SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. What this paper adds
  4. Method
  5. Results
  6. Discussion
  7. Conclusions
  8. Supporting Information

Aim  Developmental coordination disorder (DCD) is a significant disorder of childhood, characterized by core difficulties in learning fine and/or gross motor skills, and the attendant psychosocial problems. The aim of the meta-analysis presented here (the first on DCD since 1998) was to summarize trends in the literature over the past 14 years and to identify and describe the main motor control and cognitive deficits that best discriminate children with DCD from those without.

Method  A systematic review of the literature published between January 1997 and August 2011 was conducted. All available journal papers reporting a comparison between a group of children with DCD and a group of typically developing children on behavioural measures were included.

Results  One hundred and twenty-nine studies yielded 1785 effect sizes based on a total of 2797 children with DCD and 3407 typically developing children. Across all outcome measures, a moderate to large effect size was found, suggesting a generalized performance deficit in children with DCD. The pattern of deficits suggested several areas of pronounced difficulty, including internal (forward) modelling, rhythmic coordination, executive function, gait and postural control, catching and interceptive action, and aspects of sensoriperceptual function.

Interpretation  The results suggest that the predictive control of action may be a fundamental disruption in DCD, along with the ability to develop stable coordination patterns. Implications for theory development and intervention are discussed.


Abbreviation
DCD

Developmental coordination disorder

What this paper adds

  1. Top of page
  2. Abstract
  3. What this paper adds
  4. Method
  5. Results
  6. Discussion
  7. Conclusions
  8. Supporting Information
  •  The first quantitative review of the DCD literature on motor control and learning since 1998.
  •  The results show several areas of pronounced deficits in DCD.
  •  Predictive control and the ability to develop stable coordination patterns are core deficits in DCD.

Motor clumsiness (or developmental coordination disorder [DCD]) affects between 5% and 10% of all children1 (all references published online in Supporting Information). Efforts to understand the developmental precursors of DCD are important to avoid continued disruptions to skills development, secondary impacts on self-esteem and participation, and associated issues such as obesity, poor physical fitness, and social isolation.2,3 As a result, DCD has received considerable attention from researchers across disciplines including kinesiology, occupational therapy, paediatrics, physiotherapy, and psychology.4,5 Understanding the underlying mechanisms of DCD has been an important focus in an effort to better model atypical development generally, to optimize therapy, and to avoid some of the negative consequences of the disorder. Unfortunately, despite continued efforts to understand DCD, the aetiology of the disorder remains unclear.

The last comprehensive review of the literature on deficits underlying DCD was published in 1998.6 Since then, there has been a substantial growth in research designed to understand the disorder.7 Furthermore, approaches to the study of DCD have evolved rapidly since the mid-1990s. We have seen a sharp growth in research conducted from a more cognitive neuroscientific perspective, as well as in dynamic systems and hybrid approaches. This has broadened the domains of research and heralded something of a revolution in the choice of research paradigm, which is worth tracing.

The cognitivist approach, which dominated early research, seeks to understand behaviour by defining the set of internal cognitive processes that support it. The traditional metaphor used to represent the mind is the computer, generating a set of rule-governed computations. From a cognitivist perspective, simple chronometric and neuropsychological measures were dominant in DCD research up to the mid-1990s,6 and maintain some presence in the literature. Notable among these studies are those assessing aspects of executive function, drawing on the models of Baddeley8,9 and Shallice,10–12 for example. Working memory has been explored in a series of studies by Alloway et al.13–15 and by other groups,16–19 with the weight of data said to implicate visuospatial deficits. Problems in response inhibition have also been detected,20–23 as well as a more generalized impairment across set-shifting, working memory, and inhibition.16 At the level of executive attention (or supervisory control), deficits have been reported on dual-task performance,24 set-shifting,25 cognitive abilities such as planning and flexibility,26,27 and meta-cognitive awareness.28,29

Poor cross-modal integration has been implicated in a number of studies since 1996, although the exact pattern of impairment in DCD across conditions is variable.30–36 Others have reported problems in (intramodal) proprioceptive matching.32,37,38 Unfortunately, the strength of these effects has not been calculated or compared across different task conditions and processing domains.

Other aspects of sensoriperceptual dysfunction in DCD have also been reported on a consistent basis in the literature. These areas include basic visuosensory processing,39 visuospatial processing,16–19,31,40–45 tactile perception,31,46 kinaesthetic perception,32,34,47 and basic processing speed.48,49 Finally, at the level of procedural learning (a form of implicit memory), results have been mixed, with some reporting intact learning50 and others not.51

The integration of brain and behaviour under a single conceptual scheme is encapsulated by cognitive neuroscience. This approach has been revolutionary in the study of typical and atypical development, mapping the neural networks that support human cognition and action. Some of the earliest studies of DCD to adopt a cognitive neuroscience perspective appeared around the mid-1990s.52,53 The concept of predictive motor control (also known as internal forward modelling) has been particularly influential under this approach,54–59 and has been explored either directly or indirectly by several groups. Wilson et al. have used converging paradigms including motor imagery,60–66 double-step saccade task,67 and step perturbation reaching.68,69 Importantly, other groups have also used sequential reaching tasks, but with mixed results.70,71 Other methods used to explore internal modelling in DCD, and predictive control more specifically, include anticipatory postural adjustments during bimanual lifting,72 perceptual–motor adaptation,73–78 coupling of grip and load force during manual lifting,46,79 and smooth pursuit (eye) tracking.80 Deficits in covert orienting of visuospatial attention have also been taken to indicate problems of movement preparation and/or prediction,52,81 while others infer problems of inhibition.82–85 Note that force control has also been investigated more generically using isokinetic tasks86,87 and measures of peak output.88

Different aspects of computational motor control (i.e. attention, feedback, and feedforward processes) have been imbued with the language of neuroscience under the cognitive neuroscience approach. Mechanisms of control have been inferred from studies of overt orienting of attention during reaching,89,90 hand–eye coordination tasks,90 and manual steering/tracking.91–93 Feedback and feedforward processes have been implicated, but the magnitude of effects has never been compared across studies.

We have seen significant growth in studies of DCD from a dynamic systems perspective over the past 15 years. This approach has its roots in biological systems theory and ecological psychology. The main working assumption here is the interaction of multiple task, individual, and environmental constraints in the organization of movement.94 The timing and coordination of movement are viewed as emergent properties of the (individual) physical system in its interaction with the immediate environment,95 and not as centrally stored and generated outcomes. Researchers have endeavoured to describe the nature of rhythmic coordination and timing in DCD. This work includes the following: self-paced rhythmic coordination of unimanual and bimanual actions;96–98 rhythmic perceptual–motor coordination,a namely visuomotor96,98 and auditory–motor synchronization;99 stability of coupling in response to perturbation;100 and interlimb coordination between arms and legs;101,102 see also the work of Volman et al.103 Taken together, aspects of dynamic pattern stability and interlimb coupling are shown to be deficient in children with DCD. However, the magnitude of deficits across different task constraints has not been evaluated. Note also that dynamic planning (as distinct from motor execution) has been examined from a dynamical perspective and implicated in DCD.104

Hybrid approaches to DCD are those that blend conceptual schemes and paradigms from cognitivist, cognitive neuroscience, and dynamic perspectives.87,105 In other words, the conceptual framework is not pure but rather is integrative, defined by current trends in thinking across a variety of literature areas. More than ever, researchers are willing to integrate ideas from cognitive, learning, developmental, and neuroscience theories, as well as consider how the dynamics of movement might be instantiated neurally in the system.73,106 Hybrid approaches to DCD have covered different forms of manual control (reaching and catching, in particular), gait and posture, timing, and aspects of praxis (including imitation). Work on manual control includes kinematic studies of target-directed reaching under different task constraints,107–110 reach–grasp dynamics,109,111 the sensory control of manual pointing,112 head–torso–hand coordination during reaching,105 aiming using a stylus or pen,113 manual interception of simple visual stimuli,114 and graphomotor control.115–118 Prospective judgements of reaching/grasping have also been examined.34,119,120

The kinematics and dynamics of catching under different task constraints has also been an area of intense interest, mainly by Astill et al.,121–124 Deconinck et al.,125 and others.126–128 Problems of interlimb coupling have been reported, along with the temporal control of grasp or interception.

The spatiotemporal control of gait has been investigated under conditions of normal walking129–131 and under restricted vision;132 it has been suggested that individuals with DCD place greater reliance on visual control, but findings appear to be equivocal. Similarly, control of static posture has been investigated under normal conditions133 and under altered constraints and perturbation, that is, visual,133–138 physical,136,139,140 and cognitive.44,141,142 In general, postural deficits are suggested to be greater with added external constraints. Similar arguments have been made for dynamic postural control while reaching, leaning,143 or lifting.144

Motor timing issues have been reported in DCD using a blend of traditional and neuroscience methods: finger tapping paradigms,145 rhythmic finger synchronization,146 and rhythmic arm movements.147 Increasingly, timing is modelled as a distributed system of neuromuscular control, one involving primary sensorimotor cortex, posterior superior temporal gyrus, cerebellum, and supplementary motor area.148

Praxis deficits to imitation and verbal command have also been reported,149–152 suggesting disruption to motor planning. Praxis to imitation, in particular, has been linked to left parietal dysfunction. And, finally, associated movements (or motor overflow) suggest poor synergies in movement production, even for relatively well-rehearsed skills such as running;153,154 their neurological status is unclear.

The full complement of performance categories under each approach is listed in Table SI (supporting information published online), together with common outcome measures. We see, in part, the evolution in approaches to DCD, as well as the sheer range and complexity of paradigms and their associated metrics. Clearly, there is a need to synthesize this vast volume of data, drawing out the common threads in findings that will help push the field forward.

Also motivating this review was an initiative from representatives of the Association of the Scientific Medical Societies in Germany and the European Academy of Childhood Disability to develop Clinical Practice Guidelines for DCD in German-speaking countries.155 Because motor clumsiness in children is defined somewhat differently across countries, it was necessary to initiate an international consensus to confirm and/or modify a previous international consensus (London Consensus, Leeds Consensus). Chaired by Professor Rainer Blank, a panel of international researchers was established to look at three key aspects of DCD: aetiology, diagnosis, and treatment. A core aspect of examining aetiology concerned a systematic review of findings of impaired functions or underlying mechanisms.

A meta-analytical approach was used because, by combining estimates from multiple studies, meta-analysis increases the total number of primary units for analysis (i.e. participants), reduces the sampling error of an association,156 and permits evaluation of variables that moderate group differences, such as age, sex ratio, and severity of the primary condition.157 It is one of the primary means of data synthesis, and a means of building knowledge in a given domain.

The specific aims of this meta-analysis were (1) to present a quantitative review of the DCD performance literature since January 1997; (2) to identify the deficits that best discriminate between children with and without DCD; (3) to identify patterns in the constellation of deficits that suggest causal mechanisms or that fit particular accounts of the disorder; (4) to determine the moderating effect, if any, of sex and age; and (5) to identify areas where more systematic research is needed to build knowledge and inform our approaches to treatment.158

Method

  1. Top of page
  2. Abstract
  3. What this paper adds
  4. Method
  5. Results
  6. Discussion
  7. Conclusions
  8. Supporting Information

Sample of studies

A computer literature search was conducted for papers published between January 1997 and August 2011 using 14 electronic databases: MEDLINE, Cochrane Library, PubMed, CINAHL, PsycINFO, PsychLit, Science Direct, OTDBase, OTseeker, PEDRO, ERIC (EMBASE), HealthStar, Expanded Academic, and Sports Discus. These databases were selected as they represent a broad spectrum of disciplines that conduct research related to motor development and DCD. The search was conducted in English and incorporated reviews, research papers, and pilot studies. The following key terms were searched: developmental coordination disorder (DCD), motor skills disorder, clumsiness, clumsy, clumsy child syndrome, clumsy child, incoordination, dyscoordination, minimal brain dysfunction, minor neurological dysfunction/disorder, motor delay, perceptual–motor deficit/difficulties/dysfunction/impairment, developmental dyspraxia, dyspraxia, dysgraphia, developmental right hemisphere syndrome, movement disorders, motor impairment, motor skills disorder, motor coordination difficulties/problems, motor learning difficulties/problems, mild motor problems, non-verbal learning disability/disorder/dysfunction, sensorimotor difficulties, sensory integrative dysfunction, physical awkwardness, physically awkward, psychomotor disorders, deficits in attention, motor control, and perception (DAMP), and apraxia. Hand searches were made of the reference lists and bibliographic supplements of all relevant narrative reviews and research reports. Terms omitted from the search were consistent with the diagnostic exclusion criteria for DCD,159 which were cerebral palsy, stroke, traumatic brain injury, leukodystrophia, and muscular disorders.

The criteria for inclusion of a study in the review were (1) a publication date between 1997 and August 2011; (2) research that provided a comparison between children meeting a minimum definition of DCD and a typically developing control group; (3) publication in peer-reviewed journals; (4) provision of information sufficient to calculate effect size; and (5) studies that reported behavioural data that relate to processes or mechanisms of motor control, learning, or cognition. Consistent with DSM-IV-TR criteria and research definitions of motor clumsiness,95 motor-impaired children had to have no identifiable neurological disorder, an IQ within the normal range, and no gross physical or sensory impairment.

A total of 517 studies were identified from the computerized database search. Of these, 402 studies did not meet the inclusion criteria: 220 studies did not present behavioural data bearing directly on motor control, learning, or cognition in DCD, while 182 studies presented insufficient data to calculate an effect size. The final number of studies included in the meta-analysis was 129.

Coding of studies

The variables selected for coding included the test category, outcome statistics, publication year and source, and aspects of research design and method such as sample size, sex ratio, screening protocols, matching criteria, and severity of DCD. Studies were coded independently by two experts in the field. In instances of disagreement in assigning test category (<3% of all such coding decisions), a third expert was consulted and the disagreement was settled by consensus.

Test categories

A coding system was developed that best represented the main theoretical approaches to the study of DCD and research paradigms.160 In broad terms, studies in motor development and DCD (like others in the motor control and learning literature) can be grouped into four main approaches: (1) cognitivist, (2) cognitive neuroscience, (3) dynamic systems, or (4) hybrid. We acknowledge, however, that the borders can be blurred depending on the paradigm of choice and the disciplines involved in a given study. It was unavoidable that some categories represent multiple constructs given that motor performance and control is a uniquely interactive system. Notwithstanding this, we have endeavoured to code the dominant construct where possible, while acknowledging overlap. For example, the category ‘forward modelling and online control’ under the cognitive neuroscience approach (Table SI) includes paradigms that are designed to test these processes more directly at a neurocognitive level; it can be regarded as domain generalb in this sense. Aspects of forward modelling are also present within other categories that can be regarded as more domain specific,c and some are present under the hybrid approach (e.g. catching under conditions of visual occlusion). Ultimately, the coding scheme for test categories was based on an extensive literature review that included current, learned reviews of the motor control literature,161,162 recent developments in the cognitive neuroscience of action,59,163,164 the main currents of thinking in dynamic systems,165–168 and cognate literature including factor-analytic studies of psychomotor performance in children.169 Studies were then coded based on the research paradigm and how independent variables (i.e. experimental conditions) were manipulated to capture different aspects of motor control. For example, under the covert orienting of attention, conditions that use central symbolic cues capture a voluntary mode of control while peripheral cues capture an automatic mode. Dependent measure(s) from each test paradigm and conditions were grouped according to the process or underlying mechanism that was being assessed. The aim was to achieve a balance between generality and specificity. Decisions about where to clump variables and where to split were an iterative process informed by theory and expert discussion. Definitions of key test categories under each theoretical approach are provided in Table SI, together with examples of common tests/measures.

The following aspects of the outcome measures were also coded: (1) means, medians, and SDs for each data entry for the two groups; (2) results of statistical tests such as t, F, U, and p-values, and (3) the results of single-sample comparisons based on z-scores and centile ranks. If the result of a parametric test was not reported, p=0.50 was assumed. When a group difference was reported without relevant statistics, p=0.05 was assumed. If several groups of children with DCD were included in a single study, all DCD–control comparisons were coded (e.g. groups formed according to the severity of motor impairment).

Study attributes

Variables describing the individual research paper were (1) publication year and (2) journal name.

Variables regarding participant and research design factors were (1) total sample size; (2) number of participants with and without DCD; (3) number of male and female participants in each group; (4) mean age and SD of children in each group; (5) screening procedure (one-step, involving a standardized motor assessment alone, or two-step, involving teacher/instructor referral and standardized assessment); (6) motor measures (i.e. Movement Assessment Battery for Children, McCarron Assessment of Neuromuscular Dysfunction, and others including the Bruininks–Oseretsky Test of Motor Impairment); (7) DCD cut-off criterion (5th, 10th, or 15th centile); and (8) severity of DCD (estimated from motor screening data, where available).

Reliability and validity of the coding system

Validity can be maximized by conferring with other experts in the field who can cross-validate the design of the meta-analysis.158 Our coding system was validated using three experts in DCD research, two located in Australia and one in Europe. To ensure reliability of the coding, all data entries were rated independently by two other experts in child motor development. In instances where disagreement occurred, a consensus decision was reached in consultation with a third expert.

Calculation of effect size

Estimates of effect size (Cohen’s d) were calculated for each comparison between children with DCD and typically developing children. This measures the magnitude of the difference between the mean scores of criterion and comparison groups, divided by a pooled SD.170 In the absence of these data, t, F, z, χ2, and p-values were used when the sample size was available. The meta-analyses used a random-effects model based on the method of Lipsey and Wilson.171 This approach assumes that variation between effect sizes is attributable not only to random sampling error but also to other (undetermined) factors within studies; as such, two separate estimates of error are incorporated into the model. The random-effects model is preferable when effect sizes are heterogeneous within a task or construct domain; this assumption was borne out by the presence of significant heterogeneity in a number of test categories, indexed by Q-statistics. While this approach is less powerful than a fixed-effects model, it does permit greater generalization.172

Since the reliability of a given effect size is greater for larger samples, effect size was weighted by the inverse of its variance.172 In other words, a weighted effect size (dw) was calculated for each comparison between children with DCD and typically developing children. For each task category (e.g. motor imagery), a mean dw value was calculated using the SPSS macro of Lipsey and Wilson.171

Data aggregation

The inclusion of multiple dependent measures (and associated statistics) was permitted for each study. Mean effect sizes are not unduly affected by non-independence in most instances,172 and many authors choose not to weight studies according to the number of effect sizes. Pseudo-independence of effect sizes was assumed here.

Analysis of effect size

Outcome measures were analysed at the level of each finding; there were 1785 effect sizes, distributed among 134 categories (including all subcategories). Methodological variables were analysed at the study level, with up to 129 entries depending on the availability of data. A single entry for each variable from a single study was obtained by calculating the mean value.157 A moderator analysis was conducted at the study level using multiple regression.6 Mean effect size was regressed on six potential moderator variables: age (participants with DCD), number of participants with DCD, percentage of male participants with DCD, severity of DCD, matching on IQ, and year of publication.

Heterogeneity of effect size is tolerated using a random-effects model by being incorporated into estimates of effect size. Nonetheless, we did test for significant departures from homogeneity using the Q-statistic.173 Heterogeneity in a group of effect sizes can indicate that the variability in the effect sizes may be due to outliers or a moderator variable. When the distribution of effect sizes within a given category was not homogeneous, we first examined a scatterplot for outliers. An outlier was defined as any value with a z-score greater than 1.96 or less than −1.96.174 Homogeneity after removal of a small number of outliers has been shown to yield valid conceptual categories and empirical relationships.158

Interpretation of effect size

For the indices of effect sizes, a positive dw value indicates a more favourable result for the group of typically developing children while a negative value indicates a more favourable result for the group of children with DCD.6 For each test category, statistical confidence in the group effect was indicated by moderate to large effect sizes and 95% confidence intervals (CIs) that did not span zero.171 The magnitude of the mean effect size estimates (dw) were then interpreted according to the conventions of Cohen:170 0.30 (small effect size), 0.50 (moderate effect size), and 0.80 (large effect size). A fourth level of magnitude, ‘very large’, was defined notionally as dw>1.0.

Results

  1. Top of page
  2. Abstract
  3. What this paper adds
  4. Method
  5. Results
  6. Discussion
  7. Conclusions
  8. Supporting Information

Characteristics of the studies

The characteristics of the groups of children with and without DCD are shown in Table SII (supporting information published online). The median size of each group (the best estimate here of central tendency) was 16 for children with DCD and 20 for comparison groups. In general, there were twice as many males as females in groups of children with DCD. The average age of each group was 108 months (9y). The mean age of youngest reported group was 5 years of age. Only 21 studies (16%) matched DCD and comparison groups explicitly on IQ, with some version of the Wechsler Intelligence Scale for Children (3rd edition)175 used most frequently (15 times). The severity of DCD could be calculated in 79 studies (61.2%) based on motor screening data (i.e. the Movement Assessment Battery for Children and the like). There was some variability in severity, which ranged from a d value of 1.13 to 7.62. In 71 studies (55%), the full version of the Movement Assessment Battery for Children176 was reported as being used to identify children; the manual dexterity subscale was used for this purpose in one study, and the Movement Assessment Battery for Children Checklist was used to confirm DCD status in three. The McCarron Assessment of Neuromuscular Dysfunction177 was used in six studies and other motor batteries were reported in nine. Parent rating scales were used in three studies to confirm DCD status. There was insufficient information to describe accurately the method of screening in the remaining studies; however, the methods did include specialist referral and unspecified motor tests.

Correlational analysis

At the study level (k=129), the interrelationship between the potential moderator variables and mean effect size was examined. First, correlations showed no significant relationship between any moderator and mean effect size, with all p-values >0.10. Second, to explore historical trends in the research, year of publication was correlated with the other moderators. The only significant correlation was between year and matching on IQ (rpb=−0.21, p=0.015), reflecting only a slight trend for more recent studies to match explicitly on IQ less frequently. Third, of the remaining relationships, the only significant correlation was between the percentage of males in groups with DCD and sample size (r=−0.28, p<0.001); a higher percentage of males was associated with smaller groups. Overall, the magnitude of effect size was independent of the moderator variables, with no variable explaining more than 5% of the variance. The interrelationships among the moderators were limited.

Magnitude of group differences

The total number of effect sizes for all studies combined was 1785, including entries for motor and IQ scores. The weighted mean effect size (dw) across all categories was 1.11, indicating that DCD and control groups were highly discriminable, and that the effect size was large. When the motor and IQ screening measures were removed from this pool, there were 1581 effect sizes, the mean of which was large (dw=0.97). Within categories, information relevant to moderator analyses was quite limited and, hence, did not permit a meaningful analysis of other sources of influence on effect size variability.

Weighted mean effect size values within each performance category are presented in Table SIII (supporting information published online). As already mentioned, positive values indicate that the typically developing children performed better than the children with DCD. Within each domain, mean effect size values are organized in descending order of magnitude. For the test categories, Table SIII shows that these mean values range from +0.08 to +6.55. Indeed, for all 119 test categories listed in Table SIII, the children with DCD performed worse than typically developing children. Moreover, in 80 of these categories (or 67%), the magnitude of the group difference was at least large (mean dw>0.80).

There were, however, a number of categories for which the weighted mean effect size was very large (dw≥1.0) or close to this cut-off. Under the cognitivist approach, the strongest discriminators were aspects of visual processing including form detection (r=2.59), motion detection (1.34), and complex visuospatial tasks involving movement (1.27), tactile perception (1.60), executive function (seven categories ≥1.00), and procedural learning (1.50). Notably, the three main facets of executive function all had categories with large mean effect size values: inhibition, working memory, and executive attention. Under the dynamic approach, large mean effect size values were shown for categories assessing rhythmic stability under a variety of task and informational constraints (four categories ≥1.00), as well as bimanual (auditory–motor) timing (1.56). Under the cognitive neuroscience approach, large magnitude effect sizes were shown for explicit motor imagery (2.43), a number of categories assessing aspects of internal modelling (four main categories ≥1.50), inhibitory processes involved in the covert orienting of visuospatial attention (1.04), uncued overt orienting of attention (1.76 and 1.17), both ocular tracking (2.09) and manual pursuit tracking (three categories, ranging from 0.85 to 1.42), and force control (1.53 and 1.09). Under the hybrid approach, large effects were shown for a number of control factors associated with target-directed reaching (seven categories ≥1.00, and 11 >0.80), force control during target interception (1.24), aspects of control and dynamic coordination associated with catching (four categories, ranging from 1.70 to 6.55, and 9 >0.80), the spatial and temporal control of gait pattern (five categories ranging from 0.86 to 1.40), the control of postural sway under both static and dynamic conditions (four categories between 0.89 and 1.08), praxis to command (0.95), and associated movements (3.00). For the kinematic parameterization of reaching more specifically, strong effects were shown for feedforward control of head–torso–hand synergies (1.35), visual control of aiming movements (1.22), precued response time (1.53), spatiotemporal control reach and grasp phases (1.24 and 1.12 respectively), and both unilateral and bilateral movement time (1.44 and 0.94).

The categories that best discriminated between children with DCD and typically developing children are also represented in Table SIV (supporting information published online). This shows 80 categories with a weighted mean effect size (dw) of at least 0.80, listed in order of magnitude. Most notably of these, there were 20 categories with dw>1.50, and a further 38 between 1.00 and 1.50 inclusive.

Outlier analysis

For those test categories that did not achieve homogeneity, the distribution of effect sizes was plotted and outliers removed; outliers were identified for 26 categories. The results of analyses excluding outliers are given in Table SV (supporting information published online). For all categories, outliers were relatively large effect size values; hence, their removal reduced the weighted mean effect size (dw); however, in most instances the reduction was slight. For seven categories, homogeneity was achieved when the Q-statistic was recalculated. In short, the combined effect sizes for all categories in Table SV did not change dramatically after the outlier analysis.

Discussion

  1. Top of page
  2. Abstract
  3. What this paper adds
  4. Method
  5. Results
  6. Discussion
  7. Conclusions
  8. Supporting Information

There are several key findings from this meta-analysis. Across all outcome measures (excluding screening items), a relatively large effect size (dw=0.97) suggests a generalized performance deficit in children with DCD. This was somewhat larger than that reported in the 1998 meta-analysis,6 in which the mean effect size (r) of 0.37 equates to a d value of 0.80. The moderate-to-large effect sizes for perceptual processing (visuospatial and tactile) and cross-modal translation were of similar magnitude to those reported in 1998 by Wilson and McKenzie.6 This is consistent with the evolution of work in the field, which has become less exploratory and more directed at testing specific hypotheses; with tighter control, experiments tend to yield higher effect sizes. This argument is supported by the fact that the average severity of DCD in the children reported here (d=3.60) was somewhat lower than that reported in the 1998 study (d=4.69, converted from r=0.92). Whether this generalized level of dysfunction reflects a form of atypical brain development178 is an issue for continued exploration. Notwithstanding this, control deficits in DCD were pronounced in a number of broad categories or clusters: this forms the basis of our discussion. We acknowledge that some specific categories do have multiple determinants (or reflect overlapping constructs). This is unavoidable in representing a quite interactive system like motor control and learning. However, we have endeavoured to map the dominant construct in each case, while acknowledging the overlap. We identified seven main clusters under which task categories with large effect sizes could be grouped; in each instance, there were at least four task categories of very large effect size. In the body of the discussion that follows, we deal in turn with those clusters that are domain general and those that are domain specific. The former comprises (1) internal (forward) modelling, (2) rhythmic coordination, and (3) executive function, whereas the latter describes (4) control of gait and posture, (5) control of reaching, (6) catching and manual interception, and (7) aspects of sensoriperceptual function.

Internal (forward) modelling

There is converging support for deficits in the forward modelling of movement (i.e. predictive control). This hypothesis has been proposed previously by Wilson et al.179 under the internal modelling deficit hypothesis,94 with similar arguments put forward by Kagerer et al.73 What is quite compelling is the variety of paradigms and findings that contribute to this view. Taken together, it appears that children with DCD have difficulty generating or using predictive estimates of body position as a means of correcting actions in real time. This problem may also affect their ability to learn new internal models or modify existing ones over repeated learning trials.

Of the top 20 ranking categories in Table SIV, four fell under the (domain-general) area of internal modelling (specifically motor prediction). Two additional categories in the top 20 were also intimately related to motor prediction, namely motor imagery and catching under visual occlusion. Another category that reflected predictive control, but with a more moderate effect size, was prospective reaching involving the judgement of end-state comfort (0.72). Importantly, problems in motor prediction were evident under a variety of task constraints and over different timescales. For example, under tight temporal constraints, children with DCD had pronounced difficulties adjusting reaching movements to rapid visual perturbation69 and in coordinating grip and load force when making lifting movements.79 Of the former, difficulties were shown by prolonged movement time on step-perturbation reaching (dw=1.60). The ability to make these rapid (online) corrections is thought to be contingent on how well the nervous system can predict the future location of moving limbs using a forward internal model (see footnote d).54,58,164,180 Forward estimates of limb position provide a means of integrating efferent and afferent signals under tight temporal demands, thereby speeding responses to any changes in the environment during the course of movement.180 In neural terms, a functional loop between parietal cortex and the cerebellum is thought to monitor these forward estimates of limb position and correct ongoing motor commands online should the action deviate from expectations.59,181

Problems with feedforward control may also explain the pattern of results for other kinematic parameters that describe target-directed reaching. Spatiotemporal parameters for both the reach and grasp phase of movement showed high effect sizes (dw=1.24 and 1.12 respectively), as did the ability to use visual precues during the premovement phase (1.53). These effects were independent of target distance: in preliminary analyses, we collapsed these parameters over distance (loosely defined as near vs far) because no differences in effect size were evident. Group differences were somewhat higher again for manual aiming involving the translation of visual to proprioceptive information (1.22). At the level of neuromuscular control, this set of deficits suggests that slower reaching and poor trajectory control may reflect an abnormal pattern of agonist–antagonist muscle activation in DCD. These children are more reliant on slower feedback control, homing in on targets in an incremental fashion, as shown by more movement subunits and greater variability in joint and endpoint kinematics. The cerebellum is intimately involved in regulating the timing control between agonist and antagonist bursts in a predictive manner;182 more precisely, it is involved in the learning of timed motor responses, as well as online processing using feedforward information.

Forward estimates also inform the coupling of grip and load force during lifting (1.57), as well as anticipatory postural adjustments (2.20). In the case of grasp-and-lift movements, the inertial properties of the to-be-lifted object are anticipated in advance and used to programme an appropriate level of grip force that will ensure that the object does not slip during the upwards lifting phase. In DCD, safety margins tend to be higher, indicating that these children are less able to anticipate the consequences of their own movement.79 Combined with a more fundamental issue in force production per se, that is programming isokinetic force (1.53) and in generating peak torques (1.09), these children are at a distinct disadvantage when implementing and predicting force outputs.

Other high-effect-size categories that are consistent with the notion of impaired predictive control include catching under different degrees of visual occlusion (5.37), the explicit use of motor imagery (2.43), visual smooth-pursuit tracking (2.09), and manual tracking (1.42). The tracking results are particularly notable because they suggest that disrupted eye movements may be a useful marker for DCD. In the study by Langaas et al.,80 these children were less able to synchronize their eye movements to a moving target that followed a predictable sinusoidal path. Greater (unsigned) lag suggested that the predictive model for the eye movement was less well refined, regardless of whether the tracking response was ahead or behind the target. Integration of parietal cortex- and cerebellum-based control mechanisms may explain this type of performance: this is discussed further below.

Difficulties learning to develop new internal models for action are seen in the results of motor adaptation studies. The work of Kagerer et al.,73,74 using prism adaptation, shows that children with DCD are less affected by visual distortion during adaptation trials, and show reduced after-effects when the distortion is removed (0.84). Intriguingly, this effect appears to be greater when these children are exposed to a gradual visuomotor perturbation than to an abrupt one.73 Taken together, the paradigm shows significant impairment in DCD of the ability to update an existing forward model for target-directed aiming movements of the upper limb.

Finally, deficits in the explicit use of motor imagery (dw=2.43), for example on visually guided pointing,61,63 also suggest some disruption in the ability to predict the behaviour of the motor system under different task constraints. Among typically developing children, the duration of real and imagined movements was shown to increase with a reduction in target size, consistent with Fitts’s law. By comparison, among children with DCD the relationship is not preserved under imagined conditions and, furthermore, their performance is not constrained by variations in the weight of the object.63 This pattern suggests a broad-based deficit in predicting both seen and felt consequences of a prospective action. By comparison, mean effect sizes were somewhat lower (0.84) for tasks involving implicit use of motor imagery, mainly mental rotation tasks involving limb or whole-body stimuli.66 Results for mental rotation have varied somewhat according to stimulus type/complexity, the outcome measure, and the severity of DCD. However, deficits tend to be stronger in children with severe DCD than in those with mild DCD.62 Moreover, those with severe DCD are less able to benefit from explicit imagery verbal instructions in order to improve their performance. This perhaps reinforces the importance of using dynamic visual models when training movement imagery.183

Finally, it is interesting to note that for categories that represent the covert orienting of visuospatial attention, effects were fairly uniform; mean effect size ranged from 0.90 to 1.01 across tasks that enlist either voluntary or automatic modes of control. If anything, effects were marginally higher on specific measures of inhibitory control for the voluntary mode of orienting (1.04), for example the ability to resist invalid symbolic cues and to disengage attention rapidly from these locations.83

Taken together, there is converging evidence that children with DCD may have a core deficit in predictive control. In real time, a forward modelling deficit would limit both the speed and accuracy of online error correction.180 Over longer timescales, the child with DCD may be less able to train (or recalibrate) internal models for action, or may require far more practice to build an adequate model for a given movement. The control problem is even greater when biomechanical constraints change with maturation, particularly during growth spurts. These changing relationships between physical growth, biomechanics, and control are areas in need of more intensive investigation.

The dynamics of rhythmic coordination in DCD

There were five categories under rhythmic coordination that registered effect sizes between 1.04 and 2.89; four of these were in the top 20 categories listed in order of magnitude (Table SIV). Notably, four of the five represented the stability of rhythmic coordination under different task constraints: auditory–motor and visuomanual synchronization, self-paced timing, and hand–foot coordination. Stability was measured mainly by the variability of the relative phase between the limbs, critical frequency (or the point at which antiphase pattern stability is lost when movement frequency is scaled up to a metronome beat), and relaxation time (i.e. the time needed to regain a stable pattern after physical perturbation). Children with DCD generally exhibit more variability in the ability to maintain a stable coordination pattern at a constant speed;96,100 this is shown both spatially and temporally across unimanual and bimanual movements. There is some suggestion, however, that the coordination deficit is even greater when children attempt to synchronize movements to an auditory cue relative to visual.99 This view was supported by the pattern of effect sizes. Two of the three largest effects observed under rhythmic coordination were for auditory–motor synchronization stability parameters (2.89) and timing parameters (1.56).

The pattern of performance in perceptual–motor synchronization can also indicate whether children with DCD use a predictive mode of control as opposed to one based more on response feedback. As Whitall et al.99 have suggested, the former would be indicated when a child tends to move consistently in advance of the external signal. In DCD, the precise timing pattern may vary somewhat with the type of task. For unimanual tapping, faster frequencies have been shown to present more difficulty in DCD, notably a phase lag.184 But, for bimanual auditory–motor coupling, slower frequencies have been shown to be more difficult when non-homologous fingers are used: children with DCD find it more difficult to slow down their tapping rate. There was some suggestion in this study that the requirement to inhibit the contralateral homologous finger may explain this effect: inhibitory demands may be greater at lower movement frequencies. The fact remains, however, that children with DCD displayed timing deficits in both self-paced tasks and those involving external pacing (or perceptual–motor coupling). Put another way, external pacing does not improve rhythmic timing in DCD. Indeed, the ability to couple movements to an auditory entraining signal is not at all well developed.

Taken together, the rhythmic timing deficits that are apparent in DCD suggest some disruption to underlying oscillatory mechanisms and/or the ability to form and monitor internal models of action. The former is suggested for single and multilimb coordination, with the coupling of putative oscillators disrupted, perhaps at the level of the cerebellum.106 In the case of bimanual coordination, for example, two rhythmically moving fingers can be modelled by two oscillators that are coupled in a non-linear fashion, as in the Haken–Kelso–Bunz model.185 The reduced stability of coordination shown in the meta-analysis suggests that the coupling strength between component oscillators is reduced in DCD.100 Indeed, this underlying dysfunction extends to two-unit and four-unit oscillator systems given the consistent pattern of performance for bimanual and hand–foot coordination paradigms. The corticocerebellar axis has been implicated in the ability to maintain stable couplings between component oscillators.106 Dysfunction at this level is indicated in DCD. Recent functional magnetic resonance imaging data reveal hypoactivation of the parietocerebellar and frontocerebellar axes in DCD while performing a repetitive tracing task.118

From a somewhat different perspective, internal modelling involves the ability to map the rhythmic response in a feedforward manner to the required tempo. As suggested earlier, these internal models are established during the course of normal development and experience, and built up according to the particular constraints and tempo of the task at hand. For example, rhythmic movements are generally performed at a frequency that is constrained by individual biomechanics and rate-limiting factors such as the presence or absence of time pressure. The learned relationship between input and output signals is thus contextually bound, and enables the child to predict the behaviour of the system under similar constraints. In the case of externally cued action, for example, these learned perceptual–motor maps enable the child to synchronize movement to the beat by anticipating the dynamics of the limb and the timing of the stimulus.99 This type of anticipatory control is reduced in children with DCD. Whether the underlying control issue is explained by delayed development or deviance is not completely clear. We do see, however, a pattern of response lag in younger typically developing children for both in-phase and antiphase movements, suggesting that they too have not fully developed the ability to synchronize movements using a predictive (or feedforward) approach.99 Longitudinal designs involving neuroimaging are needed in this area of research to better map the corticocerebellar mechanisms that are thought to underlie rhythmic coordination and timing.

Executive function in DCD

Quite pervasive difficulties were shown across domains of executive function: working memory (1.07, averaged over visuospatial and verbal), inhibitory control (1.03), and executive attention (1.46). This finding is consistent with Piek et al.'s16 recent suggestion that generalized executive dysfunction is common in DCD. Indeed, the degree of dysfunction has been reported to be greater in children with DCD than in children with attention-deficit–hyperactivity disorder. Importantly, the deficits extend to more sophisticated aspects of cognitive control including dual-task performance (1.07 under motor load) and metacognition in a skill-learning context (1.44).

The constellation of motor and cognitive deficits in DCD is striking, and prompts us to reconsider the notion of atypical brain development.178 Kaplan’s broad argument is that dysfunction across modalities and task condition is likely to be a by-product of problems in general cortical maturation. Although the atypical brain development hypothesis draws attention to this overlap, ultimately it does little to orient theory and research to likely causes. Causation is better conceptualized within a model of brain development involving interactive specialization.186 There are two interrelated possibilities here about mechanisms of DCD: first, putative neurocognitive and neuromotor systems are not functioning as they should because of intrinsic maturational factors and, second, the delayed brain is not afforded the learning experiences (namely, stimulation) that would enable efficient couplings to form between increasingly specialized subsystems.

Motor control parameters associated with posture and gait

Seven categories in the domain-specific areas of posture and gait showed very large effect sizes. Notably, the four largest of these concerned gait: classification of gait pattern (2.15), the spatial patterning of gait under normal sensory conditions (1.39), temporal patterning under reduced vision (1.13), and gait velocity under reduced vision (1.40), although the last was based on only one value. In addition, large effect sizes in this domain covered both spatial (0.92 and 0.79) and temporal (1.13 and 0.86) aspects of control; in each case, the effects were only slightly larger under conditions of restricted visual information. With respect to posture, the three very large effect sizes all concerned the control of postural sway, both static and dynamic (1.01–1.08); in addition, a large effect size was observed for sway under perturbed conditions (0.86). Taken together, whether or not posture was tested under additional constraints (e.g. without vision or with physical perturbation) made little difference to the magnitude of group differences.

The large effect sizes for static postural control that were shown across unperturbed (1.01) and perturbed (0.89) conditions suggest a fundamental deficit in DCD. Measures of absolute sway and sway variability show that the average child with DCD has difficulty controlling the complement of joint linkages that enable them to maintain a stable alignment over their centre of mass. While some have suggested that reduced vision exacerbates this postural deficit,133 the meta-analysis suggests that the deficit is present regardless of the visual constraints. Moreover, our preliminary analyses showed no appreciable difference in the magnitude of effect size on measures of sway between visual, physical, and combined perturbation types; a generalized impairment of postural control across different task constraints was suggested. This does not rule out the possibility that a certain DCD subgroup may be more reliant on vision, for example, than others.136 However, the generalized level of postural dysfunction is a notable finding from our review.

Although not covered by the current review, physiological data can clarify the neuromuscular basis of the postural issues in DCD and should be the focus of careful analysis in the future. Alternatively, reduced postural stability and greater variability, regardless of the external constraint, may suggest higher levels of neural noise in the motor system:186,187 this too is worthy of further consideration.

One of the oft-noticed aspects of mobility in DCD is variability in cadence and rhythm that can be observed by the trained eye.188 The kinematic data support this observation. Variability in gait pattern (dw=1.39) and atypical pattern classification (2.15) suggest major anomalies in the way children with DCD acquire and automate their walking. What, then, does this result suggest about underlying processes and mechanisms? In a detailed kinematic study using Fourier analysis, Rosengren et al.130 showed that the relationship between the shank and thigh in children with DCD is highly variable and often asymmetrical between the right and left leg. This variability is likely to explain differences in mobility performance. However, whether the patterning issue is a cause or consequence of the balance and postural control issues we see in DCD remains unclear. One common denominator may be neuromuscular problems, including lower levels of absolute strength (r=0.54) and higher coactivation, reported in other work.88,110

Coordination and control of catching and interception

Four catching categories were among the 20 best discriminators assessing aspects of coordination and prediction. The ability to predict trajectory under partial visual occlusion (5.37) was particularly problematic, and serves to highlight issues with processing dynamic visual information under high temporal demands. Online control of the transport phase (2.07) and intralimb coupling (6.55; e.g. time to maximum grip aperture as a percentage of total movement time) also yielded very high effect sizes. Completing the picture were delayed initiation of the transport phase (1.07), a higher safety margin when scaling grasp aperture (1.05), poor timing of the grasp action (1.12; e.g. earlier initiation and later closure), reduced velocity of grasp opening/closing (1.23), and reduced smoothness of the grasping action (0.99). Taken together, we see evidence of separate programming for the transport and grasp phases, rather than a smooth coupling.124 Children with DCD prefer to wait before initiating transport, which then reduces the time window for its execution, with associated aberrations in grasp dynamics. In dynamic terms, separation of reach and grasp phases may be the outcome of a more fundamental issue in visuomotor control: dividing the total degrees of freedom for the task into two components is a non-optimal solution to a more basic movement issue.121 These results are consistent with high effect sizes for interceptive action using simple two-dimensional displays (force control: dw=1.24) and manual tracking under different levels of advance information (e.g. 1.42 for prediction). Pursuit tracking and catching share important neurocognitive mechanisms. In the case of pursuit tracking, we see strong preferential activation of fast visual channels involving posterior parietal cortex and bilateral cerebellum.189 The discriminating power of simple measures of interceptive action is such that it might be considered a marker for DCD.

Aspects of sensoriperceptual function

This cluster of large effect sizes includes basic visual form detection (dw=2.59) and motion detection (1.34), visuospatial processing both with (1.27) and without (0.83) motor involvement, and tactile perception (1.60). Interestingly, the magnitude of the effects for visuospatial processing was very similar to that reported in the 1998 meta-analysis by Wilson and McKenzie.6 What was intriguing in the current analysis was the very large effect sizes for basic visual processing. This data were drawn from a study by O’Brien et al.39 and involved a clinical sample of only eight children. Notably, their performance was quite homogeneous, perhaps explained by a severe level of motor dysfunction, expected of a clinical group; motor screening data are not available to confirm this, however. Notwithstanding this, it is possible that the visual processing issues we see in DCD can be traced upstream to magnocellular and parvocellular pathways, and their reciprocal connections to predictive and online control networks described earlier. Other studies are needed to fully test this argument.

The very large effect size for tactile perception is equally intriguing and may be associated with a more general problem in developing a body sense or schema (Schoemaker et al.31). Interestingly, within this category, effects were much larger for basic tactile sensory function (e.g. two-point discrimination: individual d values >3.0) than tactual form perception on the Tactual Performance Test (d values <1.0). The reason for this discrepancy remains unclear but may reflect differences in active motor versus non-motor involvement. However, it remains the case that tactile sensitivity is vital for subtle object manipulation, particularly the fast-conducting afferent pathways.190 This information informs and helps update grasp control, and may be a crucial ingredient in the ability to link different action phases.

Other specific areas of very large effect size

Several other task categories revealed very large effect sizes. The result for associated movements (dw=3.00) is consistent with the poor motor synergies that are part and parcel of DCD, even for well-practised skills such as locomotion.153,154 If anything, associated movements might be a useful diagnostic marker if suitably reliable techniques can be used to identify them. This would involve a combination of structured observational checklists and perhaps kinematic algorithms.

Procedural learning (1.50) was shown to yield consistently large effect sizes across studies. This suggests that the apparent contradiction between the study of Wilson et al.50 and Gheysen et al.51 merely reflects use of a low-power test in the former study. Intriguingly, both the basal ganglia and cerebellum are strongly implicated in procedural learning.191 That the cerebellum is implicated in a range of motor control deficits shown in the current review is an important take-home message.

Methodological considerations and future research

No meta-analytical study is immune to publication bias and uncontrolled variables.192 The former refers to the inclusion of effect sizes from published studies only, which may distort results.193 For example, it has been observed that published data tend to yield higher values than those reported in dissertations and unpublished works.157 The counterargument is that the literature search we conducted was comprehensive and spanned nearly 15 years of research. Furthermore, we saw no systematic trend for higher effect sizes among larger studies, which suggests that publication bias is unlikely to be significant in this review.173

Another issue in meta-analysis arises when effect sizes from a variety of study designs are combined in the same analysis.193 The notion of ‘mixing apples with oranges’ suggests that a summary effect may ignore important qualitative differences between individual studies. These differences include features of study design and procedure such as test environment, selection of participants, blind assessment, and so on. We acknowledged some variation across studies on these factors and elected then to use a random-effects model for the analysis of effect sizes. The alternative, fixed-effects, approach assumes that the only source of error in estimating mean effect sizes is that associated with random sampling. The random-effects approach incorporates both this source of error and that associated with other influences, including variation in methodology across studies. While confidence intervals tend to be larger under a random-effects approach for this reason, it did afford us greater generalization of results. In any case, we still observed that most test categories achieved homogeneity of effect sizes. Of the remaining, removal of outliers did achieve homogeneity in a number of instances.

The current study only examined behavioural performance measures. Neuromaturational studies (e.g. MRI, electroencephalography, and electromyography measures), indices of physical fitness (e.g. aerobic capacity), and anthropometric measures such as height and weight were all excluded. However, we referred to recent neuroimaging data to clarify the pattern of performance we saw at a behavioural level. Functional MRI data, for example, were cited to clarify possible mechanisms of dysfunction at a cortical level. Future research should endeavour to measure both behavioural and neural dynamics using parametric techniques and ‘areas-of-interest’ analysis.

Conclusions

  1. Top of page
  2. Abstract
  3. What this paper adds
  4. Method
  5. Results
  6. Discussion
  7. Conclusions
  8. Supporting Information

The meta-analysis presented here explored the pattern of performance deficit in DCD across a range of tasks measuring motor control, learning, and cognition. There was a generalized level of impairment across tasks of large magnitude (dw=0.97). This raises the possibility that the reduced ability to learn motor skills is associated with a baseline level of neuromaturational delay or dysfunction, affecting cortical networks associated with the control of action. However, there were several clusters of even more pronounced deficit in DCD: internal modelling of action (i.e. predictive control), rhythmic coordination and timing, executive function, dynamic control of posture and gait, and interceptive action (catching and manual interception). While seemingly diverse, the pattern of deficits in DCD supports several converging lines of argument about underlying mechanisms. The first argument concerns the (domain-general) concept of predictive control in its broadest sense,e both at the level of real-time control and with respect to learning and utilizing internal models for action (or predictive mapping, more specifically). The difficulties we see with motor prediction under tight temporal constraints are manifest across target-directed reaching, the coupling of grip and load force, anticipatory postural control, and interceptive actions. Problems developing new internal models for action over repeated trials were also evident during perceptual–motor adaptation. And, there is a strong suggestion that difficulties in rhythmic perceptual–motor coupling may also reflect a more fundamental issue in the development of internal maps for action. This argument is notable because it represents a prototypically hybrid view on motor control: the paradigm on which this type of work is based has its roots in the dynamical systems approach, but the interpretation draws on neurocomputational theory. The notion of increased noise within parietocerebellar networks is one possible explanation for deficits in predictive control.

The second converging argument is that a more basic issue in rhythmic coordination and timing is seen consistently, to a point where it is almost synonymous with DCD. It is perhaps then not merely a coincidence that the term ‘coordination’ is part of the diagnostic label according to DSM. At face value, the term suggests that the movement of these children lacks fluency and efficiency: the parts of an action are not put together in the right way, but rather require much effort, even for the simple skills that most children take for granted. This is apparent both to the trained eye of the clinician and to most parents and teachers. At the level of motor control, we dissect the notion of ‘coordination’ further, and see pronounced difficulties with intra- and interlimb coupling and stability. From a dynamic perspective, the control issue may reflect reduced stability of the coupling to an attractor state (Clark J, Whitall J, personal communication, 2011). However, Clark and Whitall suggest that this interpretation does little to clarify the underlying mechanism. In computational terms, children with DCD appear to have a reduced ability to form internal models for action and to use these stored estimates in a predictive manner in order to synchronize to an external entraining signal. Indeed, children with DCD tend to ‘live on feedback’ (Clark J, personal communication, 2011). Dysfunction at a corticocerebellar level is likely to explain this fundamental deficit in timing. Indeed, cerebellar dysfunction sees a constellation of deficits in timing, predictive control, fine motor coordination, and basic cognitive functions.182,194 This fits the profile of deficits that were highlighted in this meta-analysis and suggest, also, important rate-limiting factors in mainstream motor development.195

The third (domain-general) area of deficit was in executive function. Deficits here were evident in most aspects of executive control including working memory, inhibition, and executive attention. Metacognitive aspects of action planning were also affected, suggesting a generalized level of impairment in this area. That the degree of dysfunction parallels or even exceeds that which is seen in attention-deficit–hyperactivity disorder is striking and demands future investigation.

Although it is beyond the scope of this paper to comment on intervention, it is fair to say that effective remediation of DCD should be critically tied to current theory.94 What this review suggests is perhaps two main avenues for intervention research. The first concerns ways of improving predictive control. Simple forms of augmented feedback and motor imagery training may provide starting points for training internal models for action and associated body schema.183 The second avenue for intervention work concerns rhythmic coordination and timing within and between limbs, targeting cerebellar function more specifically. When concurrent augmented feedback is provided in synchrony with voluntary, rhythmic movements, the stability of coordination is often enhanced.106 This modality requires investigation in DCD.

Footnotes
  • a

    Perceptual–motor coupling here refers to rhythmic movements that are entrained to different types of external cue, mainly visual and/or auditory.

  • b

    Domain general refers to those cognitive abilities or processes that influence performance across a range of contexts or domains.

  • c

    Domain specific refers here to discrete areas of performance or skill that are thought to be under the control of more specific cognitive or control functions.

  • d

    Internal models are conceived as being of two types: so-called forward models use a copy of the motor command (namely the efference copy) to predict the future state of the moving limb(s). In contrast, the inverse model (or controller) generates the motor commands necessary to achieve a desired goal state. In computational terms, forward estimates are compared with actual sensory feedback as a way of training both motor prediction and the accuracy of inverse modelling.

  • e

    Prediction is considered part and parcel of most goal-directed activity, from low-level, real-time motor control, to complex actions coordinated over long timescales.

Supporting Information

  1. Top of page
  2. Abstract
  3. What this paper adds
  4. Method
  5. Results
  6. Discussion
  7. Conclusions
  8. Supporting Information
FilenameFormatSizeDescription
dmcn4436_sm_TablesSI-V.pdf267KSupporting info item
dmcn4436_sm_WilsononlineREFERENCES.pdf91KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.