|Home | About | Journals | Submit | Contact Us | Français|
The purpose was to conduct a structured review and meta-analysis to determine the cumulative effect of bilateral arm training on motor capabilities post stroke. Forty-eight stroke studies were selected from three databases with 25 comparisons qualifying for inclusion in our meta-analysis. We identified and coded four types of bilateral arm interventions with 366 stroke patients. A random effects model using the standardized mean difference technique determined a large and significant effect size (0.734; SE = 0.125), high fail-safe N (532), and medium variability in the studies (I2 = 63%). Moderator variable analysis on the type of bilateral training revealed two large and significant effects: (a) BATRAC (0.842; SE = 0.155) and (b) coupled bilateral and EMG-triggered neuromuscular stimulation (1.142; SE = 0.176). These novel findings provide strong evidence supporting bilateral arm training with the caveat that two coupled protocols, rhythmic alternating movements and active stimulation, are most effective.
In a 1977 Psychological Bulletin article, D. J. Glencross advocated that researchers consider integrating central and peripheral processes in the control of skilled movements (Glencross, 1977). Through the years, the central versus peripheral debate subsided; however, integrating input from both sources continues. Moreover, the exact nature of control in skilled movements still drives many research agendas. In fact, many stroke motor recovery interventions integrate input from central and peripheral sources. Re-acquiring upper extremity movements necessary for activities of daily living such as buttoning a shirt or blouse, zipping a jacket, pouring a drink, and buttering bread or toast are essential for making progress toward motor recovery.
To perform any of these four everyday tasks requires coordinating movements on two arms and hands. Thus, a leading question for stroke patients concerns bilateral arm practice: Would bilateral arm training help alleviate some motor dysfunctions and improve motor capabilities? We know that bilateral movements takes advantage of the inherent dependencies between arms; spatial and temporal dependencies (Carson, 2005; Cincotta & Ziemann, 2008; Hallett, 2001a; Hummel, et al., 2005; Lacroix, et al., 2004; Rossini, Calautti, Pauri, & Baron2, 2003). Further, symmetrical bilateral movements are known to activate similar neural distributed networks in both hemispheres. Specific activated areas include the supplementary motor area, sensorimotor cortex, cingulate motor cortex, lateral premotor cortex, superior parietal cortex, and cerebellum (Debaere, Wenderoth, Sunaert, Van Hecke, & Swinnen, 2004; Goldberg, 1985; Jancke, et al., 2000; Nachev, Kennard, & Husain, 2008; Swinnen & Wenderoth, 2004).
In spite of the inherent neural interaction patterns in the two hemispheres when both arms simultaneously move in homologous actions, consistent effective bilateral movement training findings are lacking. A comprehensive review on stroke and bilateral arm training identified contradictory findings (Carson & Swinnen, 2002; Cauraugh & Summers, 2005). Moreover, recent individual stroke rehabilitation and bilateral arm treatment studies found support (Cauraugh, Coombes, Lodha, Naik, & Summers, 2009; Cauraugh, Kim, & Summers, 2008) and failed to find support on the efficacy of bilateral training (Tijs & Matyas, 2006). Further complicating the issue is an initial meta-analysis on stroke rehabilitation and bilateral movements that reported a relatively large effect size (Stewart, Cauraugh, & Summers, 2006). However, perhaps spurious findings were found given the minimal number of studies analyzed (11), and the failure to report (a) a forest plot of the effects, (b) a funnel plot involved in publication bias, or (c) the heterogeneity of individual effect sizes (I2). These conflicting findings warrant a structured review and meta-analysis that includes new statistical techniques to determine the comprehensive effect of motor capabilities as a function of bilateral movement training. Thus, we will attempt to answer an enduring stroke rehabilitation question concerning progress toward recovery: Do bilateral movement training protocols improve motor capabilities in the upper extremities of stroke survivors?
This structured review and meta-analysis focused on studies that investigated contributions of bilateral arm training toward improving upper extremity movements post intervention. Granted, a few studies reported direct comparisons between bilateral and unilateral training, although a majority of the experiments were interested in establishing the efficacy of specific bilateral arm movement protocols versus control groups (i.e., with or without standard care). Thus, our intention was to determine the cumulative effect of bilateral arm movement training regardless of the comparison groups. Even though a considerable amount of evidence comes from unilateral training studies that followed constraint-induced movement therapy (CIMT) guidelines (e.g., EXCITE trial;(Wolf, et al., 2006; Wolf, et al., 2008), we were not concerned with directly comparing forced-use and bilateral arm training.
An exhaustive search of the literature was conducted using three databases: (a) ISI web of Knowledge, (b) PubMed Central, and (c) Cochrane Collaboration of systematic reviews. Ten primary key words/phrases guided our search: stroke, bilateral arm training, hemiplegia, hemiparesis, motor recovery/control/function, upper-extremity/limb, neurorehabilitation, bimanual coordination, coupling, and recovery protocols. References from selected studies were carefully inspected to identify studies that were not retrieved in one of our database searches. The systematic searches of the databases were conducted by two authors (NL & SN), and they identified 48 potential research studies (Cauraugh, Coombes, Lodha, Naik, & Summers, 2009; Cauraugh & Kim, 2002; Cauraugh & Kim, 2003a; Cauraugh & Kim, 2003b; Cauraugh & Kim, 2003c; Cauraugh, Kim, & Duley, 2005; Cauraugh, Kim, & Summers, 2008; Chan, Tong, & Chung, 2008; Chang, Tung, Wu, Huang, & Su, 2007; Chang, Tung, Wu, & Su, 2006; Coupar, Van Wijck, Morris, Pollock, & Langhorne, 2007; Cunningham, Stoykov, & Walter, 2002; Desrosiers, Bourbonnais, Corriveau, Gosselin, & Bravo, 2005; Garry, van Steenis, & Summers, 2005; Han, Arbib, & Schweighofer, 2008; Harris-Love, McCombe Waller, & Whitall, 2005; Hesse, Schmidt, & Werner, 2006; Hesse, Schmidt, Werner, & Bardeleben, 2003; Hesse, et al., 2007; Hesse, Schulte-Tigges, Konrad, Bardeleben, & Werner, 2003; Hesse, et al., 2005; Higgins, et al., 2006; Lewis & Byblow, 2004; Lewis & Perreault, 2007; Lin, Chang, Wu, & Chen, 2008; Luft, et al., 2004; Lum, et al., 2006; McCombe Waller, Harris-Love, Liu, & Whitall, 2006; McCombe Waller, Liu, & Whitall, 2008; McCombe Waller & Whitall, 2004; McCombe Waller & Whitall, 2005; McCombe Waller & Whitall, 2008; Morris, et al., 2008; Mudie & Matyas, 1996; Mudie & Matyas, 2000; Mudie & Matyas, 2001; Page, Levine, Teepen, & Hartman, 2008; Platz, Bock, & Prass, 2001; Rice & Newell, 2004; Richards, Senesac, Davis, Woodbury, & Nadeau, 2008; Rose & Winstein, 2005; Stewart, Cauraugh, & Summers, 2006; Stinear, Barber, Coxon, Fleming, & Byblow, 2008; Stinear & Byblow, 2004; Summers, et al., 2007; Tijs & Matyas, 2006; Ustinova, Fung, & Levin, 2006; Whitall, McCombe Waller, Silver, & Macko, 2000).
Keeping in mind that our overall research question concerned post stroke progress toward motor recovery as a function of bilateral arm training, specific inclusion/exclusion criteria for our meta-analysis were observed. In a broad inclusion approach, we cast a wide net across all three stages of stroke recovery for intervention studies that used bilateral arm movements as a training treatment. Typical time frames post stroke define the three stages of recovery: (a) acute, 0 – 1 month post stroke; (b) sub-acute, 1 – 6 months post stroke; and (c) chronic, greater than 6 months post stroke. Twenty-five studies (see Table 1) satisfied inclusion/exclusion criteria as unanimously determined by all authors. Two authors (NL & SN) independently extracted data and coded all studies for meta-analytic techniques. Our coding system recorded: (a) design with groups/subcategories, (b) sample, (c) outcome measures, (d) type of bilateral arm intervention, (e) experimental design, and (f) quality of research. A third author (JC) confirmed the data extraction, and all authors were involved in interpreting the meta-analytic results.
Twenty-four studies were excluded for one of the following four reasons: (a) case study (Hesse, et al., 2007), (b) review or meta-analysis articles (Coupar, Van Wijck, Morris, Pollock, & Langhorne, 2007; Hesse, Schmidt, & Werner, 2006; Hesse, Schmidt, Werner, & Bardeleben, 2003; McCombe Waller & Whitall, 2008; Stewart, Cauraugh, & Summers, 2006), (c) data extraction problems (Lewis & Byblow, 2004; Mudie & Matyas, 1996; Mudie & Matyas, 2000; Tijs & Matyas, 2006), and (d) missing treatment condition or lack of relevant data (Cauraugh & Kim, 2003c; Chang, Tung, Wu, & Su, 2006; Cunningham, Stoykov, & Walter, 2002; Garry, van Steenis, & Summers, 2005; Han, Arbib, & Schweighofer, 2008; Harris-Love, McCombe Waller, & Whitall, 2005; Lewis & Perreault, 2007; McCombe Waller, Harris-Love, Liu, & Whitall, 2006; Mudie & Matyas, 2001; Page, Levine, Teepen, & Hartman, 2008; Rice & Newell, 2004; Rose & Winstein, 2005; Stinear, Barber, Coxon, Fleming, & Byblow, 2008; Ustinova, Fung, & Levin, 2006).
Further, one study compared treatment and control group effects in two distinct sets of stroke patients: left versus right hemisphere location of the cerebrovascular accident (McCombe Waller & Whitall, 2005). Consequently, we considered each separate comparison as an individual study, and coded the treatment group (bilateral training) and control group (no bilateral training) comparisons in our comprehensive meta-analysis. Thus, our number of coded comparisons equaled 25. Further, control groups (no bilateral arm practice) followed diverse rehabilitation protocols including unilateral training, neuro-developmental therapy, functional movements, dose-matched therapeutic exercises, and placebo electrical stimulation.
Given that a majority of the studies reported more than one motor outcome measure, we extensively discussed establishing continuity across studies by selecting reliable and valid stroke motor outcome data. Consistent with meta-analytic recommendations, we selected only one standard stroke motor outcome measure per study to avoid biasing our findings (Rosenthal, 1995). The standard stroke assessments included: (a) upper extremity section of Fugl-Meyer Assessment, (b) Box and Block manual dexterity test, (c) Modified Motor Assessment Scale, (d) Action Research Arm Test, (e) Modified Ashworth Scale, and (f) Functional Independence Measure. In addition, if data from one of the standard assessment techniques were unavailable, then we coded kinematic and electromyography (EMG) measures such as reaction time, movement time, movement unit, and muscle activation times. Consistent with conventional meta-analytic techniques, analyzing studies with different purposes and treatment goals in synthesizing a set of literature is acceptable under certain conditions (Borenstein, Hedges, Higgins, & Rothstein, 2009; Higgins & Green, 2006). The conditions involve similarities in subjects tested, outcome measures reported, operational definitions, experimental design, and conducting technical tests beyond the overall effect size.
Rosenthal recommended two essential components for conducting meta-analyses: data synthesis and data analysis (Rosenthal, 1995). Data synthesis involves calculating an individual effect size for each study included in the meta-analysis, and data analysis involves computing an overall effect size of all studies collectively by calculating a standardized mean difference. Standardized mean difference is a robust method of determining effect sizes by computing the difference between means of two groups (e.g., treatment and control) normalized by dividing by a pooled standard deviation. All meta-analysis techniques were conducted using the Comprehensive Meta-Analysis (Biostat. Inc. Version 2) software program.
An additional meta-analytic technique included computing I2 to determine heterogeneity of the 25 individual effect sizes in our bilateral movement training data. This technique measures heterogeneity as evidenced across the outcome measures of the entire group of studies beyond statistical chance alone. I2 calculates a percentage such that lower values represent smaller variability in outcome measures of selected studies. Representative values define three heterogeneity levels: low = 25%; moderate = 50%; high = 75%.
Publication bias arises when the probability of publication of a study rises as the effect size of its findings increases. Funnel plots are used to determine the symmetry of publication bias while graphing the effect size of individual studies against the standard error associated with the study. Ideally, plotting the studies shows symmetry across studies of different size and precision with smaller and larger studies scattered uniformly at the base and apex.
Consistent with recommendations to determine stability in the meta-analytic results, we computed the classic fail-safe N analysis (Rosenthal & DiMatteo, 2001). This technique calculates the number of studies with non-significant effects required to nullify the overall effect determined in the current analysis. Larger fail-safe N values increase confidence in the overall effect and validate the stability of the current findings.
In the addition to the quantitative analysis of studies, we carried out a quality assessment of the studies. Recommendations on quality assessment include three criteria: (a) treatment randomization, (b) blinding (i.e., treatment group assignment not known by evaluators), and (c) number of drop outs or withdrawals (Higgins & Green, 2006). Table 3 displays the quality assessment values for the studies in our bilateral arm training meta-analysis.
Twenty-five bilateral arm movement training studies analyzed in a random effects model revealed a large significant cumulative effect = 0.734 (SE = 0.125; p < .0001) with a 95% confidence interval limits of 0.489 to 0.979. Individual weighted effect sizes ranged from –2.148 to 1.851. A negative effect size indicates reduced performance after bilateral training while a positive value is indicative of improvements in motor performance post bilateral intervention. Individual weighted effect sizes are shown in Table 4.
An additional meta-analytic technique is visually displaying the amount of variation between the results of the studies and the cumulative effect size for all studies on a graph. This type of graph is called a forest plot. Figure 1 shows the forest plot of the current analysis. Each horizontal line represents one study with the effect size a tick mark in the center of the line, and the 95% confidence interval at the distal of ends of each line. The numbers in the column on the far right are the individual effect sizes for each of the 25 studies in the meta-analysis. At the bottom of the plot, the black diamond shape represents the cumulative effect sizes and confidence intervals of the analyzed studies. Statistical significance is displayed by examining each line and the overall diamond in relation to the 0.00 vertical line on the x-axis. Any study that does not intercept the vertical line is either a positive or negative effect. The current forest plot is robust with 16 studies demonstrating positive effects, 1 negative effect study, and 8 with no effect. Thus, the effect sizes clearly show that post stroke rehabilitation involving bilateral arm training leads to improved motor capabilities and progress toward recovery.
Evaluation of the dispersion of effect sizes across the 25 post stroke bilateral arm training comparisons revealed an I2 = 62.81%. This indicated a relatively moderate level of heterogeneity and warranted the application of the random effects model for computing overall effect size in the current meta-analysis. This moderate consistency indicates that the cumulative effect (0.734) is robust across the domain of studies and outcome measures in our meta-analysis (Borenstein, Hedges, Higgins, & Rothstein, 2009).
Figure 2 displays the funnel plot examining publication bias by plotting treatment effect size as a function of standard error. Funnel plots evaluating bias in studies are constants in meta-analysis techniques (Higgins & Green, 2006; Borenstein et al., 2009). Each individual study is represented by a circle for the values of an effect size (x-axis) and standard error (y-axis). Because a majority of the circles of the 25 studies are on the right side of the funnel plot (right of the overall effect size vertical line: diamond on x-axis), then a potential bias effect is apparent. The closer to symmetry the circles are in a funnel plot the less concern with publication bias (i.e., only studies that find positive effects are published). The second funnel plot (see Figure 3) balances the studies toward symmetry by adding seven black circles to the left side (negative effects) of the vertical line (cumulative effect) that splits the funnel into two sides. The imputed circles are close approximations to symmetry as quantified by Duval and Tweedie's trim and fill procedure (Borenstein, Hedges, Higgins, & Rothstein, 2009). The black diamond on the x-axis is the recalculated overall effect. Technical experts agree that a perfectly symmetrical funnel plot represents a best estimate of an unbiased overall effect size (Borenstein, Hedges, Higgins, & Rothstein, 2009). Given the characteristics of the two present funnel plots, we are confident in stating that our large cumulative effect (0.734) is an unbiased estimate of stroke motor recovery progress as a function of bilateral arm training.
Rosenthal's classic fail-safe N analysis determined the number of non-significant studies necessary to negate the overall effect. Our analysis found a high fail-safe, N = 532. Thus, our meta-analysis found a robust cumulative effect. Moreover, a large number of null studies (i.e., no significant motor improvements found as a function of bilateral arm training) are required to reduce the effect of bilateral arm training interventions on motor recovery of stroke survivors.
As shown in Table 3, the 25 bilateral arm training comparisons displayed a relatively medium quality. Seventeen comparisons reported random assignment to groups whereas only eight failed to describe any randomization procedures. Fifteen studies conducted single blind experiments. Further, the low attrition rate for participants is viewed favorably.
Given that the current meta-analysis findings are based on a broad perspective of bilateral arm training techniques and supplements used during rehabilitation, we conducted a moderator variable analysis. As shown in Table 2, there are four general types of bilateral arm training techniques in the 25 comparisons: (a) pure bilateral – 6, (b) bilateral arm training with rhythmic auditory cueing (BATRAC) – 7, (c) coupled bilateral and EMG-triggered neuromuscular stimulation – 7, and (d) active and/or passive movements, including robotics – 5. These additional analyses elaborate on the contributions of each technique to our cumulative effect. Analyzing the 25 comparisons for a potential moderator variable indicated a significant overall mixed effects model equal to 0.80 (SE = 0.099, p < 0.0001; lower limit = 0.606; upper limit = 0.93). Separately, calculating the contribution of each type of bilateral arm training as moderator in a mixed effects analysis revealed two significant techniques with large effect sizes: (a) BATRAC (0.842, SE = 0.155, p < 0.0001; lower limit = 0.539; upper limit = 1.146) and (b) coupled bilateral and EMG-triggered neuromuscular stimulation (1.142, SE = 0.176, p < 0.0001; lower limit = 0.796; upper limit = 1.488). However, analysis of the active and/or passive movement training studies only indicated a relatively weak trend (0.535, SE = 0.326, p = 0.101; lower limit = −0.104; upper limit = 1.175). Finally, the pure bilateral training studies revealed a small effect (0.268, SE = 0.227) that did not reach significance (p > 0.10; lower limit = −0.178; upper limit = 0.713).
A second moderator analysis involved stroke patient's impairment levels and functional limitations: What impairment level or functional capability would benefit from bilateral arm training? Across the studies, pre-treatment impairment levels and functional limitations were assessed with a variety of tests: (a) Fugl-Meyer assessment of upper extremity, (b) Box and Block manual dexterity test, (c) Modified Ashworth, and (d) Action Research Arm Test. However, no general consensus was reported in the impairment tests, and the Fugl-Meyer scores ranged from least functional to most functional (5 – 64). Even though we were unable to conduct this planned moderator variable analysis, a partial answer on the type of stroke patient would benefit from bilateral training is that patients who experienced a mild or moderate stroke. General questions on the severity of the stroke patients in the various studies revealed a combination of mild to moderate severity, and severe to marked functional limitations.
The current meta-analysis revealed a strong bilateral arm movement training effect and improved motor capabilities post stroke. Indeed, all of the meta-analytic techniques support the conclusion that two specific bilateral arm treatments assist in making progress toward motor recovery. Moreover, a substantial number of training/treatment comparisons (N = 25) as well as participants in the treatment groups (N = 366) contribute to our robust findings. Adhering to strict inclusion/exclusion criteria, coding data rules for common motor outcome measures, and conducting five recommended meta-analytic tests, lead us to the compelling conclusion that the cumulative motor improvement evidence found for the upper extremity is real, rather than spurious.
These findings are important because the cumulative effect of bilateral arm training across the three phases of stroke recovery lends further support to the accumulating evidence that movement-based activities post stroke cause re-learning. Indeed, the current results extend the findings of three distinguished groups of stroke motor recovery researchers lead by Nudo, Cohen, and Hallett who report strong evidence that implicates neural plasticity; multiple improvements in the motor capabilities of the upper extremities after completing movement-based rehabilitation protocols (Chen, Cohen, & Hallett, 2002; Cohen LG & Hallett M, 2003; Hallett, 2001b; Hallett, 2002; Nudo, 1998; Nudo, 1999; Nudo, 2003; Nudo, 2006; Nudo, et al., 2003; Nudo & Milliken, 1996; Nudo, Milliken, Jenkins, & Merzenich, 1996). Neural plasticity refers to permanent changes in synaptic connections or neural ensembles as in establishing a habit state (Cauraugh & Summers, 2005; Cohen LG & Hallett M, 2003; Hallett, 2001b; Hallett, 2002; Hummel & Cohen, 2005; Nudo, 2006). Specifically, the current analysis identified two significant and large effect sizes involved training with BATRAC and coupled bilateral and active stimulation protocols. Movement-based experiences with either the BATRAC or coupled bilateral protocol appear to be essential and sufficient variables for improved motor capabilities.
Moreover, the large significant effects found for BATRAC and coupled bilateral training should be encouraging to therapists and rehabilitation specialists. Additional intervention options may help to significantly decrease the high percentage of stroke patients who continue to display motor dysfunctions one year post. Unfortunately, 60 – 65 % of stroke patients still have motor disabilities 12 months later (Lloyd-Jones, et al., 2008). Incorporating such efficacious bilateral arm movement training protocols into rehabilitation strategies may help to reduce the high percentage of motor disabilities. These two prominent bilateral arm training interventions represent effective strategies in helping stroke patients make progress toward motor recovery, and should be included in a comprehensive program to minimize motor dysfunctions one year post.
The current analysis focused on four specific bilateral arm training protocols, and no direct comparisons with other effective stroke rehabilitation approaches were made. Granted, quantifying motor improvements from CIMT versus bilateral arm practice would be intriguing. However, unilateral movements may generate an interhemispheric inhibition in the ipsilateral hemisphere that prevents mirror movements in the opposite upper limb. In contrast, bilateral movements activate similar neural distributed networks in both hemispheres, allowing mirror movements. Given the importance of mirror movements, future researchers will have to address the two contrasting hemispheric activation patterns (Ramachandran & Altschuler, 2009).
Concerning the meta-analysis technique, one strength is the ease of determining the contribution of potential extreme score effect sizes and calculating a new effect based on the standardized mean difference. Examining the forest plot of effect sizes (Figure 1) reveals one study (Lum, et al., 2006) with a large negative effect (−2.148). No other weighted effect size reached an extreme score area (> 2 SD), thus, we only removed the one study score from our analysis. The overall effect increased to 0.78 when 24 studies were re-analyzed. Consequently, the cumulative bilateral arm training effect is slightly larger without one extreme study with a negative value (or null findings).
Our overall effect is representative of the post stroke bilateral movement training literature. But, excluding 24 studies primarily because of missing treatment conditions (n = 14) and data extraction problems (n = 4) may raise questions about the precision of a meta-analysis. Excluding these studies when some did not support bilateral training would be a concern if the number of comparisons included in our analysis was less than 25. Further, our analysis included 366 treatment participants who experienced one of four bilateral arm practice protocols. Thus, our techniques are precise, robust, and consistent with conventional meta-analyses (Higgins & Green, 2006).
The moderator variable analysis on the type of bilateral arm training revealed two novel findings. Both the BATRAC and coupled bilateral training and EMG-triggered stimulation techniques produced significant and large effect sizes. These findings are based on seven different studies for each type of protocol. Supplementing bilateral arm movements with either rhythmically paced motion or active stimulation on the impaired arm increased the contribution of these protocols to the mixed design results. Such large and significant contributions to motor recovery were not found with pure bilateral or active and/or passive movement training protocols. Stroke researchers and rehabilitation specialists should be aware of the subtle differences in the four types of bilateral treatments as well as the effectiveness of each treatment. Moreover, likely neurological candidates for these motor improvements are bilateral neural activation patterns and associated mirror neurons (Ramachandran & Altschuler, 2009).
In closing, D. J. Glencross's 32 year old proposition on controlling skilled movements by integrating central and peripheral input still appears relevant (Glencross, 1977). As stroke patients attempt to overcome motor dysfunctions in activities of daily living, practicing bilateral arm training activates both central and peripheral input, and improvements are found.