To date, only the FRAX tool has been integrated into national clinical practice guidelines [63–71], though a recent review of updated guidelines around the world found a wide diversity of approaches [72]. Some guidelines have embraced fracture risk as the preferred decision-making approach, and others are still largely dictated by BMD, whereas others are a hybrid. The National Osteoporosis Foundation (NOF) guidelines are an example of the latter, where treatment is recommended for individuals with an osteoporotic T-score, clinical osteoporosis based upon low-trauma spine or hip fracture, or in individuals with low bone mass (osteopenia) where fracture risk under the US FRAX tools exceeds 20 % for major osteoporotic fractures or 3 % for hip fractures [63, 64]. The NOF intervention thresholds were based on a cost-effectiveness analysis [73]. The Taiwanese guidelines are similar to the NOF guidelines but do not differentiate between those with osteoporosis or osteopenia or treat based upon BMD alone; treatment is suggested for postmenopausal women with osteoporotic fracture after 50 years of age or when FRAX with BMD reveals a 10-year major osteoporotic fracture risk ≥20 % or hip fracture risk ≥3 % [74]. Canadian guidelines recommend treatment initiation based upon high-risk fracture events (hip fracture, spine fracture, or multiple fragility fractures) or where major osteoporotic fracture probability exceeds 20 % [71]. The National Osteoporosis Guideline Group (NOGG), working in collaboration with many other societies from the UK, has fully embraced fracture risk in guiding therapy and recommends that treatment be considered in women with a prior fragility fracture (BMD measurement is optional) or when fracture probability with FRAX exceeds an age-specific intervention threshold [66, 75]. The intervention threshold at each age is set at a risk equivalent to that associated with a prior fracture, resulting in a lower threshold in younger individuals and a higher threshold in older individuals. NOGG restricts BMD to individuals whose fracture risk is close to the intervention threshold: individuals with fracture risk sufficiently below or above the intervention threshold are not recommended for BMD testing. The NOGG guidelines suggested treatment if the chance of fracture in a given individual exceeded that of a subject at a similar age with a prior fragility fracture [66, 75]. A similar approach has been advocated for the European setting by the European Society for Clinical and Economic Aspects of Osteoporosis and Osteoarthritis (ESCEO) and International Osteoporosis Foundation (IOF) [76]. The UK National Clinical Guideline Centre has provided guidance on the selection and use of FRAX UK and QFracture in the care of people who may be at risk of fragility fractures [77]. The diversity of guideline approaches in part reflects regional variations in population disease burden, practice patterns, health-care priorities, and economic considerations.

The use of risk factors for case finding requires that the risk so identified is responsive to a therapeutic intervention. Table 4.1 indicates potentially reversible risk factors for each of the fracture prediction tools that are amenable to intervention, as well as those expected to be responsive to pharmacologic treatment. Some risk factors (e.g., age, sex, parental hip fracture) are clearly not amenable to intervention. Others could be specifically targeted (e.g., optimizing nutritional status and lifestyle factors, avoidance of high-risk medications, fall prevention) independent of pharmacotherapy for osteoporosis. Finally, pharmacologic intervention for osteoporosis targets one risk factor alone (e.g., femoral neck BMD) and would not affect other risk factors or fracture probability estimated in the absence of a BMD measurement.

The distinction between a reversible risk factor and reversibility of risk associated with that risk factor is important to highlight. Age is an example of an irreversible risk factor, but the fracture risk associated with older age has reversibility, i.e., the risk identified by age is amenable to therapeutic intervention. Most clinical trials that have shown efficacy recruited subjects on the basis of low BMD, but some trials have recruited patients on the basis of age, sex, a prior vertebral or hip fracture, and exposure to glucocorticoids [27, 78–81].

In the absence of a specific clinical trial, an alternative approach is to demonstrate through post hoc analyses that the presence (or absence) of a risk factor does not adversely influence therapeutic efficacy. The presence or absence of an interaction between treatment benefit and specific risk factors can help to inform the expectation of reversibility of risk. Lack of interaction implies that groups with or without the risk factors respond similarly. Therefore, even for irreversible risk factors, as long as there is no negative interaction, then this implies that the risk factor will not adversely affect treatment response. Alternatively, where a significant interaction exists, it may help to identify groups in whom greater (or lesser) benefit may be expected. These types of post hoc analyses conducted as part of pivotal clinical trials have generally not shown major interactions with the clinical risk factors used in FRAX [27]. Conversely, patients selected solely on the basis of risk factors for falling may respond less well to agents that preserve bone mass than those selected on the basis of low BMD [82]. In some cases, test for interaction may be nonsignificant due to limited power to detect such effects. Conversely, caution also needs to be exercised in interpreting results from multiple post hoc analyses due to the potential for false-positive findings [83]. For example, post hoc analyses from the FREEDOM study evaluating denosumab examined nine subgroups (age, BMI, femoral neck BMD T-score, prevalent vertebral fracture, prior non-vertebral fracture, estimated creatinine clearance, geographic region, race, and prior use of osteoporosis medications) [84]. No significant treatment interactions were observed on vertebral fracture prevention, but vertebral fracture prevention showed nominally significant interactions with BMI (treatment effective for BMI <25 kg/m² but not for higher BMI; P-interaction 0.0135), baseline femoral neck T-score (treatment effective for T-score < −2.5 but not for higher BMD; P-interaction 0.0229), and prevalent vertebral fracture (treatment effective for those without but not with fracture; P-interaction 0.0377). In contrast, the HORIZON-PFT trial found greater intravenous zoledronic acid effects on vertebral fracture risk in younger women (P-interaction 0.05), normal creatinine clearance (P-interaction 0.04), and body mass index >25 kg/m² (P-interaction 0.02), without significant treatment interactions for hip or non-vertebral fracture [85]. A meta-analysis found that the efficacy or raloxifene on vertebral fracture risk was significantly greater at lower ages [86]. Consistency of treatment effect among subgroups has been observed with alendronate, risedronate, and strontium ranelate [87–90].

Anti-fracture effect of several therapeutic agents has been analyzed based on initial FRAX fracture probabilities. Results have not been consistent between or even within classes of therapeutic agents. Greater relative risk reduction (beyond absolute risk reduction) has been observed with higher FRAX probability measurements with clodronate [91], bazedoxifene [92], and denosumab [93], but not with alendronate [94], raloxifene [86], or strontium ranelate [95]. For example, the FREEDOM study found that denosumab reduced fracture risk to a greater extent in those at moderate to high risk (Fig. 4.4) [93]. For example, at 10 % probability, denosumab decreased fracture risk by 11 % (P = 0.629), whereas at 30 % probability (90th percentile of study population), the reduction was 50 % (P = 0.001). It remains uncertain why differences would exist for some antiresorptive agents and not others. Similar type analyses with the Garvan calculator and QFracture have not yet been performed.

There is an implicit assumption that fracture prediction tools demonstrating increased fracture probability should identify individuals who will benefit from treatment with a reduction in fracture risk. If the relative risk reduction is relatively constant, then greater absolute fracture risk would imply a greater absolute benefit (equivalent to a lower number needed to treat [NNT]). Although intuitive, this is an area of controversy as no clinical trials have prospectively recruited on the basis of fracture probability. Subgroup analyses from two pivotal clinical trials, the Fracture Intervention Trial (alendronate) and the Hip Intervention Program Study Group (risedronate), showed little or no benefit from bisphosphonate therapy in subjects without low BMD [82, 94, 96].

Ultimately, if it were possible to identify factors robustly predicting greater anti-fracture effect, then this might allow for the construction of prediction tools for treatment benefit (as opposed to baseline fracture risk which may or may not always be reversible). The construct of such a tool might be quite different from the fracture probability tool used to identify baseline fracture risk and, as noted above, must consider differences between therapeutic agents. In some studies, change in BMD on therapy [97–99], bone turnover markers [100, 101], and measures of compliance/persistence on therapy have shown an association with treatment response [102]. Whether any such tool would explain sufficient variation in treatment response to be of clinical value would need to be established.

Whether FRAX can be used to assess fracture risk in patients receiving concurrent treatment for osteoporosis has been uncertain since treatment effects are not explicitly accommodated in the model [50]. However, since many individuals were initiated on treatment prior to the availability of FRAX, this potentially limits the use of important information for advising patients on their need for continued treatment or whether treatment could potentially be withdrawn. To address this issue, a large clinical cohort was linked to population-based databases to determine medication prescriptions and fracture outcomes in 35,764 women (age >50 years) and baseline BMD testing and FRAX probabilities [103]. A pharmacy database was used to categorize women using osteoporosis medication as untreated, current high adherence users (medication possession ratio [MPR] >0.80 in the year after BMD testing), current low adherence users (MPR < 0.80), and past users. FRAX and femoral neck BMD alone stratified major osteoporotic and hip fracture risk within untreated and each treated subgroup (all P-values <0.001) with similar area under the receiver operating characteristic curve (Table 4.2). Hazard ratios for prediction of major osteoporotic fractures and hip fracture using FRAX were equally strong regardless of treatment status. In untreated and in each treated subgroup, a stepwise gradient in observed 10-year major osteoporotic and hip fracture incidence was seen as a function of the predicted probability tertile (all p-values <0.001 for linear trend). Concordance (calibration) plots for major osteoporotic fractures and hip fractures showed good agreement between the predicted and observed 10-year fracture incidence in untreated women and each treated subgroup (Fig. 4.5). Only in the highest risk tertile of women highly adherent to at least 5 years of bisphosphonate use was the observed hip fracture risk significantly less than predicted, while major osteoporotic fracture risk was similar to predicted. These data suggest that the FRAX tool can be used to predict fracture probability in women currently or previously treated for osteoporosis.