Cost-effectiveness of strontium ranelate in the treatment of male osteoporosis

The simulation model is the same as the model developed for postmenopausal osteoporotic (PMO) women which has been validated [18] and used in many published health economic analyses [12, 13, 19‐23]. Recently, an updated version of the model using a 6-month cycle length has been developed [23]. This last model version was slightly revised in this study to also include a health state for venous thromboembolic events (VTEs). The model was programmed using the software TreeAge Pro 2011 (TreeAge Pro Inc., Williamston, MA, USA).

The Markov model health states are no fracture, death, VTE, hip fracture, clinical vertebral fracture, wrist fracture, other fracture and the corresponding post-fracture states. Post-fracture states were created as some parameters (e.g., fracture disutility) were estimated over a 1-year period [23]. All patients, one at a time, began in the ‘no fracture’ state and had, every 6 months, a probability of having a fracture at the hip, clinical vertebrae, wrist, or other site or of dying. Patients in a fracture state can stay in the same fracture state if they re-fracture, change to another fracture state, die or change in the next cycle to the post-fracture state. Because hip fracture is associated with extra costs in the year following the fracture that are greater than the hospitalization cost of any other fractures, patients who have had a hip fracture were only at risk for another hip fracture or dying in the first cycle following the fracture. Patients being in any post-fracture state might have a new fracture (all fracture types are possible), die or move to the ‘no fracture’ state. The probability for patients to move to the VTE health state was also considered under treatment with strontium ranelate.

A description of the different components of the model is provided below. Model data are included in Table 1. Readers are also referred to previously published research for further details and limitations of the model [17].

The incidence of hip fractures in the general men population was derived from the national database of hospital bills (average of the years 2005–2007) [2]. Since the incidence of other fractures was not known, we assumed that the age-specific ratio of index fracture to hip fracture in Belgium was the same as found in Sweden [3]. This assumption, used in the development of many FRAX® models including Belgium [24], appears to be appropriate for West European countries, the USA and Australia [25].

Age-specific mortality rates were obtained from the National Institute of Statistics. According to data from a recent meta-analysis [4], hip fractures increased male death probabilities by 5.75 in the first 6 months following the fracture, by 2.315 in the period 6–12 months and by 1.691 in subsequent years. As the increased mortality following clinical vertebral fractures has been found in many studies to be very similar than those of a hip fracture [26‐29], the same impact was assumed after hip and clinical vertebral fractures. To avoid an overestimation of the beneficial effect of treatment on mortality, it is important to take only into account excess mortality that are directly or indirectly attributable to the fractures themselves [30], which could be reduced through fracture prevention. Because excess mortality may also be attributable to comorbidities, we assumed in the model only 25 % of the excess mortality after fractures [28, 31].

A healthcare payer perspective including direct medical costs was adopted for all cost estimates, as recommended for pharmacoeconomic evaluations in Belgium [32]. Following the guideline, direct healthcare costs paid by the national health insurance and patient's out-of pocket costs were included [32]. All costs were expressed in the year 2010 using the healthcare product price index when necessary, and discount rates of 3 % for costs and of 1.5 % for health benefits were assumed for the base-case analysis also based on the Belgian guideline for pharmacoeconomic evaluations [32]. The direct hospitalisation cost of hip fracture, administrated in the first cycle following the fracture, was retrieved from the Belgian national database of hospital bills for the year 2007 [33]. It included the social security cost and the patient out-of-pocket contribution for nursing and residential fees costs only. Extra costs in the year following the hip fracture were derived from the study of Autier et al. [34], which based on a prospective controlled study including 159 women. These costs, estimated at €8,001 (expressed in €2,010), were equally distributed between the two first cycles following the fracture. Hip fractures are also associated with long-term costs. They were based on the proportion of men being institutionalized following the fracture, ranging from 6 % (for men aged 60 years) to 65 % (for those aged over 90 years) [35]. Because men might be institutionalized later in life, regardless of their hip fracture, an adjustment was made to only include long-term costs attributable to the fracture itself (see Hiligsmann et al. [18] for further explanations). The cost of non-hip fractures has never been estimated in Belgian men and these were quantified relative to hip fracture cost [36]. So, the costs of clinical vertebral, wrist and other fracture represent 17.4 %, 14.5 % and 17.4 % of the hip fracture cost, respectively. Non-hip fractures were not associated with long-term costs.

Utility values in the general men population as well as relative reductions due to fractures in the year following the fracture and in subsequent years were derived from a systematic review, which suggested reference values for countries that do not have their own database [37]. The reduction of quality-adjusted life-year (QALY) depends on fracture site but also on the number of prior fractures [18]. In the case of an occurrence of a second fracture at the same site, the impact of the first fracture event was reduced by 50 % [18]. For example, if a men with a prior hip fracture suffered another fracture, the relative reduction of utility attributable to the first hip fracture was then 0.95. For an individual with both a hip and vertebral clinical fracture, the total impact on QALY was assumed to be equal to the sum of the impacts related to each of the fractures. This last assumption is consistent with the study of Tosteson et al. [38], who suggested that the impact of the two fractures is even greater than the sum of the impacts related to each of the fractures. The model, however, does not simulate multiple fractures per 6-month cycle.

Analyses were conducted in the population from the MALEO Trial corresponding to men with mean age of 73 years, and with a bone mineral density (BMD) T-score below the threshold value for osteoporosis (i.e., BMD T-score ≤−2.5) or PVFs at baseline, in order to match the two populations for whom postmenopausal osteoporosis medications are currently reimbursed in Belgium and in most European countries. The MALEO Trial included in the Full Analysis Set (FAS) 243 men aged 65 to 90 years with osteoporosis as assessed by a mean lumbar spine BMD T-score of −2.7 [15]. The mean BMD at the femoral neck was 0.627 (g/cm²), which corresponds to a T-score of approximately −2.2.

The incidence of fracture in the general population has to be adjusted to accurately reflect the fracture risk in these populations. The relative risks of fracture were calculated from the BMD and the prevalence of vertebral fracture in the target patient groups. The relative risk for BMD was calculated using a method previously described [25]. This method estimates the risk of individuals at a threshold value or below a threshold value in comparison with that in the general population. BMD values at the femoral neck were derived from the National Health and Nutrition Examination Survey (NHANES) III [39] database and 1 standard deviation decrease in BMD was associated with an increase in age-adjusted relative risk of 1.8, 1.4 and 1.6 for clinical vertebral, wrist and other fracture, respectively [40]. For hip fracture, the age-adjusted relative risk ranged from 3.68 at 50 years to 1.93 at 85 years [41]. So, for example, the relative risks of fracture, for men aged 73 years with a BMD of 0.627 (g/cm²) at the femoral neck, were estimated at 1.683, 1.529, 1.330 and 1.440 at the hip, clinical vertebral, wrist and other fracture, respectively. The relative risks for men with PVFs were taken from a meta-analysis and were 2.3, 4.4, 1.4 and 1.8 for hip, clinical vertebral, wrist and other fractures, respectively [42]. These relative risks were reduced by 10 % each per decade above the age of 70 years [43]. An increased risk of subsequent fractures was also modelled during the simulation for men who have a prior fracture of the same location, using a previously described method [18].

The MALEO Trial has been developed in accordance with European guideline on clinical investigation of medicinal products (November 2006). This guideline deals with minimal requirement for marketing indication of a treatment in osteoporosis in men at increased risk fracture once the marketing indication in PMO women has been already granted to the same drug. The MALEO Trial is a controlled study versus placebo on the basis of calcium/vit D supplementation with BMD measure as primary efficacy criteria and a main analysis after 1 year.

In the MALEO Trial [15], a marked increase in the mean lumbar L2–L4 and femoral neck BMD was observed in men with high risk of fractures, similar to that previously observed in women (Table 2). Considering these results and the previously established relationship between change in BMD and reduction in the risk of vertebral and hip fractures with strontium ranelate in women [44, 45], a similar anti-fracture efficacy is expected in men. We therefore assumed, in the base-case analysis, the same relative risk reduction of fractures in men as those estimated in women (SOTI and TROPOS trials).

In most cost-effectiveness analyses, efficacy data were retrieved from the entire population of the randomized clinical trials and the modelers charged the full treatment cost. Although, in real-life settings, adherence is far from optimal, this assumption may be incorrect to estimate the potential economic value of a drug and probably underestimates the true underlying risk reduction with therapy since the efficacy from these trials is reduced to some degree because of non-adherence. In the SOTI and TROPOS trials, adherence was collected and per protocol studies including patients with high level of adherence were conducted. The Per Protocol Set strontium (PPS strontium) included all patients from the FAS satisfying a minimum exposure condition based on blood strontium levels criteria. In this analysis, efficacy data from intent-to-treat [5, 7] and per-protocol analyses (unpublished data, internal reports SOTI and TROPOS 3-year results) were both tested.

In the base-case analysis, fracture risk reductions were derived from the FAS of the TROPOS and SOTI trials. Strontium ranelate was assumed in this scenario to reduce the risk of hip, wrist and other non-vertebral fractures by 19 % (RR=0.81; 95 % confidence interval [CI], 0.66–0.98) using the estimated fracture risk reduction for major non-vertebral fractures [7] and the risk of clinical vertebral fracture by 38 % (RR=0.62; 95 % CI, 0.47–0.83) [5]. We took a conservative position for the efficacy of strontium ranelate on hip fracture since the results of a post hoc analysis in high-risk women aged over 74 years of age was not incorporated [7]. In the additional scenario, the efficacy of strontium ranelate on non-vertebral fractures was derived from the per-protocol study of the TROPOS Trial including 2,935 osteoporotic women above 70 years of age with high adherence. In this population, strontium ranelate was shown to reduce the risk of hip fracture, as compared to placebo and over 3 years, by 41 % (95 % CI, 5–63 %; p=0.025). The risk of any major non-vertebral fractures, used in the model for wrist and other fractures, was reduced by 35 % (95 % CI, 16–49 %; p<0.001) in the same population. In the per-protocol study conducted in the SOTI trial and including 1,076 women with a mean age of 69 years, the risk of vertebral fracture was reduced by 45 % (95 % CI, 25–57 %; p<0.001).

Patients received treatment in the base-case model for 3 years with the full effect of the treatment during the whole intervention period. After stopping therapy, the effect of strontium ranelate on fracture risk was assumed to decline linearly to zero for a period (called offset time) similar to the duration of therapy in line with a clinical study [46] and prior cost-effectiveness analyses [14]. In a sensitivity analysis, we assessed the impact of poor adherence with strontium ranelate using the same assumption than in prior cost-effectiveness analyses of strontium ranelate in postmenopausal women [12, 13]. In these analyses, adherence to strontium ranelate was similar to that observed for bisphosphonate therapy in Belgian women [47]. We therefore assumed that 30 %, 12 %, 18 % and 15 % of patients discontinued therapy after 3 months, 6 months, 1 year and 2 years, respectively. No treatment effect was assumed for patients who discontinued treatment at 3 months and offset time for non-persistent patients was assumed to be the same as their treatment period. Compliance was estimated at 70.5 % for persistent women [47]. Medication costs and fracture reduction efficacy were assumed to be proportional to compliance.

The annual cost of strontium ranelate was estimated at €477.2 (Protelos®, €109.82 for a package of 84 sachets) [48], and we assigned the cost of one physician visit (€22.67) per year of treatment and the cost of one bone density measurement (€58.05) every second year. Adverse events with strontium ranelate are usually mild and transient. In pooled data from the SOTI and TROPOS trials [5, 7], treatment with strontium ranelate, however, was associated with an increase in the annual incidence of VTE, including pulmonary embolism (PE). To account for this in the analysis, VTE was included as a health state in the model during treatment with strontium ranelate. The annual absolute risk of VTE with strontium ranelate was estimated at 0.31 % in women [5, 7]. In the model, VTE was assumed to be associated with a 10 % utility loss the first year after the event and any utility loss in the second or following years after the event, in agreement with previous health economic publications [49, 50]. The survival rate after PE was estimated at 81.6 % in the clinical trials [5, 7]. Using Belgian estimates of resource utilization based on panel experts [51], the cost of VTE was estimated at €2,622.

Microsimulations were performed to estimate the cost-effectiveness of strontium ranelate. Each model was run ten times with 200,000 trials (patients) to guarantee the stability of the results and enable variability analyses [23]. For each analysis, the incremental cost-effectiveness ratio (ICER) was computed as the difference between strontium ranelate and no treatment in terms of total costs (expressed in €2,010) divided by the difference between them in terms of effectiveness, expressed in accumulated QALYs. It represents the cost of strontium ranelate (compared with no treatment) per one QALY gained. In Belgium, as in many other countries, no threshold values for ICERs have been defined [52]. Commonly accepted thresholds for cost-effectiveness are in the range of €50,000 [11].

Uncertainty related to model parameters and assumptions was investigated using deterministic and probabilistic sensitivity analyses. Deterministic sensitivity analyses were performed to evaluate the impact of single parameter variations on the results. The baseline parameters for discount rates, fracture risk, fracture disutility, fracture cost and excess mortality were varied over plausible ranges. Changes in therapy cost, monitoring cost, adverse events, offset time and time horizon were also evaluated.

Probabilistic sensitivity analyses were performed with 200 simulations to analyze the effects of uncertainty in all model parameters simultaneously. Distributions used for key model inputs are provided in Table 1. Log-normal distributions were also assumed for fracture risk reduction with strontium ranelate. Cost-effectiveness acceptability curves were then constructed from the incremental cost and QALYs between alternatives for the 200 simulations. They show the probability that strontium ranelate is cost-effective compared with no treatment for a range of decision makers’ willingness to pay per QALY.