Smartphone-accessible AI pose estimation applied to walking videos accurately quantified gait differences across 68 older adults classified as non-frail, pre-frail, or frail. Step-time measurements correlated strongly with manual annotations (R²=0.93 at self-selected pace, R²=0.80 at fast pace), with tight Bland-Altman agreement limits (-0.082 to 0.052s). Spatiotemporal gait parameters—including step time, cadence, and walking speed—differed meaningfully across all three frailty categories, including the clinically ambiguous pre-frail group.
Frailty affects roughly 10–15% of community-dwelling older adults and is among the strongest predictors of hospitalization, disability, and mortality. Current gold-standard screening tools like the Fried phenotype or FRAIL scale require trained clinicians and in-person assessment, creating enormous bottlenecks in geriatric care. A validated video-based alternative could enable remote, continuous monitoring—particularly valuable for homebound or rural populations who rarely access specialty geriatric services. Gait analysis has long been considered a "sixth vital sign," and this approach could finally make it practical at scale.
Critical limitations warrant caution: this cross-sectional study of only 68 participants cannot establish whether gait changes predict frailty progression over time, nor can it assess whether screening with this tool improves clinical outcomes. The cohort is small and likely not demographically representative. Causal direction remains unknown. As an unreviewed preprint posted on medRxiv, these findings have not yet undergone peer review and should be considered preliminary. If validated in larger longitudinal cohorts, this technology could represent a genuinely paradigm-shifting shift in scalable frailty surveillance.