Development and validation of a machine learning model for frailty screening using claims data in Japan: the Longevity Improvement & Fair Evidence Study.

BACKGROUND: Frailty is an ageing-associated multidimensional condition linked to higher long-term care (LTC) needs, healthcare expenditures, and mortality. In Japan, the Questionnaire for Medical Checkup of Old-Old (QMCOO) is used to assess frailty, but its implementation is resource-intensive. Claims-based prediction models may offer a scalable alternative for early frailty identification. METHODS: We developed a machine learning model using administrative claims data from older adults to predict frailty status as defined by the QMCOO. In Phase 1, the model was trained and validated using data from a single municipality. In Phase 2, the model's prognostic utility for predicting all-cause mortality was evaluated using data from seven other municipalities. We applied the eXtreme Gradient Boosting algorithm, incorporating demographic variables, LTC use, comorbidities, procedures, and medical device use. Model performance was assessed mainly using the area under the receiver operating characteristic curve (ROC-AUC). Mortality risk was estimated using Kaplan-Meier method and Cox regression models. RESULTS: In Phase 1, a total of 74,148 individuals were included (development cohort: 60,930, validation cohort: 13,218). The model achieved an ROC-AUC of 0.780 in internal validation and 0.728 in external validation. In Phase 2, external validation was conducted in a new cohort of 354,815 individuals. Frailty classification was associated with significantly higher mortality in both the development (hazard ratio: 7.03, 95% confidence interval: 6.47-7.63) and validation (6.75, 6.62-6.89) cohorts. CONCLUSION: This claims-based frailty prediction model showed reasonable performance and prognostic value. It may support efficient, population-level frailty screenings where questionnaire-based assessments are impractical.