Efficacy and safety of brain–computer interface for stroke rehabilitation: an overview of systematic review

Stroke (cerebrovascular accident) results from acute vascular occlusion or rupture causing brain damage and remains a leading global cause of disability and death. Conventional rehabilitation can be monotonous and less effective for severe hemiplegia, reducing motivation and limiting recovery. Brain–computer interface (BCI) technologies provide closed-loop, intention-driven rehabilitation with real-time feedback, potentially enhancing neuroplasticity and functional recovery. Despite rapid growth in BCI research and multiple systematic reviews/meta-analyses, their methodological quality varies, and no comprehensive re-evaluation had been performed. This umbrella review aims to critically reassess existing systematic reviews/meta-analyses on BCI in stroke rehabilitation to clarify efficacy, safety, methodological rigor, and gaps to guide future research and clinical practice.

The authors note a proliferation of systematic reviews and meta-analyses on BCI for stroke rehabilitation with heterogeneous scope and methodological quality. Prior evidence consistently suggests BCI improves upper-limb motor outcomes and may enhance daily living, particularly when combined with adjuncts such as functional/neuromuscular electrical stimulation or motor imagery. However, questions remain regarding effects on speech, lower-limb function, adjunctive neuromodulation (e.g., tDCS), and durability of benefits. No prior umbrella review had comprehensively re-evaluated the methodological quality (via AMSTAR-2 and PRISMA) of these reviews; thus, this work synthesizes the state of evidence and critically appraises quality across multiple dimensions, including publication year, study design, homogeneity, and publication bias.

Design: Umbrella review (overview of systematic reviews and meta-analyses). Registration: PROSPERO CRD42023407720; Clinical trial registration also referenced CRD42024562114. Databases and dates: Systematic searches of CNKI, WanFang, CBM, Cochrane Library, PubMed, Embase, and Web of Science from inception to October 14, 2024; citation tracking was also performed. Eligibility criteria: - Population: Stroke patients of any age, sex, race, and disease course. - Intervention: BCI or BCI combined with other therapies. - Comparator: Other active treatments (e.g., sham BCI, conventional rehabilitation, robotic therapy) without BCI. - Outcomes: Primary—motor function or activities of daily living (ADL) (e.g., Fugl–Mayer Assessment [FMA], Action Research Arm Test [ARAT], Modified Barthel Index [MBI]); Secondary—brain function without restriction on assessment tools (e.g., EEG, fMRI). - Study design: Systematic reviews or meta-analyses. Exclusions: Non-English/Chinese publications; RCTs/clinical trials (primary studies); duplicates; protocols; methodological-only studies; abstracts or translations of reviews only. Study selection and data extraction: After de-duplication, two independent reviewers screened titles/abstracts and full texts, with discrepancies resolved by a third reviewer. Extracted data included publication details, participant characteristics, interventions, comparators, outcomes, and conclusions of each review. Quality assessment: Methodological quality was evaluated using AMSTAR-2 (16 items; scored Yes=1, No=0, Partial/NA=0.5) and PRISMA (27-item checklist; Yes=1, No=0). Additional four dimensions were assessed: publication year (recency weighted higher), study design (RCTs and observational studies scored 18; other controlled trials scored 14), homogeneity (p ≥ 0.01 and I² ≤ 50% scored 18; otherwise lower), and publication bias (absence of forest and funnel plots indicated high risk). Scores were visualized with radar plots to compare studies across six dimensions.

Study selection and scope: - Total records: Databases yielded 706 records (Chinese 71; English 635). Citation searching identified 202 additional records. After screening and eligibility assessment, 18 systematic reviews/meta-analyses (published 2018–2024) were included. Methodological quality: - AMSTAR-2: Scores ranged 6–13. By critical domains (items 2, 4, 7, 9, 11, 13, 15), 12 studies were rated critically low and 6 low, indicating generally moderate-to-low confidence in the findings due to methodological limitations (e.g., lack of pre-registered protocols and lists of excluded studies). - PRISMA: Scores ranged 13–27; only six reviews provided a registration number; many abstracts lacked sufficient detail. - Radar plot synthesis: 11 studies showed relatively higher overall quality across dimensions; 7 were moderate. Efficacy and outcomes: - Upper-limb motor function: Consistent improvements with BCI versus controls/conventional rehabilitation across multiple reviews (e.g., FMA-UE, ARAT, WMFT, MAL), supported by 12 systematic reviews/meta-analyses. - Activities of daily living (ADL): Several reviews reported significant improvements in MBI with BCI-based interventions. - Disease phase: Greater benefits observed in subacute stroke across several reviews; one review found similar effects in subacute and chronic phases without significant differences. - Long-term outcomes: Mixed findings; earlier reviews noted insufficient evidence, whereas the latest meta-analysis (2024) suggested favorable long-term outcomes with BCI. - Adjuncts and comparators: No additional benefit of tDCS combined with BCI over BCI alone; BCI-robot systems can improve upper-limb motor function, though some meta-analytic comparisons showed no statistical differences versus control conditions. Safety: - Overall, BCI approaches were reported as safe with good tolerability; occasional minor adverse events (e.g., headache, nausea) noted in underlying RCTs require ongoing monitoring. Evidence gaps: - Limited and preliminary data for speech/aphasia rehabilitation and for lower-limb motor recovery. - Heterogeneity in interventions, dosing (e.g., training duration), and outcome measures; optimal dosing uncertain (some evidence suggesting <12 h total training may be advantageous).

This umbrella review addressed the core question of whether BCI-based rehabilitation is effective and safe for stroke recovery and how robust current systematic evidence is. The synthesis indicates that BCI—especially for upper-limb rehabilitation and often in combination with conventional therapy or motor imagery—produces clinically meaningful gains in motor outcomes and ADL, with the strongest effects in the subacute phase and acceptable safety. However, methodological shortcomings across many included reviews temper confidence in effect estimates. The absence of added benefit from tDCS above BCI alone suggests that BCI may be the primary driver of benefit, while the role of robotic assistance remains promising but not definitively superior. Long-term efficacy appears favorable in more recent analyses but remains to be corroborated by high-quality, long-duration follow-up studies. Collectively, these findings support integrating BCI into stroke rehabilitation pathways while prioritizing more rigorous, standardized research to refine patient selection, dosing, and adjunctive strategies.

This first umbrella review re-evaluating systematic reviews/meta-analyses on BCI for stroke rehabilitation finds that existing evidence—though methodologically moderate—supports BCI-combined treatments to improve upper-limb motor function and daily living quality, particularly in subacute stroke, with good safety. Evidence is insufficient or mixed for speech improvements, lower-limb motor recovery, and long-term outcomes, though the latest meta-analysis suggests favorable long-term effects. Future work should include multicenter, rigorously designed trials and reviews with pre-registered protocols, standardized outcomes, exploration of optimal training dose/timing, targeted studies on lower-limb and language outcomes, and longer-term follow-up to strengthen clinical guidance.

- Language restriction to English and Chinese and no search for printed/grey literature may have led to omission of relevant studies. - Only 18 reviews were included, with overall moderate methodological quality (many rated low or critically low on AMSTAR-2 domains), limiting confidence in pooled conclusions. - Potential reviewer subjectivity in quality scoring and evidence appraisal may introduce bias. - Considerable variability across reviews in interventions, dosing, populations, and outcome measures; many reviews lacked protocol registration and lists of excluded studies. - Some inconsistencies and heterogeneity in included data/reporting may affect generalizability and precision of effect estimates.