Demographic information for the 40 individuals with stroke and 135 control participants is shown in Table 1. Clinical information of the LA and RA individuals was largely similar although there was a tendency for LA individuals to be more impaired. A greater number of LA participants had low CMSAa scores (score < 3; no voluntary activity) for the affected arms (n = 7) compared to RA participants (n = 4). As well, there were minor non-significant differences in for NIHSS (p = 0.95), FIM (p = 0.95), MOCA (p = 0.55), and BITC (p = 0.50) scores between LA and RA individuals.

Table 1 Demographic Information of control and stroke participants

Demographic and clinical information for healthy controls and individuals with stroke separated into left and right affected stroke. Age, Days since stroke, NIHSS, FIM-M, MoCA, BITC are displayed as mean with the minimum, maximum, and first and third interquartile below. Sex, Handedness, Stroke type, CMSA, Stroke location are shown as values split according to the subtypes identified below:

Sex [Male/Female]

Handedness [Left/Right/Ambidextrous]

Stroke type [Ischemic/Hemorrhagic/Unknown]

CMSA Scores of [1/2/3/4/5/6/7]

Stroke Location [Cortical/Subcortical/Cortical + Subcortical/Cerebellar/Brainstem/Cerebellar + Brainstem/Unknown]

There were roughly equal numbers of LA and RA participants who were able to perform the interception task. However, some individuals with stroke were unable to maintain their hand at the start position with their affected arm, and thus, performance in the task could not be assessed (LA, n = 5, average CMSAa = 2.2; RA, n = 2, average CMSAa = 2). There were less individuals with stroke unable to perform the FFIT than were able to perform voluntary actions (CMSAa score < 3). This is unlikely attributable to time differences between robot-based and clinical assessment which was only 1.7 days. A likely reason for this difference is the gravity support provided by the Kinarm Exoskeleton.

Individual FFIT performanceExemplar performance on no shift trials

Figure 2 displays hand trajectories and corresponding hand speeds for a neurologically healthy exemplar (in black), LA participant (in blue), and RA participant (in red). These plots highlight the speed (outer panels) and location (inner panels) of the participant’s movements. Figure 2A displays performance for the exemplar participants in the No Shift trials which required the individual to simply maintain their hand at the starting position to successfully intercept the target. As expected, the neurologically healthy control participant displayed minimal hand motion. The LA individual also displayed minimal hand motion for these trials. In contrast, hand motion was much larger for the RA individual, demonstrating their difficulty in holding a static position with both their left and right arms.

Fig. 2

Hand speeds and trajectories for an exemplar healthy control and participants with stroke during the FFIT. A Hand speed and trajectories for No Shift trials for a healthy control in black, left affected individual in blue, and right affected individual in red. Hand speeds are plotted with 0ms aligned to perturbation onset and 500ms at trial end represented with a dashed line. Hand trajectories are plotted from perturbation onset (0ms) to time of contact between paddle and target (open squares) or trial end (missed contact; filled squares). B Physical Shift trials for the exemplar participants where perturbation onset in hand speed plots is shown with a solid vertical line. Reaction Time is displayed as open circles. C Visual Shift trials combining both Target Shift and Paddle Shift trials for exemplars. D Task Switch trial performance for exemplars. Two dashed lines are shown to depict the Early and Late variants for this trial type. The Early Task Switch trials end at 700ms and Late end at 550ms

Exemplar performance on physical shift trials

Figure 2B highlights exemplar performance in Physical Shift trials. We found participants performed similarly whether or not visual feedback of the paddle was removed (r = 0.89, p < 0.01), and thus, our analysis grouped these trials together. The load moved the hand away from the central start location and the individual quickly returned their hand to the center to intercept the moving target. Speed profiles and hand trajectories were highly consistent across trials for the healthy control. The LA participant was able to generate responses but performed generally worse than the control participant. Their hand speeds and trajectories show consistent movements, but relative to the control, they displayed a wider distribution of Reaction Times and Endpoint Distances using both affected and unaffected arms. The RA exemplar showed large trial-to-trial variability in Physical Shift trials for both arms. Although they clearly generated responses to the mechanical perturbations, they were highly variable across the trials resulting in inconsistent Reaction Times and Endpoint Distances.

Exemplar performance on visual shift trials

We found participants with stroke performed similarly for Target and Paddle shift trials (r = 0.83, p < 0.01), and thus, these two trial types were grouped into one trial type termed Visual Shift trials. Visual Shift trials for the exemplars are shown in Fig. 2C. The control participant demonstrated distinct goal-directed movements and consistent Reaction Times and Endpoint Distances. The LA individual’s responses were less consistent than the exemplar control which was demonstrated by greater variability in Reaction Time and Endpoint Distance. The RA individual had clear difficulties responding to Visual Shift trials. In both arms, left and rightward movements are difficult to distinguish and their Reaction Times and Endpoint Distance were highly variable.

Exemplar performance on task switch trials

Finally, Task Switch trials are displayed in Fig. 2D. As the objective was to avoid the target, successful trials required hand movements away from the central location. The control participant was able to avoid the target for most trials. One can see that the control participant had greater Reaction Time variability in Task Switch trials than other trial types, highlighting the difficulty to abort the ongoing motor action of intercepting the moving target. The LA individual had difficulty responding to the stimulus change in either arm as they were unable to avoid the target in most trials. The RA individual also performed poorly for most trials, although their success in a low proportion of trials may be due to their difficulty in holding posture rather than their ability to rapidly alter their motor plan to avoid red targets.

Group performanceImpairments associated with LA and RA individuals

Table 2 displays the percentage of individuals with stroke that were impaired in Reaction Time and Endpoint Distance after controlling for age, sex, and handedness (individuals unable to perform the task were identified as impaired in each measure; LA, n = 7; RA, n = 2). A large proportion of LA individuals were impaired on Reaction Time. Using the affected arm, 85% of all LA individuals were impaired in at least one trial type with 62% impaired in Physical Shift, 69% impaired in Visual Shift, and 62% impaired in Task Switch trials. Many RA individuals, using their affected arm, were identified as impaired with 83% impaired across all trials, 67% in Physical Shift, 78% in Visual Shift, and 56% in Task Switch trials. Interestingly, RT between the LA and RA groups were not significantly different from each other (Physical Shift, p = 0.25; Visual Shift, p = 0.72; Task Switch, p = 0.82).

Table 2 Impairment for transformed FFIT parameters for LA and RA participants

Importantly, impairments were commonly observed in the ‘unaffected’ arm despite only 12.5% of individuals with stroke scoring < 6 on the CMSAa (full range of motion with near normal timing and coordination). We found 60% of LA individuals were impaired in their unaffected arm, with 40% identified as impaired in Physical Shift trials, 55% in Visual Shift trials, and 45% in Task Switch trials. For RA individuals, 55% were impaired using their unaffected arm, including 15% in Physical Shift trials, 40% in Visual Shift trials, and 40% in Task Switch trials. Unaffected arm RT for LA and RA individuals were not significantly different (Physical Shift, p = 0.06; Visual Shift, p = 0.50; Task Switch, p = 0.50).

On some trials, we did not identify a Reaction Time following the perturbation. In these trials we set Reaction Time to the maximum allotted time (1000ms). For LA individuals, the mean percentage of trials set to 1000ms was 23% (range = 2–46%) for the affected arm, and 11% (range 1–39%) for the unaffected arm. For RA individuals, the mean for the affected arm was 15% (range = 0–60%) and the unaffected arm was 4% (range = 0–10%). Mechanical disturbances were associated with the greatest percentage of trials without an identified Reaction Time (mean = 37% across all individuals and arms). In comparison, we did not identify a Reaction Time for only 3% of trials for healthy control participants.

Much like Reaction Time, impairments were commonly observed for Endpoint Distance across all trial types and for both the affected and unaffected arms. We identified 77% of LA participants with impaired Endpoint Distance in their affected arm for at least one trial type (69% in Physical Shift, 69% in Visual Shift, and 31% in Task Switch trials). For RA individuals, 83% were impaired in Endpoint Distance for the affected arm for at least one trial type (78% in Physical Shift trials, 72% in Visual Shift trials, and 39% in Task Switch trials). The Endpoint Distance for the affected arm of LA and RA groups were not significantly different (Physical Shift, p = 0.51; Visual Shift, p = 0.57; Task Switch, p = 0.88).

Impairment in Endpoint Distance was also common in the unaffected arms. For LA individuals, 75% were impaired in at least one trial type, with 55% in Physical Shift, 70% in Visual Shift, and 35% in Task Switch trials. For RA individuals, 50% were identified as impaired across all trial types, with 30% impaired in Physical Shift, 40% in Visual Shift, and 20% in Task Switch trials. Only Task Switch trials were significantly different between LA and RA individuals using their unaffected arm (Physical Shift, p = 0.13; Task Switch, p = 0.28; Task Switch, p < 0.05).

Comparison of impairments associated with the affected and unaffected arms

Figure 3 displays the untransformed Reaction Time and Endpoint Distance across our cohort of 135 controls and 40 individuals with stroke. Reaction Time performance is plotted in Fig. 3A. For control performance, one can see an increase in Reaction Time where Physical Shift trials elicit the fastest Reaction Time, Visual Shift trials elicit slower times, and Task Switch trials had the slowest times. For LA individuals, Physical Shift trials displayed a wide range of impairment with some individuals reaching the maximum allotted Reaction Time. The affected arm was significantly more impaired than the unaffected arm for Physical Shift trials (p < 0.01) but not for Visual Shift or Task Switch trials. The bottom panel displays Reaction Time of controls and RA individuals. These individuals displayed significantly greater impairment in RT for the affected rather than unaffected arm for Physical (p < 0.01), Visual (p < 0.05), and Task Switch (p < 0.05) trials.

Fig. 3

Group level Reaction Time and Endpoint Distance performance across trial types. A The Reaction Time of LA individuals (blue; top panel) for both their left (filled left-directed triangles) and right arms (unfilled right-directed triangles). Individuals that failed the BITC (presenting with visuospatial neglect) are identified as square icons. The healthy control performance in each trial type is plotted as a grey rectangle where the top and bottom of the rectangle represent the 95 and 5 percentiles of Reaction Time, respectively. Individuals with stroke were identified as impaired if their Reaction Time was greater than 95% of controls. The bottom panel presents the Reaction Time of RA individuals (red) with the left arm performance as unfilled left-directed triangles and the right arm as filled right-directed triangles. B Endpoint Distance of healthy controls, LA, and RA individuals in the same format as (A). Impairment in Endpoint Distance was identified as being greater than 95% of controls for No Shift, Physical Shift, and Visual Shift trials. Impaired Endpoint Distance for Task Switch trials was characterized by Endpoint Distance less than 5% of controls

Endpoint Distance is displayed in Fig. 3B. We identified 4 LA individuals impaired in No Shift trials, using their affected arm, highlighting an impairment in maintaining the hand at a spatial location. Of note, Endpoint Distance in Visual Shift trials was commonly impaired as most LA individuals had an Endpoint Distance near 6cm which is the maximum distance between paddle and target after the visual perturbation. Endpoint Distance was significantly more impaired for the affected arm compared to the unaffected arm for Physical (p < 0.01) and Visual Shifts (p < 0.01). The RA individuals in the bottom panel performed similarly to the LA individuals with significantly more impairment for all trial types using the affected arm rather than unaffected arm, except for No Shift trials (Physical, p < 0.01; Visual, p < 0.01; Task Switch, p < 0.05).

Impairments associated with different sensory modalities

To identify impairments specific to one sensory system, we compared Reaction Time between different trial types (Fig. 4). The left panel of Fig. 4A displays the Reaction Time impairment for LA and RA individuals in Physical and Visual Shift trials. In both Physical and Visual Shift trials, 15 (71%) and 8 (38%) LA individuals were using their affected and unaffected arms, respectively. In contrast, 10 (53%) RA individuals were impaired in both trial types in the affected arm, whereas only 3 (16%) were impaired in the unaffected arm. There were 3 LA individuals and 4 RA individuals displaying Reaction Time impairment in visual but not proprioceptive feedback using their unaffected arm. The right panel of Fig. 4A compared Visual Shift and Task Switch trials. Again, most individuals were impaired in both trial types and only a few individuals impaired in only Visual Shift or Task Switch trials. Figure 4B highlights the same comparisons as in Fig. 4A but for Endpoint Distance. For both left and right panels, performance in Endpoint Distance also displayed similar patterns of impairments with the majority impaired in both trial types. This reflects the high correlation between Endpoint Distance and Reaction Time in each trial type (Physical Shift: r = 0.95, p < 0.01; Visual Shift: r = 0.91, p < 0.01; Task Switch: r = − 0.77, p < 0.01). Interestingly, there were many individuals with Endpoint Distance impairment in only Visual Shift trials and not Task Switch Trials (LA, n = 5; RA, n = 7).

Fig. 4

Group level comparison between trial types. A Comparison of trial types for Reaction Time. The left panel displays Reaction Time for the Physical Shift and Visual Shift trial types with triangle colour and direction denoting affected arm and assessed arm, respectively, in the same format as Fig. 3. The healthy control region is displayed as a grey box with dashed lines representing the 95 percentile of control performance. The right panel shows the comparison between Visual Shift and Task Switch trials. B Trial type comparison for Endpoint Distance. The left panel compares Endpoint Distance for Physical Shift and Visual Shift trials in the same format as the left panel in (A). The right panel compares Visual Shift and Task Switch trials where impaired Visual Shift Endpoint Distance is greater than 95% of controls but impairment for Task Switch trials was identified as performance less than 5% of controls

Comparison of FFIT performance and clinical measures

To test the relationship between FFIT performance and the clinical measure scores of individuals with stroke, we performed Spearman rank-order correlations. For LA individuals, we found moderate correlations between Reaction Time impairment and CMSAa scores for Physical Shift (r =− 0.70, p < 0.05 RA participants had Reaction Time impairment that was moderately correlated with CMSAa scores for Visual Shift trials (r = − 0.53, p < 0.05) and Task Switch trials (r =− 0.52, p < 0.05). We identified no significant correlations between Reaction Time impairment with the MoCA or FIM-M and only near significant correlations with the BITC (Visual Shift trials: r =− 0.67, p = 0.055).

Endpoint Distance impairment was significantly correlated with the CMSAa. For LA individuals, Physical (r = − 0.69, p < 0.01) and Visual Shift (r = − 0.62, p < 0.05) trials had moderate correlations between Endpoint Distance and CMSAa. RA individuals only had moderately significant correlations between CMSAa and Physical Shift trials (r = − 0.60, p < 0.05). Further, we noted 4 individuals with stroke had CMSAa scores of 6 and 7 (full range of motion with near normal timing and coordination) but were still identified as impaired in Reaction Time and Endpoint Distance.