Latent profile analysis and analytical model construction based on heterogeneity of intrinsic capacity in Chinese older adults

Background Intrinsic capacity (IC) is a critical multidimensional indicator of healthy aging; however, current research on IC primarily emphasizes its longitudinal trajectories, with limited recognition of the heterogeneity within older adults and a notable lack of targeted assessment instruments. This study aimed to identify distinct IC profiles among Chinese older adults and explored an interpretable, web-based analytical model to enable rapid profiles classification. Methods A cross-sectional survey was conducted across 24 provinces in China, enrolling 4,508 older adults from community and institutional care settings via stratified sampling. IC scores were assessed using the MNA-SF, SPPB, and GDS-15. Latent profile analysis (LPA) was employed to delineate IC subgroups. Feature selection integrated univariate regression, random forest, and LASSO regression. An XGBoost algorithm was subsequently trained (70%) to predict profile membership and validated on an independent test set (30%). Model interpretability was enhanced using SHAP and XGBoost gain values. Furthermore, an interactive web-based platform was developed using Python (version 3.9), JavaScript (version 7), and MySQL (version 8.0). Results Based on their five-domain IC scores, two latent profiles were identified: “Overall Low IC” and “Overall High IC” (Entropy = 0.846, P < 0.05). Key predictors selected included self-rated health (SRH) ( OR = 1.26, 95% CI : 1.17–1.36), age ( OR = 0.87, 95% CI : 0.80–0.95), presence of multiple chronic diseases ( OR = 1.15, 95% CI : 1.05–1.25), chest tightness and pain ( OR = 1.28, 95% CI : 1.02–1.62), and mobility issues ( OR = 2.54, 95% CI : 2.08–3.11). The XGBoost model demonstrated robust discriminative performance for the “Overall High IC” profile, achieving an AUC of 0.910 (95% CI : 0.898–0.921) and accuracy of 0.874 (95% CI : 0.862–0.885) in the training set, and an AUC of 0.853 (95% CI : 0.828–0.875) with accuracy of 0.819 (95% CI : 0.800–0.838) in the test set. Decision curve analysis (DCA) confirmed a clinical net benefit within a threshold probability range of 4%−97%. The deployed web-based platform generates stratified predictions within 3 s and demonstrates clinical interpretability at the individual level. Conclusion This study delineates two IC profiles among Chinese older adults. The XGBoost-based model was explored for rapid IC profile classification. These findings highlight heterogeneity in IC and support the feasibility of a simplified digital approach, with external validation warranted.