Using Machine Learning With Wearable Devices to Advance Research and Patient Care Using Spinal Cord Injury as a Model

Wearable devices (WDs) and machine learning (ML) are increasingly used to monitor physiological signals outside of traditional clinical environments, creating new opportunities for real-time, personalized insights. Yet the practical and ethical integration of these technologies into clinical research remains underdeveloped, especially for populations with complex, unstable physiology such as persons with spinal cord injury. In this perspective, we argue that the promise of WDs and ML can only be realized through deliberate alignment between device capabilities, physiological relevance, analytic rigor, and clinical context. Using spinal cord injury as a case example, we highlight the limitations of current measurement tools for capturing autonomic and sleep dysfunction, the challenges of interpreting high-frequency wearable data, and the need for customized ML approaches that account for individual variability and contextual noise. We present a conceptual framework to guide the responsible design, interpretation, and deployment of WD-ML systems in rehabilitation research and practice. This includes strategies for addressing missing data, signal artifacts, confounding, and bias, as well as for ensuring interpretability, data privacy, and clinical relevance. Ultimately, this paper calls for interdisciplinary collaboration, linguistic transparency, and critical engagement with emerging technologies to ensure that innovation in wearable analytics leads to equitable, actionable, and patient-centered care.