Table 1 Summary of studies that compared EA to training without error modification

Abdollahi et al. [52]

Crossover randomized controlled trial.

Trial registered NCT01574495

27 in total, E = 13, C = 14 (before crossover)

Ages: 36–88 years (mean = 57.92 ± 9.96), 12 males and 15 females, all participants suffered a single cortical or subcortical stroke at least 6 months prior to the study (mean = 82.34 ± 72.04 months, 9 hemorrhagic, 18 ischemic) FM^b scores: 15–50 (mean/SD unknown)

Virtual Reality Robotic and Optical Operations Machine (VRROOM). Phantom Premium 3.0 robot.

Experimental task: various reaching movement

Control group: received only treatment of repetitive practice

EA group:

same as the control group in addition of combined visual and haptic error augmentation

Training parameters:

60 min per session, three sessions per week, two weeks of training per phase (two). After first phase, all subjects switched to the other group.

Clinical

ROM^c, AMFM^a, WMFT FAS^d, WMFT time measure, Box and Blocks test

ROM: no significant effects AMFM: in the first phase, EA showed more improvement than control (2.08 ± 2.25 vs 0.69 ± 2.90), and this difference was significant [F(1,24) = 4.261, p < 0.05]

In the second phase, EA was still better than control (1.15 ± 2.21 vs 0.54 ± 2.30), but not significantly (numerical values not provided)

WMFT FAS: in the first phase, EA was better than control (0.11 ± 0.24 vs 0.01 ± 0.16), but level of significance unknown.

In the second phase, control was better than EA (0.14 ± 0.22 vs − 0.02 ± 0.25), but level of significance unknown.

WMFT time: in the first phase, EA was better than control (1.48 ± 8.86 vs − 0.53 ± 5.19), but level of significance unknown.

In the second phase, EA was better than control (1.19. ± 5.68 vs 0.17 ± 8.02), but level of significance unknown.

Box and Blocks test: no significant effects

AMFM:

first phase 0.53 (medium effect).

second phase 0.27 (small effect).

WMFT FAS:

first phase 0.51 (medium effect).

second phase 0.52 (medium effect)

WMFT time:

First phase

0.28 (small effect)

Second phase 0.14 (very small effect)

7/10, high quality

Givon-Mayo et al. [63]

Pilot study.

Trial registered

NCT02017093

7 in total, E = 4 and C = 3

Ages: 45–78 (mean = 59.14 ± 9.77), 8 males and 1 female. All participants sustained a stroke (1 hemorrhagic, 8 ischemic) 2 to 3 weeks prior to the study. FM scores: EA group mean = 53.25 ± 3.77

Control group

mean = 54.33 ± 3.84

DeXtreme prototype robot- 2 degrees of freedom. Free end Robot.

Experimental tasks: reaching

EA group: received error inducing forces from the robot

Control group: also attached to the robot, but did not receive forces

Training parameters:

20 min per session, three sessions per week for five weeks

Kinematic: movement velocity deviation error (cm/sec).

Clinical: FM, MAS^e

MAS: the EA group showed more improvement than the control group (3.2 ± 2.6 vs 1.7 ± 3.2), but the level of significant unknown

FM: no significant changes

Velocity deviation error: the EA group showed significant (p < 0.05) more improvement than the control group (− 16.8 ± 3.8 vs − 4.7 ± 3.8)

MAS: 0.51 (medium effect)

Velocity deviation error:

3.2 (very large effect)

3/10, poor quality

Huang and Patton [59]

Crossover design.

Trial not registered

30 in total, Participants were randomly assigned to either one control group or two experimental groups, but the numbers are unknown.

Mean age = 52.0 ± 8.2, all participants suffered from a chronic stroke (mean = 102.0 ± 84.0 months). Clinical assessment results prior to the study were not available

A planar force feedback device. The subject’s arm was supported by a low-friction, low-impedance mechanism

Experimental tasks: circular movement task

Control group practiced on the training device in null-field conditions in all sessions.

The two experimental groups: trained in a null field condition in the first session, then received either EA force alone or EA force combined with positive limb inertia in the next session. They switched to the other condition in third session

Training parameters:

Two hours per session, three sessions in total

Kinematic: radial deviation (distance between the handle and template circular track) (mm)

When evaluated in the next session, the control showed no significant improvement (0.7 mm ± 2.3, 95% confidence interval: − 0.4 to 1.8). The EA group showed the largest significant improvement (1.4 ± 2.7, CI: 0.2 to 3.0) while the combined EA with inertia group showed non-significant improvement (1.1 ± 2.7, CI: 0 to 2.2)

EA compared to control: 0.28 (small effect).

EA compared to combined EA with inertia: 0.11 (very small effect)

3/10, poor quality

Majeed et al. [62]

Randomized comparative experiment.

Trial not registered.

28 in total, participants were randomized into experimental and control groups based on blocks of FM scores

Ages: 26–78 (mean = 55.38, SD unknown), 17 males and 11 females, all participants suffered from a cortical chronic stroke (mean/SD unknown), Upper extremity FM score: 25–49 (block randomized into both groups, mean/SD unknown)

Three-dimentional haptic/ graphic system called the Virtual Reality Robotic and Optical Operations Machine (VRROOM)

Experimental tasks: reaching

Control group: received only treatment of repetitive practice

EA group:

same as the control group in addition of combined visual and haptic error augmentation

Training parameters:

45 min per session, three sessions per week, two weeks of training

Clinical:

AMFM

At the end of 2 weeks of training, no significant difference was found between EA and control groups in improvement of AMFM (numerical data not provided).

At one-week follow-up, EA group showed more retention than control group (2.60 ± 3.50 vs − 0.1 ± 6.98), but the level of significant known

From the end of training to one-week follow-up:

0.52 (medium effect)

6/10, high quality

Patton et al. [53]

Randomized controlled trial.

Trial not registered

15 in total. E = 12, C = 9 (6 subjects returned for a second visit, so they served as their own control)

Ages: 30–76 (E: mean = 50.66 ± 13.08;

C: mean = 50.77 ± 12.16), 9 males and 6 females, all participants suffered from a chronic stroke (E: mean = 77.25 ± 40.85 months;

C: mean = 99.89 ± 44.13) prior to the experiment. FM score (E: mean = 34.36 ± 12.23; C: mean = 35.0 ± 11.79

A two degrees-of-freedom robot

Experimental tasks: reaching EA group: received EA forces from the robotic while doing repetitive practice.

Control group: Same training but without EA forces.

Training parameters:

One single session of three hours and consisted of 744 movements

Kinematic: size of movement error (change in %) Clinical:

AMFM

AMFM: the EA group had a greater improvement than control group (1.6 ± 2.6, p = 0.06 vs 0.4 ± 1.1, p > 0.27), but the results were not significant

Movement error size: the EA group had a greater improvement than control group (− 45.2 ± 80.6 vs − 11.1 ± 48), but the levels of significance unknown

AMFM: 0.65 (medium effect)

Movement error size: 0.53 (medium effect)

5/10, fair quality

Rozario et al. [60]

Crossover design.

Trial not registered

10 in total

Stroke group (before cross-over)

EA = 3

Control = 2

Healthy group 5

Stroke: ages: 36–69 (mean = 55.0 ± 12.1), 4 males and 1 female, suffered a single cortical stroke for more than 6 months (mean/SD unknown), AMFM:

EA group mean = 35.33 ± 8.14

Control group mean = 43.50 ± 2.12

Healthy: ages: 19–27 (mean/SD unknown)

A 6-degree of freedom PHANTOM Premium 3.0 robot

Experimental tasks: various reaching movement.

Control group: received only treatment of repetitive practice

EA group:

same as the control group in addition of combined visual and haptic error augmentation

Healthy group: did not receive any treatment, only data collected

Training parameters:

40 min per session, three sessions per week, and two weeks per phase (two). After first phase, all subjects switched groups.

Kinematic: ROM errors (m)

Clinical:

AMFM, WMFT FAS, WMFT time, Box and Blocks test

Clinical test: no noticeable changes in any of the clinical tests (numerical data not provided).

ROM errors: In the first phase, EA showed more improvement than control group (0.08 ± 0.08 vs 0.04 ± 0.04), level of significant unknown.

In the second phase, EA showed no improvement but control group showed deterioration (0 ± 0 vs − 0.02 ± 0.03).

More errors seen in stroke subjects than healthy subjects.

ROM errors: first phase 0.75 (medium effect).

Second phase 1.05 (large effect)

4/10, fair quality