Longevity & AgingMachine Learning Frailty Index Built from Claims Data Predicts Mortality With High Accuracy
Researchers in Taiwan developed and validated a machine learning-based frailty index derived entirely from health insurance claims data — including diagnoses, medications, and procedures — in adults aged 50 and older. Using an extreme gradient boosting algorithm trained on over 3,700 individuals and validated in more than 5,400 others, the resulting 63-item index predicted 1-, 3-, and 5-year mortality with accuracy comparable to a gold-standard survey-based frailty measure. It also outperformed an established multimorbidity frailty index in predicting disability and healthcare utilization. Because the index relies solely on routinely collected administrative data, it could enable large-scale, automated frailty screening without requiring clinical assessments, opening the door to earlier intervention in at-risk populations.
