Limb Accelerations Measure on Admission to Inpatient Rehabilitation as Predictors of Daily Stepping Six-Months Following Spinal Cord Injury

Research Objectives To evaluate the additive predictive value of limb acceleration features measured during acute rehabilitation in determining walking ability at six-months post spinal cord injury (SCI). Unlike previous models that focused solely on capacity (can someone walk), this study examining actual daily mobility as measured by daily stepping. Design Longitudinal cohort pilot study. Setting SCI inpatient rehabilitation unit. Participants Nineteen participants with incomplete SCI, ages 18-70. Participants walking 150 ft with minimal or no assistance at admission were excluded. Interventions Participants wore Actigraph sensors on the non-dominant wrist and bilateral ankles. Limb accelerations were extracted during sleep for the initial week of inpatient rehabilitation. Daily step counts were extracted from daytime movement at 6-months post-SCI. A support vector machine (SVM) model was built using limb accelerations, admission clinical assessments (strength, sensation, AIS), and demographic data (age, sex, height, weight) collected in the first week as predictors. The model was trained using three-fold cross-validation and backward elimination, and the model accuracy and key features were reported. Main Outcome Measures Daily stepping was averaged across days and categorized into 4 categories: non-ambulatory (< 100 steps), household ambulator (100 - 2,200 steps), limited community ambulator (2,200 - 5,000 steps), and unlimited community ambulator (>5,000 steps). Results The SVM model demonstrated a good performance of 84.2% in predicting categories of daily stepping; the inclusion of limb accelerations results in a 15.8% increase in performance compared to the model with only clinical and demographic measures. Key subsets of predictors were identified, which included limb acceleration features related to the magnitude and smoothness of movements. Conclusions This pilot study demonstrated the additive predictive value of limb acceleration features to determine daily mobility at 6-months post-SCI. While further studies are needed to explore the potential of limb acceleration in predicting longer-term recovery and larger samples, these preliminary findings support the potential for low-cost wearable sensor data to improve our understanding of mobility prognosis following SCI. Author(s) Disclosures None.