Brain-computer interface for upper limb functional recovery post-stroke: a meta-analysis of improvement in upper limb motor function and changes in neurophysiological markers.

A systematic review and meta-analysis (PROSPERO-registered, PRISMA-guided) pooled 16 randomised controlled trials in which stroke survivors received BCI-based therapy — using EEG, functional near-infrared spectroscopy, or surface electromyography, with or without functional electrical stimulation — compared against conventional rehabilitation or a sham control. The primary outcome was upper limb motor function measured by the Fugl-Meyer Assessment for Upper Extremity; secondary outcomes were the Action Research Arm Test, the Modified Barthel Index (daily living), and neurophysiological markers. A random-effects model was used for pooling. Two reviewers independently screened studies and extracted data; three researchers assessed risk of bias with the Cochrane RoB 2.0 tool.

The primary outcome (FMA-UE) favoured BCI over conventional rehabilitation with a moderate effect (SMD = 0.49, 95% CI 0.26 to 0.73, p = 0.037; I² = 43.4%). Secondary outcomes also favoured BCI: ARAT (SMD = 0.53, 95% CI 0.14 to 0.91, p = 0.035), Modified Barthel Index for daily living (SMD = 0.86, 95% CI 0.59 to 1.12), and neurophysiological markers (SMD = 0.53, 95% CI 0.08 to 0.98, p = 0.021). In stroke-phase subgroups, the subacute phase showed the largest effect (SMD = 0.62, p = 0.031), the acute phase was non-significant with a wide confidence interval (SMD = 0.45, 95% CI −0.07 to 0.96), and the chronic phase did not reach significance (SMD = 0.40, p = 0.152). By session count, 20–30 BCI sessions produced a larger effect (SMD = 0.58) than 10–15 sessions (SMD = 0.37), though the difference between the two was not statistically significant. Among BCI modalities, motor-imagination BCI combined with functional electrical stimulation showed the largest effect (SMD = 0.60), followed by non-motor-imagination BCI (SMD = 0.50) and motor-imagination BCI with conventional rehab (SMD = 0.46); differences between modalities were not statistically significant.

If you are rehabilitating a stroke patient in the subacute window (roughly 2–8 weeks post-stroke), this meta-analysis suggests that adding BCI training to your existing upper limb programme may produce a moderate additional gain in FMA-UE and ARAT scores beyond conventional therapy alone. In the chronic phase the same addition did not reach significance, so the evidence does not support expecting a measurable motor benefit there. Some BCI protocols in the included trials also delivered functional electrical stimulation, so the pooled effect reflects a combined package rather than BCI in isolation. The 14-of-16 blinding gap means the true effect may be smaller than reported.

The Modified Barthel Index (a daily-living measure) showed a larger pooled effect (SMD = 0.86, 95% CI 0.59 to 1.12) than the motor-function scores, suggesting that BCI-assisted training may translate into meaningful gains in a patient's ability to feed, dress, and manage personal care. This is most relevant for subacute-phase patients in your caseload. As with the PT reading, the effect is an adjunct to, not a substitute for, the functional task practice you already deliver, and the chronic-phase data do not support a clear ADL benefit.

Only 16 trials were included, and most had small samples (typically fewer than 30 per group), so the pooled estimates have wide confidence intervals and limited precision. 14 of 16 trials could not blind participants or therapists to the intervention, which is inherent to BCI equipment but means the observed benefit may be inflated by expectation and attention effects. The chronic-phase subgroup did not reach statistical significance (p = 0.152), and the acute-phase subgroup had a wide confidence interval spanning zero, so the evidence for benefit in those phases is weak. Key intervention parameters (task difficulty, feedback mode, electrode placement) varied widely across the included trials, making it impossible to identify which BCI protocol features drive the effect. The analysis compares BCI against conventional rehabilitation only; it does not address BCI combined with neuromodulation (TMS, tDCS), which may interact differently. The authors note they did not perform a formal quantitative publication-bias test (e.g., funnel-plot regression) for every outcome, though funnel plots for the primary outcome showed only mild asymmetry.

Do not read this as evidence that BCI works equally well at every stage after stroke. The chronic-phase subgroup did not reach statistical significance (p = 0.152), and 14 of the 16 included trials could not blind patients or therapists to the intervention, which inflates the apparent benefit. The overall p-value for the primary outcome (0.037) is also close to the conventional threshold, so the effect should be treated as a modest, phase-dependent adjunct rather than a proven standalone treatment.