Un enfoque de aprendizaje automático para predecir las necesidades de cuidados a largo plazo e identificar los...

A machine learning model using Extreme Gradient Boosting (XGBoost) was developed to predict long-term nursing care needs among older adults, based on comprehensive claims and health checkup data from Japan's public insurance system. The model demonstrated strong predictive performance (AUC: 0.878; sensitivity: 0.784; specificity: 0.820) on the test dataset, supporting its use for early identification of high-risk individuals. Key risk factors identified through permutation importance and marginal effect analyses included advanced age, prior care needs, neurological and gastrointestinal diseases, as well as specific medical procedures and medications. In contrast, factors such as joint replacement surgery and the use of preventive care services were associated with lower risk. Lifestyle and biochemical indicators, including slower gait speed and low LDL cholesterol, also significantly influenced risk. Constipation, osteoporosis, and lower back pain had relatively small marginal effects, but was associated with a high incidence rate. This model provides a valuable tool for extending healthy life expectancy and optimizing long-term care planning in aging populations, supporting both public health policy and personalized prevention.