Effects of brain–computer interface output modalities on motor recovery after stroke: a systematic review and network meta-analysis

Background The relative effects of different brain-computer interface output modalities for post-stroke motor rehabilitation remain unclear. Objectives To synthesize randomized evidence on brain–computer interface-assisted motor rehabilitation after stroke, with emphasis on functional electrical stimulation-triggered and robot-assisted output modalities and the certainty of evidence. Methods Sixteen randomized controlled trials published from January 2020 to October 2025 were identified from PubMed, Web of Science, Embase, IEEE Xplore, the Cochrane Library, and MEDLINE. Interventions included brain-computer interface-triggered functional electrical stimulation, brain-computer interface-controlled robotic rehabilitation, and conventional rehabilitation. The primary outcome was motor recovery measured by the Fugl-Meyer Assessment. A frequentist random-effects network meta-analysis was performed, and certainty of evidence was assessed using CINeMA. Results No trial directly compared the two active brain-computer interface modalities, resulting in a star-shaped network. Compared with conventional rehabilitation, brain-computer interface-triggered functional electrical stimulation improved motor recovery (standardized mean difference 0.57, 95% confidence interval 0.12 to 1.01), whereas brain-computer interface-controlled robotic rehabilitation showed an uncertain effect (0.17, −0.25 to 0.60). The indirect comparison numerically favored brain-computer interface-triggered functional electrical stimulation over brain-computer interface-controlled robotic rehabilitation, but the confidence interval crossed the null (0.39, −0.22 to 1.01). Certainty was limited by indirectness, imprecision, and heterogeneity. Conclusion Brain–computer interface-triggered functional electrical stimulation showed a clearer signal of benefit vs. conventional rehabilitation than brain-computer interface-controlled robotic. However, the comparison between the two active modalities relies entirely on indirect evidence from a star-shaped network, has very low confidence, and does not establish comparative superiority. P-score rankings are exploratory pending direct head-to-head trials.