Frailty Assessment Based on the Quality of Daily Walking

Abstract

Frailty is an increasingly recognized geriatric syndrome resulting in age-related decline in reserve across multiple physiologic systems. An impaired physical function is a prime indicator of frailty. In this study, we aim to implement a body-worn sensor to characterize the quantity and quality of everyday walking, and establish associations between gait impairment and frailty. Daily physical activity was acquired for 48 hours from 125 older adults (≥65 years; 44 non-frail, 60 pre-frail, and 21 frail based on the Fried gold standard) using a tri-axial accelerometer motion-sensor. Continuous purposeful walks (≥60s) without pauses were identified from time-domain acceleration data. Power spectral density (PSD) analysis was performed to define higher gait variability, which was identified by a shorter and wider PSD peak. Association between frailty and gait parameters was assessed using multivariable nominal logistic models with frailty as the dependent variable, and demographic parameters along with the gait parameters as the independent variables. Stride times, PSD gait variability, and total and maximum continuous purposeful walking duration were significantly different between non-frail and pre-frail/frail groups (p<0.05). Using a step-wise model with the above qualitative and quantitative gait parameters as predictors, the pre-frail/frail group (vs. non-frail) was identified with 71.4% sensitivity and 75.4% specificity. Everyday walking characteristics were found to be accurate determinants of frailty. Along with quantitative measures of physical activity, qualitative measures are critical elements representing the stages of frailty. In-home gait analysis is advantageous over clinical gait analysis as it enables cost- and space-effective continuous monitoring.

A Microservices e-Health System for Ecological Frailty Assessment Using Wearables

“…Tegou et al [14] applied several ML algorithms, such as k-Nearest Neighbour (kNN), Random Forests (RF) and Naïve Bayes (NB) to classify the three frailty status based on Fried criteria [4] (non-frail, pre-frail, frail), by measuring the number of transitions between rooms inside the older adult home. Schwenk et al [15] and Kumar et al [16] applied Multinomial Logistic Regression (MLR) model to discriminate between the three frailty status, by measuring the gait, balance, and physical activity (PA). Toosizadeh et al [3] used Ordinal Logistic Regression (OLR) for a frailty detection in 20 s elbow flexion.…”

“…In addition, most of these logistics models did not focus directly on detecting frailty status, instead they looked for the correlation between risk factors (such as falls, balance, physical performance and gait) and the frailty status. Also, the ecological approaches reviewed [16,18] use expensive wearables and questionnaire data such as demographic and clinical data, which do not allow the automation of the data collection. [15] To discriminate between frailty status with gait, balance or during a physical activity.…”

“…[15] To discriminate between frailty status with gait, balance or during a physical activity. No LEGSys 1 ($10,000) BalanSens 1 ($4450) None Three 2 MLR: AUC: 85.7% [16] To implement a wearable to characterize the quantity and quality of everyday walking, and to establish associations between gait impairment and frailty.…”

“…A feature is redundant when another relevant feature exists with a similar power of prediction [76]. The feature selection techniques are the filter, wrapper and embedded methods [16,77]. Filter methods identify the relevant features based on statistics, without considering a machine learning classifier to build a model.…”

“…They can provide efficient and precise psychophysiological and location values [ 13 ]. Some research works have started to measure frailty indicators [ 3 , 14 , 15 , 16 , 17 , 18 ], being the most used devices electronic stickers, bands, bracelets, smartphones and smartwatches [ 15 , 19 , 20 , 21 , 22 , 23 ]. However, these approaches only measure the physical dimension of the frailty during a BADL with only one sensor, either the accelerometer or the gyroscope.…”