Gait variability-based classification of the stages of the cognitive decline using partial least squares-discriminant analysis.

Abstract

The purpose of the present study was to examine the differences in gait variability in terms of spatiotemporal, sub-gait cycle, ground reaction force, and the joint profiles of kinematics and kinetics between older individuals with and without risk of potential cognitive impairment, and to derive the crucial features to discriminating the older adults with future risk cognitive decline by using partial least squares-discriminant analysis. A total of 90 community-dwelling older adults aged over 65 years underwent cognitive function assessment and were divided into three groups depending on cognitive assessment score. The participants' level-walking was analyzed by using three-dimensional instrumented gait analysis. The coefficient of variation was extracted and then comparatively analyzed depending on the stages of the cognitive decline. To identify the most important contributor when differentiating the older adults with a risk of future cognitive decline, partial least squares-discriminant analysis was applied, and the discriminative power of the coefficients confirmed as features of great importance were investigated via the receiver operating characteristic area under the curve. The differences in gait variability were found mainly between the suspected dementia groups and other groups, especially in joint dynamics variables. Through the partial least squares-discriminant analysis, the discriminative features were found as follows: the mid-stance, the moments, and the power in the hip, knee, and ankle joints. In addition, the discrimination model was found to differentiate well between the three groups. The classification accuracy of intact cognition, diminished cognition, and suspected dementia was 0.857, 0.710, and 0.857, respectively. These findings mean that gait variability changes according to continuous cognitive decline, especially in sub-gait cycles and joint biomechanics, and suggest that measures of variation can be used as predictors to identify older individuals with a risk of potential cognitive impairment.