Identifying the factors influencing long-term care utilization by older adults in China: machine learning analysis

To address the aging population in China, local governments began to encourage the establishment of formal care services and supplement informal care in 2015, thereby increasing the availability of different types of long-term care (LTC). In this study, nationwide survey data were accessed to identify factors influencing the types of LTC utilization by older adults in China.

Data from 2,305 eligible older adults receiving care were retrieved from the 2020 China Health and Retirement Longitudinal Survey. The independent variables were structured around the predisposing, enabling, and need factors of the Andersen healthcare utilization model, while the dependent variable was categorized by the type of long-term care utilized (informal versus formal care). The analytical pipeline included a 70/30 train-test data split, Bayesian Ridge multiple imputation for missing values, and LASSO logistic regression for feature selection. To correct for severe classification imbalance, cost-sensitive learning via algorithmic class weighting was applied. Eleven machine learning (ML) models were constructed and optimized using fivefold cross-validation within the training set. Finally, the SHapley Additive exPlanations (SHAP) interpretability framework was used on the independent test set to evaluate variable importance, dependencies, and interactions.

Descriptive chi-square test results indicated that age, marital status, social activity, smoking, income state, residence, and living arrangement significantly the type of care received. Following feature selection and evaluation across the 11 ML models, the Random Forest model achieved the highest predictive performance on the independent test set. Subsequent SHAP interpretation of the Random Forest model identified marital status, living arrangement, social activity, physical activity, age, and residence as the most associated variables for formal care utilization. Analysis of specific impacts showed that individuals who lived alone, were over the age of 80, participated in weekly physical activities, or did not participate in social activities were more inclined to use formal care. Overall, marital status, living arrangement, and social activity participation were identified as the key interacting factors associated with the types of long-term care utilized by older adults in China.

Living arrangement, social activity and residence were the most significant factors associated with the types of LTC utilization by older adults in China. Overall, enabling and predisposing factors had a greater influence than the need factors. These findings not only demonstrate the potential value of ML for LTC policy development, but also provide empirical support for the Chinese government to adopt targeted interventions that enhance LTC service accessibility and affordability.