NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

Nelson HD, Haney EM, Chou R, et al. Screening for Osteoporosis: Systematic Review to Update the 2002 U.S. Preventive Services Task Force Recommendation [Internet]. Rockville (MD): Agency for Healthcare Research and Quality (US); 2010 Jul. (Evidence Syntheses, No. 77.)

Cover of Screening for Osteoporosis

Screening for Osteoporosis: Systematic Review to Update the 2002 U.S. Preventive Services Task Force Recommendation [Internet].

Show details


Key Questions 1 and 4. Does screening for osteoporosis and low bone density reduce osteoporosis-related fractures and/or fracture-related morbidity and mortality in postmenopausal women and men age >50 years? What are the harms associated with osteoporosis screening?


We identified no trials of the effectiveness of screening and no studies evaluating potential harms from screening. Adverse outcomes from medications are addressed in Key Question 6 below.

Key Question 2. What valid and reliable risk-assessment instruments stratify women and men into risk categories for osteoporosis or fractures?


Several risk-assessment instruments have been developed to identify individuals at risk for low bone density or fractures. Thirty-three studies evaluated 21 externally validated clinical risk-assessment instruments and reported performance estimates of the area under the curve (AUC) for the receiver-operating characteristic (ROC) curve predicting either bone density or fractures. Twenty-three studies of 14 instruments to predict low BMD (T-scores ≤−2.5) reported AUC estimates ranging from 0.13 to 0.87, with most between 0.60 and 0.80. Eleven studies of 11 instruments to predict fractures reported AUC estimates from 0.48 to 0.89. Additional studies combined a risk-assessment instrument with bone densitometry, quantitative ultrasound, or radiograph finding, usually resulting in higher AUC estimates than the individual components. Although some instruments had high AUC estimates in selected studies, none demonstrated high estimates in several studies. Instruments with fewer risk factors often did as well or better than those with more and none performed consistently better than the others. Few instruments have been validated in men. No studies are available that demonstrate improved fracture outcomes when using risk-assessment instruments in clinical practice to identify individuals for screening and treatment.

Detailed Findings

Sixty-four publications evaluated risk-assessment instruments to predict either BMD52–86 or fractures.74, 87–115 Ten studies assessed the performance of risk-assessment instruments in combination with peripheral bone mass measurements to predict DXA-measured BMD61, 67, 69, 73, 76, 93 or fractures,91, 95, 97, 101 and two studies evaluated prediction of DXA-measured BMD by dental radiographs.63, 68 Three additional studies evaluated the use of risk-assessment instruments in clinical settings by measuring referrals for DXA,116 initiation of treatment and rates of hip and total fractures,117 or comparing various screening strategies in predicting fracture risk.93

Several risk-assessment instruments have been externally validated (Table 2; Appendix Table D2). Others were developed for a single study and are either internally validated or non validated (Appendix Table D1 includes all validated and non validated risk-assessment instruments).

Table 2. Performance of Externally Validated Risk-Assessment Instruments That Report AUC.

Table 2

Performance of Externally Validated Risk-Assessment Instruments That Report AUC.

Risk-Assessment Instruments Predicting Bone Density

We identified 36 studies that reported the performance of various instruments to predict BMD T-score ≤−2.5, including 23 studies of 14 externally validated instruments that report AUC values for the ROC curve52–54, 56, 57, 60–62, 65–67, 69–74, 76–82, 85 and 13 studies evaluating instruments that were not externally validated or that did not report AUC values.55, 58, 59, 63, 64, 68, 75, 76, 78, 83, 84, 86 The AUC for the ROC curve for the externally validated instruments ranged from 0.13 to 0.87.

Instruments with fewer risk factors often had similar or higher AUC estimates as than those with more risk factors. For example, the Osteoporosis Self-assessment Screening Tool (OST) includes only age and weight, has similar AUC estimates as other more complicated instruments, and has been validated in both men52, 69 and women.61, 64, 66, 67, 70, 74, 76, 77, 85 A recent meta-analysis of OST in postmenopausal women evaluated its performance in ruling out osteoporosis (T-score ≤−2.5).118 In the combined analyses, the summary negative likelihood ratio for ruling out a T-score <−2.5 in white women was 0.19 at the femoral neck (seven studies) and 0.43 (five studies) at the lumbar spine. However, the meta-analysis was limited by including studies that were published only as abstracts,119, 120 using retrospective data collection, using non-representative study populations, reporting the number of participant withdrawals inadequately, and reporting uninterpretable test results.118

Evaluations of several instruments, including simple calculated osteoporosis risk estimation (SCORE), osteoporosis risk assessment instrument (ORAI), body weight criterion, and osteoporosis index of risk (OSIRIS), have been based on cross-sectional analyses of cohort data. For instruments that were evaluated prospectively, studies were limited by including small numbers of participants or participants recruited from specialty clinics. Five studies include men.52, 69, 81, 82, 116

Risk-Assessment Instruments Predicting Fracture

We identified 30 studies reporting the performance of risk-assessment instruments to predict fractures, including 11 studies of 11 externally validated instruments that report AUC for the ROC curve74, 88, 90, 96, 98, 100, 103, 104, 112, 113, 115 and 19 studies that either did not report the AUC value or evaluated instruments that were not externally validated.87, 89, 91–95, 97, 99, 101, 102, 105–111, 114 The AUC estimates for the studies of externally validated instruments ranged from 0.48 to 0.89.

Methodologic limitations of these studies are similar to those of the BMD risk-assessment instrument studies. Two studies were cross-sectional, evaluating prevalent fractures at the same time as risk factors.114, 115 One instrument was designed to assess subclinical vertebral fractures114 identifying risk for current rather than future fractures. Other studies used prospective cohort or randomized controlled trial study designs with prospective collection of fracture data reducing potential bias. For these studies, instruments were developed from risk factors assessed at baseline.

Six studies included men and women;90, 103, 104, 109, 111, 113 all others included women only. Three large studies evaluated the FRAX instrument,104 an instrument developed and validated within the Women’s Health Initiative (WHI) cohort,112 and another from the National Osteoporosis Risk Assessment (NORA) study population.108

The World Health Organization and National Osteoporosis Foundation recently developed the FRAX instrument to predict individual fracture risks.104, 121 FRAX estimates adjust for nationality and include femoral neck BMD if available and age, sex, height, body mass index (BMI), previous fracture, family history of fracture, glucocorticoid use, current smoking status, daily alcohol use of 3 units or more, rheumatoid arthritis, and other secondary causes (insulin dependent diabetes mellitus, osteogenesis imperfecta, untreated long-standing hyperthyroidism, hypogonadism or premature menopause [<45 years], chronic malnutrition or malabsorption, and chronic liver disease). FRAX was derived from combined data from 46,340 individuals from nine different cohorts in Europe, Canada, United States (Rochester, MN), and Japan; seven of the development cohorts included men.104 Linear regression modeling identified risk factors that were subsequently tested in 230,486 individuals from 11 validation cohorts; one cohort (Miyama) included men.104 While the risk calculator is available on a website (, the source code is not accessible.

The AUC estimates for FRAX ranged between 0.54 and 0.78 for osteoporotic fractures,98, 104, 113 and 0.65 and 0.81 for hip fractures.104 We did not identify studies that prospectively tested FRAX in clinic populations or determined its effectiveness in selecting patients for therapy.

Three studies compared FRAX with simple models, such as age and BMD or age and fracture history, and found the simple models performed as well as FRAX in predicting hip and other clinical fractures98, 110 and vertebral fractures.96 Among women enrolled in SOF with risk factor assessment at baseline and 10 years of follow-up, the AUC for hip fracture was 0.75 for FRAX with femoral neck BMD included, 0.71 for FRAX without femoral neck BMD, and 0.76 for age and femoral neck BMD alone.98 The same SOF data were used to evaluate FRAX across levels of BMD to predict hip fracture. The resulting AUCs were 0.79, 0.69, 0.59 for normal, low bone density, and osteoporosis (T-score <−2.5), respectively. For predicting nonvertebral fractures, the AUCs were 0.59, 0.58, and 0.63, respectively.122

The FRAX model was also evaluated using data from the placebo group of the Fracture Intervention Trial (FIT).96 This study compared AUCs for several combinations of risk factors including FRAX with and without femoral neck BMD. Results indicated that models using baseline vertebral fractures, age, and femoral neck BMD yielded the highest AUC (0.76). In comparison, FRAX yielded an AUC of 0.71 with femoral neck BMD included, and an AUC of 0.68 without femoral neck BMD.96

Use of Risk-Assessment Instruments in Clinical Practice

Three studies evaluated the use of risk-assessment instruments in clinical practice.93, 116, 117 Women randomly sampled from member lists of a health maintenance organization were randomized to one of three screening strategies involving use of BMD testing or evaluation by risk instruments followed by BMD testing if results indicated increased risk.117 The groups included: 1) universal screening (everyone offered DXA testing), 2) SCORE (invited for DXA testing only if the SCORE result was >7), and 3) SOF criteria (invited for DXA testing only if they had five or more hip fracture risk factors). DXA testing was performed in 100 percent of the universal group, 73.8 percent of the SCORE group, and 6.9 percent of the SOF group. Osteoporosis treatment rates did not differ between groups.117

In another study, a pre-post evaluation of a screening strategy to improve referral for DXA enrolled men attending a rheumatology clinic.116 They were evaluated with a SOF-based 10-item checklist. Prior to the checklist intervention, 14 percent of men over age 65 had a prior DXA (5 percent of black and 29 percent of white men), whereas after the checklist intervention 32 percent of the men had a DXA request (23 percent of black and 46 percent of white men).116

A third study used the EPIDOS prospective cohort to compare several screening strategies in order to predict fracture risk. Participants underwent either: 1) DXA; 2) QUS; 3) QUS followed by DXA if suggested by QUS results; 4) weight and DXA measurement for those <59 kg followed by clinical risk assessment for those in the low-medium BMD category; and 5) a combined strategy with weight and QUS measurement, then hip DXA, followed by a clinical evaluation. Sensitivity was highest for the combined strategy (53 percent versus 15–36 percent for the others), although specificity was similar (80 percent versus 86–95 percent for the others).93

Risk Factors in Combination with Bone Mass Measures

Several studies assessed QUS, central DXA, or peripheral DXA in combination with risk factors to predict either BMD or fracture. Generally, these studies found that QUS in combination with clinical risk factors, with or without DXA, improved identification of individuals with osteoporosis or fractures. The Osteoporosis Risk Assessment by Composite Linear Estimate (ORACLE) risk instrument (which includes QUS) was developed, validated, and compared to QUS alone, and to OST.76 Both QUS and ORACLE had higher AUC estimates (0.81 [SE, 0.030]) than ultrasonometric bone profile index (ultrasonometric bone profile index [UBPI], 0.71 [SE, 0.034]), or the ultrasound derived T-score (0.69 [SE, 0.035]).76 The use of the stiffness index by QUS in combination with risk factors yielded a higher AUC estimate than either QUS or the risk factors alone.101 Models including QUS plus other risk factors reported AUC estimates ranging from 0.672 to 0.689.95

Combing the OST risk-assessment instrument with QUS measurements improved the AUC estimate.69 In another study, risk factors in combination with BUA performed better than risk factors alone.73

In a study comparing two ultrasound systems, the CUBA Clinical BUA had an AUC estimate of 0.766 for predicting a T-score of ≤−2.5.61 This estimate was higher than the AUC for the Sunlight Omnisense system (separately or in combination; range, 0.582 to 0.698), for all clinical risk prediction instruments tested in this cohort (OSIRIS, Study of Osteoporosis Fractures–Study Utilizing Risk Factors [SOFSURF], ORAI, OST, SCORE, body weight [pBW]) (which ranged 0.664 to 0.747), and higher than the velocity of sound by QUS at the calcaneous (0.723).61

In a study comparing several different risk instruments with both QUS (CubaClinical and Achilles) and peripheral DXA (Peripheral Instantaneous X-ray Imager [PIXI]), PIXI had the highest independent AUC at 0.80.67 When combined with the risk instruments, PIXI + OSIRIS had an AUC of 0.82.67

Measures of hip geometry by DXA (hip strength analysis [HAS], hip axis length [HAL], and compressive stress [c-stress]) were also included in predictive models.91 Models including compressive stress plus age and BMI had higher AUC estimates than these variables alone (0.875) or for age plus femoral neck BMD (0.856). However, HAS has been less reliable and its reproducibility is lower than conventional DXA.91

Two studies evaluated the use of dental radiographs for predicting osteoporosis compared to DXA.63, 68 Among women ages 45–70 years, the AUC estimate for femoral neck BMD was 0.835 using manually initialized fit of mandibular radiographs, compared to 0.861 using ORAI and 0.732 using the National Osteoporosis Foundation (NOF) index.63 For prediction of osteoporosis at any of the three sites (total hip, femoral neck, and lumbar spine), the AUC estimate for manual reading of the dental radiographs was better than automated reading, and also better than either ORAI or the NOF index. The manual reading had 94 percent sensitivity but 29.5 percent specificity.63 A separate study reported wide variation in intraobserver assessments for both the lower and upper jaw periapical radiographs. Across all observers, the diagnostic odds ratios ranged from 2.76 to 7.71 for the upper jaw and 2.20 to 15.35 for the lower jaw.68

Key Question 3a. How well does DXA predict fractures in men?


Although DXA is the current gold standard for diagnosing osteoporosis and making treatment decisions, it is an imperfect predictor of fractures. Its role in predicting fractures in men has only recently been evaluated in large studies. The Rotterdam Study is a large population-based prospective study that includes men and women and reports incident vertebral and nonvertebral fractures several years after obtaining baseline DXA. In this study, for each standard deviation reduction in femoral neck BMD, the hazard ratio for various fracture outcomes was increased to similar levels for men and women. Additional studies of DXA in men are generally consistent with these findings, although DXA of the femoral neck was associated with a higher risk for hip fracture in men enrolled in Osteoporotic Fractures in Men Study (MrOS) compared with women in SOF.

Detailed Findings

Evaluations of DXA in predicting fractures in men, and comparing men with women, were reported from two large, good-quality prospective cohort studies.123–125 The Rotterdam Study compared women and men age 55 years or older from the same community at the same time.123, 124 This study utilized a prospective, population-based cohort to investigate the incidence of and risk factors for chronic diseases including osteoporosis. A total of 4,731 women and 3,075 men obtained baseline DXA measurements of the femoral neck, and 2,022 women and 1,527 men obtained baseline lateral radiographs of the thoracolumbar spine. Nonvertebral fracture outcomes were determined an average of 6.8 years later from fracture reports provided by physicians in the community using a computerized reporting system and from reviewing hospital records. Fractures were verified by research physicians using a standardized protocol. Incident vertebral fractures were evaluated 6.3 years after the baseline examination using follow-up radiographs. Vertebral fractures were diagnosed using morphometric criteria.

Age-adjusted hazard ratios for vertebral and nonvertebral incident fractures were similar for men and women. For each gender-specific standard deviation (SD) decrease in BMD, the hazard ratio for all nonvertebral fractures was 1.4 (95 percent confidence interval [95% CI], 1.2–1.6) for men and 1.5 (95% CI, 1.4–1.6) for women, and were similar for several site-specific fractures (Table 3).123, 124 The hazard ratio for vertebral fractures was 1.8 (95% CI, 1.3–2.4) for men and 1.9 (95% CI, 1.6–2.4) for women.

Table 3. Results of the Rotterdam Study of DXA and Fractures in Men and Women.

Table 3

Results of the Rotterdam Study of DXA and Fractures in Men and Women.

The Rotterdam Study also reported that the incidence rate for nonvertebral fractures was higher for women than men in all age groups, incidence rates increased with age for both men and women at all levels of BMD, and the relative risks for nonvertebral fractures were higher in lower BMD categories. However, despite the ability of BMD to predict fractures, subjects with normal BMD also incurred fractures at fairly high incidence rates (6.6 nonvertebral fractures/1,000 person years for men; 13.4 nonvertebral fractures/1,000 person years for women).123 These findings were similar for vertebral fractures, although the incidence of vertebral fractures was also higher in individuals with previous vertebral fractures.124

A study of BMD and risk for hip and nonvertebral fractures that compared men enrolled in MrOS with women in SOF reported similar results as the Rotterdam Study.125 However, in this study, DXA of the total hip or femoral neck was associated with a higher risk for hip fracture in men (femoral neck RH, 3.68 [95% CI, 2.68 to 5.05]) than women (femoral neck RH, 2.48 [95% CI, 2.09 to 2.95]). Subjects in MrOS and SOF were older than those in the Rotterdam Study, men and women were recruited from different geographic regions in the United States, and they were followed for approximately 4 years but at different times. Additional studies of the performance of DXA in predicting fractures in men are consistent with the findings of the Rotterdam Study and MrOS.126–128 Variations in estimates are likely due to the different patient populations enrolled in the studies, study designs, and other factors.

Key Question 3b. How well do peripheral bone measurement tests predict fractures?


Several peripheral bone measurement tests have been developed, although clinical practice and recent research focus on QUS of the calcaneous (heel). Large studies of postmenopausal women and men indicate that QUS obtained at the calcaneus using various types of devices can predict fractures as well as DXA of the femoral neck, hip, or spine, although variation exists across studies. However, QUS is not a good predictor of DXA as determined by a recent meta-analysis that indicated AUC estimates of 0.74–0.77 depending on the QUS parameter used. Also, it is unclear how results of QUS can be used to select individuals for drug therapies that were proven efficacious based on DXA criteria.

Detailed Findings

Postmenopausal Women

Several large studies evaluated the performance of various bone measurement tests in predicting fractures in women.129–135 Although results vary, overall, DXA and QUS have similar AUC estimates and odds ratios for fracture outcomes (Table 4). For all fractures combined, AUC estimates range from 0.59–0.66 and ORs from 1.81–2.16 for DXA of the femoral neck. For QUS, AUC estimates are approximately 0.60, and ORs range from 1.26–2.25. In one study that included DXA of the distal radius, the AUC estimate was 0.64 (95% CI, 0.59–0.68) and OR for all fractures 1.47 (95% CI, 1.28–1.68).132

Table 4. Recent Studies Comparing Performance of Bone Measurement Tests in Predicting Fractures.

Table 4

Recent Studies Comparing Performance of Bone Measurement Tests in Predicting Fractures.


Studies evaluating the performance of bone measurement tests in predicting fractures in men examined the same technologies used for women (Table 4).126–128, 131, 136 Results are similar for DXA and QUS. For hip fractures specifically, DXA of the femoral neck is associated with higher risk ratios than QUS for men and women in most studies.

QUS Compared to DXA

QUS predicts most fractures as well as DXA and offers distinct advantages, such as lower cost, portability, ease of use, and avoidance of ionizing radiation. However, it is not clear how to apply the results of QUS testing to patient management. Currently, standardized diagnostic criteria for osteoporosis uses DXA not QUS cutpoints, and clinical trials of drug therapies used DXA testing in its selection criteria. To be clinically useful, QUS results would need to be similar to DXA.

To address this issue, a systematic review and meta-analysis of the accuracy QUS compared to DXA in identifying patients with osteoporosis evaluated 25 studies published prior to October 2005.137 Included studies evaluated several parameters including BUA, SOS, QUI, and stiffness. Studies varied by subject characteristics, such as location (Europe, United States, Asia), sample size (110–722), prevalence of osteoporosis using DXA criteria (7–38 percent), age (46–64 years), and sex. No studies described the race or ethnicity of subjects. Studies also varied in their use of ultrasound devices, DXA references sites (lumbar spine, femoral neck, total hip), and reference populations to determine T-scores (manufacturers, national, local). All of these factors are important sources of heterogeneity. Potential sources of bias identified in the systematic review include insufficient information to determine participant selection methods, time between QUS and DXA, and whether QUS and DXA results were interpreted independently of each other.

Eleven studies in the systematic review contributed to a summary ROC curve for the QUS index parameter.137 Results for all studies indicated AUC 0.76 (95% CI, 0.72–0.79), and results specifically for postmenopausal women were AUC 0.75 (95% CI, 0.66–0.82). These results were similar for the other QUS parameters (broadband attenuation AUC, 0.77 [95% CI, 0.73–0.81]; SOS and VOS AUC, 0.74 [95% CI, 0.71–0.77]; and stiffness AUC, 0.79 [95% CI, 0.71–0.86]).

Summary estimates of the sensitivity and specificity for the QUS Index parameter indicated wide ranges of sensitivity and specificity at various T-score thresholds.137 For example, for the QUS index parameter T-score cutoff threshold of −1 that is commonly used in screening, sensitivity was 79 percent (95% CI, 69–86) and specificity was 58 percent (95% CI, 44–70) for identifying individuals with DXA T-scores ≤−2.5 at the hip or spine. These values changed at different cutoffs, but at no cutoff were the sensitivity and specificity both high.

Key Question 3c. What is the evidence to determine screening intervals for osteoporosis and low bone density?


In a large good-quality prospective cohort study of 4,124 women age ≥65 years from SOF, repeating a BMD measurement up to 8 years after an initial measurement did not significantly change AUC and risk ratio estimates for nonvertebral, hip, or vertebral fractures.138 No studies of screening intervals have been conducted in men or other groups of women.

Key Question 5. Do medications for osteoporosis and low bone density reduce osteoporosis-related fracture rates and/or fracture-related morbidity and mortality in the target populations?


For postmenopausal women without previous fractures, trials indicate that bisphosphonates, parathyroid hormone, raloxifene, and estrogen reduce primary vertebral fractures. Bisphosphonates reduce primary nonvertebral fractures in sensitivity analysis. No trials report effects on fracture-related morbidity and mortality. The only trial that stratified results according to baseline BMD reported reduced fractures only for women with baseline T-scores ≤−2.5.50

More trials have been published that focus on secondary prevention in postmenopausal women, and several systematic reviews and meta-analyses include both primary and secondary prevention trials. For secondary prevention in postmenopausal women, the bisphosphonates alendronate, etidronate, and risedronate are similarly effective at decreasing vertebral fractures compared to placebo. Alendronate and risedronate, but not etidronate, also reduce nonvertebral fractures including hip fractures. Evidence for the newer bisphosphonates zoledronic acid and ibandronate is consistent with evidence for the other bisphosphonates. Of the other medications, parathyroid hormone, calcitonin, and raloxifene reduce vertebral fractures, and parathyroid hormone reduces nonvertebral fractures.

For men, there are no primary prevention trials of bisphosphonates. Based on two secondary prevention trials, alendronate reduces the risk of vertebral fractures compared to placebo, but not nonvertebral fractures. A single trial of parathyroid hormone reported a trend towards decreased vertebral and nonvertebral fractures, but the number of fractures was small and results did not reach statistical significance. There were no trials of other agents with fracture outcomes in men. No trials report other fracture-related morbidity or mortality outcomes.

Detailed Findings

See Appendix D for detailed evidence, quality, and supplemental tables.

Primary Prevention Trials

Postmenopausal women

Bisphosphonates. Fifteen placebo-controlled RCTs of bisphosphonates met inclusion criteria (Table 5, Appendix Tables D3 and D4), including seven trials of alendronate,47, 50, 139–143 three etidronate,144–146 four risedronate,41, 147–149 and one zoledronic acid.150 Excluded trials are listed in Appendix Table D5. FIT met criteria for good-quality.50 Of 13 trials rated fair-quality, eight lacked information on randomization, allocation concealment, or outcomes blinding41, 142–144, 146, 148–150; and five trials did not report intention-to-treat analysis or blinding of providers.47, 139, 140, 145, 147 One poor-quality trial did not report blinding, intention-to-treat analysis, or attrition.141

Table 5. Placebo-controlled Primary Prevention Trials of Medications.

Table 5

Placebo-controlled Primary Prevention Trials of Medications.

In 11 trials, mean baseline femoral neck BMD (or total hip BMD if femoral neck BMD was not available) T-scores were −1.0 to −2.547, 50, 139–141, 143–145, 148–150; one trial enrolled women with T-scores <−2.541; and three trials enrolled women with T-scores >−1.0.142, 146, 147 Five trials excluded or did not enroll women with previous vertebral fractures50, 139, 140, 144, 150; two trials enrolled >20 percent of participants with previous vertebral fractures but reported results in the subgroup of women without prior fractures41, 47; and the remainder did not report the proportion of women with previous fractures. The mean age of participants was <65 years in all of the trials except FIT (mean age 68 years).50 FIT enrolled over 4,000 patients, followed them for four years, and was the only trial designed to evaluate fracture rates as a primary outcome.50 All but three other trials41, 47, 142 randomized fewer than 200 participants, followed them for 1–2 years, and evaluated change in BMD as the primary outcome.

Rates of new vertebral fractures ranged from 0 to 24 percent for bisphosphonates and from 0 to 28 percent for placebo in 12 trials reporting this outcome (Table 5).47, 50, 139–142, 144–150 Rates of fractures may have varied because of differences in baseline BMD, other risk factors for osteoporotic fractures, duration of follow-up, and methods used to identify new fractures (e.g., actively soliciting symptoms and/or routine x-rays versus symptomatic or passive reporting only). Six trials reported no vertebral fractures in either bisphosphonate- or placebo-treated patients139, 140, 142, 144, 149, 150; and three of these trials identified new vertebral fractures clinically (i.e., did not perform routine spine radiography to identify fractures), potentially missing asymptomatic fractures.139, 149, 150

Bisphosphonates reduced vertebral fractures compared with placebo (relative risk [RR], 0.66 [95% CI, 0.50–0.89]; I2, 0 percent; seven trials) (Table 6, Appendix Figure C1).47, 50, 141, 145–148 Five trials recorded zero vertebral fractures and did not contribute to the pooled estimate in the primary analysis.139, 140, 142, 144, 149, 150 Excluding one trial that identified only one new clinical vertebral fracture and did not perform routine spine radiography to identify additional fractures did not change results.146 Results based on alternative methods for pooling were nearly identical (Table 7). FIT, the large (n=4,432) 4-year trial of alendronate, contributed two-thirds of the total number of patients (n=6,782) and vertebral fractures (169) in the analysis (RR, 0.55 [95% CI, 0.38–0.80]).50 Subgroup analyses of the other individual bisphosphonates evaluated in these trials (etidronate, risedronate, or zoledronic acid) were limited by small numbers of fractures (range, 0 to 20 events) for drugs other than alendronate. Removing the poor-quality trial did not significantly change estimates.141 Including all trials, the absolute risk for vertebral fracture was 1.9 percent for bisphosphonates compared to 3.1 percent for placebo. Based on FIT alone, the number needed to treat (NNT) was 60 to prevent one or more vertebral fractures (3.8 versus 2.1 percent).

Table 6. Fracture Outcomes of Placebo-controlled Primary Prevention Trials.

Table 6

Fracture Outcomes of Placebo-controlled Primary Prevention Trials.

Table 7. Sensitivity Analysis for Trials With Few, Rare, or Zero Fracture Events.

Table 7

Sensitivity Analysis for Trials With Few, Rare, or Zero Fracture Events.

Total nonvertebral fractures were reported in 10 trials.50, 139, 142, 143, 145–150 Rates of any fracture (vertebral or nonvertebral) could be estimated from nine trials, though in most cases we had to assume that fractures at different sites occurred in different patients.50, 139, 142, 145–150 One trial reported no fractures with either alendronate or placebo.139 In the other trials, nonvertebral fracture rates ranged from 0 to 12 percent for subjects randomized to bisphosphonates and 2 to 13 percent for those randomized to placebo. Similar ranges were observed for rates of any fracture.

For total nonvertebral fractures, a pooled analysis of trials indicated no statistically significant effects for bisphosphonates compared with placebo (RR, 0.83 [95% CI, 0.64–1.08]; I2, 15 percent; nine trials), although trends favored the bisphosphonates (Table 6, Appendix Figure C2).50, 142, 143, 145–150 Differences were also not significant for alendronate specifically (RR, 1.08 [95% CI, 0.62–1.88]; I2, 67 percent; two trials).50, 142 Subgroup analyses of other bisphosphonates were limited by small numbers of fractures (range, 5 to 18 events). One trial recorded zero nonvertebral fractures and did not contribute to the primary analysis.139 Results were statistically significant when estimated using alternative pooling methods (Peto OR, 0.84 [95% CI, 0.72–0.98]; fixed effects Mantel Haenszel with inverse sample size continuity correction RR, 0.86 [95% CI, 0.74–0.99]) (Table 7). For any type of fracture (vertebral and nonvertebral), results were similar (RR, 0.89 [95% CI, 0.77–1.03]; I2, 0 percent; eight trials) (Appendix Figure C3).50, 142, 145–150 As in the analysis of vertebral fractures, FIT heavily influenced results (RR for nonvertebral fractures, 0.89 [95% CI, 0.76–1.04]; RR for any type of fracture, 1.08 [95% CI, 0.62–1.88]).50 Results for hip, wrist, or ankle fractures showed no statistically significant differences between bisphosphonates and placebo, but were limited by small numbers of fractures (Table 6, Appendix Figures C4, C5, and C6).

For the sensitivity analysis based on a broader definition for primary prevention, we added five trials that enrolled up to 40 percent of patients with baseline vertebral compression fractures38, 40, 45, 47, 48 and one trial that enrolled patients with a mean baseline BMD T-score of −4.3 (baseline fractures not reported).46 Estimates for vertebral fracture were similar to the primary analysis, and the estimate for hip fracture remained statistically non-significant (Appendix Table D6 and Appendix Figures C7 and C8). Although the result for hip fractures neared statistical significance (RR 0.65 [95% CI, 0.42–1.01]), only five additional hip fractures were included in the sensitivity analysis.40, 47 The point estimate for total nonvertebral fractures also remained similar, but reached statistical significance with the inclusion of the additional trials (RR, 0.82 [95% CI, 0.69–0.96]; I2, 5 percent; 14 trials) (Appendix Figure C9).38, 40, 45–48, 50, 142, 145–150 This was primarily due to the addition to the analysis of a large trial (83 of the 136 additional events in the sensitivity analysis were reported by this trial) with a vertebral fracture prevalence just over our threshold for inclusion as a primary prevention trial (21 percent).47 A sensitivity analysis that only added this trial would have resulted in borderline statistical significance (RR, 0.84 [95% CI, 0.70–1.00]). We could not adequately assess whether estimates of bisphosphonates for fracture efficacy varied between trials according to the mean baseline BMD of participants. For vertebral fracture, bisphosphonates were only superior to placebo in the subgroup of trials that enrolled patients with a mean femoral BMD T-score of −2.0 or worse (RR, 0.55 [95% CI, 0.38–0.80]), but this estimate is based solely on FIT50 (Appendix Figure C10). There was no difference between bisphosphonates and placebo in seven trials that enrolled patients with mean femoral BMD T-score of −1.0 to −2.0 (RR, 0.93 [95% CI, 0.49–1.76]), but only 28 vertebral fractures were reported in three trials.141, 145, 148, 149 For all nonvertebral fractures, there was no difference between bisphosphonates and placebo for any subgroup of trials stratified according to mean femoral BMD T-score (Appendix Figure C11). Hip fractures were only reported in three trials that each enrolled patients with mean femoral BMD T-score of −2.0 or worse.41, 50, 143

FIT was the only individual trial to report results stratified according to baseline BMD.50 It found that alendronate was associated with decreased risk of any clinical fracture (RR, 0.64 [95% CI, 0.50–0.82]) and vertebral fracture (RR, 0.50 [95% CI, 0.31–0.82]) in women with baseline femoral neck T-scores <−2.5, with a NNT of about 15 and 34, respectively. In women with T-scores between −1.6 and −2.0 or −2.0 and −2.5, there was a non-statistically significant trend towards decreased risk of vertebral fracture (RR, 0.82 [95% CI, 0.33–2.07] and RR, 0.54 [95% CI, 0.28–1.04], respectively), but no effect on any clinical fracture (RR, 1.14 [95% CI, 0.82–1.60] and RR, 1.03 [95% CI, 0.77–1.39], respectively).

Parathyroid hormone. One large, fair-quality (n=2,532) RCT evaluated effects of parathyroid hormone on risk of fractures after 18 months in postmenopausal women with BMD T-score <−3.0 and no prevalent vertebral fractures (81 percent of participants), or a T-score <−2.5 and one to four prevalent fractures (19 percent) (Table 5).151 For women without a baseline fracture, parathyroid hormone decreased the risk of new vertebral fractures from 2.1 to 0.7 percent (RR, 0.32 [95% CI, 0.14–0.75]) with a NNT of 71 (42 to 248). Among all participants, there was no difference in risk of new nonvertebral fracture (RR, 0.97 [95% CI, 0.71–1.33]).

Testosterone and calcitonin. We identified no trials that evaluated efficacy of testosterone or calcitonin for primary prevention of fractures.

Raloxifene. The Multiple Outcomes of Raloxifene (MORE) trial included women with BMD T-scores <−2.5 with or without previous vertebral fractures (37 percent with prior fractures).152 Raloxifene reduced vertebral fractures (RR, 0.60 [95% CI, 0.53–0.69]), but not nonvertebral or hip fractures compared to placebo (Table 5).152 Results were similar for women with and without prior vertebral fractures and for women using two different doses of raloxifene (60 or 120 mg/day).152, 153

The Raloxifene Use for the Heart (RUTH) trial was designed primarily to determine the effects of raloxifene on coronary events and invasive breast cancer, and fractures were secondary outcomes (Table 5).154 Participants were selected for these trials based on cardiac risk factors rather than BMD or fracture status. RUTH reported reduced clinical vertebral fractures (RR, 0.65 [95% CI, 0.47–0.89]), but not nonvertebral fractures (RR, 0.96 [95% CI, 0.84–1.09]) among raloxifene users compared to placebo, consistent with results of MORE.154 A meta-analysis of both trials provided estimates for vertebral (RR, 0.61 [95% CI, 0.54–0.69)] and nonvertebral fractures (RR, 0.97 [95% CI, 0.87–1.09]) (Table 6).155, 156

Estrogen with and without progestin. The WHI trial is the largest prevention trial of estrogen (conjugated equine estrogen [CEE]) with and without progestin (medroxyprogesterone acetate [MPA]) reporting fracture outcomes in postmenopausal women. The estrogen with progestin trial reported reduced risks for clinical vertebral (RR, 0.65 [95% CI, 0.46–0.92]), hip (RR, 0.67 [95% CI, 0.47–0.96]), wrist (RR, 0.71 [95% CI, 0.59–0.85]), and all fractures combined (RR, 0.76 [95% CI, 0.69–0.83]) for estrogen with progestin users compared to placebo (Table 6).157 These results are statistically significant when using the nominal confidence intervals (nCI), but are not significant when using adjusted confidence intervals (aCI) (hip fracture RR, 0.67 [95% aCI, 0.41–1.10]).

All women in the estrogen only WHI trial had prior hysterectomies and differed from women in the estrogen with progestin trial by a number of other characteristics.158 These subject differences compromise direct comparisons between trials, although fracture outcomes are similar. Women using estrogen had reduced risks compared to placebo for clinical vertebral (RR, 0.62 [95% nCI, 0.42–0.93; 95% aCI, 0.34–1.13]), hip (RR, 0.61 [95% nCI, 0.41–0.91; 95% aCI, 0.33–1.11]), and all fractures combined (RR, 0.70 [95% nCI, 0.63–0.79; 95% aCI, 0.59–0.83]) (Table 6).158 Significance levels vary, however, depending on whether nominal or adjusted approaches are used.


The only primary prevention trial for men evaluated parathyroid hormone; we identified no trials of bisphosphonates, calcitonin, testosterone, or other agents.

Parathyroid hormone. A good-quality randomized, placebo-controlled trial evaluated effects of parathyroid hormone on risk of fractures after 11 months in men with osteoporosis (baseline BMD lumbar spine T-scores, −2.0 to −2.4) (Table 6).159 Results indicated a trend towards reduced risk of vertebral (RR, 0.49 [95% CI, 0.22–1.09]) and nonvertebral (RR, 0.51 [95% CI, 0.10–2.48]) fractures with parathyroid hormone, but the number of fractures was small and results did not reach statistical significance.159, 160

Systematic Reviews of Primary and Secondary Prevention Trials

Several existing systematic reviews of osteoporosis treatments include analyses that pooled results of primary and secondary prevention trials as well as results for men and women. Such evidence may not be fully applicable to screening for primary prevention of osteoporotic fractures in individuals without prior fractures, but may help inform estimates of treatment efficacy.

Bisphosphonates. We identified three good-quality161–163 and one fair-quality164 systematic reviews on effects of bisphosphonates on fractures (Table 8). All of the systematic reviews included trials enrolling patients with previous vertebral or nonvertebral fractures. Three of the systematic reviews classified trials that enrolled patients with a BMD T-score <−2.0 to be “secondary prevention” trials even if patients had no prior fracture (i.e., they used a more restrictive definition for primary prevention than we did).161–163 Most of the trials were not designed with sufficient statistical power to assess fracture rates as a primary outcome.

Table 8. Summary of Fracture Risks From Published Meta-analyses of Primary and Secondary Prevention Trials of Bisphosphonates.

Table 8

Summary of Fracture Risks From Published Meta-analyses of Primary and Secondary Prevention Trials of Bisphosphonates.

Three systematic reviews of alendronate,162 etidronate,163 and risedronate161 in postmenopausal women each found the bisphosphonate associated with a statistically significant decreased risk of vertebral fracture compared to placebo (Table 8, Appendix Tables D7 and D8). Relative risk point estimates ranged from 0.55 to 0.63. Statistically significant but smaller effects on nonvertebral and hip fracture were observed with alendronate (RR, 0.84 [95% CI, 0.74–0.94] and RR, 0.61 [95% CI, 0.40–0.92], respectively) and risedronate (RR, 0.80 [95% CI, 0.72–0.90] and RR, 0.74 [95% CI, 0.59–0.94], respectively), but not etidronate.

A fourth systematic review focused on effects of alendronate in men with osteoporosis (about half with vertebral fractures at baseline).164 In two trials (n=375),165, 166 alendronate was associated with a decreased risk of vertebral fractures (OR, 0.35 [95% CI, 0.17–0.77]) and a non-statistically significant trend towards decreased risk of nonvertebral fractures (OR, 0.73 [95% CI, 0.32–1.67]). We found similar results based on relative risk estimates (rather than odds ratios) using a random effects model (RR, 0.41 [95% CI, 0.21–0.80] for vertebral fracture and RR, 0.75 [95% CI, 0.35–1.60] for nonvertebral fracture) (Appendix Figures C12 and C13). These estimates are consistent with those observed in the systematic review of alendronate for postmenopausal women.162

Two large, placebo-controlled trials evaluated effects of ibandronate on fractures in postmenopausal women.167, 168 One trial (n=2,862; 54 percent with prior vertebral fracture) found that relatively low-dose intravenous ibandronate had no statistically significant effect on fracture risk.168 After three years, rates of vertebral fractures were 9.2 percent for intravenous ibandronate 1 mg every 3 months, 8.7 percent for 0.5 mg every 3 months, and 10.7 percent for placebo. Rates of any clinical fracture were 10.8 percent, 10.2 percent, and 12.6 percent, respectively. The second trial (n=2,946; all with prior vertebral fractures) found relatively higher oral doses of ibandronate associated with a statistically significant, approximately 50 percent reduction in risk of vertebral fractures, but had no statistically significant effect on the rate of any clinical osteoporotic fracture or clinical nonvertebral fracture.167 Rates of all new vertebral fractures were 4.7 percent for oral ibandronate 2.5 mg daily, 4.9 percent for 20 mg every other day for 12 doses each month, and 9.6 percent for placebo, and rates of acute clinical vertebral fractures were 5.1 percent, 5.8 percent, and 10.4 percent, respectively. We excluded a meta-analysis of individual patient data from four large (n=8,710) Phase III trials,167–172 including the two placebo-controlled trials,167, 168 because it pooled data across placebo-and active-controlled trials, did not report search methods, and failed to assess quality of included trials.173

Zoledronic acid. Two large, placebo-controlled trials evaluated effects of zoledronic acid on risk of new fractures in postmenopausal women (n=3,889; two-thirds with baseline vertebral fracture)174 and in women (75 percent) or men (25 percent) following a hip fracture (n=1,065).175 Both found that zoledronic acid reduced the risk of vertebral fracture (RR, 0.30 [95% CI, 0.24–0.38] and hazard ratio [HR], 0.54 [95% CI, 0.32–0.92], respectively), nonvertebral fracture (HR, 0.75 [95% CI, 0.64–0.87] and HR, 0.73 [95% CI, 0.55–0.98], respectively), and hip fracture (HR, 0.59 [95% CI, 0.42–0.83] and HR, 0.70 [95% CI, 0.41–1.19]) compared to placebo.

Calcitonin. A fair-quality systematic review found calcitonin for postmenopausal osteoporosis significantly reduced the risk of vertebral fracture risk compared to placebo (RR, 0.46 [95% CI, 0.25–0.87]).176 Although the pooled estimate was based on data from four trials,177–180 one trial (the Prevent Recurrence of Osteoporotic Fractures [PROOF] trial) contributed 1,108 of the 1,404 patients included in the analysis.177 Estimates of treatment benefit were less pronounced in the PROOF trial (RR, 0.79 [95% CI, 0.62–1.00]) compared to the pooled estimate. Effects of calcitonin on nonvertebral fractures were not statistically significant (RR, 0.52 [95% CI, 0.22–1.23]; three trials177, 179, 181). The trials included in the pooled analyses had methodological shortcomings, including high loss to follow-up, which ranged from 18.7 to 59.3 percent (in PROOF).

Parathyroid hormone. A good-quality systematic review found parathyroid hormone to be associated with a significant reduction in both vertebral (RR, 0.37 [95% CI, 0.28–0.47]; four trials160, 182–184) and nonvertebral (RR, 0.62 [95% CI, 0.46–0.82]; two trials159, 184) fractures compared to placebo in men or women.185 Only one of the four trials scored 4 or higher on the 5-point Jadad scale.159

In the two trials that evaluated women, we calculated estimates for vertebral (RR, 0.35 [95% CI, 0.25–0.47]; I2=0; two trials182, 184) and nonvertebral fractures (RR, 0.60 [95% CI, 0.43–0.85]; one trial184) that were very similar to estimates based on all trials (Appendix Figures C14 and C15). One of the two trials that evaluated men was very small (n=18) and did not contribute significantly to results.183 The other trial (n=437) is described in the section on primary prevention studies.

Testosterone. A good-quality systematic review identified no trials of testosterone therapy that reported fracture outcomes.186 We found no relevant trials of testosterone therapy not included in the systematic review.

Relative effectiveness of osteoporosis drugs. A fair-quality systematic review found no differences in fracture outcomes in trials comparing bisphosphonates versus estrogen (six trials), bisphosphonates versus parathyroid hormone (one trial), or bisphosphonates versus SERMs (three trials).187 Estimates from all of the head-to-head trials were imprecise, because none of the head-to-head trials were large enough to evaluate fracture rates as a primary outcome. A large (n=43,135), good-quality cohort study based on administrative claims data found no differences in nonvertebral fractures between risedronate, raloxifene, and alendronate users.188 Patients who received calcitonin experienced more nonvertebral fractures than those who received alendronate (HR, 1.40 [95% CI, 1.20–1.63]). In the subgroup of patients with a fracture history, raloxifene recipients experienced more nonvertebral fractures than alendronate recipients (HR, 1.78 [95% CI, 1.20–2.63]).

Key Question 6. What are the harms associated with medications for osteoporosis and low bone density?


A summary of evidence for major adverse outcomes of medications based on published, randomized placebo-controlled trials and systematic reviews is described in Table 9.

Table 9. Adverse Health Outcomes From Medication Studies.

Table 9

Adverse Health Outcomes From Medication Studies.

Evidence from good-quality systematic reviews of alendronate,162 etidronate,163 and risedronate,161 and large trials of ibandronate and zoledronic acid found no differences between any bisphosphonate and placebo in rates of withdrawal or withdrawals due to adverse events. There are case reports of serious upper gastrointestinal adverse events such as perforations, ulcers, bleeds, esophagitis, or esophageal ulceration with all bisphosphonates, but there is no clear increased risk when compared to placebo, given that they are taken in accordance with current recommendations to prevent esophagitis. Evidence on risk of atrial fibrillation with bisphosphonates is mixed, with some studies showing increased risk174, 189 and other showing no increased risk.175, 190, 191 A review by the FDA on atrial fibrillation risk is ongoing, but found no evidence of an increased risk from placebo-controlled trials.192 There are case reports of osteonecrosis of the jaw in patients taking bisphosphonates for osteoporosis, primarily in individuals with cancer receiving intravenous doses higher than that used for osteoporosis treatment or prevention.193 Although the incidence appears to be very low, there is no reliable evidence for estimating the incidence of osteonecrosis. There are also case reports of severe musculoskeletal symptoms with all of the bisphosphonates; atypical, low-energy fractures of the femoral diaphysis in long-term users of alendronate; and esophageal adenocarcinoma.

Evidence on harms associated with calcitonin, parathyroid hormone, and testosterone for treatment of osteoporosis is extremely limited due to sparse data from relatively small numbers of trials and inconsistent reporting of adverse events.

Raloxifene users have more thromboembolic events compared to placebo. Estrogen with progestin increases thromboembolic events, stroke, coronary heart disease among older users, and breast cancer. Estrogen alone increases thromboembolic events and stroke.

Detailed Findings

Interpreting evidence on harms is challenging because of differences in how assiduously adverse events were sought, differences in how adverse events were defined, and because many trials did not report specific adverse events of interest. We included evidence on adverse events from studies of both primary and secondary prevention.


Overall withdrawals and withdrawals due to adverse events

Three good-quality systematic reviews found no differences between alendronate,162 etidronate,163 and risedronate161 versus placebo in rates of overall withdrawals or withdrawals due to adverse events. There was also no difference between zoledronic acid and placebo in overall withdrawals or withdrawal due to adverse events in two large pivotal trials,174, 175 or between ibandronate and placebo in three large trials.168, 194, 195

Gastrointestinal adverse events

A systematic review found etidronate and pamidronate associated with an increased risk of mild upper gastrointestinal (GI) events (acid reflux, esophageal irritation, nausea, vomiting, and heartburn) compared to placebo (OR, 1.33 [95% CI, 1.21–1.46]; 18 studies, and OR, 3.14 [95% CI, 1.93–5.21]; seven studies, respectively).187 A number of the etidronate and pamidronate studies that showed increased risk of GI events were older studies, when clinical awareness of methods for administering bisphosphonates to reduce GI adverse effects may have been limited. The systematic review found no differences between alendronate, ibandronate, risedronate, or zoledronic acid compared to placebo in risk of mild upper GI events.

Esophageal ulcerations and other serious upper gastrointestinal complications have been reported with all bisphosphonates. For example, a postmarketing surveillance study published in 1996, before preventive dosing measures were widely instituted for bisphosphonates, reported serious or severe esophageal adverse events in 51 of 470,000 patients who received alendronate.196 The systematic review187 found etidronate associated with higher odds of perforations, ulcerations, and bleeds compared to placebo or non-use of etidronate in three studies (OR, 1.32 [95% CI, 1.04–1.67]), and a higher risk of esophageal ulceration in one study (OR, 0.33 [95% CI, 0.14–0.74]). However, almost all of the data (371 of 373 total cases of esophagitis/esophageal ulcers or peptic ulcers) on serious GI events associated with etidronate came from one large (n=24,000) postmarketing cohort study.197 In this study, etidronate was associated with an increased risk of serious GI adverse events only when the control group included individuals both with and without osteoporosis. When the control group was restricted to individuals with osteoporosis not taking a bisphosphonate, cyclical etidronate was not associated with a higher risk of esophagitis/esophageal ulcers (1.2 versus 1.2 percent) or peptic ulcers (0.7 versus 0.7 percent).197

No other bisphosphonate was associated with a higher rate of esophageal ulcerations or other serious upper GI complications compared to placebo.187, 198 The systematic review found daily ibandronate to be associated with a lower rate of perforations, ulcers, and bleeds compared to placebo in two trials.187 However, the estimate was primarily based on a single trial that reported nearly all of the events, and the overall number of events was low (10 cases of duodenal ulcer in nearly 2,000 patients randomized to ibandronate 2.5 mg daily or placebo).194

The FDA recently issued a report summarizing 54 cases of esophageal adenocarcinoma associated with bisphosphonate (primarily alendronate) use, and called for studies investigating a possible association.199

Cardiovascular adverse events

The large (n=7,714) Health Outcomes and Reduced Incidence with Zoledronic Acid Once Yearly [HORIZON] Pivotal Fracture Trial of once-yearly zoledronic acid for postmenopausal osteoporosis reported an increased risk of serious atrial fibrillation compared to placebo, with an absolute increased risk of 0.8 percent (1.3 percent or 40/4,862 versus 0.5 percent or 20/3,852; p<0.001), but not an increased risk of any (serious or non-serious) atrial fibrillation (2.4 percent versus 1.9 percent; p=0.12).174 The smaller HORIZON Recurrent Fracture Trial did not find zoledronic acid associated with increased risk of either serious (1.1 percent or 12/1,054 versus 1.3 percent or 14/1,057; p=0.84) or any (2.8 percent or 29/1,054 versus 2.6 percent or 27/1,057) atrial fibrillation.175 Following publication of the HORIZON trials, the authors of the FIT trial (n=6,459) pointed out in a letter to the editor that data submitted to the FDA (but not reported in the journal publication of FIT) showed alendronate to be associated with a non-statistically significant trend towards increased risk for serious atrial fibrillation (1.5 percent versus 1.0 percent; HR, 1.51 [95% CI, 0.97–2.40]), although, as in the HORIZON Pivotal Fracture Trial, there was no difference in risk of any atrial fibrillation (HR, 1.14 [95% CI, 0.83–1.57]).200 The HORIZON and FIT trials used blinded adjudication to verify potential cases of atrial fibrillation. A pooled analysis of five trials found risedronate 2.5 mg or 5 mg associated with a similar risk of non-adjudicated serious or any atrial fibrillation compared to placebo (0.5 percent or 24/4,998 versus 0.6 percent or 29/5,020 versus 0.5 percent or 24/5,048; p=0.49 for serious atrial fibrillation; and 1.3 percent or 66/4,998 versus 1.4 percent or 70/5,020 versus 1.4 percent or 70/5,048; p=1.0).190 The quality of this analysis is difficult to assess because the data are presented as a letter to the editor, with no description of the methods used.

Two population-based case-control studies reached conflicting conclusions regarding the association between bisphosphonate use in women and atrial fibrillation.189, 191 The larger of the two studies (13,586 cases and 68,054 controls in Denmark) found no association between current or former bisphosphonate use (primarily etidronate and alendronate) versus no use (adjusted RR, 0.95 [95% CI, 0.84–1.07] and 1.04 [95% CI, 0.90–1.21], respectively).191 A smaller Washington state study (719 cases and 966 controls) found any use (past or current) of alendronate associated with an increased risk of atrial fibrillation compared to no use (OR, 1.86 [95% CI, 1.09–3.15]).189 This study identified and verified atrial fibrillation and other variables by review of clinical records, supplemented by patient interviews. The Danish study relied on information available from administrative databases (e.g., discharge diagnoses of atrial fibrillation and other medical conditions). The studies also differed in terms of which variables were adjusted for in the analysis. The Washington state study adjusted for age, treated hypertension, calendar year, and the diagnostic of osteoporosis and any cardiovascular disease, and the Danish study adjusted for age, presence of various hospital diagnoses, use of various drugs, and diagnosis of alcoholism or acute alcohol intoxication.

The FDA issued an interim report of an ongoing review on risk of atrial fibrillation associated with bisphosphonates in November 2008.192 Based on data from nearly 20,000 patients treated with bisphosphonates in placebo-controlled trials, it found no clear association between bisphosphonate exposure and the rate of serious or non-serious atrial fibrillation. The absolute difference in event rates between each of the bisphosphonates and placebo arms varied from 0 to 3 per 1,000.

Musculoskeletal adverse events

A systematic review found zoledronic acid associated with a higher odds of musculoskeletal events (muscular and joint pain, arthritis, and muscle cramps) compared to placebo (OR, 4.52 [95% CI, 3.48–5.43]; three trials).187 Risedronate was associated with a lower odds of musculoskeletal events compared to placebo (OR, 0.40 [95% CI, 0.29–0.54]; nine trials). Most of the nine trials included in this analysis enrolled patients with secondary osteoporosis or with a previous fracture. However, three trials included at least some patients with primary osteoporosis.148, 166, 201 One of these trials found a significant improvement in severity of back pain among risedronate patients relative to placebo,166 but there were no differences in incidence of musculoskeletal pain between risedronate and placebo in the other two trials.148, 201 Case reports of atypical, low-energy fractures of the femoral diaphysis in long-term users of alendronate have also been reported, though the incidence is unknown.202–204 There are case reports of severe musculoskeletal pain with all bisphosphonates, including risedronate, that may be reversible after discontinuing the medication.


A FDA report summarized data from 151 case reports of osteonecrosis of the jaw through 2003.193 The vast majority (139 cases) occurred in cancer patients who received high-dose intravenous pamidronate or zoledronic acid. Only 12 cases were reported in patients who received alendronate for osteoporosis. No evidence exists to reliably estimate the incidence of osteonecrosis in patients taking standard doses of bisphosphonates for osteoporosis. The HORIZON Pivotal Fracture Trial (n=7,714) identified one case of possible osteonecrosis of the jaw in patients receiving intravenous zoledronic acid and in one patient receiving placebo, based on pre-defined criteria (exposed bone in the maxillofacial area with delayed healing for more than six weeks despite appropriate care) applied by an independent, blinded adjudication committee.205 Osteonecrosis was not evaluated or reported in other trials of bisphosphonates.


A systematic review identified five large studies of administrative databases that found that adherence rates were about 10 percent higher with weekly compared to daily bisphosphonates.187 Even with weekly bisphosphonates, adherence rates range from 45 to 69 percent. Three other studies included in the systematic review found that rates of fracture prevention consistently correlated with levels of adherence to therapy.

Calcitonin, Parathyroid Hormone, and Testosterone

Evidence on harms associated with calcitonin, parathyroid hormone, and testosterone for treatment of osteoporosis is limited by relatively small numbers of trials and inconsistent reporting of adverse events. A systematic review found that calcitonin did not increase risk of acute coronary syndrome compared to placebo (OR, 0.98 [95% CI, 0.07–13.7]; three trials).187 It also found that calcitonin, testosterone, and parathyroid hormone were not associated with increased risk of cancer, although estimates were very imprecise. Neither calcitonin nor parathyroid hormone was associated with increased risk of mild gastrointestinal events. No evidence exists to estimate risk of serious gastrointestinal events.


A meta-analysis of trials of raloxifene reports statistically significant elevated risks for thromboembolic events (RR, 1.60 [95% CI, 1.15–2.23]; two trials)155, 156 (Table 9). Risks for coronary heart disease, stroke, endometrial cancer, and all cause death are similar for raloxifene and placebo.155, 156 Raloxifene significantly reduces risk for invasive breast cancer in women without preexisting breast cancer (RR, 0.53 [95% CI, 0.34–0.84]; two trials).155, 156 Several additional symptoms are associated with raloxifene use including, most commonly, influenza syndrome, leg cramps, peripheral edema, and hot flashes.152–154


The WHI primary prevention trial provides the most complete data about adverse outcomes of estrogen with and without concurrent use of progestin compared to placebo. Results have been reported in numerous publications since the main trial results were released in 2002.206 Coronary heart disease and breast cancer were the main outcome measures of the WHI, and the estrogen with progestin trial was discontinued early when safety parameters for breast cancer were exceeded in the treatment group (HR, 1.24 [95% CI, 1.01–1.54])207 (Table 9). Coronary heart disease events were also increased in the estrogen with progestin trial (HR, 1.24 [95% CI, 1.00–1.54]).208 However, secondary analysis of WHI data suggested that women starting hormone therapy within 10 years from the onset of menopause had a reduced risk of coronary heart disease compared with those who started later.209 Neither breast cancer210 nor coronary heart disease211 were increased among estrogen users in the estrogen alone trial.

Thromboembolic events were significantly elevated among estrogen users compared to placebo in both trials,212, 213 similar to results from raloxifene trials (Table 9). Risks for strokes were also higher in estrogen users for both trials,158, 214 although the level of significance varied if using nominal versus adjusted confidence intervals. Estrogen with progestin did not increase risk for endometrial cancer215 and reduced risk for colon cancer212 compared to placebo. Women using estrogen alone had similar all cause death and colon cancer outcomes as women using placebo.158


  • PubReader
  • Print View
  • Cite this Page
  • PDF version of this title (3.1M)

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...