Dementia Specific Gait Profile: A Computational Approach Using Signal Detection Theory and Introduction of an Index to Assess Response to Cognitive Perturbations

Abstract

Elderly diagnosed with dementia are three times more likely to fall and over three times more likely to have severe injury compared to cognitively unimpaired elderly. Consequently, there is a need to identify biomarkers that can facilitate early detection, diagnosis, and progression of dementia. One of the characteristics of dementia is the inability to allocate attentional resources to concurrent tasks. Consequently, recent studies have used walking gait in conjunction with another cognitive or motor task to identify biomarkers related to the disease. However, in every study all temporalspatial gait descriptors are being evaluated and, typically, the nonspecific velocity, double limb support, and stride variability are reported as significant. The purpose, therefore, of this investigation was to use a computational approach to first establish a dementiaspecific gait profile irrespective of walking condition (talking, without talking) using the minimum number of temporalspatial gait descriptors, second to investigate the effect of condition, and third to investigate the effect of an everyday realistic cognitive perturbation, resulting in potential falls, by constructing an index of responsiveness. Six normal elderly and seven diagnosed with dementia walked on an instrumented walkway: (i) without talking, (ii) conversing with an investigator, and (iii) conversing with an investigator, but including as part of the conversation a cognitive perturbation in the form of an unexpected direct question. To accomplish the first two goals we implemented signal detection theory combined with receiver operator characteristic curves. Based on these results we constructed the index of responsiveness that we compared between the two cohorts. Only six of thirteen gait variables were needed to distinguish individuals with dementia from normally aging, irrespective of whether gait was used as a standalone task, i.e., without talking, or under a dualtask paradigm, i.e., combined with a conversation. Double limb support was the most sensitive variable to describe adaptation to walking condition. The index of responsiveness was significantly larger for individuals with dementia. The six discriminating temporalspatial gait descriptors provide new focus for health care professionals involved in diagnosis and treatment of elderly with dementia. The index of responsiveness can be used to describe a bandwidth of safety, identifying individuals with dementia at risk of falling.