Predicting Future Long-Term Care Dependency with Explainable Artificial Intelligence Using Administrative Healthcare Data

Preventing future long-term care dependency is a major public health challenge in aging societies. We developed an explainable artificial intelligence (AI) model to predict future long-term care dependency among adults aged ≥65 years using large-scale administrative healthcare data from Japan's public insurance system. The model demonstrated good discriminative performance in the randomly split internal test cohort (AUC:0.870, sensitivity: 0.790, specificity: 0.800). Age and baseline long-term care certification status were the strongest predictors. Neurological disorders, dementia-related conditions, cerebrovascular diseases, and other chronic diseases also contributed substantially to the predicted risk. Conversely, the model identified factors associated with reduced predicted risk, including hypercholesterolemia, allergic rhinitis, and chronic periodontitis. The model identified both high-risk individuals and disease-specific factors associated with predicted long-term care risk by quantifying feature importance and impact. Unlike conventional black-box models, the proposed model enables transparent interpretation by providing not only predicted risk values but also the factors associated with the predicted risk. This capability may improve the clinical interpretability and practical applicability of AI-based risk prediction for long-term care prevention.