Behavioral Outcomes Following Brain-Computer Interface Intervention for Upper Extremity Rehabilitation in Stroke: A Randomized Controlled Trial

Abstract

Stroke is a leading cause of persistent upper extremity (UE) motor disability in adults. Brain-computer interface (BCI) intervention has demonstrated potential as a motor rehabilitation strategy for stroke survivors. This sub-analysis of ongoing clinical trial (NCT02098265) examines rehabilitative efficacy of this BCI design and seeks to identify stroke participant characteristics associated with behavioral improvement. Stroke participants (n = 21) with UE impairment were assessed using Action Research Arm Test (ARAT) and measures of function. Nine participants completed three assessments during the experimental BCI intervention period and at 1-month follow-up. Twelve other participants first completed three assessments over a parallel time-matched control period and then crossed over into the BCI intervention condition 1-month later. Participants who realized positive change (≥1 point) in total ARAT performance of the stroke affected UE between the first and third assessments of the intervention period were dichotomized as "responders" (<1 = "non-responders") and similarly analyzed. Of the 14 participants with room for ARAT improvement, 64% (9/14) showed some positive change at completion and approximately 43% (6/14) of the participants had changes of minimal detectable change (MDC = 3 pts) or minimally clinical important difference (MCID = 5.7 points). Participants with room for improvement in the primary outcome measure made significant mean gains in ARAT_total score at completion (ΔARAT_total = 2, p = 0.028) and 1-month follow-up (ΔARAT_total = 3.4, p = 0.0010), controlling for severity, gender, chronicity, and concordance. Secondary outcome measures, SIS_mobility, SIS_adl, SIS_strength, and 9HPT_affected, also showed significant improvement over time during intervention. Participants in intervention through follow-up showed a significantly increased improvement rate in SIS_strength compared to controls (p = 0.0117), controlling for severity, chronicity, gender, as well as the individual effects of time and intervention type. Participants who best responded to BCI intervention, as evaluated by ARAT score improvement, showed significantly increased outcome values through completion and follow-up for SIS_mobility (p = 0.0002, p = 0.002) and SIS_strength (p = 0.04995, p = 0.0483). These findings may suggest possible secondary outcome measure patterns indicative of increased improvement resulting from this BCI intervention regimen as well as demonstrating primary efficacy of this BCI design for treatment of UE impairment in stroke survivors. Clinical Trial Registration: ClinicalTrials.gov, NCT02098265.

Keywords: brain–computer interface (BCI); hemiparesis; motor function; recovery; rehabilitation; stroke; upper extremity.

FIGURE 1

Study design. The time-points at which neuroimaging data were collected are represented by Tl, control baseline 1; T2, control baseline 2; T3, control baseline 3; T4, therapy baseline; T5, mid-therapy; T6, post-therapy; and T7, one-month post-therapy. While the crossover control group (DTG) completed visits T1–T7, the immediate therapy group (ITG) completed visits T4–T7 only.

FIGURE 2

BCI intervention block design: (1) A pre-session open-loop screening task of two attempted and then two imagined grasping tasks (left, right, rest) is used to set control features (BCI classifier) for the forthcoming intervention task (Cursor Task). (2) The closed-loop cursor and target (visual only) intervention condition consists of at least 10 runs of 10 trials of attempted grasping movements for the purpose of guiding a virtual cursor (Ball) either left, or right as cued by the target (Goal) presentation on the horizontal edge of the screen. (3) Following 10 successfully completed runs of the visual only condition, adjuvant stimuli are added to enrich the feedback environment and facilitate volitional movement of the affected extremity (grasping). Subsequent runs are attempted at the preferred pace of the participant, completing as many runs as time allows. (4) With 15 min remaining in the 2-h intervention session, the participant is switched into the post-session open-loop screening task of two imagined and then two attempted grasping tasks (left, right, rest).

FIGURE 3

Intervention vs. control and responder vs. non-responder plots. Four of the most notably significant relationships are plotted with boxplots of all patient data overlaid by simple linear best fit lines to depict general trends in the data. A and B specifically demonstrate differences in the data between all controls (in red) and all interventions (in blue) whereas C and D represent trends in the data between responders (in orange) and non-responders (in green). (A) Although the improvement rate in ARAT for subjects in intervention was not significantly higher than controls, participants in intervention did significantly improve over time, and the trend of the boxplot medians suggests a possible continuation of improvement through follow-up not present in the control period. (B) Participants in intervention significantly improved faster over time in SIS_strength than those in the control period despite both groups starting at similar levels of ability. (C, D) Responders demonstrated significantly higher average SIS_mobility and SIS_strength scores than non-responders. This suggests patients with lower SIS_mobility and SIS_strength scores may not benefit from BCI intervention as well as those with higher scores.