Analysis

Discrepancies in predicted fracture risk in elderly people

BMJ 2013; 346 doi: http://dx.doi.org/10.1136/bmj.e8669 (Published 21 January 2013) Cite this as: BMJ 2013;346:e8669
  1. Mark J Bolland, senior research fellow1,
  2. Rod Jackson, professor of epidemiology 2,
  3. Greg D Gamble, research fellow1,
  4. Andrew Grey, associate professor of medicine1
  1. 1Department of Medicine, University of Auckland, Private Bag 92019, Auckland 1142, New Zealand
  2. 2Section of Epidemiology and Biostatistics, University of Auckland
  1. Correspondence to: M Bolland m.bolland{at}auckland.ac.nz
  • Accepted 3 December 2012

Calculators used to estimate fracture risk in elderly people give widely different results. Mark Bolland and colleagues argue that treatment decisions should be based on calculators using 3-5 year estimates of risk

Doctors are increasingly using calculators to estimate the absolute risk of fracture in their patients, but we have noticed large differences between risk estimates from different calculators. Specifically, the FRAX calculator can give surprisingly low 10 year estimates of fracture risk in older patients with known risk factors, often much lower than those generated by other fracture risk calculators. When considered alongside guidelines for treatment, these lower estimates could lead to elderly people not being offered treatment. We explain the reasons for the differences in estimates and discuss the implications for clinicians.

Fracture risk calculators

Several calculators have been developed to estimate risk of fracture, of which FRAX, the Garvan calculator, and QFracture, are most widely used (table 1). They are an important advance in the care of patients with osteoporosis because management decisions no longer have to be based solely on bone mineral density. Each calculator was derived from prospective cohort studies, has been validated by independent investigators, and has moderate predictive ability for osteoporotic fractures and moderate to good predictive ability for hip fractures.1 2 3 4 The important differences are the time frame for the estimation of risk and the approach to the competing risk of mortality. The Garvan calculator estimates five and 10 year risks of fracture, QFracture estimates risk of fracture for any whole year between one and 10 years, and FRAX estimates only 10 year risk of fracture. Only FRAX takes account of the competing risk of mortality.

Unlike the Garvan and QFracture calculators, the equations and algorithms underpinning FRAX and the specific methods used to take account of mortality have not been published. This led to its exclusion from a recent systematic review of prognostic models for fracture risk.5 We argue that there are also good reasons for not using it to estimate risk in people over the age of 65.

Table 1

Comparison of calculators for fracture risk

View this table:

Time frame of risk estimation

To be useful in clinical practice, calculators need to give risk estimates for a period over which interventions produce substantial benefit. In clinical trials of potent antiresorptive drugs, the cumulative incidence of fracture differs between the treated group and the placebo group by 12 months and the differences are highly significant at three years.6 7 8 Furthermore, the cumulative incidence of fractures in both the treatment and placebo groups is approximately linear, suggesting that antiresorptive treatment leads to a rapid reduction in the rate of fractures and that the size of the reduction in risk remains relatively constant over the short term. As many elderly patients are at high absolute risk of fracture, treatment can therefore produce clinically meaningful benefits over relatively short time frames.

For example, an elderly person with a risk of fracture of 4% a year could expect a 33% reduction in risk if treated with zoledronate.6 Table 2 (model 1) shows that treatment would lead to an absolute risk reduction of 1.3% at 1 year with the number needed to treat (NNT) to prevent one fracture being 76. By three years, the absolute reduction in fracture risk is 3.8% and the NNT is 26. Thus, for elderly people, it makes clinical sense that fracture risk estimates be generated for short periods, such as 3-5 years. Adopting this approach would also align with current recommendations for an initial 3-5 year course of antiresorptive therapy, then a reassessment of fracture risk and the need for ongoing therapy.9

Table 2

 Effects of different time frames and competing risk of mortality on estimated absolute risk and absolute treatment benefits assuming a constant absolute annual fracture risk of 4% and a 33% relative reduction in risk

View this table:

Incorporation of competing mortality risk

The competing risk of mortality is an important consideration in older people. While many elderly people are at high risk of fracture, a high risk of death means that a substantial proportion will die without sustaining a fracture. Thus when applied to groups of elderly people, long term estimates of fracture risk that do not incorporate mortality risk will overestimate the true risk of fracture because they assume complete survival. However, over a short time frame, the overestimation is minimal because mortality will be low.10 Fracture risk calculators that incorporate mortality risk will be more accurate at a population level over long time frames and will be useful for epidemiological studies and planning healthcare.

The figure shows the effect of incorporating the risk of mortality on fracture prediction. The top panels show the relation between age and the incidence of hip fracture in the United States and United Kingdom.11 12 After age 70, the incidence of hip fracture roughly doubles with each five year increase in age. By holding all other variables in the fracture risk calculator constant, we can determine the contribution of age to estimated fracture risk in each calculator. The bottom panels in the figure show that the estimated 10 year risk of hip fracture for men and women with low bone mineral density (FRAX-UK, Garvan) or a low body mass index (QFracture) differs between the calculators. For Garvan and QFracture, the estimated 10 year risk of hip fracture increases threefold to fivefold after age 70, whereas with FRAX, it increases by less than twofold for women and does not change for men.

Figure1

Incidence of hip fracture in men and women from Olmstead County, US, and Edinburgh, UK (top),11 12 and effect of age on fracture risk estimated by Garvan, FRAX, and QFracture risk calculators. For women, the values used in the calculators were height 165 cm, weight 65 kg, no clinical risk factors, femoral neck bone density T score −3. For men, the values used were: height 175 cm, weight 75 kg, no clinical risk factors, femoral neck bone density T score −2.5. Bone mineral density is not incorporated in QFracture so a body mass index of 20 was used to simulate low bone mineral density for both men and women

When predicting fractures in an individual elderly patient, the advantages of estimating 10 year risk and incorporating mortality risk are not obvious and there are clear disadvantages. Clinicians will need to know the average survival for someone of the same age and sex as their patient and also be able to estimate how long the patient is likely to survive. Fracture risk will be underestimated if a patient is considered likely to survive longer than average and overestimated if predicted survival is shorter than average. Unfortunately, physicians do not predict mortality accurately, even in high risk patients,13 14 and predictive models of survival perform only modestly and have limited clinical value.15

The adjustment for mortality risk in FRAX is based on mortality rates for the population. Thus, the average risk of mortality in a population is applied to an individual, but some individual risk factors that are known to strongly predict mortality, such as recent cardiovascular events or cancer, are not taken into account, which may affect the accuracy of estimates. On the other hand, calculators that do not incorporate mortality risk will overestimate fracture risk for individuals unlikely to survive the entire time period. However, given the problems of using population estimates of mortality and the likelihood that short term risk is more clinically relevant in elderly patients, the benefits of incorporating the competing risk of mortality are outweighed by the disadvantages.

Implications for clinicians

The clinical effect of the differences among fracture risk calculators in the time frame and the handling of mortality risk is greatest in people aged over 65. The potential benefits of treatment can be obscured by calculators such as FRAX that use both a 10 year time frame and incorporate mortality risk. Influential organisations have included 10 year estimates of fracture risk generated using FRAX in their guidelines for osteoporosis treatment. The US National Osteoporosis Foundation recommends treatment of individuals without osteoporosis by bone mineral density criteria who have an estimated 10 year fracture risk ≥20%,16 and the UK National Osteoporosis Guidelines Group recommends treatment of individuals if the 10 year probability of fracture exceeds the average risk of people of the same age and sex who have had a previous osteoporotic fracture,17 18 roughly translating to a ≥20% 10 year risk at age 70 and ≥30% risk at 80.17 If FRAX is used to determine the risk, older people would need to have much higher short term risks of fracture to exceed these thresholds than if the Garvan or QFracture calculators were used because FRAX includes the competing risk of mortality.

Table 2 shows that an older patient with substantial absolute fracture risk reduction (2.9%-3.8%) at three years may not meet current thresholds for treatment based on 10 year risks calculated using FRAX or other models that incorporate mortality. In contrast, younger patients at lower, short term risk of fracture, with less potential to benefit in the short term, could meet these treatment thresholds because of their much lower risk of mortality over 10 years. Basing management advice on FRAX estimates potentially precludes effective treatment of elderly patients, as table 3 shows. Using the National Osteoporosis Guidelines Group guidelines (with FRAX-UK), older men and women are recommended reassurance despite being at substantially higher risk of fracture than younger people in whom treatment is recommended. Using the Garvan calculator would not create this discrepancy.

Table 3

 Four clinical cases showing the effect of age and fracture risk algorithm on treatment recommendations

View this table:

Conclusions

The inclusion of the competing risk of mortality in fracture risk calculators leads to substantially lower 10 year estimates of fracture risk in elderly men and women than estimates generated by calculators that do not incorporate mortality. Furthermore, the combination of a 10 year time frame and adjustments for mortality risk may obscure the short term benefits of treatment, leading to undertreatment of older people at high fracture risk. These problems might have contributed to the finding that applying UK guidelines to a cohort of elderly women identified only a minority of fracture cases.19

We suggest that for elderly patients, clinicians use calculators that provide fracture risk estimates for shorter time periods. Three to five year estimates are clinically relevant in these patients because treatment benefits are substantial within three years; moreover this shorter period aligns with current recommendations for initial duration of antiresorptive drugs. If the expected survival is very short (<2 years) then risk of death should be considered, although it is unknown whether fracture calculators provide accurate risk estimates in such patients.

Notes

Cite this as: BMJ 2013;346:e8669

Footnotes

  • Contributors and sources: MJB and AG are endocrinologists and undertake research into the management of osteoporosis. RJ is an epidemiologist whose main research interest is risk prediction. GDG is a biostatistician. AG, MJB, and GDG conceived of the paper. MB drafted the paper. RJ helped to redraft the paper. MJB and GDG carried out the analyses. All authors critically reviewed the manuscript. MB is the guarantor.

  • Competing interests: All authors have completed the ICMJE unified disclosure form competing interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare no support from any organisation for the submitted work; no financial relationships with any organisation that might have an interest in the submitted work in the previous three years; and no other relationships or activities that could appear to have influenced the submitted work.

  • Provenance and peer review: Not commissioned; externally peer reviewed.

References