A Prediction Model for Mild Cognitive Impairment Using Random Forests

Abstract

Dementia is a geriatric disease which has emerged as a serious social and economic problem in an aging society and early diagnosis is very important for it. Especially, early diagnosis and early intervention of Mild Cognitive Impairment (MCI) which is the preliminary stage of dementia can reduce the onset rate of dementia. This study developed MCI prediction model for the Korean elderly in local communities and provides a basic material for the prevention of cognitive impairment. The subjects of this study were 3,240 elderly (1,502 males, 1,738 females) in local communities over the age of 65 who participated in the Korean Longitudinal Survey of Aging (close) conducted in 2012. The outcome was defined as having MCI and set as explanatory variables were gender, age, level of education, level of income, marital status, smoking, drinking habits, regular exercise more than once a week, monthly average hours of participation in social activities, subjective health, diabetes and high blood pressure. The random Forests algorithm was used to develop a prediction model and the result was compared with logistic regression model and decision tree model. As the result of this study, significant predictors of MCI were age, gender, level of education, level of income, subjective health, marital status, smoking, drinking, regular exercise and high blood pressure. In addition, Random Forests Model was more accurate than the logistic regression model and decision tree model. Based on these results, it is necessary to build monitoring system which can diagnose MCI at an early stage.