Action observation training for rehabilitation in brain injuries: a systematic review and meta-analysis

Background To systematically review and analyse the effects of Action Observation Training on adults and children with brain damage. Methods Seven electronic databases (Cochrane, EBSCO, Embase, Eric, PubMed, Scopus and Web of Science) were searched up to 16 September 2018 to select Randomized Controlled Trials focused on adults and children with brain damage that included AOT training on upper and/or lower limb carried out for at least 1 week. Identification of studies and data extraction was conducted with two reviewers working independently. Oxford Centre for Evidence-based Medicine (March2009) – Levels of Evidence and Physiotherapy Evidence Database scale were used to grade studies. The data collected from the articles were analysed using software R, version 3.4.3. Hedge’s g values were calculated and effect size estimates were pooled across studies. Separate meta-analyses were carried out for each ICF domain (i.e. body function and activity) for upper and lower limb. Results Out of the 210 records identified after removing duplicates, 22 were selected for systematic review and 19 were included in the meta-analysis. Thirteen studies included in the meta-analysis focused on upper limb rehabilitation (4 in children and 9 in adults) and 6 on lower limb rehabilitation (only studies in adults). A total of 626 patients were included in the meta-analysis. An overall statistically significant effect size was found for upper limb body function (0.44, 95% CI: [0.24, 0.64], p < 0.001) and upper limb activity domain (0.47, 95% CI: [0.30, 0.64], p < 0.001). For lower limb, only the activity domain was analysed, revealing a statistically significant overall effect size (0.56, 95% CI: [0.28, 0.84], p < 0.001). Conclusions Action Observation Training (AOT) is an innovative rehabilitation tool for individuals with brain damage, which shows promising results in improving the activity domain for upper and lower limbs, and also the body function domain for the upper limb. However, the examined studies lack uniformity and further well-designed, larger controlled trials are necessary to determine the most suitable type of AOT particularly in children. Systematic review registration CRD42019119600.


Introduction
Action observation therapy (AOT) is a novel rehabilitation strategy for both adults and children. It involves observation of meaningful actions with the intention to imitate and then performing those actions. AOT is based on neurophysiological knowledge that observation of a goal-directed action [1,2] activates the same neural substrate, called the Mirror Neuron System, as does the physical execution of the observed action.
AOT has been investigated for its potential benefits for children with cerebral palsy (CP) [3][4][5], adult stroke patients [6,7], individuals suffering from Parkinson's [8] and Alzheimer's disease [9]. The use of AOT in rehabilitation programs may have top-down effects involving higher-level networks that impact peripheral circuits, e.g. central movement planning areas, motor areas and peripheral structures [10]. To our knowledge, only few systematic reviews have explored AOT effectiveness on upper and lower limb rehabilitation. One was carried out in both neurological and orthopedic diseases [11]. Another was carried out on patients with stroke and explored AOT enhancement in motor function and upper limb motor performance [12]. Others have only explored the effectiveness of AOT on limb pain [13] and in Parkinson disease [8].
However, in previous reviews, no meta-analysis including studies on children and on lower limb was carried out. Moreover, the data were not analyzed taking account of the International Classification of Functional Disability and Health (ICF) framework. The ICF, with its multidimensional nature, provides an international framework for measuring and documenting health outcomes at the body function and structure level as well as for activities and participation.
This review addresses clinical research questions related to: i) how many studies focused on the rehabilitation of the upper or lower limb, ii) how many studies were conducted on adults and on children, iii) what type of training was conducted, where did it take place, how long did it last and did it influence the outcome, iv) what are the effects of AOT on upper and lower limb measures according to ICF domains, in adults and children with brain damage.

Design
A systematic review and meta-analysis following the guidelines of Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [14] were conducted. Identification of studies that met the review criteria, assessment of methodological quality and data extraction was undertaken by two reviewers (BB, EB), working independently. Any disagreements were resolved through consensus or, when necessary, by a third reviewer (GS).
This systematic review was registered on PROSPERO (CRD42019119600).

Inclusion and exclusion criteria (review criteria)
The criteria used to select articles were: i) participants were children or adults with brain damage; ii) investigated AOT training on upper and/or lower limb that was carried out for at least for 1 week; iii) randomized controlled trials.
The exclusion criteria were: i) articles written in languages unknown to the authors (i.e. Chinese, Persian); ii) participants with Parkinson's disease; iii) reports in the form of abstracts, reviews, theses or conference papers; iv) AOT carried out with only the observational element and not followed by action and v) grey literature.
The literature search was conducted using seven electronic databases: PubMed, EBSCO, Cochrane, Scopus, Web of Science, Embase and Eric. The search dates were from database inception to 16th September 2018. The search used the following terms: ("Brain injury" OR "cerebral injury" OR "cerebrum lesion" OR "left hemisphere injury" OR "right hemisphere" OR "brain damage" OR "brain lesion" OR "stroke" OR "cerebral palsy" OR hemipleg*) AND "action observation" AND ("training" OR "treatment" OR "trial").

Identification of relevant articles
Two reviewers (BB, EB), independently, screened the titles and abstracts of identified articles. Duplicates were removed. All articles that probably or possibly fulfilled the study criteria were taken forward for full text screening. Each reviewer, again working independently, then examined the full text of articles to assess whether they met the study criteria. All articles that met the study criteria were included in this systematic review.

Assessment of methodological quality
The methodological quality of the included studies was assessed according to the latest versions of: the Oxford Centre for Evidence-based Medicine (CEBM, March 2009) -Levels of Evidence [15] and the Physiotherapy Evidence Database scale [16]. Reviewers worked independently as described above.

Data extraction
The two reviewers independently recorded for each included study: CEBM level; PEDro score; study aim/s; diagnosis; sample size; mean age of participants; setting; duration and intensity of training; type of AOT; video perspectives; other treatments provided to the experimental group; and the type of intervention and other concurrent treatments provided for the control group. Data was dichotomized by section (upper or lower limb) because different aims and outcome measures were used. Outcome measures regarding upper and lower limb were divided into ICF domains. If an outcome measure involved more than one domain, the outcome measure was classified within the most representative domain [17].

Meta-analyses
Study outcome measures, results and findings of examined studies were analysed. The data collected from the articles were analysed using software R, version 3.4.3. Hedge's g values were calculated and, according to Cohen [18], values of effect sizes between 0.2 and 0.5 were considered "small", between 0.5 and 0.8 "medium", and > 0.8 "large".
Effect size estimates were pooled across studies to obtain an overall effect size. Some of the studies included different outcomes, that could be correlated [19]. A multivariate random-effect linear model was used to conduct a meta-analysis, where covariance matrix was explicitly provided to the model. Separate meta-analyses were carried out for each ICF domain (i.e. body function and activity) for upper and lower limb.

Identification of included studies
The database search identified 534 articles, of which 210 remained after duplicates were removed. Of these, 168 records were excluded after the titles and abstracts were screened. Of the 42 full papers that were read 20 did not meet the inclusion criteria. The following studies were excluded: i) two studies had AOT training of less than 1 week; ii) one paper included healthy participants; iii) four articles were written in Chinese and two in Persian; iv) seven articles were merely abstracts; v) one article compared two types of AOT; vi) three articles were not RCTs. Consequently, 22 studies were included in this review ( Fig. 1).

Studies focused on upper limb Sample participants
Studies on upper limb were carried out on very heterogeneous samples. Two studies included chronic stroke patients (> 6 months duration) [6,27]; two enrolled only patients with first-ever stroke, 30 days (±7) after the onset of the event with ischemia or primary haemorrhage [26,29]; one study included adults who had suffered a stroke 3 to 31 days prior to recruitment [28]; and the remaining three studies [1,21,23] enrolled subjects within 6 months of stroke. Two studies did not specify whether patients were in their subacute or chronic poststroke phase [24,25]. Moreover, regarding studies on children, two focused on children with Unilateral Cerebral Palsy (UCP) [5] while children with unilateral and bilateral CP were included in two other studies [3,20]. It is important to highlight that all children with CP had a cognitive level within normal limits for verbal functions and did not present any sensory impairments [3,5,20,22].

AOT training and control conditions
Videos showing various actions [1,23,27] or videos of daily routines [6,26,29] were used in the experimental group. Actions demonstrated in two studies [26,29] were both unimanual and bimanual. In four studies [1,6,21,26], the difficulty of the proposed actions increased incrementally during treatment. Participants performing the action on the video were healthy men or women in four studies [20,23,26,29].The type of model performing the action was not explicitly mentioned in the others. In one study [5]), separate videos were produced for patients with left or right hemiplegia.
Some of the studies specify the perspective from which actions were performed. In one study [23] three perspectives were provided simultaneously: front, side and top. In another study [1], actions were seen from "straight on, right above and right inside", whereas in yet another [27] the video was shot from the front. In three studies [5,26,29] actions were observed from a first-person perspective. Three [6] reported that actions had been recorded from different perspectives (one [6] specified that 3 perspectives had been used) but failed to mention which ones, and in another two, no mention was made at all of the type or number of perspectives [24,25]. Two studies did not use videos but life demonstration to show actions [22].
In the home-based study [22], a parent performed the action while sitting next to the child on the lessaffected side facing in the same direction, so that the child observed the hand movements from an egocentric viewpoint, whereas in the in-patient study [28] the therapist sat next to the participant on his/her affected side, demonstrating the action to be performed.
Control groups performed actions without observation [23][24][25], or they observed videos, images, or sequences of geometric symbols [6] which showed a  [26,29] and performed the same actions as the experimental group. In one study, children were asked to play computer games [5].
In all the studies where AOT observation phase was conducted using videos a significant change in at least one outcome measure was found. On the contrary, the only two studies [22,28] where, instead of the videos, patients observed a person performing an action, reported no significant functional improvements, neither in adults with stroke in an early phase (mean 18.70 days) [28] nor in children with UCP [22].

Duration of experimental and control intervention
Duration of studies carried out on adults varied. Four studies lasted 4 weeks [6,23,26,29], while most of the others were carried out over a 3-week period [3,20,22,27,28]. Two were 8 weeks [1,21] long and the longest one lasted 12 weeks [22]. In all the studies, except two [1,21], training took place 5 days a week.

Outcome measures
In  [3,5,20,22]. In adults five studies used the Fugl Meyer Assessment [1,21,23,26,29] or the modified Ashworth Scale (MAS) [1,23]. A further study used the Motricity Index (MI) [28], while a kinematic analysis was carried out in one study [25]. All the studies focused on ICF activity domain used various outcome measures. The same outcome measure was applied in a maximum of three studies. Three of the four studies on children had at least one ICF activity domain outcome measure i.e. Assisting Hand Assessment [5,20,22], ABILHAND-kids [5,22]. Three studies in adults used the Box and Block Test [23,26,29], two used the Frenchay Arm Test [6,29,38] and three used the Wolf Motor Function Test (WMFT) [6,21,24]. Three studies also used the Barthel index (BI) [1,21,23] (in two articles [23] a modified version was used). Other outcome measures can be seen in Tables 3 and 4.
Only one study [6] analyzed participation domain using the Stroke Impact Scale (SIS) (see Table 5).

Meta-analysis of studies investigating AOT for upper limb rehabilitation
Of the 14 AOT studies on upper limb, only those with clinical standardized measures were included. For this reason one article [25] was excluded.
For body function domain, we analysed nine studies, including nine outcome measures on a total sample of 360 patients (169 allocated in AOT group). According to  (Fig. 2). Twelve studies were analysed in the activity domain, with 11 outcome measures and a total of 462 patients (219 allocated in AOT group). The multivariate random-effect model returned an overall statistically significant effect size (p < 0.001) of 0.47 (95% CI: [0.30, 0.64]).

Studies focused on lower limb Sample participants
All eight studies were carried out on adult patients diagnosed with chronic stroke.

AOT training and control conditions
All the studies on lower limb showed videos to the participants. Videoclips entailed walking in different locations and on different surfaces in two studies [32,34], while an exercise of weight shift to the affected side was included in another [35]. In one study [30], participants observed Sit To Walk (STW) video tasks and imitated the actions. Action observation tasks consisted of 16 STW tasks in which difficulty and conditions were adjusted to patient functional status and level. In two studies [36,37], several stages in the video included trunk flexion, trunk rotation, sit to stand, and stepping over obstacles to enhance balance and gait ability. In another one [33], there were three stages of an active assistive exercise: the first showed knee joint extensor and dorsiflexor training, the second knee joint flexor and dorsiflexor training and the third hip joint flexor training.
Generally, the models were healthy male/female adults. Separate videos were produced for patients with left or right hemiplegia in two studies [33,37] . The perspective was specified only in two studies [33,35] and speed of sequence was reproduced in fast and slow motion in the front, back and side views in one study [35]. The action was presented at normal speed and half the normal speed in another study [32].
The type of treatment offered to control groups, when present, varied: four groups watched videos showing static landscapes [30,32,35] or nature pictures not related to walking [34]. In one study, the control group [33] performed mirror therapy and physical training of the same movements of AOT, while in another, only action observation was conducted without any physical training. In two other studies [36,37], where two control groups were present, one participated in a motor imagery program and did physical training similar to AOT group, while the other performed only physical training.
In one study [31], AOT was combined with electromyography-triggered functional electric stimulation (ETFES) in order to improve voluntary functional movement which was compared to training of subjects in a control group who underwent functional electric stimulation (FES).

Meta-analysis results on studies on lower limb rehabilitation
Of the eight included articles that focused on the lower limb, two articles [31,36] were omitted from the metaanalysis because they did not use clinical standardized measures. All of the outcome measures for the lower limb across the six studies were in the ICF activity domain. The six studies used seven different outcome measures. In the multivariate random-effect meta-analyses, the overall effect size was statistically significant (p < 0.001), estimated as 0.56 (95% CI: [0.28, 0.84]) (Fig. 3).

Main findings
Twenty-two RCTs with a total of 748 patients were included in this review. Of these 14 focused on upper limb and 8 on lower limb rehabilitation. Four studies were carried out in children, 18 in adults. The selected articles focused only on AOT as a rehabilitation tool in stroke and CP patients. There were no RCT studies using AOT training for other brain injuries such as multiple sclerosis or acquired or traumatic brain injury. AOT observation phase was mainly conducted using videos and all these studies reported a significant change in at least one outcome measure. On the contrary, the only two studies [22,28] where, instead of the videos, patients observed a person performing an action, reported no significant functional improvements, neither in adults with stroke in an early phase (mean 18.70 days) [28] nor in children with UCP [22]. Moreover, videos are both easier to standardize and allow a broader range of patients to benefit from therapy. However, a very recent pilot study on 12 children with CP [39] suggested that live AOT is more effective than video AOT. We did not include the study in the current review because there was no control group but two types of AOT were carried out and compared. Other studies with larger sample sizes and long-term follow up are necessary to identify which is the best approach.
From a qualitative analysis of data, the use of different perspectives in AOT videos did not allow us to understand whether some perspectives are better than other, nor to assess whether the type of perspective used is relevant. This could be related to the lack of standardization of the perspectives used or to other characteristics of the study. Given the variability of the perspectives used in videos, a standardization of these variables is needed to provide the most effective AOT.
One study [28] showed greater improvement in the control group than in the AOT for one outcome measure (Action Research Arm Test) which was used only in this study.  [32,37] EMG [31] MRCP [31] H-reflex [31] balance parameters [33,34] EEG [36] Weight Distribution Index [30] Limit Of Stability [30,31] ABC3 [2] FRT [37] 10MWT [32,34,35] TUG [30,34,37] figure of 8 walking test [35] dynamic gait index [30,35] Community walking test [32] mE-FAP [33] WAQ [37] FAC [37] EMG  In this review, evidence of AOT effectiveness on motor functions, compared in the majority of the studies to physical therapy, was found on both children with CP and post-stroke adults. In the metaanalysis AOT significantly improves body function and activity domains with small and medium effect size for upper limb and lower limb, respectively.
Therefore, it could be hypothesized that the observation has a crucial add on effect to the motor activity, that is the main ingredient of conventional therapy.

Comparison with early reviews
Comparing this review to the literature, there are two systematic reviews [11] [12] analysing AOT studies on patients with neurological diseases. The first review by Sarasso [11] also includes Parkinson's and orthopedic diseases, while the review by Borges [12] is carried out on patients with stroke but focuses only on upper limb. This review adds seven articles [20,22] on AOT in both children and adults to the previous ones. However, three articles included in the previous reviews were not acquired by our search strategy [40] and three studies did not match our inclusion criteria [41][42][43]: the training lasted less than a week in one articles [42], while the observation of the action was simultaneous to the practice in the other two [41,43].
The most recent review [12] included studies up to September 2017 while our review involves studies up to September 2018.
The conclusion of the previous reviews [11,12] suggested the efficacy of AOT in improving motor functions either in neurological and orthopedic diseases and of the upper limb in adults with stroke. Our findings corroborate and extend the previous ones. We were able to identify a larger number of studies in which AOT was used to rehabilitate not only the upper limb but also the lower limb of adults and children with brain injuries. The sample size of the present review, compared to the previous reviews, was the largest also including a meta-analysis for upper and lower limb. Moreover, we evaluated the effectiveness of AOT according to different ICF domains.
Comparing our results with the previous reviews, we also acknowledge the lack of dosage uniformity as highlighted in the previous reviews. Nonetheless, most studies lasted 3 to 4 weeks and sessions were about 30 min. However, even though a metanalysis comparing dose of treatment was carried out on upper limb by Borges [12],showing no significant difference, the attention span of children and adults should be considered when deciding duration and type of treatment.
It would also be useful to understand if there is a minimum threshold before an effect is produced on mirror neuron system and if a minimum duration is necessary to maintain the effect over time. Only some studies [5,20,22,26,27,29,31] have a follow-up assessment, so studies to understand the long-term effects after AOT are needed.
Regarding the differences between video and operator observation highlighted in the previous review, a further study [22] recorded no significant change in the outcome measures, in contrast with a recent article [39] which suggested that live AOT is better than video AOT. However, the articles enrolled small samples and further studies are needed. Finally, even though a recent review and multiple studies focused on Parkinson's disease, the role of AOT in Parkinson's rehabilitation is outside the aim of this review.

Limitations of this review
The samples recruited in most RCTs were small (only five studies enrolled more than 50 patients [1,21,22,26,29] and studies adopted different inclusion and exclusion criteria, resulting in very heterogeneous populations. Moreover, in the studies on children, different types of CP were included. In addition to this, the sample selection was different, and this could affect the results of the papers and, thence, the finding of our work.
A potential limitation of this study is the risk of selection bias: the papers were identified through searches of selected databases, no reference lists of relevant papers were screened, no search for grey literature was conducted and papers published in Chinese and Persian were not included. These two issues are quite relevant because unpublished papers could have reported results in contrast with positive findings on the same topic, even if some studies included in the current review [22,28] did not report significant results on the efficacy of AOT. In addition, we did not manage to translate papers not written in English language, which potentially can add information for the current review. In the future, if there will be a growing interest in conducting studies on AOT, an update of the present review could confirm or redefine the current findings.

Strength of this review
We have analysed for the first time the effectiveness of AOT training in relation to ICF model, which is the most updated and international common framework for evaluating different disabilities, planning and measuring effects of different rehabilitation approaches. Moreover, the overall grade of recommendation based on CEBM model was A (i.e. consistent level-one study) since most studies reviewed were level one for both upper limb and lower limb.

Recommendation for clinical practice
The studies on AOT are mainly carried out in the research field. However, they give insights for application in clinical practice. The AOT results suggest that the core of rehabilitation intervention should spent time in the observation of the activities before their execution. Moreover, the repetition of the motor activity should be followed by the observation of the proposed model in order that the patient can match the observation with his/her performance. The observation of a motor activity followed by practice can be easily applied by the therapists in the rehabilitation service as the he content of the exercises commonly provided in rehabilitation setting can be easily implemented in the framework of AOT focusing the rehabilitation in a more "top-down" perspective.

Recommendation for future research
Future well-designed and sufficiently powered studies on AOT in brain injuries and multiple sclerosis have to be encouraged both in adults and even more in children. Larger scale studies should select homogeneous populations in children (e.g. AOT effect on a sample of patients with UCP, rather than a larger sample that included children with both unilateral and bilateral CP) and should investigate AOT effect on lower limb rehabilitation in children.
Moreover, thanks to the type of training which relies on the content of actions to be observed and on patient motivation to carefully observe to imitate and actively replicate the actions, AOT can be easily carried out at home. Well-standardized home-based studies need to be encouraged, as these would reduce not only hospital stays, travelling and waiting time for therapy, but would also allow for a much greater number of patients to benefit from this treatment. With this in mind, a recent trial [44] studies the effectiveness of home-based upper-limb AOT in children with UCP employing the latest technologies. However, further studies comparing different settings (e.g. hospital versus home) are needed. In addition, AOT video should be standardized as far as the perspective used (first-person or other) and length; moreover, mainly for children, the attention span should be considered when deciding duration and type of video and treatment.
Finally, larger controlled trials are necessary to determine the most suitable type of AOT regarding environment, treatment, control group and outcome measures in order to promote functional improvement of upper limb and lower limb, particularly in children.

Conclusion
In conclusion, this is the first systematic review in which the effectiveness of AOT, separately both on the lower and upper limb function, is explored, also through a metaanalysis based on the ICF framework for the analysis of its efficacy. In particular, the findings are very promising, because data suggest the use of AOT for improving the activity domain for upper and lower limb, and also the body function domain for the upper limb. However, suitably powered RCTs on more homogeneous and larger samples, by means of valid and reliable paradigm and outcome measures, are required to confirm the real efficacy of AOT. A strong design comparing different lengths of AOT treatment and this novel approach with other types of rehabilitation is needed to demonstrate the specific role of AOT to replace or to be added to traditional rehabilitation.