Because of this, we also undertook analyses where models were compared at relevant clinical intervention threshold ( Fig. 1). Kanis et al. [23] also criticized comparison of “home selleck chemicals llc grown” models with the FRAX® tool using the population

from which the “home grown” model was derived. This is a relevant concern as the best model to fit a dataset will invariably be a model developed from that particular dataset even if the diagnostic performance may not at all translate to other populations. In our study, we compared the performance of FRAX® and other models to that of age alone. This is a simple epidemiological tabulation of fracture incidence as a function of age and does not constitute a bespoke model to fit the data. Furthermore, OST, ORAI, OSIRIS and SCORE are already well validated simpler tools derived from other cohorts [15], [18], [19] and [20]. Another limitation accurately identified by Kanis et al. [23] is the comparison between predicted and observed outcomes. Since we do not have 10 years of follow-up we look at the observed fractures and compared

it with the FRAX® probability of being in risk of fracture. Moreover, we took time-to-event into account by estimating the Harrell’s C which did not influence the results. Same results were seen in the GLOW study [36]; these results also showed that AUC values and Harrell’s C values were similar for major osteoporotic fractures. Finally, FRAX® adjusts for risk of death while the other tools do not. Our findings, selleck kinase inhibitor however, were robust to competing-risks regression with both incident fractures and death as failure as alternative to Kaplan–Meier analysis. In the analyses with each tool dividing participants into those with high versus low risk of fracture we chose to use the cut-off suggested by the developers from validation studies of tools in Caucasian populations. Different cut-offs have been also recommended even among Caucasian populations from studies validating the tools but there was no clear agreement regarding cut-off values for the different tools [41], [42], ADP ribosylation factor [43] and [44]. One study by Rud et al.

[41] investigated the performance of SCORE, OST and ORAI in a Danish population. The sensitivity of SCORE, OST and ORAI was 69%, 90% and 50%, respectively, when applied as described by the developers. The authors also tried different cut-offs with higher sensitivities, but since the study only included peri- and early postmenopausal women (mean age 50.5 years) and there are no other studies on Danish women confirming the suggested cut-off from Rud et al. [41] we found it most reasonable to use the cut-offs from the developer of the tools in this study. The aim of the different tools, i.e. FRAX® with OST, ORAI, OSIRIS or SCORE, differs. FRAX® predicts the probability of fractures while ORAI, OSIRIS, OST and SCORE are designed to predict low BMD.