Accelerometry-based assessment and detection of early signs of balance deficits

Falls are the cause for more than half of the injury-related hospitalizations among older people. Accurate assessment of individuals’ fall risk could enable targeted interventions to reduce the risk. This paper presents a novel method for using wearable accelerometers to detect early signs of deficits in balance from gait. Gait acceleration data were analyzed from 35 healthy female participants (73.86±5.40 years). The data were collected with waist-mounted accelerometer and the participants performed three supervised balance tests: Berg Balance Scale (BBS), Timed-Up-and-Go (TUG) and 4 m walk. The follow-up tests with the same protocol were performed after one year. Altogether 43 features were extracted from the accelerometer signals. Sequential forward floating selection and ten-fold cross-validation were applied to determine models for 1) estimating the outcomes of BBS, TUG and 4 m walk tests and 2) predicting decline in balance during one-year follow-up indicated as decline in BBS total score and one leg stance. Normalized root-mean-square errors (RMSE) of the assessment scale result estimates were 0.28 for BBS score, 0.18 for TUG time, and 0.22 for 4 m walk test. Area under curve (AUC) was 0.78 for predicting decline in BBS total score and 0.82 for one leg stance, respectively. The results suggest that the gait features can be used to estimate the result of a clinical balance assessment scale and predict decline in balance. A simple walk test with wearable monitoring could be applicable as an initial screening tool to identify people with early signs of balance deficits.