Predicting Poor Functional Status In Adults With Knee Osteoarthritis Using Real-life Accelerometry Data

Abstract

PURPOSE: Investigate the risk factors associated with longitudinal functional decline in people with knee OA using free-living accelerometry data. METHODS: Longitudinal data from the Osteoarthritis Initiative (OAI) accelerometer study was extracted from 1229 participants tested at baseline and 2 years, including 35 features of functional capacity (gait speed, sit-to-stand time, etc), demographic characteristics, patient-reported outcome measures (WOMAC, etc.), and accelerometry-based physical activity. Poor functional status was defined as declining into or remaining in the worst function quintile compared to baseline status for two outcome measures (WOMAC score and gait speed). A Random Forest classifier was trained to predict individuals’ functional status. To explore the feature importance in model prediction, the model prediction was further interpreted using a Shapley additive explanation algorithm. RESULTS: For poor functional status in gait speed (Fig.1A), the top predictor is baseline low minutes in the performance moderate/vigorous activity range (CPM 2500+) followed by prolonged baseline sit-to-stand performance. Other significant contributing factors include high WOMAC score, low minutes in the performance light activity range (CPM 800-2499). The overall prediction accuracy is 82%. For poor WOMAC status prediction (Fig.1B), the top predictors are impaired baseline sit-to-stand and gait speed, as well as accelerated decline in gait speed. The overall prediction accuracy is 79%. CONCLUSIONS: Accelerometry-based measures of physical activity were identified as key indicators for the decline in gait speed over time. Whereas only functional capacity measures (gait speed and sit-to-stand performance) were identified as key indicators for the longitudinal decline in the WOMAC score. These findings may enable early detection and intervention for functional decline prevention in knee OA patients.