Physical performance clustering as a marker of frailty, fall risk, and cognitive health in older adults

The heterogeneity of physical decline in older adults necessitates multidimensional assessment strategies. This study investigates whether unsupervised classification based on physical performance measures (PPMs) can identify clinically relevant profiles linked to frailty, fall risk, and cognitive health. A total of 368 patients were evaluated on 23 physical performance measures (PPMs), including gait assessed with the Optogait system, isokinetic force and power measured with the Biodex dynamometer, body composition assessed with the QuadScan 4000, and grip strength tested with a hand dynamometer. These 23 PPMs were used to separate the population into clusters using a unsupervised K-means approach. Subsequently, differences in demographic and clinical characteristics (sex, Fried’s frailty phenotype, falls, memory complaints, and living situation) between clusters were analyzed using ANOVA and chi-squared tests. Three clusters emerged: High (HPP), Moderate (MPP), and Low Physical Performance (LPP). These differed significantly across most PPMs, especially muscular power and joint moments. Clinically, LPP included more frail individuals, fallers, and those living alone, while HPP predominantly comprised younger, non-frail men. Memory complaints did not significantly differ across groups. Vmax and Vopt, velocity-based measures, did not contribute meaningfully to clustering. Clustering based on multidimensional PPMs effectively stratified older adults by physical and clinical characteristics, supporting its utility for identifying at-risk subgroups. Muscular power and joint moments were the most discriminative features, reinforcing the emerging concept of “ powerpenia ” as a critical marker of functional decline. This method may inform targeted prevention strategies in geriatric care. No. 2015-A01188-41 registered on October 6, 2015, and approved on December 17, 2015. (Clinical trial registration number is NCT02690402).