Toward the era of precision fracture risk assessment

Fragility fracture poses a major public health problem because of its high prevalence and related morbidity, mortality, and worsened quality of life. From the age of 50, approximately 1 of 2 women and approximately 1 of 3 men will sustain a fracture during their remaining lifetime (1). In women, the remaining lifetime risk of hip fracture is equivalent to or higher than the risk of invasive breast cancer (1, 2), and in men, the risk of hip and clinical vertebral fractures (17%) is comparable to the risk of prostate cancer (2, 3). More importantly, fracture contributes to shortened life expectancy, and the decrement is greater in men than in women (4). Those who survive a fracture often develop 1 or more of chronic pain, increased dependence, and reduced quality of life (5). Although it is critical to treat individuals, who have had 1 fracture to try to prevent this downward spiral, it is arguably more important, from a public health perspective, to prevent the first fracture; and so prevent the cascade of adverse events triggered by the fracture. A strategy of primary prevention is to identify high-risk individuals based on known risk factors, and intervene to reduce their risk at the individual level (6).

Within any population, the risk of fracture varies remarkably between individuals, depending on their risk profile which typically includes gender, age, a personal history of prior fracture, a family history of fracture, fall, comorbidities, and bone mineral density (BMD). In this issue, Nethander and colleagues (7) show that the difference in genetic makeup between individuals also underpins the high individual-level variation in fracture susceptibility. This finding feds into the anticipation that the era of precision fracture risk assessment is gradually coming of age.

In the study, Nethander and colleagues generated a weighted polygenic risk score (PRS) that was aggregated from the effect size of 1103 single nucleotide polymorphism (SNPs). The SNPs were originally associated with eBMD (an estimated BMD from quantitative ultrasound measurement [QUS] of the heel bone) in a large-scale genomewide association study (8). The PRS was statistically associated with the risk of forearm and vertebral fractures, with average odds ratios ranging between 1.32 and 1.46 (7). This finding not only adds to the growing literature of the association between PRS and fracture risk, but also highlights an important clue regarding QUS as an antecedent of fracture. At present, QUS or eBMD is neither used for the diagnosis of osteoporosis nor fracture risk assessment, despite the fact that QUS is associated with fracture risk (9). One could argue that using eBMD is not a measured BMD, and that is a potential weakness. However, this study suggests that the osteoporosis community should take eBMD or its associated PRS more seriously as an indicator of fracture risk.

There is little doubt that fracture risk is heritable. In a previous study, we have created a PRS that we called “osteogenomic profile” from 62 BMD-associated SNPs, and found that each unit increase in PRS was associated with a hazard ratio of 1.20 (95% confidence interval, 1.04-1.38) for total fracture, and this association was independent of age, prior fracture, fall, and femoral neck BMD. In the MrOS cohort, a genetic profiling of 63 SNPs was also associated with the risk of all fractures (10). Two studies in postmenopausal women of Korean background reported that a PRS of 39 SNPs in 30 genomic loci improved the precision of nonvertebral fracture prediction (11), whereas a PRS of 35 SNPs was significantly associated with the risk of vertebral fracture (11, 12) in patients on bisphosphonate treatment. Taken together, these results confirm that the between-individual variability in fracture risks is partly attributable to genetic differences among individuals.

How can we translate the PRS-fracture association into utility for clinical practice? One could argue that PRS be used as a predictor of fracture risk and/or as a quantitative index of family history of fracture. However, given the modest association with fracture (evidenced by a substantial overlap in PRS distribution between fracture and nonfracture groups), PRS alone is unlikely to be a major indicator for fracture risk prediction. However, PRS can have clinical utility when it is combined with clinical risk factors that are currently incorporated into individualized risk assessment models such as the Garvan Fracture Risk Calculator (13, 14) and the FRAX model (15). Indeed, our previous work has suggested that in terms of fracture discrimination, the area under the receiver operating characteristic (ROC) curve for PRS alone was 0.61, but when it was added on top of the Garvan Fracture Risk Calculator, the area under the ROC increased to 0.71 (16). Moreover, we found that the incorporation of PRS into the existing Garvan Fracture Risk Calculator, the correct reclassification of fracture versus nonfracture ranged from 12% for hip fracture to 23% for wrist fracture (16). Thus, PRS could provide additional prognostic information to that obtained from clinical risk factors, particularly in women in the top or bottom deciles of fracture risk (17), and help stratify individuals by fracture status.

PRS can also be used in place of family history in the assessment of fracture risk. Family history of fracture, particularly a family history of hip fracture, is a recognized risk factor for fracture. However, as qualatative information, family history not only depends upon its accuracy and indeed parental survival but also is an unreliable indicator that poorly captures the polygenic nature of risk. PRS can be viewed as the weighted sum of risk alleles that are combinatorially unique to an individual, and that makes PRS an attractive quantitative index of family history.

It is worth emphasizing that PRS is not and cannot be used as a diagnostic test. Like other risk assessment tools (e.g., Garvan, FRAX), PRS can only indicate the risk of fracture, but cannot categorically determine whether an individual will or will not have a fracture. However, the advantage of risk assessment via PRS is that it allows a life-time prediction well before the onset of fracture. Although there is no “genetic therapy” for individuals at high risk of fracture, this salient lifetime prediction may raise immediacy of primary and secondary prevention behaviors in those at heightened genetic risk.

The emergence of PRS has raised some broad theoretical issues. Several PRSs have been constructed to quantify genetic liability for fracture risk prediction, but their validity has neither been studied nor explored. Because each of the PRSs was constructed from a selected number of SNPs, and the selection was mainly based on P value thresholds, it remains largely unknown whether these PRSs capture all aspects of genetic liability (i.e., content validity). It seems that a PRS with a large number of SNPs performs better than the one with fewer SNPs. In the Nethander and colleagues’ study, the 1103 eBMD-based PRS explained 17% of variance in QUS, whereas the 49 femoral neck BMD-based PRS explained only ~4% of variance in femoral neck BMD (7). Previous simulation studies also suggest that by relaxing P value threshold and hence including more SNPs selection, while accounting for genomic correlation, may improve risk prediction performance (18). Moreover, rare variants (i.e., allelic frequencies < 1%) which are not identified by genomewide association study may account for a significant proportion of variance in eBMD and BMD (19). Still, the proportion of variance in fracture risk explained by PRSs is still much lower than the index of heritability, suggesting that the content validity of PRSs requires further research.

The criterion validity of PRS as a predictive test also requires more consideration. In the PRS context, criterion validity refers to the correlation between a polygenic risk score and an existing valid test such as bone mineral density. This is a challenging issue, because as mentioned previously, all SNP-based PRSs are modestly correlated with BMD, suggesting that PRSs have low level of criterion validity. However, that previous PRSs were associated with fracture risk independent of BMD implies that PRSs capture information about bone health that is not measured by BMD. For example, Nethander and colleagues (7) observe that the QUS-based PRS was significantly associated with trabecular bone parameters, implying that PRS may capture some aspects of trabecular bone microarchitecture.

It is highly possible that the osteoporosis field will have multiple PRS for different ethnicities. So far, most genomewide association studies have been performed primarily in Caucasian populations. The SNPs identified in Caucasian populations are likely to have different allelic frequency among non-Caucasians. Moreover, the effect sizes estimated based on Caucasian populations can be different from those observed in non-Caucasians. Thus, a PRS derived from a Caucasian population may not be applicable to non-Caucasians. We need more genomewide association studies to clarify ethnic-specific PRS in non-Caucasian populations.

The assessment of fracture risk, or for that matter any disease risk, should be individualized. This is true because each individual is unique, and there exists no “average individual” in a population. The uniqueness of an individual can be defined in terms of clinical risk factors, but it can also be refined in terms of PRS. Two individuals with the same BMD, same age, same gender, and both may have a history of fracture, but they can have quite different future fracture risk, because they likely have different PRS. However, it should be recognized that although the estimated risk of fracture may be unique, it is not a constant, because PRS may be updated as more putative variants are identified.

In summary, although PRS, as a single indicator, has insufficient predictive utility, it can have prognostic value particularly in moderate- to high-risk individuals when combined with clinical risk factors. In the era of genomic-driven health care and big data, it is very likely that PRS will have a place in risk assessment and risk prediction. With the current technology, it is quite feasible to generate PRS with hundreds of thousands or even millions of SNPs for less than $100; thus, the implementation of PRS into clinical practice is economically and technically feasible. Now is the time to start planning for the next level of fracture risk assessment: precision risk assessment.

Acknowledgments

The author thanks Professor John A. Eisman of the Garvan Institute of Medical Research for his thoughtful comments that led to improvement of the manuscript.

Financial Support: This work is supported in part by a grant from the Amgen Competitive Grant Program (2019).

Additional Information

Disclosure Summary: The author has no disclosures.

References and Notes