Cost-effectiveness of romosozumab for the treatment of postmenopausal women at very high risk of fracture in Canada

The modelled patient population consisted of a hypothetical cohort of postmenopausal osteoporotic women who are at very high risk for future fracture. The model included patients with a mean age of 74 years (based on participants in the ARCH trial [19]). In line with the American Association of Clinical Endocrinologists/American College of Endocrinology (AACE/ACE) [7] and Endocrine Society Guidelines [8] definitions of very high risk for future fracture, the population consisted of patients with a femoral neck BMD T score ≤  − 2.5 and a history of fragility fracture. Since clinical evidence has shown that the distribution of single and multiple fractures was approximately even, the assumption was made that 50% of patients had a single previous fracture, and 50% had multiple prior fractures [22]. Considering that treatment for osteoporosis influences the risk of fractures and mortality [2], the model used a lifetime time horizon to capture all relevant benefits and costs associated with treatment [23]. All costs and health outcomes were discounted at an annual rate of 1.5% as recommended by the Canadian Agency for Drugs and Technologies in Health (CADTH) Guidelines for the Economic Evaluation of Health Technologies [23]. Alternative discount rates (e.g., 0, 3%) were considered in scenario analyses. In the reference case, the analysis was conducted from the perspective of the public healthcare payer in Canada, with a societal perspective used in a scenario analysis.

In the reference case analysis, the model included the following interventions:

Alendronate and risedronate were selected as comparators because they comprise the large majority of antiresorptive prescriptions in Canada [13]. Patients in the romosozumab/alendronate arm received 12 months of romosozumab (consistent with the duration specified in the product label [24]), sequenced to alendronate, which aligns with the clinical evidence from the ARCH trial [19]. Patients in all arms were assumed to be treated for a total of 5 years; a commonly recommended duration for pharmacological osteoporosis therapy [25]. The model assumed that patients were persistent with therapy in all three arms over the 5-year treatment period.

A Markov cohort state transition model with a 6-month cycle length was used to assess costs and quality-adjusted life-years (QALYs) associated with romosozumab sequenced to alendronate (romosozumab/alendronate) compared with alendronate alone or risedronate alone. The model structure is based on the model developed by the International Osteoporosis Foundation, which has been widely used as a basis for economic analyses of osteoporosis [26‐32]. Markov models are considered appropriate methodologies for this therapeutic area, considering that osteoporosis is a chronic condition and involves a continuous risk over time [28‐33].

Seven Markov health states were considered (Fig. 1): at risk of fracture (i.e., baseline), clinical vertebral fracture, post clinical vertebral fracture, hip fracture, post-hip fracture, “other” fragility fracture (i.e., non-vertebral non-hip fragility fracture), and death. All patients were assumed to be at risk of fracture at baseline. During each cycle, patients had a probability of sustaining a fracture, remaining in the baseline state, or dying. Patients who sustained a fracture transitioned to any of the three health states depending on the fracture type (hip, vertebral, or other fracture). After 1 year in the “other” fracture state, patients who did not sustain another fracture returned to the baseline at-risk health state. After 1 year in the hip and vertebral fracture states, patients who did not sustain another fracture transitioned to the “post-hip fracture” and “post-vertebral fracture” states, respectively.

The model was built with a hierarchical structure based on the severity of fracture types, with hip being the most severe, followed by vertebral and “other.” The structure assumed that patients were only allowed to transition to more severe health states. For example, patients who sustained a hip fracture could not subsequently transition to vertebral or “other” fracture states. This assumption allowed capturing long-term costs and HRQoL with the post-hip or post-vertebral fracture states [29, 31]. However, the hierarchical structure does not explicitly count subsequent fractures further down in the hierarchy, and therefore the model may result in an underestimation of fracture incidence. To correct for this, as in previous adaptations, lower hierarchy fractures were estimated separately by multiplying the number of subjects in each higher hierarchy state with the incidence rate of the lower hierarchy fracture type in the model population [20, 27].

The risk of sustaining a fracture in the model depended on (i) the risk for an individual in the general population of incurring a fracture, (ii) the increased fracture risk associated with osteoporosis, and (iii) a risk reduction, if any, attributed to treatment. The general population risk depended on age and gender. The risk of fracture relative to the general population depended on age, bone mineral density (BMD), and prior fracture prevalence. Age-specific general population fracture rates were taken from Canadian sources (as shown in Appendix Table 5). These values were linearly interpolated or extrapolated as required to produce fracture rates for each year of age (Appendix Table 5). To estimate fracture risks in untreated patients with PMO at very high risk for future fracture, general population fracture rates were adjusted for the lower BMD T scores and higher prevalence of previous fracture in the modelled population. Fracture risks for the target population were adjusted with relative risks of subsequent fracture in patients with a prior vertebral fracture and relative risks of subsequent fracture per standard deviation decline in BMD, as described in previous economic evaluations [20, 30].

For patients receiving treatment, treatment efficacy data (relative risks of fracture) were applied to fracture rates in untreated very high-risk postmenopausal osteoporotic patients. Efficacy data used in the model provided relative risks (RRs) of hip, new vertebral, and nonvertebral fracture separately. RRs of new vertebral fracture were used to inform treatment-specific efficacy in preventing vertebral fractures, while RRs of nonvertebral fracture were used to inform efficacy for “other” fractures (comprising wrist, and all other non-hip, non-vertebral fractures). Hip fracture efficacy was directly informed by treatment-specific RRs of hip fracture.

For the alendronate and risedronate arms, RRs of hip, vertebral, and nonvertebral fracture versus placebo were applied to fracture rates in untreated patients for the duration of treatment (5 years). These data were obtained from a network meta-analysis of randomized controlled trials (Table 1) [34]. For the romosozumab/alendronate arm, RRs of fracture versus placebo were established indirectly from two sources: a comparison of romosozumab with alendronate from the ARCH trial [19, 35] and the comparison of alendronate with placebo [34]. Because of their different modes of action, patterns of treatment benefit over time likely differ for regimens containing a bone-forming agent and regimens consisting of an antiresorptive agent alone. Therefore, RRs of fracture for romosozumab/alendronate versus alendronate were calculated time-dependently. To do this, parametric survival curves were fit to time-to-event data for hip and nonvertebral fractures from the ARCH trial. For each regimen and fracture type, fracture incidence in each 6-month period was calculated from the selected parametric survival functions. The survival functions were selected based on best fit, defined by the Akaike information criterion (AIC). A scenario analysis was considered to select survival functions based on the Bayesian information criterion (BIC) (Appendix Table 6). These values were used to calculate RRs of hip and nonvertebral fracture for romosozumab/alendronate versus alendronate in each 6-month model cycle over the 5-year treatment period (Table 1). Survival models were fitted separately by arm to allow changing fracture incidence over time.

The fracture reduction benefit of pharmacological osteoporosis treatment does not disappear immediately following discontinuation, but rather persists for some time (i.e., the “offset time”). A clinical study in which patients received 5 years’ alendronate treatment followed by 5 years’ placebo found that mean BMD remained at or above pre-treatment levels [36], suggesting that treatment benefit persists for a substantial period. Therefore, the assumption was made that the fracture reduction benefit of treatment declines linearly to 0 over the length of time for which a patient was treated. That is, the treatment offset period lasts for 5 years.

General population all-cause mortality was informed by life tables for females in Canada from Statistics Canada [37]. The model also accounted for the increased risk of mortality following a fracture. Two key assumptions were made regarding mortality following osteoporosis-related fractures: (i) 30% of the excess mortality following a fracture was attributable to the fracture itself, in line with previous analyses [29‐31] and (ii) the increased risk of mortality after hip and vertebral fractures was assumed to last for 8 years as per previous analyses [29‐31]. This duration of excess mortality only applied to hip and vertebral fractures as other fractures were assumed to only have effects in the first year of fracture. Age-specific RRs of mortality in the first year after hip, vertebral, and other fracture, and in the second and following years after hip and vertebral fracture for Canadian women were sourced from Morin et al. [2]. (Appendix Table 7).

To account for the HRQoL loss due to fracture, in the first year after hip, vertebral, and other fracture, and for the second and subsequent years after hip and vertebral fracture, utility multipliers were applied to utilities of the general population. Data specific to subsequent “other” fractures were not available. Therefore, the utility multiplier in the first year after “other” fracture was assumed to correspond to that of a distal forearm fracture. These values were taken from Svedbom et al. [38], an analysis of HRQoL from the International Costs and Utilities Related to Osteoporotic Fractures Study (ICUROS), which recorded HRQoL before and after fracture, for different fracture locations. In a scenario analysis, Canadian-specific HRQoL inputs were derived from HRQoL at different time points to estimate disutilities associated with fractures [39]. Health-related quality of life and utility parameters are summarized in Table 2.

The model included drug acquisition costs, treatment monitoring/administration costs, direct medical costs due to fracture, and long-term care costs (Table 2). Additionally, broader societal costs (including lost productivity costs and patient out-of-pocket costs) were included in a scenario analysis. Lost productivity was estimated based on the mean hourly wage of females 55 years or older, working full and part time in Canada. The value of lost productivity associated with fractures was estimated based on the average time off from work due to each fracture type (Appendix Table 8). For out-of-pocket costs, an assumption was made that BMD measurements, physician visits, and nurse visits were associated with a $20 parking and travel fee. All costs were expressed in Canadian dollars (CAD), inflated to 2020 values where required using the Consumer Price Index (CPI) for Canada [46]. Drug acquisition costs were obtained from the manufacturer for romosozumab and from the list prices on the Ontario Drug Benefit Formulary [41] for antiresorptive agents, using the lowest available unit price for the weekly oral dose of 70 mg and 35 mg for alendronate and risedronate respectively. A conservative assumption was made that patients were fully persistent with therapy in all three arms over the 5-year treatment period, due to the current lack of real-world persistence data for romosozumab sequenced to an antiresorptive and the inherent limitations of discontinuation data from RCTs. In reality, it is known that persistence with osteoporosis treatments is imperfect [47]. Discontinuation data from the randomized controlled ARCH trial are unlikely to represent rates in practice, and therefore full persistence with all treatments was assumed in the model. Wholesaler upcharge and pharmacist dispensing fees were not considered.

For treatment monitoring/administration costs, the model assumed that patients receiving treatment incurred the cost of a physician visit once a year, and the cost of a BMD measurement every 2 years, as per Ontario Health Insurance Plan Schedule of Benefits and Fees [42]. The model assumed that 85% of patients treated with romosozumab required a monthly nurse visit for subcutaneous injection administration. Based on feedback from Canadian clinicians, it was conservatively assumed that the remaining 15% of patients would self-administer romosozumab requiring 2 nurse visits in total (one training visit and one follow-up visit). The cost per nurse visit was estimated assuming a 20-min appointment and based on an hourly wage of $46.31 as per the 2020 Ontario Nurses’ Association Collective Agreement [48]. Age-specific costs in the first year after hip, vertebral, and other fracture in Canada were informed by Metge et al. [43]. Values used in the model comprised total incremental healthcare costs for the year following fracture, versus the year pre-fracture. The cost of “other” fracture was calculated as a weighted average of wrist and humerus fracture.

The model estimated total discounted lifetime costs and QALYs for each intervention, with cost-effectiveness assessed by dominance and in terms of incremental cost-utility ratios (ICURs). Reference case results were assessed through a probabilistic model with 5000 stochastic iterations, where parameters were varied simultaneously according to distributions representing their uncertainty [49]. Cost-effectiveness acceptability curves (CEACs) were derived to summarize the proportion of probabilistic iterations in which each comparator was cost-effective across a range of willingness to pay per QALY-gained thresholds. In addition, sensitivity analyses using the deterministic model were performed to assess the sensitivity of results to changes in individual parameters. Parameters were varied using published confidence intervals or standard errors, where available, and by 25% above and below point estimates where measures of uncertainty were unavailable. Cost-effectiveness was assessed with the incremental net monetary benefit (INMB) for each deterministic sensitivity analysis.

Scenario analyses were conducted to assess the impact of using alternative model assumptions. For each scenario analysis, 5000 iterations of the probabilistic model were conducted. The first scenario tested model structural uncertainty by assuming an alternative treatment sequence, where romosozumab sequenced to risedronate was compared with alendronate alone and risedronate alone. The efficacy of romosozumab/risedronate was assumed to be equivalent to that of romosozumab/alendronate in the first two cycles of the model (i.e., for the duration of romosozumab treatment). For cycles 3 to 10, cycle-specific efficacy of romosozumab/risedronate versus placebo was estimated by applying RRs of fracture for romosozumab/alendronate versus alendronate to RRs for risedronate versus placebo [34]. Additional scenario analyses around modelling assumptions were considered to account for a societal perspective and indirect costs, annual discount rates of 0 or 3%, treatment efficacy rates estimated from parametric models, modified treatment offset time (1 year), increased excess mortality duration, Canadian fracture disutilities, alternative cost for risedronate (Actonel DR), and reduced duration of fracture reduction benefits associated with romosozumab/alendronate. Cost-effectiveness was estimated across the three comparators for each scenario.

The total and disaggregated reference case model results are presented in Table 3. Romosozumab/alendronate yielded the most discounted QALYs and lowest total cost (8.454 and $86,314, respectively), and was associated with a lifetime gain of 0.103 and 0.127 QALYs and a cost reduction of $343 and $3805, relative to alendronate and risedronate, respectively. Although drug costs were highest for romosozumab/alendronate ($8259 vs. $521 for alendronate and $491 for risedronate), the total cost incurred by patients treated with romosozumab/alendronate was lower relative to alendronate and risedronate ($86,314, $86,656, $90,119, respectively). Therefore, the improvements in QALYs and cost reductions for romosozumab/alendronate were driven by a reduction of the expected number of fractures (2561 per 1000 patients versus 2700 for alendronate and 2724 for risedronate). Consequently, alendronate and risedronate were dominated by romosozumab/alendronate. Despite similar annual drug costs of alendronate and risedronate, the total discounted incremental costs of romosozumab/alendronate compared to risedronate (− $3805) was lower than the comparison with alendronate (− $343) due to the differences in hip fracture incidence rates.

Results are also shown for all three comparators as cost-effectiveness acceptability curves (CEACs) (Fig. 2) and a scatter plot of the 5000 iterations (Appendix Figure 3). The number of iterations in which each intervention was found to be cost-effective is displayed over a range of cost per QALY-gained thresholds. These results show that, at a willingness to pay threshold of $50,000 per QALY for example, romosozumab/alendronate was the optimal intervention (in terms of net monetary benefit) in 92.0% of probabilistic iterations. Additionally, romosozumab/alendronate had the highest probability of being cost-effective, relative to alendronate and risedronate, at any willingness to pay threshold value.

Deterministic sensitivity analysis results demonstrated that the model outcomes were most sensitive to fracture reduction efficacy and parameters relating to the cost of long-term care and proportion of patients entering long-term care after hip fracture. However, romosozumab/alendronate was consistently cost-effective, as it represented the highest NMB relative to alendronate and risedronate for each run of the one-way sensitivity analysis (see Tornado diagrams in Appendix Figures 4 and 5).

The incremental costs, QALYs, and ICUR of romosozumab/alendronate relative to alendronate and risedronate for all scenarios are presented in Table 4. Most scenarios yielded similar results to the reference case, where romosozumab/alendronate yielded additional QALYs and fewer costs relative to alendronate and risedronate (i.e., remained the dominant intervention). However, the first scenario analysis that considered a romosozumab/risedronate sequence, represented additional costs ($1322) relative to alendronate alone (ICUR of $14,209 per QALY gained). Therefore, although not dominant, romosozumab/risedronate would be cost-effective relative to alendronate at a willingness to pay threshold of $50,000 per additional QALY gained. Finally, alendronate also had fewer costs compared with romosozumab/alendronate when treatment offset time was defined at 1 year (ICUR = $21,321). Risedronate was consistently dominated throughout all scenario analyses.

This economic evaluation assessed the lifetime cost-effectiveness of 1 year of romosozumab sequenced to 4 years of alendronate versus alendronate alone (5 years) and risedronate alone (5 years), for the treatment of osteoporosis in postmenopausal women in Canada with a history of osteoporotic fracture and who are at very high risk for future fracture. The Markov model employed clinical and economic inputs to estimate the incidence of several fracture types and their associated effects under specific lines of treatment. Treatment with romosozumab/alendronate was associated with the most QALYs and lowest costs, relative to the comparators. Despite having a higher drug and treatment management cost, romosozumab/alendronate produced an overall cost reduction versus both comparators from a healthcare payer perspective, due to a reduction in overall fractures and fracture-related costs. Furthermore, deterministic, probabilistic, and scenario analyses evaluated the robustness of the conclusions after accounting for structural and parameter uncertainty. In general, the parameters which had the largest impact on cost-effectiveness results were those relating to treatment efficacy as these inputs drive differences in fracture incidence between arms. Romosozumab/alendronate had the highest probability of being cost-effective at a willingness to pay threshold of $50,000 per additional QALY gained. Whereas, the current first-line treatments, alendronate or risedronate, were the optimal intervention (in terms of net monetary benefit) in less than 10% of probabilistic iterations.

Although the cost-effectiveness of romosozumab has been previously estimated in different countries, this evaluation contributes novel evidence to the pharmacoeconomic literature of osteoporosis in Canada. A study conducted by Hagino et al. employed a similar model to demonstrate the cost-effectiveness of romosozumab in the Japanese context [20]. However, romosozumab was compared with teriparatide, both sequenced to alendronate, in women with severe PMO previously treated with bisphosphonates. Additionally, it employed BMD efficacy data from the STRUCTURE trial rather than fracture outcomes to inform relative efficacy. Another study, conducted by Sӧreskog et al., assessed the cost-effectiveness of romosozumab/alendronate compared with alendronate alone from a Swedish societal perspective [21]. Romosozumab followed by an antiresorptive was cost-effective compared to an antiresorptive alone despite the substantial price difference; however, the study was primarily designed to present a novel cost-effectiveness model framework that incorporated recency of fracture and treatment sequencing. Whereas, our evaluation provides evidence of the cost-effectiveness of romosozumab relative to an additional comparator (i.e., risedronate), and considering an alternative treatment sequence (romosozumab/risedronate). The results of these economic evaluations suggest that romosozumab has a relatively high probability of being cost-effective relative to bone-forming agents and antiresorptive agents, in three different countries. Furthermore, we acknowledge that different model types, such as patient-level simulation models, have been used to assess the cost-effectiveness of bone-builders [21]. However, previous cost-effectiveness results were consistent with the cohort modelling approach which we applied.

As with all analyses based on economic models, this evaluation has a number of limitations. First, the hierarchical nature of the model can lead to an underestimation of the number of vertebral and other fractures. However, an adjustment function was introduced to correct for the omitted lower hierarchy fractures. Second, there is uncertainty in the duration of treatment “offset time”—the duration of fracture reduction benefit after treatment discontinuation. While the duration of the offset time was based on clinical evidence [36], it is not possible to precisely quantify the duration of the offset time. This assumption was tested through scenario analyses. Third, there is uncertainty in the duration of excess mortality following fracture. This uncertainty was also tested in scenario analyses, with results showing that romosozumab/alendronate remained cost-effective in all cases. Fourth, international data were used where appropriate local data were not available. These included HRQoL loss due to fracture, the baseline age of the population, and RRs used to adjust general population fracture rates for fracture history and BMD. Fifth, the efficacy of romosozumab sequenced to risedronate (included as a scenario analysis) assumed that the additional fracture reduction benefit of romosozumab/risedronate versus risedronate was equivalent to that of romosozumab/alendronate versus alendronate. However, this cannot be confirmed in the absence of direct RCT evidence. In addition, the lack of real-world persistence data for romosozumab and comparators is an inherent limitation of the analysis. Finally, the model extrapolated the efficacy data for romosozumab/alendronate versus alendronate observed in the ARCH trial over the 5-year treatment period.

Despite the limitations, this analysis provides clear evidence of the cost-effectiveness of romosozumab sequenced to alendronate versus alendronate and risedronate alone, with results that are robust to alternative deterministic and scenario analyses. There are several areas where further research is required to facilitate future cost-effectiveness analyses of romosozumab and osteoporosis treatments in general. Research is required to quantify the fracture reduction efficacy of romosozumab sequenced to an antiresorptive agent beyond the duration observed in the ARCH trial. Furthermore, osteoporosis models usually lack empirical data on the duration of fracture reduction benefit after treatment cessation. In this analysis, although romosozumab/alendronate remained cost-effective when the treatment offset time was reduced to 1 year, the number of fractures avoided versus antiresorptive agents alone was reduced, resulting in a positive ICUR of $9408 versus alendronate. Considering that results are sensitive to treatment offset time, further research in this area would help reduce this uncertainty. Third, research into the real-world persistence associated with romosozumab and subsequent antiresorptive treatment is needed. Given its longer dosing interval, it is likely that the persistence of romosozumab would be superior to that of oral alendronate and risedronate [47]. However, quantification of this persistence is required to inform economic analyses. Additionally, local data sources are required to inform the model, specifically around HRQoL. Canadian-specific HRQoL following fracture was estimated by Tarride et al. [39]; however, the population was restricted to patients in long-term care or receiving home care. Although these data were not used to inform the base case, they were tested in a scenario analysis. As such, further local evidence would improve the generalizability of future economic analyses of osteoporosis therapies in Canada and settings with close similarities in their health system such as the UK, Australia, and Western Europe. Lastly, further research on the effectiveness of treatment in other populations (i.e., male osteoporosis, patients with secondary forms of osteoporosis) could inform the potential to extrapolate the results of this analysis.