Robot-assisted vs. conventional gait training for balance and gait speed after stroke: An updated systematic review...

Balance and gait impairments are common disabling consequences of stroke and central targets of rehabilitation. Robot-assisted gait training (RAGT) can deliver repetitive, task-specific, and quantifiable walking practice, yet its comparative effectiveness over conventional gait-oriented training remains uncertain.

We searched 8 English- and Chinese-language databases from inception through June 2, 2026 for randomized controlled trials comparing lower-limb RAGT with conventional therapeutic training after stroke. Primary outcomes were balance, assessed with the Berg Balance Scale (BBS) and Timed Up and Go (TUG) test, and gait speed. Reports from the same or potentially overlapping trials were linked to minimize participant double-counting. Change-from-baseline mean differences (MDs) were synthesized using restricted maximum likelihood random-effects models with Hartung-Knapp-adjusted 95% confidence intervals (95%CIs). Heterogeneity, 95% prediction intervals, sensitivity analyses, small-study effects, and Grading of Recommendations, Assessment, Development and Evaluation (GRADE) certainty of evidence were evaluated.

We included 258 eligible reports representing 240 trial-level units. Post-intervention pooled estimates favored RAGT for BBS (MD = 6.03 points; 95%CI: 5.14-6.93), direction-reversed TUG (MD = 3.04 s; 95%CI: 1.02-5.06), and gait speed (MD = 0.080 m/s; 95%CI: 0.061-0.100). However, between-study heterogeneity was very high across primary outcomes (I² = 89.8%-98.1%), all 95% prediction intervals included the null, and GRADE certainty was low. Sensitivity analyses attenuated key estimates: excluding add-on/extra-RAGT designs reduced the BBS and gait-speed effects, and restriction to reports with PEDro scores ≥ 6 attenuated all 3 primary-outcome estimates, with the TUG 95%CI including the null. Durability evidence was sparse, and most follow-up primary-outcome estimates included the null.

Low-certainty evidence suggests that RAGT may provide post-intervention improvements in balance and gait speed after stroke, but their magnitude, consistency, clinical importance, and durability remain uncertain. High heterogeneity, null-inclusive prediction intervals, attenuation in sensitivity analyses, and limited follow-up data argue against interpreting RAGT as clinically superior to conventional therapy. RAGT may be considered an individualized adjunct to gait-oriented rehabilitation when it increases task-specific practice without replacing therapist-directed care.