Motor Learning in Older Adults with Mild Cognitive Impairment: A Systematic Review

Abstract

The purpose of this systematic review was to synthesize the current evidence on motor learning in mild cognitive impairment (MCI). A search of five databases returned a total of 6058 references, 10 of which met criteria for inclusion in this review. The existing evidence was notably variable with an overall moderate risk of bias. Eight articles compared behavioral motor learning outcomes in MCI and age matched, non-cognitively impaired (NCI) samples. In 37.5% of these studies, the degree of motor skill acquisition in the MCI group was statistically significantly less than in the NCI group. Skill retention was only compared between MCI and NCI samples in one article, which reported a relative reduction in MCI group performance following a 24-h, no-practice delay. Importantly, none of the included articles examined motor skill transfer. We discuss possible sources of heterogeneity among collective findings including variability in motor tasks, outcome measurement, and research design. Further research is needed to support a comprehensive understanding of motor learning in the early stages of age-related cognitive decline. Future investigations should emphasize functional motor tasks and clinically relevant learning outcomes, including retention and transfer of motor skills, while controlling for potentially confounding factors such as motivation and sleep performance. This systematic review was registered with PROSPERO international prospective register of systematic reviews (registration ID CRD42023417329).

Introduction

The global population is aging, contributing to an increased prevalence, burden, and duration of age-linked neurocognitive disorders such as mild cognitive impairment (MCI) (Rajan et al., 2021; Kassebaum, 2022). Currently, there are an estimated 12.2 million diagnosed cases of MCI in the U.S. with a projected increase to 21.5 million cases within the next 50 years (Rajan et al., 2021). MCI is characterized by age-related cognitive decline that often precedes dementia and is therefore regarded as a key marker for early cognitive intervention (Peterson et al., 2014). Peterson et al. (2014) clinical criteria for MCI, which resemble the DSM-5-TR criteria for mild neurocognitive disorder, include a change in cognitive status in the absence of dementia with (1) subjective cognitive complaint, (2) objective cognitive impairment, and (3) preserved independence in functional activities.

Several underlying etiologies have been identified in MCI, contributing to diverse clinical presentation and course (American Psychiatric Association [APA], 2022). Neurodegenerative etiologies include Alzheimer’s disease (AD), Lewy body disease, Parkinson’s disease (PD), Huntington’s disease, prion disease, and frontotemporal degeneration. Other etiologies include vascular disease, traumatic brain injury, substance/medication use, HIV infection, and multiple causes. Classically, MCI has been subtyped by the category and number of cognitive domain(s) impacted (Peterson et al., 2014). Primary subtypes include amnestic (aMCI) and non-amnestic (naMCI) MCI, and secondary subtypes include single-domain (sd) and multi-domain (md) (e.g., aMCI-sd). Recently, alternate classification models have been proposed to better predict MCI conversion to dementia using single-feature and multi-modal data (Machulda et al., 2017; Kikuchi et al., 2022).

A traditional paradigm has posited distinct trajectories for primary MCI subtypes, with aMCI conventionally viewed as the predominant precursor to Alzheimer’s dementia. In line with this proposal, Csukly et al. (2016) reported greater relative atrophy of medial temporal lobe (MTL) structures in aMCI; however, a recent review of differential and subtype-specific neuroimaging abnormalities in MCI revealed that MTL and white matter alterations manifest across amnestic and non-amnestic phenotypes, though the magnitude of these structural changes appears more pronounced in aMCI (Yeung, 2022). Similarly, the progression of MCI to dementia is seemingly nuanced; several predictors of conversion have been identified beyond primary subtype classification including multi-domain involvement, comorbid conditions, and demographic variables (Fischer et al., 2007; Chen et al., 2022 [review]; McGrattan et al., 2022 [review]; Menegon et al., 2024).

Nonetheless, neural signatures in MCI seemingly vary by etiology (Stephan et al., 2012 [review]); Yi et al., 2012; Wolz et al., 2016; Sasikumar et al., 2020 [review]). MCI due to AD is associated with abnormal amyloid deposition and decreased hippocampal volume (Wolz et al., 2016). In contrast, amyloid burden is not significantly elevated in PD-MCI though hippocampal atrophy is reported (Sasikumar et al., 2020). Gray matter atrophy in frontal and subcortical (caudate) regions is common to both PD-MCI and vascular MCI (Yi et al., 2012). Across etiologies, brain changes typically become more pronounced with symptom advancement to dementia, associated with disease advancement (neurodegenerative) or cumulative effects (vascular).

While MCI is distinguished from dementia, in part, by relatively intact functional performance, MCI appears to correspond with motor impairments beyond those expected in healthy aging. Impairments in static balance, gait speed, grip strength, and fine motor control have been reported (Bahureksa et al., 2017 [review]; Koppelmans et al., 2022 [review]). Notably, motor signatures in MCI may vary by primary subtype (Montero-Odasso et al., 2014). In a recent systematic review of clinical practice guidelines and consensus statements in MCI, Chen et al. (2021) found that one-third of consensus documents recommended screening patients’ motoric function, highlighting the clinical relevance of motor behaviors in cognitive aging.

Changes to motor performance and quality of life in age-related cognitive decline may be addressed through physical and occupational therapies (Zhu et al., 2015 [review]; Saúde et al., 2023 [review]). Moreover, older adults with cognitive decline may be referred to physical rehabilitation to address co-occurring medical conditions given that MCI is linked to several health risk factors such as sedentary lifestyle and sleep-disordered breathing with common comorbidities of vascular disease, diabetes, and depression (Rajan et al., 2021). Indeed, persons with MCI are more likely than non-cognitively impaired (NCI) age-matched controls to utilize inpatient, outpatient, and home health services (Lin et al., 2016). However, the effectiveness of physical training and rehabilitation in preventing motor dysfunction and benefitting functional performance in neurocognitive populations has been previously debated (Zhu et al., 2015; Kiper et al., 2022). The pathological, often progressive involvement of motor-associated cortical and subcortical areas in neurocognitive disorders may present barriers to integrating motor practice for long-term adjustment of motor behavior. Therefore, when considering the suitability of various motor interventions for older adults with MCI it is important to assess how cognitive aging affects motor learning capacity.

Motor learning has been defined, and consequently quantified, in various ways, likely contributing to heterogeneity among research findings within given populations (Ranganathan et al., 2022). Here, motor learning is defined as a relatively permanent change in motor performance resulting from practice or experience (Schmidt, 1976; Wolpert & Flanagan, 2010). Motor learning relies on the cohesion of several underlying cognitive processes including attentional states, movement planning, sensory integration, and feedback interpretation (Ronsse et al., 2011; Hudson & Landy, 2012; Song, 2019; Therrien & Wong, 2022). In dementias due to neurodegenerative diseases, such as AD or PD, cognitive processes underpinning motor learning are progressively disrupted (Aslan et al., 2021 [review]). Relatedly, behavioral motor learning outcomes including skill acquisition, retention, and transfer can be impaired in moderate to severe dementia, and the use of dementia-centered motor training approaches (e.g., implicit strategies) have been recommended (Ries et al., 2018 [review]; Palma et al., 2023 [review]).

In MCI, early structural changes can affect brain regions that are critical for motor learning processes. Frontal gray matter alterations may affect the initial “cognitive” phase of skill learning, during which anterior brain structures significantly contribute to conscious problem-solving and movement planning (Fitts & Posner, 1967; Durand-Ruel et al., 2023). Likewise, hippocampal and striatal changes might hinder motor memory consolidation, essential for skill refinement and retention (Albouy et al., 2013). Implications of neural changes are likely task specific; Krakauer et al. (2019) assert that explicit features of adaptation and sequence tasks have notable frontal and hippocampal dependencies. Still some research in amnestic populations, such as early work involving the patient H.M. (Milner, 1962), indicates that procedural learning remains relatively intact despite degradation (or in H.M.’s case, removal) of medial temporal lobe structures (Ruan et al., 2016 [review]; Ledig et al., 2018). Such variability in findings across cognitively impaired samples might be attributed to differences in underlying neural involvement, suggestive of the value in considering motor learning evidence according to cognitive diagnosis and severity.

Objective

There is a growing body of literature investigating motor learning in MCI populations. This research, however, has yet to be comprehensively and independently synthesized via systematic review. Therefore, the aim of this review was to systematically outline the current evidence on motor learning behavioral outcomes in older adults with MCI. Emphasis was placed on the presentation of all available evidence. Accordingly, articles with varied comparator groups, including healthy older adults and older adults with other age-linked neurological disorders, were considered. Summarized findings are intended to inform future research and elucidate the effectiveness of motor training in cognitively impaired older adults with potential application in rehabilitation, sport, and music.

Methods

Search Strategy

This systematic review was reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Page et al., 2021). Details of the protocol for this systematic review were registered on PROSPERO (registration ID CRD42023417329). Records were identified from five databases: MEDLINE (Epub Ahead of Print, In-Process and Other Non-Indexed Citations, Daily; Ovid interface), Embase (Elsevier interface), Cochrane Library (Cochrane Database of Systematic Reviews, Cochrane Central Register of Controlled Trials, Cochrane Methodology Register; Wiley interface), Scopus (Elsevier interface), and CINAHL (EBSCOhost interface). The initial search was conducted on September 19, 2022, and a search update was conducted on January 29, 2024.

All searches included indexed terms and free text words related to the primary search terms “cognitive impairment” (i.e., neurocognitive disorder, cognitive disorder, cognitive deficit, cognition disorder, cognitive dysfunction, dementia, amnestic, benign senescent forgetfulness, age-associated memory impairment, and age-related cognitive decline) and “motor learning” (i.e., motor skill learning, motor skill acquisition, motor skill retention, motor skill practice, movement skills, motor skill transfer, motor memory, procedural learning, motor consolidation, motor function, motor control, motor adaptation, and plasticity). Searches were supplemented by screening reference lists of identified studies, previous reviews, and known publications. Details of the search strategy by database are outlined in Supplemental Table 1 (see “Supplemental Information”).

Eligibility Criteria

Records included in this review were required to be full-text, peer-reviewed original research articles published in English. No publication date restrictions were applied. Randomized controlled trials, prospective cohort studies, and cross-sectional designs were considered for inclusion. Conference proceedings, abstracts, protocols, book chapters, and review articles were excluded from this review. Additional inclusion criteria were (1) study population of older adults (mean minimum age of 60 years at time of recruitment) with MCI and (2) at least one study outcome related to a behavioral measure of task-specific motor skill learning.

Population

Operationalized definitions of MCI have varied widely in the literature, contributing to discrepancies in research outcomes (Peterson et al., 2014). For this review, studies were considered for inclusion only if MCI was characterized (1) by qualified medical professional(s) (e.g., clinical neuropsychologist) or medical institution(s) or (2) with stated adherence to established MCI classification criteria. Here, established MCI classification criteria were considered those resembling Peterson, 2004 (or revised versions), the National Institute on Aging-Alzheimer’s Association (NIA-AA, Albert et al., 2011, 2013), the International Working Group on Mild Cognitive Impairment (Winblad et al., 2004), or the DSM- 5 (APA, 2013). Therefore, cognitively impaired samples classified exclusively by performance on a brief cognitive screen (e.g., Mini-Mental State Exam [MMSE, Folstein et al., 1975], Montreal Cognitive Assessment [MoCA, Nasreddine et al., 2005]) were excluded. In instances where a cognitively impaired sample was comprised of persons with MCI or dementia, we included only those studies in which the sample had > 50% MCI or in which subgroup analyses were performed. This review excluded studies involving non-human animals and older adult MCI samples with noted comorbidities (other than possible, probable, or known MCI etiology). To promote consideration of all current evidence, we included no criterion related to sub-classification of MCI; however, we acknowledge that motor learning outcomes may vary by subtype (i.e., aMCI/naMCI, sd/md) and have referenced these classifications in the reporting of results when available.

Outcomes

Eligible studies reported on at least one behavioral motor learning outcome. Motor learning was considered a subset of the broader category ‘procedural learning,’ which encompasses both motor and nonmotor tasks (Knowlton et al., 2017). Therefore, motor learning studies were distinguished from other procedural learning studies through consideration of task features. Behavioral outcomes of interest included motor skill acquisition, retention, consolidation, and transfer. More specifically, any study that measured change in task-specific motor performance upon repeated experience with a motor behavior and discussed results in the context of motor skill learning was considered for inclusion. As such, motor adaptation paradigms, which capture early error-driven motoric responses typically following task or environmental perturbation, also met this criteria (Izawa et al., 2008; Wei & Körding, 2009). Studies investigating motor learning through neural monitoring techniques alone, without mention of behavioral outcomes, were not included in this review. Additionally, research on the general effects of physical exercise or specific physical interventions (e.g., yoga, tai chi) in MCI were excluded.

Study Selection

Covidence systematic review software (Veritas Health Innovation, Melbourne, Australia) was used for study screening and selection. Duplicate records were removed automatically by the software. Title and abstracts were screened for relevance by two independent reviewers (K.T. & R.A.), and selected articles then underwent full-text review. Full-text eligibility was assessed, with consensus reached, by two assigned reviewers (K.T., R.A., E.S., or S.E.). Discrepancies were resolved through discussion or by a third reviewer as necessary.

Data Extraction

Articles meeting inclusion criteria underwent data extraction in Covidence. Each article was assigned to two independent reviewers (K.T., R.A., E.S., or S.E.) for data extraction. Data were collected using a standardized data extraction table with the following variables: first author, publication year, sample size, baseline participant and comparator characteristics (i.e., age, gender, cognitive status), characterization of cognitive impairment (i.e., diagnosis, classification criteria, or cognitive assessment), motor task, and outcomes. Discrepancies in collected data were resolved through consensus of assigned reviewers.

Quality Assessment

Two evaluators independently assessed the studies’ quality using the National Institutes of Health (NIH) Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies. The tool consists of fourteen questions (see Supplemental Table 2), each designed to evaluate a specific aspect of the study's methodology that could potentially introduce bias. As none of the included studies employed longitudinal design or blinding, Items 10 and 12 were omitted due to lack of relevance. The assessors scored the studies based on the number of “no” or “not specified” responses on any of the questions, such that studies with one “no” or “not specified” response were considered good quality, studies with two to four such responses were considered fair quality, and studies with more than four were considered poor quality. Consensus on study quality was reached through discussion between initial assessors with resolution by a third, independent reviewer as necessary.

Results

Study Selection

Details of the systematic searches are outlined in Fig. 1. A total of 6058 references were returned from Medline (1943), EMBASE (1703), Cochrane (628), Scopus (1126), and CINAHL (658) with 4090 references remaining after de-duplication in Covidence (Veritas Health Innovation, Melbourne, Australia). Upon title and abstract screen, 3899 articles were deemed unfit for inclusion. Of the 188 full-text reports assessed for eligibility, 68 and 76 articles did not meet population and outcome criteria, respectively. An additional 34 texts were excluded for not being peer-reviewed, full research articles available in English language. In total, 10 studies met criteria for inclusion in this review.

Fig. 1.

PRISMA flow diagram for searches of databases. Diagram retrieved from Page et al. (2021)

Interrater reliability (IRR) for the data extraction phase was evaluated using Cohen’s kappa (κ). A standardized coding scheme was developed to classify free-text data extraction responses into nominal categories. For data extraction items with multiple predetermined response options, agreement was assessed at the item level using a binary (yes/no) framework for each selection. Across all reviewer pairs, weighted κ = 0.774, indicating substantial agreement on data extraction from selected studies (Landis & Koch, 1977). Final extracted data had 100% agreement with all discrepancies resolved through discussion between reviewer pairs.

Study Characteristics

Included articles are summarized in Table 1. Detailed descriptions of motor tasks are provided in Supplemental Table 3. Screened studies that approached, but did not meet, inclusion criteria are listed in Supplemental Table 4. Study findings are primarily organized by behavioral motor learning outcome variables in the categories of 1) acquisition (Table 2) and 2) retention (Table 3). Note, findings are organized in accordance with definitions provided in this review (see “Acquisition” and “Retention and Transfer”) and may not reflect the original language of study authors (e.g., “retention” versus “one-week practice effects” [Schaefer & Duff, 2017]). Results are further organized into the motor task categories defined by Ranganathan et al. (2021) including (1) sequence, (2) adaptation, (3) coordination, and (4) applied tasks.

Table 1.

Summary of included articles

Author, year	Sample(s) (group – na | M age)	MCI criteriab	Motor task(s)c	Motor task category	Learning outcome variable(s)
Yan & Dick, 2006	MCI (amnestic) – 13 (f), 16 (m) | 73.9; NCI – 23 (f), 8 (m) | 77.4; Dementia (AD) – 10 (f), 18 (m) | 71.6	Peterson et al., 1999	Rapid aiming to fixed target	Coordination	Acquisition
Negash et al., 2007	MCI – 24 | 77.1; NCI (ϵ 4 carriers) – 11 | 78.5; NCI (ϵ 4 non-carriers) – 13 | 74.5	Peterson et al., 1999	SRT with implicit sequence	Sequence	Acquisition
Ameli et al., 2011	MCI (amnestic) – 3(f), 5 (m) | 67; Dementia (AD) – 4 (f), 4 (m) | 65	Peterson et al., 1999	Visuomotor mapping of object mass	Adaptation	Acquisition
Gobel et al., 2013	MCI (amnestic) – 8 (f), 5 (m) | 76; NCI – 14 (f), 6 (m) | 71.2; Parkinsons – 6 (f), 11(m) | 63.3	Peterson, 2004	Serial interception with implicit sequence	Sequence	Acquisition
Nemeth et al., 2013	MCI – 14 (f), 3 (m) | 61.8; NCI – 15 (f), 2 (m) | 57.8	Consensus by clinical neurologists and neuropsychologist	ASRT with implicit sequence	Sequence	Acquisition; retention
Schaefer & Duff, 2017	MCI (amnestic) – 26 (f), 28 (m) | 75.8	Winblad et al., 2004	Reach with spoon to-from targets	Applied	Acquisition; retention
Vecchio et al., 2018	MCI (amnestic) – 47 | 72.8; NCI – 22 | 68.3; Dementia (AD) – 17 | 70.1	Peterson et al., 2001; Portet et al., 2006	Visuomotor rotation of trackball cursor between central and radial targets	Adaptation	Acquisition
Hong et al., 2020	MCI – 7 (f), 13 (m) | 66.7; NCI – 14 (f), 8 (m) | 67.6;	Jak et al., 2009	SRT with implicit sequence	Sequence	Acquisition
De Wit et al., 2023	MCI (amnestic) – 81 | 73.4; NCI – 85 | 69.6	Albert et al., 2011	SRT with sequence; Mirror star tracing task	Sequence; Adaptation	Acquisition
Malek-Ahmadi et al., 2023	MCI (amnestic) – 20 (f), 15 (m) | 73.4; NCI – 33 (f), 21 (m) | 72.5; Dementia (AD) –18 (f), 14 (m) | 77.8	ADNI2, 2020	Reach with spoon to-from targets	Applied	Acquisition

NCI non-cognitively impaired older adults, AD Alzheimer’s disease, ε4 APOE allele, mixed CI mixed cognitively impaired sample, NINCDS-ADRDA National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association, NIA-AA National Institute on Aging – Alzheimer’s Association, RT reaction time, SRT serial RT, ASRT alternating SRT.

^aGroup sample size separated by biological sex, (f) = female and (m) = male, when reported.

^bEstablished MCI criteria used for subject inclusion, when reported. If established criteria are not noted, MCI characterization procedures meeting eligibility for this review are listed.

^cMore detailed descriptions of motor tasks are provided in Supplemental Table 3

Table 2.

Summary of motor skill acquisition outcomes

Author, year	$\text{Sample}\left(\text{s}\right)\left(\begin{array}{c}\mathit{\text{group}}\\ \left[n\right]\end{array}\right)$	Visit designb	Motor task(s)	Motor task outcome variable(s)	Within-groupc results	Between-groupd results	Effect size(s)e
Sequence Tasks
Negash et al., 2007	MCI (24) NCI – ϵ4 carriers (11) NCI – ϵ4 non-carriers (13)	Single-visit	SRT with implicit sequence	RT (block, patterned blocks, reversed blocks)	Overall skill acquisition – Improved RT with practice Implicit sequence acquisition – Acquired sequence-specific performance advantage	Overall skill acquisition – NCIϵ4 carriers = NCIϵ4 non-carriers = MCI Implicit sequence acquisition – NCIϵ4 carriers = NCIϵ4 non-carriers = MCI	–-
Gobel et al., 2013	aMCI (13) NCI (20) PD (17)	Single-visit	Serial interception with implicit sequence	Percent correct responses Cue speed (adapted to control task difficulty)	Acquired sequence-specific performance advantage	NCI = aMCI aMCI > PD	–-
Nemeth et al., 2013	MCI (17) NCI (17)	Multi-visit	ASRT with implicit sequence	RT (epoch, high frequency triplets, low frequency triplets)	Overall skill acquisition – Improved RT with practice Implicit sequence acquisition – Did not acquire sequence-specific performance advantage	Overall skill acquisition – NCI = MCI Implicit sequence acquisition – NCI > MCI	Overall skill acquisition – (EPOCH × GROUP interaction) η2 = 0.07 Implicit sequence acquisition – (TRIPLET × GROUP interaction) η2 = 0.21
Hong et al., 2020	MCI (20) NCI (22)	Single-visit	SRT with implicit sequence	RT (random block, repeated block)	Overall skill acquisition – Improved RT with practice in early blocks Implicit sequence acquisition – Acquired sequence-specific performance advantage in late blocks	Overall skill acquisition – NCI = MCI Implicit sequence acquisition – NCI = MCI	–-
De Wit et al., 2023	aMCI (81) NCI (85)	Single-visit	SRT with sequence	RT	–-	NCI = aMCI	g = 0.175h
Adaptation Tasks
Ameli et al., 2011	aMCI (8) AD (8)	Single-visit	Visuo-motor mapping of object mass	First peak in rate of grip force increase Peak in grip force Peak acceleration Peak in lift force	No change in ability to use color cue to scale grip force to object mass with practice	aMCI = AD	–-
Vecchio et al., 2018	aMCI (47) NCI (22) AD (17)	Single-visit	Visuomotor rotation of cursor between central and radial targets	Number of targets reached	Improved number of targets reached with practice	NCI > aMCI, AD	–-
De Wit et al., 2023 *	aMCI (81) NCI (85)	Single-visit	Mirror star tracing task (MTT)	MT Error	–-	NCI = aMCIg	MTT acquisition, error slope – g = −0.230h MTT acquisition, time slope – g = 0.243h
Coordination Tasks
Yan & Dick, 2006	aMCI (29) NCI (31) AD (28)	Multi-visit	Rapid aiming to fixed target	MT Jerk Percent primary submovementf	Improved MT, jerk, and percent primary submovement with practice	NCI = aMCI	–-
Applied Tasks
Schaefer & Duff, 2017	aMCI (54)	Multi-visit	Reach with spoon to-from targets	MT	Improved MT with practice	–-	η2 = 0.17
Malek-Ahmadi et al., 2023	aMCI (35) NCI (54) AD (32)	Single-visit	Reach with spoon to-from targets	MT	–-	NCI > aMCI, ADi Raw, but not age-adjusted, ISD showed acceptable diagnostic accuracy (AUC > 0.70) for differentiating NCI vs MCI	–-

MCI mild cognitive impairment, aMCI amnestic MCI, NCI non-cognitively impaired older adults, AD Alzheimer’s disease dementia, ε4 APOE allele, mixed CI mixed cognitively impaired sample; PD Parkinson’s disease, MT movement time, RT reaction time, ISD intrasubject standard deviation. –- data not available.

^*De Wit et al. (2023) article involved two motor tasks so is listed in table twice (results divided by task category).

^aTotal group sample size.

^bSingle-or-multi-visit design with reference to motor task performance (i.e., not including screening or imaging visits during which the motor task was not performed).

^cWithin-group findings only reported for MCI (or mixed CI) sample.

^dGroup compared using =, >, < when applicable based on statistical significance at study-defined α (α =0.05 unless otherwise indicated).

^eEffect sizes correspond to MCI within-group and/or MCI vs. NCI between-group comparisons (reported in table) only, when available.

^fProportion of total movement time contained between reach onset and first zero velocity crossing.

^gBased on reported lack of statistical significance, but authors note Hedge’s g was larger than the a priori set boundary of a meaningful between-group difference for MTT error slope (NCI < MCI) and MTT time slopes (NCI > MCI) after controlling for age and level of education.

^hAge and education adjusted value (authors also provide g without co-variates).

ⁱBased on raw ISD.

Table 3.

Summary of motor skill retention outcomes

Author, year	$\text{Sample}\left(\text{s}\right)\left(\begin{array}{c}\mathit{\text{group}}\\ \left[n\right]\end{array}\right)$	Motor task(s)	Visit designb	Motor task outcome variable(s)	Learning variable(s)	Within groupc results	Between groupd results	Effect size(s)e
Sequence tasks
Nemeth et al., 2013	MCI (17) NCI (17)	ASRT with implicit sequence	Multi-visit	ASRT (epoch, high frequency triplets, low frequency triplets)	24-h consolidation	Overall skill consolidation – Maintained RT across sessions Implicit sequence consolidation – Maintained sequence specific knowledge across sessions	Overall skill consolidation – NCI > MCIf Implicit sequence consolidation – NCI = MCI	Overall skill consolidation – (EPOCH × GROUP interaction) η2 = 0.12 Implicit sequence consolidation – (TRIPLET × EPOCH × GROUP) η2 = 0.05
Coordination tasks
Schaefer & Duff, 2017	aMCI (54)	Reach with spoon to-from targets	Multi-visit	MT	Retention after 1 week	Retained some MT improvements (Trial 1 Visit 1 < Trial 1 Visit 2 < Trail 10 Visit 1)	–-	η2 = 0.03

MCI mild cognitive impairment, aMCI amnestic MCI, NCI non-cognitively impaired older adults, AD Alzheimer’s disease dementia, ε4 APOE allele, mixed CI mixed cognitively impaired sample, PD Parkinson’s disease, MT movement time.

^aTotal group sample size.

^bSingle-or-multi-visit design with reference to motor task performance (i.e., not including screening or imaging visits during which the motor task was not performed).

^cWithin-group findings only reported for MCI (or mixed CI) sample.

^dGroup compared using =, >, < when applicable based on statistical significance at study-defined α (a = 0.05 unless otherwise indicated).

^eEffect sizes reported for MCI within-group and MCI vs. NCI between-group comparison only, when available.

^fAccounting for first and second part of block.

Quality Assessment

Quality assessment results are summarized in Table 4. Seven studies (Yan & Dick, 2006; Negash et al., 2007; Ameli et al., 2011; Nemeth et al., 2013; Schaefer & Duff, 2017; Vecchio et al., 2018; De Wit et al., 2023) were graded fair quality and three studies (Gobel et al., 2013; Hong et al., 2020; Malek-Ahmadi et al., 2023) were graded high quality. No studies were graded poor quality. All included reports clearly defined a research objective (Item 1), specified and defined the study population (Item 2), took place over a reasonable timeline (Item 7), consistently implemented valid and reliable exposure and outcome measures (Items 9 & 11), and lost no more than 20% of participants to follow-up (Item 13). Only five studies (Gobel et al., 2013; Nemeth et al., 2013; Schaefer & Duff, 2017; Hong et al., 2020; De Wit et al., 2023) provided variance and effect size estimates (Item 5). Reporting on participation rate was inconsistent (Item 3) and was only clearly specified in two studies (Hong et al., 2020; Malek-Ahmadi et al., 2023). Most studies addressed confounders by matching participants’ sex, age, and education level; however, two studies were deemed to have an inadequate adjustment for confounders (Item 14).

Table 4.

Summary of quality assessment

Consensus response for each relevant item on the quality assessment tool is indicated by , , and for “Yes”, “No”, and “Not Specified”, respectively. Overall score of 1 considered “good” quality, 2–4 “fair” quality, and > 4 “poor quality”. Note, quality assessment items 10 and 12 were excluded due to lack of relevance

Participant Characteristics

A total of 729 subjects were included across 10 studies. Of these, 340 subjects were classified as having MCI. Amnestic MCI subtype was specified in seven studies (Yan & Dick, 2006; Ameli et al., 2011; Gobel et al., 2013; Schaefer & Duff, 2017; Vecchio et al., 2018; De Wit et al., 2023; Malek-Ahmadi et al., 2023); subtype was unspecified in the remaining three studies. Peterson et al. (1999, 2001, 2004) criteria was most frequently used to classify the MCI sample. Information on cognitive screening measures and neuropsychological batteries administered to study participants, including mean scores by sample, are listed in Supplemental Table 5.

Common comparator groups included non-cognitively impaired (NCI) older adults, totaling 287 subjects, and older adults with dementia due to Alzheimer’s disease (AD), totaling 85 subjects. Only one study (Negash et al., 2007) categorized the NCI sample by Apolipoprotein E (APOE) genotype, a genetic factor that may affect risk of developing Alzheimer’s disease. Additionally, Gobel et al. (2013) included a comparator sample of older adults with Parkinson’s disease. Note, comparisons between MCI and NCI samples were emphasized in the reporting of results.

Motor Tasks

In total, motor learning outcomes were examined for 10 unique tasks. Using Ranganathan et al. (2021) model, motor tasks were grouped into sequence, adaptation, coordination, and applied task categories. Although some tasks could be considered under multiple categories, each task was assigned to only the most relevant category for clarity of results reporting.

Sequence Tasks

Most included studies examined motor sequence learning (Negash et al., 2007; Gobel et al., 2013; Nemeth et al., 2013; Hong et al., 2020; De Wit et al., 2023). All sequence tasks measured timing and accuracy of (physical or touch-screen) “key-press” responses which corresponded with set stimuli, presented in undisclosed sequences.

Key-press sequence tasks were most commonly employed with traditional or alternating serial reaction time (SRT) designs (Negash et al., 2007; Nemeth et al., 2013; Hong et al., 2020; De Wit et al., 2023). SRT designs typically require participants to perform a specific motor behavior, as quickly and accurately as possible, in response to a fixed set of visual or auditory stimuli. In addition to reaction time, SRT designs can be used to probe implicit motor sequence learning by comparing change in accuracy and response time on pseudorandom stimuli blocks and undisclosed sequence blocks over time. While sequenced SRT designs could be classified under multiple motor task categories given their outcomes relate to accuracy, reaction time, and implicit pattern learning, these designs are organized as sequence tasks in the present review. Of the included SRT designs, some authors (e.g., Negash et al., 2007) assessed learning of deterministic motor sequences while others (e.g., Nemeth et al., 2013) examined learning of probabilistic motor sequence elements. Uniquely, Gobel et al. (2013) studied sequence learning through a key-press design that diverged from other SRT approaches in this review; specifically, subjects were asked to precisely time a key press to the instant a descending on-screen cue entered a target zone. Still, similar to other tasks in this category, Gobel et al.’s (2013) behavioral outcome variables involved timing and accuracy of key press response.

Generally, on key-press sequence tasks, authors reported comparable learning outcomes between MCI and NCI samples. Specifically, most studies (Negash et al., 2007; Gobel et al., 2013; Hong et al., 2020; De Wit et al., 2023) found robust motor sequence learning in older adults with MCI despite their slower overall motor response times compared to NCI controls (see Supplemental Table 6). However, Nemeth et al. (2013), who employed an alternating SRT (ASRT) design, noted that probabilistic sequence learning outcomes in aMCI varied between early and late trials of a given block. These authors concluded that older adults with aMCI showed sequence learning similar to NCI controls on late-block, but not early-block, trials, suggesting possible impairment of motor sequence reactivation processes in aMCI.

Adaptation Tasks

Ranganathan et al. (2021) define motor adaptation tasks as those requiring individuals to respond to perturbations of typically well-learned movements. Three of the included studies assessed visuomotor adaptation. Both Vecchio et al. (2018) and De Wit et al. (2023) investigated the learning of upper extremity motor control with reliance on rotated visual feedback. Vecchio et al. (2018) compared the abilities of aMCI, NCI, and AD groups to control cursor movement between central and radial targets using a 90-degree rotated trackball. The behavioral outcome of interest included the number of targets reached. De Wit et al. (2023) compared aMCI and NCI groups’ learning on the mirror tracing task (MTT). The MTT involves tracing an image of a five-point star, as quickly and accurately as possible, using only visual feedback from a mirror (direct vision to hand and stimulus occluded). This task was classically administered to the amnestic patient H.M. (Corkin, 1968) and has since been widely administered as a measure of procedural learning.

The remaining task in this category examined visuomotor adaptation in MCI and AD samples using an arbitrary visuomotor mapping design (Ameli et al., 2011). This study assessed whether subjects could learn to modulate precision grip force for two different masses according to undisclosed assigned color cues only (vision occluded to information relevant to object mass). Peak grip force, peak acceleration, and peak lift force were compared across groups on colorcued and uncued (control) trials.

Findings from these three studies provide mixed evidence on learning of visuomotor tasks in MCI. Vecchio et al. (2018) found that, while performance on a visuomotor rotation task improved with practice in aMCI, practice-effects were more pronounced in the NCI sample; though, a between-group effect size was not reported. In contrast, De Wit et al. (2023) found “variable but overall trivial” learning differences between NCI and aMCI groups on a visuomotor rotation task (p. 113). Although these authors indicated that, after controlling for age and level of education, Hedge’s g was larger than their a priori set boundary for meaningful between-group difference for both time slope (g = 0.243) and error slope (g = −0.230), these differences were in opposite directions. That is, the MCI sample displayed less improvement in movement time but more improvement in movement accuracy compared to the NCI controls on the MTT task. Notably, the authors did not include a speed-accuracy tradeoff function. On the visuomotor mapping task, Ameli et al. (2011) found that aMCI and AD samples appropriately scaled grip force according to mass of the preceding lift; however, evidence did not support learning of arbitrary visuomotor associations in either clinical population. Conclusive findings in this category are limited by the small number of studies, lack of NCI comparator group in one study (Ameli et al., 2011), and general task heterogeneity.

Coordination Tasks and Applied Tasks

Coordination tasks involve production of a spatiotemporal pattern and typically require greater than one degree of freedom of movement (Ranganathan et al., 2021). Yan and Dick (2006) studied a dominant upper extremity coordination task involving rapid linear aiming between a fixed proximal and a distal target using a stylus pen. Generally, these authors reported that MCI and NCI samples made similar improvements in movement time, jerk, and percent primary submovement (defined as the proportion of total movement time contained between reach onset and first zero velocity crossing) with practice; although, effect sizes for within-group improvements were not reported.

Reaching behavior was further explored in two studies using a task with greater ecological validity. Schaefer and Duff (2017) found that, with practice, older adults with MCI had meaningful reductions in movement time on an applied reaching task, evidenced by a moderate within-group preto-posttest effect size (η² = 0.17). The task, adapted from the simulated feeding subtest of the Jebsen-Taylor Hand Function Test (Jebsen et al., 1969), requires use of the non-dominant upper extremity to acquire and transport raw kidney beans between central and radial targets as quickly and accurately as possible. When learning on this task was compared between MCI and NCI samples, however, the MCI group showed statistically greater variability of reach across training trials, potentially indicating group differences in learning mechanisms, strategy selection, or strategy persistence (Malek-Ahmadi et al., 2023).

Behavioral Motor Learning Outcomes

The included studies measured change in task-specific performance upon repeated experience with a motor task and discussed outcomes in the context of motor learning. Collective findings are organized into the categories of (1) motor skill acquisition and (2) motor skill retention and transfer.

Acquisition

For the purpose of this review, acquisition outcomes broadly refer to measures of online, so-called “practice-dependent” changes in motor performance that are attributed to active skill rehearsal. As such, results pertaining to online measures of motor adaptation are included here as acquisition outcomes. Motor skill adaptation has historically been defined as early, error-driven change in motor behavior usually occurring in relation to task or environmental perturbation (Izawa et al., 2008; Wei & Körding, 2009). The term acquisition, in contrast, has been used more generally to describe both initial and more gradual, practice-dependent improvements in motor behavior that are not ascribed to processes of biological maturation and that are often measured independent of task or environmental perturbation. The term acquisition as it relates to motor learning might be best reserved for multi-session studies in which some degree of permanence of the change in motor behavior can be established (i.e., skill learning); however, when acquisition is measured as a change in performance across consecutive online periods (e.g., multiple practice sessions) results contain influence of offline processes unless periods of non-practice are accounted for in the calculation of outcomes. Note that, in alignment with the motor task categories proposed by Ranganathan et al. (2021), the term “adaptation task” is used here to describe a category of motor tasks in which there is some response to perturbation of typically well-learned movements, regardless of temporal characteristics of practice or outcome measure.

Acquisition outcomes can be quantified using various approaches including (1) baseline-to-post-practice difference values (e.g., Keith et al., 2022) which are sometimes called”change scores,” (2) slopes of performance models or “learning curves” (e.g., De Wit et al., 2023), and (3) within-subject variabilities such as intrasubject standard deviation (ISD) (e.g., Malek-Ahmadi et al., 2023) calculated across online performance.

Table 2 shows a summary of acquisition outcomes. All of the included studies reported on motor skill acquisition. Eight studies compared acquisition outcomes between MCI and NCI samples, four of which involved comparator groups beyond NCI (Yan & Dick, 2006, Vecchio et al., 2018, Malek-Ahmadi et al., 2023 [Alzheimer’s dementia] and Gobel et al., 2013 [Parkinson’s disease]). The remaining studies either conducted only within-group analyses of acquisition outcomes (Schaefer & Duff, 2017) or compared MCI acquisition outcomes to an Alzheimer’s dementia sample only (Ameli et al., 2011).

Of the eight studies comparing acquisition in MCI and NCI samples, three reported a reduced degree of skill acquisition in the MCI group (Nemeth et al., 2013; Vecchio et al., 2018; Malek-Ahmadi et al., 2023). Motor task category and calculation of acquisition outcomes differed across these reports. Specifically, MCI samples displayed reduced or more variable acquisition of key press reaction time (moderate to large effect, η² = 0.21), center out visuomotor rotation with trackball (effect size not reported), and applied reaching tasks (Nemeth et al., 2013; Vecchio et al., 2018; Malek-Ahmadi et al., 2023) quantified reaching acquisition outcomes using ISD, and only raw, not age-adjusted, ISD showed acceptable diagnostic accuracy (AUC > 0.70) for differentiating MCI and NCI samples.

In contrast, five studies reported overall comparable acquisition outcomes between NCI and MCI samples on tasks including sequenced reaction time (three studies), visuomotor rotation, and fixed point-to-point reach (Yan & Dick, 2006; Negash et al., 2007; Gobel et al., 2013; Hong et al., 2020; De Wit et al., 2023). Although, De Wit et al. (2023) noted that while the p-values of between-group comparisons on the visuomotor rotation (mirror tracing task) did not meet the study-defined significance level, effect size values for these comparisons did fall within an a-priori-established range of a meaningful effect (i.e., Hedges’ g ≥ |2.00|). Additionally, Nemeth et al. (2013) concluded that MCI and NCI samples showed similar improvements in general reaction time on the SRT paradigm despite differences in acquisition of probabilistic sequence knowledge.

Importantly, while Schaefer and Duff (2017) and Ameli et al. (2011) did not include an NCI control group, these authors mention that NCI samples have demonstrated intact learning in prior investigations of the applied reach and arbitrary visuomotor mapping task, respectively. However, direct between-group comparisons across same-task studies that may otherwise methodologically differ were not regarded as conclusive evidence in the present review, due to marked potential for bias.

Retention and Transfer

Retention outcomes measure persistence of a motor performance following a period of non-practice (“retention interval”) (Schmidt et al., 2019). Although some authors consider the assessment of behavioral performance directly following skill rehearsal (i.e., immediate post-tests) as measures of “immediate retention”, only data collected after some no-practice delay were considered for inclusion in the present review (Schmidt et al., 2019). Under this definition, motor skill retention is influenced both by initial online acquisition and by offline processes such as motor memory consolidation (Schmid et al., 2020 [review]). Results specific to overnight consolidation of motor skills are organized under general retention outcomes here, given the paucity of literature. Notably, while retention tends to imply skill maintenance, processes of consolidation can actually result in spontaneous skill improvements over offline periods (Schmid et al., 2020). Transfer outcomes measure application of a rehearsed motor behavior in novel context (Mussgens et al., 2015; Schmidt et al., 2019). Motor skill transfer has been described for task or domain (e.g., near transfer, far transfer) and effector (e.g., intermanual transfer).

None of the included studies examined motor skill transfer. Two studies investigated motor skill retention in MCI (Nemeth et al., 2013; Schaefer & Duff, 2017). Of these, only Nemeth et al. (2013) compared retention outcomes between MCI and NCI samples. These authors reported that general reaction time improvements on an SRT task were maintained overnight in MCI; however, degree of consolidation in the 24-h offline period was statistically significantly less in MCI than in NCI controls (moderate to large effect,η² = 0.12). In contrast, Schaefer and Duff (2017) found a statistically significant within-group difference (small effect, η² = 0.03) in movement time (MT) of an applied reach at baseline and one-week follow-up (i.e., MT at baseline > MT at one-week follow-up), which the authors interpreted as preservation of trained skill over time. Importantly, this study lacked a comparator group, and did not compare performance between final training trials and retention trials, limiting interpretation of findings.

Discussion

This review outlined current evidence on motor learning in MCI. While previous reviews have summarized motor performance (Koppelmans et al., 2022) and learning (Palma et al., 2023) in mild and major neurocognitive disorders, the results of this review add to the literature a systematic synthesis of motor skill acquisition and retention in older adults with MCI classified in accordance with established clinical criteria. In addition to distinctions in population eligibility, this review diverges from a review by De Wit et al. (2021) by narrowing focus on motor learning, rather than the broader category of procedural learning which involves non-motor tasks. Only studies investigating motor tasks and a strictly-defined MCI population were considered for inclusion in the present review. Results are intended to inform motor training and physical rehabilitation in clinically-defined MCI.

In total, 10 studies were included. A summary of findings is presented in Fig. 2. Results of included studies were considerably heterogenous. Nearly 40% of the studies comparing motor skill acquisition in MCI and NCI samples reported some impairment of skill acquisition in the MCI sample (Nemeth et al., 2013; Vecchio et al., 2018; Malek-Ahmadi et al., 2023). Similarly, the single article comparing motor skill retention in MCI and NCI samples reported worse outcomes in MCI; however, a within-group investigation of retention in MCI suggested some preservation of trained skill overtime (Nemeth et al., 2013; Schaefer & Duff, 2017). Importantly, we acknowledge that a lack of statistical significance does not indicate lack of motor learning impairment, perse.

Fig. 2.

Summary of findings. Between-group findings on motor skill acquisition and retention in mild cognitive impairment (MCI) and non-cognitively impaired (NCI) samples. No included report examined motor skill transfer. Group comparisons based on statistical significance at study-defined α (a = 0.05 unless otherwise indicated). Articles referenced: 1. Negash et al. (2007); 2. Gobel et al. (2013); 3. Hong et al., (2020); 4. De Wit et al. (2023); 5. Yan and Dick (2006); 6. Nemeth et al. (2013); 7. Vecchio et al. (2018); 8. Malek-Ahmadi et al. (2023)

Variability in findings is likely influenced, in part, by differences in motor tasks and outcome measurement across studies. For example, differences in motor task complexity, even within a task category, could explain conflicting outcomes. Complex motor skills have been defined as those that cannot be mastered in a single session, have several degrees of freedom, and are ecologically valid (Wulf & Shea, 2002). Under this definition, authors who reported comparable learning in MCI and NCI samples tended to test relatively simple motor skills such as deterministic sequences and fixed, point-to-point reach; in contrast, authors who reported impaired learning in MCI assessed more complex skills such as probabilistic sequences, applied reach, and visuomotor transformation. Though arguably, variability in study findings may be better explained, generally, by differences in the degree of cognitive workload (mental effort) induced by motor tasks than by the tasks resemblance to some defined set of complex features (Ranchet et al., 2017).

While valuable for reconciling disparate findings, meta-analysis was not attempted with these data due to clinical and methodological heterogeneity. Deeks et al. (2023) suggest that meta-analysis should only be considered when samples, approaches, and outcome measures are similar enough to provide meaningful conclusions. Reports in this review have notable variability in (1) clinical and control group samples (i.e., clinical samples: unspecified MCI, aMCI, and mixed cognitively impaired samples; control samples: NCI with ϵ 4 carrier, non-carrier, and unspecified carrier status), (2) motor ‘intervention’ both in terms of motor task (e.g., rapid key press versus applied reach) and duration of task practice (single-session versus multi-session), and (3) quantification of outcome variables (e.g., as a change score, slope of a learning curve, or within-subject standard deviation). To address the limitation of not completing a meta-analysis, we provide effect sizes to contextualize statistical comparisons when available. Meta-analyses of motor learning in MCI should be pursued once research demonstrates greater homogeneity.

Beyond variability in findings, the ability to make conclusive statements is limited by a general paucity of research. Little of the included research investigated retention or multi-session outcomes, and no study examined motor skill transfer. Importantly, such outcomes have high relevance to motor training in real-world settings. Training in music, sport, and motor rehabilitation often involves more than one session of rehearsal, raising considerations of motor memory reactivation and consolidation. Moreover, the ability to utilize a motor skill at a later timepoint or in a novel context is foundational to the functional relevance of skill learning. It is worth noting that some studies investigating such outcomes were excluded from this review (see Supplemental Table 4). Namely, Yan et al. (2010) and Schaefer et al. (2015) investigated 24-h consolidation and 1-month retention of reaching tasks, respectively, in cognitively impaired older adult samples; however, these articles were not included here as their classification of cognitive impairment did not meet eligibility criteria.

Future studies should build on the included research in MCI samples classified using established criteria and specified clinical subtype (i.e., aMCI/naMCI, sd/md) when possible. MCI is a stand-alone classification, distinct from dementia, that can remain stable overtime. An estimated 14–38% of cases of MCI do not progress, a statistic which highlights the value of investigating motor learning in this population independently from dementia (Lopez et al., 2012; Peterson et al., 2018). If grouping older adults with MCI and dementia into a single cognitively impaired research sample, it may be useful to consider underlying etiology and to complete sub-group analyses.

Notably, this review did not limit inclusion of articles to those with healthy controls. Diverse comparator samples were considered in an effort to elucidate whether motor learning capacity parallels the continuum of cognitive and motoric presentations in aging. Comparison of motor learning behaviors in MCI against those in better-researched, relevant patient populations (e.g., PD and AD) could inform motor rehabilitation in MCI. Although debated, clinical subtypes of MCI might differentially progress. aMCI-md may more commonly progress to Alzheimer’s dementia, while naMCI may more likely to advance to Parkinson’s or vascular dementia (Jungwirth et al., 2012; Tifratene et al., 2015; Dugger et al., 2015; Csukly et al., 2016; Caffò et al., 2022). Given current evidence suggests AD and PD populations benefit from divergent motor training approaches, aMCI and naMCI subtypes may have unique motor learning needs (van Halteren-van Tilborg et al., 2007; Olson et al., 2019). Motor training approaches and practice schedules with evidenced application in AD and PD populations may or may not have relevance in MCI. The present review is limited in the discussion of outcomes by clinical subtype and etiologic classifications due to limited reporting; organization of findings by such subgroups would likely clarify conclusions.

Definitions of outcomes of interest in this review (e.g., acquisition, retention, transfer) differ elsewhere in the literature. In most instances, study designs of the included articles allowed for suitable categorization of outcomes into “Acquisition” and/or “Retention and Transfer” categories as defined in this review with one notable exception. Findings by Yan and Dick (2006) were treated as “Acquisition” outcomes despite likely influence of offline periods which appear unadjusted for in the multi-session design. This categorization may misrepresent the MCI samples capacity for online practice gains and should be considered a limitation of this review.

Other limitations include potential for study selection and outcome reporting biases. Efforts were taken to minimize such biases including prospectively registering the protocol in PROSPERO database, adhering to PRISMA review guidelines, and reporting outcomes regardless of statistical significance. Review inclusion criteria or search strategy limitations may limit generalizability of findings. Specifically, we acknowledge the possibility for language bias with the exclusion of articles not available in English. Additionally, we were unable to retrieve three manuscripts and, at times, were unsuccessful in contacting corresponding authors, resulting in the exclusion of potentially eligible data. We acknowledge that the studies included in this review may not represent the full range of available evidence. Inadvertently, all included articles examined upper extremity learning. Reports in non-human subjects were excluded from analysis, notwithstanding evidence of impaired motor learning in murine models of dementia and potential exacerbation of these deficits under pharmacological intervention (Ewers et al., 2006; Wagner et al., 2019; Niikura et al., 2022).

Collective findings of the included studies may support the possibility of a motor learning impairment in MCI; however, the methodologies and results of current literature vary substantially. This observation is congruent with the conclusions of a scoping review by Palma et al. (2023) and expands on this work by identifying additional articles specific to MCI that meet rigorous classification criteria. Results of this review elucidate future research directions specific to early cognitive decline in aging. The effectiveness of motor rehearsal in aging, including that in physical and occupational therapies, may be influenced by presence of MCI; however, the extent to which learning outcomes are affected likely varies by motor task and training approach. Current evidence is insufficient to establish conclusive associations between MCI and motor learning processes in the context of physical rehabilitation, thereby limiting evidenced-based recommendations to optimize motor training in clinical settings.

In general, additional research is needed to clarify the effect of MCI on the various stages of motor learning and to elucidate contributing mechanisms, if any, according to clinical presentation and underlying pathology. Future studies should evaluate the extent to which task-specific explicit and implicit processes are affected by early structural and functional brain changes in MCI. Additionally, future research should identify variables moderating differences in existing research findings such as task complexity, training approach, and training outcome with consideration of factors such as changes in sleep, use of feedback, motivation, and self-efficacy in MCI.

Conclusion

The current evidence on motor learning in MCI is variable but may suggest a relationship between cognitive status and learning outcomes in aging. Some studies reported impaired motor learning outcomes in MCI while others reported comparable outcomes in MCI and NCI samples. Variability in study findings may be attributed to differences in research design or motor task complexity. Quality assessment of the literature revealed an overall moderate risk of bias, and we therefore advise that our interpretations of the collective findings be considered with caution. Further research is recommended to support a more comprehensive understanding of motor learning capacity across the cognitive aging continuum. Future investigations should emphasize functionally relevant motor skills, training schedules, and learning outcomes (e.g., delayed retention and skill transfer) and provide effect sizes to contextualize statistical comparisons when possible.

Supplementary Material

The online version contains supplementary material available at https://doi.org/10.1007/s11065-025-09661-x.

Data Availability

Datasets and tables generated during the review are available in Open Science Framework repository https://osf.io/bjs3c/?view_only=7c784f8aa87b42009249e8583a18fe1c