transfer” OR “inter limb transfer” OR “contralateral strength gain” OR “unilateral strength gain” OR “transfer of motor skill” OR “bilateral transfer” OR “unimpaired arm”). The second combined the themes: stroke and mirror therapy (e.g., “mirror therapy” OR “mirror visual feedback” OR “mirror box”). To ensure saturation, the reference lists of studies that met the eligibility criteria and the most recent relevant Cochrane review (Thieme et al., 2018) were also hand searched. The two authors independently screened the studies identified. Any disagreements were resolved by discussion. If it was not possible to establish eligibility using the published work, an e-mail request for further information was sent to the corresponding author. We intended initially to exclude studies in which “conventional therapy” (CT) was delivered in conjunction with MT. As only a single case study remained after it was applied, this criterion was removed. Significantly, therefore, every intervention designated as MT (the case study was ineligible) also included CT.

Effect sizes (ESs) for each functional outcome were calculated independently by the two authors. Our preferred approach was to calculate ESs for repeated measures. This permits the analysis of the change by tracking the same individual over time (i.e., from pre- to post-intervention). With respect to the CE studies, the data required for this approach were either presented in the paper or provided by the corresponding author in response to an e-mail request. Unfortunately, this could not be said of studies in which MT was applied. None of the corresponding authors replied to requests for the relevant information. The ES was therefore defined as the difference between the respective pre- and post-intervention mean values divided by the standard deviation of the measurements obtained prior to the intervention. It is usually assumed that this “standardiser” (i.e., the pre-intervention standard deviation) will be consistent across studies. Hedges’ g was used as the ES statistic, as it outperforms Cohen’s d when sample sizes are below 20. The meta-analyses were performed using Robust Variance Estimates.

An answer to the question: It was apparent that when the upper limb was engaged, training of the less-impaired arm alone improved the performance of the more-impaired limb in all but a few individuals (Figure 1A–C). Nonetheless, it does appear initially, that the extent of this “cross education” effect (Figure 1D) is smaller than the magnitude of the benefit attributable to MT (Figure 2A). There are however several reasons to hesitate before making this inference.

Figure 1.

Cross education effects in people living with stroke.

In panels A–C, each line represents the pre-intervention to post-intervention difference for one individual ((A) Urbin et al., 2015; (B) De Luca et al., 2017; (C) Sun et al., 2018). For ease of interpretation, and to provide a clearer picture of the variance, all individuals whose scores are depicted in the respective panels have the same mean score (pre- and post-intervention combined). Lines passing through grey quadrants correspond to individuals who exhibited a positive change in motor function. (D) Forest plots of the meta-analysis corresponding to the influence of cross education on motor function. The analyses were conducted using Robust Variance Estimation. 10MWT: 10 m walk test; 95% CI: 95% confidence interval; ARAT: action research arm test; AROM: active range of motion; BBS: Berg balance scale; df: degrees of freedom; FAC: functional ambulation category; FAX: functional ambulation category; FMA-UE: Fugl-Meyer assessment – upper extremity subsection; modified FM: Fugl-Meyer assessment (excluding balance and upper limb motor components); TUG: timed up and go; WMFT: Wolf motor function test. All studies scored at least 7/9 on the Newcastle Ottawa Scale (NOS). The PMID (PubMed ID) is given for each study included in the meta-analysis. Unpublished data.

Figure 2.

Forest plots of the meta-analysis corresponding to the influence of mirror therapy on motor function (A) and the influence of conventional therapy on motor function (B).

The analyses were conducted using Robust Variance Estimation. 10MWT: 10 m walk test; 95% CI: 95% confidence interval; BBS: Berg balance scale; BBT: box and block tests; df: degrees of freedom; FIM: functional independent measure; FMA-gross: Fugl-Meyer assessment for gross manual skills; FMA-LE: Fugl-Meyer assessment – lower extremity subsection; FMA-UE: Fugl-Meyer assessment – upper extremity subsection; mEFAP: modified Emory Functional Ambulation Profile; MFT: manual function test; RVGA: Rivermead visual gait assessment; WMFT: Wolf motor function test. All studies scored at least 7/9 on the Newcastle Ottawa Scale. If a PMID (PubMed ID) was not available for a study included in the meta-analyses, the Digital Object Identifier (DOI) is given. Unpublished data.

The most obvious consideration is that in every instance, participants in receipt of MT were also given the advantage of CT. It is possible to gauge the contribution of CT to the magnitude of the effect attributed to MT, by examining those studies in which a separate CT only group was included (Figure 2B). In the six studies so identified, the ES reported for MT (i.e., combined with CT) was 0.64, which is comparable to the 0.60 value obtained for the full set of 11 MT studies. The effect size reported for CT alone was 0.39. It is thus readily apparent that the sum of the ES (0.20) obtained for the CE studies (which did not include additional therapy) and the ES derived for CT, is effectively the same (0.59) as the ES estimated for the (11) MT studies (in which additional therapy was given). Only two CE studies in which there was additional therapy were identified. An ES of 0.53 was calculated for the upper limb (Choi et al., 2019). The range of values estimated for the lower limb was 0.22–0.36 (Park et al., 2021).

Conspicuously, the standard deviations (i.e., the dominator used in the ES calculations) of the pre-intervention measurements were markedly lower for the MT studies, than for CE studies that comprised only movement of the less-impaired limb (and lower than for CT only cohorts). This difference can be quantified by calculating (sample size adjusted) coefficients of variation (CVs). On average, the adjusted CV of the pre-intervention measurements in CE studies (54.2) was approximately 1.6 times that of the measurements recorded in MT studies (34.5). To obtain the same ESs, the magnitude of the (pre- to post-intervention) change in the outcome measure (i.e., the numerator in the effect size calculation) for the CE studies would therefore have to be 1.6 times those reported for MT. If the ES for the CE studies is scaled in accordance with the CV calculated for the MT studies, it increases to 0.31. The sum (0.81) of this value, and the CV scaled ES calculated for CT alone (0.50), comfortably surpasses the ES obtained for MT plus CT calculated either for all MT studies (0.60) or for those that also included a CT only control group (CV scaled to 0.62).

Conclusions: There is presently insufficient evidence to support the assumption that a mirror is necessary to realize the benefits derived through what is known as mirror therapy. Our analyses indicate that the therapeutic effects of unilateral MT for chronic stroke survivors are not differentiated clearly from gains that are realized via the phenomenon of cross education and which may be achieved simply through the active engagement of the less impaired limb.

As far as we are aware, the impact of MT only (i.e., without additional CT) and of training of the less-impaired limb only (i.e., without CT or the use of a mirror) have been compared directly on only one occasion. In an unusual feature of this study, the participants undertook both upper and lower limb training (K. Monaghan, personal communication, 13^th March 2021). There were no reliable differences in the impact of these forms of therapy when either the upper (Ehrensberger et al., 2019; Simpson et al., 2019) or lower (Simpson et al., 2019) limb was assessed. This is consistent with a failure to obtain reliable additional effects attributable to the presence of a mirror in individuals without brain injury (Chen et al., 2019).

There is a further striking difference between the studies in which MT was applied (with CT) and those that involved only training of the less-impaired limb. In the MT studies, the mean total duration of therapy was approximately 1800 minutes per person, of which ~964 minutes consisted of therapy that directly engaged the more-impaired limb. In the CE studies, the mean total duration of therapy was 478 minutes per person. While it is not possible to gauge directly the therapeutic impact of such differences, it seems probable that a comparable training dose would elevate the magnitude of the benefits derived through training of the less-impaired limb.

Further consideration should therefore be given to the exploitation of CE as an element of therapeutic interventions to restore functional movement capacity in stroke survivors. There is a particular requirement to conduct studies in which the dose of CE therapy is at least comparable to that which has been employed in the evaluation of MT (~1800 minutes per person), and ideally very much greater. There also exists the opportunity to determine whether the therapeutic potential of CE is put to best use by interleaving focused and high-intensity, but intermittent, training of the less-impaired limb with conventional therapy. Indeed, this approach would appear to offer specific benefits when seeking to increase the overall dose of therapy, since peripheral fatigue of the more-impaired limb will not be the limiting factor.

Given the small scale of the studies considered in this analysis, it was not possible to account for variables that may mediate therapeutic impacts, such as age, pharmacological interventions, or the patterns of comorbidity that arise from varying constellations of brain damage. For a therapeutic intervention to be used optimally in clinical practice, mechanistic knowledge (from functional analyses of brain damage) is at least as important as evidence of an effect. Most therapies used to restore motor function following stroke have several constituent elements. Mirror therapy is a case in point. It was first in vogue as a treatment for phantom limb pain and then adopted to treat other forms of unilateral dysfunction. It should not simply be assumed that elements of this therapy that are instrumental in the context of pain relief are those, which are most efficacious in restoring motor function following stroke, nor that the most conspicuous element (the presence of the mirror) is necessarily the most effective. Although the provision of mirrored visual feedback gives rise to patterns of brain activity that are distinct from those observed in its absence (Deconinck et al., 2015), there remain few clear indications that this differentiated activity is causally related to neural adaptations that provide the basis for improvements in functional movement capacity. Indeed, both the mechanistic knowledge that is available currently and the present analysis suggest that the mirror may be superfluous.