Elsevier

Bone

Volume 26, Issue 4, April 2000, Pages 387-391
Bone

Original Articles
Prediction of fracture from low bone mineral density measurements overestimates risk

https://doi.org/10.1016/S8756-3282(00)00238-6Get rights and content

Abstract

There is a well-established relationship between bone mineral density (BMD) and fracture risk. Estimates of the relative risk of fracture from BMD have been derived mainly from short-term studies in which the correlation between BMD at assessment and BMD in later life ranged from 0.8 to 0.9. Because individuals lose bone mineral at different rates throughout later life, the long-term predictive value of low BMD is likely to decrease progressively with time. This article examines and formalizes the relationship between current BMD, correlation coefficients, and long-term risk. The loss of predictive value has important implications for early assessment and supports the view that measurements should be optimally targeted at the time interventions are contemplated and, when necessary, repeated in later life.

Introduction

Of the osteoporotic fractures that arise, the greatest morbidity and mortality is associated with hip fracture.23 Hip fracture also accounts for the greatest burden to health services in terms of direct costs, and to society in health economic terms.1, 21 Because bone mineral density (BMD) measurements aid in the prediction of hip fractures and other osteoporotic fractures,17 there has been a great deal of interest in the use of BMD measurements and other risk assessments to identify individuals at risk for fracture in the hope that interventions can be undertaken in the most cost-effective manner.

There have been many agents developed for clinical use that are capable of modulating bone mineral density in individuals with osteoporosis and, in many cases, the effects on bone mineral density have been shown to be associated with a significant decrease in the risk of osteoporotic fractures.14, 19 Most of these treatments have been given to relatively young women, often with an average age range of 50–65 years. The average age of hip fracture in many countries, however, is ≥80 years,5 15–30 years after most interventions are contemplated. Thus, most hip fractures are likely to occur 15–30 years after assessment of bone mineral density. For these reasons, it is important to assess the effects of risk assessment over a lifetime, rather than over a few years, in the natural history of osteoporosis. In an earlier study we showed that lifetime risk of fracture is underestimated when current trends in mortality are ignored.20 In this study we show that estimates of risk based on low BMD produced from relatively short-term studies overestimate the risk over the lifetime of patients.

Section snippets

Methods and assumptions

Bone mineral density at any given age is assumed to be normally distributed. The risk of hip fracture is presumed to increase by a constant fraction for each standard deviation decrease in bone mineral density. In this article we model apparent gradients of risk of 1.4- and 2.6-fold increases in risk for each standard deviation decrease in bone mineral density. The former approximates the lower estimates of the risk of hip fracture provided by single-photon absorptiometry at the forearm from

Results

The effect of time on correlation coefficients between two sets of BMD measurements is shown in Table 1 using observed coefficients ranging from 0.4 to 0.8. Over all ranges of assumptions and intervals, correlation coefficients are expected to decline with time. For example, where studies undertaken for 5 years show a correlation coefficient of 0.8, the true correlation coefficient 15 years later is likely to be 0.41. The effect is proportionately greater with poorer correlation coefficients.

Discussion

Current definitions of osteoporosis incorporate low bone mineral density as a cornerstone.2 Diagnostic criteria for osteoporosis have been developed based on the relationship of a given bone mineral density to that of the young, healthy population.23 The diagnosis of osteoporosis is only of relevance where it helps in deciding on management, or gives useful prognostic information. With regard to the latter, there are many prospective observational studies that have assessed the utility of bone

Acknowledgements

We are grateful to Lilly Research Centre for their support of this work.

References (23)

  • E.A. Chrischilles et al.

    Costs and health benefits effects of osteoporotic fractures

    Bone

    (1994)
  • J. Falch et al.

    Perimenopausal appendicular bone lossA ten year prospective study

    Bone

    (1990)
  • Diagnosis, prophylaxis and treatment of osteoporosis

    Am J Med

    (1993)
  • J.W. Davis et al.

    Comparison of cross-sectional and longitudinal measurements of age related changes in bone mass

    J Bone Miner Res

    (1989)
  • J.W. Davis et al.

    The long-term precision of bone loss rate measurements among postmenopausal women

    Calcif Tissue Int

    (1991)
  • L. Elffors et al.

    The variable incidence of hip fracture in southern EuropeThe MEDOS study

    Osteopor Int

    (1994)
  • P. Gardsell et al.

    The predictive value of bone loss for fragility fractures in womenA longitudinal study over 15 years

    Calcif Tissue Int

    (1991)
  • M.A. Hansen et al.

    Spontaneous postmenopausal bone loss in different skeletal areas followed up for 15 years

    J Bone Miner Res

    (1995)
  • M.A. Hansen et al.

    Role of peak bone mass and bone loss in postmenopausal osteoporosis12 year study

    Br Med J

    (1991)
  • Y.F. He et al.

    When should bone density measurements be repeated?

    Calcif Tissue Int

    (1994)
  • C. Huang et al.

    Short term and long term fracture prediction by bone mass measurementA prospective study

    J Bone Miner Res

    (1998)
  • Cited by (54)

    • Composition and characteristics of trabecular bone in osteoporosis and osteoarthritis

      2020, Bone
      Citation Excerpt :

      Therefore, patients with similar low bone densities could have significant microstructural differences. This could explain why the fracture predictive value of DXA scan decreases with age as BMD progressively declines [28,29]. In this study, no differences were found between the two groups regarding the HA crystal c-axis and a-axis size.

    • Cost-effectiveness of different strategies for selecting and treating individuals at increased risk of osteoporosis or osteopenia: A systematic review

      2012, Value in Health
      Citation Excerpt :

      There are many other factors besides BMD, however, that determine fracture risk, for example, variability of bone microarchitecture, liability to fall [7], or bone markers (i.e., serum and urine markers of bone turnover) [8]. Because the correlation between BMD in a person today and 15 years or more later is poor [9] and because many fractures occur among persons with normal bone density [10], screen-and-treat strategies based on BMD alone have been either not cost-effective or cost-effective only within certain ranges of drug costs, age, and societal willingness to pay [11]. Therefore, researchers have included several prescreening tools for the measurement of BMD.

    • Compressive axial mechanical properties of rat bone as functions of bone volume fraction, apparent density and micro-ct based mineral density

      2010, Journal of Biomechanics
      Citation Excerpt :

      Virtual alternatives to mechanical testing often account for the contribution of both tissue material and geometric properties to bone strength. Currently used methods of virtual biomechanics can be categorized into patient specific finite element analysis (FEA) (Martin, 1991; Bessho et al., 2007; Bonnick, 2007) and direct strength assessment based on Dual-energy X-ray Absorptiometry (DXA) (Leichter et al., 1982; Kanis et al., 2000) or quantitative computed tomography (QCT) (Gasser, 1995; Jamsa et al., 1998; Lochmuller et al., 2002; Buckley et al., 2007). While patient specific FEA accounts for changes in tissue material and geometric properties, its main drawback is the cost associated with it, as this method requires large computational power and expert manpower to generate an appropriate model from medical imaging data (Guldberg et al., 1998).

    View all citing articles on Scopus
    View full text