The distribution of real-world gait speed and its relation to gait bout length in people with and without gait impairment

Real-world walking has been suggested to consist of two distinct constructs: slower walking ensembled in short walking bouts, and faster walking during longer walking bouts. Here, we evaluated the presence of a bimodal gait speed distribution reflecting these constructs in people with and without gait impairment. Remote walking data were analyzed from four cohorts: healthy adults (n = 16), chronic stroke (n = 20), end-stage knee osteoarthritis (n = 21), and incomplete spinal cord injury (n = 16). Participants wore a sensor on each foot and one on the lower back for three days. Gaussian mixture modeling was used to characterize each individual's gait speed distribution from all recorded strides. Bimodality was quantified using Ashman's D, with D > 2 indicating well-separated components. For participants with D > 2, bout lengths were compared between the slower and faster component. Well-separated components were found in 12 out of 73 participants (16%), with no evident differences between people with and without gait impairment. Overall, mean gait speed increased while variability decreased with longer bout length. Among participants with bimodal distributions with D > 2, bout lengths did not differ significantly between components (slow speed component: 11 [IQR: 7.5, 31.8] strides vs. fast speed component: 34.5 [IQR: 20.8, 122] strides; p = 0.056). Real-world gait speed distributions were typically not bimodal, and gait bout length could not explain bimodality. Together, our findings do not support the idea that distinct components of the gait speed distribution reflect short versus long walking bouts.