Physical Medicine

AI for gait and balance in neurorehab is promising but unvalidated

Scoping review of 18 studies (2009–2025) found AI mainly used for prognostic tasks in stroke, PD, and MS. Common methods included Random Forest, SVM, logistic regression, and XGBoost, with most work originating from Asia. Crucially, none of the studies performed prospective or external validation on independent datasets. Clinical translation is therefore limited and current models should not guide care without independent validation. 1

Mobility Plan QI in a Swiss geriatric clinic showed feasible uptake with mixed staff engagement

Implementation across 69 plans showed high fidelity (median 75%) and adoption of at least one activity (median 88%). Therapists reported acceptable, appropriate, and feasible implementation; nurses reported lower acceptability and feasibility. Median daily step counts were descriptively higher during implementation, but the design prevents causal attribution. Work to improve interdisciplinary engagement before expecting consistent patient mobility gains. 2

AI/ML RCTs in geriatrics show clinical benefits but raise generalizability concerns

Editorial summary identified four RCTs of AI/ML interventions in adults ≥65 showing benefits in postoperative recovery, psychosocial outcomes, and preventive uptake. Interventions included personalized smartphone programs, algorithm-guided paired rehab, robotics, and chatbots. Authors caution that trial populations may be healthier or more motivated, limiting generalizability and equity. Long-term outcomes, digital literacy, and safety monitoring are needed before broad implementation. 3