Robot-Assisted Upper-Limb Rehabilitation After Stroke: A Systematic Review and Meta-Analysis of Cortical...

Robot-assisted upper-limb rehabilitation is widely used after stroke, but links to neuroplasticity biomarkers remain uncertain.

To synthesize biomarker evidence after robot-assisted upper-limb training and quantify post-intervention effects versus non-robot comparators.

We searched PubMed, Web of Science, Embase, Cochrane Library, EBSCOhost, and Scopus from inception to December 31, 2025. Randomized controlled trials (RCTs) and nonrandomized intervention studies enrolling adults with stroke were eligible if they reported at least one neuroplasticity biomarker. Prespecified comparison classes were used; only the primary contrast (robot-assisted training vs. non-robot control) was pooled at post using DerSimonian-Laird random-effects models. Effects were expressed as standardized mean differences (SMDs; Hedges' g) or mean differences (MDs) with 95% confidence intervals (CIs).

Fifty-five studies, including 24 RCTs, were included. Four RCTs contributed to the primary quantitative synthesis, and each pooled endpoint included two or three trials. Robot-assisted training was associated with improved ipsilesional resting motor threshold (RMT; SMD 1.77, 95% CI: 0.46-3.08; k = 2) and upper-limb impairment (Fugl-Meyer assessment-upper extremity [FMA-UE]; MD 4.48 points, 95% CI: 0.33-8.62; k = 3). Effects were uncertain for ipsilesional motor-evoked potential (MEP) amplitude (SMD 0.52, 95% CI: -0.31-1.34; k = 2) and activities of daily living (ADL). Heterogeneity was moderate to substantial, prediction intervals crossed the null, and certainty was low or very low.

Robot-assisted training may improve impairment and corticospinal excitability, but randomized biomarker evidence remains sparse and heterogeneous. Biomarkers should be interpreted as exploratory observations, not validated surrogates or causal evidence of cortical restitution.