Gait and sensory parameters based machine learning classification for detecting cognitive impairment in older adults.

Abstract

Detecting mild cognitive impairment in its early stages can increase access to treatment and allow care planning. However, it is still challenging as many older adults do not preemptively seek a neuropsychological assessment. To address this issue, methods for detecting cognitive impairment without cognitive testing should be explored. The present study designed machine learning algorithms based solely on gait and sensory parameters and assessed their ability to discern older individuals with suspected cognitive impairment from those with healthy cognition. Community-dwelling men older than sixty-five (n = 94) underwent cognitive, sensory, and gait function assessments. Based on the cognitive evaluation, they were divided into the non-cognitively impaired group (n = 65) and the suspected impaired cognition group (n = 29). Machine learning models were trained and compared in terms of diagnostic accuracy to discern the group suspected of having cognitive impairment from the non-cognitively impaired group. Among the machine learning algorithms, a support vector machine and an automated machine learning model showed the highest ability in classifying older individuals with suspected cognitive impairment from those with healthy cognition with an accuracy of 82.8 %. The gait and hearing parameters of older individuals with suspected cognitive impairment differed significantly from those of cognitively healthy older adults. By utilizing these parameters, the present research presented the possibility of developing a fast and simple screening method for detecting early cognitive impairment without needing neuropsychological testing.