Cost-effectiveness of zoledronic acid compared with sequential denosumab/alendronate for older osteoporotic women in Japan

We updated a Markov microsimulation model validated in previous work [2, 11‐13] to perform a cost-effectiveness analysis among hypothetical cohorts of community-dwelling osteoporotic women in Japan without prior fragility fracture, at various ages of therapy initiation (65, 70, 75, and 80 years). We estimated quality-adjusted life-years (QALYs) and total costs in 2020 Japanese yen (¥). For ease of interpretation, we converted these results to US dollars ($) at a rate of ¥105 to $1, which approximates the current exchange rate as of December 2020 [14]. We obtained ICERs, representing cost per QALY gained for one strategy compared with the others, over a lifetime horizon (until a participant reached age 105 years, or died).

Japan developed a universal healthcare insurance system in 1961, and separately launched a mandatory public long-term care insurance system in 2000. Those age 65 and older are eligible for long-term care services, including not only institutional care (e.g., long-term admission or short-term stay in a long-term care facility) but also community- and home-based care (e.g., adult day care, outpatient rehabilitation, home help, or home-visit nursing) [15]. Thus, for the base case, we evaluated cost-effectiveness from the public healthcare and long-term care payer’s perspective (i.e., the perspective of a single payer responsible for both public healthcare costs and long-term care costs) [13]. The public healthcare payer’s perspective (including public healthcare costs, but not including long-term care costs) was adopted as a sub-analysis (Table 1).

The willingness-to-pay threshold was set to ¥5 million ($47,500) per QALY in the base case [2]. In deterministic and probabilistic sensitivity analyses, we also evaluated a willingness-to-pay threshold of ¥10 million ($95,000) per QALY [13, 16]. We discounted all costs and health benefits at 2% per year for the base case [17]. This study’s reporting followed the Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement and recommendations for the conduct of economic evaluation in osteoporosis [18, 19] (Supplemental Tables 1, 2).

We performed an extensive systematic review of all the parameters in the model (Table 2). Inputs were derived from peer-reviewed literature (e.g., meta-analyses of randomized controlled trials or observational studies, individual observational studies, and cost-effectiveness analyses) and websites (e.g., statistics reports from the Ministry of Health, Labour and Welfare, drug prices, and currency exchange rate) that were considered the most relevant (e.g., Japanese population), high-quality, and up-to-date estimates. We used our own assumptions only if no reliable published estimate was available. We used TreeAge Pro Healthcare 2020 (TreeAge Software Inc., Williamstown, MA, USA) to program the model.

Each cycle lasts 1 year, and every participant may sustain a hip or clinical vertebral fracture during each cycle. A participant can sustain only one fracture per cycle and can have a maximum of two hip fractures and an unlimited number of clinical vertebral fractures over the entire time horizon. We used tracker variables for treatments and fractures to incorporate memory of previous events in each individual from one cycle to the next in the model [2, 11‐13].

The target population was postmenopausal osteoporotic women in Japan without prior fragility fracture. Consistent with the Japanese guidelines, osteoporosis was defined as a T-score ≤  − 2.5, or bone mineral density (BMD) ≤ 70% of the young adult mean (YAM), for the lumbar spine or hip (either femoral neck or total hip) as measured by dual-energy X-ray absorptiometry (DXA) [20]. In the model, individuals had already received a DXA scan and had been diagnosed with osteoporosis based on the DXA scan result.

We compared the cost-effectiveness of intravenous zoledronic acid annually for 3 years compared with sequential denosumab/alendronate (i.e., subcutaneous denosumab every 6 months for 3 years followed by weekly oral alendronate for 3 years). Data from a recent systematic review and network meta-analysis of randomized controlled trials were used to obtain the efficacy of zoledronic acid, denosumab, and alendronate compared with placebo in reducing the risks of fragility fractures for those with osteoporosis [21]. Persistence rates with zoledronic acid and denosumab were based on meta-analyses of observational studies [22]. These meta-analyses showed that the cumulative persistence rates with zoledronic acid with a permissive gap of 90 days at the second dose (i.e., getting at 12 months, lasting till 24 months) and the third dose (i.e., getting at 24 months, lasting till 36 months) were 52 and 36%, respectively. As zoledronic acid is given once yearly, we assumed that the persistence rate with zoledronic acid was 100% at the end of the first year. The same study showed that, allowing for a permissive gap of 60 days, the cumulative persistence rates with denosumab at the second dose (i.e., getting at 6 months, lasting till 12 months) at the fourth dose (i.e., getting at 18 months, lasting till 24 months) and at the sixth dose (i.e., getting at 30 months, lasting till 36 months) were 81, 55, and 26%, respectively. The cumulative persistence rate at the second and fourth doses was based on a meta-analysis of 14 and 9 studies, respectively. The rate at the sixth dose was, however, based on a single study performed in Czech Republic rather than a meta-analysis of multiple studies, which raises the possibility that the result is not generalizable. Therefore, we conservatively assumed the same persistence rate from the second dose to the fourth dose applied to the persistence rate from the fourth dose to the sixth dose, making the rate at the sixth dose 37% (the 95% CI for this latter rate was estimated based on the 95% CI at the second year) [22].

A small retrospective observational study at a single institution in Japan (n = 102) showed cumulative persistence rates with denosumab of 94 (year 1), 92 (year 2), and 87% (year 3) for ages 65, 70, and 75, and 83 (year 1), 71 (year 2), and 59% (year 3) at age 80 [23]. This Japanese study was published after the meta-analysis had been conducted and was not included in the meta-analysis of the persistence rates of denosumab above [22]. As this small single-center study seemed to lack generalizability and we also could not find a counterpart study that examined the cumulative persistence rates of zoledronic acid in the Japanese setting, we evaluated these higher cumulative persistence rates of denosumab in a deterministic sensitivity analysis. In an additional deterministic sensitivity analysis, higher cumulative persistence rates of zoledronic acid were included in addition to the higher cumulative persistence rates of denosumab based on this study, assuming that the same ratios of the cumulative persistence rates of zoledronic acid to denosumab based on the previously noted meta-analysis were applied [22]. Those persistent with zoledronic acid or denosumab with the permissive gaps above were by definition also adherent to the medication.

The cumulative persistence rates with weekly bisphosphonates were estimated to be approximately 55, 39, and 28% at the end of the first, second, or third year, respectively, with a permissive gap of 30 days, and the adherence rates with weekly bisphosphonates were estimated to be 70.6 and 60.9% in the first and fifth year, respectively [24]. We assumed a linear decline in the adherence rates between the first and fifth year. Adherence rates with oral bisphosphonates were higher in clinical trials (mostly greater than 80%, as high as 100%) than observational studies that reflected actual clinical settings [25]. We estimated the relative effectiveness of alendronate in the community by assuming a linear relationship between relative risk reduction and adherence [2, 11‐13].

We assumed that zoledronic acid and alendronate had efficacy from the first year through the end of the duration of the treatment (i.e., 3 years) and the risk for fractures after completing therapy returned to rates in the absence of the treatment after the same number of years as the treatment was given, in a gradual linear fashion (i.e., offset effects were assumed to be proportional to the treatment periods’ length) [2, 11‐13]. Accumulating evidence has shown that after discontinuing denosumab, there is an increased risk of vertebral fractures within a relatively short period compared with those who continued denosumab [3‐5]. Therefore, we assumed that the therapeutic effect of denosumab wore off rapidly at the end of the cycle when one received the final dose, and returned to the baseline risk without treatment. To keep the model parsimonious, we assumed that each individual obtained benefits of fracture prevention if she persisted in taking the treatment at the end of each cycle (i.e., 1 year). Those who were not persistent with denosumab did not start on alendronate after they stopped taking denosumab in this model.

We used the EuroQol five dimension scale (EQ-5D) and assumed that disutilities (i.e., losses in health-related quality of life) associated with hip and clinical vertebral fractures were highest in the year immediately following the fracture, but persisted for the rest of life [32‐34].

We divided costs into formal healthcare sector and non-healthcare sector costs and provided an impact inventory (Table 1) [35]. We assumed that costs were identical regardless of age.

For base case analyses, we ran the model with 100,000 trials (100,000 individuals through the model one at a time). We performed deterministic (one-way) sensitivity analyses to evaluate the robustness of the results across a range of values for critical model parameters (Table 2). Although these treatments’ optimal durations have not been determined, those at high risk for osteoporotic fractures may benefit from up to 6 years of use of zoledronic acid, or up to 10 years’ use of denosumab or alendronate. We therefore performed a deterministic sensitivity analysis with longer durations of treatments (i.e., zoledronic acid for 6 years on, 2 years off, and 6 years on versus denosumab for 10 years followed by alendronate for 10 years). For this analysis, we assumed the same relative rate of persistence with zoledronic acid or denosumab in the fourth year and beyond as in the third year. This resulted in cumulative persistence rates of zoledronic acid decreasing to 25, 17, and 12% in the fourth, fifth, and sixth year, respectively, and cumulative persistence rates with denosumab decreasing to 25, 17, 12, 8, 5, 4, and 2% in the fourth through tenth years, respectively. We assumed that those who took alendronate for 7 years continued to take alendronate for up to 10 years (i.e., no dropout from eighth year onward except for death) with the same adherence rate as the fifth year from the sixth year onward [24]. We also performed two additional sensitivity analyses, in which (1) we assumed the same excess mortality associated with clinical vertebral fracture as with hip fracture, and (2) we assumed that those who received either zoledronic acid or denosumab had physician visits and blood tests twice a year.

Next, we performed probabilistic sensitivity analyses, in which parameter values were randomly selected from their probability distributions for uncertain key model inputs. Monte Carlo simulation was performed with 1000 simulations and 100,000 trials per simulation. To verify the model’s accuracy, we initially included a “no-intervention” arm in calculating mortality and fracture rates in the model.

Our model predicted that the probabilities of dying by age 105 with different starting ages (i.e., 65, 70, 75, or 80) were greater than 99%, consistent with the 2018 Japanese life table [29]. Our model also predicted that without treatment, the probabilities after the starting ages of having at least one hip fracture were 21% at ages 65 or 70, and 20% at ages 75 or 80, respectively. The probabilities of having at least one clinical vertebral fracture were 44% at age 65 or 70, 43% at age 75, and 39% at age 80, respectively.

Zoledronic acid was cost-saving (i.e., more effective and less expensive) compared with sequential denosumab/alendronate at all ages examined. Costs and effectiveness at various starting ages were as follows: age 65, $23,710 and 14.219 QALYs for zoledronic acid and $24,160 and 14.218 QALYs for denosumab/alendronate; age 70, $24,050 and 11.628 QALYs for zoledronic acid and $24,540 and 11.626 QALYs for denosumab/alendronate; age 75, $22,930 and 9.201 QALYs for zoledronic acid and $23,500 and 9.196 QALYs for denosumab/alendronate; and age 80, $19,650 and 6.942 QALYs for zoledronic acid and $20,290 and 6.935 QALYs for denosumab/alendronate. From the public healthcare payer’s perspective, the conclusions remained the same (Table 3).

Results were sensitive to changes in the efficacy of zoledronic acid for reduction of clinical vertebral or hip fracture. The ICERs for denosumab/alendronate compared to zoledronic acid became less than the willingness-to-pay threshold of ¥5 million ($47,500) per QALY only at ages 75 and 80 with lower efficacy of zoledronic acid for reduction of clinical vertebral fracture (Fig. 2). Results were also sensitive to cumulative persistence rates (Table 4). If the cumulative persistence rates of denosumab were based on a small observational study in Japan (a rate higher than the upper value of the 95% CI of the meta-analysis results used for the base case), the ICERs became less than the willingness-to-pay threshold of 5million/QALY at ages 70 and 75. If we further assumed that the cumulative persistence rates of both zoledronic acid and denosumab were higher based on the Japanese study, the ICERs became greater than the willingness-to-pay threshold of ¥5 million/QALY at ages 70 and 75. Otherwise, in deterministic sensitivity analyses, zoledronic acid remained cost-saving compared with sequential denosumab, except that for some unfavorable values of zoledronic acid at age 65, zoledronic acid was not cost-saving, but denosumab/alendronate did not become cost-effective even at the willingness-to-pay threshold of ¥10 million/QALY. Finally, at all ages examined, zoledronic acid remained cost-saving with the specified sensitivity analyses for longer planned durations of treatment, assuming the same excess mortality associated with clinical vertebral fracture as with hip fracture, or assuming those who took either zoledronic acid or denosumab had physician visits and blood tests twice a year.

The probabilities of zoledronic acid being cost-effective were 100% for ages 65 and 70, and 98% for ages 75 and 80, respectively, at a willingness-to-pay threshold of ¥5 million ($47,500) per QALY. The probabilities were 100% for age 65, 99% for age 70, and 97% for ages 75 and 80, respectively, at a willingness-to-pay threshold of ¥10 million ($95,000) per QALY.