Robust Unsupervised Analysis of Longitudinal Functional Trajectories in Older Inpatients Reveals Stable Recovery Profiles: The AIRCOT Study

Functional decline in older adults represents a major challenge in geriatric rehabilitation, and machine learning (ML) clustering techniques may help identify functional recovery profiles and support personalized rehabilitation strategies. This study characterizes functional recovery profiles in older adults admitted for rehabilitation using unsupervised clustering techniques. A retrospective longitudinal study was conducted including 957 older adults admitted to a geriatric rehabilitation unit between 2019 and 2025. Clinical and functional variables, including the Modified Barthel Index (MBI), Daniels and Worthingham’s Muscle Testing, and Functional Ambulation Category, were collected from medical records. Considering the baseline functional status and the temporal changes in MBI scores, a clustering of the functional trajectories has been performed using a k-means algorithm based on the silhouette score. The robustness of the resulting segmentation has been evaluated by comparing alternative partitions obtained from bootstrap-sampling, hierarchical clustering, and the centroids of Gaussian Mixture Models. Four distinct functional recovery profiles were identified, showing different trajectories of independence, ambulation, and muscle strength during rehabilitation. Two clusters demonstrated favorable recovery and higher functional resilience, whereas the remaining profiles were characterized by chronic impairment or severe damage with limited recovery. These findings support the usefulness of unsupervised ML clustering techniques for identifying clinically meaningful recovery profiles and may facilitate patient stratification and individualized intervention planning in geriatric rehabilitation.