Speed-Biased Training Temporarily Improves Motor Performance of the Paretic Arm Compared to Accuracy-Biased Training in Chronic Stroke Survivors: The Phase 1 FAST Randomized Clinical Trial

Abstract

Background

Despite the prevalence of upper extremity (UE) limitations after stroke, few training interventions prioritize fast movement speeds during rehabilitation.

Objectives

To compare the effects of an equivalent dose (in the number of trials) of speed versus accuracy training in chronic stroke with mild-to-moderate impairments who have no direct cerebellar damage.

Patients and Methods

In this Phase-1 randomized controlled trial, we randomized 42 participants to either a speed or an accuracy arm-movement training condition. Participants moved their paretic hands through complex tracks, with 2080 trials in 4 sessions within a week. Speed and accuracy were manipulated by displaying 5 cm-wide or 1.25 cm-wide tracks or providing feedback based on average speed and accuracy in the Speed and Accuracy groups, respectively. We measured changes in kinematics in a 3-target test, in the speed-accuracy trade-off in a modified Fitts’ test of the paretic arm during goal-directed reaching, and clinical outcomes (ie, UE Fugl-Meyer, Action Research Arm Test, and Box and Block Test) at 3 days and 1-month post-training.

Results

Speed training led to significantly faster and smoother movements with more symmetric reach velocity profiles at the 3-day post-test, consistent with better feedforward control. Speed training temporarily improved the speed-accuracy tradeoff. At 1 month, however, most gains in the 3-target test and in the modified Fitts’ test were lost.

Conclusion

Speed training led to greater gains in kinematics of goal-directed actions than accuracy training, notably in a 3-day post-test. Our results suggest that training programs with high repetitions of fast movements may improve paretic arm reaching performance. The trial is registered at ClinicalTrials.gov under ID NCT05013762.

Keywords

chronic stroke rehabilitation upper extremity kinematics speed-accuracy tradeoff

Introduction

Paretic arm performance after stroke is often slow and fragmented; however, few training interventions prioritize faster movement speeds during rehabilitation. We contend that a potential reason for the lack of superiority of previous interventions compared to respective control groups^1-4 is that movements during training were performed at insufficient speeds. In contrast, speed training, in which the participants repeatedly generate fast goal-directed movements, holds promise for retraining upper extremity (UE) movements in chronic stroke individuals.

In rapid-reaching movements, feedforward control generates the initial motor command impulse that rapidly moves the hand toward the target.⁵ In the feedback error learning theory,⁶ feedback controllers simultaneously control movements and provide error signals to calibrate feedforward controllers.⁶ Thus, fast, but not slow, movements generate large feedback errors that improve feedforward controllers. Individuals with chronic stroke demonstrate deficits in feedforward control,^7-11 and increased reliance on feedback control, with reaching movements characterized by reduced speeds, prolonged deceleration phases, and multiple peaks in velocity profiles.^11,12-16 Because relearning lost motor functions in chronic stroke is thought to parallel motor learning in neuro-typical participants¹⁷ (at least when the structures for motor learning are spared by the lesion), we predicted that speed training would be effective in restoring feedforward control of UE movements post-stroke.

We previously investigated the effect of 2 sessions of speed training with the paretic arm in individuals with chronic stroke and mild to moderate impairments.^18,19 In Park et al,¹⁹ following 1200 fast-reach movements, training gains generalized over a large workspace, with significant and durable (1 month) movement time and smoothness improvements. In Kantak et al,¹⁸ 300 training trials on a complex tracking task led to significant and durable (1 month) improvements in the speed-accuracy trade-off.

Because these 2 studies did not include a control group, it is unclear whether the improvements were due to fast movements or the large number of movements. Here, using training tracks similar to Kantak et al,¹⁸ we conducted a Phase-1 FAST randomized controlled trial by randomizing participants with chronic stroke, and moderate to mild impairments to either a Speed training condition or an Accuracy training condition. Participants in both groups performed 520 trials per session, with 4 sessions within a week. In the Speed condition, the track width was wider (5 cm), and feedback rewarded short movement times. In the Accuracy condition, the tracks were narrower (1.25 cm), and feedback rewarded accuracy. Kinematic testing on a reaching transfer task and a speed-accuracy trade-off Fitts’ task, as well as clinical assessments (UE Fugl Meyer [UEFM], Action Research Arm Test [ARAT], and Box and Block test), were performed before training (pre-test) and after 3 days and 1 month following training.

We hypothesized that following speed training, reaching movements in the reaching transfer task would be more compatible with improved feedforward control: the movements would be faster, smoother, and exhibit more symmetrical velocity profiles. In addition, we expected that speed training would lead to a larger reduction in Fitt’s slope, that is, in the rate of change in movement time as a function of target difficulty. Finally, we also expected greater improvements in secondary clinical outcome measures for the Speed group.

Methods

Participants

The inclusion criteria were: a unilateral hemorrhagic or ischemic stroke that spared the cerebellum over 180 days before enrolment; mild to moderate upper-extremity impairments measured by the UEFM scale (range = 21-57)²⁰; ability to slide the more affected hand 25cm on a table within 5 sec without assistance and with trunk constrained; no additional neurological diagnoses; and no magnetic resonance imaging (MRI) contraindications. The exclusion criteria were inability to follow a 2-step command, hemi-spatial neglect (>4% of lines left uncrossed on Albert’s line test), severe proprioceptive impairment (0 on part H of the UEFM score), and severe pain or an upper-extremity orthopedic disorder. Recruitment was limited to unilateral strokes (possibly more than 1, but on the same side). All participants signed an informed consent approved by the Internal Review Board of Casa Colina Hospital before participation (IRB00002372). The trial, registered at ClinicalTrials.gov (ID NCT05013762 on 08/20/2021), enrolled the first participant on 08/23/2021.

Study Design

Participants visited the laboratory 9 times over 6 to 8 weeks (Figure 1A). Participants were tested for inclusion and exclusion criteria on Visit 1. Within 2 weeks of Visit 1, participants performed baseline kinematic assessments of the ipsilesional, less-affected (LA), and the contralesional, more-affected arms during Visits 2 and 3, respectively (pre-test). Hand trajectory kinematics were assessed during 2-dimensional planar reaching movements (see below for details). Visits 2 and 3 also included additional clinical assessments (ARAT, Box and Block, and Modified Ashworth Scale), and an MRI scan to confirm there was no evidence of a primary cerebellar stroke. Within 2 weeks following the third visit, participants underwent 4 training sessions within a week in either speed or accuracy conditions using the same apparatus as for testing. Participants performed the kinematic test and the clinical assessments of the more affected arm again 3 days (visit 8) and 1 month (visit 9) post-training.

Figure 1.

Design. Study timeline (A), experimental setup (overhead perspective) (B), examples of hand path trajectory for the Speed training (C) and Accuracy training conditions (D). The blue line shows the hand trajectory in 1 trial for a representative participant in the Speed group (panel A, Participant 1. UEFM 47), and for a representative participant in the Accuracy group (Panel B Participant 12, UEFM 36). Note that the tracks do not have the same lengths because they are scaled to the participants’ maximum reach (see Methods Section). A. Study timeline. B. Overview of the set-up. C. Example of a track and trajectory in the Speed condition. D. Example of a track and trajectory in the Accuracy condition.

During each training session, participants completed 8 training blocks consisting of 65 trials (520 trials) of tracking a complex task (see below). A magnetic sensor (3D Guidance Model 800 Sensor) was placed on top of the index fingernail of the more affected hand. Sensor data were recorded at 255 Hz.

Participants were randomly assigned to each group via an adaptive randomization procedure, as in Woytowicz et al,²¹ that aimed to balance age, sex, time since stroke, stroke lesion side, and average movement time with the more affected arm. Both the testers/assessors and participants were blinded to group assignment.

All experimenters were experienced occupational or physical therapists with 1 year or more of experience in neurorehabilitation. Assessors involved with the UEFM were all trained and certified according to best practice.²²

Training Protocol

Training Methods Common to Both Groups

Participants sat in a chair and in front of a table with a restraining belt to minimize trunk movements. The chair’s height was adjusted to align the tabletop with the xiphoid process. Wrist movements were restrained via a wrist brace, and finger movements via a splint. To reduce friction, the hand was covered with a tubular bandage (see Figure 1B).

At each trial, 1 of 8 possible track patterns (4 originals tracks and 4 flipped along the y-axis) was projected on the table from an overhead projector. Participants were instructed to slide their hands horizontally on the tabletop, moving their index fingers along the displayed track as rapidly and accurately as possible. With wrist and fingers splinted, actions predominantly involved shoulder and elbow movements. The tracks’ start box was at 20 cm in front of the xiphoid process. The tracks’ lengths were adjusted to 85% of the individual participant’s maximal paretic arm reach along an oblique 45° path in the transverse plane. Consequently, some participants had shorter tracks than others (based on both impairments and anatomy). Because all participants could complete all training trials (except for 1 participant on day 1 of training; see Results), this challenge level seemed reasonable.

A trial was successful if completed within 10 seconds and with a trajectory at least 85% of the trajectory within the track. An auditory cue sound was played after the hand stayed 0.1 seconds in the finish box to ensure a complete stop at the target. After each trial, a feedback score (depending on the group; see below) and a corresponding feedback message cue (out of 5 possible cues) were displayed based on the comparison of the current trial performance with previous trials. Specifically, performance on the last 20 successful trials was quantized into 5 ranges (20% percentiles), and a post-trial feedback message based on these ranges was provided (see detailed methods and feedback cues in Park et al¹⁹) A total score (sum of the trial scores) was also provided.

Methods Specific to Each Training Group

In the Speed group, participants were instructed to follow 5-cm wide tracks (see Figure 1D). The trial-by-trial and total feedback scores and feedback cues were based on movement time, computed from when the finger left the start box until it reached the finish box and stayed in it for 0.1 seconds. The trajectory was displayed at the end of the trial for 3 seconds.

In the Accuracy group, the width of the tracks was 1.25 cm (see Figure 1C). The trial-by-trial and total feedback scores and feedback cues were based on the within-track error, which was computed based on the percentage of samples within the boundary of the track (in-track percentage). The trajectory was displayed in real-time to improve accuracy during movement and at the end of the trial for 3 seconds.

Movement Assessments

Transfer of Performance With the 3-Target Reach Test

Following 32 familiarization trials, participants performed 10 reaches to each of 3 targets as fast and accurately as possible. The targets (black circle of 4 cm diameter) were located at 45°, 90°, and 135° on an arc of 20 cm radius centered on the starting position (green circle 5cm diameter).

Transfer of Speed and Accuracy Trade-Off With a Fitts’ Test

We tested whether the increase in speed-accuracy trade-off during training transfers to reaching movements with a modified Fitts’ task, similar to McCrea and Eng.¹² Participants were presented with 1 of 9 targets at 90° from the start. Three target distances, 10, 20, and 25 cm, and 3 diameters, 2, 4, and 8 cm, resulted in 9 possible targets, with 6 trials per target.

Data Processing and Analyses

All sensor data were filtered with a Butterworth filter with a 10 Hz cut-off frequency. For all training trials, we computed movement time, MT (from cursor leaving the start box to entry in the finish box), mean speed (trajectory length divided by MT), mean error (mean trajectory deviation from the track center), and training speed-accuracy tradeoff (training SAT, mean speed divided by mean error).²³

In the 3-target transfer test, MT was measured from the time the tangential velocity first exceeded 5% peak velocity to the time it fell below 5% of peak velocity. Smoothness was assessed by calculating the number of peaks in the tangential velocity profile,¹⁹ using the MATLAB function findpeaks(). The symmetry of the velocity profile was assessed as the time of the maximum tangential velocity divided by MT, termed the velocity symmetry index. The movement phase before the peak velocity is predominantly associated with feedforward control, and the post-peak phase with feedback control; thus, an increased velocity symmetry index (up to a maximum of 0.5) indicates improved feedforward control.^5,11,24 In the speed-accuracy tradeoff Fitts’ test, MT was calculated from when the participant left the start position to when they reached the target and the velocity was less than 5% of the peak velocity.

Statistical Analyses

We conducted mixed-effect models to examine the impact of training. We analyzed (1) average movement speed, (2) tracking error as the mean absolute value of the deviation from track midline from start to stop, and (3) training speed-accuracy tradeoff SAT as a function of trials and training conditions and their interaction as fixed effects.

Similarly, all testing variables were assessed with mixed-effect models. For the 3-target transfer tests, we analyzed (1) MT, (2) the number of peaks in the velocity profile, and (3) the velocity profile symmetry index. Testing days (pre-, 3-day-post, and 1-month post), groups, and day-by-group interactions were entered as fixed effects.

For the Fitts’s test, the speed-accuracy trade-off was assessed as a linear relationship between movement time and the log ratio of the movement distance and movement error tolerance:

MT = a + b \log_{2} (\frac{2 D}{W})

(1)

where MT is the movement time, a is the intercept, b is the slope, D is the target’s distance from the start position, and W is the target diameter.¹² $ID = \log_{2} (\frac{2 D}{W})$ is the index of difficulty. The slope was used as a measure of the tradeoff. An increased slope has been observed for the more affected arm compared to the less affected arm.¹² We estimated Fitts’ slope via a mixed effect model of the MT where difficulty ID, testing days, and the group were included as fixed effects.

Secondary clinical outcome data (UEFM, ARAT, and Box and Block) were analyzed using mixed-effect models, with days and conditions as fixed effects.

We refer the reader to Supplemental material: detailed statistical analyses for information on fitting the mixed effect models, rationale, methods for removing a small number of outlier trials due to abnormal movement times induced by the protocol, and reporting the results after correcting for multiple comparisons. The data used in this manuscript are available upon reasonable request from the corresponding author.

Results

Demographic and Overall Training and Testing Completion

Forty-two individuals (18 Females; median age = 61 years; range = 34-83 years) with mild to moderate UE impairments due to a unilateral stroke were randomized into either the Speed (N = 21) or Accuracy group and completed training (N = 21; see Supplemental Table S1 and Supplemental Consort diagram). All 42 participants completed the 2 baseline tests and the 3-day post-test, while all but 2 in the accuracy group completed the 1-month post-test. However, we analyzed data from all 42 participants, replacing the missing data in the 1-month test for these two participants with missing values. The median duration since the last stroke was 1.2 years (range = 0.5-21 years), and the median UEFM score was 44.5 (range = 26-57). No participants showed scores greater than 2 on the Modified Ashworth scale for elbow flexors at any visit. Table 1 demonstrates that the adaptive randomization method aimed at balancing UEFM, age, sex, time since stroke, and lesioned hemisphere resulted in balanced groups for all variables (P > .05) except for age (P = .03). A stroke imaging expert confirmed on MRI that none of the participants had sustained a cerebellar stroke.

Table 1.

Summary of participant characteristics.

Variables	Speed group	Accuracy group	Between-group difference (P levels)
N	21	21
Age (y)	56.9 (3.1)	65.8 (2.3)	.03*
Sex (M,F)	12,9	12,9	>.05
Hand dominance prior to stroke	20R, 1L	19R, 2L	>.05
Time since stroke	22.1 (5.2)	35.3 (11.6)	.3
Lesioned hemisphere	12LCVA, 9RCVA	10LCVA, 11RCVA	>.05
Paretic side	12R, 9L	10R, 11L	>.05
UEFM	42 (2.1)	43 (1.6)	.52
Average baseline MT for paretic arm	0.85 (0.09)	0.77 (0.05)	.43

indicates P < 0.05.

All participants completed 4 training days and 520 trials in each training session, totaling 2080 movements, except for 1 participant in the Speed group who performed 411 movements on day 1. The average session lasted 118 ± 37 minutes (range = 66-185 minutes) in the Speed group and 178 ± 37 minutes (range = 97-224 minutes) in the Accuracy group. Because of a reporting error, the UEFM score from 1 Accuracy group participant was missing in the 3-day post-test.

Effect of Speed and Accuracy Conditions on Movement Time and Speed-Accuracy Trade-Off During Training

As expected from our design, the Speed group showed higher mean speed and mean tracking error during training than the Accuracy group (Figure 2A and B). From the first trial, the mean speed was significantly higher in the Speed group (mean mixed model intercept 30.1 cm/s vs 9.0 cm/s for Accuracy P < .0001). It increased at 0.010 cm/s per trial (P < .0001) over practice, reaching a maximum mean speed of 51.1 cm/s at the end of training. In contrast, the speed did not change in the Accuracy group (P = .69), resulting in a significant interaction (P < .0001; Figure 2A). Because of the large speed and the wider tracks, the mean tracking error was larger in the Speed condition (mean 1.1 cm) than in the Accuracy condition (mean 0.55 cm, P < .0001; Figure 2B).

Figure 2.

Evolution of mean movement speed (A), mean tracking error (B), and speed accuracy tradeoff SAT (C) across 2080 trials in 4 sessions for the speed and Accuracy groups. a.u.: arbitrary units. The p values in the legends indicate whether the slopes are significantly different from 0. A. Tracking speed during training. B. Tracking error during training. C. Speed accuracy trade-off during training.

Due to faster speed and a relatively smaller increase in error, the training SAT was larger initially in the Speed group (30.7 s⁻¹) than in the Accuracy group (19.5 s⁻¹; P < .0001). The SAT increased at a rate of 0.0065 s⁻¹ (P < .0001) over all trials in the Speed group. In the Accuracy group, SAT increased more slowly at 0.0021 s⁻¹ on average (P = .009), with a significant interaction between groups (P < .001; Figure 2C).

Speed Training Reduces Movement Time and Number of Peaks and Increases Symmetry in Velocity Profiles in 3-Target Reaching Transfer Test

Figure 3 shows examples of reaching movements from the pre-training, 3-day, and 1-month post-training reaching transfer tests for representative participants from Speed and Accuracy groups.

Figure 3.

Examples of hand paths and tangential hand velocities before and after training for a participant post-stroke in the (A) Speed group and a participant in the (B) Accuracy group in baseline (Pre), 3-day (3D), and 1-month (1M) post-tests. First row: Hand path. Second row: Tangential velocities. Notice how training reduces the movement time, increases the speed, increases smoothness (fewer peaks in velocity profiles), and makes the velocity profiles more symmetrical.

Reaching MT changed with training in both groups (effect of day, P = .0013). From baseline to 3-day post-test, MT decreased by ~16% in the Speed group (from 0.76 ± 0.4 to 0.64 ± 0.04; P < .000) (Figure 4A) but increased in the Accuracy group (from 0.64 ± .05 to 0.71 ± 0.05; P < .0001), resulting in a significant interaction (P < .0001). However, the gains in the Speed group were largely lost at the 1-month post-test, as MT increased to 0.73 ± 0.08 seconds (P < .0001 between 3 days and 1 month; P = .03 between baseline and 1 month). Despite this loss, there was still a small but significantly greater decrease in MT in speed training, resulting in a significant interaction between baseline and 1 month (P = .034; Figure 4B).

Figure 4.

Changes and between-group comparison for the kinematic variables in the 3-target reach test. (A and B) movement time. (C and D) number of peaks in the velocity profile. (E and F) velocity symmetry index, where the index is defined as the ratio of the time of the maximum velocity peak divided by the movement time. Left: each dot is the average for each participant, with the bars showing the estimated marginal mean with the SE. Right: Interaction plots, with the horizontal bars and P values showing the interaction effect day × group for each time period (Pre- to 3D, 3D to 1 month, and Pre to 1 month).

Movement smoothness changed due to training overall (P < .0001), with a significant difference between groups (interaction; P < .0001). Training increased movement smoothness in the Speed group, evidenced by fewer peaks in the velocity profile. Between baseline and 3-day post-test, the number of peaks decreased by ~19% in the Speed group (from 1.88 ± 0.12 to 1.51 ± 0.12; P < .0001) (Figure 4C) but did not change significantly in the Accuracy group (from 1.98 ± 0.05 to 1.88 ± 0.05; P > .05), resulting in a significant interaction (P < .0001). Some of the gains in Speed training were lost at the 1-month post-test, as the number of peaks increased to 1.66 ± 0.12 (P = .0023 between 3 days and 1 month), although the gain between baseline and 1 month was still significant (P < .0001).

The velocity symmetry index also changed with training days (P < .0001), with a significant difference between groups (interaction: P < .0001). Following training, velocity profiles were more symmetrical, with a greater effect on the Speed group (Figure 4E). Speed training resulted in an increase in the velocity symmetry index from 0.38 ± 0.01 at baseline to 0.43 ± 0.01 (P < .0001) at 3 days, which is not different from the index for the less-affected (LA) arm (0.44 ± 0.02; P = .47). Some loss of symmetry occurred at the 1-month test (1 month: 0.41 ± 0.01; P = .002 between 3 days and 1 month), although the gain between baseline and 1 month was still significant (P < .0001). Accuracy training resulted in no significant change in the velocity symmetry index at 3 days and 1 month (P > .05). There was a significant interaction from baseline to 3-day (P < .0001), but there was no significant difference between the groups from baseline to 1 month following the intervention (interaction effect P > .05; Figure 4).

Speed Training Decreases Fitt’s Slope in the Speed Condition

As expected, an increase in the index of difficulty ID was accompanied by increased MT, evidenced by the positive Fitts’ slope across days and groups (P < .0001). Training days and groups had a differential effect on the slope (3-way interaction, day, group, and ID P = .003). In the Speed group, we observed a significant decrease in the slope from 0.123 ± 0.013 at baseline to 0.095 ± 0.013 at the 3-day test (P = .01), with some (non-significant) increase in the 1-month post-test (0.11 ± 0.014, P > .05, Figure 5A). In the Accuracy group, there was a (non-significant) increase in the slope from 0.130 ± 0.014 at baseline to 0.148 ± 0.014 at the 3-day test (P > .05), but a return near baseline levels at 1-month post-test (0.137 ± 0.02, P > .05, Figure 5D and E). Speed training thus resulted in a significantly greater decrease in Fitt’s slope from baseline to the 3-day post-test than did accuracy training (interaction effect P = .0024), but no significant change between the groups from baseline and 1 month following the intervention (interaction effect P > .05) (Figure 5B).

Figure 5.

Evolution of the estimated slope in the speed accuracy transfer Fitts’ test, for the Speed and Accuracy conditions. (A) Within-group change in slopes for both conditions in pre-, 3-day post-, and 1 month-post tests. The error bars show the group marginal means and SE of the slopes. (B) Group interaction effects for each time period (Pre- to 3D, 3D to 1 month and Pre to 1 month).

Clinical Assessments

Training improved clinical outcomes (UEFM, P = .0054, Supplemental Figure S1; ARAT, P = .000038, Supplemental Figure S2; Box and Block, P < .0001, Supplemental Figure S3), with no significant group, or day × group interactions (all P > .05). See details in Supplemental Results: Clinical assessments.

Discussion

We designed a UE motor task with a simple manipulation of visual display, an increase in the track width, which led to a significant increase in movement speed. Four sessions of speed or accuracy training with 520 movements/session led to significant differences in kinematic outcomes for the more affected arm that largely persisted to the 3-day post-test. In the 3-target transfer test, the Speed group showed larger decreases in MT, improvements in smoothness (reduced number of peaks), and reductions in velocity symmetry index. Remarkably, the velocity symmetry index at 3 days post-training was similar to that of the less-affected arm.

In the Fitts’ test, where precise reaching adjustments are required, MT reflects the quality of both the feedforward commands and the subsequent feedback corrections.^25,26 As shown in Figure 5, a comparison of the Fitts’s slopes for the less affected and more affected arms confirmed previous findings that stroke increases the slope of the Fitts law.^11,12 Post-stroke, increased neuromotor noise in the initial feedforward impulses results in a larger disparity between the planned and the actual movement. Thus, a conservative strategy is to minimize feedforward control to accommodate the accuracy constraints; this results in the need to wait for feedback information to run subsequent corrective sub-movements to reach the target accurately.¹¹ This leads to an increase in MT and, thus, the Fitts’ slope. Speed training led to a decrease in the Fitts’ slope at 3 days post, which suggests a training-induced shift from feedback to feedforward control.¹²

Surprisingly, little research has examined how motor learning impacts Fitts’ law. Two studies with neurotypical participants reported that the speed-accuracy trade-off changes with practice,^27,28 primarily through decreased trial-to-trial variability and an increase in movement smoothness. In chronic stroke, a study on grip control training found an average decrease in Fitts’ slope of 26% at high difficulty levels after training, indicating improved grasp control.²⁹ Thus, our study adds to this small body of literature and shows that speed but not accuracy training decreases Fitts’s slope in post-stroke individuals.

Taken together, these results show that speed training leads to at least short-term improvements in the skilled arm reaching in chronic stroke individuals with mild to moderate impairments. Importantly, this study demonstrated that the capacity to move faster, smoother, and with a more symmetrical velocity profile is not lost after a stroke – see also DeJong et al.³⁰

Our theoretical framework of feedback error learning posits that increasing errors facilitates (re-) tuning of the feedforward controller, with greater errors providing the signals needed to learn corrections for amplitude and timing of interaction torques during movements. This approach resembles error augmentation (EA), where feedback is artificially amplified, which may improve motor performance in chronic stroke.^31-34 Unlike EA, however, fast movements in the Speed condition generate interaction torques that are difficult to control, making the resultant error feedback more appropriate for learning to compensate for these torques.

In contrast to these error-based paradigm, “error-less” training has also been shown to improve motor performance in non-disabled and post-stroke individuals by mostly recruiting an implicit learning process.^35,36 Error-less practice could improve performance via use-dependent learning, which could reduce the variability of feedforward control. Alternatively, repeated positive rewards in error-less learning could enhance motor command and decrease the need for exploration noise in reward-based learning. Future research is needed on how a combination of these error-based and reward-based mechanisms can improve arm-reaching performance post-stroke.³⁷

However, in our study, most gains in Speed training were lost 1 month after training. This contrasted with our previous work where, following speed training, patients retained improvements over 1 month.¹⁹ While surprising, these differences in our findings may be due to differences in participants, training, feedback, and testing conditions. For example, in our previous study with fast-reaching movement training,¹⁹ the speed of movements could be twice that in the current study. Future trials should further investigate whether reducing the frequency of feedback³⁸ or increasing the duration of speed training leads to longer-lasting effects, or possibly whether low use of the paretic UE in daily activity led to a decrease in performance.^39-42 Perhaps a longer duration of in-clinic Speed training paired with a technology-assisted, self-training home program would foster better durability and generalizability in the wild.³⁹

Speed training sessions were on average ~60 minutes shorter than accuracy sessions (118 minutes vs 178 minutes; P < .05). This emphasizes that speed training may be more efficient and effective in driving motor control and clinical changes in stroke survivors. However, the substantial difference in average session duration may have contributed to fatigue, either from more intensive exercises in the Speed group or more extended sessions in the Accuracy group. It is, therefore, possible that a more distributed practice schedule would be more effective. Nevertheless, we have previously shown that whereas extensive speed training leads to a decrement in performance within (long) sessions, learning, as measured by a delayed retention test, is independent of this fatigue effect.⁴³

Similar to our previous study,¹⁹ in which we showed an increase in the Box and Block scores due to speed reach training, there were increases in the clinical outcome measures due to training (UEFM, ARAT, and Box and Block Test). However, we found no significant group interaction in the 3 assessments. Note that while ARAT and Box and Block predominantly rely on hand function, which was not engaged in our paradigm, they also engage the proximal arm. For example, in the Box and Block, the transferring motion involves some proximal arm movements. Further, previous reports show some improvement in distal function following proximal arm training; see, for example, McCombe Waller and Whitall.⁴⁴

Limitations and Future Work

Here, data for training and testing are reported only for the end-point kinematics. Previous studies have stressed measures of movement quality to identify motor compensations, for example, Cirstea and Levin⁴⁵ and Nibras et al.⁴⁶ Thus, we cannot ascertain at this time if the changes in end-point performance are associated with true changes in intersegmental kinematics or compensation.⁴⁶

To address the above, we have conducted 3D reaching tests with and without trunk restraint. We have also conducted an arm choice test, a fast elbow reaching test, an elbow perturbation test, and a bilateral proprioception test; the results of all these tests will be reported in secondary outcome manuscripts. However, we note that these tests constitute a non-insignificant amount of practice (at least 266 active movements, including 150 fast elbow movements, and 90 passive movements) that may have reduced the difference in outcomes between the 2 groups.

While speed training improved short-term performance and transfer, movements may be disrupted at higher speeds due to abnormal regulation of the stretch reflex.^47,48 Active actions modulate the range at which abnormal stretch reflex is expressed; however, how these processes operate in multi-joint actions⁴⁵ and modulate over training are key questions for future research.

Recovery of the UE following stroke is influenced by numerous factors.^49-53 While encouraging, it is unclear which patients with stroke will most benefit from speed training. For example, reciprocal connections between cerebellum and cerebral cortex, and sensory processing likely implement error-based learning mechanisms needed for retraining feedforward control.^54-58 Future analyses are needed to understand the possible influences of connectivity and sensory deficits on gains during speed training.

In conclusion, our study shows that in individuals with moderate to mild post-stroke impairments, not all repetitions during rehabilitation are equally effective in restoring motor performance: faster movements during training lead to better movement quality, at least temporarily. Our results suggest that therapists should emphasize speed to improve reaching movements during UE rehabilitation, encouraging patients to move as quickly as possible, even to some detriment of accuracy. This approach is easy to implement in therapy settings and offers a cost-effective strategy to enhance UE recovery.

Supplemental Material

sj-docx-1-nnr-10.1177_15459683251331582 – Supplemental material for Speed-Biased Training Temporarily Improves Motor Performance of the Paretic Arm Compared to Accuracy-Biased Training in Chronic Stroke Survivors: The Phase 1 FAST Randomized Clinical Trial

Supplemental material, sj-docx-1-nnr-10.1177_15459683251331582 for Speed-Biased Training Temporarily Improves Motor Performance of the Paretic Arm Compared to Accuracy-Biased Training in Chronic Stroke Survivors: The Phase 1 FAST Randomized Clinical Trial by Yannick Darmon, Shailesh Kantak, Hannah Cone, Niko Fullmer, Debra Ouellette, Carolee Winstein, Emily R. Rosario and Nicolas Schweighofer in Neurorehabilitation and Neural Repair

Footnotes

Acknowledgements

We thank Emily Kieffer, Madison Gerber, Daniel Humphrey, Lana Ignacio, and all other clinical staff who helped with recruitment and data collection, Marika Demers for help with implementation of the UEFM standardization, Amy Zheng for help with randomization, Amy Zheng and Gary Jensen for help with brain imaging, and Alec Roig for technical assistance.

Author Contributions

Yannick Darmon: Conceptualization; Data curation; Formal analysis; Methodology; Project administration; Software; Supervision; Visualization; Writing—original draft; and Writing—review & editing. Shailesh Kantak: Conceptualization; Methodology; Writing—original draft; and Writing—review & editing. Hannah Cone: Investigation; Methodology; and Writing—review & editing. Niko Fullmer: Data curation; Project administration; and Writing—review & editing. Debra Ouellette: Investigation and Writing—review & editing. Carolee Winstein: Conceptualization; Methodology; Writing—original draft; and Writing—review & editing. Emily R Rosario: Conceptualization; Data curation; Investigation; Methodology; Project administration; Resources; Supervision; Validation; Writing—original draft; and Writing—review & editing. Nicolas Schweighofer: Conceptualization; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Supervision; Validation; Visualization; Writing—original draft; and Writing—review & editing.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was funded by grant NIH R21NS120274 to NS.

ORCID iDs

Shailesh Kantak

Carolee Winstein

Nicolas Schweighofer

Supplementary material for this article is available on the Neurorehabilitation & Neural Repair website along with the online version of this article.

References

Winstein

Wolf

Dromerick

, et al. Effect of a task-oriented rehabilitation program on upper extremity recovery following motor stroke: the ICARE randomized clinical trial. JAMA. 2016;315(6):571-581. doi:10.1001/jama.2016.0276

Lang

Strube

Bland

, et al. Dose response of task-specific upper limb training in people at least 6 months poststroke: a phase II, single-blind, randomized, controlled trial. Ann Neurol. 2016;80(3):342-354. doi:10.1002/ana.24734

Krakauer

Kitago

Goldsmith

, et al. Comparing a novel neuroanimation experience to conventional therapy for high-dose intensive upper-limb training in subacute stroke: the SMARTS2 randomized trial. Neurorehabil Neural Repair. 2021;35(5):393-405. doi:10.1177/15459683211000730

Rodgers

Bosomworth

Krebs

, et al. Robot assisted training for the upper limb after stroke (RATULS): a multicentre randomised controlled trial. Lancet. 2019;394(10192):51-62. doi:10.1016/S0140-6736(19)31055-4

Elliott

Hansen

Grierson

Lyons

Bennett

Hayes

SJ.

Goal-directed aiming: two components but multiple processes. Psychol Bull. 2010;136(6):1023-1044. doi:10.1037/a0020958

Kawato

Internal models for motor control and trajectory planning. Cur Opin Neurobiol. 1999;9(6):718-727. doi:10.1016/S0959-4388(99)00028-8

Beer

Dewald

Rymer

WZ.

Deficits in the coordination of multijoint arm movements in patients with hemiparesis: evidence for disturbed control of limb dynamics. Exp Brain Res. 2000;131(3):305-319.

Laczko

Scheidt

Simo

Piovesan

Inter-joint coordination deficits revealed in the decomposition of endpoint jerk during goal-directed arm movement after stroke. IEEE Trans Neural Syst Rehabil Eng. 2017;25(7):798-810. doi:10.1109/TNSRE.2017.2652393

Raj

Dounskaia

Clark

Sethi

Effect of stroke on joint control during reach-to-grasp: a preliminary study. J Motor Behav. 2020;52(3):294-310. doi:10.1080/00222895.2019.1615861

10.

Takahashi

Reinkensmeyer

DJ.

Hemiparetic stroke impairs anticipatory control of arm movement. Exp Brain Res. 2003;149(2):131-140. doi:10.1007/s00221-002-1340-1

11.

Mottet

van Dokkum

Froger

Gouaich

Laffont

Trajectory formation principles are the same after mild or moderate stroke. PLoS ONE. 2017;12(3):e0173674. doi:10.1371/journal.pone.0173674

12.

McCrea

Eng

JJ.

Consequences of increased neuromotor noise for reaching movements in persons with stroke. Exp Brain Res. 2005;162(1):70-77. doi:10.1007/s00221-004-2106-8

13.

Murphy

Willén

Sunnerhagen

KS.

Kinematic variables quantifying upper-extremity performance after stroke during reaching and drinking from a glass. Neurorehabil Neural Repair. 2011;25(1):71-80. doi:10.1177/1545968310370748

14.

Roby-Brami

Jacobs

Bennis

Levin

MF.

Hand orientation for grasping and arm joint rotation patterns in healthy subjects and hemiparetic stroke patients. Brain Res. 2003;969(1):217-229. doi:10.1016/S0006-8993(03)02334-5

15.

Stewart

Gordon

Winstein

CJ.

Control of reach extent with the paretic and nonparetic arms after unilateral sensorimotor stroke II: planning and adjustments to control movement distance. Exp Brain Res. 2014;232(11):3431-3443. doi:10.1007/s00221-014-4025-7

16.

Stewart

Gordon

Winstein

CJ.

Control of reach extent with the paretic and nonparetic arms after unilateral sensorimotor stroke: kinematic differences based on side of brain damage. Exp Brain Res. 2014;232(7):2407-2419. doi:10.1007/s00221-014-3938-5

17.

Kitago

Krakauer

JW.

Chapter 8 - Motor learning principles for neurorehabilitation. In: Barnes

Good

, eds. Handbook of Clinical Neurology. Elsevier; 2013:93-103.

18.

Kantak

Zahedi

McGrath

Complex skill training transfers to improved performance and control of simpler tasks after stroke. Phys Ther. 2017;97(7):718-728. doi:10.1093/ptj/pzx042

19.

Park

Kim

Winstein

Gordon

Schweighofer

Short-duration and intensive training improves long-term reaching performance in individuals with chronic stroke. Neurorehabil Neural Repair. 2016;30(6):551-561. doi:10.1177/15459683156069901545968315606990

20.

Woytowicz

Rietschel

Goodman

, et al. Determining levels of upper extremity movement impairment by applying a cluster analysis to the Fugl-Meyer assessment of the upper extremity in chronic stroke. Arch Phys Med Rehabil. 2017;98(3):456-462. doi:10.1016/j.apmr.2016.06.023

21.

Grischott

The Shiny balancer: software and imbalance criteria for optimally balanced treatment allocation in small RCTs and cRCTs. BMC Med Res Methodol. 2018;18(1):108. doi:10.1186/s12874-018-0551-5

22.

See

Dodakian

Chou

, et al. A standardized approach to the Fugl-Meyer assessment and its implications for clinical trials. Neurorehabil Neural Repair. 2013;27(8):732-741. doi:10.1177/1545968313491000

23.

Riga

Gathy

Ghinet

, et al. Evidence of motor skill learning in acute stroke patients without lesions to the Thalamus and internal capsule. Stroke. 2022;53(7):2361-2368. doi:10.1161/STROKEAHA.121.035494

24.

Meyer

Smith

Wright

CE.

Models for the speed and accuracy of aimed movements. Psychol Rev. 1982;89(5):449-482.

25.

Fitts

PM.

The information capacity of the human motor system in controlling the amplitude of movement. J Exp Psychol. 1954;47(6):381-391.

26.

Meyer

Abrams

Kornblum

Wright

Keith Smith

JE.

Optimality in human motor performance: ideal control of rapid aimed movements. Psychol Rev. 1988;95:340-370. doi:10.1037/0033-295X.95.3.340

27.

Abrams

Pratt

Rapid aimed limb movements: differential effects of practice on component submovements. J Mot Behav. 1993;25(4):288-298. doi:10.1080/00222895.1993.9941650

28.

Shmuelof

Krakauer

Mazzoni

How is a motor skill learned? Change and invariance at the levels of task success and trajectory control. J Neurophysiol. 2012;108(2):578-594. doi:10.1152/jn.00856.2011

29.

Kim

Wininger

Craelius

Training grip control with a Fitts’ paradigm: a pilot study in chronic stroke. J Hand Ther. 2010;23(1):63-72. doi:10.1016/j.jht.2009.10.004

30.

DeJong

Schaefer

Lang

CE.

Need for speed: better movement quality during faster task performance after stroke. Neurorehabil Neural Repair. 2012;26(4):362-373. doi:10.1177/1545968311425926

31.

Patton

Stoykov

Kovic

Mussa-Ivaldi

FA.

Evaluation of robotic training forces that either enhance or reduce error in chronic hemiparetic stroke survivors. Exp Brain Res. 2006;168(3):368-383. doi:10.1007/s00221-005-0097-8

32.

Israely

Carmeli

Error augmentation as a possible technique for improving upper extremity motor performance after a stroke: a systematic review. Top Stroke Rehabil. 2016;23(2):116-125. doi:10.1179/1945511915Y.0000000007

33.

Liu

Lamontagne

The effects of error-augmentation versus error-reduction paradigms in robotic therapy to enhance upper extremity performance and recovery post-stroke: a systematic review. J Neuroeng Rehabil. 2018;15(1):65. doi:10.1186/s12984-018-0408-5

34.

Rajda

Feldman

Boudrias

, et al. The use of enhanced intrinsic feedback for motor learning in stroke survivors: clinical trial protocol. Contemp Clin Trials. 2023;130:107236. doi:10.1016/j.cct.2023.107236

35.

Poolton

Masters

Maxwell

JP.

The relationship between initial errorless learning conditions and subsequent performance. Hum Mov Sci. 2005;24(3):362-378. doi:10.1016/j.humov.2005.06.006

36.

Maxwell

Masters

Kerr

Weedon

The implicit benefit of learning without errors. Q J Exp Psychol A. 2001;54(4):1049-1068. doi:10.1080/713756014

37.

Truong

Ruffino

Crognier

Paizis

Crognier

Papaxanthis

Error-based and reinforcement learning in basketball free throw shooting. Sci Rep. 2023;13(1):499. doi:10.1038/s41598-022-26568-2

38.

Winstein

Schmidt

RA.

Reduced frequency of knowledge of results enhances motor skill learning. J Exp Psychol Learn Memory Cogn. 1990;16:677

39.

Ballester

Winstein

Schweighofer

Virtuous and vicious cycles of arm use and function post-stroke. Front Neurol. 2022;13:804211. doi:10.3389/fneur.2022.804211

40.

Han

Arbib

Schweighofer

Stroke rehabilitation reaches a threshold. PLOS Comput Biol. 2008;4(8):e1000133. doi:10.1371/journal.pcbi.1000133

41.

Hidaka

Han

Wolf

Winstein

Schweighofer

Use it and improve it or lose it: interactions between arm function and use in humans post-stroke. PLoS Comput Biol. 2012;8(2):e1002343. doi:10.1371/journal.pcbi.1002343PCOMPBIOL-D-10-00044 [pii]

42.

Schweighofer

Han

Wolf

Arbib

Winstein

CJ.

A functional threshold for long-term use of hand and arm function can be determined: predictions from a computational model and supporting data from the extremity constraint-induced therapy evaluation (EXCITE) trial. Phys Ther. 2009;89(12):1327-1336. doi:10.2522/ptj.20080402

43.

Park

Schweighofer

Nonlinear mixed-effects model reveals a distinction between learning and performance in intensive reach training post-stroke. J Neuroeng Rehabil. 2017;14(1):21. doi:10.1186/s12984-017-0233-2

44.

McCombe Waller

Whitall

. Fine motor control in adults with and without chronic hemiparesis: baseline comparison to nondisabled adults and effects of bilateral arm training. Arch Phys Med Rehabil. 2004;85(7):1076-1083. doi:10.1016/j.apmr.2003.10.020

45.

Cirstea

Levin

MF.

Compensatory strategies for reaching in stroke. Brain. 2000;123(5):940-953. doi:10.1093/brain/123.5.940

46.

Nibras

Liu

Mottet

, et al. Dissociating sensorimotor recovery and compensation during exoskeleton training following stroke. Front Hum Neurosc. 2021;15:645021.

47.

Subramanian

Feldman

Levin

MF.

Spasticity may obscure motor learning ability after stroke. J Neurophysiol. 2018;119(1):5-20. doi:10.1152/jn.00362.2017

48.

Turpin

Feldman

Levin

MF.

Stretch-reflex threshold modulation during active elbow movements in post-stroke survivors with spasticity. Clin Neurophysiol. 2017;128(10):1891-1897. doi:10.1016/j.clinph.2017.07.411

49.

Shelton

FdNAP

Reding

MJ.

Effect of lesion location on upper limb motor recovery after stroke. Stroke. 2001;32(1):107-112. doi:10.1161/01.STR.32.1.107

50.

Zhu

Lindenberg

Alexander

Schlaug

Lesion load of the corticospinal tract predicts motor impairment in chronic stroke. Stroke. 2010;41(5):910-915. doi:10.1161/STROKEAHA.109.577023

51.

Kim

Schweighofer

Haldar

Leahy

Winstein

CJ.

Corticospinal tract microstructure predicts distal arm motor improvements in chronic stroke. J Neurol Phys Ther. 2021;45(4):273-281. doi:10.1097/NPT.0000000000000363

52.

Meyer

Karttunen

Thijs

Feys

Verheyden

How do somatosensory deficits in the arm and hand relate to upper limb impairment, activity, and participation problems after stroke? A systematic review. Phys Ther. 2014;94(9):1220-1231. doi:10.2522/ptj.20130271

53.

Vidoni

Boyd

LA.

Preserved motor learning after stroke is related to the degree of proprioceptive deficit. Behav Brain Funct. 2009;5(1):36. doi:10.1186/1744-9081-5-36

54.

Guder

Frey

Backhaus

, et al. The influence of cortico-cerebellar structural connectivity on cortical excitability in chronic stroke. Cereb Cortex. 2020;30(3):1330-1344. doi:10.1093/cercor/bhz169

55.

Schulz

Frey

Koch

, et al. Cortico-cerebellar structural connectivity is related to residual motor output in chronic stroke. Cereb Cortex. 2017;27(1):635-645. doi:10.1093/cercor/bhv251

56.

Park

C-H

Chang

Ohn

, et al. Longitudinal changes of resting-state functional connectivity during motor recovery after stroke. Stroke. 2011;42(5):1357-1362. doi:10.1161/STROKEAHA.110.596155

57.

Várkuti

Guan

Pan

, et al. Resting state changes in functional connectivity correlate with movement recovery for BCI and robot-assisted upper-extremity training after stroke. Neurorehabil Neural Repair. 2013;27(1):53-62. doi:10.1177/1545968312445910

58.

Wang

Chen

, et al. Dynamic functional reorganization of the motor execution network after stroke. Brain. 2010;133(4):1224-1238. doi:10.1093/brain/awq043

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.82 MB