Efficacy of Action Observation Therapy on Cognitive Function in Stroke: A Systematic Review and Meta-Analysis

Stroke is a leading cause of neurological disability and mortality globally, with approximately one-third of survivors experiencing post-stroke cognitive impairment that negatively impacts functional recovery and quality of life. Cognitive rehabilitation aims to leverage neuroplasticity to restore function, and Action Observation Therapy (AOT)—which activates the mirror neuron system by having individuals observe goal-directed actions—has been proposed as a promising approach to enhance cognition. Given mixed findings across trials and the need for accessible, home-based interventions, this systematic review and meta-analysis aimed to evaluate the efficacy of AOT on clinical cognitive outcomes in adults after stroke compared with non-AOT interventions.

Background literature indicates that AOT activates an action observation network and the mirror neuron system (including premotor and parietal regions and prefrontal areas), potentially strengthening neural circuits and enhancing neuroplasticity relevant to cognition. Prior studies on AOT in stroke rehabilitation have largely focused on motor recovery, with emerging but inconsistent evidence for cognitive benefits. The review highlights that Montreal Cognitive Assessment (MoCA) may be more sensitive than MMSE for detecting post-stroke cognitive impairment. Previous systematic reviews noted variable effects likely due to heterogeneity in interventions, populations, and outcome measures, motivating a focused synthesis on cognitive outcomes after stroke.

The review followed PRISMA guidelines and was prospectively registered in PROSPERO (CRD42024571694). Databases searched from inception to May 6, 2024 included PubMed, Cochrane Library, Embase, Web of Science, EBSCO, CNKI, WanFang, and VIP, with supplementary checks of references and grey literature. Search terms combined cognition-related terms (e.g., cognition, executive function, attention, memory) with stroke terms and action observation. Inclusion criteria: adults (>18 years) with stroke; intervention involving AOT (alone or combined); comparator with non-AOT interventions (e.g., usual/standard care, placebo); randomized controlled trials; outcomes including cognitive assessments (MoCA and/or MMSE). Exclusions included insufficient data (after unsuccessful author contact), non-English/Chinese languages, descriptive or conference papers, and other non-RCT designs. Two reviewers independently screened, extracted data (study and participant characteristics, intervention details, outcomes), and assessed risk of bias using Cochrane ROB2, resolving disagreements with a third reviewer. Evidence certainty was appraised with GRADE, considering risk of bias, inconsistency, indirectness, imprecision, and other factors. Statistical analysis used RevMan 5.4 and Stata 18.0. Continuous outcomes were synthesized as mean difference (MD) or standardized mean difference with 95% CIs using random-effects models. DerSimonian-Laird random-effects results were converted post hoc to Hartung-Knapp-Sidik-Jonkman (HKSJ) estimates for more accurate error rates with few studies. Heterogeneity was assessed by I² with standard thresholds. Subgroup analysis and meta-regression explored sources of heterogeneity, examining intervention characteristics, minutes per session, number of weeks, overall minutes of therapy, and frequency. Sensitivity analyses sequentially omitted individual studies. Publication bias was assessed with funnel plots and Egger’s regression test.

Six RCTs (total n=400) were included; five reported MoCA outcomes and one reported only MMSE. Session lengths were 20–30 minutes, with total durations of 3–8 weeks. The random-effects meta-analysis of MoCA (5 studies; n=286) favored AOT over controls for improving cognitive function: MD = 2.71 (95% CI 1.28 to 4.13), I² = 90%, p < 0.01. After conversion to the HKSJ method, the 95% CI was 0.89 to 4.53; model t-statistic = 4.14, p < 0.01. Meta-regression did not identify significant moderators of heterogeneity (p > 0.05) for intervention characteristics, minutes per session, number of weeks, overall minutes of therapy, or frequency. Publication bias was not indicated by Egger’s test (t = 0.47, p = 0.67). Sensitivity analysis showed that omitting Wu et al. (2022) reduced heterogeneity substantially (I² = 1%, p = 0.39) while maintaining a significant pooled effect (MD = 3.00, 95% CI 2.56 to 3.44, p < 0.01), suggesting this study as a major source of heterogeneity. GRADE assessment rated the certainty of evidence as low (downgraded for risk of bias and imprecision).

The findings indicate that AOT produces clinically meaningful improvements in post-stroke cognitive function compared with non-AOT controls, addressing the study question regarding AOT’s efficacy for cognition. The discussion links these effects to the activation of the mirror neuron system and broader action observation network, including premotor, parietal, and prefrontal regions critical for working memory, attention, and action understanding—mechanisms that may enhance neuroplasticity and cognitive processing. Despite significant pooled benefits, high heterogeneity remained largely unexplained by meta-regression, likely due to the small number of studies, study-level variability (e.g., stroke type and stage), and limited statistical power. The results underscore the promise of AOT as an accessible, noninvasive, home-practicable cognitive intervention while emphasizing the need for rigorous, homogeneous trials to strengthen causal inference and generalizability.

AOT appears effective for improving cognitive function after stroke, but current evidence certainty is low and findings should be interpreted cautiously due to heterogeneity. Future research should include larger, high-quality, multicenter RCTs with rigorous randomization and allocation concealment, blinding of outcome assessment, adequate sample sizes, standardized intervention protocols, and extended follow-up to evaluate durability. Studies focusing on stroke subtypes and disease stages and incorporating objective neurophysiological or imaging measures are recommended to clarify mechanisms and optimize intervention parameters.

The review did not report inter-rater agreement for study selection, risk of bias, or GRADE assessments. Small sample sizes in included trials may over- or underestimate treatment effects, contributing to low certainty. Due to insufficient longitudinal data, analyses focused on immediate post-intervention outcomes, precluding conclusions about long-term effects of AOT on cognition.