Prediction of Cognitive Impairment Using Sleep Lifelog Data and LSTM Model

Rapid elderly population growth has increased the number of patients with cognitive impairment (CI). Early detection and ongoing medical treatment can slow CI progression and significantly reduce the cost of managing patients. However, distinguishing CI from natural cognitive decline associated with aging is challenging. Previous studies conducted to identify patients with CI using lifelog data did not consider changes in lifelog data over time because each data point was learned individually. This study introduces a model that predicts patients with CI based on sleep lifelog data and analyzes significant sleep factors that influence cognitive decline. This study followed three steps: (1) collecting sleep lifelog data from elderly Korean people and reconstructing sleep lifelog data as time-series data; (2) building a model to classify CI using a time series of sleep lifelog data and a long short-term memory model; and (3) identifying sleep factors that influence the onset of CI using an explainable AI algorithm. The proposed CI classification model achieved a sensitivity of 0.89, a specificity of 0.80, and an area under the receiver operating characteristic curve of 0.92. This study will facilitate the noninvasive screening, diagnosis, and continuous monitoring of CI in the elderly.