The independence of deficits in position sense and visually guided reaching following stroke

Robotic device

The robotic assessment was performed using the KINARM exoskeleton robotic device (BKIN Technologies Ltd., Kingston, Ontario) [35]. Subjects sat in a modified wheelchair base with each arm supported in the horizontal plane by the robotic exoskeleton. The device permits arm movements in the horizontal plane, monitors shoulder and elbow motion and can apply mechanical loads at the shoulder and/or elbow. Subjects were allowed free head movement and viewed an augmented reality system that could display visual targets in the same plane of the arms. Details of the tasks and analyses used in this study have been previously described [24, 25].

Assessment of position sense: Arm-position matching task

With the arms and hands occluded from vision, the robot passively moved one hand (passive arm) to one of nine spatial locations separated by 10 cm (Figure 1A,B). When the robot completed the passive movement, subjects actively moved their opposite hand (active arm) in an attempt to mirror match the passive hand position. Subjects were permitted as much time as necessary to match the passive hand location before cuing the experimenter to trigger the next trial. Each subject completed 6 randomized blocks for a total of 54 trials. For subjects with stroke, the robot passively moved the arm on their affected side. Controls were tested on both arms.

Three parameters quantified position sense: 1) trial-to-trial Variability of the active hand (Var_xy), 2) Contraction/Expansion of the spatial area enclosed by the active hand relative to the spatial area enclosed by the passive hand (Area_xy), and 3) Systematic Shifts between the passive and active hands (Shift_xy). The formulae for these parameters have been previously described and provide reliable, 2-dimensional quantification of common deficits in position sense, and may describe different physiologic/functional attributes of limb position sense [24]. For this study, Contraction/Expansion was normalized by control performance to create positive scores for both contraction and expansion. Values between 0 and 1 indicated performance within the 95% confidence bounds of non-disabled controls, whereas values greater than 1 indicated abnormal contraction or expansion.

Assessment of motor performance: visually guided reaching task

With full vision of the arms and hands, subjects were instructed to reach “as quickly and accurately as possible” from a central target (near the centre of the arm’s workspace) to one of eight peripheral targets located 10 cm away (Figure 1C,D). The robot provided no active resistance or assistance for the reaching movements. Each trial began with the subjects holding their index finger tip at the central target for 1250–1750 ms before the peripheral target was illuminated thereby signaling to initiate their reaching movement. Subjects were provided 3000 ms to complete each reach. Each subject completed 8 randomized blocks. Two catch trials, in which no peripheral target was displayed, were included in each block. In total, each subject completed 80 trials with each arm.

We report five movement parameters, one from each of five attributes of sensorimotor control: a) postural control, b) reaction time, c) initial movement d) corrective movements, and e) total movement metrics. The algorithms for determining all parameters have been previously described in detail [25]. Postural control was characterized by postural hand speed (PS) for the 500ms prior to peripheral target illumination. Reaction time (RT) was the time from peripheral target onset until movement onset. The initial phases of movement from movement onset to the first speed minimum were quantified using initial movement direction error (IDE). IDE is the angular deviation between a straight line from the central to peripheral target and the actual path taken in the initial phase of movement (first minima of hand speed after movement onset). Corrective movements after their initial response were quantified by recording the number of speed peaks per movement (NSP). The total movement metric used was total movement time (MT). These parameters provided the best combination of high reliability and sensitivity (ie. the ability to detect the presence of an impairment relative to normal behaviour) for each attribute [25].

Clinical assessment

The clinical assessment consisted of a broad range of sensory, motor, and functional evaluations. These measures were chosen as the majority of them are used in day-to-day practice in our stroke rehabilitation centres.

Handedness was documented with the Modified Edinburgh Handedness inventory [36]. Subjects were considered Right handed (R) if they scored >50, Left handed (L) if they scored <−50 and Mixed handedness (M) if they scored >−50 and <50.

Position sense was assessed clinically with the Thumb Localizer Test (TLT) [16], chosen because of its previous use quantifying whole-limb position sense in several studies of subjects with stroke [14, 37–43]. As described by Hirayama [16], the subject, with eyes closed, extends one thumb and the examiner moves that arm to a position in front of the subject at or above eye level, lateral to the midline. The subject is then asked to pinch the extended thumb with the opposite thumb and forefinger. Subjects are scored 0 (accurately does the task) to 3 (subject is unable to find his or her thumb and does not climb up the affected arm to locate it). Subjects with stroke localized the thumb on their affected arm whereas controls localized the thumb on their dominant arm.

Spasticity was assessed using the Modified Ashworth Scale [44]. In this assessment, the examiner moves the subject’s limb through a range of motion and evaluates the resistance to passive movement. Scores range from 0 (no increase in muscle tone) to 4 (affected part is rigid in flexion or extension). In the present manuscript we report the MAS for the elbow flexors.

Motor impairment was assessed using the Purdue Pegboard (PPB) (LaFayette Instrument Co., LaFayette, IN, USA) [45]. In this assessment the subject attempts to place as many pegs as possible in a board in a 30 second time window with one hand. While this is traditionally touted as a test of fine motor skills, the subject is required to use the proximal upper extremity to get the hand to the correct position to retrieve and insert each peg.

The Chedoke-McMaster Stroke Assessment - Impairment Inventory (CMSA) [46] was performed to stage the arm and hand. This tool relies on the concept of stages of motor recovery, first introduced by Twitchell [1], whereby a patient is thought to progress through a sequence of stages from flaccid paralysis (Stage 1) to normal movement (Stage 7) as they recover from a stroke. Individuals are scored based on their ability to perform a set of predetermined movements in front of an examiner. The authors of the tool indicate that the “Stages of Motor Recovery” serve as a measurement for the amount of neurological impairment.

Functional abilities were documented with the Functional Independence Measure (FIM) [47]. This 18 item scale scores individuals across a number of items. There are 13 items considered to be motor tasks (eating, grooming, bathing, upper and lower body dressing, toileting, bladder and bowel management, bed to chair transfers, toilet and shower transfers, locomotion and the ability to do stairs). There are 5 items considered to be cognitive tasks (cognitive comprehension, expression, social interaction, problem solving and memory). Each item is ranked on a 7-point ordinal scale that ranges from 1 (total dependence) to 7 (complete independence). In the present manuscript we present the total FIM score (measured out of 126) and the FIM self-care subscore (FIMsc – measured out of 42, items include: eating, grooming, bathing, dressing above the waist, dressing below the waist and toileting).

Visual acuity was assessed with a Snellen eye chart and visual fields were examined using confrontation [48]. Visuospatial attention was assessed using the conventional subset of the Behavioural Inattention Task (BIT) [49]. The BIT is a standardized assessment for visuospatial inattention (also known as hemispatial neglect) with a number of pencil and paper tests including line, letter and star cancellation, line bisection, figure copying and drawing.

Stroke lesion locations were recorded by one of the study physicians with experience in neuroimaging (S.P.D). Clinical computed tomography (CT) or magnetic resonance imaging (MRI) scans and the clinical reports generated by a neuroradiologist were evaluated. From this a standard form was used to document the side of the lesion, the vascular territory (eg. right middle cerebral artery), and brain region (egs. cortical, subcortical, brainstem, cerebellum).

Non-disabled controls completed the same assessments as subjects with stroke with a few exceptions. No control subject performed the FIM or BIT. A subset of non-disabled controls performed the TLT, MAS and CMSA to obtain some idea about the amount of variability amongst controls in these clinical measures.

A trained study physician or a physiotherapist with expertise in stroke rehabilitation performed all clinical assessments.

Statistical comparisons

Statistical Analyses were performed using MATLAB (Mathworks, Inc., Massachusetts, USA). Subjects with stroke were flagged for impairments on robotic parameters when their performance fell outside a parameter’s normative reference range, defined as the 95% confidence boundaries of control performance taking age [50–53], sex [54–56], and test-arm (dominant/non-dominant) [57–62] into consideration. Our primary analysis was concerned with the performance of the most affected arm in subjects with stroke. To establish each parameter’s normative reference range, control values were tested for normality (Lillifors test, P< 0.01) and transformed with a log, square root, or inverse transform to obtain normality when necessary. Regression analyses were then used to model the effects of age, gender and test-arm when creating normative reference ranges for each stroke subject.

Impairments on the parameters from each task were used to classify whether a subject had performed “normally” or “abnormally” on either the arm position matching task, the visually-guided reaching task, both or neither. Fisher’s exact probability tests and Pearson’s correlation tests were used to quantify relationships between robotic and clinical measures. The p value of significance on these tests was set at p < 0.01. Bonferroni corrections were applied to tests with multiple comparisons.

Individual subject exemplars

Figure 1A,B shows arm position matching results of a healthy 67 year old female control subject. The robot passively moved the left arm and the subject actively moved the right arm to the mirror location in space. As with most non-disabled controls, variability about each target is relatively small (< 5 cm) and there is little spatial contraction/expansion or systematic shift. Figure 1C,D depicts reaching performance of the same subject. The modest curvatures in the hand-paths (Figure 1C) and moderate variability in the velocity profiles (Figure 1D) are typical for a non-disabled control subject of this age.

Figure 2 illustrates the performance of subjects with stroke in the matching and reaching tasks. These exemplar subjects illustrate that statistically established deficits in the matching task are not necessarily coupled with significant deficits in the reaching task, and vice versa.

Figure 2A,B displays two exemplar subjects with stroke whose performance was not statistically different from non-disabled controls. Figure 2A shows a right-affected subject with an ischemic stroke of the left pontine artery 28 days prior to robotic testing. A right-affected subject who experienced a hemorrhagic stroke of the left basilar artery 29 days before robotic assessment is illustrated in Figure 2B. Despite their different stroke types, their reaching and matching performance is qualitatively and quantitatively similar to controls.

Other subjects with stroke performed abnormally on the matching task but normally on the reaching task (Figure 2C,D). Figure 2C depicts a right-affected subject with an ischemic stroke of the left anterior cerebral artery 9 days prior to robotic testing. This subject demonstrates a significant systematic shift in the matching task, yet is unimpaired on the reaching task. Figure 2D depicts data from a left-affected subject who had an ischemic stroke of the right middle cerebral artery (MCA) 59 days before robotic examination. This subject exhibits significantly increased variability in the matching task, but normal reaching behaviour with the affected left arm.

Another group of subjects with stroke performed normally on the matching task, but abnormally on the reaching task (Figure 2E,F). Figure 2E displays a left-affected subject with an ischemic stroke of the left pontine artery 24 days before robotic testing. Figure 2F illustrates a left-affected subject with an ischemic stroke of the right MCA 24 days prior to robotic assessment. These subjects perform qualitatively and quantitatively similar to controls in the matching task. However, there are clear abnormalities in the hand paths (middle) and multiple peaks in the velocity profile (right) consistent with deficits in reaching.

A final group of subjects with stroke performed abnormally on both robotic tasks. Figure 2G depicts this pattern in a subject with an ischemic left MCA stroke 15 days prior to robotic examination. Figure 2H shows a subject with an ischemic right PCA stroke 19 days prior to robotic assessment.

Stroke group data

A more detailed demonstration of this independence is seen by examining the categorical relationships (Fisher’s exact probability tests) and correlation coefficients (Pearson’s correlations) of individual parameters within and between the tasks (Table 4). In the matching task, Fisher’s tests demonstrated significant categorical relationships between Variability, Contraction/Expansion, and Systematic Shifts (All P < 0.0014). Additionally, Pearson’s correlations revealed significant relationships between Variability and Shift (P < 0.0014) with a correlation of 0.36. Figure 3A illustrates the relationship between Variability and Systematic Shifts. Far more subjects failed (diamonds) or passed (circles) on both parameters than expected by random chance, whereas far fewer subjects had deficits in only one of the two parameters (squares, triangles).

	Varxy	Areaxy	Shiftxy	PS	RT	IDE	NSP	MT
Var xy		0.18	†0.36	0.01	0.29	0.13	0.21	0.19
Area xy	†<10 -5		0.13	0.23	-0.02	0.12	0.09	0.01
Shift xy	†<10 -3	†<10 -3		0.05	0.14	0.28	0.27	0.28
PS	0.301	0.024	†<10 -3		–0.04	0.12	0.03	–0.23
RT	0.004	0.128	0.003	0.014		†0.36	0.28	†0.36
IDE	0.031	0.326	0.024	0.069	†<10 -3		†0.62	†0.67
NSP	0.006	0.080	0.087	0.378	0.027	†<10 -5		†0.78
MT	0.047	0.341	0.033	0.119	†<10 -3	†<10 -8	†<10 -8

Lower left: Probabilities of independence between categorical contingencies (Fisher’s exact probability tests). Upper right: Correlations between measures (Pearson coefficients). †Significant contingencies and correlations, P < 0.0014 (Bonferroni’s method is used to control for multiple comparisons; α=.01, n = 7).

Within the reaching task, a similar trend was observed across parameters except for relationships with PS (Table 4). The relationship between the parameters MT and IDE is shown in Figure 3B. Similar to the matching task, more subjects exhibited deficits (diamonds) or were normal (circles) on both parameters and fewer subjects showed deficits in only one of the two parameters (squares, triangles) than would be expected by chance alone (Fisher’s test, P < 0.0014). Additionally, there was a significant Pearson’s correlation between these two parameters (r = 0.67, P < 0.0014). Except for the weak relationship between RT and NSP, all other relationships between RT, MT, IDE, and NSP showed statistical categorical results (Fisher’s tests, P < 0.0014).

In contrast to the relationships within each task, there was substantively less statistical interactions between the matching and reaching tasks (Table 4). The relationship between Variability (matching) and Movement Time (reaching) is illustrated in Figure 3C. The number of subjects identified as performing abnormally on both parameters, one only or neither was essentially randomly distributed (Fisher’s test, P = 0.047) and the Pearson’s correlation was not significant (r = 0.19, P > 0.0014). In fact, only one pair of reaching and matching parameters, Shift_xy and PS, exhibited a significant categorical relationship (Fisher’s test, P < 0.0014) and none of the matching and reaching parameters demonstrated a significant Pearson’s correlation (all P > 0.0014).

Effect of visuomotor impairment of the ipsilesional arm

A small number of subjects with stroke in the present study (n = 9) demonstrated impairment in the ipsilesional (less affected) arm that caused them to have difficulty getting to visually guided reaching targets in more than 5% of trials. As part of a secondary analysis, we removed these individuals from the data set and again compared the relationship between the visually guided reaching task and the position matching task. We observed that deficits in arm position matching were still largely independent of those in visually guided reaching (Fisher’s test, P = 0.22).

Relationship between robotic assessment and clinical measures of impairment and disability

Relationship between lesion location and performance on robotic measures

Taking into account the side, vascular territory and the cortical or sub-cortical nature of the stroke lesions, we assessed the lesions to determine whether certain locations might be more commonly associated with deficits in the two tasks. A large cohort of our subjects with stroke had cortical or sub-cortical lesions of the MCA territory (29 right and 31 left). Subjects with identified deficits for reach only, match only and both tasks were 4, 3 and 21 for right MCA lesions and 13, 0 and 17 for left MCA respectively. More subjects displayed deficits in reach only with left MCA lesions compared to right MCA lesions, the reverse was true for match only deficits. However, these differences did not reach statistical difference (Fisher’s test, p = 0.98). As well, the distribution of deficits across the two tasks for subjects with lesions in the MCA territory were also not different than the distributions observed for the remaining subjects with lesions elsewhere in the brain (Fisher’s test, p = 0.11).