Estimating fall risk with inertial sensors using gait stability measures that do not require step detection

Abstract

Falls have major consequences both at societal (health-care and economy) and individual (physical and psychological) levels. Questionnaires to assess fall risk are commonly used in the clinic, but their predictive value is limited. Objective methods, suitable for clinical application, are hence needed to obtain a quantitative assessment of individual fall risk. Falls in older adults often occur during walking and trunk position is known to play a critical role in balance control. Therefore, analysis of trunk kinematics during gait could present a viable approach to the development of such methods. In this study, nonlinear measures such as harmonic ratio (HR), index of harmonicity (IH), multiscale entropy (MSE) and recurrence quantification analysis (RQA) of trunk accelerations were calculated. These measures are not dependent on step detection, a potentially critical source of error. The aim of the present study was to investigate the association between the aforementioned measures and fall history in a large sample of subjects (42 fallers and 89 non-fallers) aged 50 or older. Univariate associations with fall history were found for MSE and RQA parameters in the AP direction; the best classification results were obtained for MSE with scale factor τ = 2 and for maximum length of diagonals in RQA (72.5% and 71% correct classifications, respectively). MSE and RQA were found to be positively associated with fall history and could hence represent useful tools in the identification of subjects for fall prevention programs.

Introduction

Falls in the elderly have adverse physical and psychological consequences for the individuals, as well as substantial consequences for health-care and economy [1]. In older adults, falls often occur during walking [2]. The analysis of gait stability may allow identification of subjects at risk. However, the definition of gait stability is still not entirely clear, and many direct and indirect measures aiming to quantify this feature have been suggested in the literature [3]. Measures of trunk accelerations are crucial in the assessment of gait stability [4], [5], [6], as the trunk segment is known to play a critical role in regulating gait-related oscillations in all directions [7].

Many gait stability measures proposed in the literature are based on the identification of gait cycles [2], [8], [9], [10], [11]. Several methods for step detection have been presented in the literature [12], [13], [14], based on different techniques and sensor positioning. Errors in step detection can, however, critically affect stability outcomes, making step detection a possible intrinsic source of error. Examples are present in the literature of inability in the detection of gait events due to irregular acceleration patterns [15] and incorrect identification of acceleration peaks [16]. Gait characteristics or anomalies typical of certain pathologies (e.g. shuffling, crouched, toe gait) can result in atypical acceleration signals, determining unreliable step-detection. Assuming that such deviations are more common among people with a high fall risk, such errors may cause bias when calculating gait stability measures. Other temporal parameter detection systems, such as foot switches or pressure sensors attached to the sole, involve several problems [14] (e.g. difficulties in sensor attachment when assessing subjects with abnormal gait). To overcome this possible source of error, nonlinear analysis techniques may offer a powerful tool. In particular, some of these stability related measures do not depend on step detection. In this study the Harmonic ratio (HR) [17], [18], the Index of harmonicity (IH) [19], Multiscale entropy (MSE) [20], and Recurrence quantification analysis (RQA) [21] of trunk accelerations during gait were analysed [17], [18], [19], [20], [21], [22]. The relationship between these measures and fall risk has not been analysed and reported before.

HR, derived from trunk acceleration signals and based on amplitudes in frequency spectra, provides information on how smoothly subjects control their trunk during walking, giving an indication of whole body balance and coordination [17], [18].

Similarly to HR, IH assesses the contribution of the oscillating components to the observed coordination patterns by means of spectral analysis [19], quantifying the contribution of the stride frequency to the signal power relative to higher harmonics.

MSE quantifies the complexity of a time series at multiple spatio-temporal scales [20], since biological systems are likely to present structures at different scales.

RQA is a nonlinear technique that has been applied recently to biological time series, including gait data [22]. Based on local recurrence of data points in the reconstructed phase space, it provides a characterization of a variety of features of the time series, such as the quantification of deterministic structure and non-stationariness [21].

The aim of the present study was to investigate the association between fall history and the aforementioned measures (HR, IH, MSE and RQA) during treadmill walking in a large sample of older subjects.