The prevalence of hypertension reaches 66% in adults over age 60 years, with over half of older adults using an antihypertensive medication.1 These medications work through a variety of mechanisms, with several having a putative effect on bone metabolism. Notable examples include thiazide diuretics, which act at the distal convoluted tubule to increase calcium absorption2; loop diuretics, which act at the distal tubule to decrease calcium absorption2; beta blockers, which block input from the sympathetic nervous system, enhancing bone formation3; and angiotensin-converting enzyme (ACE) inhibitors and angiotensin-receptor blockers, which block the rennin-angiotensin system, shifting the balance toward bone formation.4
Studies of antihypertensive agents have demonstrated effects on bone mineral density (BMD). In longitudinal treatment studies performed in humans, thiazide diuretics and beta blockers have been shown to improve BMD,5–9 and treatment with loop diuretics has been shown to reduce BMD in some10, 11 but not all studies.12 Several previous epidemiologic studies investigated the association between these medications and fracture risk. While several studies found that beta blocker use was associated with a reduction in osteoporotic fracture risk,5–7, 13–15 others have found no effect with these agents.16–19
Prior studies of fracture risk typically have compared active users of one type of antihypertensive agent with nonusers, but nonusers were not restricted to subjects with elevated blood pressure. Since hypertension may affect BMD and fracture risk, these designs may not be optimal.20 In addition, nonuser comparators who are not users of other antihypertensive agents may differ across many measured and unmeasured covariates, introducing the possibility of confounding bias. This prior work has not focused on new users of antihypertensive agents, a recommended method for pharmacoepidemiology.21 New-user designs are especially important when the risk of a drug-related toxicity changes over time. Possible effects of antihypertensive agents on fracture risk likely fall into two categories based on duration of use: early effects through gait imbalance and late effects through changes in BMD. Because of this distinction, a design that is not limited to new users may introduce important bias owing to the overlap of these two effects.
We examined the risk of typical osteoporosis fractures among older adults who were new users of antihypertensive drugs.
We examined Medicare beneficiaries 65 years of age or older who also were enrolled in a state-run drug benefit program for low-income older adults. Potentially eligible subjects had at least one outpatient or inpatient medical encounter with a diagnosis of hypertension (DRG 134 or ICD-9 401.xx, 402.xx, 403.xx, 404.xx, or 405.xx), had not filled a prescription for any antihypertensive agent in the prior 30 days, and had consistent health care system use during this period (see Supplemental Table S1 for a listing of antihypertensive agents considered and their categories). From this group, we identified subjects who began treatment with a single antihypertensive agent. Therapy with a combination agent or the initiation of agents from two different categories was excluded because of the difficulty with interpretation. Patient identifying information was encrypted before the data were provided to the research team. This work was approved by the Partners Healthcare Institutional Review Board.
Antihypertensive medications were considered based on class or mechanism of action—ACE inhibitors, angiotensin-receptor blockers, beta blockers, calcium channel blockers, loop diuretics, and thiazide diuretics—and not the specific agents. Only new use was considered, and study follow-up began with the date of the first antihypertensive prescription filling (index date). Follow-up continued until 30 days after the last available pill was taken, based on the day supply from the last drug dispensing. Follow-up was truncated with the first of any censoring events, including death, osteoporotic fracture (see below), nursing home admission, or end of study database. Nursing home admission was considered a censoring event because drug use in the nursing home may not be recorded accurately in claims data.
In sensitivity analyses, the initial prescriptions for antihypertensive medications were categorized according to dosage using the World Health Organization (WHO) system for defined daily dosages.22 In addition, we examined different periods during follow-up—days 1 to 90, 91 to 180, 181 to 365, and more than 365. By breaking follow-up time into these different periods, we thought that we might be able to observe “early” and/or “late” effects of these agents. It is possible that early effects would be related to gait or balance issues and late effects related to BMD changes.
We studied four typical osteoporotic fractures well defined in health care utilization data: hip, distal forearm, humerus, and pelvis. We did not have radiographs available to confirm fractures or to assess for asymptomatic fractures. We used previously developed algorithms (Supplemental Table S2) that have been shown to have high positive predictive values for incident fractures.23, 24
The covariates were defined based on information from health care utilization data, including medical diagnoses, surgical procedures, and pharmacy utilization. No information was available on blood pressure readings or BMD values. However, we did have information on many relevant covariates (listed in Table 1) from health care utilization data during the 12 months preceding the index date (see Supplemental Table S3 for details). These data include age, gender, race, Charlson comorbidity score adapted for administrative data,25 number of physician visits, acute-care hospitalizations, number of different medications, osteoporosis diagnoses and medications, prior fractures, BMD testing, use of medications with fracture associations (eg, oral steroids, anticonvulsants, benzodiazepines, selective serotonin reuptake inhibitors, and proton pump inhibitors), diagnoses associated with falls (eg, Parkinson disease and Alzheimer disease), and prior history of falls.
Table 1. Baseline Characteristics of Subjects With Hypertension Included in Study Cohort
We examined the baseline characteristics of subjects starting the six antihypertensive classes of interest. Fracture outcomes were defined during the study follow-up period, and incidence rates with 95% confidence intervals (CIs) were calculated. Calcium channel blockers were chosen as the reference exposure, and Cox proportional hazards regression models were constructed to assess the relative risk of fracture associated with each of the antihypertensive classes of interest. Initially, unadjusted models were examined. Age and gender were added to the model next. Finally, we added all the potential confounders to the fully adjusted models. No selection process was applied because all covariates were hypothesized to be related to the outcome and possibly the exposure. Secondary analyses examined adjusted hazard ratios (HRs) by dosage and duration of antihypertensive agent use.
We identified 1,048,813 episodes of eligible antihypertensive drug use (monotherapy without evidence of recent antihypertensive drug use). From this pool of potentially eligible courses of antihypertensive drug use, episodes were excluded because of inconsistent health care system use (n = 560,747), initiation of multiple antihypertensive agents (n = 75,717), and age younger than 65 years or ineligibility at the index date (n = 36,288). This left a final study cohort of 376,061 patients (Fig. 1).
For each of the antihypertensive classes, large numbers of subjects were studied (Table 1). The median age of subjects was approximately 80 years, and the vast majority were white women. The antihypertensive class initiated varied by race, with 83% of calcium channel blockers versus 91% of beta blocker and loop diuretic users being white. The percentages of patients with a diagnosis of osteoporosis, use of a medication for osteoporosis, and prior fracture were similar across antihypertensive agent classes. Oral steroid use and a history of falls were more common among loop diuretic users than with other antihypertensive agent classes. In addition, loop diuretic users were hospitalized more frequently and filled prescriptions for more medications than users of other antihypertensive agent classes. Subjects were followed for a median of 70 days (interquartile range 45 to 177 days) after starting antihypertensive medication.
We observed large numbers of fractures during follow-up across all antihypertensive agent classes (Table 2). In this older, predominantly white female study population, the composite fracture rate was 35.2 per 1000 person-years (95% CI 34.4–36.1). The composite rate varied across antihypertensive agent classes, increased with age, and was about twice as high for women than for men. Incidence rates by age and gender were consistent with the overall population (Fig. 2), with increasing rates by age and for women.
Table 2. Fracture Rates per 1000 Person-Years (95% CI) by Anatomic Site
Calcium channel blocker
Note: Composite refers to hip, wrist, humerus, or pelvis fractures. Median follow-up time for the composite fracture outcome was calcium channel blockers: 90 days; angiotensin-receptor blockers: 89 days; ACE inhibitors: 86 days; loop diuretics: 60 days; beta blockers: 61 days; and thiazide diuretics: 76 days.
By anatomic site
The age- and gender-adjusted and fully adjusted Cox proportional hazards models gave very similar results (Table 3). Thiazide diuretics showed a significantly reduced risk compared with calcium channel blockers for the composite fracture outcome. Loop diuretics did not increase fracture risk, and beta blockers did not reduce risk. Angiotensin-receptor blocker use was associated with a large and significant reduction in fracture risk. These trends were observed across fractures at most anatomic sites but not all.
bFully adjusted model includes all variables included in Table 1.
Composite fracture outcome
Anatomic sites (fully adjusted)
The sensitivity analyses by dosage and duration agreed with the main findings (Fig. 3). Loop diuretic use showed a steady increase in fracture risk with increasing dosage, and thiazide diuretics showed a steady decline in fracture risk with increasing dosage. The effect of angiotensin-receptor blockers was inconsistent across dosage categories. Low- and medium-dosage beta blocker use showed a trend toward a reduced fracture risk, but high-dosage use was associated with an increased fracture risk. ACE inhibitors showed no trend over dosage categories. The duration analyses demonstrated that use of loop diuretics over 1 year was associated with an increased fracture risk, whereas shorter-term use was not. Thiazide diuretic use seemed to reduce fracture risk after day 90 but not before. The effect of angiotensin-receptor blockers appeared to strengthen over the first year of use but not thereafter.
Many antihypertensive medications have been linked to osteoporosis and fractures. Their pharmacologic mechanisms suggest a possible link, and substantial data support the association. However, the prior human research has several limitations, such as reliance on nonuser reference groups, mixing new and ongoing users, and limited confounder control. We studied fracture risk among a very large cohort of elderly Medicare beneficiaries in the United States and found that fracture risk varied significantly across antihypertensive agent categories. Angiotensin-receptor blockers and thiazide diuretics were consistently associated with reduced fracture risk compared with calcium channel blockers. Other agents were associated with no increase or decrease in fracture risk.
The strengths of our study include the definition of antihypertensive drug exposure. Our drug-exposure data are based on actual prescription-filling information rather than self-report. While we do not know if subjects actually used their filled perscriptions, these types of data generally are thought to be one of the strongest sources of drug information.26 The accuracy of these data is supported by the consistency of effect demonstrated with varying duration of prescriptions, as shown in Fig. 3B. We defined the index date at the beginning of antihypertensive agent use. Such a new-user design is preferred in pharmacoepidemiology because it allows one to better assess for early and/or late effects.21 In addition, we carefully defined our antihypertensive agent categories, excluding combination therapy, where it would be impossible to attribute fracture risk to one category or another.
There also are important limitations to our methods. Fractures were identified based on diagnosis and procedure codes, in the absence of medical records or radiographs. We employed algorithms that have high positive predictive values, but there is undoubtedly some degree of misclassification, albeit likely random. Similarly, covariate definitions were based on drug, diagnosis, and procedure codes. We did not have actual BMD data and are limited in historical information beyond 1 year prior to the start of follow-up. The study findings may not generalize beyond our cohort, which was 80% women, reflecting the composition of the pharmacy benefit programs from which our study cohort was drawn. Finally, the number of subjects in the longer-duration categories was limited, and thus the duration analyses should be considered exploratory.
The results showing a protective effect of thiazide diuretic agree with prior observational and experimental data. Prior data on loop diuretic and fracture risk have been mixed, and our results suggest several possible reasons. First, the loop diuretic users demonstrated many possible confounders at baseline. After adjusting for them, there was no overall increase in fracture risk. However, higher-dosage use and longer-duration use both demonstrated a trend toward an increased fracture risk with loop diuretics. Prior beta blocker studies in humans have shown mixed results, with about half showing a reduced risk and half showing no difference in risk.5–7, 13–19 We found no difference in fracture risk. Prior studies were inconsistent in their adjustment for confounders. None used a new-user design, and many that found a significant reduction in fracture risk compared beta blocker users with nonusers; a reference group of nonusers may not be appropriate.
The two most probable mechanisms by which antihypertensive medications increase fracture risk are (1) falls related to relative hypotension, which would be expected to be an immediate effect, and 2) direct effects on bone metabolism that decrease BMD over a more prolonged period. While the increased risk of fracture with higher doses of loop diuretics may suggest falls related to overly aggressive antihypertensive treatment, the analyses of duration of treatment show that risk is higher with longer-term treatment with loop diuretics. Accordingly, our findings are more consistent with an effect of loop diuretics on BMD over time. Conversely, the increasingly protective effect of angiotensin-receptor blockers over time may suggest an improvement in BMD, although the mechanism by which this might occur is not known.
We found no prior human studies of angiotensin-receptor blockers and BMD or fracture risk in the literature. However, there are data from a rat ovariectomy model where administration of angiotensin II increased the activity of tartrate-resistant acid phosphatase (TRACP), a marker of bone resorption. This increase in TRACP activity was accompanied by a significant decrease in BMD. In this same set of analyses, hypertensive rats treated with an angiotensin-receptor blocker experienced less decrease in BMD and less increase in TRACP activity and urinary deoxypyridinoline.27
In conclusion, we found a variation in fracture risk across new users of antihypertensive agents. Our findings were very similar to prior work showing a reduced fracture risk with thiazide diuretics. The findings related to loop diuretics and beta blockers agree with certain prior studies but not all. The fracture reduction associated with angiotensin-receptor blockers suggests a possible mechanism for affecting fracture risk. The strengthening of the effect with longer duration use may speak to a possible effect on BMD worth investigating. While the relative risks that we observed were small, antihypertensive agent use in older adults is widespread; thus even small differences in risk translate into many possible fractures. Clinicians and patients may want to take these possible differences in fracture risk into account when choosing between categories of medications.
All the authors state that they have no conflicts of interest.
DHS is supported by the NIH for work on osteoporosis (P60 AR 047782).