From Promise to Practice: AI-Enabled Telerehabilitation and the Evidence-to-Implementation Gap in Resource-Constrained Health Systems AI-Enabled Telerehabilitation – Evidence-to-Implementation Gap
DOI:
https://doi.org/10.5281/zenodo.22224408Keywords:
AI, Telerehabilitation, Rehabilitation, Digital Health, ImplemEntation Science, Health Equity, LMC, PakistanAbstract
Background: Access to rehabilitation is still restricted in many low- and middle-income countries (LMICs) and artificial intelligence (AI) is becoming more widely used in remote rehabilitation. The main public health question is not about the feasibility of implementing AI-based telerehabilitation, but about the evidence base for safe, equitable and sustainable implementation at scale.
Methods: This critical narrative review collates literature from PubMed/MEDLINE, Scopus and additional web literature searches, focusing on literature from 2024 to 2026, with the inclusion of foundational implementation literature. Eligible interventions included remote/home-based rehabilitation plus a demonstrated AI, machine-learning or deep-learning element. The synthesis included clinical evidence, added value of AI, validity in real-world, implementation readiness, equity, safety and scale-up, in the context of Pakistan.
Results:Primary studies indicate positive outcomes with specific AI-supported telerehabilitation applications, such as computer vision exercise support and AI-guided programmes. Most studies are small and heterogeneous, and few studies sufficiently account for the effects of AI while controlling for remote delivery, therapist contact, monitoring, feedback, adherence support, and adherence dose. Home validation evidence, health-economic value, governance and sustainable scale evidence are significantly less robust. No Pakistan specific AI based Telerehabilitation trial was found.
Conclusions: AI-based telerehabilitation has strong potential but is not yet backed by robust evidence for widespread implementation within health systems with limited resources. Implementing should be done via evidence thresholds for clinical benefit, independent AI-added value, real-world validity, safety, equity, affordability, workforce compatibility and sustainability.
