Vertebral Fracture Assessment (VFA) With a Densitometer Predicts Future Fractures in Elderly Women Unselected for Osteoporosis


  • The authors state that they have no conflicts of interest.


Low radiation dose imaging of the lateral spine acquired with a bone densitometer for vertebral fracture assessment (VFA) has great potential for clinical use. We have undertaken an evaluation of VFA in a prospective population cohort of elderly women to examine the prevalence of vertebral fractures, their ability to predict incident fractures, and their use in targeting therapy. Women (n = 5157) ≥75 yr of age living in the general community in the United Kingdom underwent posteroanterior and lateral imaging of the spine (T4–L4) with a densitometer (Hologic QDR4500A) at entry to a randomized, double-blind, controlled trial of 800 mg oral clodronate (Bonefos) or matching placebo daily over 3 yr. The women were identified from general practice registers and recruited by letter of invitation regardless of skeletal status. The proportion of vertebrae interpretable varied from 98.2% at T12 to 57.1% at T4, with >92% interpretable at levels between T8 and L3. As judged by BMD at the total hip, 19.6% of the women had osteoporosis, and the prevalence of vertebral fracture was 14.5%. Women with one or more vertebral fractures had a relative risk (RR) for incident osteoporotic fractures of 2.01 (95% CI, 1.64–2.47). The RR for hip fractures was 2.29 (95% CI, 1.63–3.21). After adjustment for age, femoral neck BMD, weight, and treatment, the RR was 1.50 (95% CI, 1.21–1.86) for osteoporotic fractures, with similar results for hip fractures (RR, 1.41; 95% CI, 0.99–2.02). For women with two or more vertebral fractures, the adjusted RRs were 1.97 (95% CI, 1.24–2.72) and 1.86 (95% CI, 1.14–3.03) for osteoporotic and hip fractures, respectively. We conclude that VFA can frequently detect vertebral fractures in a population cohort of elderly women. These fractures, like radiographic fractures, predict future clinical fractures independent of age, weight, and BMD. Having multiple vertebral fractures was associated with greater risk of incident osteoporotic fractures and hip fractures.


Vertebral fractures are the most common osteoporotic fracture but are frequently undiagnosed, and only about one third come to clinical attention.math image However, both radiographically detected and clinical vertebral fractures are associated with significant future fracture risk,math image morbidity,math image and mortality,math image and some pharmacological agents have been shown to reduce future fracture risk in patients selected solely on the basis of radiographically detected vertebral fractures.math image Economic analyses with some of the available pharmacological treatments have shown that it is cost effective in many countries to treat older women with one or more radiographically detected vertebral fractures.math image These observations indicate that there are benefits to be had from improved detection of vertebral fracture.

Vertebral fracture assessment (VFA), which uses lateral spine images acquired with a bone densitometer, has great potential for clinical use in detecting vertebral fractures. VFA can be performed at the point of care at the same time and with the same equipment as used for BMD measurement, providing greater convenience for the patient. Additionally, VFA uses much less radiation than lateral radiographs.math image

Several studies have examined the concordance of VFA and lateral radiographs and found moderately good agreement.math image The aims of this analysis of a prospective, population-based study of elderly women recruited to a double-blind intervention study were to examine the prevalence of vertebral fractures by VFA, their ability to predict incident fractures, and their use in targeting therapy.


The study was a double-blind, prospective, randomized, placebo-controlled, single center study in community-dwelling women ≥75 yr of age. The aims of the study were to determine risk factors for fracture in elderly women in the United Kingdom and to examine the efficacy of the bisphosphonate, clodronate. Participants were recruited randomly from general practice lists in South Yorkshire and North Derbyshire by letter of invitation and did not need to have proven osteoporosis or any other known risk factors for fracture. The population and methods have been fully described previously.math image

Subjects for this analysis comprised 5212 women recruited to the main study population (367 were recruited to a pilot study). Follow-up data were available in 99.8% of the participants. The duration of the treatment period was 3 yr. Women completing the 3-yr study were followed under blinded conditions until the last patient had completed the treatment period. For the evaluation of safety, events were collected over the entire study period (median duration, 4.0 yr).

As expected in a volunteer population, a healthy participant bias was observed, in part related to the lack of recruitment of women from nursing homes or sheltered accommodation. The cohort had a mortality and hip fracture rate approximately one half that of the age-matched population.math image Similarly, the observed prevalence of osteoporosis at baseline determined by femoral neck BMD was 23.1% versus an expected prevalence of ∼35%.math image

Baseline assessments

All baseline assessments were carried out during a single visit to the WHO Collaborating Center for Metabolic Bone Diseases in Sheffield. All contacts thereafter were conducted in the community by a team of study nurses at 6-mo intervals to acquire data on fractures, adverse events, and hospitalizations and to collect and dispense study medication.

At the initial visit, each participant underwent a detailed and comprehensive assessment of their general health including self-reported health status measured with the EuroQOL visual analog scale (VAS)math image and fracture history. At study entry, participants had their BMD measured at the hip with a Hologic QDR 4500A with analysis software version 8. This same densitometer acquired paired posteroanterior and lateral VFA images of the thoracic and lumbar spine (T4–L4) in the supine position. Both the single-energy and the “fast” dual-energy acquisitions modes were acquired for the lateral images.math image

None of the results of the baseline assessments of fracture risk, including BMD values or vertebral fracture assessment results, were communicated to the participants.

Prevalent vertebral fracture assessment

After a brief 2-day training program using sample images of normal and abnormal VFA images, the technicians performing the VFA scans initially classified the scans as “normal” or no fracture (63%); “equivocal,” indicating a possible fracture (23%); or “abnormal,” indicating a probable fracture (14%). In all equivocal and abnormal scans, the technician performed quantitative vertebral morphometry by placing six points on each of the assessable vertebrae. These six points marked the anterior, middle, and posterior locations of the inferior and superior endplates in a manner described by Rea et al.math image Additionally, quantitative vertebral morphometry was undertaken in a randomly selected subset of “normal” scans.

For the current analysis, vertebral height ratios measured on the VFA images from this study population were used to derive normative data using the previously described trimming algorithm of Melton et al.math image Vertebrae were classified as either fractured or nonfractured based on the McCloskey algorithm.math image The method identifies a fracture if two criteria are fulfilled. First, there must be a 3 SD reduction in one of the ratios of measured vertebral heights (anterior-posterior, middle-posterior, or posterior-adjacent posterior) compared with a reference database. Additionally, there must be a 3 SD reduction in a similar ratio calculated using the “predicted posterior height” instead of the measured posterior height. The predicted posterior height is calculated from the patient's adjacent vertebrae. The McCloskey algorithm has been extensively validated with radiographs but has not previously been used with VFA. Prevalent vertebral fractures detected by the McCloskey algorithm predict future fracture riskmath image and are associated with increased mortality.math image

All scans with quantitative morphometry, including a random selection of 210 of the scans deemed to be normal by the scanning technicians, were visually assessed using softcopy by a single experienced reader (EMC) for correct point placement, suitability of the vertebrae for evaluation, and to exclude deformities not caused by fracture. This visual assessment provided final adjudication for vertebral fractures.

Incident fractures

The collection and validation of incident fractures in this study are reported elsewhere.math image The accuracy of fracture reporting was >98%, with none of the incident hip fractures being missed.math image Verified fractures that had presented symptomatically and triggered radiological study were included in the statistical analyses and defined as “clinical fractures.” Of these clinical fractures, further categories were derived including osteoporotic fractures (excluding fractures of the skull, nose, face, hand, finger, feet, toe, ankle, patella) and hip fractures alone.

Study treatment

After randomization, the women received either clodronate 800 mg daily (two BONEFOS 400-mg tablets once daily or one tablet twice daily) or an identical placebo. The intervention was continued for 3 yr. Concomitant supplementation with calcium and vitamin D was not given.

Statistical analysis

The statistical analyses presented in this paper were conducted independently from the study sponsors by one of the authors (EMC). The analyses of baseline risk factors and incident fractures were conducted by logistic regression and Cox regression, with both analyses giving very similar results. The results presented here were derived from logistic regression. The relationship between baseline vertebral fracture and incident fractures was expressed as a relative risk (RR) and were also adjusted for the covariates; baseline age, femoral neck BMD, and weight. Previous analysis showed a significant effect of clodronate on osteoporotic fracture incidence but not hip fracture incidence. The analysis of the treatment efficacy of clodronate on osteoporotic fracture incidence was undertaken by Cox regression modeling that included treatment as a factor, and baseline age, femoral neck BMD, and weight as covariates in addition to prevalent vertebral fracture. All analyses were undertaken on an intent-to-treat basis so that all fractures were included in the analyses regardless of whether the women were taking study medication or not.


Of the 5212 women included in the baseline assessment, VFA was undertaken in 5191 (99.6%). Visual assessment of these 5191 images by the scanning technicians classified 63% of the scans as normal, 23% as equivocal, and 14% as abnormal (Fig. 1). After quantitative morphometry and visual assessment by an experienced reader (EMC), a further 35 scans (30 in the equivocal group) were determined to be not evaluable because of poor image quality (Fig. 2) Therefore, scans were interpretable in 5156 of the total cohort (98.9%).

Figure FIG. 1..

Study flowchart.

Figure FIG. 2..

(A) Classification pathway for abnormal scans. (b) Classification pathway for equivocal scans. (c) Classification pathway for normal scans.

The percentage of vertebrae suitable for assessment varied from 98.2% at T12 to 57.1% at T4, with >92% of scans interpretable at between T8 and L3 (Fig. 3). Five or more vertebrae were assessable in 99.8% and 10 or more in 93.9% of the 5157 women. Fractures were most common at T12 and L1 and the least commonly fractured vertebra was T4. This result was similar when the analysis was restricted to those women (53%) where all 13 vertebrae were interpretable (data not shown).

Figure FIG. 3..

Percentage of vertebrae interpretable and percentage of fractures by vertebral level.

In those women in whom the scans underwent formal morphometry (i.e., excluding those women with scans deemed to be normal in whom formal morphometry was not undertaken), the fracture prevalence in women with 10 or more vertebrae interpretable was not statistically significantly different from the prevalence in the 15% of women where there were <10 vertebrae interpretable (30.8% versus 36.5%, respectively; p = 0.06).

The observed prevalence of women with one or more vertebral fractures detected by VFA using this study design was 14.5%, with 4.3% having two or more vertebral fractures. Subjects with one or more prevalent vertebral fractures had statistically significant differences in baseline characteristics compared with those without a prevalent vertebral fracture. Specifically, they were older, had decreased standing height, lower weight, lower hip BMD, and decreased self-reported health status (Table 1). These differences were observed in those with only one prevalent vertebral fracture and were even more marked in those with two or more (Table 1).

Table Table 1.. Baseline Characteristics of Study Participants With and Without Prevalent Vertebral Fractures
original image

Prediction of future fracture risk by VFA

During a median of 4 yr of follow-up, 744 women sustained a new clinical fracture. Of these, 620 sustained one or more osteoporotic fractures including 180 women who sustained an incident low-trauma hip fracture. Those with one or more prevalent vertebral fractures at baseline had significantly increased RR of incident fracture (Table 2). The RRs for all clinical fractures, all osteoporotic fractures, and hip fracture were 1.84, 2.01, and 2.29, respectively. Apart from the prediction of hip fracture by a single vertebral fracture, all of the RRs remained statistically significant after adjustment for age, weight, and treatment effect (Table 2). The RR was somewhat attenuated after further adjustment for femoral neck BMD but remained statistically significant for all clinical fractures (RR = 1.42) and osteoporotic fractures (RR = 1.50) and of borderline significance for hip fracture (Table 2).

Table Table 2.. RR of Fracture for Those With One or More Prevalent Vertebral Fractures Detected by VFA Compared With Those Without a Prevalent Vertebral Fracture
original image

The RR was greater in those with two or more prevalent vertebral fractures compared with those without fracture for all outcome fracture categories, both before and after adjustment. As in the case of women with one or more vertebral fracture, RRs were lower when adjusted for age, femoral neck BMD, weight, and treatment effect but remained significantly higher than unity.

An increased mortality risk was observed in those with a prevalent vertebral fracture (RR, 1.43; 95% CI, 1.00–2.05), but this effect no longer remained significant after adjustment for age, femoral neck BMD, weight, and treatment (RR, 1.09; 95% CI, 0.75–1.58).

Efficacy of clodronate on the risk of any osteoporotic fracture

During the 3 yr of the treatment phase, women in the clodronate group had a reduced risk of an incident osteoporotic fracture compared with women given placebo (RR, 0.76; 95% CI, 0.62, 0.92). This effect remained robust after adjustment for baseline age, femoral neck BMD, weight, and prevalent vertebral fracture (RR, 0.77; 95% CI, 0.61–0.96). The efficacy of the bisphosphonate to reduce fracture incidence was similar in those with and without prevalent vertebral fracture with no significant interaction between prevalent vertebral fracture and treatment efficacy (p = 0.89; Fig. 4) in the multivariate model. Thus, the number needed to treat (NNT) to prevent one women having an incident osteoporotic fracture during 3 yr decreased with increasing number of prevalent vertebral fractures from 59 in women without prevalent vertebral fracture to 22 in those with at least one vertebral fracture, and 16 in those with two or more vertebral fractures (Fig. 4).

Figure FIG. 4..

Incidence of osteoporotic fractures for the placebo and clodronate-treated groups with respect to baseline vertebral fracture status.


The significance of vertebral fractures in predicting future fracture risk is well established. This is the first, large-scale, prospective study to measure the prevalence of vertebral fractures detected by VFA in an unselected population and to assess the number of scans that were interpretable. Nearly 99% of the participants had VFA scans of adequate quality, and >92% of the vertebrae between T8 and L3 were assessable. At the vertebrae where fractures were most common (T12 and L1), the percentage of vertebrae assessable was 98.2% and 97.6%, respectively, similar to the results reported by Rea et al.math image in a smaller, selected population, but measured with the same equipment.

There has been much discussion about the definition of vertebral fracture, with early studies usually showing a relatively high prevalence, largely caused by the use of high-sensitivity but low-specificity techniques. Whereas the decision about prevalent vertebral fractures in this study was largely caused by visual assessment, the fractures also largely fulfilled a well-established morphometric definition of vertebral fracture, designed to maintain specificity but to improve specificity. This method used differs from that currently recommended for clinical reading of VFA scans, which is a semiquantitative visual assessment by a trained physician or radiologist.math image However, the McCloskey method is well validated in large cohorts and has been compared with other methods, both visual and morphometric.math image Moreover, in this study, trained technicians first did a visual assessment, and after morphometric assessment, the scans were visually assessed and adjudicated by an experienced physician, thus in some ways replicating current clinical practice for scans that were labeled as equivocal or abnormal by the technician. Further support that the vertebral fractures detected by VFA in this study are of clinical relevance is their association with the typical clinical correlates of radiographically detected prevalent vertebral fractures such as increased age, decreased height, decreased BMD, decreased health status, and increased mortality. These clinical features were more marked in women with more than one vertebral fracture as was the fracture risk and adds support to the definition and relevance of the prevalent vertebral fractures detected by VFA.

The vertebral fracture prevalence at this age in the general population would be expected to be [sim]25%, and the observed prevalence in this study was 42% lower. However, this is entirely consistent with the observed healthy participant bias, which was evident in that the prevalence of osteoporosis by total hip BMD was 35% lower and mortality was 50% lower than the population estimate. However, two other possibilities should be kept in consideration for the lower observed prevalence of vertebral fracture. Morphometric and expert review was undertaken in only a subset of the normal group, and the proportion of subjects with prevalent vertebral fractures in this subpopulation was 3.3%. If all scans classified as normal by visual assessment of the scanning technician had been analyzed more fully, the overall detected prevalence of vertebral fracture would have been ∼2% higher (16.5%) if all images had undergone morphometric analysis and physician review. Additionally, it may be that some vertebral fractures that are detected by radiographs are undetected by VFA, resulting in a lower reported prevalence. Previous studies comparing VFA to radiographs indicate that mild radiographic vertebral fractures are less well visualized by VFA and may be missed.math image

Our study is the first to show that prevalent vertebral fractures detected with VFA are significant predictors of future clinical fractures. The magnitude of the excess risk was in agreement with a previous meta-analysis of prevalent clinical and morphometric vertebral fractures,math image where prevalent vertebral fractures were associated with RRs of 1.9 (95% CI, 1.7, 2.3) and 2.3 (95% CI, 2.0, 2.8) for all clinical fractures and hip fractures, respectively.math image As previously observed to be the case with radiographic vertebral fractures,math image two or more fractures were associated with an even greater risk of future fractures. A significant portion of excess future fracture risk remained after adjustment for age, femoral neck BMD, weight, and treatment effect. The presence or absence of vertebral fracture did not have an observable effect on treatment efficacy. However, those with a vertebral fracture on VFA are at higher risk of fracture and thus the NNT to prevent a future fracture is much lower in this subgroup.

This study has several limitations. The results may not be directly applicable to other types of bone densitometers with different imaging technology. Current users of Hologic densitometers typically rely in clinical practice on only a single-energy lateral scan. However, Rea et al.math image directly compared the lateral single-energy and “fast” dual-energy scans used in this study and found that, at every vertebral level, the single-energy scan had more interpretable vertebrae and better long-term precision for point placement than the “fast” dual-energy scan.

An additional limitation is that there were no spinal radiographs for computation of sensitivity and specificity. Finally, this cohort was elderly women, and these results may not be applicable in men or younger women.

In conclusion, VFA commonly detects vertebral fractures in a population cohort of elderly women. VFA detected fractures have similar characteristics in terms of clinical correlates and fracture prediction as radiographically detected vertebral fractures. The relative risk of incident clinical fractures for those with VFA detected vertebral fractures remains significant after adjustment for age, femoral neck BMD, weight, and treatment effect. Multiple vertebral fractures were associated with greater clinical fracture risk and also predicted hip fractures after adjustment for age, femoral neck BMD, weight, and treatment effect. VFA is a more convenient and lower radiation dose alternative to radiographs for incident fracture risk prediction.


The authors acknowledge the help of Kevin Wilson, PhD (Hologic) in the preparation of the manuscript and Linda Kersh for input to technician training for reading of the VFA scans and morphometric point placements. The original study was funded by the Medical Research Council in the UK and Bayer Schering Pharma Oy, Helsinki, Finland.