Statins inhibit an enzyme in the mevalonate pathway and therefore may affect bone. In this first study on both symptomatic and nonsymptomatic vertebral fractures in the elderly (N = 3469), we show that long-term statin use is significantly associated with a 50% lower vertebral fracture risk. Randomized trials on statins and fractures, carried out in populations at risk for fractures, are needed.
Introduction: Statins are cholesterol-lowering agents that could potentially affect bone. Previous studies on statin use and fracture risk reported contradictory results and did not include both symptomatic and nonsymptomatic vertebral fractures.
Materials and Methods: To examine the association between statin use, vertebral fractures, and lumbar spine BMD, we performed a prospective population-based cohort study in men and women (N = 3469) ≥55 years of age. These individuals had both baseline and follow-up spinal X-rays available. Statin use was obtained from detailed computerized pharmacy data, and the total number of days of exposure before second X-ray was calculated. A multivariate logistic regression model was fitted to calculate odds ratios and CIs.
Results: During a mean follow-up of 6.5 years, 176 incident vertebral fractures occurred. There were 508 statin users and 16 exposed cases. The adjusted relative risk for incident vertebral fracture in users of statins (compared with nonusers) was 0.58 (95% CI, 0.34-0.99). The relative risk decreased on higher cumulative use to 0.52 (95% CI, 0.28-0.97) for use for more than 365 days during the study period. Use of (the hydrophilic statin) pravastatin and use of nonstatin cholesterol-lowering drugs was not significantly associated with vertebral fracture risk. Statin use was not significantly associated with lumbar spine BMD.
Conclusion: Statin use is associated with a lower risk of vertebral fracture. Randomized clinical trials in a population at risk for fracture are needed to examine this association.
STATINS ARE 3-hydroxy-3-methylglutaryl coenzyme A (HMG CoA) reductase inhibitors that are used to treat hypercholesterolemia. By inhibition of the enzyme that catalyzes the conversion of HMG CoA to mevalonate in cholesterol synthesis, statins lower serum low-density lipoprotein (LDL) cholesterol levels. Inhibition of the mevalonate pathway may also have positive effects on bone.(1) Statins were shown to increase expression of bone morphogenetic protein-2 (BMP-2) in bone cells, thereby increasing bone formation.(1–4) Statins may also play a role in reducing bone resorption by inhibiting osteoclast differentiation.(1,5)
Studies on use of statins and fractures or BMD have reported contradictory results. In most studies, an association between statin use and lower fracture incidence was found,(6–10) but not all reports confirmed this association.(11–14) The studies on lumbar spine BMD and statin use were contradictory as well.(10,15–21) One study also examined the association between statin use and symptomatic vertebral fractures,(12) and another study included data on morphometric vertebral fractures in women with low bone mass and statin use.(22) However, to our knowledge, there have not been reports on the incidence of all (non)symptomatic vertebral deformities among statin users in the general population. Vertebral fractures are the most common and typical fractures in osteoporosis patients and are associated with increased morbidity and mortality.(23,24) Most vertebral fractures are not symptomatic and are not, in contrast to other fractures, related to falls. Because comparisons of multiple radiographs, taken at different points in time, are essential to study the incidence of vertebral fractures, this type of fracture is not well studied. Therefore, we examined the association between the use of statins and the incidence of vertebral fractures in a prospective population-based cohort study with baseline and follow-up radiographs of the lateral spine available.
MATERIALS AND METHODS
The Rotterdam Study is a population-based cohort study designed to assess the occurrence and determinants of diseases in an aging population. The cohort includes 3105 men and 4878 women ≥55 years of age (78% of the eligible population), who lived for at least 1 year in a defined district in Rotterdam in The Netherlands.(25) All participants gave written informed consent to retrieve all relevant medical information from treating physicians. The Medical Ethics Committee of the Erasmus MC approved the study. Baseline measurements were obtained from 1990 to 1993 and consisted of a home interview and research center visits for physical examinations. The third examination phase took place from 1997 to 1999. For this study, all participants were included for whom spinal radiographs were available both at baseline and at the third examination.
In the research area, there are seven fully computerized pharmacies that are all linked to one network. During the study, all participants filled their prescriptions in one of these seven pharmacies. Information on all dispensed drugs starting January 1, 1991 is available in computerized format on a day-to-day basis. The data consist of information on the date of prescribing, the total amount of drug units per prescription, the prescribed daily number of units, the product name of the drugs, and the Anatomical Therapeutic Chemical (ATC) code.(26)
Because we studied all vertebral deformities, including nonsymptomatic fractures, we did not have an exact date of occurrence of fracture. Therefore, we added the total number of days of statin use between the two radiographs. For participants (n = 566) who had their first radiograph taken before January 1, 1991, we added the number of days of statin use from that date onward. Patients were classified as statin users if they received at least one prescription for statins between the baseline and follow-up radiograph. To investigate effects of increasing cumulative exposure, we defined a priori three mutually exclusive intervals of statin use: no use, short-term use (1-365 days), and long-term use (>365 days). We expressed the mean prescribed daily dose during the study period as a proportion of the defined daily dose. One defined daily dose of statins equals the standard recommended adult daily dose for treatment of hypercholesterolemia in The Netherlands. In contrast to lipophilic statins (defined as all statins except pravastatin, which included atorvastatin, simvastatin, fluvastatin, and lovastatin in our study), the hydrophilic pravastatin was previously shown to not induce BMP-2.(4) Therefore, we distinguished pravastatin use from use of all other (lipophilic) statins in separate analyses.
Vertebral deformities were assessed as described previously.(27) In short, all radiographs of the third examination phase were evaluated morphometrically in Sheffield using the McCloskey-Kanis method.(28) 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 prevalent fracture. However, if the specific vertebra was determined to be normal at baseline, it was considered an incident fracture. All vertebral deformities were confirmed using visual interpretation by an expert in the field to rule out artifacts and other etiologies such as pathological fractures.
Lumbar spine (L2-L4) BMD was measured by DXA (Lunar DPX-L densitometer) during the third examination phase as described previously.(29)
During a baseline home interview, trained interviewers gathered information on medical history, risk factors for chronic diseases, medication use, and habitual diet. Among other information, potential risk factor information such as smoking habits and age at menopause was gathered. Lower limb disability was assessed using a modified version of the Stanford Health Assessment Questionnaire(30) and by calculating the mean score of answers to questions concerning rising, walking, bending, and getting in and out of a car.(31) A score of more than one indicates disability.
After the home interview, the participants were invited to visit the research center for clinical examinations and laboratory assessments. Nonfasting blood samples were drawn. Cognitive impairment was measured using the Mini Mental State Examination.(32) We computed the 5-year Framingham cardiovascular disease risk score for all participants with help of a previously published algorithm.(33) This score predicts 5-year cardiovascular disease risk and can be used as a measure of indication for statin therapy. Use of other medication was extracted from the pharmacy database.
Demographic and clinical characteristics of participants with and without vertebral fractures and statin users versus nonusers were compared with Student's t-test and Pearson's χ2 test.
To examine the association between vertebral fractures and statin use, we used a logistic regression model with incident vertebral fractures as the dependent variable and with exposure to statins as the independent variable. In subsequent models, we adjusted for age, gender, and the number of days of available pharmacy data. Other co-factors were also included if they caused a change in the risk estimate of use of statins at least 10% or were biologically plausible. Tests of significance for the ordered variable of the categories of statin use were considered to be tests for trends of increasing duration of use. Effect modification by age, gender, and BMD at baseline was investigated by stratified analysis.
The association between lumbar spine BMD and statin use was investigated with ANOVA. We computed crude and adjusted means of BMD for categories of statin use. The category reflecting no use of statins was used as the reference category for significance tests.
Multiple imputation was used to impute missing information for confounding variables. Five imputation values were calculated based on the posterior predictive distribution of the missing values, and five complete data sets were created. On each complete set, the statistical analyses were performed, and the point estimates of the five data sets were combined to form one summary statistic as the average of the five components. The variance of the summary statistic is calculated from the within-imputation variance and the between-imputation variance. The combined variance accounts for the uncertainty introduced by estimating the missing values.(34) In the original data set, there were no missing values for age, gender, and length of follow-up, and 170 (4.9%), 235 (7.3%), and 19 (0.5%) missing values for diabetes mellitus, cardiovascular disease risk, and body mass index (BMI), respectively. SPSS 11.0 for Windows (SPSS, Chicago, IL, USA) and SAS (procedures mi, mianalyze, and logistic, version 8.2; SAS, Cary, NC, USA) were used for the analyses.
At the third examination phase, nearly 2000 participants of the original 7983 had died, and 1260 participants were too old or disabled to visit the center, or they refused to come. For 3469 of 4730 participants in this examination phase, both baseline and follow-up radiographs were available (1498 men and 1971 women), and for 3525 participants (1512 men and 2013 women), lumbar spine BMD measurements were available. During the study period with a mean follow-up of 6.50 years (range, 3-9 years), 176 individuals suffered a new vertebral fracture. Of these new fractures, 48 occurred in participants with a vertebral fracture present at baseline. In total, 508 participants of the vertebral fracture analyses used a statin on ≥1 day during the study period, and among cases, there were 16 exposed participants (5 short-term users and 11 long-term users). There were 359 subjects who used simvastatin, 63 subjects who used fluvastatin, 70 who used atorvastatin, and 106 who used pravastatin. Some subjects took more than one type of statin. Of the 3525 participants with BMD measurements, 511 were statin users.
Characteristics of participants with and without vertebral fractures and of participants who did and did not use statins during follow-up, respectively, are described in Table 1. As expected, participants with a vertebral fracture were older, more often female, and more often had a history of fractures. They had a lower BMD in the lumbar spine and a lower BMI. Statin users were younger, more often had a history of hypertension and diabetes mellitus, and more often were smokers. The 5-year cardiovascular disease risk was significantly higher for statin users, but it did not differ between fracture cases and noncases.
Table Table 1.. Baseline Characteristics of the Study Population
Use of statins on ≥1 day of the period between the two radiographs was associated with a lower incidence of vertebral fractures. The incidence of vertebral fractures was ∼40% decreased in statin users (Table 2). After adjustments for age, gender, length of available pharmacy data between the radiographs, diabetes mellitus, BMI, and the natural logarithm of the 5-year risk of cardiovascular disease (a proxy for indication for statin therapy), statin use was still associated with a significant risk reduction (odds ratio [OR]: 0.58; 95% CI, 0.34-0.99). Additional adjustments for prevalent vertebral fractures (present at baseline), baseline lumbar spine BMD, use of other medications (thiazides, estrogens), lower limb disability, smoking, or presence of other diseases, such as thyroid disease, cholesterol levels, and Mini Mental State Evaluation (MMSE) score, did not change the association essentially.
Table Table 2.. Risk of Incident Vertebral Fracture With Use of Statins
When we further examined the association between statin use and vertebral fractures in terms of duration of statin use, we found a statistically significant risk reduction associated with long-term statin use (OR, 0.51; 95% CI, 0.28-0.95). After adjustments, the OR for long-term use did not essentially change. There was a significant trend in decreasing ORs with increasing duration of use. In contrast with users of lipophilic statins (five short-term exposed cases, six long-term exposed cases) who had a risk reduction of ∼60% after statin use for >1 year (Fig. 1), participants that used pravastatin (four exposed cases) did not have a reduced risk of vertebral fracture (OR, 1.26; 95% CI, 0.44-3.53). The median dose of statin users was 0.7 defined daily dosages (DDD). Over the study period, most statin users had a mean dose over the study period below 1 DDD. Therefore, the variation in doses taken by the participants was too low to study dose effects.
When we examined use of other lipid-lowering drugs (e.g., fibrates and nicotinic acids), we did not observe a significant association between use and vertebral fracture incidence (adjusted OR, 0.85; 95% CI, 0.37-1.97). Among participants who did not take any cholesterol-lowering drug during the study period, cholesterol levels at baseline, categorized in tertiles, were not significantly associated with vertebral fractures (highest versus lowest level: adjusted OR, 1.19; 95% CI, 0.78-1.83; trend p = 0.44).
Stratification on age at baseline (≤65 versus >65 years) by gender and on lumbar spine BMD at baseline (≤1.09 versus >1.09 g/cm2) did not reveal any effect modification by these factors.
We conducted additional analyses to examine the potential effect of misclassification of exposure. All participants who used statins or other lipid-lowering drugs at baseline were excluded. These participants may have used statins before baseline, and therefore, could have been assigned to the wrong duration category. Exclusion of these subjects (n = 105) did not change the risk estimates essentially (any use OR, 0.62; 95% CI, 0.35-1.11; long-term use OR, 0.52; 95% CI, 0.28-0.99). After exclusion of subjects with a prevalent vertebral fracture at baseline, no essential changes in the estimates were observed.
Lumbar spine BMD was not associated with statin use (Table 3). Although mean BMD was highest among persons with the longest duration of statin use, there was no significant trend for increasing BMD with increasing duration of use for either all statins or lipophilic statins.
Table Table 3.. Mean Lumbar Spine BMD for Categories of Statin Use
Our results show that use of statins is associated with a decreased risk of incident vertebral fractures. When statins are used for ≥1 year, this risk is reduced by ∼50%. Pravastatin, in contrast to other statins, seems to lack this protective effect, although we emphasize that absolute numbers of fracture cases were low. We did not observe a significant relationship between statin use and lumbar spine BMD.
Our results are in line with most other observational studies of statin use and (nonvertebral) fracture incidence. Several studies detected a trend toward a lower risk of fracture of ∼40-60%.(7–10) At variance with those studies, van Staa et al.(12) did not confirm this association with fractures but reported an OR of 1.15 (95% CI, 0.62-2.14) for vertebral fractures. However, vertebral fractures, as assessed in that study, were symptomatic in contrast to our study in which we examined all participants for vertebral deformities with follow-up radiographs. Because only one-third of all vertebral fractures come to medical attention,(35) it is possible that the above-mentioned study was influenced by diagnostic bias. In secondary analyses of two randomized trials on statin use and mortality, no association was found between statin use and fractures reported as adverse events.(11,13) When we recalculated the data available from one of these studies (the 4S Study),(13) we found a relative risk of vertebral fracture with simvastatin use of 0.47 for subjects >60 years of age, which is in line with our data. Pravastatin use, investigated in the other trial, is reported to have less effect on bone than use of other statins, which was also in line with our study. There have been several studies on the association between lumbar spine BMD and statin use. Some studies did not find an association,(10,16,19–21) whereas other studies did report an association between lumbar spine BMD and statin use.(15,17,18) In our study, we did not observe an association between statin use and lumbar spine BMD. In elderly people, osteoarthritis, which is associated with an increased BMD, occurs often in the spine, and because our population consists of the elderly, we expect that this could have influenced our results.
Previous studies on statin use and risk of fracture were criticized, because risk estimates were not adjusted for common confounders such as BMI, diabetes mellitus, and dementia. In our study, we had extensive information on potential confounders, and therefore, could adjust for actual confounders. Vertebral fractures, in contrast with other fractures, are not strictly related to falls. Differences in activity levels and fall incidents between statin users and nonusers are, therefore, not a likely explanation for our results.
Because our participants had to survive until the third examination phase, they were possibly healthier than the general population of the same age. This does not necessarily influence the association between statin use and vertebral fractures. Statin use was ascertained from pharmacy records and therefore not subject to recall bias or subject to assumptions about duration of use such as in studies that only had data on baseline statin use. Misclassification of exposure, however, cannot be excluded, because we do not have an exact fracture date. We expect misclassification of exposure to be randomly distributed among cases and controls, which leads rather to underestimation than to overestimation of an association. Confounding by indication would happen when physicians prescribed statins predominantly to subjects at a lower risk for vertebral fracture. In our study, statin users did not differ much from nonusers, and we adjusted for all co-variates that were different between users and nonusers in our analyses. We did not detect an association between cholesterol levels and fractures. Use of other lipid-lowering drugs was also not related to fracture incidence, which confirms the findings in previous studies.(7,9,36) Because of all above-mentioned analyses, confounding by indication is unlikely.
In conclusion, long-term statin use is associated with a 50% lower risk of vertebral fracture. Statins are designed to act in the liver, and in the future, statins with higher affinity to bone could be potentially useful in the treatment of osteoporosis and fracture prevention. At present, there is not enough evidence to prescribe statins for the indication of osteoporosis. Randomized trials designed to investigate this subject and carried out in the proper population (e.g., patients at high risk for fracture) could be helpful to solve this issue.
Dr Sturkenboom receives funding from Pfizer Inc. All other authors have no conflict of interest.