Validation of digital health measures for upper extremity function after cervical spinal cord injury

Introduction: Restoration of upper extremity (UE) function remains a top rehabilitation priority for individuals with cervical spinal cord injury (CSCI). Conventional clinical outcome assessments (COAs) are the primary tools for evaluating recovery and intervention effectiveness, but provide only brief, episodic snapshots of function, yielding a fragmented view of how recovery translates into real-world UE activity. Digital health technologies (DHTs) such as wearable sensors enable passive, continuous monitoring of naturalistic activity, offering a way to bridge the gap between clinician-observed capacity and daily performance. However, despite their promise, sensor-derived metrics remain insufficiently validated against established benchmarks in CSCI and are rarely linked to patient-centered outcomes, which has limited their clinical adoption. Methods: This dissertation used an integrated, multi-modal framework to improve the interpretability and patient-centeredness of outcome assessment in CSCI. In Aims 1–3A, individuals with CSCI wore a single wrist-worn inertial sensor on their most-used UE continuously for approximately four weeks in inpatient and community settings. Accelerometry-derived metrics, aggregated across the monitoring period, were compared with three COA types: performance-based (Aim 1), clinician-reported (Aim 2), and patient-reported (Aim 3A) outcomes. Two classes of accelerometry-derived metrics—volume-based (capturing the total volume or quanta of activity) and pattern-based (capturing the structure of activity)—were evaluated. To address the broader challenge of establishing the patient-centricity of digital measures and improving the patient-reported outcomes (PROs) used to assess them, Aim 3B leveraged Reddit as a complementary source of patient-centered data. Natural language processing (NLP) methods were applied to more than 30,000 submissions to characterize the thematic structure of user-generated content relevant to CSCI. Results: Volume-based metrics, including activity counts, emerged as strong surrogate metrics of UE function, demonstrating moderate-to-strong correlations with UE capacity and performance across COA types but negligible associations with pain, fatigue, and quality of life. Pattern-based metrics, such as autocorrelation, showed moderate relationships with fine motor ability, suggesting that autocorrelation is a potential surrogate metric of hand function. In predictive analyses, most sensor metrics did not improve forecasts of later function after accounting for baseline ability; however, activity counts contributed unique explanatory value for self-care outcomes. NLP analyses identified themes convergent with recognized patient-important outcomes (PIOs), with UE function emerging as a central, cross-cutting theme, particularly in discussions of assistive technology, mobility, regenerative treatments, and reconstructive surgery. Conclusion: These findings support the use of sensor-derived metrics as surrogate indicators of UE function and demonstrate that NLP-based methods provide a scalable means of collecting patient-centered data. Together, these approaches establish a pathway for grounding digital health measures in patient experience and enhancing the interpretability and relevance of DHT-derived outcomes in CSCI rehabilitation research.