Performance of FRAX in older adults with frailty: the Framingham Heart Study

Summary We compared the performance of FRAX according to frailty status in 3554 individuals from the Framingham Study. During 10-year follow-up, 6.9% and 3.0% of participants with and without frailty experienced MOF. Discrimination profiles were lower in participants with frailty compared to those without, but they improved when FRAX included BMD. Introduction Frailty increases fracture risk. FRAX was developed to predict fractures but never validated in individuals with frailty. We aimed to compare the predictive performance of FRAX (v4.3) in individuals with and without frailty. Methods We conducted a cohort study using the Framingham Heart Study. Frailty was defined by the Fried phenotype. Major osteoporotic fractures (MOF) were ascertained from medical records during 10-year follow-up. To evaluate discrimination and calibration of FRAX, we calculated the area-under-the-receiver-operating characteristics curves (AUC) using logistic regression models and observed-to-predicted fracture probabilities. Analyses were stratified by frailty status. Results Frailty was present in 550/3554 (15.5%) of participants. Participants with frailty were older (81.1 vs. 67.6 years), female (68.6% vs. 55.1%), and had greater mean FRAX scores (MOF: 15.9% vs. 10.1%) than participants without frailty. During follow-up, 38 participants with frailty (6.9%) and 91 without (3.0%) had MOFs. The AUC for FRAX (without BMD) was lower in participants with frailty (0.584; 95% CI 0.504–0.663) compared to those without (0.695; 95% CI 0.649–0.741); p value = 0.02. Among participants with frailty, the AUC improved when FRAX included BMD (AUC 0.658, p value < 0.01). FRAX overestimated MOF risk, with larger overestimations in individuals without frailty. Performance of FRAX for hip fracture was similar. Conclusion FRAX may have been less able to identify frail individuals at risk for fracture, as compared with individuals without frailty, unless information on BMD is available. This suggests that BMD captures features important for fracture prediction in frail persons. Future fracture prediction models should be developed among persons with frailty.