Cost-effectiveness of buffered soluble alendronate 70 mg effervescent tablet for the treatment of postmenopausal women with osteoporosis in Italy

A Markov microsimulation model with a 6-month cycle length was used to allow tracking patient characteristics and individual disease histories (e.g. fractures) and avoid unnecessary transition restrictions. The model health states were ‘no fracture’ (where all individuals begin), ‘death’, ‘hip fracture’, ‘clinical vertebral fracture’, ‘wrist fracture’ and ‘other fracture’. The ‘other fracture’ state includes other osteoporotic fractures as defined by the IOF-EFPIA report [17]. Patients could experience multiple fractures at the same site or multiple sites. Discount rates of 3% for both costs and health benefits were used in line with the Italian guideline for economic evaluations.

Analyses were conducted at different age (60–80 years) in two populations with high risk of fragility fractures in line with reimbursement conditions in Italy (Nota 79): (a) in postmenopausal female patients with existing vertebral fracture, and (b) those with osteoporosis as defined by T-score ≤ − 3.0 at the femoral neck and without vertebral fracture.

Information concerning all hospitalizations occurring in Italy is registered in hospital discharge records, which are collected at central level by the Italian Ministry of Health (National Hospitalization Database). The incidence of hip fractures in the general Italian population was derived from this database for the year 2017 for different age groups (65–69, 70–74, 75–79, 80–84, 85–89, 90–94 and 95+ years) [18]. As these recent data did not include women aged lower than 65 years, the ratio of hip fractures between 60–64 and 65–69 years from Piscitelli et al. [19] for the year 2008 was applied to derive incidence rate for the age group 60–64 years. To estimate the incidence of clinical vertebral and wrist fractures, ratios between hip fracture and clinical vertebral/wrist fractures from an older Italian database from the year 2008 were applied [20]. As no data for other fractures are available at national level in Italy, we applied the age-specific ratio incidence from other countries in line with the methodology used by the IOF-EFPIA report [17].

Initial probabilities were then adjusted to accurately reflect the fracture risk in the target population in comparison with that of the general population using previously validated methods [21, 22]. The methods calculated the relative risk of individuals below the threshold value (i.e. BMD T-score ≤ − 3.0) and of individuals with prevalent vertebral fractures compared with that of the general population. Fracture risk was also adjusted when a new fracture occurred during the simulation process in line with studies suggesting an increased risk for fracture after a prior fracture [21].

The age-specific mortality rates for the general population were derived from the Italian Institute of Statistics for the year 2018. An increased mortality after hip fracture and clinical vertebral fracture was modelled in line with previous studies [23]. Because excess mortality may also be attributable to comorbidities, we further took into account that only 25% of the excess mortality following fractures was attributable to the fractures themselves [24, 25].

The healthcare decision maker perspective was used for the cost estimation. All costs were expressed in €2019 and adjusted using the national price index. The cost of hip fractures was derived from the study of Piscitelli et al. with data from the year 2014 [26]. In the absence of local data, the costs of non-hip fractures were quantified relative to hip fracture in line with the assumption used in the IOF-EFPIA report [17]. This assumption is conservative compared to a previous cost-effectiveness study conducted in Italy [27]. Long-term hip fracture costs were based on the proportion of patients being institutionalized following the hip fracture, estimated at 10% in Italy [19, 26], while the annual cost of being in the nursing home was estimated at €41,300 [28].

Utility values were derived from the International Costs and Utilities Related to Osteoporotic Fractures Study (ICUROS) study [29]. This study is the largest study assessing the quality of life of patients with fractures from 11 countries including 2,808 patients, and collected EQ-5D data at different time points after fractures and a recall before fractures. Baseline utility data (for patients without fractures) were derived from the utility estimation before fractures. Decreases in quality of life as a consequence of fracture were included as utility multipliers for the first year following fracture and in subsequent years. Since other fractures were not included in the ICUROS study, estimate from a previous systematic review was used [30]. An additional effect on utility after multiple fractures was modelled [13, 14].

In base case, the effects of treatment on the risk of fractures (expressed as relative risks, RR) were derived from the most recent network meta-analysis of the National Institute for Health and Care Excellence (NICE) in the UK (ta464) [31]. In the absence of studies suggesting a clear and significant difference between bisphosphonates, and in line with the NICE committee discussion regarding bisphosphonates (appraisal TA464) suggesting that the efficacy estimates of the oral and intravenous bisphosphonates should be pooled for each fracture site, results of the network meta-analysis of pooled bisphosphonates data were used for all bisphosphonates including buffered soluble ALN, generic ALN and zoledronic acid. After stopping medication, it was assumed a linear decrease of the effects for a duration similar to the duration of therapy, in line with previous economic analyses of oral bisphosphonates [10] and clinical data [32]. The effect of denosumab on fracture risk was derived from the FREEDOM trial [33]. In line with recent evidence suggesting an immediate bone loss after denosumab discontinuation [34], an immediate loss of treatment effect following treatment discontinuation was assumed.

Two treatment duration scenarios (1 and 3 years) were assessed and adjusted by medication persistence using a previously used methodology [35, 36]. Medication persistence was measured as the probability to be on treatment at different time points, and has been shown to be the driver of adherence in economic evaluations in osteoporosis [37]. In the first scenario, treatment duration was set-up to 1 year in line with persistence data available for buffered soluble ALN derived from an Italian prospective observational study [38] (see later). In line with previous literature [39‐41], it was assumed that 90% and 80% of patients under denosumab are persistent at 6 and 12 months respectively. For patients who discontinued therapy, treatment cost was stopped and the offset time period started immediately. For those who discontinued oral/effervescent bisphosphonates within 6 months, no treatment effect was received, because at least 6 months of treatment is necessary to reduce the risk of fractures. In the second scenario, a maximum of 3-year treatment duration was assumed for all treatments to better reflect clinical practices. Real-world persistence data with oral bisphosphonates were derived from a recent systematic review suggesting that the mean persistence was 53% at 6 months, 46% at 1 year, 37% at 2 years and 31% at 3 years [42]. Another systematic review of articles published up to September 2018 identified ten studies reporting zoledronic acid persistence rates with mean persistence rates of 52% and 36% for second and third dose, respectively [41]. Denosumab was reported in 19 studies, with mean persistence rates of 81%, 55% and 35% at second (year 1), fourth and fifth doses (year 2 and year 2.5) [41]. The (relative) positive effect of buffered soluble ALN on persistence at year 1 compared to oral ALN was assumed to be maintained up to 3 years, leading to persistence rates for buffered soluble ALN of 83% at 6 months, 67% at 1 year, 61% at 2 years and 58% at 3 years. Persistence data for both treatment duration scenarios could be found in Table 2.

The treatments cost included drug costs and costs for assessment. Drug costs were derived from official listings from February 2020. We also assigned the cost of one physician visit (€20.66) every 6 months of treatment (for persistent patients) and the cost of one bone density measurement (€43.36) at the start of treatment.

In line with previous economic analyses, the risks of gastrointestinal effects with alendronate and cellulitis with denosumab were considered. The assumptions relating to gastrointestinal effects were chosen to be similar to those used by the NICE appraisal. It was estimated that patients treated with ALN required 0.041 extra physician consultations during the first cycle (6 months) and 0.021 physician consultations during the following cycles on treatment, as well as a proton-pump inhibitor (PPI) for each visit. Despite studies suggested lower frequency of GI disorders with buffered soluble ALN, we conservatively assumed than buffered soluble ALN and generic ALN are associated with similar side effects in the base case. The rate of skin infections, including cellulitis, was reported more frequently with denosumab in the FREEDOM trial, i.e. 0.0031 annually, and was included in the analysis.

The Italian observational study included postmenopausal women from a standardized clinical database with BMD T-score < − 2.5, or between − 2 and − 2.5 and at least one vertebral fracture, starting buffered soluble ALN between July 2015 and June 2016. A historical cohort comprised of randomly selected and age-matched women on conventional ALN tablet was used as a control. Persistence at 6 and 12 months was estimated between the two groups.

A total of 1,000,000 trials was run for each analysis. Total costs, disaggregated costs (i.e. treatment costs and fracture-related costs) and accumulated QALYs were estimated for each treatment. If buffered soluble ALN is associated with more QALYs and lower costs than an alternative treatment, buffered soluble ALN is considered dominant or cost-saving (when compared to no treatment). If buffered soluble ALN provides more QALYs and more costs, then we computed the incremental cost-effectiveness ratio (ICER) defined as the difference between buffered soluble ALN and the comparator treatment in terms of total costs (expressed in €2019) divided by the difference between them in terms of QALYs. If the ICER is above the cost-effectiveness threshold, then the cost is too high for the benefits and the intervention is not considered as cost-effective. In Italy, no specific threshold is actually used for defining cost-effectiveness. Borgström et al. [43] have suggested a threshold for QALY equal to two times the gross domestic product per capita for industrialized countries (±€70,000 in Italy). This assumption has been used for defining fracture risk thresholds in several countries [44].

Sensitivity analyses were then performed to assess the impact of model parameters on the results. One-way sensitivity analyses assessed the impact of single parameters on the results and were conducted on discount rates, fracture costs, fracture risks, fracture disutility, mortality and treatment costs. Sensitivity analyses were also conducted on varying by ± 25% the discontinuation rates of buffered soluble ALN and by reducing by half side effects associated with buffered soluble ALN. We also conducted another sensitivity analysis where treatment-specific efficacy data were derived from other sources. For ALN (used in our study for both generic ALN and buffered soluble ALN), data from the article of Black et al. (2000) were used [45]. Data of women with existing vertebral fracture were specifically used for this population, and data from women without vertebral fracture and femoral T-score < − 2.5 were used in the model for women with a T-score < − 3.0. The effects of zoledronic acid were derived from the HORIZON-FT trial [46], and the effects of denosumab were derived from a post hoc analysis of the FREEDOM trial in women with higher risk of fractures [47].

Finally, probabilistic sensitivity analyses were undertaken to examine the effect of the joint uncertainty surrounding the model variables. Nearly all parameters were modified simultaneously over plausible range of values, following guidelines and in line with previous studies (see Appendix 1 Table 2). For each probabilistic sensitivity analysis, the model was run 200 times based on runs of 50,000 patients per treatment arm. Results were presented in the form of cost-effectiveness acceptability curves that show the probability of being cost-effective as a function of the decision maker’s willingness to pay per QALY gained.

A total of 360 postmenopausal women were included in the retrospective study; 144 were treated with buffered soluble ALN and 216 with conventional ALN tablet. The study revealed that a significantly higher number of women were persistent at 6 months and 12 months with buffered soluble ALN (91% and 81% respectively) compared to 75% and 69% of patients with conventional ALN tablet, respectively.

Table 3 presents the total and disaggregated healthcare costs, accumulated QALYs and the ICER (expressed in cost per QALY gained) of buffered soluble ALN compared with no treatment, generic ALN, denosumab and zoledronic acid in women aged 70 years. In the 1-year treatment scenario, in women with prevalent vertebral fractures, the incremental treatment cost (including drug cost adjusted by persistence and monitoring costs) between buffered soluble ALN and generic alendronate was €50, while the improved persistence of buffered soluble ALN leads to a saving of €39 resulting from more prevented fractures due to the improved persistence. The incremental total healthcare cost was thus estimated at +€11 (50–39) and buffered soluble ALN was associated with a QALY gain of 0.0028. The cost per QALY gained of buffered soluble ALN compared to generic alendronate was thus estimated at €4,028 (11/0.0028) per QALY gained. Compared to DMAB, buffered soluble ALN was associated with more QALYs and lower costs, being therefore dominant. Zoledronic acid is associated with more QALY than buffered soluble ALN; however, the cost per QALY gained of zoledronic acid (€121,514 per QALY gained) is higher than commonly accepted cost-effectiveness thresholds, meaning that zoledronic acid is not cost-effective compared to buffered soluble ALN. In women with BMD T-score ≤ − 3.0, buffered soluble ALN was dominant (more QALYs and lower costs) than generic alendronate and denosumab. In addition, buffered soluble ALN was cost-saving (more QALYs, lower total costs) compared to no treatment, and the cost per QALY gained of zoledronic compared to buffered soluble ALN was also higher than cost-effectiveness threshold, meaning that buffered soluble ALN is the most cost-effective option. In the 3-year treatment scenario, buffered soluble ALN was shown to be dominant compared to both denosumab and zoledronic acid in both populations. Buffered soluble ALN was also shown to be cost-saving and dominant compared to generic ALN in women with BMD T-score ≤ − 3.0.

In Table 4, the ICERs of buffered soluble ALN compared to all alternative treatments are presented for other ages ranging from 60 to 80 years. Appendix 2 Tables 1and 2 a-f provide the lifetime costs and QALYs for all these age-specific simulations. Compared to denosumab, buffered soluble ALN was always dominant (more QALYs, lower costs). The cost per QALY gained of buffered soluble ALN compared to generic ALN and no treatment falls always below €20,000 per QALY gained. In women aged 75 years and older with prevalent vertebral fractures and in women aged 65 years and older with T-score ≤ − 3.0, buffered soluble ALN was even shown to be dominant (more QALYs, lower costs) compared to generic alendronate and no treatment. In the 1-year treatment scenario, zoledronic acid was associated with more QALY than buffered soluble ALN but the cost per QALY gained of zoledronic acid compared to buffered soluble ALN was always higher than €70,000 per QALY gained, meaning that zoledronic acid is not cost-effective, while buffered soluble ALN was shown to be dominant compared to zoledronic acid in the 3-year treatment scenario.

Table 5 reports the results of the one-way sensitivity analyses in women aged 70 years with BMD T-score ≤ − 3.0 in the 1-year treatment scenario. In most analyses, buffered soluble ALN remained dominant (more QALYs, lower costs) compared to generic ALN and denosumab, and cost-saving compared to no treatment. The ICERs of buffered soluble ALN were shown to be in particular affected by fracture costs and discontinuation rates of buffered soluble ALN. In all the sensitivity analyses, buffered soluble ALN remained cost-effective compared to all treatments except when using other treatment-specific efficacy data where the ICER of zoledronic acid fall below the threshold of €70,000 per QALY gained.

The results of the probabilistic sensitivity analyses are provided in Fig. 1 where the cost-effectiveness acceptability curves show the probability that each intervention is cost-effective for different willingness to pay of decision makers per QALY gained. The curves suggest that buffered soluble ALN is the most cost-effective intervention for willingness to pay between €5,000 and €75,000 per QALY gained in the 1-year treatment xscenario. At a threshold of €45,000 per QALY gained, in women aged 70 years with BMD T-score ≤ − 3.0, buffered soluble ALN was cost-effective in 56% of the simulations compared to 6% for DMAB, 10% for generic alendronate and 26% for zoledronic acid. In the 3-year treatment scenario, buffered soluble ALN was the most cost-effective treatment for any willingness to pay.

Cost-effectiveness acceptability curve of buffered soluble ALN compared to individual treatments could be found in Appendix 2 Fig. 2 a, b, c, d for women with BMD T-score ≤ − 3.0 aged 70 years. Buffered soluble ALN was the most cost-effective intervention compared to all individual treatments. By example, at a threshold of €45,000 per QALY gained, in the 1-year treatment scenario, buffered soluble ALN was cost-effective in 100% of the simulations compared to no treatment, in 84% compared to generic alendronate, in 92% compared to denosumab and 71% compared to zoledronic acid.