Introduction
Osteoporosis is a common condition, whose prevalence is increasing with the progressively aging population in Asia. Fragility fractures are the main consequence of osteoporosis, which, as well as resulting in increased mortality and diminished health-related quality of life (HRQoL), also produce a substantial economic burden [
1].
Denosumab is a fully human monoclonal antibody that targets and inhibits RANK-L, a ligand which stimulates bone resorption, with high specificity. Denosumab is currently approved as a treatment for postmenopausal osteoporosis in more than 80 healthcare markets globally [
2‐
5], including in Taiwan, where it has been marketed since 2011 [
6], and is one of the most frequently used antiresorptive treatments under Taiwan’s National Health Insurance system [
7]. As with most pharmacological treatments, compliance with osteoporosis therapies has been shown to be an important factor in achieving treatment success [
8].
During the past decade, several economic models have been published evaluating the cost-effectiveness of denosumab compared with either no treatment or other pharmacological therapies [
9‐
12]. All these studies used efficacy data from randomized clinical trials (RCT), or network meta-analyses of RCTs, to inform the fracture reduction effects of pharmacological therapy, thereby making the assumption that the treatment efficacy observed in trials translates to clinical practice. To address this potential limitation, the cost-effectiveness of denosumab was assessed using treatment effectiveness data from a robust real-world study.
A retrospective database study evaluating the effectiveness and safety of denosumab in clinical practice among women with postmenopausal osteoporosis in Taiwan and Hong Kong using National Health Insurance claims data was completed in 2019 [
13], referred to the Taiwan Real-World Study (TWRWS) hereinafter. Employing the propensity score (PS) matching approach, this study compared the real-world clinical fracture incidence in patients who received at least 2 doses of denosumab (referred to as the “treatment cohort”) and patients who discontinued after 1 dose of denosumab (referred to as “off-treatment cohort”) between 2012 and 2016, using an index date of 225 days after the initial dose. The off-treatment cohort was selected to minimize confounding relating to the initial treatment decision, and because this patient group can be considered a proxy for a “no treatment” cohort, given that any fracture reduction benefit from the initial denosumab dose is unlikely to persist beyond the study index date. PS matching was undertaken based on 59 baseline variables, all of which were balanced between cohorts following the matching process, defined as exhibiting a standardized mean difference < 0.1. Results of the study showed that the risk of hip fracture, clinical vertebral fracture, and non-vertebral fracture was reduced in the Taiwanese treatment cohort by 38% (hazard ratio, 0.62 [95% CI: 0.52, 0.75]), 37% (hazard ratio, 0.63 [95% CI: 0.52, 0.75]), and 38% (hazard ratio, 0.62 [95% CI: 0.53, 0.73), respectively.
The objective of this economic analysis is to evaluate the cost-effectiveness of continued treatment with denosumab versus discontinuation of denosumab after one dose in postmenopausal women with osteoporosis in Taiwan, using fracture reduction effectiveness outcomes from the TWRWS. The study design was selected to align with the design of the TWRWS as closely as possible, and therefore assessed the cost-effectiveness of persistence with denosumab versus non-persistence, rather than the cost-effectiveness of treatment versus no treatment per se. However, given that patients in the “off-treatment cohort” received only a single dose of denosumab, outcomes are also likely to represent a reasonable proxy for the latter comparison. To the best of the authors’ knowledge, this study is the first to use denosumab treatment effectiveness data from a real-world study to inform an economic analysis.
Discussion
This study assessed the cost-effectiveness of continued treatment with denosumab (the “treatment cohort”) with discontinuation of denosumab after one dose (the “off-treatment cohort”), using real-world effectiveness and cost data from Taiwan’s NHIRD. Results show that, compared with discontinuation after one dose, continued treatment with denosumab produces a gain of 0.042 QALYs, due to the fracture reduction benefit of treatment, and an incremental cost of USD $704 (despite a reduction in fracture costs), due to the higher drug cost. Continued denosumab treatment therefore produces an ICER of $16,743 per QALY, indicating that this strategy is cost-effective at a cost per QALY threshold equivalent to Taiwan’s GDP per capita (USD $30,038 [
26]). As suggested by the World Health Organization [
27], in settings without an explicit willingness-to-pay threshold, interventions with an ICER below three times local GDP per capita are commonly considered cost-effective, and those with an ICER below one times GDP per capita considered highly cost-effective. Scenario analysis and probabilistic sensitivity analysis show that cost-effectiveness results are generally stable to variation in model assumptions and parameters.
The modelling approach used in this study is generally consistent with previous work in the field. The model structure has been used to evaluate the cost-effectiveness of denosumab in various settings, including Sweden [
9,
29], the USA [
10], Canada [
30], and Thailand [
17]. Like a number of prior economic evaluations of osteoporosis therapies [
31‐
33], this structure is based on the widely accepted International Osteoporosis Foundation reference model.
To the best of the authors’ knowledge, this study is distinct from previous economic evaluations of denosumab in two ways. First, model inputs were taken from a single real-world source (the NHIRD) wherever possible. In the current study, real-world
effectiveness data were used to inform relative fracture incidence between arms. By contrast, previous economic analyses [
9,
10,
17,
29,
30] used treatment
efficacy data from RCTs, or network meta-analyses of RCTs, which predominantly relied on the FREEDOM trial [
2] (the pivotal phase III denosumab RCT) for the fracture reduction benefit of denosumab. Furthermore, in the current study, fracture rates, treatment discontinuation data, fracture costs, and excess mortality were all informed by the NHIRD, whereas previous evaluations used data from a variety of sources. Using real-world data addresses issues with the generalizability of RCT evidence to clinical practice. For instance, real-world patients are likely to exhibit lower treatment compliance, and are likely to be more diverse in characteristics than participants in RCTs. Additionally, according to reimbursement guidance from Taiwan’s National Insurance Program, only patients with a prior fracture are eligible for osteoporosis treatment [
34]. This is reflected in the substantially higher proportion of patients with a fracture at baseline in the TWRWS, compared with the FREEDOM trial [
2].
A second way in which the current study differs from previous analyses is the choice of comparators. This comparison is consistent with the study design of the TWRWS, which assessed fracture outcomes in patients who received at least two doses of denosumab versus those who discontinued denosumab after one dose to minimize confounding; a real-world study comparing denosumab-treated patients with untreated patients would be subject to bias, since the two patient groups are likely to differ in terms of characteristics and risk factors. To align as closely with the available evidence as possible, the current evaluation modelled the two cohorts as they were defined in the TWRWS study, and therefore assessed the cost-effectiveness of persistence with denosumab, rather than using the “off-treatment” cohort to represent a “no treatment” arm. However, considering that the residual fracture reduction benefit from the initial dose of denosumab in the off-treatment TWRWS cohort is likely to be minimal, outcomes of this evaluation are also likely to represent a reasonable proxy for the real-world cost-effectiveness of denosumab versus no treatment.
As with all economic evaluations, this study has several limitations. First, while model parameters were sourced from NHIRD data wherever available, this was not possible in all cases. Of note, general population health-related quality of life (HRQoL) scores were informed by mainland Chinese data. Although the populations of mainland China and Taiwan are similar, healthcare systems, and potentially health state valuations, differ between the two locations. Therefore, further research is required to produce Taiwan-specific data to inform future economic analyses.
Second, the Taiwan real-world study did not include all types of fragility fracture; only hip, vertebral, humerus, and distal forearm fractures were considered. Although these are regarded as the most important fracture types (in terms of cost and impact on HRQoL and mortality), only including these locations means that the incidence of “other” fractures (i.e., non-hip, non-vertebral fractures) in the model is likely to be underestimated. This omission leads to an understatement of the burden of disease overall, and likely provides a conservative estimate of the cost-effectiveness of continued denosumab treatment; lower fracture rates lead to a smaller absolute fracture reduction for the treatment cohort, and therefore a smaller QALY gain and cost reduction from avoided fractures.
Third, this evaluation assumes that fracture rates in untreated patients remain constant over time, due to a lack of age-specific fracture data from the Taiwan real-world study. In reality, fracture incidence generally increases with age [
28]. In a scenario analysis, the impact of this assumption was explored by adjusting fracture rates for age, using external data on age-specific fracture incidence in Swedish women [
28]. Results of this scenario indicate that the assumption of constant fracture rates is likely to underestimate the cost-effectiveness of continued denosumab treatment.
Fourth, the index date of the Taiwan real-world study does not align exactly with the cycle length of the model. Study outcomes were recorded starting from 6 months and 45 days after the first dose of denosumab, whereas each cycle of the model begins at the time of denosumab administration. However, given the chronic nature of osteoporosis and the relatively long model cycle length, this discrepancy is unlikely to materially affect results.
Finally, as with any evaluation based on observational effectiveness data, it is impossible to rule out residual bias due to unobserved confounders. However, the design of the TWRWS (where both cohorts were selected based on treatment with denosumab, and propensity score matching was conducted using a substantial number of covariates) is likely to minimize the risk of confounding. Information on patients’ smoking and alcohol use, body mass index (BMI), and bone mineral density (BMD) were not available from Taiwan’s NHIRD, and therefore could not be included as covariates in the propensity score matching process. However, the prevalence of smoking and alcohol use is less than 5% among the elderly in Taiwan [
35], and a subset analysis using records from the Chang Gung Research Database [
36] showed that BMI and BMD were balanced between cohorts at baseline. Therefore, it is unlikely that these unobserved variables were associated with a substantial confounding effect.
Despite these limitations, this study provides clear evidence that continued treatment with denosumab is cost-effective compared with discontinuation after one dose, from the perspective of Taiwan’s National Health Insurance system, at a cost per QALY threshold equivalent to GPD per capita. Probabilistic sensitivity analysis and scenario analyses highlight that model conclusions are generally robust to uncertainty. Further research in this area is required, both in providing suitable Taiwan-specific data for future economic evaluations of osteoporosis treatments and in providing real-world evidence of denosumab’s effectiveness in other settings.
Publisher's note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.