Effect of robotic assistance on upper-limb sensorimotor recovery: a systematic review and meta-analysis

Abstract

Robot-assisted upper-limb neurorehabilitation enables several beneficial therapy components, including repetitive task-specific practice, high training intensity, real-time feedback, gamification, and continuous physical interaction with the patient’s limb. However, conventional comparisons of robot-assisted and therapist-delivered rehabilitation cannot determine whether the physical assistance provided by the robot contributes independently to recovery. This systematic review and meta-analysis, therefore, focused specifically on randomized controlled trials (RCTs) of upper-limb rehabilitation in which experimental and comparator interventions were designed to be similar except for the presence, or level, or mode of robotic assistance. PubMed, Scopus, and Google Scholar were searched through November 2025 for eligible studies. Study screening and filtration were performed using Covidence software. Thirteen studies met the eligibility criteria for qualitative synthesis, and nine studies involving 272 stroke survivors provided Fugl-Meyer Assessment (FMA) data for quantitative synthesis. The standardized mean difference (SMD) between the control and the experimental group for change in pre-to-post intervention FMA score was calculated to perform meta-analysis. Heterogeneity computation, sensitivity analysis, risk of bias, publication bias, and certainty of evidence evaluation were also performed. The main meta-analysis pooled effect (SMD = 0.425) favored the experimental robot-assisted condition across disease stages, assistance levels, and targeted limb segments. However, this overall effect was imprecise (95% CI: [−0.007, 0.858]) and showed substantial heterogeneity (62.62%). In studies restricted to chronic stroke, the pooled effect was larger (SMD = 0.53), although imprecision [0.067, 0.992] and heterogeneity (60.72%) remained similar. Estimates for shoulder-elbow and hand-focused interventions were small and imprecise too. Sensitivity analyses showed that the direction of the pooled effect was stable across various considerations. The certainty of evidence was judged to be low, primarily because of risk of bias concerns (mainly lack of double-blinding), inconsistency, and imprecision. Exploratory meta-regression did not identify clear study-level moderators of between-study heterogeneity. In addition, no significant publication bias was identified. Although this study provides preliminary evidence suggesting a positive independent influence of the robotic assistance on upper-limb recovery, the limited number of studies with their small sample sizes and substantial heterogeneity preclude a firm conclusion. Larger, adequately powered RCTs that isolate robotic physical assistance while standardizing other intervention components are needed to establish its independent causal contribution to upper-limb sensorimotor recovery.