Machine Learning–Based Biomechanical and Clinical Assessment of the Five Times Chair Rise Test in Older Adults: Sensor-Based Functional Aging Study

Abstract Background Maintaining functional ability is essential for healthy aging. The Five Times Chair Rise Test (5CRT) is widely used to assess lower-extremity function, yet conventional stopwatch timing provides limited insight into movement quality and compensatory strategies. Sensor-based biomechanics combined with machine learning (ML) enables detailed analysis of movement dynamics and supports digital health approaches for early detection and monitoring of functional decline. However, the clinical relevance of ML-based 5CRT movement patterns and their relationship to established functional outcomes and aging trajectories remains insufficiently understood. Objective This study aimed (1) to classify 5CRT movement patterns using an established dynamic time warping (DTW)-based clustering workflow; (2) to examine associations between ML-based movement clusters and multidimensional indicators of strength, mobility, and functional performance across 24 months; and (3) to evaluate whether ML-based movement classification provides clinically meaningful differentiation beyond conventional stopwatch-based 5CRT timing within a digital health assessment context. Methods In this longitudinal observational study, community-dwelling adults aged ≥70 years completed standardized functional assessments at baseline, 6 months, and 24 months, including Handgrip Strength, Stair Climb Power Test, Timed “Up & Go” Test, Short Physical Performance Battery (including 5CRT), 4-meter Gait Speed, Balance Testing, 6-minute Walking Test, and Countermovement Jump. The assessments were recorded using a force plate and depth camera to capture biomechanical time-series data. DTW and k-means clustering were applied to ground reaction force signals to identify movement patterns. Results Among 172 participants included in the final analysis (mean age 76.9, SD 3.40 years; 96/172, 56% female), 2 distinct movement clusters were identified. Cluster 1 demonstrated faster chair rise performance and lower stabilization forces, whereas cluster 2 exhibited slower execution and increased stabilization demands. Across baseline, 6-month, and 24-month follow-up assessments, cluster 1 consistently showed higher muscle strength, greater leg power, faster mobility, higher gait speed, and longer walking distance than cluster 2 (all P <.01). At 24 months, large between-cluster differences were observed for the Timed “Up & Go” Test, 5CRT duration, gait speed, 6-minute walk distance, and Short Physical Performance Battery score. Conventional stopwatch scoring categorized most participants as high functioning, whereas ML-based clustering revealed substantial functional heterogeneity and clinically meaningful information not reflected by timing alone. Conclusions ML-based analysis of sensor-based chair rise performance identifies functional performance profiles beyond conventional timing metrics. Integrating biomechanical movement analytics into digital health assessment frameworks may improve early detection of mobility decline and support longitudinal monitoring and personalized intervention planning in older adults.