Effects of brain-computer interface based training on post-stroke upper-limb rehabilitation: a meta-analysis.

The trials were small and did not agree with each other: for the main outcome, the measured disagreement between trials was 70%, so the pooled average is not what happens to any one patient. Most of the training-dose comparisons rested on fewer than five trials, which the authors themselves say is too few for a subgroup result to be relied on. The funnel plot was asymmetric, which suggests trials that found no benefit were less likely to be published. Every trial gave BCI training alongside conventional rehabilitation, and several combined it with electrical stimulation, a robot or visual feedback, so no trial shows how much of the improvement came from the BCI itself rather than from the other treatments given at the same time. The daily-living result flipped from no significant difference to a significant difference when one trial was removed, so that finding is not stable. The included trials did not record what medication patients were taking, so the authors could not check whether drugs influenced the results. The paper reports 18 trials as good quality and 5 as fair quality, which adds up to more trials than the 21 it included.

The article text supplied here contains no funding statement and no competing interests declaration, so it is not possible to say from this text who paid for or designed the work.

Do not treat BCI-based training as an established treatment for upper-limb recovery after stroke. The 3.69-point average gain came from 21 mostly small trials that disagreed with each other, the funnel plot suggests trials with negative results went unpublished, and the benefit did not extend to independence in daily activities.