Use of age-dependent FRAX-based intervention thresholds for Singapore

Osteoporosis is a common, chronic, and costly condition; the annual economic burden in Singapore associated with fragility fractures was estimated at approximately € 118 million in 2017 and is forecast to increase to € 186.9 million by 2035 [1]. In Europe, the annual cost of fractures associated with osteoporosis exceeded € 37 billion in 2010 [2]. Disability due to fragility fractures was greater than that caused by any single cancer, with the exception of lung cancer and was comparable or greater than that caused by a variety of chronic noncommunicable diseases, such as rheumatoid arthritis-, asthma-, or high blood pressure-related heart disease [3]. Fortunately, a wide range of treatments is available that improve bone mass and decrease the risk of fractures associated with osteoporosis [4]. The use of such interventions by healthcare practitioners is assisted by instruments that assess patients’ fracture risk to optimize clinical decisions about prevention and treatment. The most widely used web-based tool FRAX® (https://​www.​sheffield.​ac.​uk/​FRAX/​) meets these requirements and computes the 10-year probability of fragility fractures based on several common clinical risk factors and, optionally, a bone densitometry result obtained from dual x-ray absorptiometry (DXA) [5, 6]. FRAX models are available for 66 countries covering more than 80% of the world population at risk [7] and have been incorporated into more than 100 guidelines worldwide [8].

A country-specific FRAX model was developed for Singapore which was launched in December 2010. Whereas the model should enhance accuracy of determining fracture probability among the Singaporean population, guidance is not yet available to make decisions about treatment [9]. The aim of the present study was to explore a potential assessment pathway for treatment and characteristics of women selected for treatment in Singapore based on FRAX.

The baseline characteristics are given in Table 1. The prevalence of the simulated variables was similar to the age-matched prevalence of these risk factors from Ms. Os (Table 4, Appendix A). Similarly, age-matched BMD values in the SCHS cohort were similar to those in the Ms. Os cohort (Table 5, Appendix A). In the SCHS cohort, the Osteoporosis Self-Assessment Tool for Asians (OSTA) yielded a sensitivity of 85% and specificity of 50% based on BMD at the femoral neck.

Individual probabilities of hip fracture and a major osteoporotic fracture (with and without BMD) are given in Table 1. The mean probability of a major fracture was 7.5% and, for a hip fracture, was 2.8% when calculated without BMD. Probabilities calculated with BMD were similar. As expected, average fracture probabilities increased progressively with age. For a major osteoporotic fracture, the 10-year probability rose from 2.1% in the age category 50–54 years to 24.3% for the ages 80–84 years.

Hip fracture incidence was recorded during an average of 9.1 years with a maximum of 11.5 years of follow-up. During this period with 266,025 person years of observation, 789 women experienced a first hip fracture (2.7%). FRAX-based hip fracture probabilities predicted 808 hip fractures. There was a close correspondence between FRAX-based hip fracture probability and observed hip fracture rates (Fig. 1).

In this report, we present intervention thresholds and BMD assessment thresholds based on fracture probability using FRAX. The approach used was similar to that adopted by the National Osteoporosis Guideline Group (NOGG) in the UK and more recently in European guidelines [4, 6, 18] but applied to the FRAX model for Singapore. Thus, the intervention threshold was set at the fracture probability equivalent to a woman from Singapore with a prior fragility fracture. The rationale is that if women with a prior fragility fracture are considered eligible for treatment, as commonly considered and, indeed, recommended in the present Singapore guidance [9], then women without fracture but with equivalent probabilities should also be eligible for treatment. The use of a single intervention threshold, despite ethnic-specific FRAX models, is appropriate in the sense that intervention is recommended at an identical fracture probability irrespective of ethnicity.

The starting point in the assessment of women is the presence of a clinical risk factor that alerts the physician to consider osteoporosis. The opportunistic case finding strategy arises because screening the general population is not widely recommended in Asia or Europe, though advocated in North America [12, 21]. In those eligible for assessment and in common with the NOGG guidelines, we limited the use of BMD testing. Individuals with fracture probabilities equivalent to, or lower than, those of women with no clinical risk factors (as used in FRAX) would not be assessed by BMD. At the other extreme, BMD testing was not universally recommended in individuals at high risk. The rationale is that reclassification of risk with the addition of a BMD test (from high risk to low risk and vice versa) is high when fracture probabilities estimated without BMD are close to the intervention threshold, but the likelihood of reclassification decreases the further away the probability estimate is from the intervention threshold [26]. The approach used has been well validated in the UK and Canada [14, 26‐29].

The attraction of this approach is that this makes efficient use of BMD resources. The strategy implies, however, that patients at high risk, but identified without BMD, would respond to pharmacological intervention. The evidence that such patients respond to treatment is strong [29‐35]. A principal reason is that BMD values are low in patients identified with FRAX but without a BMD test [29]. Overall, the mean T-score in women eligible for treatment and selected with FRAX was −2.94 (see Table 3).

In the present study, we have focused on intervention thresholds based on 10-year probabilities of a major osteoporotic fracture. There is, in principle, no reason why a strategy should not be based on the probability of hip fracture. Indeed, screening on this basis has recently been shown to decrease the incidence of hip fracture in the UK [36]. We have also assumed that measurements of BMD are included in the strategy. Where facilities for BMD testing are wanting, FRAX without BMD provides similar predictive value as BMD without FRAX [37]. Nevertheless, the combination of FRAX with BMD where appropriate provides the optimal strategy.

The implementation of this strategy is expected to lessen unnecessary treatment of individuals at low fracture risk and better direct treatments to those at high risk than treatment decisions based only on the measurement of BMD [37]. Implementation will, however, raise immediate problems in that current guidance for treatment in Singapore, and many other countries, is led by measurement of BMD. For example, patients are eligible for treatment with a T-score of −2.5 SD or lower. Thus, it will be important that healthcare agencies are involved in any implementation process.

There are a number of potential limitations of the present study to consider. First, although the cohort was large, it may not be representative of the Singaporean population. A recruitment bias towards healthier individuals is expected to preferentially lower fracture probabilities when BMD is included in the FRAX calculation. It is of interest that fracture probabilities were very similar when calculated with or without BMD, supporting a view that such bias is likely to be small. A more robust argument that biases were small was the close agreement between hip fracture incidence in the SCHS cohort and that predicted from the Singaporean FRAX model. Unfortunately, other outcome fractures were not available from the SCHS cohort to check the predictive value of probability estimates of a major osteoporotic fracture. Another important limitation was that not all FRAX variables were documented in the SCHS cohort and the missing values were simulated using regression equations derived from a Chinese cohort in Hong Kong. The adequacy of the simulations is supported by the similar prevalence of clinical risk factors in the Hong Kong and SCHS cohorts. Moreover, we tested the validity of the simulation of BMD by computing the sensitivity and specificity of OSTA. The sensitivity of 85% and specificity of 50% were very comparable with published estimates of 91 and 45%, respectively [16]. Meta-analyses of studies evaluating OSTA in Caucasian populations using the same cut off threshold of < 1 to identify postmenopausal women with osteoporosis at the femoral neck provided summary sensitivity and specificity estimates of 89% (95%CI 82–96%) and 41 (95%CI 23–59%), respectively [38]. These considerations suggest that the SCHS cohort was representative of the Singaporean population and that the treatment pathways are applicable to the general population.

The present study has shown that it is possible to apply FRAX-based assessment guidelines using the same principles that have been applied to guidelines elsewhere but tailored to the epidemiology of Singapore. The approach to intervention thresholds is based on the principles of case finding and does not consider a health economic perspective. Although the approach has been shown to be cost-effective in a UK setting [39], cost-effectiveness will necessarily differ in the context of Singapore because of different fracture risks and costs. It will be important therefore to underpin these guidelines with an economic assessment. Overcoming these hurdles will, however, improve the delivery of healthcare to those most at need.