This study was sponsored by Amgen. Drs. Danese and Bauer were consultants for Amgen. Dr. Bauer received research funding from Novartis, Amgen, and Procter & Gamble and is a consultant for Amgen, Zelos, and Merck.
Effect of Adherence on Lifetime Fractures in Osteoporotic Women Treated With Daily and Weekly Bisphosphonates†
Article first published online: 4 MAY 2009
Copyright © 2009 ASBMR
Journal of Bone and Mineral Research
Volume 24, Issue 11, pages 1819–1826, November 2009
How to Cite
Danese, M. D., Badamgarav, E. and Bauer, D. C. (2009), Effect of Adherence on Lifetime Fractures in Osteoporotic Women Treated With Daily and Weekly Bisphosphonates. J Bone Miner Res, 24: 1819–1826. doi: 10.1359/jbmr.090506
Published online on May 4, 2009
- Issue published online: 4 DEC 2009
- Article first published online: 4 MAY 2009
- Manuscript Accepted: 1 MAY 2009
- Manuscript Revised: 10 MAR 2009
- Manuscript Received: 10 OCT 2008
Patients miss doses of their osteoporosis medications, or stop taking them altogether, for a variety of reasons. Whereas the reasons have been well-studied, their consequences, at the population level, have not. The goal of this study was to estimate the number of fractures that could be prevented with optimal adherence compared with usual adherence to daily and weekly bisphosphonates in the United States (US). We developed a simulation of adherence to bisphosphonate therapy in the US. The model samples women by age and BMD from nationally representative US distributions, and tracks them over time assuming they are treated with a daily or weekly bisphosphonate. The model simulates two adherence scenarios: usual adherence and optimal adherence. The differences in fracture rates between these scenarios, as well as the medication and fracture costs, are estimated with the model. Approximately 258 (95% interval, 194–324) lifetime fractures can be prevented with optimal adherence per 1,000 bisphosphonate-treated women. For optimal adherence, these results translate to an additional lifetime medication cost of $3,800 and a lifetime savings in fracture-related costs of $2,100, for an expected net cost of $1,700 (95% interval, −$4,100 to $3,300) per woman over her lifetime. These results suggest that in patients taking daily or weekly bisphosphonate therapy, a substantial number of fractures occur that are attributable to less than optimal adherence. These results show that there is implicit value to improving adherence, both from a financial and clinical perspective.
Patients miss medication doses for a variety of reasons including side effects, cost, inconvenience or difficulties with the route of administration, and simple forgetfulness.(1,2) A variety of terms are used to describe missed doses, including adherence, persistence, and compliance.(3) Simply put, adherence reflects both whether all doses have been taken (compliance) and the duration over which a person is considered to be taking medication (persistence). In osteoporosis, both aspects of adherence have been found to be suboptimal.(4)
When fractures occur, they can be associated with high costs of treatment, lower quality of life, and increased risk of death, factors that affect patients, providers, and payers alike.(5–8) Since the introduction of daily, and then weekly, bisphosphonates for the prevention of fractures in postmenopausal women, there has been a concerted effort to influence adherence with the hope that, if adherence improved, fracture risk reduction would also improve.(2,9–11) More recently, evidence showing that worse adherence is associated with both lower BMD gain and higher fracture rates has made it a key issue in osteoporosis treatment for providers and health systems.(12–17)
Models geared toward guiding coverage decisions for populations, such as cost-effectiveness analyses of osteoporosis treatment, generally have not accounted for adherence explicitly, although this is changing.(5,18–25) To date, the importance of missed doses has not been systematically evaluated, and the magnitude of the problem has not been evaluated at a population level. The goal of this study was to determine how many fractures might be prevented with optimal adherence (i.e., including both compliance and persistence) compared with usual adherence in osteoporotic women treated with daily or weekly bisphosphonates in the United States.
MATERIALS AND METHODS
We constructed a Monte Carlo simulation model using TreeAge Pro (TreeAge Software, Williamstown, MA, USA) to compare two scenarios: usual adherence and optimal adherence. The usual adherence scenario was defined as the current state of adherence with daily and weekly bisphosphonate therapy in the United States. This state reflected a distribution of both compliance levels and times until discontinuation. In contrast, the optimal adherence scenario reflected a very high level of adherence (90% or better compliance and no discontinuation; Fig. 1).
The model was constructed as a simulation of bisphosphonate treatment and its effects on hip, vertebral, and wrist fractures in a sample of newly diagnosed individuals who are representative of the U.S. female population >50 yr of age.(26) Patient age was sampled from the distribution of ages in U.S. females >50 yr of age based on projections from the U.S. Census. In addition, specific age cohorts (55, 65, and 75 yr) were also modeled. All patients were modeled until age 100 or death.
Based on the sampled age, the patient's BMD at the femoral neck was determined by sampling from the age-specific distribution of BMD values published from the Third National Health and Nutrition Examination Survey (NHANES III).(27) Based on this population-based BMD value, a T-score was calculated by comparing the sampled value to the value for white women 20–29 yr of age from NHANES III.(28) If the BMD value indicated osteoporosis (T-score ≤ −2.5), the patient was assumed to be treated for osteoporosis and included in the simulation. This T-score, after being converted to a Z score, was used to estimate the patients underlying, age-specific fracture rate (see more details below).
Compliance and persistence with therapy
The usual adherence scenario, which included estimates of both compliance and persistence, was modeled based on published U.S. data from Huybrechts et al.(13) This report was used because it provided both compliance estimates using repeated medication possession ratio (MPR; the proportion of days covered by dispensed prescriptions) estimates, and relative risk estimates for the association between compliance and fracture risk. Compliance was estimated using an MPR that was recalculated on a monthly basis using four groups: >90%, 80–90%, 50–80%, and <50%.
Patients were started in our simulation after 6 mo of therapy to match the construction of the risk model from Huybrechts et al. (i.e., fractures during the first 6 mo were excluded from their relative risk analyses). Over time, the likelihood of being in one of the four compliance groups, or having discontinued, was determined randomly for each simulated patient based on the aforementioned study. Patients were assumed to remain in their compliance group for the remainder of the simulation unless they discontinued therapy: data on transitions among compliance groups over time were not available. Patients in the optimal adherence scenario were assumed to remain highly compliant (>90% of doses taken) and to remain on therapy until the end of the simulation. This assumption was relaxed to reflect 10 yr of therapy in sensitivity analyses. Table 1 shows estimates of compliance and persistence rates.
Distributions, relative risks, and costs
Distributions were available or could be created for most inputs in the model, allowing for the effects of uncertainty in these parameter estimates to be incorporated into the analyses. Distributions for age and BMD were incorporated into the base case analyses to reflect the heterogeneity within the U.S. population of women ≥50 yr of age with osteoporosis. Distributions for relative risks were incorporated to assess uncertainty in the model using triangular distributions (sampled on the log scale). These included estimates for the relative risk of fracture for various compliance (MPR) levels, the relative risk reductions in fracture from bisphosphonate therapy among other inputs, and the cost of fractures (but not the cost of bisphosphonates; Table 2).
Bisphosphonate treatment was defined as daily or weekly alendronate or risedronate. This was done to match the study on which the compliance rates were based, even though weekly regimens have been shown to be associated with better adherence than daily.(13,29,30) Relative risk (RR) reductions for therapy were taken from a meta-analysis of alendronate therapy by Cranney et al.(31) Recent analyses suggest that there are no substantive differences between alendronate and risedronate based on available comparative data.(1) Furthermore, the available data suggest that the risk reductions are comparable for these agents.(1,31–33) In addition to these specified distributions, randomly occurring events based on probabilities were used for discrete events, including the assignment of compliance group, death, discontinuation of bisphosphonate therapy, and occurrence of fracture. Table 2 shows a complete listing of probabilities and distributions.
Costs in the base case were limited to the drug cost and the cost of fracture. Adverse event costs were added as a sensitivity analysis. Fracture costs were based on site-specific United U.S.-based estimates for the 1-yr direct medical cost of fracture, inflated to 2008 using the medical component of consumer price index.(34) Medication costs were based on 2008 wholesale acquisition costs (WACs), and assumed an equal market share between branded, generic daily, and generic weekly bisphosphonates. (There were no meaningful differences in the monthly costs of the branded bisphosphonates.) Adverse event costs for gastrointestinal complications in sensitivity analyses were adapted from a cost-effectiveness model from Liu et al.(21) and inflated to 2008 dollars. All costs were discounted at 3%/yr. Because our primary aim was to estimate the lifetime number of fractures prevented (i.e., not the present value of the fractures), we did not discount clinical outcomes. Tables 2 and 3 show a list of costs.
Fracture risk estimation
Fracture events were included for hip fractures, clinical vertebral fractures, and all other fractures (distal forearm, shaft/distal humerus, proximal humerus, pelvis, clavicle/scapula, and leg). Population-based first fracture rates from Rochester, MN,(35) were converted to values specific to the sampled patient's age and BMD using an approach implemented by Schousboe et al.(22) and DeLaet et al.(36) This approach was chosen because it permits the calculation of fracture risk for any Z score as a function of the underlying relationship between a Z score and the relative risk of fracture (exponential), the known distribution of Z scores (normal), and the relationship between age and fracture rate in the population (monotonically increasing step function). Additional fracture types were not included in the model.
Model details and assumptions
Several simplifications and assumptions were required to develop a reasonable model. These are listed below.
• The study by Huybrechts et al.(13) also included patients taking hormone replacement therapy. We assumed that the results would be representative for bisphosphonate users, in part because the authors noted that there were no notable differences among therapies in the analyses of compliance and fracture risk.
• We assumed that the relationship between compliance and fracture risk would be the same for each fracture in the model because fractures of all types were used to derive the estimates.
• The number of fractures of any kind was limited to two.
• Patients were assumed to be fracture free at the time of initiating therapy.
• The efficacy of therapy after discontinuation was attenuated linearly over time. The time until complete attenuation depended on the duration of therapy before discontinuation.
• Patients who discontinued therapy before 60 mo would take between 6 and 60 mo to return to their baseline fracture rate for their age.
• Patients who discontinued after 60 mo would take between 60 mo and their total time on therapy to return to their baseline fracture rate for their age.
• The effects of fractures on mortality were not modeled in the base case but were incorporated into sensitivity analyses.(37)
• The cost of fracture after 1 yr and the quality of life decrements associated with any fracture were not incorporated.
• No additional costs to achieve optimal adherence were included.
• All patients with T-score ≤ −2.5 were treated with bisphosphonates.
For the primary analyses of fractures prevented, we used second-order simulations to calculate estimates that incorporate uncertainty. This approach yields bootstrap-like CIs that result from allowing all model inputs with distributions to vary simultaneously. Exploratory first-order simulations established that model fracture rates stabilized after ∼5000 simulated patients. Exploratory second-order simulations of 5000 simulated patients established that differences in fracture rates stabilized after ∼250 replications of 5000 simulated patients; therefore, the final results were run using 400 replications of 5000 simulated patients. For these results, we report the middle 95% of the distribution for all results (95% interval).
For our sensitivity analysis of changing the T-score to −2.0, the model was run similarly to the primary analyses. For other one-way sensitivity analyses of age at initiation of therapy and bisphosphonate cost, where we were only assessing the uncertainty resulting from the variable being studied, we ran the model for 10,000 simulated patients. In these models, all distributions used on the second-order analyses were also included within the first-order analysis, except for the one being varied deterministically. Therefore, there is no sampling interval reported for these results-only the mean result is reported at each level of the age or cost factor being tested. Because Monte Carlo sampling is inherently random, there are slight differences between results for the 10,000 patient sensitivity analyses and the primary analyses using 400 replications of 5000 patients.
In a simulated population-based random sample of women ≥50 yr of age, the average age was 65.2 yr, and the average follow-up time was 237 mo (19.8 yr), with a range of 231–242 mo across the replications. In the optimal adherence group, there were 258 total fractures prevented per 1000 women with osteoporosis treated for a lifetime compared with the usual adherence group. The 95% interval for this outcome ranged from 194 to 324 fractures per 1000 women when uncertainty about the inputs was incorporated by second-order simulation (Fig. 2). Table 4 shows the results by fracture type.
For each fracture type, the estimated mean fracture rates in the usual adherence group were greater than in the optimal adherence group. The fracture rate in the optimal adherence group was 23% lower when considering all fractures across all ages (95% interval, 16–30%). By fracture type, optimal adherence was lower by 20% (95% interval, 8–30%) for hip fractures, 24% (95% interval, 16–32%) for vertebral fractures, and 25% (95% interval, 12–40%) for all other fractures combined. Similar results were seen across age groups, with higher fracture rates in the usual adherence group. Table 5 shows fracture rates for each fracture type and age group.
In terms of fracture costs in the U.S. population, the expected total cost of medication and fractures in the usual adherence group was $16,600 per patient over a lifetime (95% interval, $7000 to $37,500). In the optimal adherence group, the range was higher at $18,300 (95% interval, $9200 to $35,200). The difference between usual and optimal was $1700 (95% interval, −$4100 to $3300), with optimal adherence more expensive, on average, than usual adherence (Fig. 3). When separating costs into those for fractures and those for medication, the optimal adherence strategy was associated with a reduction in the expected lifetime cost of fractures. However, the optimal adherence group was also associated with an increase in the expected cost of medication (Table 5).
In sensitivity analyses of age, bisphosphonate cost, and the T-score threshold, the results were similar to the base case (Table 6). Fracture rates increased with age, and the difference in total cost became smaller. There were slight variations in the total cost difference for the entire population depending on whether branded, generic weekly, or a mixture (the base case assumption) was assumed. When the least expensive option, generic weekly alendronate, was used, the total cost difference became smaller ($800), and when branded was used, the difference in cost became larger ($3000). Using a T-score threshold of −2.0 resulted in fewer fractures and a greater difference in total cost.
The incorporation of higher mortality risk after fracture resulted in an 11% decline in the total number of fractures regardless of adherence. Accordingly, the relative reduction in fractures with optimal adherence was essentially unchanged (23%). Furthermore, when a 10-yr maximum time on therapy was considered, the fracture rate in the overall population increased in both adherence groups and eliminated much of the benefit of adherence. In contrast, when the same 10-yr maximum was applied to 85-yr-old patients, the results were affected to a much smaller degree. Finally, the incorporation of adverse event costs increased the cost of optimal adherence relative to usual adherence. Table 6 shows sensitivity analysis results.
These results suggest that, in patients taking daily or weekly bisphosphonate therapy, a substantial number of fractures occur that are attributable to less than optimal adherence. Whereas the value of avoiding these fractures from a clinical perspective is simple and clear, it is less so from a strictly financial perspective. To avoid these fractures with current branded therapies at current prices would be expensive, averaging $1700 per patient over her lifetime to pay for the improvement in compliance and persistence. However, ∼55% of the added medication cost (in present dollars) would be offset by the savings from the avoided fractures.
Adherence to medications is a complex issue, and many factors contribute to whether a patient takes her medication. As many as 60% of elderly patients with chronic disease are poorly adherent to treatment,(38) and many programs have been initiated in an attempt to improve compliance.(39) However, a review of current programs concluded that these programs are “complex, labor-intensive, and not predictably effective.”(39) Additionally, the benefit of these programs may not be long lasting. A randomized trial assessing the ability of a pharmacist intervention program to improve adherence to heart failure medications reported improvement in adherence and outcomes, but the program's benefit required constant intervention. The cost per patient to implement the program in 2004 was $205.(40) If incorporated into our model, such an expense on an annual basis would widen the difference between the usual and optimal adherence groups. In addition, there is no evidence that adherence would approach the level of optimal adherence, as defined in our model. These results suggest that treatments that are inherently easier to take and that foster better adherence have implicit value over and above current daily or weekly treatment options.
There are a number of important assumptions introduced to create this model as well as limitations to our resulting analyses. We selected our base case analysis to be conservative (i.e., to underestimate the effects of adherence) where possible. We structured our analyses to represent the fracture risk in a random sample of U.S. women with osteoporosis, assuming they are all equally likely to receive treatment. In reality, patients who receive treatment are likely to have lower BMD values and have a higher risk of fracture.(41) We limited the number of fractures to two of each type. We assumed no prevalent fractures at baseline, which lowered the underlying fracture rate and reduced the potential benefit of better adherence. Also, whereas we included eight fractures (by pooling six of them into a single group for computational efficiency), we did not consider every possible fracture site. We selected a T-score threshold of −2.5, which is in contrast to the guidelines from the National Osteoporosis Foundation, which recommends −2.0.(42) However, this did not have much of an effect on the results; as expected, there were fewer fractures to offset the additional medication costs from better adherence.
Moreover, our analyses assume that patients who take >90% of their medication will achieve reductions in fracture risk based on pooled data from published clinical trials. This seems plausible because trials are likely to reflect the highest achievable rate of compliance in actual practice. However, because most of the trial results were analyzed on an intent-to-treat basis, and because the actual compliance in the trials was frequently <90%, the estimates of efficacy from these trials were already attenuated to some degree because of noncompliance and dropout.(1) Therefore, we underestimated the underlying risk reduction with therapy and the benefit of better adherence.
Finally, we do not know the relationship between the duration of bisphosphonate therapy and the duration of antifracture efficacy. Based on recent data from the FLEX study,(43) we allowed patients who received at least 5 yr of bisphosphonate therapy to retain the associated fracture risk reduction for at least 5 yr, and we allowed patients on treatment for shorter periods to retain the risk reduction for 27 mo on average. On a related note, in sensitivity analyses, shorter (10-yr) durations of therapy reduced the benefits of adherence substantially for the whole population. However, this was not the case in older patients who have much higher fracture risks. Hence, a limited duration of treatment-even with optimal adherence-would not be expected to benefit younger patients unless the effects of therapy after discontinuation persist for decades and protect women in their years of highest risk. Whereas the relationship is still not well studied, it is unlikely that the model understates the efficacy of bisphosphonates after discontinuation. When taken together, all of these decisions suggest that our analyses understate the actual effects of adherence.
On the other hand, it is possible that the relationship between fracture risk and compliance may be confounded by a patient's underlying risk aversion.(44) That is, risk aversion may be associated with both better compliance and a healthier lifestyle, and the healthier lifestyle may be associated with fewer adverse outcomes. A recent inquiry into this area in postacute myocardial infarction care has suggested that this may be a drug class-specific phenomenon.(44) In osteoporosis, the limited available evidence does not suggest that there is a strong “healthy adherer” effect, at least for patients taking calcium supplementation. In a 5-yr randomized trial of calcium supplementation on fracture risk in older women (RR = 0.87 for any fracture), when the compliant patients were compared with similarly compliant placebo patients, the benefit of therapy was significant (RR = 0.66).(45) When the comparison was made between the noncompliant treated and placebo groups, there was no treatment benefit (RR = 1.09). This improved efficacy for treatment relative to placebo in similarly compliant patients suggests that not all of the relationship between compliance and fracture risk can be attributed to self-selection.
The ability to create a lifetime model was limited by the available long-term data on adherence. The compliance and persistence data from Huybrechts et al. were collected over 5 yr, and the average follow-up period for the patients in the analyses was 1.7 yr. It is possible that patients who discontinue therapy might restart treatment at a later date.(46) Therefore, this study may not properly reflect lifetime use and may underestimate the usual state of adherence. However, the 5-yr persistence estimate of 55% in this study is much higher than estimates in many other studies,(1) which may mitigate this problem to some degree. On a related note, if the efficacy of bisphosphonates were to change with time, the simulation results would be affected. This was not incorporated into the model.
Our approach to the cost side of the model was intentionally simplistic. We only included drug costs and 1 yr of fracture-related costs (which were sampled to reflect the heterogeneity of fracture severity). At older ages in particular, the economic effects of fractures (rehabilitation or nursing home care) may be incurred for several years, so the savings from prevented fractures may be higher at older ages. As discussed above, we did not include any costs for assisting a patient with continuing therapy for her remaining lifetime. It is likely that, for such a goal to be achieved with these medications, substantial costs of additional monitoring and management would be incurred and additional adverse events might be elicited. As seen in our sensitivity analyses, when adverse event costs were added to the model, the cost of optimal adherence increased. However, it should also be noted that optimal adherence includes taking medication properly, which could lower adverse event costs; hence, this sensitivity analysis may overstate the adverse event burden.
The results also suggest that the introduction of generic alendronate makes better adherence less expensive. As treatment becomes less expensive, it is more likely that cost savings from prevented fractures will offset the additional medication costs that result from better adherence, a result shown in the sensitivity analyses. It might be possible to use the expected drug cost savings to justify additional or improved adherence programs while keeping the cost of osteoporosis care unchanged.
In sum, improving the current state of adherence with bisphosphonates may have important clinical benefits to patients by reducing fractures. This may be achieved in a variety of ways, including developing adherence programs and improved medications. However, regardless of the approach taken, such improvements will be important to eliminating the hidden costs-both clinical and financial-of poor adherence.
We thank David Macarios for invaluable assistance in conceptualizing this research project and helpful comments along the way.
- 12007 Comparative Effectiveness of Treatments to Prevent Fractures in Men and Women With Low Bone Density or Osteoporosis. Comparative Effectiveness Review No. 12. (Prepared by Southern California/RAND Evidence-based Practice Center under contract No. 290–02–0003). Agency for Healthcare Research and Quality, Rockville, MD, USA., , , , , , , , , , , , , , , ,
- 26U.S. Census Bureau 2006 Annual Estimates of the Population by Sex and Five-Year Age Groups for the United States: April 1, 2000 to July 1, 2005. Available online at http://www.census.gov/. Accessed May 10, 2006.
- 34U.S. Department of Labor Consumer Price Index—All Urban Cosumers Medical Care; 1995 to 1997. Available online at http://www.bls.gov. Accessed May 2, 2008.
- 42National Osteoporosis Foundation 2003 Physician's Guide to Prevention and Treatment of Osteoporosis. National Osteoporosis Foundation, Washington, DC, USA.
- 47National Center for Health Statistics 2004 National Vital Statistics Reports. Probability of Death During 1-Year Intervals, 53rd ed. National Center for Health Statistics, Hyattsville, MD, USA.
- 52DMD America I. Analysource online. Available online at http://www.analysource.com. Accessed on May 2, 2008.