Whom to treat? The contribution of vertebral X-rays to risk-based algorithms for fracture prediction. Results from the European Prospective Osteoporosis Study

We’re sorry, something doesn't seem to be working properly.

Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.

Abstract

Introduction

Vertebral fracture is a strong risk factor for future spine and hip fractures; yet recent data suggest that only 5–20% of subjects with a spine fracture are identified in primary care. We aimed to develop easily applicable algorithms predicting a high risk of future spine fracture in men and women over 50 years of age.

Methods

Data was analysed from 5,561 men and women aged 50+ years participating in the European Prospective Osteoporosis Study (EPOS). Lateral thoracic and lumbar spine radiographs were taken at baseline and at an average of 3.8 years later. These were evaluated by an experienced radiologist. The risk of a new (incident) vertebral fracture was modelled as a function of age, number of prevalent vertebral fractures, height loss, sex and other fracture history reported by the subject, including limb fractures occurring between X-rays. Receiver Operating Characteristic (ROC) curves were used to compare the predictive ability of models.

Results

In a negative binomial regression model without baseline X-ray data, the risk of incident vertebral fracture significantly increased with age [RR 1.74, 95% CI (1.44, 2.10) per decade], height loss [1.08 (1.04, 1.12) per cm decrease], female sex [1.48 (1.05, 2.09)], and recalled fracture history; [1.65 (1.15, 2.38) to 3.03 (1.66, 5.54)] according to fracture site. Baseline radiological assessment of prevalent vertebral fracture significantly improved the areas subtended by ROC curves from 0.71 (0.67, 0.74) to 0.74 (0.70, 0.77) P=0.013 for predicting 1+ incident fracture; and from 0.74 (0.67, 0.81) to 0.83 (0.76, 0.90) P=0.001 for 2+ incident fractures. Age, sex and height loss remained independently predictive. The relative risk of a new vertebral fracture increased with the number of prevalent vertebral fractures present from 3.08 (2.10, 4.52) for 1 fracture to 9.36 (5.72, 15.32) for 3+. At a specificity of 90%, the model including X-ray data improved the sensitivity for predicting 2+ and 1+ incident fractures by 6 and 4 fold respectively compared with random guessing. At 75% specificity the improvements were 3.2 and 2.4 fold respectively. With the modelling restricted to the subjects who had BMD measurements (n=2,409), the AUC for predicting 1+ vs. 0 incident vertebral fractures improved from 0.72 (0.66, 0.79) to 0.76 (0.71, 0.82) upon adding femoral neck BMD (P=0.010).

Conclusion

We conclude that for those with existing vertebral fractures, an accurately read spine X-ray will form a central component in future algorithms for targeting treatment, especially to the most vulnerable. The sensitivity of this approach to identifying vertebral fracture cases requiring anti-osteoporosis treatment, even when X-rays are ordered highly selectively, exceeds by a large margin the current standard of practice as recorded anywhere in the world.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2

References

  1. 1.

    WHO Study Group (1994) Assessment of fracture risk and its application to screening for postmenopausal osteoporosis. World Health Organisation Technical Report, vol. 843, Geneva

    Google Scholar 

  2. 2.

    National Osteoporosis Foundation (1998) Osteoporosis: review of the evidence for prevention, diagnosis and treatment and cost-effectiveness analyses. The basis for a guideline for the medical management of osteoporosis. Osteoporos Int 8[Suppl 4]:S1–S80

    Google Scholar 

  3. 3.

    Hauselmann HJ, Rizolli R (2003) A comprehensive review of treatments for postmenopausal osteoporosis. Osteoporos Int 14:2–12

    PubMed  Article  CAS  Google Scholar 

  4. 4.

    McCreadie BR, Goldstein SA (2000) Biomechanics of fracture: is bone mineral density sufficient to assess risk? J Bone Miner Res 15:2305–2308

    PubMed  Article  CAS  Google Scholar 

  5. 5.

    Seeman E (1997) From density to structure: Growing up and growing old on the surfaces of bone. J Bone Miner Res 12:509–521

    PubMed  Article  CAS  Google Scholar 

  6. 6.

    Rowe RE, Cooper CC (2000) Osteoporosis services in secondary care: a UK survey. J R Soc Med 93:22–24

    PubMed  CAS  Google Scholar 

  7. 7.

    Klotzbuecher CM, Ross PD, Landsman PB, Abbott TA III, Berger M (2000) Patients with prior fractures have an increased risk of future fractures: a summary of the literature and statistical synthesis. J Bone Miner Res 15:721–727

    PubMed  Article  CAS  Google Scholar 

  8. 8.

    Watts NB, Josse RG, Hamdy RC, Hughes RA, Manhart MD, Barton I et al (2003) Risedronate prevents new vertebral fractures in postmenopausal women at high risk. J Clin Endocrinol Metab 88:542–549

    PubMed  Article  CAS  Google Scholar 

  9. 9.

    Lunt M, O’Neill TW, Felsenberg D, Reeve J, Kanis JA, Cooper C et al (2003) Characteristics of a prevalent vertebral deformity predict subsequent vertebral fracture: results from the European Prospective Osteoporosis Study. Bone 33:505–513

    PubMed  Article  Google Scholar 

  10. 10.

    Reeve J, Lunt M, Felsenberg D, Silman AJ, Scheidt-Nave C, Poor G et al (2003) Determinants of the size of incident vertebral deformities in European men and women in the 6th –9th decades of age: the European Prospective Osteoporosis Study (EPOS). J Bone Miner Res 18:1664–1673

    PubMed  Article  CAS  Google Scholar 

  11. 11.

    Felsenberg D, Silman AJ, Lunt M, Armbrecht G, Ismail AA, Finn JD et al (2002) Incidence of vertebral fractures in Europe: results from the European Prospective Osteoporosis Study (EPOS). J Bone Miner Res 17:716–724

    Article  Google Scholar 

  12. 12.

    Felsenberg D, Wieland E, Gowin W, Armbrecht G, Bolze X, Khorassani A et al (1998) Morphometrische Analyse von Rontgenbildern der Wirbelsaule zur Diagnose einer osteoporotischen Fraktur. Med Klinik 93[Suppl 2]:26–30

    Article  Google Scholar 

  13. 13.

    Ismail AA, O’Neill TW, Cockerill W, Finn JD, Cannata JB, Hoszowski K et al (2000) Validity of self-report of fractures: results from a prospective study in men and women across Europe. Osteoporos Int 11:248–254

    PubMed  Article  CAS  Google Scholar 

  14. 14.

    O’Neill TW, Cooper C, Cannata JB, Diaz Lopez JB, Hoszowski K, Johnell O et al (1994) Reproducibility of a questionnaire on risk factors for osteoporosis in a multicentre prevalence survey: the European Vertebral Osteoporosis Study. Int J Epidemiol 23:559–565

    PubMed  Article  CAS  Google Scholar 

  15. 15.

    Cameron CA, Trivedi PK (1998) Regression analysis of count data. In: Hammond P, Holly A (eds) Econometric society monographs. Cambridge University Press, pp 70–85

  16. 16.

    Looker AC, Orwoll ES, Johnston CCJr, Lindsay RL, Wahner HW, Dunn WL et al (1997) Prevalence of low femoral bone density in older US adults from NHANES III. J Bone Miner Res 12:1761–1768

    PubMed  Article  CAS  Google Scholar 

  17. 17.

    Dequeker J, Pearson J, Reeve J, Henley M, Bright J, Felsenberg D et al (1995) Dual X-ray absorptiometry: cross-calibration and normative reference ranges for the spine: results of a European community concerted action. Bone 17:247–254

    PubMed  Article  CAS  Google Scholar 

  18. 18.

    Kaptoge SK, Armbrecht G, Felsenberg D, Lunt M, O'Neill TW, Silman AJ, et al (2004) When should the doctor order a spine X-ray? Identifying vertebral fractures for osteoporosis care: results from the European Prospective Osteoporosis Study (EPOS). J Bone Miner Res 19:1982–1993

    PubMed  Article  Google Scholar 

  19. 19.

    O’Neill TW, Cockerill W, Matthis C, Raspe HH, Lunt M, Cooper C et al (2004) Back pain, disability, and radiographic vertebral fracture in European women: a prospective study. Osteoporos Int 15:760–765

    PubMed  Article  CAS  Google Scholar 

  20. 20.

    van Staa TP, Dennison EM, Leufkens HGM, Cooper C (2001) Epidemiology of fractures in England and Wales. Bone 29(517–522)

    PubMed  Article  Google Scholar 

  21. 21.

    Gehlbach SH, Bigelow C, Heimisdottir M, May S, Walker M, Kirkwood JR (2000) Recognition of vertebral fracture in a clinical setting. Osteoporos Int 11:577–582

    PubMed  Article  CAS  Google Scholar 

  22. 22.

    Sahota O, Worley A, Hosking DJ (2000) An audit of current clinical practice in management of osteoporosis in Nottingham. J Public Health Med 22:466–472

    PubMed  Article  CAS  Google Scholar 

  23. 23.

    Cummings SR, Bates D, Black DM (2002) Clinical use of bone densitometry. JAMA 288:1889–1900

    PubMed  Article  Google Scholar 

  24. 24.

    Kanis JA, Delmas P, Burckhardt P, Cooper C, Torgerson D (1997) Guidelines for diagnosis and management of osteoporosis. Osteoporos Int 7:390–406

    PubMed  Article  CAS  Google Scholar 

  25. 25.

    Cooper C, Atkinson EJ, O’Fallon WM, Melton LJ III (1992) Incidence of clinically diagnosed vertebral fractures: a population-based study in Rochester, Minnesota, 1985–1989. J Bone Miner Res 7:221–227

    PubMed  CAS  Article  Google Scholar 

  26. 26.

    Kanis JA, Johnell O, Oden A, Borgstrom F, Zethraeus N, De Laet C et al (2004)The risk and burden of vertebral fractures in Sweden. Osteoporos Int 15:20–26

    PubMed  Article  CAS  Google Scholar 

  27. 27.

    Melton LJ III, Lane AW, Cooper C, Eastell R, O’Fallon WM, Riggs BL (1993) Prevalence and incidence of vertebral deformities. Osteoporos Int 3:113–119

    PubMed  Article  Google Scholar 

  28. 28.

    Delmas PD, Genant HK, Crans GG, Stock JL, Wong M, Siris E et al (2003) Severity of prevalent vertebral fractures and the risk of subsequent vertebral and nonvertebral fractures: results from the MORE trial. Bone 33:522–532

    PubMed  Article  CAS  Google Scholar 

  29. 29.

    Roy DK, O’Neill TW, Finn JD, Lunt M, Silman AJ, Felsenberg D et al (2003) Determinants of incident vertebral fracture in men and women: results from the European Prospective Osteoporosis Study (EPOS). Osteoporos Int 14:19–26

    PubMed  Article  CAS  Google Scholar 

  30. 30.

    Kanis JA, Johnell O, De Laet C, Johansson H, Oden A, Delmas P et al (2004) A meta-analysis of previous fracture and subsequent fracture risk. Bone 35:375–382

    PubMed  Article  CAS  Google Scholar 

  31. 31.

    Kanis JA, Johansson H, Oden A, Johnell O, de Laet C, Melton LJ et al (2004) A meta-analysis of prior corticosteroid use and fracture risk. J Bone Miner Res 19:893–899

    PubMed  Article  Google Scholar 

Download references

Acknowledgements

The study was financially supported by a European Union Concerted Action Grant under Biomed-1 BMH1CT920182) and also by EU grants C1PDCT925102, ERBC1PDCT 930105 and 940229. The central coordination was also supported by the UK Arthritis Research Campaign, the Medical Research Council (G9321536) and the European Foundation for Osteoporosis and Bone Disease. The EU's PECO program linked to BIOMED 1 funded in part the participation of the Budapest, Warsaw, Prague, Piestany, Szczecin and Moscow centres. Data collection from Zagreb was supported by a grant from the Wellcome Trust. The central X-ray evaluation was generously sponsored by the Bundesministerium fur Forschung and Technologie, Germany. Local or National research funds supported the participation of: Austria: University Hospital, Graz; Belgium: University Hospital, Leuven; Croatia: Clinical Hospital, Zagreb; Czech Republic: Charles University, Prague; Germany: Behring Hospital, Berlin; Humboldt University, Berlin; Ruhr University, Bochum; Medical Academy, Erfurt; University of Heidelberg; Clinic for Internal Medicine, Jena; Institute of Social Medicine, Lubeck; Greece: University of Athens; Hungary: National Institute of Rheumatology and Physiotherapy, Budapest; Italy: University of Siena; The Netherlands: Erasmus University, Rotterdam; Portugal: Hospital de Angra do Herismo, Azores; Hospital de San Joao, Oporto; Poland: PKP Hospital, Warsaw; University School of Medicine, Szczecin; Russia: Institute of Rheumatology, Moscow; Medical Institute, Yaroslavl; Slovakia: Institute of Rheumatic Diseases, Piestany; Spain: Asturia General Hospital, Oviedo; Sweden: Lund University, Malmö; United Kingdom: University of Aberdeen; Royal National Hospital for Rheumatic Diseases, Bath; University of Sheffield; University of Southampton; Royal Cornwall Hospital, Truro. The EVOS-EPOS study also received targeted start-up and continuation support from the European Foundation For Osteoporosis (now the International Osteoporosis Foundation). SK was the recipient of an ECTS Young Investigator award.

Author information

Affiliations

Authors

Corresponding author

Correspondence to S. Kaptoge.

Additional information

A.J. Silman and J. Reeve are the EU Grant holders and Project Leaders.

Appendices

Appendix 1

Negative binomial regression

Negative binomial regression is useful for modelling count data when there is evidence of over-dispersion, meaning that the variance of the count is greater than the mean (a condition, for example, that must be fulfilled if Poisson regression is to be used). The negative binomial model estimates the variance as an algebraic function of the mean (a variance function) and therefore provides more conservative estimates of standard errors and confidence intervals than Poisson regression.

In general, we observe outcome variables y 1 ,y 2 , ···, y n with y i =0,1,2,··· incident fractures for i=1,2,···, n participants and for the i th participant we are interested in measuring the effect of k explanatory variables \(x_{i} = {\left( {x_{{i1}} ,x_{{i2}} , \cdots ,x_{{ik}} } \right)}^{\prime }.\) Using the exponential mean function, the conditional mean μ i (which in this paper is the expected number of incident fractures given covariates) is given by:

$$E{\left[ {y_{i} \left| {x_{i} } \right.} \right]} = \mu _{i} = \exp {\left( {x^{\prime }_{i} \beta + offset} \right)}$$

and the unknown vector of coefficients β is estimated from the data by maximum likelihood. The offset term is a variable whose coefficient is constrained to be 1 and is often a denominator to standardise the observed counts (in this case follow-up time). The variance in the Poisson model is μ i , whereas in the negative binomial model the variance is given by:

$$\omega _{i} = \mu _{i} {\left( {1 + \alpha } \right)} = \mu _{i} \delta$$

where the term \(\delta = {\left( {1 + \alpha } \right)}\) is known as the dispersion, and the negative binomial model reduces to the Poisson model if α=0. For our incident vertebral fracture data, the value of α was 1.05 [95% CI (0.76, 1.45); p<0.0001] in a model not having covariates and was 0.99 [(0.71, 1.37);p<0.0001] in Model 4, giving over-dispersion of 2.05 and 1.99 respectively. After modelling the expected counts as a function of the explanatory variables as above and obtaining the conditional mean μ i and the dispersion parameter \(\delta = {\left( {1 + \alpha } \right)}\), the predicted probability of having 0, 1, 2, ... incident vertebral fractures can then be calculated using the negative binomial probability density function

$$P{\left( {y_{i} = y\left| x \right._{i} } \right)} = {\left( {\frac{\delta }{{\delta + \mu _{i} }}} \right)}^{\partial } \frac{{\Gamma {\left( {\delta + y} \right)}}}{{\Gamma {\left( {y + 1} \right)}\Gamma {\left( \delta \right)}}}{\left( {\frac{{\mu _{i} }}{{\delta + \mu _{i} }}} \right)}^{y} \,y = 0,1,2, \cdots $$

Where Γ is known as the gamma function and is defined by the integral

$$\Gamma {\left( x \right)} = {\int\limits_0^\infty {e^{{ - t}} t^{{x - 1}} dt\quad x > 0} }$$

Appendix 2

A clinical algorithm for patients with lateral X-rays of the thoracic and lumbar spine to guide treatment decisions in clinical practice

As discussed, when presented with a patient without a prior X-ray of the dorsal and lumbar spine, it might be appropriate to order an X-ray to identify vertebral fractures that have already occurred, using the algorithm below from Kaptoge et al. [18] to guide the decision to order the X-ray:

$$\begin{array}{*{20}l} {{{\text{Prevalent}}\,fx\,{\text{Score}}\,{\text{Score}}\,{\text{Female}} = {{\left( {5 \times age + 6 \times htloss - 2 \times weight} \right)}} \mathord{\left/ {\vphantom {{{\left( {5 \times age + 6 \times htloss - 2 \times weight} \right)}} {10}}} \right. \kern-\nulldelimiterspace} {10}} \hfill} \\ {{\begin{array}{*{20}l} {{ + \left\{ {\begin{array}{*{20}l} {{20\,{\text{if}}\,hvert = Yes} \hfill} \\ {{0\,{\text{if}}\,hvert = No} \hfill} \\ \end{array} } \right.} \hfill} & {{ + \left\{ {\begin{array}{*{20}l} {{6\,{\text{if}}\,hotherfx = Yes} \hfill} \\ {{0\,{\text{if}}\,hotherfx = No} \hfill} \\ \end{array} } \right.} \hfill} \\ \end{array} } \hfill} \\ \end{array} $$
$$\begin{array}{*{20}l} {{{\text{Prevalent}}\,fx\,{\text{Score}}\,{\text{Score}}\,{\text{Male}} = {20 + {\left( {3 \times age + 6 \times htloss - 2 \times weight} \right)}} \mathord{\left/ {\vphantom {{20 + {\left( {3 \times age + 6 \times htloss - 2 \times weight} \right)}} {10}}} \right. \kern-\nulldelimiterspace} {10}} \hfill} \\ {{\begin{array}{*{20}l} {{ + \left\{ {\begin{array}{*{20}l} {{16\,{\text{if}}\,hvert = Yes} \hfill} \\ {{0\,{\text{if}}\,hvert = No} \hfill} \\ \end{array} } \right.} \hfill} & {{ + \left\{ {\begin{array}{*{20}l} {{4\,{\text{if}}\,hotherfx = Yes} \hfill} \\ {{0\,{\text{if}}\,hotherfx = No} \hfill} \\ \end{array} } \right.} \hfill} \\ \end{array} } \hfill} \\ \end{array} $$

Where hvert=Yes when a positive history of a vertebral fracture is obtained from the patient but the X-ray cannot be located and hotherfx=Yes indicates an affirmative answer to the question: have you had a fracture elsewhere than in the spine since reaching adult life? A score of >23.1 in women or >27.9 in men was found to yield one fracture case for every five subjects X-rayed and raising the threshold would increase the proportion of positive X-rays while lowering it would increase the absolute yield [18].

For an individual with known values of age, sex, height loss in centimetres since age 25 years (htloss), and the number of prevalent vertebral fractures (prevfx), a simplified score with integer coefficients was derived as:

$$Model4\,Simple\,Score = age + htloss + \left\{ {\begin{array}{*{20}l} {{15\,{\text{if}}\,sex = Female} \hfill} \\ {{0\,{\text{if}}\,sex = Male} \hfill} \\ \end{array} + } \right.\left\{ {\begin{array}{*{20}l} {{0\,{\text{if}}\,prevfx = 0} \hfill} \\ {{25\,{\text{if}}\,prevfx = 1} \hfill} \\ {{25\,{\text{if}}\,prevfx = 2} \hfill} \\ {{50\,{\text{if}}\,prevfx = 3 + } \hfill} \\ \end{array} } \right.$$

The Pearson correlation between the simplified linear predictor score with the score estimated from the actual model coefficients was 0.99, preserving exactly the rank ordering of individuals. Table A1 shows the score’s predictive accuracy. It was found that a decision criterion based on cut-off score category ≥7 (i.e. simple score >81.3) usefully predicted 1+ incident vertebral fractures. The sensitivity was 63% and the specificity was 71% with a positive likelihood ratio of 2.17. The algorithm’s performance was almost unaffected by assuming that X-rays were available selectively, as suggested by our previous algorithm [18], with the sensitivity becoming 57%, the specificity 75% and the positive likelihood ratio 2.28 at a saving of X-rays not done in 65% of subjects. The sensitivity of this approach to identifying vertebral fractures in the over 50’s population exceeds by a large margin the current standard of practice as recorded anywhere in the world.

Table A1

Sensitivities, specificities, likelihood ratios and predictive values from using different cut-points of the simplified linear predictor score based on age, sex and X-ray assessment of prevalent vertebral fractures to predict one or more (1+) incident vertebral fractures. Area under the ROC curve (AUC)=0.73, 95% CI (0.70, 0.77) in full dataset and in dataset derived from subjects selected for X-ray

Score category (min, max) n N (%) Passing X-ray thresholda Decision cut point Sensitivity Specificity LR(x,y)b LR+c LR–c PPVd NPVd
Algorithm developed assuming baseline X-rays were available for everyone
0 (28.4, 56.9) 556 11 (2%) (>=0) 100% 0% 0.10 1.00   3%  
 1 (56.9, 63.1) 556 50 (9%) (>=1) 99% 10% 0.48 1.10 0.10 3% 100%
 2 (63.1, 67.7) 555 83 (15%) (>=2) 94% 20% 0.37 1.18 0.29 4% 99%
 3 (67.7, 70.6) 557 96 (17%) (>=3) 90% 31% 0.48 1.30 0.32 4% 99%
 4 (70.6, 73.7) 556 128 (23%) (>=4) 85% 41% 0.81 1.44 0.36 4% 99%
 5 (73.7, 77.3) 556 180 (32%) (>=5) 77% 51% 0.81 1.58 0.45 5% 99%
 6 (77.3, 81.3) 556 205 (37%) (>=6) 69% 61% 0.64 1.77 0.51 5% 98%
 7 (81.3, 86.3) 556 326 (59%) (>=7) 63% 71% 1.37 2.17 0.52 6% 98%
 8 (86.3, 94.2) 556 441 (79%) (>=8) 49% 81% 1.31 2.59 0.63 7% 98%
 9 (94.2, 155.6) 557 440 (79%) (>=9) 36% 91% 3.98 3.97 0.70 11% 98%
     (>9) 0% 100%    1.00   97%
  1. aThreshold set to identify one case of previously unknown prevalent vertebral fracture for every five subjects X-rayed
  2. bLikelihood ratio comparing probability of finding a person with an incident fracture in that score interval versus probability of finding a person without an incident fracture in that same score interval
  3. cLR+, Positive likelihood ratio; LR, negative likelihood ratio
  4. dPPV, Positive predictive value; NPV, negative predictive value. Both calculated based on a 3% incidence of 1+ vertebral fracture within 3.8 years of follow-up

Rights and permissions

Reprints and Permissions

About this article

Cite this article

Kaptoge, S., Armbrecht, G., Felsenberg, D. et al. Whom to treat? The contribution of vertebral X-rays to risk-based algorithms for fracture prediction. Results from the European Prospective Osteoporosis Study. Osteoporos Int 17, 1369–1381 (2006). https://doi.org/10.1007/s00198-005-0067-9

Download citation

Keywords

  • Algorithm
  • Osteoporosis diagnosis
  • Osteoporosis treatment
  • Radiograph
  • Spine X-ray
  • Vertebral fracture