Machine learning prediction model for cognitive dysfunction in older adults

Abstract

AbstractBackgroundCognitive decline is a typical aging‐related condition in older adults. Maintaining or improving the cognitive function of older adults has been a major focus of older adults’ welfare policy. Examining predictors of deteriorating older adults’ cognitive decline is critical to support healthcare policies. This study aims to identify predictors of cognitive function in community‐dwelling older adults using machine learning algorithms.MethodWe extracted 6,940 older adults’ demographic and clinical characteristics from the 2018 Korean Longitudinal Study of Aging (KLoSA). We utilized the final sample of 4,282 older adults by removing the missing observations in the study variables (n = 2,658, 38.3%). The dependent variable was the Mini‐Mental State Examination‐K (MMSE‐K), and the independent variables included demographic characteristics, grip strength, the Center for Epidemiologic Studies Depression Scale (CES‐D) score, life habit, presence of chronic disease, and social participations. We used two machine learning (ML) algorithms: random forest and classification and regression tree (CART) to predict older adults’ cognitive function. Mean absolute error (MAE) and root mean square error (RMSE) were used to compare the model accuracy of those machine learning algorithms.ResultThe mean age was 75.7 years (SD = 8.1) with more females in the sample (n = 2,376, 55.5%). The random forest model performed better than CART (MSE: 3.0 and 3.3 and RMSE: 4.0 and 4.4 for random forest and CART sequentially). The random forest model identified the CES‐D, age, grip strength, educational attainment, self‐reported health, residential area, and social participations as predictors of older adults’ cognitive function (the degrees of predictor importance were in order).ConclusionThis study presents a list of predictors for cognitive function in community‐dwelling older adults. Among various predictors, depression in older adults was the most influential variable in predicting older adults’ cognitive function.