A clustering-based trajectory analytics of functional loss and recovery among older adults

Objectives: Functional loss and recovery in older adults are heterogeneous, with important implications for independence, care needs, and survival. While Activities of Daily Living (ADLs) are routinely assessed, most existing approaches reduce them to summary scores, thereby losing information about the order, timing, duration, and recurrence of functional change. We introduce a novel trajectory analytics framework designed to identify clinically interpretable trajectory phenotypes and evaluate their association with mortality. Materials and methods: We analyzed 1.3 million ADL assessments from 265,530 residents in U.S. Veterans Affairs nursing homes. A hybrid trajectory clustering framework was developed, combining spell-based sequence construction, optimal-matching-based dissimilarity, and scalable quality-guided clustering. Sequence comparison was designed to preserve temporal structure while remaining computationally feasible for large-scale longitudinal data. Candidate clustering solutions were evaluated using multiple quality metrics, and mortality differences across clusters were examined using Kaplan-Meier estimation and Cox proportional hazards models adjusted for age and sex. Results: Thirteen distinct ADL trajectory clusters were identified from about 110,000 unique trajectories, differing in dominant disability states, duration, recurrence, and mortality risk. Short, severe trajectories showed the highest early mortality, whereas longer, milder trajectories involving limited impairments, such as walking or bathing, were more stable and had lower mortality. The highest-risk cluster showed approximately 48% cumulative mortality within the first year and remained associated with the greatest hazard of death after adjustment. Sex-based sensitivity analyses showed broadly similar mortality ordering across males and females, although female-specific estimates were less precise because of the smaller sample size. Discussion and conclusion: The proposed framework reveals substantial heterogeneity in functional trajectories and their prognostic implications, providing an interpretable and computationally efficient tool for large-scale ADL trajectory clustering. These findings support trajectory-based phenotyping as a useful approach for personalized care planning, targeted resource allocation, and policymaking in long-term care settings.