Predictive factors of functional recovery in older inpatients: AIRCOT Study.

PURPOSE: Geriatric functional recovery units need improved prognostic tools to anticipate rehabilitation outcomes. Although machine learning offers promising possibilities, thorough exploratory analyses are essential before clinical use. This study examines determinants of functional recovery and develops predictive models for the Modified Barthel Index (MBI) at discharge, aiming to identify key variables, determine the most relevant predictors, and generate initial models of MBI at discharge. MATERIALS AND METHODS: A retrospective longitudinal study was conducted with 957 patients admitted in a Functional Recovery Unit. Sociodemographic, clinical, functional and psychosocial variables were analyzed. Functional assessments included MBI, muscle strength, and ambulation capacity. Descriptive analyses, association tests, and linear models - including multiple regression and Least Absolute Shrinkage and Selection Operator (LASSO) regularization, were applied to predict MBI at discharge. RESULTS: Participants had a mean age of 82.4 years, with a predominance of women and high rates of acquired brain injury and hip fracture. Functional recovery was primarily associated with pre-admission and admission MBI scores, trunk muscle tone, cognitive impairment, and time since injury. Complex models outperformed simpler ones, with the LASSO model achieving R2 = 0.849. The strongest predictors were previous MBI, MBI at admission, trunk muscle tone, motor aphasia, and muscle strength. CONCLUSIONS: Functional status before and at admission is the most reliable predictor of autonomy at discharge. While some comorbidities and biomarkers showed inconsistent associations, multivariable models improved predictive accuracy and may support individualized clinical decision-making. Further research should incorporate advanced machine learning techniques and evaluate their clinical applicability.