Soft Robotic Glove Versus GRASP in Subacute Stroke: A Dose and Selection Question

Clinical Question (PICO)

Among medically stable adults within 6 months of ischemic or hemorrhagic stroke with unilateral arm impairment, does an occupational-therapy-assistant-supervised sensor-assisted soft robotic (SSR) hand glove, added to conventional occupational therapy, improve Fugl-Meyer Assessment of the Upper Extremity (FMA-UE) more than the same duration of supervised Graded Repetitive Arm Supplementary Programme (GRASP), also added to conventional therapy? The primary comparison was the overall randomized cohort at the end of training (T2); no subgroup benefit was prespecified. A combined severity/exposure subgroup was explored after the fact.

Bottom Line

SSR did not demonstrate superiority to supervised GRASP: the adjusted between-group difference in total FMA-UE at T2 was 2.27 points (95% CI −2.64 to 7.18; p = 0.357). Recruitment stopped at 61 randomized versus 130 planned; an encouraging distal-hand result in 38 selected participants was post-hoc and transient, not a reason to replace established training.

Design

Trial type: Two-arm, parallel, superiority randomized controlled trial; outcome assessors blinded, participants and therapists unblinded.
N: 61 randomized; 58 included in modified intention-to-treat (mITT) primary analysis.
- SSR: 31 randomized, 28 in mITT (2 could not achieve glove fit; 1 missed T2 after an unrelated medical event).
- Supervised GRASP: 30 randomized, 30 in mITT.
Randomization: 1:1 computer-generated variable blocks of 2, 4, and 6; balanced by center, no other stratification; sequence held by an independent coordinator.
Setting: Alexandra Hospital and Jurong Community Hospital inpatient rehabilitation units, Singapore.
Enrollment: July 2023–May 2025; stopped early when the funding period ended.
Follow-up: T2 after up to 3 weeks of training; T3 at 3 months postintervention (50/58 mITT patients with T3 outcome data).
Analysis: mITT with available T2 FMA-UE, linear mixed model adjusting for baseline score and center, including group, time and group×time; a sensitivity model included all 61 randomized with baseline observation carried forward. The planned sample was 106 evaluable participants (130 allowing for loss), assuming a 6-point between-group FMA-UE change difference and 80% power.
Primary outcome: FMA-UE total, 0–66 points (higher is better), at T2; T3 assessed durability.

Population

Inclusion Criteria

Age 21–90 years, medically stable, within 6 months of ischemic or hemorrhagic stroke, with unilateral upper-limb impairment.
FMA-UE 8–47/66, or 48–56/66 if wrist and hand performance was not at ceiling.
Able to consent, follow commands, and sit upright for at least 45 minutes.

Exclusion Criteria

Recurrent stroke; substantial cognitive or communication impairment; severe depression or active psychiatric disorder.
Unstable medical illness or expected survival under 1 year; affected-hand skin breakdown, finger amputation, adhesive allergy, or severe distal spasticity (Modified Ashworth Scale >3).

Baseline Characteristics (mITT)

Mean age 61.9 years in SSR versus 63.9 in control; 20/28 versus 21/30 male.
Ischemic stroke: 22/28 (78.6%) versus 25/30 (83.3%); mean days since onset 18.2 versus 21.3.
Mean total FMA-UE 29.3 versus 29.2; distal subscore 8.9 versus 9.7. This was a mixed-severity inpatient sample, not exclusively severe hand paresis.

Interventions

SSR: EsoGLOVE pneumatic, sensor-assisted glove with passive and active-assisted finger flexion/extension, grasp–release, and individualized object tasks; occupational therapist prescribed progression, assistant delivered 30–45 minutes/day, 5 days/week, target 10–15 sessions over up to 3 weeks, in addition to conventional occupational therapy. Mean delivered dose 11.9 sessions and approximately 96 device-logged repetitions per session across modes.
Control: Supervised, graded GRASP hand and arm exercises with a positioning-only neoprene wrist support, 30–45 minutes/day on the same schedule, in addition to approximately 1 hour of conventional occupational therapy. Mean delivered dose also 11.9 sessions. The neoprene support did not actively assist the hand.

Outcomes

Primary Outcome (total FMA-UE, 0–66, adjusted between-group difference SSR minus control):

All mITT participants, T2: adjusted means 44.7 versus 42.4; difference 2.27 points (95% CI −2.64 to 7.18; p = 0.357). The overall group×time interaction was not significant (p = 0.776).
All mITT participants, T3: difference 3.10 points (95% CI −2.95 to 9.15; p = 0.308).
Prespecified subgroups: None reporting a confirmatory differential treatment effect. The analysis restricted to baseline FMA-UE <35 and at least 10 sessions was post-hoc (n = 38; SSR 18, control 20). Its total FMA-UE difference at T2 was 4.08 (95% CI −2.75 to 10.90; p = 0.234), not significant.

Secondary Outcomes:

Distal FMA-UE, all mITT, T2: difference 0.86 (95% CI −1.59 to 3.30; p = 0.486).
Post-hoc severity-plus-exposure subgroup, distal FMA-UE, T2: difference 3.06 (95% CI 0.02 to 6.09; p = 0.049), but at T3 2.50 (95% CI −2.07 to 7.08; p = 0.273). The subgroup group×time interaction for distal FMA-UE was p = 0.804; the isolated T2 contrast is exploratory, not interaction evidence.
All mITT, T2: Stroke Upper Limb Capacity Scale difference 0.32 (95% CI −0.72 to 1.37; p = 0.537); Stroke Impact Scale-16 (SIS-16) difference 1.72 (95% CI −3.19 to 6.63; p = 0.485); DuruÖz Hand Index difference 1.38 (95% CI −9.37 to 12.13; p = 0.798). No demonstrated advantage in activity or patient-reported function.
Clinically important change, all mITT, T2: total FMA-UE improvement ≥12.4 points in 17/28 (60.7%) versus 16/30 (53.3%); adjusted OR 1.32 (95% CI 0.43 to 4.03; p = 0.623). In the selected subgroup, SIS-16 improvement ≥9.4 points occurred in 14/18 (77.8%) versus 11/20 (55.0%); adjusted OR 6.52 (95% CI 1.00 to 42.58; p = 0.050) with absolute risk difference 22.8 percentage points (95% CI −7.2 to 47.4)—borderline and explicitly not a confirmatory positive result.

Adverse Events / Safety:

No intervention-related adverse events or device complications were reported. Two randomized SSR participants could not use the glove because it did not fit their hands, so safety/feasibility is not the same as universal usability. Two recurrent strokes, one in each group, led to T3 exclusions; not attributed to the intervention. The study did not assess hemorrhage/ICH as a therapeutic safety endpoint.

Figures

Sensor-assisted glove during grasp, pinch and release training — *Figure 1.* Setup of sensor-assisted soft robotic glove. Participants perform (a) grasping, (b) pinching, and (c) release while following on-screen instructions.

CONSORT participant flow and retention diagram — *Figure 2.* CONSORT flow diagram of recruitment and retention; SSR, sensor-assisted soft robotics; T2, postintervention; T3, 3-month follow-up; FMA-UE, Fugl-Meyer Assessment of the Upper Extremity.

Between-group outcome estimates with confidence intervals — *Figure 3.* Between-group differences in selected outcomes. Baseline values are unadjusted means; later values are model-estimated means, with 95% confidence intervals. The asterisk marks the exploratory post-hoc distal-hand contrast (p = 0.049).

Criticisms

The trial stopped after 61/130 planned randomized because funding ended. The primary 95% CI (−2.64 to 7.18) allows both no benefit and a clinically relevant benefit; a nonsignificant p value does not prove equivalence.
Three of 61 randomized patients were excluded from the primary mITT analysis; two exclusions were specifically due to glove fit, a device-dependent postrandomization loss. Eight of 58 lacked T3 data; an MCAR test (p = 0.249) does not establish absence of attrition bias.
The attractive p = 0.049 comes from an unplanned intersection of baseline severity and completing ≥10 sessions. Completion is postrandomization, risking selection bias; multiple outcomes and time points were examined. No corresponding significant treatment-by-time interaction or durable effect was seen.
The comparator was actively supervised GRASP, not no therapy: both groups received similarly timed extra sessions plus usual occupational therapy. This isolates incremental device benefit better than an inactive comparison, but the study did not measure GRASP repetition counts or cost-effectiveness, and cannot establish dose equivalence or a staffing advantage.
Single-country, two-unit inpatient setting and narrow enrollment/fit criteria limit generalization to home therapy, other staffing models, and patients whose hands cannot accommodate the glove. One device co-developer has related patents; the device company provided equipment and funding.

Funding

Enterprise Singapore grant 2009TC3N and Roceso Technologies supported the study; Roceso supplied the glove. Authors state funders did not participate in design, data collection, analysis, interpretation, or manuscript writing. Coauthor Jeong Hoon Lim is a named inventor on related glove patents. Registered as NCT05995665.

The paper

Authors. Tay KWE, Tam PK, et al.
Title. Efficacy of a soft robotic exoskeleton for upper-limb rehabilitation in subacute stroke inpatient rehabilitation: a randomised controlled trial.
Journal. Journal of NeuroEngineering and Rehabilitation.
Year. 2026.
DOI. 10.1186/s12984-026-02133-0
PMCID. PMC13589368

Deep Dive — click to expand

What this is

This September 2026 inpatient rehabilitation trial asks whether a soft robotic hand glove buys more recovery than therapist-prescribed, assistant-supervised GRASP when both are added to usual occupational therapy for subacute stroke. It did not show superiority on the prespecified overall FMA-UE endpoint: 2.27 points in favor of the glove, CI crossing zero, after only 61 of 130 intended participants were randomized. A distal-hand difference appeared only in a small, selected analysis at treatment end; it was not sustained at 3 months.

1. Shadow Audit

The headline technological contrast is smaller than it sounds: both groups received an additional 30–45 minutes a day and averaged 11.9 sessions; the control was actively supervised, not untreated. The funding-limited sample matters as much as p = 0.357: the primary interval reaches 7.18 FMA-UE points of possible benefit, so this is not proof that no worthwhile effect exists. Conversely, the two postrandomization losses from poor glove fit are real-world costs of deploying a hand-sized device, even though the report found no device-related medical adverse events. A patent-holding coauthor and company device support belong in the same appraisal as the efficacy signal.

2. Inversion Engine

To turn the neutral overall finding into evidence for the planned advantage, a new trial would need to move the estimated between-arm effect beyond chance and toward the authors’ 6-point planning difference. Here the point estimate is 2.27, or 3.73 points short of that assumed effect; its CI spans −2.64 to 7.18. To turn the subgroup finding into a treatment-selection rule, severity must be specified before randomization, therapy exposure must not be used to select survivors after allocation, and a distal-hand advantage must survive multiplicity control and follow-up. The observed 3.06-point T2 subgroup difference sits below the reported 4.9-point individual distal FMA-UE MCID; those are different estimands and should not be equated.

3. Second-Order Catalyst

The first practical experiment is not a hospital-wide glove purchase. A rehab unit that already runs supervised GRASP could test whether people with low baseline FMA-UE and poor voluntary finger movement can use a fitting glove to complete otherwise impossible grasp–release repetitions, while tracking training time, adherence, meaningful hand use, and cost. That would be a monitored feasibility pathway, not a guideline change. A prespecified multicenter trial of that subgroup, with comparable practice dose and blinded function assessments at 3 months, should precede procurement claims.

4. Asymmetric Leverage

The possible payoff is concentrated where spontaneous hand practice is hardest: patients with more severe deficits may gain access to repetitions they cannot perform unaided. But the 38-person, selected subgroup yields a wide distal T2 CI of 0.02 to 6.09, and no clear activity advantage; across the full 58-person mITT cohort, the distal difference is only 0.86 (CI −1.59 to 3.30). The attractive hardware narrative may have a small incremental return if inexpensive, supervised GRASP produces similar outcomes at the same added therapy time. Staffing savings and cost-effectiveness remain unmeasured rather than established.

5. Paradigm Destroyer

The reflex to retire here is “more advanced device means better rehabilitation.” Tomorrow’s protocol wording: “Offer structured, supervised, task-specific upper-limb practice during inpatient subacute stroke rehabilitation; do not replace GRASP with a robotic glove on the basis of this trial. Consider the glove only as an individualized, fit-checked way to deliver practice when voluntary hand movement limits participation, with outcomes and cost recorded.”

MVP — Minimum Viable Proof

Show, in an adequately powered, assessor-blinded trial with baseline severity stratified before randomization, that the glove improves a prespecified distal-hand or real-world activity endpoint against an equally supervised, equal-time GRASP program at 3 months, without fit-related attrition undermining the result. A repeat isolated p = 0.049 among ≥10-session completers would not suffice.

Best Combination

The earlier GRASP literature made supervised, repetitive arm practice a credible active comparator; previous glove studies often had shorter follow-up or weaker conventional-therapy comparators. This trial adds a useful constraint to the field: test incremental benefit over a reproducible active program, not just a device-plus-therapy versus usual care contrast. Its signal of benefit in severe distal deficits should be combined with baseline-severity and dose hypotheses from prior robotic-rehabilitation trials, then prospectively tested rather than pooled as a proven indication.

Overvalue Warning

First, no significant overall difference is not noninferiority or equivalence: the trial was under-recruited and its primary CI admits a meaningful advantage. Second, the distal p = 0.049 is not a validated responder subgroup: it came after the fact from severity plus treatment completion, with no sustained significant T3 benefit or demonstrated improvement on activity outcomes.

Educational use notice. The Time Is Brain Review is dedicated to medical education. Figures from published articles are embedded for teaching purposes, with attribution to the original publication. We do not claim copyright over any embedded figure; all rights remain with the original publisher and authors under the article's published license. If you are a rights holder and would like a figure removed, please contact the editorial team. See our editorial notice for the full policy.