You are here

Classification of frailty and falls history using a combination of sensor-based mobility assessments


BR Greene, EP Doheny, RA Kenny, Brian Caulfield

Publication Type: 
Refereed Original Article
Frailty is an important geriatric syndrome strongly linked to falls risk as well as increased mortality and morbidity. Taken alone, falls are the most common cause of injury and hospitalization and one of the principal causes of death and disability in older adults worldwide. Reliable determination of older adults' frailty state in concert with their falls risk could lead to targeted intervention and improved quality of care. We report a mobile assessment platform employing inertial and pressure sensors to quantify the balance and mobility of older adults using three physical assessments (timed up and go (TUG), five times sit to stand (FTSS) and quiet standing balance). This study examines the utility of each individual assessment, and the novel combination of assessments, to screen for frailty and falls risk in older adults.Data were acquired from inertial and pressure sensors during TUG, FTSS and balance assessments using a touchscreen mobile device, from 124 community dwelling older adults (mean age 75.9 ± 6.6 years, 91 female). Participants were given a comprehensive geriatric assessment which included questions on falls and frailty. Methods based on support vector machines (SVM) were developed using sensor-derived features from each physical assessment to classify patients at risk of falls risk and frailty.In classifying falls history, combining sensor data from the TUG, Balance and FTSS tests to a single classifier model per gender yielded mean cross-validated classification accuracy of 87.58% (95% CI: 84.47-91.03%) for the male model and 78.11% (95% CI: 75.38-81.10%) for the female model. These results compared well or exceeded those for classifier models for each test taken individually. Similarly, when classifying frailty status, combining sensor data from the TUG, balance and FTSS tests to a single classifier model per gender, yielded mean cross-validated classification accuracy of 93.94% (95% CI: 91.16-96.51%) for the male model and 84.14% (95% CI: 82.11-86.33%) for the female model (mean 89.04%) which compared well or exceeded results for physical tests taken individually.Results suggest that the combination of these three tests, quantified using body-worn inertial sensors, could lead to improved methods for assessing frailty and falls risk
Digital Object Identifer (DOI): 
Publication Status: 
Publication Date: 
Physiological Measurement 2014
Research Group: 
National University of Ireland, Dublin (UCD)
Open access repository: