Comparison of bilateral and unilateral upper limb training in people with stroke: A systematic review and meta-analysis

Background and objectives Bilateral upper limb training (BULT) and unilateral upper limb training (UULT) are two effective strategies for the recovery of upper limb motor function after stroke. This meta-analysis aimed to compare the improvements in motor impairment and functional performances of people with stroke after BULT and UULT. Research design and methods This systematic review and meta-analysis identified 21 randomized controlled trials (RCTs) met the eligibility criteria from CINAHL, Medline, Embase, Cochrane Library and PubMed. The outcome measures were the Fugl-Meyer Assessment of Upper Extremity (FMA-UE), Wolf Motor Function Test (WMFT), Action Research Arm Test (ARAT) and Box and Block Test (BBT), which are validated measures of upper limb function. Results Twenty-one studies involving 842 subjects with stroke were included. Compared with UULT, BULT yielded a significantly greater mean difference (MD) in the FMA-UE (MD = 2.21, 95% Confidence Interval (CI), 0.12 to 4.30, p = 0.04; I2 = 86%, p<0.001). However, a comparison of BULT and UULT yielded insignificant mean difference (MD) in terms of the time required to complete the WMFT (MD = 0.44; 95%CI, -2.22 to 3.10, p = 0.75; I2 = 55%, p = 0.06) and standard mean difference (SMD) in terms of the functional ability scores on the WMFT, ARAT and BBT (SMD = 0.25; 95%CI, -0.02 to 0.52, p = 0.07; I2 = 54%, p = 0.02). Discussion and implications Compared to UULT, BULT yielded superior improvements in the improving motor impairment of people with stroke, as measured by the FMA-UE. However, these strategies did not yield significant differences in terms of the functional performance of people with stroke, as measured by the WMFT, ARAT and BBT. More comparative studies of the effects of BULT and UULT are needed to increase the reliability of these conclusions.


Discussion and implications
Compared to UULT, BULT yielded superior improvements in the improving motor impairment of people with stroke, as measured by the FMA-UE. However, these strategies did not yield significant differences in terms of the functional performance of people with stroke, as measured by the WMFT, ARAT and BBT. More comparative studies of the effects of BULT and UULT are needed to increase the reliability of these conclusions.

Introduction
Stroke often causes upper limb motor function deficits. Accordingly, people with stroke tend to be more reliant on others during their daily lives [1]. Severe motor impairments in the upper limbs were found to persist for 6 months after stroke in a third of people with stroke [2]. More than half of the activities of daily living (e.g. dressing, feeding and cooking) rely on upper limb functions [3]. Therefore, motor impairment in the upper limb presents a significant barrier to reintegration into society [4].
Unilateral upper limb training (UULT), a common rehabilitative strategy for people with stroke, includes repetitive task-related training [5][6][7]and constraint-induced movement training (CIMT) [8][9][10][11][12][13]. During CIMT, the subjects are required to wear a constraint mitten on the unaffected upper limb and to perform intensive training with affected side for at least 6 hours per day. Compared with CIMT, task-related training is a less intensive form of goal-directed training. This form uses various types of motor tasks to help subjects to derive optimal control strategies for solving problems related to motor abilities. Wolf et al. [8]compared the effects of a 14-session CIMT program with dose-matched conventional therapy (day treatment program, outpatient visits, physiotherapy and occupational therapy) on the motor outcomes of people with sub-acute stroke. In that study, CIMT induced a significantly greater increase in the Quality of Movement in Motor Activity Log (MAL) score (p<0.01) and a significant reduction in the time needed to complete the Wolf Motor Function Test (WMFT) (p<0.01), compared to the conventional treatment program. Additionally, task-related training was also found to be superior to conventional therapy for rehabilitating upper limb function. Narayan et al. [7]found that 20 sessions of task-related upper limb training, including reaching and lifting objects with different shapes using the affected upper limb, were superior to dose-matched neurodevelopmental-based therapy as measured by the Fugl-Meyer Assessment of Upper Extremity (FMA-UE), WMFT, MAL and Action Research Arm Test (ARAT).
Bilateral upper limb training (BULT) is another stroke motor rehabilitation strategy in which the subjects are required to perform motor tasks with both upper limbs. Here, the unimpaired limb is used to increase the functional recovery of the impaired limb by facilitating coupling effects between the two limbs [14]. BULT includes bilateral functional task training [15][16][17][18][19], bilateral robotic-assisted training [20][21][22] and bilateral arm training with rhythmic cueing [23][24][25]. Several studies have indicated the superiority of BULT over various conventional therapies (including neurodevelopmental therapy [18,26], occupational therapy [19,27,28], physiotherapy [17, 21, 28] and unilateral robotic-assisted training [20]) for improving the FMA-UE, WMFT, ARAT and MAL and the ranges of motion (ROM) of the shoulder, elbow and wrist joints in people with stroke.
Several clinical trials have also demonstrated the ability of BULT to improve hemiplegic arm functions [29][30][31][32][33]. In two systematic reviews [29,32], combinations of BULT with other therapies, such as electrical stimulation and auditory rhythmic cueing, effectively increased the functional WMFT scores of patients with acute to chronic stroke immediately after completion of the intervention. However, those reviews [29, 32] included single-group pre-post studies [34][35][36][37][38][39]. Moreover, the reviews [29, 32] did not directly compare the results of BULT and UULT in people with stroke and thus, were unable to demonstrate which approach more effectively improved the performance of the paretic upper limb, based on functional scales such as the FMA-UE, WMFT and ARAT.
By contrast, two recently published meta-analyses [33,40] compare the abilities of BULT and UULT to improve the FMA-UE, WMFT, ARAT and MAL in people with stroke. Van Delden et al. [33] categorized studies according to the motor impairment level, as measured by the FMA-UE, Brunnstorm Stage, WMFT and ARAT. The results showed that UULT and BULT yielded similar improvement in the FMA-UE, WMFT, ARAT and MAL scores of people with stroke. Lee et al. [40] compared the effects of BULT with those of unilateral task-related training and CIMT during upper limb rehabilitation after stroke. Notably, CIMT was more effective than BULT in improving the WMFT and ARAT score. However, this finding should be interpreted cautiously because only three studies [16,41,42] were included in the metaanalysis.
Although CIMT can be used to train the paretic upper limb intensively, a direct comparison of the effects of CIMT and BULT on the rehabilitation of upper limb motor function in people with stroke may be inappropriate. First, CIMT requires the subjects to wear a constraint mitten on the unaffected upper limb and to perform intensive training for at least 6 hours per day [43]. Page et al. [44] found that about 68% of subjects with stroke were unable to complete the full schedule of CIMT because of the training requirement and restrictive device. According to Blanton et al. [45], only 20-25% of patients with chronic stroke benefited from CIMT because of the tight training schedule and potential risk induced by the restricted training plan. By contrast, BULT has a lower training intensity. Patients are expected to complete approximately 1 to 2 hours of training per session on 3 to 5 days per week [24,46], in contrast to the CIMT schedule of 6 hours of supervised task practice on each of 14 consecutive days [10,47,48]. Second, most CIMT studies applied stringent inclusion criteria, including at least 10 degrees of wrist extension, thumb abduction and finger extension on the affected side [49]. Compared with CIMT, BULT only requires people with stroke to maintain volitional control of the nonparetic arm, to be capable of flexing the paretic arm and shoulder and to have maintained the residual grip function of the paretic hand [18,25]. CIMT is only applied to stroke survivors with mild to moderate levels of upper limb dysfunction. Thus, the exclusion of CIMT from a comparison of the effects of BULT and UULT in people with stroke would improve the validity of the quantitative results.
Although several meta-analyses [33, 40] have compared the effects of BULT and UULT in people with stroke according to the FMA-UE, WMFT and ARAT, these meta-analyses treated CIMT as a subtype of UULT and included it in comparison with BULT. To the best of our knowledge, no studies have excluded CIMT when comparing the effects of BULT and UULT in people with stroke. This systematic review and meta-analytical review aimed to evaluate the available randomized controlled trials (RCTs) that compared the effects of BULT and UULT, but excluding CIMT on improvements in the FMA-UE, WMFT and ARAT score of people after stroke.

Study selection criteria
An exhaustive search of the literature was conducted to identify publications related to the effectiveness of BULT. The CINAHL, Medline, Embase, Cochrane Library and PubMed databases were searched systematically through April 2018 using the keywords listed in Table 1.
All identified full-text English language journal articles were screened independently by the two reviewers (PM and PK). The reference lists of the selected articles were then examined to identify additional potential articles. The inclusion criteria were applied to identify studies that (1) were randomized control trials; (2) reported quantitative behavior outcome measures; (3) had investigated the effects of interventions on upper limb function; (4) included an intervention group with bilateral movement training; (5) an intervention group with unilateral movement training or conventional occupational therapy or physiotherapy; and (6) included people with stroke. The following exclusion criteria were also applied: (1) the use of BULT in both the experimental and control groups; (2) failure to provide data on the outcome measures; (3) the studies with was a single session design; (4) systematic review or meta-analyses; and (5) inclusion of CIMT as the UULT.

Risk of bias
The methodological quality of the studies was assessed using the Cochrane Collaboration tool for assessing the risk of bias [50]. Studies that provided information clearly (i.e., randomized, subject-blinded) were rated as low risk for the corresponding items. If the study provided the information against the assessed items (i.e. non-randomized, not blinded), the study was rated as high risk for those items. If no information suitable for our judgment process was provided, the study was rated as unclear.

Data synthesis and analysis
Two reviewers (PM and PK) extracted the participants' demographic information (age, gender, post-stroke duration post stroke, type of lesion and side of hemiparesis), details of the intervention (type, intensity and duration) and outcome measures to identify the study characteristics. The third reviewer (SMN) made judgments if discrepancies occurred between the two reviewers. The International Classification of Functioning, Disability and Health (ICF), which is regarded as the international standard for evaluations of health and disability, was used to assess motor impairment and functional performance. The ICF can facilitate a more comprehensive understanding of the effectiveness of bilateral movement training during stroke rehabilitation and an optimal bilateral movement training scheme for improving upper limb function.
According to the ICF framework, the outcome measure for the meta-analysis was divided into two domains: (1) motor impairment and (2) functional performance. The effect size of each outcome was computed by calculating the mean difference (MD), standard mean difference (SMD) and 95% confidence interval (CI), as appropriate. If a study did not provide the standard deviation (SD) of the MD or SMD, this value was estimated using the following formula, with the correlation coefficient (Corr) set to 0.8 [50]: SD ¼ ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi ffi The results of the meta-analysis were then visualized using a forest plot (Review Manager 5.3; The Nordic Cochrane Centre, Cochrane Collaboration, Copenhagen, Denmark). To compare the effects of BULT and UULT during different phases of stroke, the included studies were classified as acute (mean post-stroke duration<1 month), sub-acute (mean post-stroke duration>1month to <1 year), chronic (mean post-stroke duration>1 year) or not reported.
To investigate the influence of treatment dosage on the effect size estimates, meta-regression analyses were performed using STATA 12.0 (Stata Corporation, College Station, TX, USA).

Publication bias
Egger's test is more frequently used than other tests to detect publication bias in a meta-analysis [51]. Accordingly, Egger's test [52] was used to detect the probability of publication bias in this study.

Heterogeneity test
The Higgins I 2 index was used to evaluate the heterogeneity of the studies. The I 2 boundary was set at 50%. A random effect model was used when I 2 >50%, indicating heterogeneity. A fixed effect model was used when I 2 <50%, indicating homogeneity (51).

Study identification
The search strategy yielded 828 citations on April 10, 2018. After excluding duplicated articles, 375 potentially relevant articles were subjected to further screening via a review of the abstracts. During this meta-analysis, the third reviewer made judgments on two of the screened articles [15,20] that were ultimately included. Finally, 21 full-text articles with 842 subjects fulfilled the eligibility criteria of the review. Details of the studies' identification, screening, eligibility and inclusion criteria are shown in Fig 1.

Characteristics of the intervention
Nine studies [15-18, 21, 27, 53, 54, 57] compared the effects of bilateral functional task training (e.g., folding a towel, lifting two cups and picking up two pegs bilaterally) and unilateral taskrelated training. Four studies [17, 21, 53, 54] compared these effects in people with acute stroke. Two studies [16, 18]compared the effects of bilateral functional task training and dosematched neurodevelopmental therapy, weight-bearing exercises and unilateral functional task training. Two studies [15,57]

Meta-analysis
Based on the measurement items categorized by the previous studies [59,60], this review compared the overall effects of BULT and UULT on (1) improved motor impairment by pooling  Table 4.

Fig 2. Risk of bias summary: A review of the authors' judgments about the risk of bias of each item in each included study.
https://doi.org/10.1371/journal.pone.0216357.g002   The meta regression indicated that the number of training sessions (p = 0.947), total duration of training (p = 0.217) and duration of training per session (p = 0.316) had no significant impact on the effect size of FMA-UE.

Publication bias
According to the results of Egger's test (Fig 3), no significant publication bias was observed in the meta-analysis of FMA-UE (p = 0.774) or of WMFT, ARAT or BBT (p = 0.950).

Functional performance outcomes
In this study, improvements in functional performance were assessed using the time component of WMFT and the WMFT, ARAT and BBT scores.

Assessment tool Description
Motor Impairment FMA-UE FMA-UE is a stroke-specific assessment tool to measure the upper limb motor impairment [61], which included shoulder-arm, wrist, hand and coordination. There were 33 items, which scoring on an ordinal scale from 0 to 2. The total score was ranged from 0 to 66.

Functional Performance
WMFT WMFT is used to assess the upper limb motor function in people with stroke [62], which included 2 strength-based tasks and 15 functional-based tasks. The 15 function-based tasks were assessed by the time taken to complete each task and the quality rating of the use of the affected hand in attempting each task. The functional task was graded from 0 to 5. The total score was ranged from 0 to 75.
ARAT ARAT is used to assess the upper limb function of grasping, gripping, pinching and gross arm movement [63]. This ordinal scale consists of 19 items. The quality of the performance on each item was rated from 0 to 3 points. The total score was ranged from 0 to 57.

BBT
BBT is a measure of gross manual dexterity for handicapped people [64]. The test will count the number of wooden blocks that can be transported form one compartment of a box to another within 1 min. The more block being transported indicated a better functional performance.  Five studies [16, 23-25, 56] were used to evaluate the effect on improvement in the time required for the WMFT (Fig 6). A comparison of BULT and UULT revealed no significant

Discussion
This is the first systematic review and meta-analysis to compare the effects of BULT and UULT on motor impairment and functional performance in people with stroke after excluding CIMT. The exclusion of CIMT may provide a more realistic overview of the effects of actual rehabilitation efforts. The meta-analysis examined the pooled results of 21 RCTs including 842 subjects with stroke. According to this analysis, BULT yielded significantly greater improvement in the FMA-UE, compared to UULT. However, no significant differences in the change of functional performance as indicated by the WMFT, ARAT and BBT were observed between BULT and UULT.

Motor impairment
Our results are partially consistent with the findings of a review by Coupar et al. [30]. In that review [30] of two sets of four studies, the authors found that BULT was more effective than the usual care in terms of improving the FMA-UE scores in people with stroke. However, Coupar and colleagues [30] found no differential effects between BULT and UULT in terms of improving the FMA-UE score [17,26,65,66]. Our inclusion and exclusion criteria enabled us to include 4 studies [17, 23, 26, 54]from the review by Coupar and 12 other RCTs in our metaanalysis. Thus, our review may have elicited more robust estimations of the overall effects of BULT and UULT on improvments in the FMA-UE.
Two reviews [33, 40] found that BULT and UULT yielded similar reductions in motor impairment in people with stroke, as indicated by the FMA-UE score. However, these results should be interpreted cautiously because van Delden [25,42,65,66] and Lee [25,42,65,67] each reviewed four studies to estimate the effects of BULT and UULT on the FMA-UE. In contrast, our meta-analysis on FMA-UE was based on 16 RCTs, and included a larger sample size. Our review would therefore have a stronger power to detect a difference between BULT and UULT in terms of an improved FMA-UE score. The reviews by Lee [40] and van Delden [33] also included two studies [42,66] that compared the effects of BULT and CIMT on improved FMA-UE scores. In our study, we only compared BULT with task-related training. This Our review demonstrated a significantly greater improvement in the FMA-UE scores of people with stroke after BULT, compared with UULT. Moreover, we found no significant association of the training dosage with the improvements in FMA-UE. Therefore, the different levels of improvement between BULT and UULT was mainly attributable to the types of intervention (BULT/UULT). The excitability typically decreases in the lesional cerebral hemisphere and increases in the contralesional hemisphere [68][69][70]. TMS studies [67,[71][72][73] have indicated that this restoration of interhemispheric imbalance positively correlates with motor recovery after stroke. Accordingly, the different levels of improvement in the FMA-UE scores after BULT and UULT may be related to the use of different mechanisms to facilitate the reorganization in the lesional hemisphere. UULT is based on the principle of activation in the lesional hemisphere via assisted or resisted unilateral training of the paretic limb [8,74,75]. Compared with UULT, BULT activates similar neural networks in the bilateral hemispheres during the simultaneous activation of homologous muscle groups [14,[76][77][78]. Studies have shown that BULT can activate the distributed corticospinal pathway bilaterally via ipsilateral corticospinal fibers [79][80][81], contralateral corticospinal fibers [82][83][84] and the corpus callosum [85,86]. Compared with UULT, BULT may evoke greater activation of the lesional hemisphere by recruiting more neural pathways [14]. In the BULT group, increased activation in the lesional hemisphere might have led to greater improvements in motor impairment, as indicated by the FMA-UE scores.

Functional performance
Consistent with the findings of previous reviews [33, 40], our meta-analysis indicated that BULT was not superior to UULT in terms of improving the functional performance of people with stroke, as measured by the WMFT, ARAT and BBT. Although we observed a significantly greater improvement in the FMA-UE with BULT than with UULT, we observed no significant differences in functional performance between the training strategies. Buchner [87] reported a non-linear relationship between leg strength and gait speed in people with stroke. This curve revealed a positive slope that gradually decreased to zero; in other words, the curve eventually reached a plateau, as the leg strength increased. A muscle strength threshold is required to perform each type of activity. However, increased strength does not result in an improved gait speed until a certain threshold is reached [88]. Similarly, the improvement in the FMA-UE score and the functional performance may also exhibit a non-linear relationship. Although previous studies [62,[89][90][91] reported a moderate to good correlation between the FMA-UE and functional performance-related scales (e.g., WMFT, ARAT and BBT), the non-linear relationship observed between these fators in our meta-analysis may reflect an inability of the stroke survivors to achieve the motor control threshold needed to perform the functional tasks. Although the FMA-UE improved to a significantly greater level in the BULT group when compared with the UULT group, this significant improvement may have been insufficient to yield a significant improvement in functional performance.
Our meta-analysis revealed that bilateral functional task training tended to yield a larger SMD for improving functional performance, compared to bilateral robotic-assisted training and bilateral arm training with auditory cueing. However, this result should be interpreted cautiously because although our meta-analysis included 9 of studies [15-18, 21, 27, 53, 54, 57] on bilateral functional task training (the largest proportion), it also included only 7 studies [19,20,22,24,26,46,55] of bilateral robotic-assisted training and 5 studies [23,25,28,56,58] of bilateral arm training with auditory cueing. Therefore, the inclusion of more studies of bilateral robotic-assisted training and bilateral arm training with auditory cueing would support more robust conclusions regarding the effects of the different types of BULT. Moreover, we used WMFT, ARAT and BBT to estimate improvements in functional performance. The items included in these evaluation tools, such as gripping objects, folding a towel and lifting objects, are similar to the components of bilateral functional task training. Accordingly, there may be a stronger learning effect in the studies that investigated bilateral functional task training, compared to those that investigated bilateral robotic-assisted training and bilateral arm training with auditory cueing.
This systematic review had several limitations. First, we only examined 21 studies with 842 subjects. This sample size may not have been sufficiently large to detect significant differences in the functional performance outcomes. Second, the results of this review may not be generalizable to all stroke survivors. In addition to the 13 studies [15, 16, 18, 19, 22-25, 28, 55-58] of people with chronic stroke included in this study, only 4 studies [17, 21, 53, 54] of people with acute stroke and 2 studies [26, 46] of people with sub-acute stroke investigated the effects of BULT. Therefore, the true effect of BULT may be underestimated because of the small number of included non-chronic stroke studies. The inclusion of more studies with subjects in different phases of stroke would increase the generalizability of our conclusions. Third, most studies of Asian populations reported a significant improvement after BULT, compared with UULT. By contrast, most studies conducted in western countries reported insignificant differences between BULT and UULT. However, the reason underlying this discrepancy remains unclear. In future reviews, clear methodological information and a larger sample size may help to explain this phenomenon. Fourth, we only evaluated the immediate effects of the outcome measures. Our meta-analysis did not calculate the carry-over effects of BULT and UULT in terms of improving the FMA-UE, WMFT, ARAT and BBT score, as only 29% of the studies [17, 26, 28, 46, 53, 57] included in our review provided data from the follow-up assessments. Thus, our findings may not provide sufficient power to estimate the carry-over effects of BULT and UULT in people with stroke. These carry-over effects should be explored further. Fifth, the studies included in this review did not classify the severity of the motor impairment experienced by people with stroke. Therefore, we did not have sufficient information to analyze the effects of BULT and UULT with respect to the severity of the motor impairment.

Conclusions
Both BULT and UULT can help to improve motor impairment and functional performance after stroke. Notably, BULT was superior to UULT in terms of improving motor impairment after stroke, as measured by the FMA-UE. However, BULT and UULT yielded similar effects on functional performance in people with stroke, as measured by the WMFT, ARAT and BBT.