Osteoporosis International

, Volume 23, Issue 1, pp 53–58

Predictors of new and severe vertebral fractures: results from the HORIZON Pivotal Fracture Trial


  • R. Wustrack
    • Department of Epidemiology and BiostatisticsUniversity of California, San Francisco
  • E. Seeman
    • Austin HealthUniversity of Melbourne
  • C. Bucci-Rechtweg
    • Novartis Pharmaceuticals Corp
  • S. Burch
    • Department of Epidemiology and BiostatisticsUniversity of California, San Francisco
  • L. Palermo
    • Department of Epidemiology and BiostatisticsUniversity of California, San Francisco
    • Department of Epidemiology and BiostatisticsUniversity of California, San Francisco
Original Article

DOI: 10.1007/s00198-011-1664-4

Cite this article as:
Wustrack, R., Seeman, E., Bucci-Rechtweg, C. et al. Osteoporos Int (2012) 23: 53. doi:10.1007/s00198-011-1664-4



We examined prevalent and recent vertebral fractures in 1 year as predictors of new vertebral fractures over subsequent 2 years using data from RCT placebo patients. We found that prevalent and recent vertebral fractures strongly and independently predicted subsequent vertebral fractures including those which were severe.


While several studies have shown that prevalent vertebral fractures (pVFx) increase the risk of new vertebral fractures (VFx), the impact of recent vertebral fractures on future fractures is less studied.


Data from the placebo arm of the HORIZON Pivotal Fracture Trial, an international trial of zoledronic acid in postmenopausal, osteoporotic women between 65 and 85 years, were used. We included the subset of 2677 women with annual spinal radiographs to study the impact of vertebral fractures in year 1 (Y1 VF) on those occurring in years 2 and 3 using morphometric and semiquantitative (SQ) criteria. In addition, a subset of severe VFx was defined using SQ criteria. Logistic regression examined the impact of pVFx and Y1 VF on all incident VFx and on severe incident VFx.


Two hundred fourty-five (9.1%) women sustained a new VFx in years 2–3. VFx risk in years 2–3 was 3.9% in those without pVFx or VFy1 and 29.8% in those with both risk factors. Both pVF and VFy1 remained independent predictors for future VF when they were both entered into a logistic regression model (odds ratio (OR) = 3.3; 95% confidence interval (CI), 2.3–4.7; OR = 3.7, 95% CI, 2.3, 5.8, respectively). ORs were similar after adjustment. Of the total number of women, 4.1% had severe VFx. PVFx and Y1 VF were also significant predictors of severe VFx; however, Y1 VF appeared more strongly predictive of severe VFx.


Prevalent and incident vertebral fractures are highly predictive of subsequent new and severe vertebral fractures. Women with both of these risk factors are likely to benefit from anti-osteoporosis treatment.


FractureOsteoporosisPredictorsVertebral fracture


Fragility fractures are a growing public health concern currently affecting over 44 million people in the USA and impacting the US health care system by incurring a direct cost of US $19 billion in 2005 [1] (www.nof.org, accessed August 13, 2010). While hip fractures are the most devastating consequences of osteoporosis with 20–30% mortality within the first year [2], vertebral compression fractures can negatively impact health, overall function, and quality of life (www.nof.org, August 13, 2010). In addition, there is evidence to suggest that vertebral fractures have a greater overall disutility [3] compared to non-vertebral fragility fractures. Only a third of vertebral fractures come to clinical attention; however, both silent and clinical vertebral fractures can adversely affect quality of life [36]. Additionally, the presence of either a silent or clinical vertebral fracture increases the risk of sustaining subsequent vertebral compression fractures and non-vertebral fragility fractures [3, 713].

While many studies have examined risk factors for hip fractures [7, 14, 15], fewer studies have focused on the predictors of vertebral fractures. Age and low BMD have been established as risk factors for incident vertebral fractures [8], and several authors have pointed out the significance of a prior vertebral fracture in predicting a future vertebral fracture as well as non-vertebral and hip fractures [2, 8, 11, 12]. Previous studies have shown that both mild and severe prevalent vertebral fractures predict future fractures [12, 16], and that a recent vertebral fracture can increase a future fracture within 2 years [12, 16]. In addition, a spinal deformity index, incorporating overall severity of all prevalent vertebral fractures, has been shown to be a good predictor of incident vertebral fractures [17, 18]. However, these studies have not demonstrated whether a baseline or incident vertebral fracture is a greater risk factor for future fractures.

While the severity of a prevalent vertebral fracture has been shown to predict future fractures [9, 19], little has been done to determine predictors of severe incident vertebral fractures. This is relevant because the severity of a vertebral fracture not only predicts future fractures but can also cause a greater reduction in quality of life and health [6].

We hypothesized that a prevalent and an incident vertebral fracture interacted to increase the risk for sustaining a new vertebral fracture and increased the risk for sustaining more severe vertebral deformities in women with osteoporosis. Thus, the specific aims of this analysis were (1) to identify predictors of new vertebral fractures, (2) to define predictors of severe vertebral fractures, and (3) to determine how these established risk factors interact to predict future and severe fractures.

Materials and methods

We performed a secondary analysis among women in the placebo arm of the Health Outcomes and Reduced Incidence with Zoledronic Acid Once Yearly (HORIZON) Pivotal Fracture Trial, a 3-year international trial. The study design of the HORIZON Pivotal Fracture Trial has been described previously [20]. Over 7,000 postmenopausal women from 11 international centers were randomized to placebo or once-yearly zoledronic acid 5 mg and followed for 3 years. All patients enrolled in this trial had osteoporosis as defined by a T-score of ≤2.5 measured at the hip or a T-score of ≤1.5 and a history of two mild or one moderate vertebral fracture. Patients were stratified into one of two strata based on baseline osteoporosis medication use. Stratum 1 included patients who were not taking any osteoporosis medications at time of randomization or met a defined washout criterion. Patients assigned to stratum 2 were taking or had recently been taking allowed osteoporosis mediations at the time of randomization. Standardized lateral spine radiographs were obtained at baseline and at 12, 24, and 36 months in all patients in stratum 1 and at baseline and 36 months only for those in stratum 2. Since we were interested in new fractures in the last 2 years of the trial, only the patients in stratum 1 with annual radiographs were included in this analysis. The analysis was limited to those patients randomized to placebo.

Definition of vertebral fractures

Prevalent vertebral fractures at study baseline

Expert readers in a blinded fashion read all spinal lateral radiographs. Each vertebra from T4–L4 was included in quantitative morphometric analysis by an expert reader at a central imaging laboratory (Synarc). A prevalent vertebral fracture, i.e., a vertebral fracture at baseline, was defined if any one of the ratios of morphometric vertebral heights was at least 3 SD below the vertebra-specific mean height ratio for that level [20, 21]. Radiographs with a moderate or severe fracture by semiquantitative (SQ) evaluation, but not meeting morphometric criteria, required confirmation as fractured by a senior radiologist using SQ methods [21, 22].

Incident vertebral fractures

An incident vertebral fracture was defined as a new vertebral fracture sustained between serial radiographs. It was defined as a loss of vertebral height of at least 20% and 4 mm from the previous radiograph by quantitative morphometry. Any fractures identified by quantitative morphometry were then confirmed using SQ analysis: if the vertebral body also had an increase of one or more severity grade, it was considered an incident vertebral fracture [23].

Severe vertebral fractures

Severe incident vertebral fractures were defined based on the SQ system only and defined over the final 2 years of the study. Each vertebral fracture was given an SQ grade ranging from 0 = normal to 3 = severe [21]. A grade change was then defined by comparing each follow-up radiograph to the prior radiograph. Severe incident vertebral fractures in an individual were defined using two different algorithms and analyzed separately for each. Firstly, we calculated the sum of the SQ changes over the 2 years across the 13 vertebrae. A woman with a severe vertebral fracture was defined in whom the sum was 3 or greater. Secondly, a severe incident fracture was defined by the largest SQ change in any one vertebra. For this definition, a woman with a severe fracture was one in whom the maximum grade change was 3.

Statistical analysis

Patients who had a baseline radiograph, a radiograph at 12 months, and a final radiograph at 24 or 36 months were included in the analysis. The primary outcome for this analysis was a new incident vertebral fracture in the last 2 years of the trial. The absolute risk of vertebral fractures in years 2–3 by baseline and year 1 fracture status was determined. Then, several prediction models were created. The primary predictors were baseline prevalent vertebral fracture and year 1 incident vertebral fracture. We tested whether or not these predictors were independent of each other using univariate logistic models and then using models with both variables included. Previously identified predictors of incident vertebral fractures were then included in a multivariable adjusted model together with an incident vertebral fracture in year 1 (Y1 VF) and a baseline, prevalence vertebral fracture (BL VF). The odds ratios (ORs) and 95% confidence intervals (95% CI) were reported for all variables. In a second analysis, severe fractures, both an SQ grade change of 3 and summed SQ grade change of ≥3, were considered as the primary outcomes in two parallel analyses. Similar models were created to determine significant predictors of sustaining a severe vertebral fracture.


Of the 3,039 patients enrolled in stratum 1 and randomized to the placebo group, 2,677 had available radiographs (88%) at baseline, year 1, and at least one radiograph in years 2 or 3. Table 1 shows the baseline characteristics of these patients; 63.5% had at least one vertebral fracture at baseline, and 4.0% had an incident vertebral fracture in year 1 (Table 2).
Table 1

Baseline characteristics and vertebral fractures in year for stratum 1; placebo participants included in this analysis

Stratum 1, placebo


Average age


Femoral neck T-score


24 (0.9%)

>−2.5 to −1.5

726 (27.2%)


1,918 (71.9%)

Femoral neck BMD


Prevalent vertebral fracture at baseline (pVFx)


978 (36.5%)


766 (28.6%)


933 (34.9%)

Incident vertebral fracture in year 1


2,571 (96%)


106 (4.0%)

Table 2

Incident vertebral fractures in years 2–3 (n = 2,677)

Any incident fracture


2,432 (90.9%)


245 (9.1%)

Severe incident fracture (based on sum semiquantitative grades ≥3)


2,567 (95.9%)


110 (4.1%)

Severe incident fracture (based on one vertebra with semiquantitative change of ≥3)


2,586 (96.7%)


91 (3.3%)

Among the 2,677 subjects, 245 (9.2%) sustained a new vertebral fracture (VF) in the second or third year of the trial (Table 2). Of the 245 women who sustained a new VF in the second or third year of the trial, 79 had an incident fracture in the first year, and 59 of those also had a prevalent fracture at baseline. The risk of sustaining a VF in the second or third year of the trial in participants without a baseline VF or a new VF in the first year was 3.9% (Fig. 1). The risk was 11.2% in women with a baseline VF only and 9.1% in those women who sustained a new VF in the first year, but no baseline VF. The risk in patients with both a baseline VF and a new VF in the first year was 29.8%, nearly eight times the risk among those with neither risk factor (Fig. 1).
Fig. 1

Incidence of new vertebral fracture in the last 2 years of the study stratified by the risk factors: baseline vertebral fracture and incident vertebral fracture in year 1 of the study

Both a baseline VF and a new VF in the first year remained as strong and independent predictors for future VFs when they were each entered into a univariate logistic regression model (OR, 3.3; 95% CI, 2.3–4.7 and OR, 3.7; 95% CI, 2.3–5.8, respectively). These risks remained significant and were only slightly attenuated after adjustment for each other and for other confounding variables including age, BMI, height, smoking status, and BMD (Table 3).
Table 3

Baseline and new vertebral fractures (year 1) as predictors of new vertebral fracture over subsequent 2 years


Odds ratio and 95% CI for logistic regression models


Unadjusted univariate

Unadjusted (both in model)

Adjusted for other risk factorsa

Any new VF

Baseline vertebral fracture

3.3 (2.3–4.7)

3.2 (2.2–4.5)

2.8 (1.9–4.0)

New vertebral fracture year 1

3.7 (2.3–5.8)

3.3 (2.0–5.2)

3.1 (1.9–5.0)

Severe fractures defined as sum of SQ grades ≥3

Baseline vertebral fracture

2.2 (1.4–3.6)

2.1 (1.3–3.4)

1.8 (1.1–2.8)

New vertebral fracture year 1

3.6 (1.9–6.6)

3.2 (1.7–6.0)

3.0 (1.6–5.8)

Severe fractures: at least one vertebra SQ change ≥3

Baseline vertebral fracture

2.1 (1.3–3.5)

2.0 (1.2–3.3)

1.7 (1.0–2.8)

New vertebral fracture year 1

3.2 (1.6–6.4)

2.9 (1.5–5.8)

2.7 (1.3–5.4)

aAge, BMI, height, smoking, region, BMD. Age, smoking, BMI, not significant. Region, BMD and height, p < 0.05

*p < 0.5, p < .01

Predictors of severe vertebral fractures

Of the 2,677 patients with radiographs, 110 individuals (4.1% of the total 2677 or 44.9% of those with ≥1 VFx in year 2/3) had a summed SQ grade change of ≥3 while 28 women (1% of total, 11% of those with ≥1 Vfx) had a summed grade change ≥5 (Fig. 2). Two patients had a sum grade change of 10. Ninety-one (3.3%) women had a single vertebral fracture with a maximum SQ grade change of 3 in Y2/3 (Table 2, Fig. 3).
Fig. 2

Sum of semiquantitative (SQ) grade change of incident vertebral fractures during years 2 and 3 of the study among 245 women with ≥1 VFx in years 2 and 3. The sum of the SQ grades is calculated as the sum of SQ changes across all 13 vertebrae assessed

Fig. 3

Maximum semiquantitative grade change of incident fractures during years 2 or 3 of the study among 245 women with ≥1 vertebral fracture in years 2 and 3. The maximum SQ grade change is the largest change seen in any one vertebra

Both prevalent and an incident VF in year 1 were independent predictors of sustaining a severe VF (maximum SQ grade change of 3; OR, 2.1; 95% CI, 1.3–3.5 and 3.2; 95% CI, 1.6–6.4, respectively; Table 3). When entered into a multivariate adjusted model with age, BMD, BMI, height, and smoking as other predictors, new VF Y1 remained statistically significant, while a BL VF was associated (OR, 2.7; 95% CI, 1.3–5.4 and 1.7; 95% CI, 1.0–2.8, respectively). When a sum SQ grade change of ≥3 was used to define severe vertebral fractures, a new VF Y1 continued to be a significant predictor with an OR of 3.0 (95% CI, 1.6–5.8). A prevalent vertebral fracture also was a significant predictor with an OR of 1.8 (95% CI, 1.1–2.8). Both new Y1 VF and BL VF remained significant for predicting severe VF when entered into a multivariate adjusted model (Table 3).


We report that among patients with previous fractures, already at very high risk of sustaining future vertebral fractures, a new vertebral fracture further amplifies that risk by threefold. Thus, combining the risk factors of an existing plus a new vertebral fracture creates a high-risk gradient: the risk of sustaining another vertebral fracture within 2 years with both risk factors was almost eightfold higher compared to women with neither risk factor (29.8% versus 3.9%). The one previous study of this question reported the period of high risk to extend to 1 year beyond the new fractures [12]. We confirm the increased risk followed a recent VF and extend this work showing that this increased risk continues for at least 2 years and also applies to incident severe fractures.

Our results are consistent with several previous studies confirming that women with an existing, prevalent vertebral fracture are at very high risk of new vertebral fracture. We report prevalent vertebral fractures carried an adjusted OR of 2.8 (95% CI, 1.9–4.0) in predicting new fractures, which agrees with several studies that have cited relative risks in the range of 2.5–3.7 [8, 9, 11, 12].

Lindsay et al. examined the impact of recent vertebral fracture on future vertebral fracture risk and reported a relative risk over 9 (9.3; 95% CI, 1.2–71.6, p = 0.03) [12]. Roux et al. studied the ability of mild prevalent vertebral fractures to predict incident fractures. They showed that mild prevalent vertebral fractures do predict new vertebral fractures (relative risk (RR) 1.8, 1.3–2.4, p < 0.001), but not as much as severe fractures (RR 2.7, 2.3–3.3, p < 0.001) [16]. Our results are more similar to Roux compared to the high relative risk shown by Lindsay. This could be due to a number of differences between studies including the selection of patients, adjustment for covariates, and the length of follow-up (2 years in our study, 1 year in Lindsay’s study, and 4 years in Roux’s study). The two studies also used different definitions of incident fracture: we used a definition of a decrease in vertebral height by 20% and 4 mm and then confirmed the presence of a new fracture using a change of at least 1 grade using the semiquantitative grading scale. Lindsay and colleagues defined an incident fracture as a 15% height vertebral decrease in the previous radiograph, and Roux used a change of at least 1 using semiquantitative analysis alone. The study by Lindsay et al. is small, and the relative risk has a wide confidence internal (1.2 to 71.6) which is consistent with our results. Most importantly, all three studies agree qualitatively in showing that patients with recent vertebral fractures are at extraordinarily high risk of new vertebral fractures in the next 1 to 2 years, independently of previous vertebral fractures.

This study extends the results previously reported and is the first to provide data focused specifically on prediction of severe incident fractures. We found that regardless of which definition of severity we used for incident vertebral fractures, a recent vertebral fracture was a strong predictor of an incident severe fracture. Two studies have examined the severity of prevalent vertebral fractures to predict any incident vertebral fracture (including mild, moderate, and severe) and found that severe prevalent fractures predicted all incident fractures [9, 19]. Minne, Crans and Kerkeni proposed that a spinal deformity index (SDI) be used to summarize in a single numerical value the number and severity of vertebral fractures from spine radiographs obtained at a single point in time. In two different publications, they found that the SDI was a good predictor of incident vertebral fractures [17, 18, 24]. However, our study is the first to use SDI in a longitudinal way to define severity of incident vertebral fractures and to examine predictors of severe incident fractures. Since severe incident fractures are associated with worse back pain, worse quality of life, and higher mortality than mild vertebral fractures [9, 20, 25], it is important to define those at risk of more severe future fractures.

Despite the known ability of anti-osteoporosis therapies to greatly reduce future vertebral fracture risk [20, 2631], many studies report that patients are not being diagnosed as osteoporotic after sustaining a vertebral fracture and are left untreated [1, 2, 3234]. Our results show that patients who sustain one fracture are much more likely to sustain multiple future fractures within a short time period, and that both prevalent vertebral fractures and recent incident vertebral fractures are strong and independent predictors of any incident vertebral fractures as well as severe incident vertebral fractures. For example, the combined risk of sustaining a new fracture within 2 years is nearly 30% among patients who have both a history of a vertebral fracture as well as a recent vertebral fracture. Our study allows clinicians to identify a group of particularly high-risk women for whom treatment is particularly urgent. We hope that clinicians will recognize the extremely high risk for these patients and ensure that they are properly treated before they sustain multiple and more severe vertebral compression fractures.


The HORIZON-PFT was funded by Novartis Pharmaceuticals. Funding for this analysis of vertebral fracture prediction was provided by Medtronic.

Conflicts of interest

Dennis Black has research support from Novartis Pharmaceuticals.

Copyright information

© International Osteoporosis Foundation and National Osteoporosis Foundation 2011