Clinical Scorecard: The Role of Artificial Intelligence in Rehabilitation: An Overview of Clinical Efficacy, Practical Application, Safety, and Equity Across Various Modalities and Environments

At a Glance

Key Highlights

• Technology-assisted training (robotics with or without VR) shows reproducible activity improvement post-stroke upper limb, but effects on impairment and independence are inconsistent under dose-matched and blinded conditions.
• AI-enabled tools face performance drops from development to deployment, especially brain-computer interfaces and computer-vision movement evaluation, limiting immediate clinical impact.
• Safety events are generally mild and rare (~0.3% in telerehabilitation sessions), but usability, adherence, equity, and cost data are underreported, especially in home and hybrid settings.

Guideline-Based Recommendations

Diagnosis

• Require external, multi-site validation with declared lab-to-clinic performance loss before clinical adoption.
• Ensure local calibration and impact evaluation of imaging-based decision support tools prior to pathway changes.

Management

• Adopt an adjunct-first posture integrating AI with standard rehabilitation care.
• Use technology-assisted training to increase task-specific practice dose, especially post-stroke upper limb rehabilitation.
• Gate adoption by minimum clinically important difference–anchored benefit under dose symmetry and blinded assessment.

Monitoring & Follow-up

• Implement continuous post-market monitoring for performance, safety, equity, and fairness.
• Standardize safety and usability data capture, particularly for home and hybrid delivery models.
• Establish operational monitoring frameworks to sustain accuracy and equity under data shifts.

Risks

• Performance degradation when AI models are transported across different sites and populations.
• Under-measurement of usability, adherence, equity, and cost, risking inequitable access and outcomes.
• Insufficient reporting of adverse events and lack of standardized safety protocols in unsupervised environments.

Patient & Prescribing Data

Primarily adults with neuro- and musculoskeletal impairments, including post-stroke patients; some pediatric populations in cerebral palsy and speech/language rehabilitation

AI and technology-assisted modalities can extend rehabilitation reach and personalize care but require rigorous validation, dose matching, and blinded assessment to confirm clinical benefits.

Clinical Best Practices

• Conduct pragmatic, multi-site, assessor-blinded, dose-matched trials to establish efficacy.
• Ensure subgroup fairness with mitigation strategies to address equity concerns.
• Prepare AI systems for regulation, change control, and cybersecurity prior to deployment.
• Develop a public, living evidence atlas to track evolving AI rehabilitation evidence.
• Adopt standardized safety practices including explicit hold criteria and structured adverse-event capture.

References

• World Health Organization Rehabilitation Report 2023
• Contemporary Evaluations in Clinical AI and Performance Decay
• Safety and Usability in Telerehabilitation