Study Paradigms and Principles Investigated in Motor Learning Research After Stroke: A Scoping Review.

OBJECTIVES: To (1) characterize study paradigms used to investigate motor learning (ML) poststroke and (2) summarize the effects of different ML principles in promoting skill acquisition and retention. Our secondary objective is to evaluate the clinical utility of ML principles on stroke rehabilitation. DATA SOURCES: Medline, Excerpta Medica Database, Allied and Complementary Medicine, Cumulative Index to Nursing and Allied Health Literature, and Cochrane Central Register of Controlled Trials were searched from inception on October 24, 2018 and repeated on June 23, 2020. Scopus was searched on January 24, 2019 and July 22, 2020 to identify additional studies. STUDY SELECTION: Our search included keywords and concepts to represent stroke and "motor learning. An iterative process was used to generate study selection criteria. Three authors independently completed title, abstract, and full-text screening. DATA EXTRACTION: Three reviewers independently completed data extraction. DATA SYNTHESIS: The Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension guidelines for scoping reviews were used to guide our synthesis. Thirty-nine studies were included. Study designs were heterogeneous, including variability in tasks practiced, acquisition parameters, and retention intervals. ML principles investigated included practice complexity, feedback, motor imagery, mental practice, action observation, implicit and explicit information, aerobic exercise, and neurostimulation. An additional 2 patient-related factors that influence ML were included: stroke characteristics and sleep. Practice complexity, feedback, and mental practice/action observation most consistently promoted ML, while provision of explicit information and more severe strokes were detrimental to ML. Other factors (ie, sleep, practice structure, aerobic exercise, neurostimulation) had a less clear influence on learning.
CONCLUSIONS: Improved consistency of reporting in ML studies is needed to improve study comparability and facilitate meta-analyses to better understand the influence of ML principles on learning poststroke. Knowledge of ML principles and patient-related factors that influence ML, with clinical judgment can guide neurologic rehabilitation delivery to improve patient motor outcomes.

One of the most debilitating deficits poststroke is motor impairment., To address this, rehabilitation therapists help individuals practice movements, enabling them to relearn functions affected by their stroke., It is important for the individual's recovery and function that this “relearning” leads to a permanent change in motor behavior. For example, clinicians may facilitate a better sit-to-stand during therapy, yet the ultimate goal is for the person to stand up independently in their daily life. This rehabilitation process exemplifies motor learning (ML) that is defined as a group of internal processes associated with practice that leads to a relatively permanent change in the capacity for skilled movement. ML is a proposed model for stroke rehabilitation.

Since 2010, a surge of randomized controlled trials has evaluated the effectiveness of different interventions for stroke recovery, leading to best practice guidelines recommending the use of specific therapies (eg, virtual reality, task-specific practice) to improve motor recovery. Although there is strong support for these interventions, it remains unclear if there are optimal conditions in which they can be implemented to maximize effectiveness. Consideration of not only the type of therapy but also the conditions under which the therapy is delivered has the potential to improve motor recovery poststroke. These conditions can be referred to as ML principles.

We define ML principles as conditions during task practice that influence learning and can be applied by rehabilitation clinicians; they are not targeted for a specific movement or therapeutic approach/intervention. As principles of ML can affect learning, they should be used to guide the implementation of neurorehabilitation to promote long-term recovery. The influence of these principles on learning can be scientifically evaluated using ML paradigms. ML paradigms typically have 3 distinct phases: acquisition session(s) when individuals practice a motor skill, retention session(s) where motor skill performance is evaluated after a period of no practice, and transfer testing whereby one tests whether improvements in the practiced skill translates to another similar skill or the same skill under different conditions. Conditions of acquisition sessions (ie, ML principles) influence the trajectory, magnitude, and permanence of learning (assessed by retention testing), and have physiological effects through neuroplasticity and neural recovery. Furthermore, ML principles that promote motor acquisition (eg, blocked practice) may be different than features that are beneficial for motor retention (eg, random practice).,

The methods used to investigate ML can affect the interpretation of study results which in turn influences clinical application. However, knowledge about the methods or paradigms used to investigate ML in stroke is limited. A detailed understanding of ML study protocols is necessary to inform stroke rehabilitation. In addition, although previous papers have summarized ML findings poststroke,,, and Maier et al have recently published a conceptual analysis demonstrating how some ML principles may influence neurorehabilitation efforts, there has been no systematic or comprehensive synthesis of the ML literature involving persons poststroke. For clinicians to be able to use the evidence to inform their practice, they need to be able to easily identify and compare all factors that can influence learning. Therefore, the primary objectives of our study are to (1) summarize and describe the approaches and methods used to investigate ML poststroke and (2) summarize the effects of different ML principles in promoting skill acquisition and retention in individuals’ poststroke. To facilitate the translation of ML research into clinical practice, our secondary objective was to evaluate the clinical utility of each ML principle for stroke rehabilitation.

Methods

The methodology for this review is based on the recommendations of Levac et al and the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension guidelines for scoping reviews. Three authors (S.G., T.S., L.C.) performed article screening, data extraction, quality assessment, and determined ease of clinical application. Discrepancies were resolved through discussion, and a fourth author (K.P.) was consulted if needed.

Data sources

We searched 5 databases for studies investigating ML in individuals poststroke on October 24, 2018 and again on June 23, 2020 from inception: Medline (1946 to present), Excerpta Medica Database (1947 to June 22, 2020), Allied and Complementary Medicine (1985 to June 2020), Cumulative Index to Nurisng and Allied Health Litrerature (1981 to present), and the Cochrane Central Register of Controlled Trials (2014 to present). Keywords and Medical Subjects Headings terms for the concepts of “stroke” and “motor learning” were searched using the AND operator, with human and English filters applied. The initial search strategy was created for Medline (table 1) in collaboration with an information specialist and translated for the other databases. A second information specialist evaluated the initial and translated searches and made suggestions to improve precision and sensitivity.

Table 1

Search strategy for Medline

No.	Searches	Results (October 24, 2018)	Results (June 23, 2020)
1	Cerebrovascular disorders/	45,411	46,338
2	Stroke rehabilitation/	11,212	13,166
3	Exp stroke/	117,205	122,630
4	(stroke* or poststroke or transient isch* or TIA or cerebrovascular* or CVA).ti,ab,kf.	266,053	11,630
5	((cerebral or cerebellar or brain* or vertebrobasilar or intracerebral) adj5 (infarct$ or isch?emi* or thrombo* or emboli* or apoplexy or occlus*)).ti,ab,kf.	91,346	99,567
6	((cerebral or intracerebral or intracranial or brain* or cerebellar) adj5 (h?emorrhage or h?ematoma* or bleed*)).ti,ab,kf.	44,664	49,589
7	exp hemiplegia/ or exp paresis/	18,552	19,384
8	(hemipleg* or hemipar* or paresis or paretic).ti,ab,kf.	33,874	36,283
9	or/1-8 [**stroke]	411,790	456,028
10	((motor or skill) adj5 (learn* or relearn* or acquisition or reacquisition)).ti,ab,kf.	12,733	14,585
11	((motor or skill) adj5 (retention* or transfer*)).ti,ab,kf.	2008	2292
12	Motor activity/	93,056	96,406
13	Motor skills/	22,652	24,183
14	Psychomotor performance/	59,569	63,629
15	Learning/	59,114	65,169
16	(12 or 13 or 14) and 15	6700	7223
17	10 or 11 or 16 [**motor learning]	18,233	20,573
18	9 and 17	1053	1212
19	18 not (exp animals/ not humans.sh.)	928	1073
20	Limit 19 to English	884	1028
21	(201810* or 201811* or 201812* or 2019* or 2020*).dt.		2,309,454
22	20 and 21		147

Study selection

Early versions of our selection criteria were trialed on a small sample of articles. Variability in study methods and definitions of ML necessitated 10 iterations to refine the selection criteria. Studies were included if they met the final inclusion criteria: (1) individuals poststroke older than 18 years; (2) investigated a ML principle; (3) investigated the learning of a clearly defined motor task; (4) measured motor task performance using the same practice conditions at multiple points throughout acquisition; (5) had a retention interval longer than the interval between each acquisition block (if 1 acquisition session) or each session (if multiple acquisition sessions); (6) the learned motor task was performed and measured at retention under different conditions from the acquisition session (eg, removal of feedback); and (7) the study was written in English. Studies were excluded if (1) the motor task practiced was related to communication, swallowing, eye movements, or perception (attention or sensation) or they were a (2) conference proceeding, review article, thesis, commentary, or protocol. No predefined set of ML principles were used to ensure the review captured the full breadth of the literature.

Titles and abstracts were screened initially, followed by full-text screen of the included abstracts. Full-text exclusion reasons were documented.

The studies cited in the included articles were searched in Scopus (January 24, 2019 and July 22, 2020). All new studies identified were screened with the process outlined above.

Data extraction

A data extraction template was created in Excel (version 16.26)a and piloted on 9 studies. Remaining studies were divided between reviewers; 1 extracted the data and the second reviewed the extraction for accuracy. Corresponding authors of included studies were contacted to clarify or gather ambiguous or missing data (3 instances).

The following data were extracted from each study: (1) study demographics: inclusion and exclusion criteria, sample size, age, sex, paretic side, stage of recovery; (2) ML paradigm: motor task practiced, acquisition conditions, dose of practice, retentional interval, and differences between acquisition and retention testing. We also evaluated if authors explicitly stated that the purpose of their study was to evaluate ML; and (3) ML principle: what was investigated (eg, feedback) and manipulated (eg, frequency of feedback). The ML principle was considered the independent variable for each study

Ease of clinical application

ML principles were rated on the ease of application in a clinical setting considering the equipment and technology employed in the studies, and whether these were easily accessible in clinical environments with limited financial resources and/or personnel. This scale was used to determine potential clinical utility and implementation of each ML principle. Separate ratings were given for (1) the ability to duplicate study paradigms in clinical settings and (2) the ease of applying the general principle of ML in clinical practice. Possible ratings were easy, moderate, and hard, and were determined through discussion with members of the research team, all of whom have clinical research experience.

Study quality

Risk of bias was assessed with the appropriate National Institutes of Health National Heart, Lung and Blood Institute quality assessment tool based on study methodology. One reviewer assessed study quality and a second reviewer checked for assessment accuracy and fit of the tool used.

Results

Selection of sources of evidence

In total, 7043 studies were identified through our database searches. After screening, 39 articles met our inclusion criteria (fig 1).

Fig 1

Search strategy and results based on Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for scoping reviews flow diagram.

Study demographics

The 39 included studies were published between 1999 and 2020. Participant demographics are in table 2. Sample sizes of people with stroke ranged from 4 to 91. Thirty-one studies (79%) included individuals with chronic stroke,21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48,57, 58, 59 6 studies (15%) enrolled individuals at least 3 months poststroke (late subacute),52, 53, 54, 55, 56, 57 and 2 studies (5%) enrolled participants starting in the early subacute phase., One-third of studies (n=13)21, 22, 23,,,,,,49, 48, 49,, included a healthy control group. Studies commonly excluded participants with cognitive impairments (n=31, 79%),21, 22, 23, 24, 25, 26, 27,,30, 31, 32, 33, 34,,,,,43, 44, 45, 46,48, 49, 50,52, 53, 54, 55, 56, 57, 58 or severe motor deficits (n=19, 49%).,,,,34, 35, 36, 37,,,,50, 51, 52, 53, 54, 55, 56,

Table 2

Demographics of participants in the included studies

First Author	Year	Study Country	Groups (n)	Mean Age ± SD, y	Sex, M/F	Stage of Recovery*	Paretic Side, L/R
Backhaus34	2018	Switzerland	Short nap (10)Long nap (10)No nap (10)	60.0±12.166.3±5.559.7±5.5	7/39/15/5	Chronic	6/44/64/6
Bonni35	2020	Italy	Stroke (8)†	54.1±11.5	6/2	Chronic	5/3
Bonuzzi21	2016	Brazil	Stroke (20)‡	65.2±9.3	12/8	Chronic	10/10
Bonuzzi36	2020	Brazil	R hemisphere (10)L hemisphere (10)	62.8±9.867.5±8.8	7/35/5	Chronic	10/00/10
Boyd48	2003	USA	Explicit info (5)No explicit info (5)	59.0±10.558.6±19.2	2/34/1	Chronic	3/21/4
Boyd57	2004	USA	Explicit info (5)No explicit info (5)	51.0±9.858.2±14.6	4/13/2	Chronic	4/14/1
Boyd47	2006	USA	BG explicit info (5)SM explicit info (5)BG no explicit info (5)SM no explicit info (5)	51.0±9.859.0±10.558.2±14.658.6±19.2	4/12/33/24/1	Chronic	4/13/24/11/4
Brodie24	2014	Canada	Active rTMS (10)§Sham rTMS (5)§	64.5±NR67.2±NR	8/23/2	Chronic	NRNR
Brodie25	2014	Canada	rTMS (11)Sham rTMS (11)	65.1±6.867.4±6.7	9/28/3	Chronic	5/65/6
Carey29	2007	USA	Track (10)‡	65.9±7.4	9/1	Chronic	5/5
Charalambous37	2018	USA	Treadmill walking (12)Cycle ergometer (12)Active control (13)	55.1±16.062.2±10.157.5±9.0	7/57/59/4	Chronic	7/58/46/7
Cirstea49	2003	Canada	Mild–moderate (10)||Moderate–severe (10)||	54.4±20.152.5±11.6	5/58/2	Late subacute–chronic	0/100/10
Dobkin55	2010	11 countries	Daily reinforcement (88)No reinforcement (91)	62.9±12.665.1±11.9	59%/41%¶	Early subacute	49%/51%¶
Doost38	2019	Belgium	Stroke (21)†	65.1±8.0	16/5	Chronic	12/9
Guttman31	2012	Israel	Stroke (13)†	68.9±4.9	10/3	Chronic	5/8
Hamoudi50	2018	Germany	Real tDCS (18)Sham tDCS (18)No training/tDCS (14)	61.6±361.9±364.7±2	12/615/38/6	Late subacute–chronic	10/89/97/7
Helm39	2020	USA	Variable (16)Constant (16)	58.7±11.362.3±9.7	12/47/9	Chronic	NRNR
Jo54	2020	Korea	Random (7)Blocked (7)	57.7±7.463.1±8.0	5/24/3	Late subacute–chronic	0/70/7
Lefebvre27	2013	Belgium	Stroke (18)†	61±9	12/6	Chronic	10/8
Lefebvre58	2015	Belgium	Stroke (19)†	65±10	16/3	Chronic	5/14
Lefebvre26	2017	Belgium	Stroke (22)†	64.7±9.8	18/4	Chronic	NR
Malouin45	2009	Canada	Mental practice (5)Cognitive practice (3)No training (4)	61.3±7.261.0±8.561. 8±9.5	3/23/04/0	Chronic	5/02/13/1
Nepveu40	2017	Canada	HIIT (11)No exercise (11)	64.7±11.665.0±11.3	6/510/1	Chronic	1/103/7/1#
Neva41	2019	Canada	M1 cTBS (12)S1 cTBS (13)Sham cTBS (12)	62.3±9.766.5±13.068.2±9.1	9/311/29/3	Chronic	6/66/75/7
Orrell22	2006	UK	Discovery learning (5)Errorless learning (5)	49.20±15.7154.60±12.16	4/15/0	Chronic	2/2/1⁎⁎3/1/1#
Ploughman56	2018	Canada	Stroke (10)†	58.2±14.9	4/6	Early subacute–chronic	2/8
Pohl46	2001	USA	Stroke (47)	71±6	29/18	Chronic	NR
Pollock59	2014	Canada	Stroke (4)	61.75±6.75	4/0	Chronic	3/1
Quattrocchi32	2017	UK	Reward (15)Punishment (15)Neutral (15)	58.9±3.156. 3±3.458.5±3.6	10/57/89/6	Chronic	8/79/69/6
Schuster52	2012	UK and Switzerland	Embedded imagery (13)Added imagery (12)No imagery (14)	65. 8±10.259.7±13.064.4±6.8	10/37/510/4	Late subacute–chronic	4/95/78/6
Schweighofer53	2011	USA	Blocked (12)Random (13)	61.25±13.9254.58±13.39	8/49/4	Late subacute–chronic	6/64/9
Siengsukon23	2009	USA/Canada	Explicit info—sleep (10)Explicit info—no sleep (10)Implicit info—sleep (10)Implicit info—no sleep (10)	62.2±10.359.8±13.762.9±10.565.4±15.4	6/45/56/43/7	Chronic	6/43/77/37/3
Takeuchi42	2012	Japan	rTMS unaffected (9)tDCS affected (9)Combined rTMS and tDCS (9)	64.0±5.863.4±6.757.0±10.2	6/35/46/3	Chronic	6/35/43/6
Tretriluxana51	2014	Thailand	Dyad (10)Individual (10)	50-70††50-70††	NR	Late subacute–chronic	NR
Tretriluxana30	2015	Thailand	6-min observation (6)1-min observation (6)	60.67±2.8164.83±7.52	NR	Chronic	NR
Vliet33	2017	Netherlands	Short-lasting online (20)Long-lasting offline (18)Short-lasting offline (21)Sham (21)	64±1159±960±862±11	15/512/68/1314/7	Chronic	11/911/79/1211/10
Wadden44	2019	Canada	M1 cTBS (9)S1 cTBS (11)Sham cTBS (8)	60.2±10.367.2±16.168.5±11.2	21/7||	Chronic	NR
Winstein43	1999	USA	Stroke (40)¶	57.1±11.1	26/14	Chronic	20/20
Zimerman28	2012	Germany	Stroke (12)†	58.3±NR	6/6	Chronic	7/5

NOTE. Only groups that matched the inclusion criteria are reported in this table.

Abbreviations: BG, basal ganglia; F, female; HIIT, high-intensity interval training; L, left; M, male; NR, not reported; R, right; SM, sensorimotor; tDCS, transcranial direct current stimulation; UK, United Kingdom; USA, United States of America.

Time poststroke is classifed based on the following categoreies: acute (<1wk), early subacute (1wk-3mo), late subacute (3-6mo), and chronic (>6mo).

Crossover study with all participants information grouped together.

Study contained other groups not reported.

Classified based on initial grouping.

Separated by Fugl-Meyer score: mild-moderate group range=63-50, moderate-severe group range=46-5.

Demographics of all stroke groups reported together.

Bilateral lesions.

Cerebellum lesion.

Reported as a range.

Methods and approaches used to investigate ML in people poststroke

Despite all included studies using a ML paradigm, only 64% stated that the purpose of their study was to investigate motor (skill) learning (n=19, 49%)22, 23, 24, 25, 26, 27,,37, 38, 39, 40,,48, 49, 50, 51,,, or learning of a motor task (n=6, 15%).,,,,, Details about the ML paradigms used are summarized in table 3.

Table 3

Motor learning paradigms

First Author	Year	Motor Task	Motor Task Performance Measure	Structure of Acquisition	Retention Interval
Upper-extremity motor tasks
Backhaus34	2018	Visuomotor adaptation task (joystick to targets with 110° rotation)	Accuracy (targets hit)	• Length of trial: 150 s* • No. of trials/sessions: 6 • No. of sessions: 1	45 min-2 h; 1 d
Bonni35	2020	Visuomotor adaptation task (joystick to target with 30° rotation)	Movement (angular) error	• No. of trials/blocks: 152 • No. of blocks:1 • No. of sessions: 1	45 min
Brodie24	2014	STT (move cursor between targets)	Response time; peak velocity; cumulative distance	• No. of trials/blocks: 72† • No. of blocks: 6 • No. of sessions: 6	1-3 d
Brodie25	2014	STT (move cursor between targets)	Reponses time (reaction and movement time combined)	• No. of trials/blocks: 72† • No. of blocks: 6 • No. of sessions: 5	1 d
Boyd48	2003	SRTT (press key corresponding to cued target)	Response time (with related change score)	• No. of trials/blocks: 10 • No. of blocks/sessions: 5 • No. of sessions: 3§	1 d
Boyd57	2004	CTT (matching rotation of lever to cued pattern)	Tracking error; lag time; tracking accuracy	• No. of trials/blocks: 10 • No. of blocks/sessions: 5 • No. of sessions: 3	1 d
Boyd47	2006	SRTT (press key corresponding to cued target); CTT (matching rotation of lever to cued pattern)	SRTT—response time; CTT—tracking errorTo compare tasks—change score	• No. of trials/blocks: 10 • No. of blocks/sessions: 5 • No. of sessions: 3	1 d
Carey29	2007	Matching finger and wrist potion to cued pattern	Accuracy score	• No. of reps/blocks: 3 • No. of blocks/sessions: 60 • No. of sessions: 10	3 mo
Cirstea49	2003	Reaching to target	Movement precision; movement time; movement segmentation; kinematics	• No. of trials/sessions: 70 • No. of sessions: 1	10 min
Doost38	2019	Bimanual circuit game (move cursor along complex path)	Bimanual speed-accuracy trade-off, bimanual coordination	• Length of trial: 30 s • No. of trials/sessions: 30 • No. of sessions: 1	1 wk
Hamoudi50	2018	Matching pinch force to cued pattern	Speed; accuracy (error rate)	• No. of reps/blocks: 20 • No. of blocks: 5 • No. of sessions: 5	3, 24, 52, 80, and 108 d
Jo54	2020	Spooning task	Movement time	• No. of trials/blocks: 15 • No. of blocks: 3 • No. of sessions: 9	3 wk
Lefebvre27	2013	Circuit game (move cursor along complex path)	Learning index (based on velocity and error)	• Length of trial: 30 s • No. of trials/sessions: 30 • No. of sessions: 1	30 min,|| 60 min, 1 wk
Lefebvre58	2015	Circuit game (move cursor along complex path)	Learning index (based on velocity and error)	• Length of trial: 30 s • No. of trials/sessions: 30 • No. of sessions: 1	1 wk
Lefebvre26	2017	Circuit game (move cursor along complex path)	Learning index (based on velocity and error)	• Length of trial: 30 s • No. of trials/sessions: 30 • No. of sessions: 1	30 min, 1 h, 1 wk
Nepveu40	2017	Handgrip task (match grip force to target)	Accuracy (time on target)	• No. of trials/blocks: 20 • No. of blocks: 5 • No. of sessions: 1	1 d
Neva41	2019	STT (move cursor between targets)	Total response time (reaction and movement time combined)	• No. of trials/blocks: 111† • No. of blocks: 4 • No. of sessions: 5	1 d
Pohl46	2001	Matching closing different switches to cued patterns	Sequence response time	• No. of trials/blocks: 10 • No. of blocks: 8† • No. of sessions: 1	1 d
Quattrocchi32	2017	Planar reaching to target with a force field perturbation	Difference between target angle and angular hand position (ie, angular error) at peak outward velocity	• No. of trials/blocks: 50 • No. of blocks/sessions: 7 • No. of sessions: 1	1 d
Schweighofer53	2011	Matching grip force to cued pattern	Normalized error of force trajectory	• No. of trials/blocks: 50 OR 3|| • No. of blocks/sessions: 3 or 50|| • No. of sessions: 1	1 d
Siengsukon23	2009	CTT (matching joystick position to cued pattern)	Tracking accuracy	• No. of trials/blocks: 10 • No. of blocks: 10 • No. of sessions: 1	12 h
Takeuchi42	2012	Pinching task	Pinch force; bimanual coordination	• Length of session: 15 min • No. of sessions: 1	1 wk
Tretriluxana51	2014	Bimanual cup stacking	Movement time; reaction time	• No. of trials/blocks: 5 • No. of blocks/sessions: 4 • Observation-practice ratio: 1:1 (if applicable) • No. of sessions: 1	1 d
Tretriluxana30	2015	Bimanual cup stacking	Movement time; reaction time	• No. (length) of trials: 4 (6min) or 24 (1min) • Length of block: 6 min • Observation-practice ratio: 1:1 • No. of sessions: 1	1 d
Vliet33	2017	Circuit game (move cursor along complex path)	Motor skill change (based on speed and errors)	• No. of trials/bocks: 5 • No. of blocks/sessions: 9 • No. of sessions: 1	1 wk
Wadden44	2019	STT (move cursor between targets)	Total response time (reaction and movement time combined)	• No. of trials/blocks: 102† • No. of blocks: 4 • No. of sessions: 5	1 d
Winstein43	1999	Matching planar elbow flexion/extension movements to cued pattern	Average difference between movement pattern and cued pattern; variable error (consistency)	• No. of trials/bocks: 99 • No. of blocks/sessions: 2 • No. of sessions: 1	1 d
Zimerman28	2012	SRTT (press key corresponding to cued target)	No. of correct sequences	• Length per trial: 3 min • No. of trials/sessions: 5 • No. of sessions: 1	90 min, 1 d, 3 mo
Lower-extremity motor tasks
Bonuzzi21	2016	Weight shifting to targets in standing	Complexity of game; no. of errors; no. of correct weight shifts	• Length of session: 30 min • No. of sessions: 4	1 wk
Bonuzzi36	2020	Weight shifting to targets in standing	Complexity of game; no. of errors; no. of correct weight shifts	• Length of trial: 10 min • No. of trials/sessions: 3 • No. of sessions: 4	1 wk
Charalambous37	2018	Gait (split-belt treadmill)	Step length symmetry index	• Length of session: 15 min • No. of sessions: 1	1 d
Dobkin55	2010	Walking as quickly as possible	Gait speed	• No. of trials/sessions: 1¶ • No. of sessions: daily to 8 wk	3, 6 mo§
Helm39	2020	Gait (split-belt treadmill)	Step length symmetry, limb phase symmetry	• Length sessions: 15 min • No. of sessions: 1	1 d
Malouin45	2009	Sit to stand	Loading of paretic leg (% of body weight)	• Length of session: 1 h • Imagery-practice ratio: CD# • No. of sessions: 12	3 wk
Orrell22	2006	Symmetry of standing; ability to keep board standing on stable	Degree the board tilts from horizontal	• Length of trial: 60 s • No. of trials/session: 24 • No. of sessions: 1	15 min, 1 wk
Ploughman56	2018	Gait	Cadence, velocity, % in double support, step length symmetry	• No. of trials: 4 passes • No. of sessions: 1	5, 20 min
Pollock59	2014	Reactive stepping in response to leaning outside base of support	Step velocity; step length	• No. of trials/blocks: 60 • No. of blocks/sessions: 2 • No. of sessions: 12	1 y
Schuster52	2012	Lie down on floor and then stand up	Speed	• Length of session: 45-50 min¶ • Imagery-practice ratio: 5:1 • No. of sessions: 6	2 wk
Upper- and lower-extremity motor tasks
Guttman31	2012	Sit to stand, reach to grasp	Time to stand; maximum reaching velocity	• Length of session: 15 min (imagery) • No. of sessions: 12	4 wk

Abbreviations: CD, cannot determine; CTT, continuous tracking task; SRTT, serial reaction time task; STT, serial targeting task.

Includes a block with random rotations.

Includes both random and repeated sequences.

‡One group only physical practiced during the first 2 sessions.

Data for this retention interval not reported.

Based on if groups are in randomized or blocked practice.

Practice embedded into PT session.

Repetitions varied per individual.

Motor task

Motor tasks practiced were broadly separated into upper-extremity (n=28, 72%) and lower-extremity (n=10, 26%) tasks. Seventeen upper-extremity studies included tasks whereby individuals moved to match a visually cued sequence.23, 24, 25,,,,,,,,,,,,,, In 5 of these studies participants moved along a predefined path,,,,, in 2 studies participants reached to a target,, in 2 studies participants completed bimanual cup stacking,, in 1 study participants practiced spooning food, and in 1 study participants practiced a pinching task. Of the 10 lower-extremity studies, 4 included balance tasks (weight shifting,, symmetry of standing, reactive balance), 4 investigated gait,,,, and 2 examined transitional movements (sit-to-stand, stand-to-lie-to-stand). One study included both upper-extremity (reach-to-grasp) and lower-extremity (sit-to-stand) tasks.

Measurement of motor task performance

Of the 28 upper-extremity studies, 9 quantified task performance by movement and/or reaction time,,,,,,,,, 9 measured the accuracy and/or error of movement,,,,,,,,, and 9 used a combination of both to investigate the speed-accuracy trade-off,26, 27,,,,,,, with 5 of these studies combining speed/accuracy values to create a single score.26, 27,,, One upper-extremity study compared the force and coordination between limbs. In the 11 studies that examined learning a lower-extremity motor task, 2 investigated the symmetry of limb loading,, 2 examined time to completion of functional movement,, 5 included spatiotemporal measures of stepping or walking,,,,, and 2 evaluated the accuracy of weight shifting toward targets in standing.,

Structure of acquisition

Dose of practice was reported by describing the number of trials/repetitions (n=23, 59%; range, 1-152), the time for each trial (n=9, 23%; range, 30s-6min), or duration of an acquisition session (n =7, 18%; range, 15min-1h). Most studies (n=25, 64%)22, 23, 24, 25, 26,28, 29, 30,33, 34, 35,,,,,,,46, 47, 48,,,,, had multiple practice blocks increasing the total dose of practice. Just over half of the studies (n=21, 54%),,26, 27,,,,,,37, 38, 39, 40,,,,,,,, had a single acquisition session, with 1 study having 56 sessions. Breaks were quantified as times between trials/repetitions (n=5, 13%; range, 10s-2 min),,,,, intervals between blocks (n=16, 41%, range, 30s-1h),,26, 27,,,,,,,,,,57, 58, 59, or time between sessions (n=2, 5%; range,1-3d)., Four studies (10%) outlined the presence of a rest period, without defining the duration or timing.,,, Finally, within motor imagery, mental practice, or action observation studies (n=5),,,,, 4 compared the amount of imagery/observation to physical practice.,,, Two studies had equal amounts of physical and observation/imagery practice (ie, 1:1 ratio),, 1 study included 5 imagery trials for every physical practice, and 1 study had no physical practice trials (imagery only).

Retention interval

The interval between the final acquisition session and the retention session of the included studies ranged from 10 minutes to 1 year. The most common retention interval was 1 day (n=16; 41%).,,,,,39, 40, 41,,,46, 47, 48,,, Two studies (5%) had variable retention intervals of 45 minutes to 2 hours or 1-3 days. Seven studies (18%) included multiple retention intervals.26, 27, 28, 29,,,

Principles influencing ML in stroke

Studies that manipulated similar ML principles were grouped to determine trends across studies. During the screening and subsequent data extraction, 2 additional patient-related factors emerged that did not strictly fit the definition of ML principles, however, were determined to influence ML. These factors could not necessarily be manipulated by the therapist but should be considered when delivering rehabilitation to improve ML. Details describing how ML was analyzed for each study with results are in supplemental table S1 (available online only at http://www.archives-pmr.org/). Summary details about the influence of ML principles on motor acquisition and retention with clinical utility ratings are in table 4.

Table 4

Summary table on the effect of ML principles on acquisition and retention of motor skills poststroke

First Author, Year	ML Principle Category	ML Manipulation Details	Effect on Acquisition	Effect on Retention	Simplified Conclusion	Ease of Implementation*
Bonuzzi, 202035	Stroke location	Hemisphere damaged	+†	+†	Side of stroke lesion does not affect ML.	ML: easyParadigm: moderate
Cirstea, 200349	Stroke severity	Severity of motor impairment	+†	+	Stroke severity influences the aspect of motor performance that changes.	ML: easyParadigm: easy
Pohl, 200146	Stroke severity	Severity of motor impairment	+‡	+‡,§	Greater motor performance improvements with mild vs moderate stroke.	ML: easyParadigm: hard
Bonuzzi, 201621	Task complexity and structure of practice	Task complexity increased during acquisition	+§	+§	Increasing task difficulty is effective in promoting motor improvements poststroke.	ML: easyParadigm: moderate
Helm, 202039	Task complexity and structure of practice	Practice structure (constant vs variable practice)	+†	+†	Both constant and variable practice can promote motor adaptation improvements.	ML: easyParadigm: hard
Jo, 202054	Task complexity and structure of practice	Practice structure (blocked vs random practice)	−†	−†	There was no difference in blocked or random practice on motor acquisition or retention.	ML: easyParadigm: easy
Orrell, 200622	Task complexity and structure of practice	Errorless vs discovery learning	CD	+†	Both errorless and discovery learning can promote motor performance improvements.	ML: easyParadigm: easy
Pollock, 201459	Task complexity and structure of practice	Task complexity increased during acquisition	+	+/−	Increasing task difficulty is effective in promoting motor improvements poststroke.	ML: easyParadigm: Moderate
Schweighofer, 201153	Task complexity and structure of practice	Practice structure (blocked vs random practice)	+†	+‡	Random practice structure is more effective than blocked practice in promoting motor improvements that are maintained overtime.	ML: easyParadigm: hard
Carey, 200729	Feedback	Feedback through telerehabilitation	+	+	Providing feedback is effective in promoting motor improvements poststroke.	ML: easyParadigm: moderate
Dobkin, 201055	Feedback	Daily feedback vs no feedback	+‡	+‡	Providing daily feedback is effective in promoting motor improvements poststroke.	ML: easyParadigm: easy
Ploughman, 201856	Feedback	Mode of feedback (tactile vs verbal)	+‡/†	−	More features of movement are improved with tactile compared to verbal feedback; however, improvements are not maintained long term.	ML: easyParadigm: easy
Quattrocchi, 201732	Feedback	Content of feedback (reward, punishment, neutral)	+†	+†	Positive and negative feedback is more effective at promoting motor improvement than neutral feedback.	ML: easyParadigm: hard
Winstein, 199943	Feedback	Frequency of feedback (100% vs 67%)	+†	−	Feedback helps promote ML; however, the frequency of feedback did not make a significant difference.	ML: easyParadigm: hard
Guttman, 201231	Motor imagery, mental practice, and action observation	Mental imagery without physical practice	+	+/−	Mental imagery without physical practice is effective in promoting motor improvements poststroke.	ML: easyParadigm: easy
Malouin, 200945	Motor imagery, mental practice, and action observation	Mental practice vs cognitive practice vs control	+‡	+	Mental practice is more effective in promoting motor improvements than practicing unrelated cognitive tasks.	ML: easyParadigm: easy
Schuster, 201252	Motor imagery, mental practice, and action observation	Timing of mental imagery (embedded vs consecutive, vs control)	+†	+†	The timing of mental imagery has no influence on ML.	ML: easyParadigm: easy
Tretriluxana, 201454	Motor imagery, mental practice, and action observation	Action observation with physical practice, vs physical practice alone	+†,‡	+†,‡	Action observation is effective in promoting improvements in movement time, but not reaction time poststroke.	ML: easyParadigm: easy
Tretriluxana, 201530	Motor imagery, mental practice, and action observation	Duration of action observation (6 vs 1min)	+†	+‡	Longer duration of observation is most effective in promoting motor improvements poststroke	ML: easyParadigm: easy
Boyd, 200348	Implicit vs Explicit Information	Explicit information vs no explicit information	+‡	NR‡	Provision of explicit information about the task is detrimental to ML poststroke.	ML: easyParadigm: moderate
Boyd, 200457	Implicit vs explicit information	Explicit information vs no explicit information	+†,‡	+†,‡	Provision of explicit information about the task is detrimental to ML poststroke.	ML: easyParadigm: moderate
Boyd, 200647	Implicit vs explicit information	Explicit information vs no explicit information, comparing lesion location	+‡	+‡,§	Regardless of lesion location or type of task, provision of explicit information about the task is detrimental to ML poststroke.	ML: easyParadigm: moderate
Charalambous, 201837	Aerobic exercise	Treadmill vs cycle ergometer vs active control	+†	+†	Neither intensity nor timing of exercise (as a primer) improves ML poststroke.	ML: moderateParadigm: moderate
Nepveu, 201740	Aerobic exercise	Exercise vs no exercise	+†	+/−‡	High-intensity interval training after motor training improves motor retention.	ML: moderateParadigm: hard
Backhaus, 201834	Sleep	Short vs long term napping, vs no napping	+†	+/−†	Napping does not enhance long-term retention poststroke.	ML: easyParadigm: moderate
Siengsukon, 200923	Sleep	Sleep vs no sleep	+CD	+‡	Sleep promotes the maintenance of motor improvements.	ML: easyParadigm: moderate
Bonni, 202035	Neurostimulation	Active vs sham iTBS over lateral cerebellum, prior to task practice	+‡	NR‡	iTBS prior over the lateral cerebellum prior to training improves ML poststroke.	ML: hardParadigm: hard
Brodie, 201424	Neurostimulation	Active vs sham rTMS over ipsilesional-S1, prior to task practice	+‡	+†,‡	Active rTMS over the ipsilesional-S1 prior to training improves ML poststroke.	ML: hardParadigm: hard
Brodie, 201425	Neurostimulation	Active vs sham rTMS over ipsilesional-S1, prior to task practice	+‡	+†,‡	Active rTMS over the ipsilesional-S1 prior to training improves ML poststroke.	ML: hardParadigm: hard
Doost, 201938	Neurostimulation	Active vs sham dual-tDCS over ipsilesional-M1, midacquisition	+†	+†	Active dual-tDCS does not enhance ML compared to sham stimulation.	ML: hardParadigm: hard
Hamoudi, 201850	Neurostimulation	Active vs sham tDCS over ipsilesional-M1, mid acquisition	+‡	+†	Active tDCS over the ipsilesional-M1 during training improves acquisition of motor skills poststroke.	ML: hardParadigm: hard
Lefebvre, 201327	Neurostimulation	Active vs sham dual-tDCS over ipsilesional-M1, mid acquisition	+‡	+‡	Active tDCS over the ipsilesional-M1 during training improves acquisition and retention of motor skills poststroke	ML: hardParadigm: hard
Lefebvre, 201558	Neurostimulation	Active vs sham dual-tDCS over ipsilesional-M1, mid acquisition	+‡	+‡	Active tDCS over the ipsilesional-M1 during training improves acquisition and retention of motor skills poststroke.	ML: hardParadigm: hard
Lefebvre, 201726	Neurostimulation	Active vs sham dual-tDCS over ipsilesional-M1, mid acquisition	+‡,§	+‡	Active tDCS over the ipsilesional-M1 during training improves acquisition and retention of motor skills poststroke	ML: hardParadigm: hard
Neva, 201941	Neurostimulation	Contralesional M1 vs S1 vs sham cTBS, prior to practice	NR†	+†	cTBS does not enhance motor acquisition or retention, regardless of the location of stimulation.	ML: hardParadigm: hard
Takeuchi, 201242	Neurostimulation	rTMS over contralesional vs tDCS over ipsilesional vs combined rTMS-tDCS	+/−‡	+/−‡	Combination of rTMS-tDCS may help promote ML more than a single type of stimulation.	ML: hardParadigm: hard
Vliet, 201733	Neurostimulation	Active vs sham tDCS over ipsilesional-M1, within various durations and timing	+†	+†	The amount or timing of bihemispheric tDCS does not influence the amount ML.	ML: hardParadigm: hard
Wadden, 201944	Neurostimulation	Contralesional M1 vs S1 vs sham cTBS, prior to practice	+†,‡	NR†	cTBS does not enhance motor acquisition or retention, regardless of the location of stimulation	ML: hardParadigm: hard
Zimerman, 201228	Neurostimulation	Active vs sham cathodal tDCS over contralesional-M1 during learning	+‡	+†,‡	Active cathodal tDCS over contralesion-M1 during training improves acquisition and shorter-term (up to 1d) retention but not long-term (3mo) retention of motor tasks.	ML: hardParadigm: hard

NOTE. + is improvement in motor performance during acquisition, maintenance in motor improvements at retention; − is no motor improvement noted during acquisition, loss of improvement at retention.

Abbreviations: CD, cannot determine; NR, not reported.

For the ease of implementation, “ML” is the ease of applying the general prinipcle of ML in clinical practice; and “paradigm” is the ease of duplicating the study paradigm in clinical settings.

No significant ML group differences.

Significant ML group differences.

Based-on reviewers’ observation of figure.

Patient-related factors

Stroke characteristics

Stroke characteristics evaluated were stroke severity and location. Different clinical outcome measures (ie, the Orpington Prognostic Scale and the Fugl-Meyer) were used to evaluate stroke severity. Two studies, showed that stroke severity relates to the magnitude of ML and influenced which features of task performance improved. Right versus left hemispheric strokes were compared in 1 study; stroke location did not significantly affect ML. Motor severity and lesion location were evaluated as principles easy to consider when implementing ML in clinical settings.

Sleep

Two studies investigated the effect of sleep on motor performance. One study found that a full night of sleep between acquisition and retention sessions enhanced motor retention of both implicitly and explicitly cued tasks. In contrast, the second study found that daytime napping does not enhance ML. Sleep was rated easy to monitor in a clinical setting.

ML paradigm conditions

Practice complexity

Two studies investigated the effects of increasing task complexity throughout the practice trials and found that this facilitated ML., One study evaluated if learning without error (less complex) was more effective than learning through error (more complex), but there was no significant difference in motor performance at retention. Practice complexity was manipulated using constant/blocked versus variable/random practice in 3 studies. Random/variable practice promoted maintenance of motor improvements in 1 study, but was not significantly different from constant/blocked practice in 2 studies., Study paradigms for practice complexity varied in rating for reproducibility; overall, however, it was determined to be easy to implement clinically.

Feedback

Feedback was investigated in 5 studies. Two studies, compared feedback to a no-feedback control; both studies found feedback facilitated ML. Furthermore, the mode (tactile vs verbal), frequency (100% vs 67%), and content (reward and punishment) of feedback influenced motor acquisition, but had smaller and less consistent effects on retention. ML paradigms varied in difficulty to duplicate; yet, it was consistently rated that it would be easy to manipulate how feedback is provided in clinical settings.

Motor imagery/mental practice and action observation

Five studies with components outside of typical physical practice—motor imagery, mental practice, and action observation—were grouped together. These principles were rated as easy to implement clinically and were all found to have a positive effect on ML.,,,, All studies had participants imagine/observe functional motor tasks. Improvements in motor performance were found irrespective of timing of mental imagery or if physical practice was performed. One study found that longer sessions of observation and practice promoted greater motor improvements during retention testing.

Implicit and explicit information

Three studies compared the effects of explicit versus implicit cues. All studies showed a negative effect of providing explicit information (ie, description of movement sequences to be learned) on ML.,, The task paradigm in these studies required technology that resulted in a rating of moderate difficulty to duplicate; however, providing implicit over explicit cues was rated easily implementable clinically.

Aerobic exercise

Two studies evaluated the effect of aerobic exercise either before or after motor task practice on ML with conflicting results. One study investigating a locomotor learning task found no effect of exercise on ML, whereas another study using a hand grasp-force task found that high-intensity interval training enhanced retention performance. Raters determined that it is moderately difficult to apply high-intensity aerobic exercise enhance ML in clinical practice.

Neurostimulation

Thirteen studies investigated the effect of different types of neurostimulation on ML including transcranial direct current stimulation (tDCS), repetitive transcranial magnetic stimulation (rTMS), continuous theta burst stimulation (cTBS), and intermittent theta burst stimulation (iTBS). Seven studies investigated the effect of 1mA active tDCS compared to sham stimulation during task practice. Montages included dual or bihemispheric stimulation with the anode over ispislesional-M1,,,, a single site of anodal stimulation over ispislesional-M1,, or a single site of cathodal stimulation over contralesional-M1. Most studies found a positive effect of tDCS on acquisition,,, and, at minimum, active stimulation helped improve short-term retention of motor skills.26, 27, 28, However, 2 studies showed no significant group differences between active and sham stimulation groups., Two studies, which included the same participants, investigated the influence of 5Hz rTMS over ipsilesional-S1 prior to task practice., rTMS promoted acquisition and retention of motor skills to a greater extent than sham stimulation. One study compared rTMS to tDCS, to the combination of both prior to practice, and found that rTMS-only initially decreased motor performance after stimulation. However, groups that contained rTMS maintained motor improvements better at retention. Two studies investigated the use of cTBS over S1 versus M1 stimulation prior to practice, neither enhanced ML greater than sham stimulation., Finally, iTBS over the cerebellum was found to enhance motor acquisition and retention greater than sham stimulation. Based on the technology, personnel, and training required it was rated hard to implement neurostimulation in clinical settings.

Study quality evaluation ratings are in table 5. Nineteen studies were evaluated using the tool for controlled interventions; 9 studies received a rating of good,,,, and 10 studies received a rating of fair.,,, Nine studies that were assessed used the pre-post study with no control intervention tool; 7 had a good rating,,26, 27, 28,,, and 2 had a fair rating. Finally, 11 studies were assessed using the tool for observational cohort and cross-sectional studies; 5 studies were rated good,,,,, and 6 studies were rated fair.,,,,,

Table 5

Risk of bias using the National Institutes of Health National Heart, Lung and Blood Institute quality assessment tools

Controlled Intervention Studies
First Author, Year	Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9	Q10	Q11	Q12	Q13	Q14	Overall Rating
	Described as Random-ized	Suitable Random-ization	Group Allocation Concealed	Study Participants and Providers Blinded	Assessors Blinded	Groups Similar at Baseline	Dropout <20%	Difference in Dropout Between Group<15%	High Adherence to Protocols	Other Treatment Avoided/Similar in Groups	Reliable and Valid Measure-ment Tools	Sample Size Large Enough for 80% Power	Outcomes Reported /Subgroups Analyzed Prespecified	Participants Analyzed Based on Initial Group Allocation
Backhaus, 201834	Y	Y	Y	NA	NA	Y	N	N	N	Y	Y	N	Y	Y	Fair
Bonni, 202035	Y	NR	NR	NA	NA	Y	NR	NR	Y	NR	Y	NR	Y	Y	Fair
Bonuzzi, 202036	N	NA	NA	NA	NA	Y	Y	Y	Y	Y	Y	NR	Y	Y	Good
Brodie, 201424	Y	Y	NR	NA	NA	Y	Y	Y	Y	Y	Y	NR	Y	Y	Good
Brodie, 201425	Y	Y	NR	NA	NA	Y	Y	Y	Y	Y	Y	N	Y	Y	Good
Charalambos, 201837	N	NR	NA	NA	NA	Y	NR	NR	Y	Y	Y	NR	Y	Y	Fair
Dobkin, 201055	Y	Y	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Y	Y	Fair
Doost,201938	Y	Y	NR	Y	NA	Y	Y	Y	Y	Y	Y	NR	Y	Y	Good
Hamoudi, 201850	Y	Y	NR	NR	NA	Y	Y	Y	Y	NR	Y	Y	Y	Y	Good
Helm, 202039	Y	NR	NR	N	N	Y	NR	NR	Y	NR	Y	Y	Y	Y	Fair
Jo, 202054	Y	N	N	N	N	Y	Y	Y	Y	Y	N	NR	Y	Y	Fair
Malouin, 200945	Y	Y	NR	NR	Y	NR	NR	Y	Y	Y	N	Y	Y	Y	Fair
Nepveu, 201740	Y	NR	NA	NA	NA	Y	Y	Y	Y	Y	Y	NR	Y	Y	Good
Neva, 201941	Y	Y	N	N	Y	Y	Y	Y	Y	NR	Y	NR	Y	Y	Good
Schuster, 201252	Y	Y	Y	N	Y	N	Y	Y	Y	Y	Y	Y	Y	Y	Fair
Siengsukon, 200923	Y	N	NR	NR	NA	Y	Y	Y	Y	Y	Y	N	Y	Y	Fair
Takeuchi, 201242	Y	NR	NR	NA	NA	Y	NR	NR	Y	Y	NR	NR	Y	Y	Fair
Vliet, 201733	Y	Y	CD	Y	Y	N	Y	Y	Y	NR	Y	Y	Y	Y	Good
Wadden, 201944	Y	Y	Y	NA	NA	Y	NR	NR	Y	Y	Y	NR	Y	Y	Good

Abbreviations: CD, cannot determine; N, no; NA, not applicable; NR, not reported; Y, yes.

Discussion

This scoping review found fair to good quality evidence regarding the influence of ML principles on the acquisition and retention of motor skills poststroke. Overall, we found variability in how ML paradigms were implemented in stroke research. Key ML principles and patient-related factors with consistent evidence of influence on ML were stroke severity, practice complexity, feedback, mental practice, action observation, implicit and explicit information. These principles, in addition to sleep, and practice schedule, were all rated as easy to apply in the clinical setting and thus have clinical utility. The evidence for the effect of sleep, practice schedules, aerobic exercise, and neurostimulation on ML is less clear. Based on the additional personnel and equipment required, aerobic exercise was determined moderately difficult to apply clinically, whereas all types of neurostimulation were determined to be difficult to apply clinically.

Variable methods and paradigms used to investigate ML poststroke

Most of the ML studies (n=31, 79%) were conducted with people with chronic stroke. This may limit clinical application because the subacute stage is when inpatient rehabilitation primarily occurs and where potential for neuroplasticity is highest. The type and measurement of motor tasks varied across studies. Most studies (n=28, 74%) evaluated ML in the upper extremity with novel tasks, which may reduce the ecological validity for rehabilitation poststroke. In contrast, lower-extremity (n=10, 26%) tasks were functional including walking, balance, and transitional movements. It is possible that ML is different for common lower-extremity tasks compared to novel upper-extremity tasks because task experience (ie, novice vs expert) influences learning in neurotypical populations. Because ML groupings consisted of both upper- and lower-extremity studies to meet study objectives, task differences may limit the comparability of studies. One study that evaluated ML in the upper- and lower-extremity showed that individuals improved motor performance for both tasks. However, both tasks were functional movements (reach-to-grasp and sit-to-stand) and the authors did not compare the magnitude or trajectory of learning between tasks.

Measurement of motor performance also differed between upper- and lower-extremity studies. Speed and/or accuracy measures were often used for upper-extremity tasks, whereas lower-extremity measurements were more variable. Studies assessing multiple components of motor performance found that movement accuracy and timing improved at the detriment of optimal kinematics. This phenomenon has also been observed in gait; although people with stroke can increase their gait velocity, they do so with a more atypical gait pattern. Thus, how motor performance is measured influences whether a change will be considered a positive ML outcome. The definition of motor improvement should be considered carefully when interpreting study findings, especially when applied clinically.

Magnitude of motor improvement exhibited by individuals with and without neurologic conditions is tightly linked to the amount of practice,, thus it should be considered when interpreting study results. Structure and reported dose of practice during acquisition varied greatly across the included studies. Most studies (n=23, 59%) reported the exact number of task repetitions, with additional information about the distribution of repetitions within sessions. However, some studies only provided the duration of a practice session, meaning the exact number of movements executed likely varied between participants. Previous work has shown that age, side affected, time poststroke, and stroke severity cannot be used to understand the varying amount of practice within a physiotherapy session. It is therefore challenging to determine how the dose of practice in the included studies affected our results. Finally, most of the studies had a single acquisition practice session. This was more common among upper-extremity studies (n=18 of 28) than lower-extremity studies (n=4 of 10). Considering the dose-response relation, it is unsurprising that many studies found only modest improvements in motor performance.

Clinicians aim to facilitate motor improvement during therapy that has lasting effect on a client's everyday function after rehabilitation is complete. Therefore, research that aims to investigate ML principles to be applied in stroke rehabilitation practice must include retention sessions. This is because retention testing enables researchers and clinicians to draw conclusions about ML (ie, relatively permanent changes in motor behavior) and distinguish it from motor performance which refers to temporary fluctuations in behavior that can be observed and measured during or immediately after the acquisition session. However, of the studies that reached full-text screening stage, 168 of 306 (55%) were excluded from our review for not meeting retention criterion. This included not having retention testing, or not having a change of practice condition at retention testing as recommended for ML studies. Within these excluded studies, additional ML principles (eg, focus of attention) were investigated in persons with stroke. However, the clinical implications of these findings for ML and stroke recovery were unclear because of a lack of retention testing. It is essential that future ML studies in stroke include retention testing to ensure conclusions about ML can be drawn to facilitate clinical application of the results. It is also important to consider the retention interval when evaluating permanence of motor improvement, as motor performance can decay as retention intervals increase. Of the included studies, 51% (n=20) had a retention interval of 1 day or shorter. Furthermore, of the 6 studies that showed improvements in task performance during acquisition, and had a retention interval of 3 weeks or longer,,,,,, 4 studies maintained improvements over the retention period,,,, and 2 studies showed a decline in task performance or a loss of any group differences during retention., Therefore, longer retention intervals or multiple retention sessions may provide meaningful information related to short- versus long-term recovery of motor skills poststroke.

Influence of ML principles on skill acquisition and retention and their clinical application

The large variability in ML paradigms made meta-analysis infeasible and complicated comparisons of study findings. However, we used qualitative summaries of trends across studies and clinical utility and implementation ratings to investigate how ML principles might be incorporated into clinical practice.

The heterogenous clinical presentation of stroke requires clinicians to individualize therapy to maximize outcomes. It was unsurprising to learn that stroke presentation also influences ML. Therefore, although clinicians do not manipulate stroke severity, because it is inherent to the client, it is important that stroke characteristics are considered when creating rehabilitation programs. We found that regardless of stroke severity or location, all individuals improved with task-specific practice and therefore should have access to rehabilitation. Although stroke location did not affect ML, stroke severity influenced the magnitude and mechanism of motor improvements., Individuals with moderate-severe strokes often improved motor performance through compensation (ie, increased trunk flexion during reaching). Thus, it is important to be mindful of how motor performance is measured within a study because the measure of task completion or movement quality can influence how clinicians use study results to guide their own practice with patients. Overall, it is recommended to consider the balance between task completion and movement quality in combination with patient goals and stroke characteristics when structuring therapy sessions to facilitate ML and recovery.

Sleep, like stroke severity, is not a factor easily manipulated by rehabilitation therapists. However, it is estimated that sleep is impaired in up to 78% of persons with stroke and thus is an important consideration when implementing a rehabilitation program. We found conflicting evidence about the effect of sleep on ML., This conflict may be because study protocol factors (ie, time of testing and task features) that mediate the influence of sleep on learning. However, a small meta-analysis looking at sleep on ML found that overall sleep does enhance ML in persons with stroke or other brain lesions, more so than healthy adults. Therefore, despite conflicting results in our study, and considering the relative ease of administering sleep questionnaires, clinicians may wish to evaluate sleep in persons with stroke and monitor the association with the motor outcomes of their interventions.

Our review found benefits to progressing the challenge of a task as people learn; this likely facilitates neuroadaptation. However, it is important to control the amount of challenge because individuals with stroke can perform better in transfer conditions when error is minimized. Therefore, clinicians should carefully titrate task difficulty to the individual's skill level to optimize ML; something that is feasible for clinicians to do. One example of a method to monitor task challenge is the NASA Task Load Index (NASA-TLX), a scale that is used to measure the work load efforts of a task on multiple domains including mental demands, physical demands, temporal demands, performance, effort, and frustration. ML research with neurotypical adults has found that optimal ML occurs with a NASA-TLX score of 51.5. Future research should determine if a NASA-TLX score of 51.5 is optimal to promote ML in the stroke population as well.

Practice complexity can also be manipulated through the structure (ie, variability) of practice. Motor performance during acquisition with random practice is similar to blocked practice,,, but 1 study exhibited less decay at retention with random practice. This is consistent with other ML studies in neurotypical adults and people poststroke, and is supported by the forgetting-reconstruction hypothesis, whereby each time an individual repeats a task they make stronger memory representations of the task that are easier to recall. This provides evidence that clinicians should consider how to disperse exercises/movements within a single therapy session. For example, for perturbation based-balance training, clinicians can randomize the direction of their external perturbations, instead of grouping all perturbations by direction, to increase the unpredictability and practice complexity. Together these studies highlight the importance of continuing to progress the difficulty of a task, making sure that the clients find the task challenging but not beyond their skill level, as well as varying task practice to facilitate improved long-term retention and generalization of motor performance.

It is clear from the 3 controlled studies that augmented feedback significantly improves both motor acquisition and retention poststroke.,, These benefits remain even when delivered remotely through telerehabilitation as seen in 1 included study, a promising finding for communities with limited rehabilitation resources. Our review found good quality evidence that certain features of feedback mediate ML responses. Meaningful feedback (rewarding or punishing) is more effective than neutral feedback. In addition, 100% feedback trended to be more effective for motor acquisition and retention compared to faded feedback on 67% of trials, which differs from ML in neurotypical populations. Finally, tactile feedback was associated with better motor acquisition compared to verbal cueing during practice in 1 study; differences were not maintained at retention. This may be a result of the small dose of practice in this study, so definitive conclusions about mode of feedback cannot be drawn at this time. In summary, feedback is beneficial for ML poststroke; however, further investigation is required to understand the optimal mode of delivery, frequency, and content because this information can guide clinician's provision of feedback to clients.

This review found fair to good quality evidence that motor imagery facilitates motor acquisition even with limited physical task practice.,, There were conflicting conclusions about the effect of motor imagery/mental practice on skill retention,,, which may be because of differences in acquisition parameters. Motor imagery can be easily applied in a clinical setting and it enables task practice when individuals may be too tired for physical activity or if they want to practice a motor task that is beyond their current capabilities. Improvements are thought to be mediated through similar cortical networks active during imagined compared to physical movement. Therefore, motor imagery/mental practice should be a technique that clinicians consider implementing to help facilitate ML.

This review also found action observation is beneficial for ML poststroke, which is consistent with results in neurotypical adults. Practice combined with observation was found to be more effective than physical practice alone for equal duration. In addition, longer observation was more effective than multiple shorter observation sessions, which may be because of certain mirror neurons only discharging after repeated observation. These studies provide evidence that individuals may benefit from repeated demonstration of movements to help them achieve motor goals.

Our review found strong evidence that regardless of the location of stroke (ie, basal ganglia or sensorimotor cortex) or type of task practiced (ie, continuous or discrete tasks), explicit provision of the sequence of task to be learned was detrimental to ML.,, This is opposite to the facilitatory effect of explicit cues in neurotypical adults. The cognitive load of remembering explicit instructions during task practice (ie, dual-task) may exceed the capacity of individuals with stroke and interfere with ML. An alternative explanation is that explicit information inhibits the creation of implicit memories, especially with more severe strokes. These consistent conclusions suggest that clinicians should be cognizant about how they instruct clients to complete an exercise/movement. As suggested by Boyd et al, this may involve orienting clients focus on components of movement that are atypical (eg, telling a client their knees fully extend too early during sit-to-stand) yet allow them to make their own corrections to improve the sequence of movement without explicit direction.

We found preliminary conflicting evidence on the effects of aerobic exercise on ML. One study of good quality found that high-intensity interval training after task practice modestly improved the neuroplastic response by mediating the amount of interhemispheric inhibition, thus enhancing ML. In contrast, the second included exercise study did not find a beneficial effect of high- compared to low-intensity exercise, though this could be a result of all groups showing adaptation to the split-belt treadmill paradigm used. Many studies have shown that aerobic exercise promotes neuroplasticity by increasing the production of brain derived neurotrophic factor in rats, healthy adults, and persons with stroke. Nonetheless, more work is needed to understand the benefits of aerobic exercise for ML before it is widely implemented clinically for ML purposes specifically.

There was good quality evidence that neurostimulation can enhance ML poststroke using various montages and techniques, before (for rTMS, iTBS) or during (for tDCS) practice. Compared to sham stimulation, cTBS was not found to enhance ML. Despite the benefits, it was consistently determined that neurostimulation is difficult to implement clinically. Future research evaluating how to improve the feasibility of neurostimulation implementation clinically is warranted.

Study limitations

Although our study consolidates information related to ML poststroke, there are limitations to this review. We acknowledge that there are ML principles (eg, focus of attention) that were not evaluated owing to a lack of suitable studies available in the stroke literature. In addition, some factors influencing ML were evaluated by a single study (eg, lesioned hemisphere) in this review, limiting the strength of our conclusions regarding clinical effect. Many of the included studies implemented a pre-post study design to investigate ML principles. Although this study design shows the magnitude ML, it fails to provide a full picture about the trajectory of learning. Understanding how people learn (illustrated by ML curves with multiple measures of motor performance throughout acquisition) in addition to the magnitude of learning can provide clinically useful information about the optimal dose or length of intervention. Finally, many included studies exhibited selection bias by excluding individuals with cognitive impairment or severe motor dysfunction. This limits generalizability of study findings because approximately 85% of people poststroke have cognitive impairment.,

Two ways to address some of the limitations in this review are to make the identification of ML studies easier by explicitly stating objectives related to ML, and have future studies be more consistent in implementation and reporting of ML paradigms and principles. Current poststroke ML studies have heterogenous designs, making it challenging to compare study findings. Variability also exists in the outcomes used to evaluate ML. For example, some studies evaluated retention session by comparing motor performance at retention to the baseline performance values (savings), whereas other studies reported differences from retention to end of acquisition training (forgetting). This variability leads to challenges interpreting ML study results, which may explain why many clinicians recognize the importance of ML principles on recovery but do not use them to guide their practice. Creating ML reporting guidelines, similar to the Consensus on Exercise Reporting Template used for the reporting of exercise programs, may address this issue.

Conclusions

Reporting guidelines for ML studies would be beneficial to enable easier comparison and interpretation of study findings as well as facilitate future meta-analyses to better inform clinical practice. Despite differences between ML study paradigms and clinical rehabilitation practice, the goal of permanent gains in skilled motor performance remains consistent. This review identified consistent evidence that ML poststroke is influenced by stroke severity, task complexity, motor imagery, action observation, and feedback that could be easily implemented in clinical practice. Other ML principles or patient-related factors influencing ML with conflicting evidence on ML effect—ie, sleep, practice schedules, and aerobic exercise—are worth further investigation given they also would be relatively easily applied clinically. This review also identified a considerable amount of research on various types of neurostimulation, yet the resulting effects on ML are varied and clinical implementation will be a challenge. Research on people in the subacute stage of recovery, with severe impairment, or cognitive deficits would be valuable for this research field.

Supplier

a. Excel, version 16.26; Microsoft Corporation.