Background: Limited treatments for patients with amyotrophic lateral sclerosis (ALS) are partially attributed to the lack of meaningful and sensitive endpoints for ALS clinical trials. Novel digital endpoints enabled by wearable digital health technologies (DHTs) provide objective and continuous measures for tracking ALS progression with minimal patient burden, with great potential to improve and accelerate clinical development.
Objective: To evaluate the accuracy of DHT-derived measures of gait in individuals with ALS.
Methods: Nine adults with ALS participated (age: 59.66 ± 15.19 years, height: 1.8 ± 0.13 m, weight: 88.07 ± 18.86 kg). Participants performed 1-minute walking trials on an instrumented treadmill. Five GT9X (ActiGraph, L.L.C., Pensacola, FL) sensors were placed on wrists, ankles, and lumbar spine. Participants were also placed with retroreflective markers on anatomical landmarks of the lower legs. Algorithms specific to DHT-sensor locations were applied to derive step count, cadence, distance, and gait speed. A 3D motion capture of the lower legs estimated spatiotemporal gait parameters as the reference data. DHT-derived estimated measures were compared against the reference data to calculate mean absolute percent error (MAPE) and agreement metrics (Pearson’s r, Bland-Altman statistics). MAPE <15% is considered an acceptable accuracy for outcome measures.
Results: For the step count, the wrist (MAPE <9%) and the lumbar (MAPE <6%) showed high accuracies. For cadence, the wrist placement showed MAPE < 6% while the lumbar placement showed MAPE < 2%. For gait speed, the wrist showed poor accuracy (MAPE=51.16%) while the lumbar showed moderate/good accuracy with MAPE of 15.45%. Pearson’s statistics showed that lumbar-derived gait measures significantly correlated with the reference values (step count: r=0.95, p=0.012; cadence: r=0.99, p<0.05; speed: r=0.88, p=0.048), however, only step count showed a significant correlation (r=0.75, p<0.05) with the ground truth for the wrist.
Conclusions: Lumbar-derived measures of gait were found to be the most accurate. Cadence was found to be the most accurate measure, while gait speed was found to be the least accurate, irrespective of the sensor placement. Elevated errors were observed in wrist-derived measures due to the dampened wrist movement while using the railing support while walking. Future validation efforts on a larger sample are needed.