Predicting cognitive function among Chinese community-dwelling older adults: A supervised machine learning approach.

Abstract

Identifying cognitive impairment early enough could support timely intervention of cognitive impairment and facilitate successful cognitive aging. We aimed to build more precise prediction models for cognitive function using less variable input among Chinese community-dwelling older adults. We used data from a prospective cohort of 13,906 older adults aged 60 years and above from the nationally representative China Health and Retirement Longitudinal Study (CHARLS) 2011-2020. The Gradient Boosting Classifier (GBC) and gradient boosting regressor (GBR) models were used to predict an individual's current cognitive function. For future cognition prediction, we trained GBR models to analyze the prediction error over the years. Among 68 features, ten features were finally selected to develop the model: education attainment, childhood friendship, age, instrumental activities of daily living (IADLs), hukou type, mobility, sleep duration, gender, residence, and social participation. Our model exhibited robust performance in predicting current and future cognitive function. When an individual's current cognitive function was assessed as a dichotomous classification of cognitive impairment presence, the GBC model achieved an area under the receiver operating characteristic (ROC) of 0.832. When the outcome was forecasted as a continuous variable, the model achieved a root mean square error (RMSE) loss of 3.356 in the test set. For predicting future cognition, models taking into account the current cognitive state demonstrated superior performance. Our study offers a practical tool to aid in the early identification of cognitive impairment, thus supporting timely interventions in the community environment and potentially contributing to successful cognitive aging.