How well Garvan and FRAX models predict fracture?
Osteoporosis Int has just published an interesting paper that answers the above question. The authors conclude that "Overall, the FRAX and Garvan calculators underestimated major osteoporotic fracture and fragility fracture, particularly in indiviuals with osteopenia or osteoporosis. Hip fractures were predicted better by both calculators" (1). However, a close reading of the data reveals substantial differences in prognostic performace between the two predictive models.
The study was based on the Geelong Osteoporosis Study (GOS) cohort. The cohort included 809 women and 821 men aged between 50 and 90 years at study entry. The investigators applied the Garvan and FRAX models to estimate 10-year risk of fragility fracture and hip fracture on each participant in the study. Key results of the study can be summarized by the following table.
As can be seen from the above summary, the Garvan model consistently performed better than the FRAX model in terms of fracture prediction in men as well as in women. When data from the two sexes are combined, out of 100 fragility fractures, the Garvan model identified 77 individuals at high risk; the FRAX model identified only 41. For hip fracture, out of 100 actual fractures, the Garvan model predicted 120 (over-estimation) and the FRAX model predicted 51 (under-estimation).
When the analysis was limited to osteopenic sample, the Garvan model again outperformed the FRAX model. For instance, out of 100 fragility fractures, the Garvan model predicted 83 people at risk, but the FRAX model identified only 40. For hip fracture, out of 100 fractures, the Garvan model slightly overpredicted (106), but the FRAX model predicted only 48 people at high risk.
Consequently, the sensitivity of the Garvan model was greater than FRAX's. For example, in women, the sensitivity of total fracture for Garvan (77.2%) was significantly higher than FRAX (23.5%). Moreover, the sensitivity of hip fracture prediction for Garvan (~93%) was higher than FRAX (71%).
In men, the results were rather strange: the sensitivity of total fracture prediction for FRAX was only 1.4%, substantially lower than the Garvan model (61.5%). Again, the sensitivity of hip fracture in men for FRAX (41%) was also substantially lower that Garvan's (76.5%).
Overall, these results indicate that the Garvan model could identify people at high risk of fracture more accurately than the FRAX model. These results are actually consistent with a previous study in New Zealand, where the Garvan model predicted almost 100% of 279 fracture cases, but FRAX underestimated the actual fracture cases by 50% (2). The next question is: does bisphosphonates or denosumab reduce the risk of fracture among those with high risk of fracture (by the Garvan model)?
From Bolland et al (2).
I am looking for collaboration to validate and compare the prognostic performance of the Garvan and FRAX models in fracture risk assessment. If you are interested in collaborating with me, please drop me an email.
(1) Holloway-Kew, et al. How well do the FRAX (Australia) and Garvan calculators predict incident fractures? Data from the Geelong Osteoporosis Study. Osteoporosis Int 2019.
(2) Bolland MJ, et al. Evaluation of the FRAX and Garvan Fracture Risk Calculators in Older Women. J Bone Miner Res 2011; 26:420-427.