Dr. Schousboe has received research support from Eli Lilly & Company and consulting fees from Roche. Dr. Hochberg has received research support, has consulted for, or received honoraria from Amgen, Eli Lilly & Company, Merck, Novaratis Pharma, Roche Pharmaceutical, Zelos, Procter & Gamble, and GSK. Dr. Cummings has received research support, has consulted for, or received honoraria from Amgen, Eli Lilly & Company, Pfizer, Zelos, Procter & Gamble, and GSK. All other authors state that they have no conflicts of interest.
FRAX and Risk of Vertebral Fractures: The Fracture Intervention Trial†
Article first published online: 4 MAY 2009
Copyright © 2009 ASBMR
Journal of Bone and Mineral Research
Volume 24, Issue 11, pages 1793–1799, November 2009
How to Cite
Donaldson, M. G., Palermo, L., Schousboe, J. T., Ensrud, K. E., Hochberg, M. C. and Cummings, S. R. (2009), FRAX and Risk of Vertebral Fractures: The Fracture Intervention Trial. J Bone Miner Res, 24: 1793–1799. doi: 10.1359/jbmr.090511
Published online on May 4, 2009
- Issue published online: 4 DEC 2009
- Article first published online: 4 MAY 2009
- Manuscript Accepted: 1 MAY 2009
- Manuscript Revised: 18 FEB 2009
- Manuscript Received: 12 DEC 2008
- vertebral fracture;
- risk factors;
The validity of the WHO 10-yr probability of major osteoporotic fracture model (FRAX) for prediction of vertebral fracture has not been tested. We analyzed how well FRAX for major osteoporotic fractures, with and without femoral neck BMD (FN BMD), predicted the risk of vertebral fracture. We also compared the predictive validity of FRAX, FN BMD, and prevalent vertebral fracture detected by radiographs at baseline alone or in combination to predict future vertebral fracture. We analyzed data from the placebo groups of FIT (3.8-yr follow-up, n = 3221) with ORs and areas under receiver operating characteristics (ROC) curves (AUC). FRAX with and without FN BMD predicted incident radiographic vertebral fracture. The AUC was significantly greater for FRAX with FN BMD (AUC = 0.71) than FRAX without FN BMD (AUC = 0.68; p = 0.002). Prevalent vertebral fracture plus age and FN BMD (AUC = 0.76) predicted incident radiographic vertebral fracture as well as a combination of prevalent vertebral fracture and FRAX with FN BMD (AUC = 0.75; p = 0.76). However, baseline vertebral fracture status plus age and FN BMD (AUC = 0.76) predicted incident radiographic vertebral fracture significantly better than FRAX with FN BMD (AUC = 0.71; p = 0.0017). FRAX for major osteoporotic fractures (with and without FN BMD) predicts vertebral fracture. However, once FN BMD and age are known, the eight additional risk factors in FRAX do not significantly improve the prediction of vertebral fracture. A combination of baseline radiographic vertebral fracture, FN BMD, and age is the strongest predictor of future vertebral fracture.
Vertebral fractures are the most common type of fracture in older adults,(1) and prior studies suggest that only one fourth to one third of these fractures come to medical attention.(2,3) Vertebral fractures are associated with increased risk of back pain and functional limitation.(4) Women with existing radiographically detected vertebral fractures are four to five times more likely to suffer another vertebral fracture and are also at increased risk for hip fracture(5) and other nonspine fractures(5,6) compared with women without a vertebral fracture. Furthermore, radiographic vertebral fractures are associated with increased mortality.(7,8)
Approximately one half of all clinical fractures occur in persons without osteoporosis based on BMD.(9) Thus, there is a need to identify other risk factors that may help to identify persons at highest risk for fracture and therefore identify those who warrant further assessment and treatment. Prospective cohort studies have consistently shown that low BMD, age, and prevalent radiographic vertebral fracture are factors that significantly increase the risk for incident radiographic vertebral fracture.(10,11) However, other risk factors such as smoking, alcohol consumption, and family history of a fracture have been inconsistently associated with vertebral fracture.(10–12) To address the issue of identifying persons at high risk for hip fracture and other fractures associated with osteoporosis, the WHO developed two 10-yr probability of fracture models (FRAX): one for hip fracture and one for major osteoporotic fracture (hip, spine [clinical], wrist, or humerus).(13,14) FRAX uses nine clinical risk factors to estimate the 10-yr probability of fracture: age, sex, body mass index, parental history of hip fracture, exposure to systemic glucocorticoids, history of prior fragility fracture, current smoking, three or more units of alcohol per day, and the presence of secondary osteoporosis. The 10-yr probability of fracture using FRAX can be calculated with or without BMD at the femoral neck (FN). FRAX with and without BMD has been shown to predict hip and major osteoporotic fractures.(15)
The value of FRAX for prediction of radiographic vertebral fracture has not been tested. The primary aim of this study was to determine whether the 10-yr probability of a major osteoporotic fracture, estimated by FRAX with and without FN BMD, can predict radiographic vertebral fracture. We also aimed to compare the predictive validity of a model containing FRAX with that of models containing age, femoral neck BMD, spine BMD, self-reported fracture history, and prevalent radiographic vertebral fracture alone or in combination. We addressed these aims using data from the placebo groups of the Fracture Intervention Trial.(2,16)
MATERIALS AND METHODS
Details of the trial protocols and selection of participants for the Fracture Intervention Studies have been described previously.(2,16,17) Briefly, women 55–81 yr of age who had been postmenopausal for at least 2 yr and had low femoral neck BMD (BMD ≤ 0.68 g/cm2) were enrolled at 11 clinical centers in the United States. Women were divided in two study groups based on the presence or absence of an existing radiographic vertebral fracture. A woman was classified as having an existing vertebral fracture if any of the ratios of vertebral heights was >3 SD below the mean population norm for that vertebral level.(1,18) Women were followed up for incident clinical fractures and new radiographic vertebral fractures as defined by a combination of semiquantitative radiographic reading and morphometry. In this analysis, the outcome of interest was incident radiographic vertebral fractures and major osteoporotic fracture as defined by FRAX. Their ascertainment is described in detail below.
Women who had at least one prevalent radiographic vertebral fracture (n = 2027) were enrolled in the vertebral fracture arm (FIT-VF) and were randomly allocated to placebo (n = 1005) or to alendronate (n = 1022). Women in the FIT-VF arm were followed up for 36 mo. Women who did not have a prevalent radiographic vertebral fracture (n = 4432) were enrolled in the clinical fracture arm (FIT-CF) and were randomly allocated to placebo (n = 2218) or to alendronate (n = 2214). Women in the FIT-CF arm were followed up for 48 mo. To address the aims of this research question, we pooled participants from the placebo groups of the FIT-VF and FIT-CF study arms (n = 3223).
Assessment of risk factors
At baseline in both FIT studies, BMD was measured at the hip, posterior-anterior spine, and lateral spine on all participants by DXA (QDR-2000; Hologic, Waltham, MA, USA). Follow-up measurements were made each year.
Risk factors for fracture included in FRAX were measured at baseline including height, weight, and questionnaires to ascertain smoking status, maternal history of hip fractures, history of a prior fragility fracture, consumption of three or more units of alcohol per day, and presence of rheumatoid arthritis. FRAX for U.S. white women was calculated by the WHO guidelines (28 September 2006). Although FRAX estimates for both hip fracture and major osteoporotic fracture (hip, clinical spine, humerus, or wrist) were calculated, we used the FRAX estimate of major osteoporotic fracture in all analyses. Finally, clinical and morphometric vertebral fractures can be included in the “history of prior fragility fracture,” but as defined previously, the FRAX 10-yr estimate of major osteoporotic fracture only includes clinical spine fractures.
Lateral radiographs were obtained at baseline and 24 and 36 mo after randomization in the FIT-VF arm and at baseline and 48 mo in the FIT-CF arm. Vertebral morphometry was done with a translucent digitizer and cursor.(19) Six points defining the anterior, middle, and posterior heights were marked on each vertebra.(17)
A woman was classified as having a baseline radiographic vertebral fracture if any of the ratios of vertebral heights was >3 SD below the mean population norm for that vertebral level.(1,2,17,18) A new radiographic vertebral fracture was defined as a decrease of 20% and at least 4 mm in any vertebral height from the baseline radiograph to that taken at the end of the study. Each fracture was confirmed by a repeat digitization of the involved vertebral body. In addition, vertebrae classified as having a new fracture by morphometry but not judged to be fractured by the morphometry technicians were reviewed by the study radiologist, who either confirmed or disqualified the fracture. Vertebrae judged to be fractured by the morphometry technicians, but which could not be digitized because of technical problems or other disease processes, were also reviewed by the study radiologist. If he identified a new fracture, the vertebra was classified as fractured. All evaluations by the technicians and the radiologist used a semiquantitative method.(18) Technicians were trained and certified by the radiologist before the study. The technicians and the study radiologist remained unaware of treatment allocation during their assessments.
Major osteoporotic fracture ascertainment
A fracture was defined as one diagnosed by a physician. Self-reports of fractures were confirmed by written reports of radiographs or other tests. We excluded pathologic fractures. A major osteoporotic fracture was defined as a fracture of the hip, spine (clinical), wrist, or humerus. Clinical vertebral fractures were defined as those that came to medical attention and were reported to the clinical centers by the participants. A copy of the radiograph obtained by the patient's physician was sent to the coordinating center and compared with the baseline study radiograph. Because such clinical radiographs were not standardized for morphometry, an incident clinical vertebral fracture was defined by semiquantitative reading by the study radiologist.(2)
We compared nine models for the prediction of incident radiographic vertebral fracture. We initially established the association between age, femoral neck (FN) BMD, and a combination of the two with prevalent radiographic vertebral fracture. Next we tested the association of FRAX with and without FN BMD with and without prevalent radiographic vertebral fracture. Because history of prior fracture is the most common risk factor of the WHO model,(15) we tested the use of that single risk factor in place of FRAX. We examined the association between baseline radiographic vertebral fracture with and without FN BMD. Finally, we tested whether adding FRAX to baseline radiographic vertebral fracture status and FN BMD improved prediction. We used logistic regression to calculate the OR and 95% CI. We used the C-statistic and 95% CI to evaluate the discrimination of each model. The C-statistic estimates the area under the receiver operator characteristic (ROC) curve (AUC) and indicates the model's ability to distinguish those with and without new radiographic vertebral fractures. Finally, we compared the observed risk of new radiographic vertebral fracture in quartile 4 (highest risk) versus quartile 1 (lowest risk) of the predicted probabilities of each model. We used FN BMD in our primary analyses because this is the site specified in the WHO model. Data were analyzed using SAS v 9.1 (SAS Institute, Cary, NC, USA) and STATA 9.2 (StataCorp, College Station, TX, USA). The “roccomp” procedure in STATA was used to compare AUCs using a nonparametric approach described by DeLong et al.(20) We applied the same approach for the prediction of incident major osteoporotic fractures (hip, spine [clinical], wrist, or humerus).
Complete follow-up for incident radiographic vertebral fracture was available for 3043 (94%) women. All women provided at least 1 day of follow-up for incident major osteoporotic fracture. Participants were followed for a mean of 3.8 ± 0.8 (SD) yr. Of these, 223 (7.3%) had at least one new radiographic vertebral fracture identified during the follow-up period, and 253 (7.8%) experienced a major osteoporotic fracture. Few participants reported consuming three or more alcoholic beverages per day, and none reported exposure to systemic glucocorticoids at baseline, because this was an exclusion criterion for FIT (Table 1). Our results suggest that the observed risk of morphometric vertebral fracture (7.3%) is slightly lower than the predicted risk (11.3%) based on FRAX (adjusted to reflect an average 3.8 yr of follow-up compared with 10 yr). The FRAX estimates used in this analysis are from the first release and have been shown to overestimate fracture risk in the U.S. white female population. We know that FRAX estimates will be revised (estimates will decrease by ∼20–25%). This will not affect these analyses; however, this will likely show that FRAX is well calibrated to FIT.
Incident radiographic vertebral fracture was associated with age alone (OR of 1.74 per 1 SD increase in age) and FN BMD (OR of 1.81 per 1 SD decrease in BMD). AUCs for these models (age alone and FN BMD alone) were 0.65 and 0.66, respectively (Table 2), whereas a model with both age and FN BMD yielded an AUC of 0.71. For the model of age and FN BMD together, the risk of vertebral fracture was 14.1%/4 yr in the highest quartile of predicted risk from this model compared with 1.4%/4 yr in the lowest quartile (Fig. 1).
FRAX without FN BMD was associated with incident radiographic vertebral fracture (OR of 1.76 per 1 SD increase in fracture probability) and had an AUC of 0.68. The risk of vertebral fracture was 13.3%/4 yr in the highest quartile of predicted risk compared with 1.6%/4 yr in the lowest quartile (Fig. 1). FRAX with FN BMD was also associated with incident radiographic vertebral fracture (OR of 1.92 per 1 SD increase in fracture probability) and had an AUC of 0.71 (Table 2). The risk of vertebral fracture was 15.1%/4 yr in the highest quartile of predicted risk compared with 1.8%/4 yr in the lowest quartile (Fig. 1). The AUC was significantly greater for FRAX with FN BMD compared with FRAX without FN BMD (p = 0.002).
Age and self-reported history of fracture at baseline are both components of FRAX. Self-reported history of fracture was associated with incident vertebral fracture (OR = 2.30). A model with age alone, self-reported history of fracture at baseline alone, and a combination of the two for predicting incident radiographic vertebral fracture had AUCs of 0.65, 0.60, and 0.68. Participants in the highest quartile of predicted risk using age alone had an observed vertebral fracture risk of 12.1%/4 yr. Participants who had a self-reported history of fracture at baseline had a vertebral fracture risk of 10.6%/4 yr. Combining these two factors identified a similar high-risk group as FRAX without FN BMD (13.3%). FRAX without FN BMD, however, better identified those at lowest risk (1.6%/4 yr) compared with a model of age and self-reported fracture history. However, the AUCs were not statistically significantly different between the model with self-reported history fracture and age compared with FRAX without FN BMD (p = 0.92).
To assess the contribution of clinical risk factors to the risk of incident radiographic vertebral fracture, once FN BMD and age were known, we compared a model combining FN BMD and age to a model of FRAX with FN BMD. Models of FN BMD plus age and FRAX with FN BMD predict incident radiographic vertebral fracture to a similar degree (AUCs for both equal to 0.71). The risk of vertebral fracture in the highest and lowest quartiles of predicted risk is similar for both models (∼4.0%/4 yr and 1.4%/4 yr). The AUCs were not statistically significantly different between these models (p = 0.67).
We assessed the contribution of prevalent radiographic vertebral fracture status at baseline to the risk of incident radiographic vertebral fracture. A model combining baseline vertebral fracture, FN BMD, and age was more strongly predictive of incident radiographic vertebral fracture than the previously described models (AUC = 0.76). The risk of vertebral fracture in the highest and lowest quartiles of predicted risk is 16.9%/4 yr and 0.9%/4 yr, respectively. A model combining prevalent radiographic vertebral fracture status and FRAX with FN BMD did not improve discrimination or better identify those at highest and lowest risk. The AUCs were not statistically significantly different comparing the model with vertebral fracture status at baseline, FN BMD and age with a model of FRAX including FN BMD and vertebral fracture status at baseline (p = 0.76). However, the AUC for a model combining vertebral fracture status at baseline, FN BMD, and age was significantly greater compared with the model of FRAX including FN BMD (p = 0.0017).
Finally, the addition of vertebral fracture status at baseline did not improve prediction of major osteoporotic fracture. The range in AUCs was 0.64 (for models of age alone and, FRAX without FN BMD) to ∼0.70 (for models of FN BMD plus age, FRAX with FN BMD, FN BMD plus age and self-reported history of fracture, and FN BMD plus age and baseline vertebral fracture).
This is the first study to test and compare the WHO risk index, FRAX, for predicting radiographically detected vertebral fractures. Among women ≥55 yr of age with low bone mass, of whom 30% had a radiographically detected vertebral fracture at baseline, we showed that FRAX for major osteoporotic fracture with and without FN BMD can predict vertebral fracture. Although FRAX with FN BMD performed similarly to FRAX without FN BMD, the strongest predictor of future vertebral fracture was the presence of a vertebral fracture at baseline in addition to FN BMD and age.
In the absence of baseline vertebral fracture status, FRAX with FN BMD was as accurate in terms of discrimination and identification of those at highest and lowest risk compared with a model that combined FN BMD, self-reported fracture, and age. The latter model represents three of the nine components included in FRAX. Our results suggest that when FN BMD is not available that FRAX without FN BMD is as accurate in terms of discrimination and identifying those at highest and lowest risk of vertebral fracture as FN BMD and age. Comparing FRAX without FN BMD to age alone, self-reported fracture history alone, and a combination of age and self-reported fracture history suggests that FRAX without FN BMD is primarily driven by a combination of age and self-reported fracture history. Both of these models have similar discrimination and identify similar high- and low-risk groups. Finally, our results suggest that if FN BMD, age, and self-reported fracture history are known, that little is gained by adding the other components of FRAX.
Prior studies have shown that the presence of a vertebral fracture is a very strong predictor of future vertebral fracture.(5,11,21,22) Including vertebral fracture status at baseline seems to contribute significantly to identification of postmenopausal women at high risk of incident radiographic vertebral fracture. These results support including lateral spine imaging, either X-ray or by vertebral fracture assessment using DXA,(23) among postmenopausal women who have a reasonable likelihood of having a prevalent vertebral fracture, and incorporating this information in vertebral fracture risk assessments. Our results also suggest that, after including baseline vertebral fracture status, FN BMD and age, little is gained by including the remaining risk factors included in FRAX.
FRAX performed well for predicting radiographic vertebral fractures because it includes FN BMD and age. It is not clear that the other risk factors in FRAX further contribute to improving prediction of vertebral fractures because of the inconsistent association of alcohol,(11,12) BMI,(10,12) and smoking(10,12) with incident vertebral fracture. Consistent risk factors for vertebral fracture include age(10,12) and BMD.(10,11,24,25) In our study, models of age, FN BMD, and history of any fracture performed well.
As previously reported by Kanis et al.,(15) our data support the finding that FRAX with and without FN BMD predicts major osteoporotic fracture. Given that few strong consistent risk factors for these four fractures have been reported(26–28) it is not surprising that a model of FN BMD and age performs as well as models that incorporate other risk factors, including the presence of a vertebral fracture at baseline.
Although FRAX performed as well or nearly as well as models including FN BMD for predicting vertebral fracture, these results do not by themselves support selecting postmenopausal women for pharmacologic fracture prevention therapy without the use of BMD when densitometry is available. First, risk models incorporating BMD predict hip fracture better than models without BMD.(15) Second, if low BMD is associated with nonvertebral fracture reduction efficacy or response to oral bisphosphonates, using BMD (when available) in addition to clinical risk factors to select postmenopausal women for bisphosphonate fracture prevention therapy may be preferable to using clinical risk factors alone.
Our study has several limitations. First, the participants in this study were recruited based primarily on criteria involving low BMD and, for the vertebral fracture arm, were required to have a vertebral fracture. They also had to meet other entry criteria for the trial.(17) Therefore, they are not a random sample of the population and have higher prevalence of osteoporosis and fractures than the general population. The participants in this study were older white women and therefore these results may not be generalized to other populations. FRAX requires information on parental history of hip fracture; however, only maternal history of hip fracture was available in the FIT cohort. Data from the Study of Osteoporotic fractures suggests that 15% of women reported a history of parental history of hip fracture; 13.8% is caused by maternal history of hip fracture.(29) In FIT, 15% of women reported a maternal history of hip fracture. Therefore, whereas this might slightly underestimate the proportion of women reporting a parental history of hip fracture, this is likely to be small and would not influence the study results. Women using glucocorticoids, a risk factor in the WHO model, were excluded from entry into the trial. Although participants for both FIT trials were recruited separately, the same population was targeted, and all other criteria were the same. However, the overall prevalence of chronic oral glucocorticoid use in the general postmenopausal population is low (∼1–2%),(30) and hence we believe that addition of that risk factor would be quite unlikely to alter our results or conclusions. FRAX was developed to estimate the 10-yr probability of major osteoporotic fracture and hip fracture alone. Finally, our models are conditional on the estimates of the 10-yr probability of fracture developed by FRAX.
In conclusion, we showed that, in a cohort of women with low BMD of whom 30% had a radiographically detected vertebral fracture at baseline, FRAX with FN BMD can identify those at highest and lowest risk of fracture. Although FRAX with FN BMD more accurately identifies those at high and low risk compared with FRAX without FN BMD, a simpler model with the history of fracture and age performed as well as the full index. Vertebral fracture status at baseline in combination with FN BMD and age was the strongest predictor of future vertebral fracture. When baseline vertebral fracture status, FN BMD, and age are known, added risk factors do not improve predictive accuracy for vertebral fracture.
This study was funded by Merck.
- 13World Health Organisation 2008 FRAX WHO Fracture Risk Assessment Tool. World Health Organization, Geneva, Switzerland.
- 191995 Vertebral Fracture in Osteoporosis. Radiology Research and Educational Foundation, San Francisco, CA, USA., ,