Preventive, personalized treatment of biopsychosocial (BPS) functional decline or stagnation in the rehabilitation process requires early prediction of those at risk. Ambulant monitoring of objective physical activity using wearable accelerometers could be a quantitative, noninvasive, and affordable method to gain insight into current and future biopsychosocial functioning.

In this retrospective study, Random Forest Regression was used to predict cross-sectional and longitudinal Functional Ambulation Category (FAC) and Six-item Cognitive Impairment Test (6CIT) after hip fracture using ambulatory accelerometry data. Accelerometry data and BPS functional assessments were available of 49 participants of the HIPCARE study, assessing prognostic determinants of outcome after hip fracture in the elderly.

Overall, cross-sectional FAC scores three months after hip fracture could be predicted with moderately low error, and categorized regression predictions showed high precision and recall. Cross-sectional 6CIT and both longitudinal regression models underperformed, but categorized regression predictions revealed mixed but more promising precision and recall.

It is expected that the predictive performance of models can be improved by increasing participant sample size with balanced samples over population-specific, prevalent ranges of BPS outcome scales and exploring additional machine learning models. In the future, accurate accelerometry-based predictions for individual patients needing rehabilitation could support personalized treatment and improve long-term biopsychosocial functioning.