Cognitive Function Prediction Among Chinese Community-Dwelling Older Adults: A Supervised Machine Learning Approach
Author(s)
Xin Ye, PhD.
Assistant Professor, Fudan University, Shanghai, China.
Assistant Professor, Fudan University, Shanghai, China.
OBJECTIVES: Cognitive impairment is a significant public health issue. The aim of this study was to construct more accurate prediction models for cognitive function in Chinese community-dwelling older adults, using fewer variables. This could help identify cognitive impairment early and support timely interventions for successful cognitive aging.
METHODS: Data from 12,394 older adults (39,886 observations) aged 60 and above from the China Health and Retirement Longitudinal Study (CHARLS) were used. Cognitive function was measured by composite scores of episodic memory and executive function. A total of 68 potential predictors from different life-course circumstances were considered. Recursive feature elimination with cross-validation (RFECV) and gradient boosting regressor (GBR) models were employed for feature selection and to determine feature importance. The Gradient Boosting Classifier (GBC) and GBR models were used to predict current cognitive function, and GBR models were trained to analyze future cognition prediction error.
RESULTS: Ten features were selected for the model: education attainment, childhood friendship, age, IADLs, sleep duration, gender, hukou type, mobility, residence, and social participation. The GBC model achieved an area under the curve (AUC) of 0.818 when predicting the presence of cognitive impairment. When predicting current cognitive function scores as a continuous variable, the GBR model had a RMSE loss of 3.256 in the test set. For future cognition prediction, models considering the current cognitive state outperformed those that did not. The current cognitive score was the most significant feature influencing future cognitive difference, followed by factors like childhood friendships, education attainment, and age.
CONCLUSIONS: The study developed a practical prediction model for cognitive impairment in Chinese community-dwelling older adults. The model can be a useful tool for early identification of cognitive impairment, enabling timely interventions in the community. Future research can build on this to refine models and develop data-driven interventions for better cognitive aging management.
METHODS: Data from 12,394 older adults (39,886 observations) aged 60 and above from the China Health and Retirement Longitudinal Study (CHARLS) were used. Cognitive function was measured by composite scores of episodic memory and executive function. A total of 68 potential predictors from different life-course circumstances were considered. Recursive feature elimination with cross-validation (RFECV) and gradient boosting regressor (GBR) models were employed for feature selection and to determine feature importance. The Gradient Boosting Classifier (GBC) and GBR models were used to predict current cognitive function, and GBR models were trained to analyze future cognition prediction error.
RESULTS: Ten features were selected for the model: education attainment, childhood friendship, age, IADLs, sleep duration, gender, hukou type, mobility, residence, and social participation. The GBC model achieved an area under the curve (AUC) of 0.818 when predicting the presence of cognitive impairment. When predicting current cognitive function scores as a continuous variable, the GBR model had a RMSE loss of 3.256 in the test set. For future cognition prediction, models considering the current cognitive state outperformed those that did not. The current cognitive score was the most significant feature influencing future cognitive difference, followed by factors like childhood friendships, education attainment, and age.
CONCLUSIONS: The study developed a practical prediction model for cognitive impairment in Chinese community-dwelling older adults. The model can be a useful tool for early identification of cognitive impairment, enabling timely interventions in the community. Future research can build on this to refine models and develop data-driven interventions for better cognitive aging management.
Conference/Value in Health Info
2025-09, ISPOR Real-World Evidence Summit 2025, Tokyo, Japan
Value in Health Regional, Volume 49S (September 2025)
Code
RWD291
Topic Subcategory
Health & Insurance Records Systems
Disease
SDC: Geriatrics