Comparison of Risk of Osteoporotic Fracture in Denosumab vs Alendronate Treatment Within 3 Years of Initiation

Key Points Question What is the real-world risk of fracture for patients treated with denosumab compared with those treated with alendronate? Findings In this Danish cohort study including 4624 individuals treated with denosumab and 87 731 individuals treated with alendronate, 3-year cumulative incidence of hip fracture was 3.7% and 3.1% among the denosumab and alendronate cohorts, respectively. The 3-year cumulative incidence of any fracture was 9.0% for both cohorts. Meaning In routine clinical practice, initiation of denosumab and alendronate treatments were associated with similar risks of hip and any fracture over a 3-year period.


Introduction
Osteoporosis is characterized by progressive deterioration of bone structure and decreased bone mineral density (BMD). 1 Fractures are a common manifestation of osteoporosis, 2,3 with hip fracture being the most serious. 4,5 Worldwide, osteoporosis is the most common metabolic skeletal disease, 6 affecting 200 million persons and causing nearly 9 million fractures annually. 2 Bisphosphonates are a mainstay of prevention and treatment of postmenopausal osteoporosis, with alendronate being the first-line choice, given its effectiveness and low price. 7,8 Denosumab was approved in the European Union, including Denmark, in May 2010 for the treatment of osteoporosis in postmenopausal women and in men at high fracture risk. 9 Both denosumab and alendronate are antiresorptive drugs and inhibit osteoclasts, albeit via different mechanisms; while denosumab binds the cytokine receptor activator of nuclear factor-κB ligand and thereby blocks osteoclasts' formation, maturation, activation, and survival, alendronate binds to bone mineral, where it is absorbed by mature osteoclasts, inducing osteoclast apoptosis and suppressing resorption. 10 Some adverse effects reported among bisphosphonates users, such as gastrointestinal tract effects or acute kidney injury, are rarely seen with denosumab, owing to differences in administration and clearance between the 2 agents. 11 In contrast to weekly oral alendronate treatment, denosumab treatment, administered subcutaneously every 6 months, removes the need for patient adherence for that period. 12 On the other hand, denosumab is promptly cleared after 6 months, while bisphosphonates linger in bone for many years, potentially partially compensating for suboptimal adherence. Denosumab is a cost-effective treatment for osteoporosis with €50 000 to €60 000 (US$56 232-$67 478) per quality-adjusted life-year gained 13 and is a potential alternative to alendronate in patients 75 years or older, those with renal impairment, and those with low expected adherence. 14 In randomized clinical trials (RCTs), denosumab was more efficacious than bisphosphonates in increasing bone mass among postmenopausal women. 15,16 Although there is an association of change in BMD with reduction in fracture risk, 17,18 the magnitude of the association is not well established.
The RCTs examined the risk of any fracture as a secondary outcome, finding no difference in the fracture risk between denosumab and alendronate within 1 year of treatment initiation. 15,16 However, these studies were based on a low number of fractures, and the extent of fracture reduction with denosumab compared with alendronate in routine clinical practice is unclear from the data. Using routinely collected data from population-based health registries in Denmark, we conducted a nationwide cohort study to compare risk of hip and any fracture in patients treated with denosumab and alendronate.
As this cohort study did not involve contact with patients or an intervention, it was not necessary to obtain permission from the Danish Scientific Ethical Committee. The study was approved by the Danish Data Protection Agency. Danish law does not require informed consent or ethical approval for studies based solely on registry data. This report follows the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline for cohort studies.

Patients
We used an active-comparator, new-user approach 26 to include patients taking denosumab (the denosumab cohort) and patients taking alendronate (the alendronate cohort), defined as persons with the first dispensing of denosumab or alendronate between May 26, 2010, andDecember 31, 2017. The date of the first dispensing of denosumab or alendronate during the study period was the index date in each respective cohort. We excluded patients with a recorded dispensing of any osteoporosis medication (eg, denosumab, alendronate, raloxifene, ipriflavone, strontium ranelate, other bisphosphonates, teriparatide, calcitonin) in the 12 months preceding the index date. We furthermore excluded patients younger than 50 years on the index date and those with a history of cancer or Paget disease in the 10 years before index date (both of which are alternative indications for denosumab and alendronate).

Outcome
The primary outcome was hip fracture, identified by primary or secondary diagnosis during hospitalization. The secondary outcome was any fracture, including primary or secondary diagnosis during hospitalization for hip or vertebral fractures and primary or secondary diagnosis for nonvertebral nonhip fractures recorded during any hospital encounter (inpatient, outpatient, or emergency).

Covariates
We measured the following covariates at the index date: age, sex, and comorbidities, including history of fracture, alcohol-related disorders, obesity, and diseases from the Charlson Comorbidity Index (CCI). We summarized diseases from the CCI using the Romano modification, 27 ie, excluding chronic obstructive pulmonary disease (a marker of smoking) and chronic renal impairment (a predictor of denosumab treatment), which were treated as separate covariates. We classified the CCI scores as no comorbidity (CCI score, 0), moderate comorbidity (CCI score, 1-2), or severe comorbidity (CCI score, Ն3). We also included data on dispensings of comedications affecting bone metabolism or fracture risk since 2004. Included comedications were oral corticosteroids, anticoagulants and antithrombotics, hormone replacement therapy, anxiolytics and sedatives, antipsychotics, antidepressants, statins, nonsteroid anti-inflammatory drugs, antihypertensive drugs, drugs for treatment of chronic obstructive pulmonary disease, opioids, and antithyroid drugs. History of hip and any fracture was defined using the same codes as those used to define hip and any fracture outcome.

Statistical Analysis
We tabulated the characteristics of the denosumab cohort and the alendronate cohort, then computed crude rates of fracture outcomes per 1000 person-years. Analogous to the intention-totreat approach used in RCTs, 26 a patient was considered treated with the agent initiated on the index date. We included all patient data from the index date to the date of a fracture outcome, death, emigration, or December 31, 2017, whichever occurred first. Occurrence of a fracture at a site other than the hip did not censor follow-up for the hip fracture end point. We plotted cohort-specific cumulative incidence curves for hip fracture and any fracture, considering death as a competing risk and weighting all observations by stabilized inverse probability of treatment weights (IPTWs) to balance the measured covariates. 28 Inverse probability of treatment weights were derived from propensity scores, which were computed using logistic regression as predicted probability of initiating denosumab or alendronate treatment as a function of the covariates and interaction terms of sex with comedication and of calendar year with all other covariates. Weights were stabilized by multiplying the weights in the denosumab cohort by the proportion of denosumab users in the total cohort and by multiplying the weights in the alendronate cohort by the proportion of alendronate users in the total cohort. 28 We used standardized mean differences to assess the balance of covariates in the 2 cohorts, with the aim of achieving a standardized mean difference of less than 0.1 for all covariates.

JAMA Network Open | Diabetes and Endocrinology
Furthermore, we computed risk differences (based on cumulative incidences) with 95% CIs. We used Cox proportional hazards regression to compute crude and adjusted hazard ratios (aHRs) with 95% CIs for each outcome, overall and stratified by age group, sex, and fracture history. In the stratified analyses, new stabilized IPTWs were computed within each subgroup. The proportionality of hazards assumptions was found not to be violated, according to log-log plots.
We tested the robustness of the results in a series of sensitivity analyses. First, we repeated the analyses while excluding patients with a recent fracture at the same site as a given outcome, varying the definition of recent from 30 to 730 days before the index date. Second, we used a stricter definition of treatment initiation by extending the requirement of absence of pre-index date osteoporosis therapy from 1 year to 5 years. Third, we excluded atypical femoral fractures (approximated, in the absence of specific diagnostic codes for atypical fractures or access to results of radiography, by diagnostic codes for fractures of the femoral shaft or subtrochanteric region) from the hip fracture definition and computed new aHRs for hip fracture. Fourth, we repeated the analyses while censoring follow-up at discontinuation of or switch from the initial treatment (analogous to the per-protocol approach in RCTs). Discontinuation was defined as a gap of 90 days or more between the expiration of the days supplied in a given dispensing for the denosumab cohort and as a gap of 180 days or more for the alendronate cohort, thereby accounting for differences in the pharmacokinetic properties of the 2 agents (ie, faster clearance of denosumab). 10 We tested different combinations of discontinuation-defining intervals for denosumab/alendronate: 90/90 days, 60/150 days, 120/210 days, and 120/180 days. A switch was defined as a dispensing of a different antiosteoporosis agent before discontinuation of the initial agent. Analyses were performed with SAS version 9.4 (SAS Institute Inc).

Results
We identified 4624 initiators of denosumab and 87 731 initiators of alendronate. Of the 92 355 included patients, 75 046 (81.3%) were women, and the mean (SD) age was 71 (10) years. Table 1 shows patient characteristics in the 2 cohorts before applying IPTWs. There was a sharp increase in   history of any fracture ( Table 2). Weighted cumulative incidences for hip fracture within 3 years among patients with a history of any fracture were 6.1% and 5.1% in the denosumab cohort and the alendronate cohort, respectively, with an absolute risk difference of 1.0% (95% CI, −0.7 to 2.7).

JAMA Network Open | Diabetes and Endocrinology
Weighted cumulative incidences of hip fracture within 3 years among patients without history of any fracture were 2.9% and 2.3% in the denosumab cohort and the alendronate cohort, respectively, with an absolute risk difference of 0.6% (95% CI, −0.3 to 1.5).

Any Fracture
Within 3 years of follow-up, the initiation of denosumab and alendronate were both associated with a cumulative incidence of 9.0% for any fracture. The overall aHR for denosumab vs alendronate was 0.92 (95% CI, 0.83-1.02) for any fracture ( Table 3). For denosumab vs alendronate, the aHR for any fracture was 0.84 (95% CI, 0.71-0.98) among patients with a history of any fracture and 0.77 (95% CI, 0.64-0.93) among patients without a history of any fracture. Results of stratified analyses for any fracture are presented in Table 3 and are very similar across strata.

Sensitivity Analyses
The results did not materially change after excluding patients with a recent fracture at the same site as a given outcome or after excluding atypical femoral fractures from the hip fracture outcome definition. Extending the required therapy-free period to 5 years to define new use yielded an aHR Patients switching between treatments were rare. Discontinuation was more frequent; thus, we lost approximately 30% of person-years in the denosumab group and approximately 44% for the alendronate cohort (these figures were similar for hip and any fracture). Nevertheless, the on-treatment analyses were consistent with the findings for hip fracture in the primary analysis (eTable 2 in the Supplement).

Discussion
This large nationwide historical cohort study was conducted using routinely and prospectively collected data originating from a health care system with universal coverage. Initiation of denosumab and initiation of alendronate were associated with similar risks of hip fracture or any fracture during a median follow-up of approximately 3 years, regardless of sex, age, or previous fracture.
Randomized clinical trials evaluating head-to-head efficacy of denosumab vs alendronate used BMD as the primary outcome and clinical fractures as an adverse effect or complication. 15,16 These RCTs, conducted in postmenopausal women, regardless of prior treatment for osteoporosis, both found denosumab to be more efficacious than alendronate in increasing BMD, possibly the best proxy outcome for subsequent fracture risk. Overall, 6 of 8 RCTs 29-34 included in 2 meta-analyses 15,16 reported results on risk of any fracture. Our finding of no clinically relevant differences between denosumab and alendronate in the risk of hip and any fracture is in line with findings of the RCTs. 15,16 In addition, our absolute risk estimates are similar to those from the RCTs; however, the absolute number of fractures in our study is considerably higher than the total number of fractures from RCTs included in the 2 meta-analyses. 15   alendronate is an argument for prescribing denosumab rather than alendronate to prevent hip fractures.

Limitations
Our study had limitations. The most important limitation when comparing a new-in-class agent with an established treatment is the possibility of residual confounding. For example, renal impairment was more prevalent in the denosumab cohort than in the alendronate cohort but is incompletely measured by hospital diagnoses. 35 We lacked measures of frailty, socioeconomic status, and lifestyle, which could affect our results. Furthermore, no data on BMD were available, which is likely the main driver of treatment choice. Most guidelines, including National Institute for Health and Care Excellence technology appraisal guideline 204 from 2010, 36 recommend the use of denosumab only in patients with lower BMD or in secondary fracture prevention, if first-line alendronate is not tolerated or contraindicated. Thus, confounding by indication is another limitation of our study, despite comprehensive availability of confounders and use of stabilized IPTWs in statistical analyses.
Also, despite the large sample size, it was not possible to divide patients into more than 2 age groups.

Conclusions
In this nationwide cohort study based on routinely collected data in Denmark, treatment with denosumab and alendronate were associated with similar risks of hip and any fracture over a 3-year period. Sex, age, and fracture history were not associated with patients' risk.