Exploratory 6-year prediction of LTCI certification using KCL data: A decision tree analysis in a retrospective observational study

Amid the progression of an aging society, it is essential to develop a long-term predictive model capable of distinguishing between older adults with high and low risks of requiring Long-Term Care Insurance (LTCI) services. However, previous models were limited in that they could not simultaneously examine multiple and heterogeneous relationships, despite the fact that risk factors interact in complex and multifaceted ways. In addition, the existing model was restricted to a short-term prediction span. This study aimed to develop a predictive model for LTCI certification using Kihon Checklist data and a 6-year follow-up data with a decision tree analysis. This study was designed as a retrospective cohort, longitudinal, and observational study. Data were obtained from 3263 individuals aged 65 and older in a community-dweller. The dependent variable was LTCI certification status after 6 years. Independent variables included 6 Kihon Checklist domains, age, and gender. Decision tree analysis and a confusion matrix were used to make and evaluate the model. The decision tree model identified age, gender, and depression risk as key predictive factors. The model demonstrated moderate specificity (79.4%) and high negative predictive value (97.1%) but limited sensitivity (52.2%) and positive predictive value (11.1%). These findings indicate that the model performs better in identifying individuals unlikely to require LTCI certification than in detecting future LTCI users. Although it may provide preliminary population-level risk stratification over a 6-year horizon, its predictive performance is insufficient for standalone clinical or policy decision-making, and further refinement and validation are required.