Osteoporotic fracture continues to be a critical public health problem worldwide [1
]. One main reason is that the incidence of osteoporotic fracture increases exponentially throughout one’s life [3
]. Approximately 40% of postmenopausal women will suffer at least one fracture in their lifetime [4
]. Additionally, bone fractures often lead to devastating consequences, including functional decline, prolonged disability, and death [7
]. With longevity increasing globally, the potentially high cumulative rate of osteoporosis and fractures, and the associated excess disability and mortality, will lead to an inevitable increase in social and economic burdens worldwide [8
Furthermore, osteoporosis is a silent disease because bone loss occurs without any signs or symptoms [3
]; therefore, fracture prediction becomes critically important. The Fracture Risk Assessment Tool (FRAX), which is the most widely used tool for fracture risk assessment, was developed by the Collaborating Centre for Metabolic Bone Diseases (Sheffield, UK) and is a computer-based program that computes the 10-year probability of major osteoporotic fracture (MOF, a composite of hip, humerus, forearm, and clinical vertebral fractures) and hip fracture. The FRAX can be used with or without femoral neck bone mineral density (BMD) measurement [10
]. Although FRAX improves fracture prediction over the BMD T-score method alone [11
], the FRAX performance of predicting fracture risk varies in different study populations [12
]. Hence, there is room for further improvement in fracture prediction.
FRAX was derived from nine cohorts and has been validated in 11 independent cohorts from around the world [11
]. The US FRAX was calibrated from the data of the Rochester Epidemiology Project [15
] cohort, which was composed predominantly of Caucasians [16
]. For the US minorities, the FRAX estimates were adjusted based on race-specific hip fracture incidence rates and race-specific mortality [17
]. This adjustment was not empirically based. Racial/ethnic differences that influence fracture risk were not adequately taken into account by US FRAX [18
]. Additional studies are needed to examine the performance of FRAX in US minorities.
Additionally, genetic profiling is an essential predisposition to bone deterioration and fragility fractures [19
]. Genetic factors are also determinants of bone structure [20
]. Although FRAX does not factor in genetic elements, mounting evidence shows that fracture susceptibility is genetically determined [21
]. Virtually 50% of the variance in susceptibility to fragility fracture is attributable to genetic elements [22
]. With the advancement of genomic technologies in the past two decades, major genome-wide association studies (GWASs) and genome-wide meta-analyses have successfully identified numerous genetic loci associated with fracture [15
]. To date, the largest genome-wide meta-analysis on fracture, which involved 32,961 participants, revealed 14 single nucleotide polymorphisms (SNPs) associated with fracture [15
]. However, the way in which these SNPs cause bone fragility and associated fracture remains unclear. As the allelic frequency of these discovered SNPs featured high variability in the population, and each SNP is associated with small effect size, the contribution of any single SNP to fracture susceptibility is expected to be minimal [25
]. The cumulative effects of many associated genetic variants possibly cause osteoporotic fracture [26
]. Thus polygenic scores summarized from risk alleles at each locus have commonly been employed to quantify the overall genetic effect contributing to fracture risk [28
The performance of FRAX with different genetic profiling has not been reported in the literature. In addition, the performance of FRAX in minorities of the US was rarely studied. Thus our study aimed 1) to evaluate whether FRAX performs differently in estimating the 10-year absolute probability of MOF and hip fracture in postmenopausal women with different polygenic risk scores, and 2) assess FRAX performance in the prediction of MOF hip fracture in minority women. We also examined if the interaction of race and polygene score impacts the performance of FRAX in fracture prediction.
The present study found that FRAX overestimated the risk of fracture in women aged 50–79 years, and the degree of overestimation by FRAX in the low GRS group is greater than high genetic risk groups in both outcomes of MOF and hip fracture. In the multivariate analysis, genetic profiling was further demonstrated to be a significant predictor of MOF and hip fracture, independent of FRAX probability.
Genetic factors that influence osteoporotic fracture risk have long been recognized. Genetics are determinants of bone structure and thus a predisposition to fragility. Hereditary factors contribute almost half of the variance in fracture susceptibility [22
]. However, genetic factors are not counted in the FRAX or any other existing clinical fracture risk assessment model. Since FRAX is the most commonly used fracture prediction model, determining if the performance of FRAX varies with different genetic profiling has become crucially important. The largest and most updated GWAS meta-analysis has identified 14 SNPs that are significantly associated with fracture risk at a significant genome-wide level [22
]. Although these individual SNPs have modest effect size on fracture risk, the GRS, as summarized from these individual risk SNPs, enables us to examine if FRAX performance varies with different genetic risk factors. The varied prediction performance of FRAX by GRS, as observed in our study, suggests that the accuracy of FRAX can be improved by incorporating genetic profiling. Several studies suggested that including GRS as a predictor may help improve the accuracy of various fracture prediction models. For example, GRS of 39 SNPs increased the precision of non-vertebral fracture prediction in postmenopausal Korean women [31
]. Additionally, GRS based on 63 SNPs improved the accuracy of non-trauma fracture prediction [26
]. One of our recent studies on older US men also found that GRS is one of the most important variables in MOF prediction models developed by the gradient boosting approach [32
The present study also provides compelling evidence that FRAX overestimates the risk of MOF and hip fracture in women 50–79 years old, across all racial groups, but especially in minorities. In Asian, African-American, and Hispanic women, the observed probability of fracture, in terms of both MOF and hip fracture, was significantly lower than the risk estimated by FRAX, indicating that the FRAX did not adequately capture racial and ethnic differences of fracture risk. Additionally, our multivariate analysis demonstrated that race is a significant predictor of MOF and hip fracture independent of the cumulative fracture risk estimated by FRAX, suggesting that FRAX does not have adequate adjustment for racial difference. Racial and ethnic difference that influences fracture risk not being adjusted for adequately in the FRAX has long been a concern [10
]. As we know, the US FRAX was calibrated from the Rochester Epidemiolofy Project data, composed predominantly of Caucasians. For non-Caucasian minorities, the FRAX estimates were adjusted based on race-specific hip fracture incidence rates and race-specific mortality [33
]. This adjustment for minorities in FRAX is not empirically based, thus making the prediction accuracy of FRAX increasingly uncertain, especially for MOF, a composite of hip, humerus, forearm, and clinical vertebral fractures. The current FRAX adjustment model, based on race-specific hip fracture incidence rate and race-specific mortality, remains likely to be inadequate for MOF risk estimation in minorities. In this study, we observed that the overestimated risk for MOF by FRAX was much higher than that for hip fracture, which validated that the US FRAX has not adjusted race adequately for MOF. A prior study conducted on the same WHI sample assessing the accuracy of FRAX without BMD in predicting fracture also demonstrated that the FRAX has significantly lower sensitivity in identifying incidence fractures in African-American and Hispanic women [34
]. Another study on 2266 postmenopausal women who participated in the Hong Kong osteoporosis study revealed that the predictive accuracy of FRAX with BMD was not substantially different from the model with BMD alone [35
]. Considering the generally lower incidence of fracture in Asians than in Caucasians, as well as the prominent effect of BMD in fracture prediction in the Asian population, the absence of BMD in the present study may explain the significant overestimation of fractures in this racial group. Moreover, inconsistent findings regarding the performance of FRAX without BMD was reported in several other studies. Leslie et al. observed that the fracture probability estimated without BMD overestimate risk among the general population [36
], which is consistent with findings from the present study. Other studies have reported underestimation of fracture risk by FRAX [12
], but their methodologies have lately been found to be problematic because they either compared incidence with probabilities or failed to take the competing mortality risk into account [39
When both FRAX probability and race were adjusted simultaneously in the multivariate model, the effect of GRS was reduced, which could be due to the following reasons. First, genetic profiling regarding osteoporosis or osteoporotic fracture varies in different racial groups; the effect of race and GRS on fracture could be overlapping (See Appendix A Figure A1
). Second, the genetic effect on fracture probability may not be fully captured by the limited number of discovered risk SNPs. With more fracture-related genetic components being discovered in the future, a larger effect of GRS on fracture risk prediction can be foreseen.
Limitations to this study are acknowledged. First, the WHI data we used only included women aged 50–79 years, so our findings may not apply to men or to women who are not in the study age range. Second, rare genetic variants with high effect size were not included in the present study, mainly because risk SNPs used in this analysis were identified from a GWAS meta-analysis study, which likely discovered common but not rare variants [15
]. The limited number of fracture-associated SNPs may not capture all genetic risk, which partially explained the reduced effect of GRS in the model when both FRAX probability and race were included. Third, our study only focuses on FRAX without BMD because the BMD measurement was unavailable for most of the study subjects. The performance of FRAX with BMD will be examined in a future study. Finally, the sample size of Asian and American Indian subjects was very small in this study; the results may, therefore, be underpowered.
To the best of our knowledge, this is the first study to assess FRAX performance in the prediction of MOF and hip fractures in groups with different genetic profiling and of various races. Our findings suggested genetic profiling of an individual should be considered in fracture prediction, as genetic factors have been demonstrated to be a significant risk factor for osteoporotic fracture, independent of FRAX. Our results also demonstrated that FRAX performed differently in different races, and thus the effect of race in osteoporotic fracture prediction has not heretofore adequately been taken into account by existing FRAX models. Fully integrating genetic profiling and racial factors into the existing fracture assessment model is very likely to improve the accuracy of fracture prediction. Thus, developing racial/ethnic-specific, individualized fracture risk-assessment models will provide more accurate fracture risk assessment. Further studies, especially those including men, larger samples of minorities, and more comprehensive fracture-associated genetic variants, are clearly warranted.