Risk Factors for Incident Vertebral Fractures in Men and Women: The Rotterdam Study†
The authors have no conflict of interest
Low BMD and prevalent vertebral fractures are known risk factors for incident vertebral fractures. In 3001 men and women from the Rotterdam Study, prevalent nonvertebral fractures, early menopause, current smoking, and walking aid use were also strong risk factors for incident vertebral fractures.
Introduction: Thus far, age, low BMD, and prevalent vertebral fractures are the only well-known risk factors for incident vertebral fractures. Therefore, our aim was to investigate other potential risk factors for incident vertebral fractures in the elderly.
Materials and Methods: This study was based on the Rotterdam Study, a large prospective population-based cohort study among men and women ≥55 years of age. For 3001 subjects, spinal radiographs were obtained at baseline and again ∼6.3 years later. These follow-up radiographs were scored for vertebral fractures using the McCloskey-Kanis method. Whenever a vertebral fracture was detected, the radiograph was compared with the baseline radiograph. If this fracture was not already present at baseline, it was considered incident. At baseline, information on potential risk factors was obtained.
Results: Low BMD and prevalent vertebral fractures were strong risk factors for incident vertebral fractures in both men and women (RR 2.3 [1.6-3.3] and 2.2 [0.9-5.0] for men and RR 2.1 [1.6-2.6] and 4.1 [2.5-6.7] for women, respectively). For women, age, early menopause (≤45 years of age; RR 1.0 [1.1-3.5]), current smoking (2.1 [1.2-3.5]), and walking aid use (2.5 [1.1-5.5]) were additional independent risk factors. For men, only a history of nonvertebral fractures was a significant independent risk factor (OR 2.4 [1.2-4.8]).
Conclusion: Apart from low BMD and prevalent vertebral fractures, prevalent nonvertebral fractures are associated with an increased incident vertebral fracture risk in men. In women, early menopause, current smoking, and walking aid use are additional independent risk factors for incident vertebral fractures.
VERTEBRAL FRACTURES ARE common fractures in osteoporosis, and they are associated with increased morbidity and mortality.(1–4) However, in contrast to other fractures, especially hip fractures, risk factors for incident vertebral fractures have not been extensively studied. The most important risk factors for nonvertebral fractures are age, gender, low BMD, low body weight, a history of prior fractures, and family history of fractures.(5–8) It is uncertain whether these same risk factors are equally important for vertebral fractures or whether other factors play a role. We and others have previously reported that age, low BMD, and the presence of baseline vertebral fractures are strong and independent risk factors for incident vertebral fractures.(9–11) Furthermore, Gregg et al.(12) suggested that physical activity is inversely associated with vertebral fracture incidence.
Recently, the European Prospective Osteoporosis Study (EPOS) published a study on determinants of incident vertebral fractures in men and women. In this study, age and low weight, and for women, late menarche (≥16 years of age) and use of hormone replacement therapy (HRT) determined vertebral fracture risk, whereas lifestyle factors did not.(13) In contrast to the Rotterdam Study, a population-based single-cohort study, the EPOS study is a multinational multicentered age-stratified study in which data on risk factors were gathered by an administered questionnaire. In our study, a home interview was performed by trained interviewers, the results of which were validated. Furthermore, in this study, only age-adjusted relative risks were presented. In addition, several cross-sectional studies have been performed on risk factors for prevalent vertebral fractures.(14–22) These studies suggest that age, low bone mass, smoking, alcohol intake, and low physical activity are associated with an increased risk of vertebral fractures. However, these fractures may have occurred several years before, and conclusions on causality and temporal relationships are less secure from these analyses.
In this study, we evaluated potential risk factors for incident vertebral fractures in both elderly men and women from a large prospective population-based single-cohort study.
MATERIALS AND METHODS
The Rotterdam Study is a large prospective population-based cohort study of men and women ≥55 years of age. The study objective was to investigate the incidence of, and risk factors for, chronic disabling diseases. Both the rationale and the study design have been described previously.(23) All 10,275 inhabitants of Ommoord, a district in Rotterdam, the Netherlands, were invited to participate. Of these, 7983 participated in the study (response rate, 78%).
Of the original cohort, 4216 subjects (2467 women) were available for a second follow-up visit between 1997 and 1999. Of the subjects without a second visit, 2004 subjects died (1199 women), 165 subjects were unable to come, and 1658 refused to come to the research center for various reasons (no longer being interested in participating in a study, physical complaints, etc.). All participants signed informed consent, and the Medical Ethics Committee of the Erasmus Medical Centre has approved the Rotterdam Study.
Data collection for potential risk factors
Between 1990 and 1993, an extensive baseline home interview on medical history was performed by trained interviewers. Information on baseline drug use, such as use of diuretics, systemic glucocorticoids, thyroid hormone, statins, HRT, and oral contraceptives (OC), was gathered. Total caloric intake and calcium intake during the preceding year were assessed by trained dietitians using a semiquantitative food frequency questionnaire. The conversion from foods to energy and nutrient intake was established using the Dutch Food Composition Table.(24) For the current analyses, calcium intake was adjusted for total caloric intake. Smoking habits were categorized as current, former, or never. Recent falling was described as at least one fall in the year before the baseline interview. Data on history of nonvertebral fractures at or after age 50 and use of a walking aid were obtained. Lower limb disability was assessed using a modified version of the Stanford Health Assessment Questionnaire, as described previously.(25) Information was obtained on age at and reason for menopause, defined as cessation of menses for 12 consecutive months. Natural menopause was defined as menopause occurring spontaneously, not after any intervention that would have stopped the menses. For non-natural menopause, we validated the date and indication of surgery by checking the General Practitioners' patient records and hospital discharge letters.
After the home interview, subjects visited the research center for a clinical examination, laboratory examination, radiographs of the spine, and BMD measurements. Height and weight were measured with subjects wearing indoor clothing and without shoes. Body mass index (BMI) was calculated by dividing weight by height squared (kg/m2). Subjects were classified as diabetics if they reported use of antidiabetic therapy (code A010 of the Anatomical Therapeutical Chemical classification index; WHO 1992), or if the pre- or postload serum glucose level was ≥11.1 mM. BMD measurements of the femoral neck and the lumbar spine (L2-L4) were performed by DXA (Lunar DPX-L densitometer) as described previously.(26)
Vertebral deformity assessment
Both at baseline, between 1990 and 1993, and at the second follow-up visit, between 1997 and 1999, a trained research technician obtained lateral radiographs of the thoracolumbar spine of subjects who were able to come to the research center. All radiographs were taken following a standard protocol, with a distance between source and plate of 120 cm, using a Solarize FV (General Electric CGR, Utrecht, The Netherlands). The follow-up radiographs were available for 3549 individuals (2022 women) who survived an average of 6.3 years after their baseline center visit and who were still able to come to our research center. All follow-up radiographs were evaluated morphometrically in Sheffield by the McCloskey-Kanis method, as described previously.(9,26) If a vertebral fracture was detected, the baseline radiograph was evaluated as well. If the fracture was already present at baseline, it was considered a baseline prevalent fracture. If, however, the vertebra was determined to be normal at baseline and any of the three vertebral heights (anterior, central, or posterior) showed a minimum decrease of at least 4.6 mm or 15% in absolute height on the later film, it was considered an incident fracture. All vertebral fractures were confirmed by visual interpretation by an expert in the field to rule out artifacts and other etiologies, such as pathological fractures. We excluded 80 individuals who had not attended the baseline visit. We also excluded 468 men and women for whom data on one or more risk factors were missing. Therefore, the final population for analysis consisted of 3001 individuals (1624 women) with information both on vertebral deformities and on potential risk factors. The current study population did not differ from the total of subjects who were able to come to the second follow-up center visit (Appendix).
Differences in baseline characteristics were compared using Student's t-test for continuous variables and χ2 for categorical variables. Age was evaluated in 5-year strata, whereas BMD was evaluated per gender-specific SD decrease. BMI, height, and weight were evaluated in quartiles. All potential risk factors were tested for significance in unadjusted models using logistic regression. To take account of those relatively weak potential risk factors that do not reach statistical significance at a p value of 0.05 because of low numbers, risk factors were considered statistically significant when the p value was <0.10 (two-tailed). All associations of risk factors and incident vertebral fractures that were statistically significant in the crude analyses for either men or women were evaluated for independence from age, by adjusting for age as a continuous variable in the model, using multivariable logistic regression. When still significant, additional independence from BMD and subsequently from both BMD and the prevalence of baseline prevalent vertebral fractures were tested. This was done to test whether the risk factors studied reflect a low BMD and/or the presence of vertebral fractures or whether these risk factors independently add to the etiology of incident vertebral fractures. Finally, because risk indicators remained fairly stable, although the power of the model is low because of the large amount of degrees of freedom and relatively few cases, all remaining risk factors were entered into a multivariable model to test independence from each other. All analyses were performed for men and women separately. Because incident vertebral fractures are a rare condition (158 cases in 3001 subjects who were at risk), under the rare disease assumption, all odds ratios can be considered as relative risks and are reported accordingly. SPSS 10.0 for Windows was used for all analyses.
During an average follow-up time of 6.3 years, 44 men and 113 women suffered at least one incident vertebral fracture. Table 1 shows baseline characteristics of men and women with and without incident vertebral fractures. Both men and women with incident vertebral fractures had a lower BMD at both sites measured, more often had a history of fractures, both vertebral and nonvertebral, and tended to smoke more. In addition, women with incident vertebral fractures were older, thinner, more frequently used a walking aid, and had an earlier age at natural menopause. No differences in calcium intake, adjusted for caloric intake, were observed between subjects with and without incident vertebral fractures.
Table Table 1.. Baseline Characteristic for Men and Women With and Without Incident Vertebral Fractures From the Rotterdam Study
In an unadjusted analysis, incident vertebral fracture risk increased more strongly with age in women than in men (Table 2). Similarly, low BMI and low body weight, but not height, were associated with increased incident vertebral fracture risk in women only. The presence of at least one prevalent vertebral fracture, however, was associated with an increased risk of new vertebral fractures in both sexes, as was having had a prevalent nonvertebral fracture at or after age 50 years. As reported earlier, a low BMD at both the femoral neck and lumbar spine was associated with a strongly increased vertebral fracture risk.(9) Only in women, the use of a walking aid and a moderate or severe lower limb disability were risk factors for incident vertebral fractures. Current smoking was associated with an increased vertebral fracture risk in both sexes, although statistically significant in women only.
Table Table 2.. Unadjusted Analyses of Potential Risk Factors for Incident Vertebral Fractures
For women, several estrogen-related factors were predictors for incident vertebral fractures. Women with an early natural menopause, defined as an age at natural menopause at or before age 45, had a 2.7 times increased risk of an incident vertebral fractures compared with women who had their menopause above age 50. Women in our cohort with a non-natural menopause (e.g., surgery) took HRT more often and longer (data not shown) and had a somewhat decreased vertebral fracture risk. Ever use of HRT and OC were also univariately associated with a decreased vertebral fracture risk.
Furthermore, the use of several other types of medications at baseline (diuretics, systemic glucocorticoids, thyroid hormone, and statins) was evaluated in relation to incident vertebral fractures, but because of low exposure rates, no significant associations were observed with incident vertebral fractures (data not shown).
The associations of the risk factors that were significant in the unadjusted analyses, with incident vertebral fractures, were evaluated for independence from age (Table 3). In contrast to men, a history of a nonvertebral fracture in women at or after age 50 was no longer a risk factor for incident vertebral fractures. In addition, a lower limb disability and the use of HRT or OC were no longer significantly associated with increased vertebral fracture risk in women.
Table Table 3.. Risk Factors for Incident Vertebral Fractures After Adjustment for Age
Age, BMD, and prevalent vertebral fracture-adjusted analyses
We evaluated whether the associations of the remaining risk factors with incident vertebral fractures were independent of lumbar spine BMD, and additionally, of the presence of baseline prevalent vertebral fractures (Table 4). When adjusting for BMD, neither BMI nor weight was a risk factor for incident vertebral fractures. All risk factors that were independent from BMD were also independent from the presence of a baseline prevalent vertebral fracture.
Table Table 4.. Risk Factors for Incident Vertebral Fractures After Adjustment for Lumbar Spine BMD and Additionally for Prevalent Vertebral Fractures
Full multivariable model
The power of a full multivariable model is low because of the many degrees of freedom and relatively few incident cases. However, because risk factors remained stable in subsequent steps of the analysis, all remaining risk factors were entered into a multivariable model to test independence from each other (Table 5). For women, age, low BMD, the presence of a prevalent vertebral fracture at baseline, use of a walking aid, (natural) menopause before or at the age of 45, and current smoking at baseline remained strong independent risk factors for incident vertebral fractures. For men, only a low BMD and prevalent nonvertebral and vertebral fractures were independent risk factors for incident vertebral fractures, although for the presence of prevalent vertebral fractures, this was only borderline significant, because of a low number of cases.
Table Table 5.. Multivariable Model of Independent Risk Factors for Incident Vertebral Fractures
In this study, we observed that apart from age, low BMD, and the presence of at least one baseline prevalent vertebral fracture, current smoking, use of a walking aid, and early menopause are also strong independent risk factors for incident vertebral fractures in women. In men, only a history of a prevalent nonvertebral fracture was an additional significant risk factor for incident vertebral fractures.
In this study, our aim was to investigate known risk factors for a low BMD, hip fractures, or both as potential risk factors for incident vertebral fractures.
In contrast to hip fractures, vertebral fractures mainly occur spontaneously.(27,28) Therefore, it was not surprising that recent falling before baseline was not associated with increased vertebral fracture risk. In addition, a lower limb disability was no longer associated with increased vertebral fracture risk after adjustment for age, suggesting that lower limb disability is merely a marker for older age. For men, but not for women, having a history of a nonvertebral fracture after age 50 years was a statistically significant risk factor for incident vertebral fractures. In women, only an unadjusted association was observed. The relative risk estimate dropped, and statistical significance was no longer achieved after adjustment for age. This might be explained by the fact that vertebral fractures on average occur at an earlier age than hip fractures do.(29) In addition, the risk estimates further dropped after adjustment for prevalent vertebral fractures.
In contrast to the results shown by the EPOS, in our study, current smoking at baseline was also associated with an increased incident vertebral fracture risk in both men and women, although only statistically significant in women.(13) Several studies have previously suggested an inverse relation between smoking and BMD.(30–32) However, the mechanism through which smoking influences BMD and fracture risk is not fully elucidated. Some research suggests that nicotine directly influences bone metabolism.(33) Smoking might also influence BMD and fracture risk through its effect on body weight, sex steroid hormone levels, and other hormones and enzymes involved in bone regulation or overall lifestyle.(34–37) Finally, there is indirect evidence that smoking may damage blood supply to bone.(38,39)
The effects of estrogen on bone are well established.(40–42) This study strengthens the importance of estrogen exposure further by showing that an early age at menopause is associated with an increased incident vertebral fracture risk. This effect seems limited to women whose age at menopause was before or at the age of 45 years, which was 11% of the whole female population. Women with a non-natural menopause more often took HRT, thereby supplementing their estrogen deficiency. In line with the EPOS data, long- and short-term HRT use and oral contraceptive use were all univariately associated with a decreased incident vertebral fracture risk.(13) However, in our study, these associations disappeared after adjusting for age. In the Netherlands, oral contraceptives were not frequently prescribed until the mid-1960s, resulting in only the younger women in our cohort ever using these drugs.
As mentioned earlier, previous studies on risk factors for incident vertebral fractures have been limited to age, low BMD, prevalent vertebral fractures, and physical activity.(9–12) Apart from the physical activity, on which we did not have data, results are generally very consistent between studies. However, there are some discrepancies between our study and the results of the recently published EPOS.(13) It is likely that these discrepancies can at least in part be explained by important differences in study design and in the way of gathering data. Because the populations are quite different in distribution, results cannot be generalized from one study population to another.
There are limitations to this study. Although this is a large single-cohort population-based study, some health selection bias is present. In contrast to other types of fractures, vertebral fractures primarily occur spontaneously, and only about one-third of subjects complain enough to come to clinical attention.(27,28) Thus, the only way to investigate all incident vertebral fractures is to examine baseline and follow-up radiographs of the spine. Therefore, a selection bias is unavoidable in studies on incident vertebral fractures. Because of this selection bias, the subjects included in our study may not be representative of all subjects with vertebral fractures in the general population. In addition, physical activity has been shown to be a risk factor for incident vertebral fractures.(12) Unfortunately, in our cohort, we do not have data on level of physical activity.
In conclusion, the results of this study show that in women, besides low BMD and the presence of baseline prevalent vertebral fracture, age, early menopause (at or below age 45), current smoking, and use of a walking aid are strong and independent risk factors for incident vertebral fractures. In men, a positive history of nonvertebral fractures and possibly current smoking are additional independent risk factors. Using these easily identifiable risk factors might be useful in facilitating the process of risk assessment in general practice. For future research, it would be of great value to perform a large meta-analysis on risk factors for incident vertebral fractures, combining all international studies on this topic to have the power to fully investigate all risk factors.
The authors thank L. Buist, the DXA and radiograph technician. Moreover, we thank the participating general practitioners and the many field workers at the research center in Ommoord. This study was supported by Dutch Health Research and Development Council (Zorg Onderzoek Nederland) Grant 22000011.
Table Appendix. Baseline Characteristics for Selected and Unselected Participants