Machine learning-based predictive modeling of decline in intrinsic capacity among migrant older adults with children

Background While migrating to join adult children can provide familial support, it may also pose risks to the intrinsic capacity (IC) of older adults due to disrupted social networks and environmental stressors. To address the paradox, we applied machine learning within a health ecology framework to develop risk predictive models and to rank key drivers of IC decline among migrant older adults with children (MOAC). Methods This multi-center, large-sample cross-sectional study was conducted from December 2022 to September 2023. A total of 3016 MOAC were randomly recruited across three Chinese regions (XX, XX, XX). The health ecology model was operationalized into a structured questionnaire to assess intrinsic capacity and its potential risk factors across five domains. Data were analyzed using interpretable machine learning techniques, including predictor importance ranking and partial dependence plots (PDPs). Results The top ten predictors of IC decline were identified as social network, health literacy, sleep quality, migration reason, economic level, number of chronic diseases, primary source of financial support, prior career, multi-layered medical insurance coverage, and age. The Random Forest model provided the most robust prediction of IC decline (AUC: 0.843; 95% CI: 0.820–0.867). PDPs further delineate the non-linear patterns through which these predictors influence IC decline. Conclusions The findings provide practical and actionable guidance for community healthcare providers. Screening based on non-modifiable risks (advanced age, chronic diseases, etc) enable the early identification of high-risk MOAC individuals. Building on this, tailored interventions can be developed to target modifiable factors such as health literacy and social networks, thereby promoting IC and enhancing quality of life.