Machine learning frailty index from healthcare claims predicts mortality and disability
Integrating diagnoses, medications, and procedures into a 63-item tool outperformed an established multimorbidity index in adults aged 50 and older.
Researchers evaluated middle-aged and older adults aged 50 and older by linking the Taiwan Longitudinal Study on Aging with the National Health Insurance Research Database. Using a development cohort of 3,727 participants and a validation cohort of 5,434 participants, the authors trained an extreme gradient boosting model to predict a survey-based frailty index. The resulting 63-item index integrated diagnoses, medications, and procedures. Across both cohorts, 723 and 580 deaths occurred, respectively. Each 0.1-point increase in the claims-based index was associated with higher one-, three-, and five-year mortality, with adjusted hazard ratios ranging from 1.44 to 1.81. Higher scores also predicted increased risks of disability and healthcare utilization, showing greater discrimination than an existing multimorbidity frailty index.
Why it matters
Tracking frailty using existing medical claims allows researchers and clinicians to gauge biological vulnerability across broad populations without specialized physical testing. Combining medications and procedures alongside diagnoses captures functional decline that standard diagnostic codes miss.
Caveats
The study was observational and conducted exclusively within Taiwan's national health system, which may limit generalizability to other healthcare systems with different coding practices. The model also relies on administrative records that may reflect healthcare access patterns rather than purely biological status.
The paper
Development and validation of a machine learning-based claims frailty index for predicting mortality, disability, and healthcare utilization in middle-aged and older adults
Hwang AC, Tseng SF, Chen LK et al.
Experimental Gerontology · 24 Sep 2026