How reliable is the FRAX score for assessing risk of osteoporotic fractures?

FRAX model predictions are moderately good. One commonly used measure of a prediction tool's robustness is the C-statistic, which is a generalized analogue of the area under the curve. No predictive value yields C = 0.5, while perfect prediction yields c = 1.0. There are different FRAX models for different populations and they are not equally good, in part because the data used to develop each are not equally good. There are also separate models for hip fracture and major osteoporotic fractures (MOF; hip, clinical spine, humerus, and distal radius). The hip models do better than the MOF models, with c ~0.8 and ~0.7, respectively.

It's important to appreciate the inherent limitations of both FRAX and the C statistic. FRAX focuses on a small n...