Screening fall risk with acceleration data

Purpose Prevention of falls is a crucial factor in preserving the independence and autonomy of older adults. Falls can be considered a public health problem due to their high incidence and the severity of their consequences1 and, therefore, falls pose an important matter of investigation. Among the approaches for fall prevention, wearable sensors are currently considered a viable option2, for example, to discriminate between fa Hers and non-fallers, with a previous study showing an Area Under the Roc Curve (AUC) of 0.843. However, there is still a gap in the literature on the analysis of such data, in particular for prediction of future falls, to be adopted for health care and prevention4. The objective of this study is to investigate patterns, obtained from an acceleration sensor, which could trace the risk of future falls in the elderly through variations in the Timed Up and Go (TUG) test Methods 73 non-faller community-dwelling elderly and participants of the University of the Third Age in Sao Carlos, SP, Brazil (representative sample with error=5% and power=90%), performed three variations of the TUG: (a) regular, (b) dual task motor, and (c) dual task cognitive, wearing an accelerometer in front of their center of mass. After collecting the data, each volunteer received a call at the end of 3 months to check for occurrences of falls. From the fusion of the accelerometer signal axes (X, Y and Z), obtained during the realization of the TUGs, 5 variables based on the signal frequency were computed to investigate the gait characteristics which would allow to differentiate the elderly who had a fall in the 3 months from monitoring compared to those who did not. The following frequency features were investigated: PSE (sum of the entropies of the frequencies), PSP (peaks of the frequency amplitude), PSPF (frequencies, in Hz, related to the peaks found) in three variants: PSPF1, PSPF2 and PSPF3 (which are the relative to the three first frequency amplitude peaks). A West was employed to compare the variables according to the groups: faller in 3 months and non-faller (Table /). Results & Discussion The average age of the sample was 70 years, 56% females. Among those, 7 elderly reported having suffered falls in the period of 3 months after the realization of the TUGs. According to the accelerometer-based frequency features, the second frequency peak, i.e., the PSPF2, showed a significant difference between the groups (t= -2.26, p = 0.027). We believe our result