Incident vertebral fractures and mortality in older women: a prospective study
- First Online:
- Cite this article as:
- Kado, D.M., Duong, T., Stone, K.L. et al. Osteoporos Int (2003) 14: 589. doi:10.1007/s00198-003-1412-5
- 226 Views
Older persons who have prevalent vertebral fractures have an increased risk of mortality. It is not known whether incident vertebral fractures are also associated with an increased risk of mortality. To determine whether older women with incident vertebral fractures have an increased risk of mortality, we conducted a prospective cohort study of 7233 community-dwelling older women aged 65 years or older who were enrolled in the Study of Osteoporotic Fractures. We measured incident vertebral fractures by radiographic morphometry of paired lateral spine X-rays taken an average of 3.7 years apart. We also collected information on baseline prevalent vertebral fractures; calcaneal bone density; anthropometric measures; and demographic, medical history, and lifestyle variables. Overall mortality was assessed and confirmed by receipt of death certificates. Over an average of 3.7 years, 389 (5.4%) women developed at least one incident vertebral fracture. During an additional 8 years of follow-up, 1617 (22%) women died. Women with at least one new fracture had an age-adjusted 32% increased risk of mortality (RH=1.32; 95% CI=1.10–1.58, P=0.003) compared to those without incident vertebral fractures. After adjustment for weight loss, physical frailty markers, and nine other predictors of mortality, there was no longer an independent association between incident vertebral fractures and mortality (RH=1.06; 95% CI=0.88 1.28). Older women with incident vertebral fractures have an increased risk of mortality that may be explained by weight loss and physical frailty.
KeywordsIncident vertebral fracturesMortalityOlder women
Vertebral fractures are the most common type of osteoporotic fracture affecting up to 40% of the population of women aged 80 years or older . With the average US female life expectancy approaching 80 years, osteoporotic vertebral fractures are a public health concern. Vertebral fractures are associated with adverse outcomes including back pain , physical impairment , an increased risk of future osteoporotic fractures  and an increased risk of mortality [5, 6]. Previous studies investigating vertebral fractures and mortality have examined data on prevalent vertebral fractures, but none have assessed the risk of mortality in relation to incident vertebral fractures.
Other incident osteoporotic fractures, particularly those of the hip, are associated with mortality . In most cases, the fracture itself is not the cause of mortality; rather, the observed increased mortality may be a consequence of general poor underlying health that is further compromised by the occurrence of an osteoporotic fracture. We previously reported that older women with prevalent vertebral fractures had an increased risk of mortality that was independent of 13 other predictors of mortality . We postulated that vertebral fractures might be a marker of inflammatory disease or physiologic aging, reflected possibly by increased bone turnover. Our previous reports that low bone mineral density and bone mineral density loss are also associated with increased mortality support the hypotheses of an underlying inflammatory etiology or an increased rate of physiologic aging as potential causative factors leading to earlier death [8, 9].
To determine whether older women with incident vertebral fractures experience increased mortality, we prospectively studied 7233 women aged 65 or older who had incident vertebral fractures measured from a second lateral spine radiograph. After the second radiograph, we measured mortality rates during a mean follow-up of 8 years. Expecting that, like other osteoporotic fractures, incident vertebral fractures would be associated with increased mortality, we examined markers of frailty to assess whether these markers may offer an explanation as to the possible cause of death in these women.
Materials and methods
The study subjects were participants in the Study of Osteoporotic Fractures, a prospective cohort study designed to determine the risk factors for osteoporotic fractures. Between September 1986 and October 1988, 9704 ambulatory white women aged 65 years or older without bilateral hip replacements were recruited from population-based listings in four United States cities. Of the original cohort, 9575 had baseline lateral spine radiographs that were technically acceptable for interpretation by radiographic morphometry. An average of 3.7 years later, repeat lateral spine radiographs suitable for morphometry were obtained for 7233 women (75% of the original cohort). Of those who did not return for repeat X-ray (n=2337), 1111 (48%) were dead, 805 (34%) did not return for a full visit, 307 (13%) refused a repeat X-ray, and (5%) did not get a repeat X-ray for other reasons. Each center's institutional review board approved the study protocol, and subjects provided written informed consent.
Vertebral morphometry measurements
Radiographs of the thoracic and lumbar spine were obtained in accordance with the 1995 National Osteoporosis Foundation guidelines . Trained technicians used a visual triage system to separate radiographs into normal, uncertain, or probably fractured groups based upon a semiquantitative grading scheme that categorized a patient by the most abnormal vertebral level . Uncertain grades were further categorized by the study radiologist as normal or probably fractured. Those radiograph pairs that were categorized as probably showing a fracture (42%) were then evaluated by morphometry using six-point digitization to calculate the anterior (Ha), mid- (Hm), and posterior (Hp) heights for each vertebral body from T4 to L4. Three height ratios were calculated for each vertebral level: Ha/Hp, Hm/Hp, and Hp/Hp±1 or Ha/Ha±1. A vertebra was classified as having a prevalent fracture on the baseline radiograph if any of the three height ratios were more than 3 SD below the study population-specific mean for that level . An incident fracture was defined if any of the three vertebral heights decreased by 20% or more and by at least 4 mm compared with the baseline height. Incident fractures identified by morphometry were reviewed by a radiologist to exclude imaging artifacts or conditions such as osteophytosis and Scheuermann disease; 7% were reclassified as not having an incident fracture. In a random sample of 503 women whose radiographs were triaged and then digitized, triage missed no incident fractures according to the study definition.
Questionnaire and interview
All participants completed a baseline questionnaire that ascertained their level of education and health behaviors including smoking, alcohol use, and exercise frequency (total kilojoules expended per week in the past year). They also were asked about a history of diabetes, hypertension (measured blood pressure >160/90 or thiazide use), and perceived health status compared with that of women of similar age (very poor, poor, or fair versus good or excellent). At the third clinic follow-up visit and average of 3.7 years later, they were asked about current medication use including calcium and estrogen use, whether they had difficulty standing on their feet for 2 h, and about a history of the number of falls in the past 12 months.
At the baseline clinic examination, participants had their body weight, knee height, and systolic blood pressure measured by standard techniques . Pulse was measured in the supine position for 30 s after at least 5 min of rest. Visual contrast sensitivity was measured and the scores for low and high spatial frequencies were averaged separately . Bone mineral density of the calcaneus was measured by single photon absorptiometry (OsteoAnalyzer: Siemens-Osteon, Wahiawa, Hawaii, USA).
At the third clinic visit, body weight was measured again and the subjects were asked to rise up from a chair (without using the arms) five times. Percent weight change was calculated by subtracting weight at the baseline examination from weight at the third examination and expressed as a percentage of the baseline value.
Ascertainment of death
Follow-up for vital status was 96% complete as of December 2000 and death was confirmed by receipt of death certificates.
After confirming that hazard rates for incident vertebral fractures and death were proportional, we used Cox proportional hazards models to determine the association between incident vertebral fractures and subsequent all-cause mortality. Because prevalent and incident vertebral fractures are closely related, we checked for an interaction between baseline prevalent vertebral fracture, incident vertebral fracture, and mortality. We also performed analyses stratified by baseline prevalent vertebral fracture. To investigate whether the number of incident vertebral fractures is associated with increasing risks of mortality, we conducted analyses comparing those who had 1 or ≥2 incident vertebral fractures with women with no fractures. All results are reported as relative hazards with 95% confidence intervals.
To assess for potential confounders, we considered a list of variables that may be associated with both vertebral fractures and mortality. Based upon prior knowledge, we forced the following covariates into the final multivariable models: age, prevalent vertebral fracture, heel bone mineral density, and smoking [6, 9, 15]. Candidate confounders including education, physical activity, self-reported health, body mass index, diabetes, hypertension, alcohol use, estrogen use, calcium use, pulse >80, difficulty standing on the feet for >2 h, inability to stand from a chair without aid, poor contrast sensitivity, falls, and percent weight change were further considered in multivariable modeling if they were significantly related to mortality (P<0.15). We simultaneously added all qualifying covariates to a model of incident vertebral fractures predicting mortality and then ranked them according to the P-value of each variable's estimated beta-coefficient. We removed the candidate covariate with the highest P-value and re-ran the model; if the beta-coefficient of the main predictor, incident vertebral fracture, changed by less than 10%, the candidate covariate was removed permanently from the model. We continued this procedure of ranking and re-ranking the covariates' P-values and replacing or removing each subsequent covariate until all variables had been considered. Final models were unaffected by whether or not the forced covariates (age, vertebral fracture, bone mineral density, and smoking) were included in the ranking procedure. All analyses were performed using SAS software (SAS Institute, Cary, N.C., USA).
Baseline characteristics of subjects, classified by prevalent and incident vertebral fracture status
Never fractured (n=5639)
Baseline fracture, no incident fracture (n=1207)
No baseline fracture, incident fracture (n=180)
Both baseline and incident vertebral fracture (n=207)
Age, mean±SD, years
Self-reported health (%)
Calcaneal bone mineral density
Physical activity per SD increase
Current smoker (%)
Current calcium use (%)
% weight change
Inability to rise from a chair (%)
Difficulty standing on feet for 2 h (%)
Multivariable model of incident vertebral fractures and mortalitya
95% confidence interval
Incident vertebral fracture
Prevalent vertebral fracture
% weight changec
Age per 5 years
Calcaneal bone density
Physical activity per SD increase
Current calcium use
Inability to rise from chair
Difficulty standing on feet for 2 h
Multivariable models of incident vertebral fractures and mortality stratified by prevalent vertebral fracture statusa
Study population (n)
95% confidence interval
All subjects (7044)
Only subjects without baseline vertebral fractures (5684)
Only subjects with a baseline vertebral fracture (1360)
We also examined whether having a greater number of incident vertebral fractures was associated with a greater risk of mortality. In age-adjusted models, the relative hazards increased with an increasing number of incident vertebral fractures (0, 1, ≥2), and those who had no baseline fracture followed by two or more incident fractures were at greatest risk for mortality (RH=2.04, 95% CI=1.20–3.46, P=0.008). In multivariable models adjusted for 12 covariates, even those women with two or more incident vertebral fractures no longer had a significantly increased risk of mortality.
Results stratified by number of incident vertebral fractures: final multivariable models minus frailty covariatesa
No. incident vertebral fractures (no. subjects)
1.0 (referent group)
Only subjects without baseline vertebral fracture
1.0 (referent group)
By ranking the covariates according to their effect on the estimated beta-coefficient of the main predictor, we found that percent weight loss had the greatest effect in diminishing the association between incident vertebral fractures and mortality. When removed from the model, percent weight loss decreased the β-coefficient for incident vertebral fracture 2.4-fold. Difficulty standing up from a chair five times and difficulty standing on the feet >2 h had the next strongest effects on reducing the magnitude of the association between incident vertebral fractures and mortality. Difficulty standing up from a chair reduced the β-coefficient by 90% and difficulty standing for >2 h reduced the β-coefficient by 45%. After adjustment for age, percent weight change, and inability to risk from a chair, the association between incident vertebral fractures and mortality was no longer statistically significant (RH=1.13; 95% CI=0.94–1.36, P=0.20).
Older women who have incident vertebral fractures are at increased risk for mortality. The risk increases with greater numbers of fractures and is independent of having low bone density, prevalent vertebral fractures, or poor self-reported health. However, incident vertebral fractures were not significantly associated with increased mortality after adjustment was made for markers of frailty. Whether these markers of frailty caused or resulted from incident vertebral fractures cannot be discerned by this analysis. Weight loss, an inability to rise from a chair, and difficulty standing on the feet for >2 h may be risk factors for vertebral fractures while they also are likely complications that occur after vertebral fractures. Thus, we reported the results both with and without the frailty markers included in the multivariable models.
Aside from age, weight loss was the covariate that most strongly diminished the significant association between incident vertebral fracture and mortality. These results are not surprising since weight loss, and in particular, unintentional weight loss, is associated with an increased risk of subsequent frailty fractures, including those of the hip, proximal humerus, and pelvis . Furthermore, weight loss is also a strong predictor of multiple adverse outcomes associated with aging such as bone loss , functional decline , and mortality .
It is not surprising that incident vertebral fractures do not predict mortality independent of frailty. It may be that incident vertebral fractures are markers of an increased rate of physiologic aging, possibly through heightened inflammation, dysregulated stress responses, and/or poor cellular reserve. Previously, we suggested that vertebral fractures may be a marker of physiologic aging by mechanisms that include oxidative damage and cellular senescence . Whether vertebral fractures or markers of frailty such as weight loss, these factors may be markers of an accelerated aging process. Accelerated aging is a concept supported by studies that have attempted to identify biological markers that are predictive of adverse outcomes of aging including functional decline and mortality. For example, interleukin-6, a marker of inflammation, has been linked to osteoporosis, weight loss, functional decline, and mortality .
Although frailty may explain the relationship between vertebral fractures and mortality, it did not explain the significant association between prevalent vertebral fractures and mortality (Table 2). One possible explanation is that incident vertebral fractures were assessed only once during over a decade of follow-up so that any participants who had an incident vertebral fracture after the second radiograph would have remained undetected and counted in the no fracture group. Second, there was limited power to detect a significant difference in multiple-adjusted models of incident vertebral fracture and mortality. We had about 80% power to detect multiple-adjusted relative hazards of 1.4 or greater. Third, it is possible that fracture severity or absolute numbers of fractures are the cause of the increased risk of mortality. Our previous work demonstrated that those with more severe or greater numbers of vertebral fractures had the highest mortality rates . The current analysis found that those who had no fractures at baseline, but then went on to develop two or more incident vertebral fractures had the highest age-adjusted mortality rates; however, we did not have enough data on fracture severity or multiple incident vertebral fractures to detect a continuous graded effect.
Interestingly, the risk of mortality in subjects who had baseline prevalent fractures and two or more incident vertebral fractures was not as great as those women without baseline prevalent fractures who had two or more incident fractures. One possible explanation is survival bias; women with prevalent vertebral fractures who came to the follow-up visit may have had characteristics that made them more likely to survive than those with prevalent fractures who did not return for the follow-up visit. Also, one limitation of the prevalent fracture data is that the time of fracture is not known. Vertebral fractures frequently result from osteoporosis; however, they can also result from other causes such as trauma and it may be that those that occur earlier in life do not indicate the same increased risk of mortality as those that occur later in life.
Our study had some limitations. We only studied white women so that our study findings might not apply to other populations. Second, incident vertebral fractures were assessed only once during the follow-up period. Hence, some misclassification bias likely resulted, but this bias would be towards not finding a significant association. Third, there was no distinction made between clinically apparent and silent vertebral fractures. Vertebral fractures that are clinically recognized appear to be the most strongly associated with increased mortality . Lastly, we had limited power because only 5% were identified as having incident vertebral fractures; therefore, we were unable to assess cause-specific mortality.
In summary, older women with incident vertebral fractures are at increased risk of overall mortality that seems to be explained in large degree by weight loss and markers of decreased physical function. However, it is not known whether these markers of frailty caused or resulted from incident vertebral fractures. Either way, we do not believe that incident vertebral fractures directly cause death, but rather that they are a marker for increased risk for adverse outcomes including future osteoporotic fractures. Therefore, although the increased mortality that has been observed after vertebral fractures likely reflects a complex pathogenesis, a reasonable argument can be made to treat all persons with incident vertebral fractures to prevent future morbidity and possibly reduce the risk of death.
The authors would like to thank Paul Mischel MD for his helpful comments on the manuscript and Sandra Navarro BS and Stephanie Litwack MS for their administrative and analytic support. None of authors has any financial interest in the subject matter discussed in the manuscript. However, some authors have received funding from pharmaceutical companies. This study was supported by the UCLA Claude D. Pepper Older Americans Independence Center and by Public Health Service grants AG05407, AR35582, AG05394, AM35584, AR35583 and 1K12AG01004 from the National Institutes of Health/National Institute on Aging, Bethesda, Md., USA.