Analytical and clinical validation of step counting method in people living with amyotrophic lateral sclerosis

Accelerometer-based digital measures offer a scalable and low-burden means of quantifying physical function, but existing processing algorithms may not quantify pathological gait correctly. In people living with amyotrophic lateral sclerosis (ALS), where gait patterns are slow, variable, and asymmetric, validated tools to quantify mobility are urgently needed. We proposed a step-counting algorithm designed for ankle-worn accelerometers that leverage wavelet-based decomposition to quantify heel strikes under heterogeneous gait patterns. We validated this method using five datasets comprising healthy individuals and those with ALS in controlled and semi-controlled activities, and we performed clinical validation in a free-living cohort of 305 people with ALS. We tested our method for accuracy in detecting steps and recognizing walking activity. Reference labels used for analytical validation were obtained from annotated studies or video-based ground truth. Step counting accuracy was assessed using Bland-Altman analysis while clinical validity was evaluated by comparing step counts to gross motor functioning on the ALS Functional Rating Scale-Revised (ALSFRS-R). Walking recognition was robust across walking conditions and body types; sensitivity ranged from 0.94 to 0.98, and specificity exceeded 0.95 across all evaluated datasets. The mean step counting bias was minimal (e.g., 0.44 steps), and the 95% limits of agreement were narrow (LoA = [-5.90, 5.40]) relative to reference standards, including video-annotated ground truth. Clinical validation indicated substantial differences between groups with various levels of gait impairment, e.g., participants who reported "walks with assist" on the ALSFRS-R accumulated a mean of 1283 (95% CI 1063, 1503) steps/day, while those reporting "normal" walking covered 3984 (95% CI 3537, 4432) steps/day. Our study covered analytical and clinical validation of a step-counting method developed for ankle-worn accelerometers and demonstrated its applicability to pathological gait. The method provides accurate quantification of walking activity in controlled and free-living environments, supporting its use as a digital endpoint in ALS research.