Abstract
Frailty increases the risk of adverse outcomes in older adults and is an important target for preventive care. Clinical frailty measures are established but rarely available at scale due to limited functional data in routine healthcare records. Claims-based frailty prediction models offer a pragmatic approach for population-level risk stratification but are often developed in specific healthcare systems and lack external validation. We aimed to externally validate and recalibrate the Claims-Based-Frailty-Indicator (CFI) and develop a claims-based frailty model for the German healthcare context. We used data from the Berlin Initiative Study, a cohort of older adults with assessment of the frailty phenotype, linked to individual health insurance claims and cause-of-death data. External validation of CFI was assessed by discrimination and calibration. Additionally, a new claims-based frailty model was developed. Model performance and application were evaluated through associations with mortality and hospital admission. Reporting followed TRIPOD statement. Among 1,151 participants (mean age 84.4 years; 55% women), the CFI demonstrated good discrimination (AUC 0.78) but poor calibration; recalibration of intercept improved calibration. Maximizing the Youden Index identified an optimal frailty threshold corresponding to a predicted probability of 0.39. The new model showed improved discrimination (AUC 0.82). Predicted frailty was similarly associated with mortality and hospital admissions as the frailty phenotype. Claims-based frailty models can be validated, recalibrated, and adapted to the German healthcare system. They provide a feasible approach for population-level frailty assessment and may support health services research, risk stratification, and care planning for older adults.