Revisiting successful aging through a machine learning approach to quantifying the influence of chronic diseases.

Chronic diseases are highly prevalent among older adults and may be associated with their ability to achieve successful aging, which encompasses five key components: absence of major chronic diseases, freedom from disability, high cognitive function, no depressive symptoms, and active social participation. However, many disease conditions are excluded from conventional definitions of successful aging. This study aims to quantify the predictive effects of multiple chronic diseases on overall successful aging and its five core components, thus providing new evidence to refine aging frameworks. Data were obtained from the China Health and Retirement Longitudinal Study (CHARLS) covering the years 2011, 2013, and 2015. Six common chronic diseases not traditionally included in the successful aging definition were selected: hypertension, dyslipidemia, arthritis, kidney disease, liver disease, and digestive disease. Six machine learning models were applied to construct prediction frameworks of successful aging and chronic diseases. The SHapley Additive Explanations (SHAP) and ALE (Accumulated Local Effects) analyses were used to quantify the impact of each chronic disease on the prediction of successful aging, providing both global and individual-level interpretability. Logistic regression was conducted to examine the associations between key diseases identified by SHAP and the five components of successful aging. A total of 4,385 participants were included initially in this study, and a total of 1,104 participants were selected after propensity score matching for subsequent analysis. After hyperparameter tuning with Optuna, the XGBoost model was chosen for model interpretation and prediction (F1 = 0.781, F2 = 0.891, AUC = 0.707, AUPRC = 0.724). SHAP analysis indicated that hypertension, kidney disease, and arthritis were the most influential predictors of successful aging. Additionally, SHAP results highlighted that sleep duration was also among the most important features. Subsequent logistic regression further revealed that, beyond their associations with the disease component, kidney disease and arthritis were significantly linked to depressive symptoms and cognitive function, while hypertension was strongly associated with physical functioning. Our findings highlight that several chronic diseases not traditionally included in successful aging criteria are significantly associated with aging outcomes. Extending the disease spectrum within the definitions of successful aging may enhance individual-level assessment and provide insights for future research on targeted health interventions for the older adults. Furthermore, these findings may help raise awareness of health factors associated with successful aging.