New sensor-based methods for clinical and in-home assessment of fall risk in older people

Abstract

Background. Falls are common in older people and are becoming an increasing burden on society as the population ages. An accurate assessment of fall risk could assist to identify individuals at increased risk and would enable appropriate intervention before a fall occurs. To date, the methods used to assess fall risk often lack accuracy and objectivity, require expensive equipment and specialized knowledge, and are limited to the clinical setting. Objective. The main objectives of this thesis were the development and evaluation of new sensor-based methods to accurately assess fall risk in both clinical and home settings. These new methods included the Microsoft Kinect, a low-cost consumer depth camera, and a pendant-style wearable sensor (Philips Research), comprising of an accelerometer and barometric air pressure sensor. Methods. A Kinect-based system (also referred to as the iStoppFalls assessment) for the assessment of balance, strength and reaction time through a program on the television was developed. The feasibility and validity of the Kinect-based system and wearable sensor to assess fall risk in community-living older people was investigated across three studies. In the first study, 94 older people were assessed on a Kinect-based five-times-sit-to-stand, choice stepping and choice reaching reaction time test in a clinical setting. Signal processing algorithms were developed to quantify performance on these tests by extracting temporal and spatial measurements from the Kinect skeleton data. The convergent validity of these measurements was established against traditional clinical tests and the ability to differentiate between people who are at risk for falls from people who are not at risk was investigated. In the second study, 119 older people were asked to perform typical daily activities, including postural transitions (an indicator for fall risk), while wearing the pendant device. A signal processing algorithm, based on wavelet transformations of the accelerometer and barometric air pressure sensor data, was developed to automatically detect and assess sit-to-stand movements during daily activities. In the third study, the long-term use of the Kinectbased system and wearable sensor was investigated in the homes of 62 older people over a four months study period. Participants were asked to independently perform at least one assessment session each month and to wear the pendant during the day. Interviews were conducted to investigate user experience and design guidelines for future home-based assessments were developed. Results. The Kinect-based system and the wearable sensor could be used to assess fall risk in older people in a clinical setting. Further, it was feasible to use both methods in an unsupervised home setting. Fallers performed significantly worse than non-fallers on the Kinect-based tests (p < 0.05). The proposed algorithms accurately detected these performance differences in the sensor data, supporting good discriminant validity of this new method. The wearable sensor based on the proposed algorithm accurately detected sit-to-stand movements through the monitoring of activities of daily life (sensitivity: 93.1 - 98.9%, false positive rate: