Development and validation of a successful aging prediction model for older adults in China based on health ecology theory

Zhang Zhucheng^1,2Peng Chenxi³Li Zhuo²Li Jiaqiang³Pan Yuhang^1,2Liu Ruihong^2*Chen Xiangdong^1,4*

• ¹Department of General Practice, Health Science Center, Shenzhen University, Shenzhen, Guangdong Province, China
• ²Department of Family Medicine, Shenzhen Hospital, The University of Hong Kong, Shenzhen, Guangdong Province, China
• ³Department of Biomedical Engineering, Health Science Center, Shenzhen University, Shenzhen, China
• ⁴Department of Otolaryngology Head and Neck Surgery, Shenzhen University General Hospital, Shenzhen University, Shenzhen, China

Background and aim: Accelerated aging poses significant physical, psychological, and social health challenge to Chinese. Successful aging (SA) serves as a proactive approach to population aging, reflecting individual health status and quality of life, thereby enhancing the capacity for healthy living among the elderly. However, the complexity of SA measurement methods often hinders its application in community healthcare. Currently, there is a dearth of prediction model tailored for the elderly in community. This study aimed to develop and validate a prediction model for SA in Chinese community elderly. Methods: Data were derived from the fifth wave of the China Health and Retirement Longitudinal Study (CHARLS), targeting community-dwelling elderly individuals over 60. Employing health ecology theory, we comprehensively utilized variables from community health records. The Shapley Additive exPlanation (SHAP) method identified key variables contributing to outcome prediction. An extreme gradient boosting machine learning method was used to construct the prediction model for SA in Chinese community elderly. The final model was obtained through hyperparameter adjustment via 8-fold cross-validation. The model's performance was evaluated using area under the receiver operating characteristic curves (AUROC), discriminant slope, calibration curves, decision curves, SHAP-based risk factor analysis, and comparison with other methods to assess differentiation, calibration, interpretability, and clinical utility. Results: The model incorporated variables available from community health records. SHAP indicated a robust importance ranking of variable features, with the most frequent top 16 features aligning with clinical practice, ensuring good interpretability and extensibility of the resulting prediction model. We used six machine learning methods to construct the prediction model. Among them, the extreme gradient boosting model demonstrated an AUROC of 0.78, a discrimination slope of 0.140, and a Brier score of 0.124. The proposed model is superior to other methods, and has outstanding discriminability and consistency. Decision curve analysis (DCA) indicated a higher clinical utility compared to other models. Conclusions: We proposed a prediction model for SA in Chinese community elderly based on health ecology theory and machine learning, which demonstrate excellent prediction performance, interpretability, and extensibility. The prediction model can be applied to community elderly health management, promoting SA within community elderly.