3 The manufacturer's submission

The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of denosumab and a review of this submission by the Evidence Review Group (ERG; appendix B).

3.1 The manufacturer's submission presented data for clinical effectiveness from one main randomised trial, the FREEDOM (fracture reduction evaluation of denosumab in osteoporosis every 6 months) study. This multicentre, double-blind, placebo-controlled trial enrolled 7868 postmenopausal women aged 60–90 years with T-scores of less than −2.5 SD and greater than −4.0 SD at lumbar spine, total hip, or both locations. The T-score measures bone mineral density using central (hip and/or spine) DXA scanning and is expressed as the number of standard deviations (SD) below the mean bone mineral densityof young, healthy adults of the same gender at their peak bone mass. Lower T-scores indicate lower bone mineral density. Women were randomly assigned to receive a subcutaneous injection of either 60 mg denosumab or placebo twice a year for 3 years. All participants also took daily calcium and vitamin D supplements.

3.2 The primary outcome was the incidence of new radiographically diagnosed vertebral fractures. Secondary outcomes were time to first non-vertebral fracture and time to first hip fracture. Health-related quality of life was assessed in terms of change from baseline in patient-reported outcomes using both the osteoporosis assessment questionnaire-short version (OPAQ-SV; physical function, emotional status and back pain score), and the EUROQOL-5D (EQ-5D) questionnaire.

3.3 The results of the FREEDOM study demonstrated that, based on the number of people who underwent spinal radiography at baseline and during at least one visit after baseline, the 36-month incidence of new radiographically diagnosed vertebral fractures was 2.3% (86 of 3702 women) in the denosumab group compared with 7.2% (264 of 3691 women) in the placebo group (relative risk [RR] 0.32, 95% confidence interval [CI] 0.26 to 0.41; p < 0.001). The reduction in risk was similar during each year of the trial. Similar reductions in incidence were seen for clinically diagnosed vertebral fractures (0.8% for denosumab versus 2.6% for placebo; hazard ratio [HR] 0.31, 95% CI 0.20 to 0.47; p < 0.001) and for multiple new radiographically diagnosed vertebral fractures (0.6% for denosumab versus 1.6% for placebo; RR 0.39, 95% CI 0.24 to 0.63; p < 0.001). Denosumab also reduced the risk of non-vertebral fracture (6.5% for denosumab versus 8.0% for placebo; HR 0.80, 95% CI 0.67 to 0.95; p = 0.01) and hip fracture (0.7% for denosumab versus 1.2% for placebo; HR 0.60, 95% CI 0.37 to 0.97; p = 0.04). The manufacturer stated that dropout rates were similar between groups and no imbalances were observed.

3.4 Health-related quality of life was assessed using the OPAQ-SV and EQ-5D questionnaire at baseline and every 6 months for 3 years. Among women who completed the study, completion rates for measures of health-related quality of life at year 3 were 83% for OPAQ-SV and 82% for EQ-5D. No significant differences were seen between treatment groups in measures of health-related quality of life at baseline compared with year 3, or between women without any fractures and those with incident clinical fractures. Changes from baseline to year 3 for each OPAQ-SV dimension and EQ-5D scores were positively correlated (all p < 0.0001).

3.5 A statistically significant difference was noted in skin infections, which occurred in 12 women receiving denosumab compared with one woman receiving placebo (p = 0.002). However, when all studies of denosumab were pooled in the manufacturer's meta-analysis, the overall incidences of adverse events, serious adverse events and adverse events leading to treatment withdrawal were generally similar between denosumab and placebo groups. Further safety data were available from 30 studies, giving a total of 14,000 patients, including 11,000 postmenopausal women with low bone density or osteoporosis, as well as people taking denosumab for preventing bone loss in prostate or breast cancer.

3.6 The manufacturer stated that it was mindful of the need for efficient use of NHS resources, and that, given the wide availability of generic oral bisphosphonates, denosumab was expected to be an option for women in whom oral bisphosphonates are unsuitable (reasons for unsuitability are that the woman is unable to comply with the special instructions for the administration of oral bisphosphonates, or has a contraindication to or is intolerant of oral bisphosphonates). Therefore denosumab was not expected to compete with oral bisphosphonates in clinical practice. In the absence of head-to-head clinical trials comparing denosumab with all relevant comparators (denosumab, strontium ranelate, raloxifene, teriparatide and zoledronate), the manufacturer carried out a random-effects meta-analysis of the relative risks (RRs) for all fracture endpoints directly comparing each treatment against placebo. The fracture incidence data (and RRs) for strontium ranelate for hip and wrist fracture were taken from the publication by Reginster et al. (2008) which reported 5-year data from the TROPOS study. As outlined in the table 1 below, the results of the manufacturer's meta-analysis showed that all treatments were associated with statistically significant decreases in the risk of morphometric vertebral fractures compared with placebo. Denosumab, strontium ranelate and zoledronate were associated with statistically significant decreases in the risk of clinical vertebral fractures, but raloxifene was not (no data were available for teriparatide). Similarly, denosumab, strontium ranelate, teriparatide and zoledronate were associated with statistically significant decreases in the risk of non-vertebral fractures, but raloxifene was not. Denosumab and zoledronate were associated with statistically significant decreases in the risk of hip fractures but strontium ranelate, and teriparatide were not (no data were available for raloxifene). None of the treatments were associated with a statistically significant decrease in the risk of wrist fracture.

Table 1 Manufacturer's direct comparison of each comparator with placebo from the random effects meta-analysis

Comparator

Clinically diagnosed vertebral fracture (relative risk [95% CI])

Non-vertebral fractures (relative risk [95% CI])

Hip fracture
(relative risk [95% CI])

Wrist fracture
(relative risk [95% CI])

Denosumab

0.32
(0.21 to 0.48)*

0.81
(0.69 to 0.96)*

0.61
(0.37 to 1.0)*

0.84
(0.64 to 1.1)

Zoledronate

0.23
(0.14 to 0.37)*

0.75
(0.65 to 0.87)*

0.59
(0.42 to 0.83)*

Raloxifene

0.45
(0.05 to 3.82)

0.66
(0.16 to 2.65)

Strontium ranelate

0.65
(0.50 to 0.84)*

0.88
(0.78 to 0.99)*

0.89
(0.67 to 1.2)

0.98
(0.73 to 1.31)

CI, confidence interval.

*Statistically significant (p ≤ 0.05).

3.7 The manufacturer's submission included a systematic review of the cost-effectiveness evidence for denosumab. The manufacturer carried out Markov cohort modelling to assess the cost effectiveness of denosumab against primary and secondary comparators. Primary comparators were strontium ranelate, raloxifene and no treatment (placebo). Secondary comparators were intravenous ibandronate, zoledronate and teriparatide. The manufacturer stated that denosumab is expected to be a treatment option for women with osteoporosis for whom oral bisphosphonates are unsuitable. Therefore, comparisons with oral bisphosphonates were not directly relevant to this appraisal and were included in appendices to the manufacturer's submission. The manufacturer stated that 71.6% of women receiving treatment for osteoporosis in England and Wales receive alendronate, 15.8% receive risedronate, 1.5% receive etidronate and 4.3% receive oral ibandronate, meaning that 93.2% of this population receive oral bisphosphonates (2009 figures). This means an estimated 6.8% of women receiving treatment for osteoporosis in England and Wales receive drugs other than oral bisphosphonates (2.8% strontium ranelate, 2.2% raloxifene, 0.6% intravenous ibandronate, 0.7% zoledronate, 0.2% calcitonin, 0.2% calcitriol and 0.1% teriparatide).

3.8 The manufacturer stated that persistence and compliance with oral bisphosphonates are poor because of the strict and complex dosing regimen and side effects of treatment. The manufacturer's submission stated that at least 42% of patients taking oral bisphosphonates stop within 1 year, and the median duration of treatment is estimated to be as low as 1.2 years. The manufacturer stated that few people (< 1%) permanently discontinued denosumab treatment because of treatment-related adverse events over 2–3 years in the FREEDOM study.

3.9 The model assessed the cost effectiveness of denosumab against the primary and secondary comparators for two separate cohorts. The first investigated the primary prevention of fragility fractures in women (70 years and over) with osteoporosis (T-score of −2.5 SD or below) for whom oral bisphosphonates are unsuitable. The second investigated the secondary prevention of subsequent fragility fractures in women (70 years and over) with osteoporosis (T-scores of −2.5 SD or below) and prior fragility fractures in whom oral bisphosphonates are unsuitable. The model had a cycle length of 6 months and a lifetime horizon (defined as until time of death or age of 100 years), including a half-cycle correction, with a treatment duration of 5 years.

3.10 The model included six discrete health states: well, hip fracture, clinically diagnosed vertebral fracture, wrist fracture, other types of fracture (pelvic, femur shaft, tibia, fibular, humerus, scapula, clavicle, rib or sternum), and death. It included two additional health states (post-hip fracture and post-vertebral fracture) to account for the long-term costs and effects associated with these fractures (no long-term costs or effects were assumed for women with wrist or other fractures). When a fracture occurred, women were modelled to remain in the respective fracture state for two cycles (1 year). After this period, women with a wrist fracture or other types of fracture were modelled to return to the well state. Women with a vertebral fracture or hip fracture were modelled to enter a post-fracture state. Women who had a vertebral fracture could no longer incur a wrist fracture or other type of osteoporotic fracture (other than a subsequent vertebral fracture or hip fracture). Women who had a hip fracture could only incur further hip fractures. The manufacturer's model was not a treatment-sequencing model because of the lack of clinical evidence for such use.

3.11 The manufacturer's base-case analysis assumed that women continued osteoporosis therapy for 5 years, and costs and quality-adjusted life years (QALYs) were tracked over the lifetime of the cohorts (consistent with economic modelling in 'Alendronate, etidronate, risedronate, raloxifene and strontium ranelate for the primary prevention of osteoporotic fragility fractures in postmenopausal women' [NICE technology appraisal guidance 160] and 'Alendronate, etidronate, risedronate, raloxifene, strontium ranelate and teriparatide for the secondary prevention of osteoporotic fragility fractures in postmenopausal women' [NICE technology appraisal guidance 161]). This assumption was examined in a sensitivity analysis.

3.12 Subgroup analysis was undertaken for women with and without prior fracture by age (55–75, 5-year age bands) and T-score (between −2.5 to −4.0 SD). Sensitivity analysis assessed the effect on cost effectiveness of the presence or absence of additional independent clinical risk factors for fracture in women of 70 years of age, with and without prior fragility fractures. Sensitivity analysis also assessed the effects on cost effectiveness of differences in treatment persistence and compliance.

3.13 In the manufacturer's base-case analysis, fracture risks were estimated on the basis of epidemiological literature, and were based on three main elements: general population fracture risk, increased fracture risk associated with osteoporosis, and risk reduction attributed to treatment (if any). A systematic review of the literature was undertaken to identify appropriate UK studies or systematic reviews for all three model parameters. Age-specific fracture risks were estimated for women in the general population (using a study by Singer et al. [1998] to estimate risk of wrist and hip fractures, and a study by Kanis et al. [2000] to derive estimates for the incidence of clinically diagnosed vertebral and other fractures). Next, age-matched Z-scores (that is, the estimate of the number of SD below the mean bone mineral density of the general population for the patient's age and sex) were estimated for a cohort with osteoporosis using the National Health and Nutrition Examination Survey (NHANES) III database. Evidence from the systematic review was then used to attribute age-specific relative risks for the different types of fracture. Treatment was modelled to continue for 5 years by applying relative risks to the estimated baseline risks of fracture in the cohort with osteoporosis. An assumption was made that, on stopping treatment after 5 years, women would return in a linear fashion to baseline risk levels over 1 year (a return to baseline over 5 years was assumed in NICE technology appraisal guidance 160 and 161). The relative risks of fracture for each treatment for clinical vertebral, hip and wrist fractures were estimated from the manufacturer's direct comparison for each treatment against placebo if data were available. If evidence was not available for a comparator, the following explicit assumptions were made:

  • that for interventions without data for the relative risk of clinical vertebral fracture, this was equivalent to the relative risk of morphometric vertebral fracture

  • that the relative risk for interventions for which data for wrist and hip fractures were not available was 1.00.

  • that since no efficacy evidence was identified for intravenous ibandronate compared with placebo, efficacy was equivalent to that of oral ibandronate

  • that the relative risk for other fractures was 1.00 for all treatments, because 'other fracture' was not defined consistently across studies.

3.14 The model accounted for observed increases in the risk of mortality after fracture by applying relative risks for mortality obtained from a review of the literature. An increased risk was modelled for the first year and subsequent years after hip fracture or vertebral fracture. For other types of fracture, women were modelled to be at increased risk of mortality for 1 year only. The relative risks of mortality after all types of fracture were adjusted downwards to account for the observation that a proportion of mortality after fracture is explained by comorbidity. It was assumed that 30% of all mortality after all types of fracture is causally related, which is consistent with similar assumptions in NICE technology appraisal guidance 160 and 161.

3.15 The manufacturer's model also took into account persistence and compliance. Persistence is defined as the duration of time from start to end of therapy, and compliance is defined as conforming to the recommendations made by the provider with respect to timing, dosage and frequency of medication taking. Persistence and compliance were assumed to be 100% for the 5-year treatment period for all modelled treatments. Sensitivity analysis was carried out for oral therapies and denosumab.

3.16 Women completed the EQ-5D questionnaire in the FREEDOM study, but the number of fracture events with associated EQ-5D scores recorded was low and the trial design precluded assessment of health status immediately after fracture events. Therefore, evidence from the manufacturer's systematic review of the literature on health-related quality of life in osteoporosis was considered to be more meaningful and was applied in the economic analysis. The disutilities associated with fracture were obtained from a systematic review of the literature and applied to population norms in the form of utility multipliers. Utility loss associated with hip and vertebral fractures was modelled in a two-stage process, with a larger decrease in the first year after fracture and an ongoing but less severe utility loss in subsequent years. Utility multipliers for the first and subsequent years after hip fracture were obtained from a meta-analysis of studies using the EQ-5D responses. Utility loss associated with clinically diagnosed vertebral fracture was estimated separately for women managed in hospital and in primary care. The disutilities for women in hospital were derived from the EQ-5D scores of a cohort that were predominantly in hospital. The disutilities for women who were not in hospital were obtained from cohorts with prevalent morphometric fractures. Utility multipliers associated with wrist fracture were also obtained from the literature and applied in the model for 1 year after the event. Because of an absence of evidence, the same multiplier and the same approach were also used to model utility loss associated with other types of fractures. Finally, utility losses associated with selected adverse events were also included in the model.

3.17 Treatment costs and quality-of-life losses associated with wrist fracture or other types of fracture were modelled to last 1 year. Clinically diagnosed vertebral fractures and hip fractures were modelled to incur ongoing costs and loss of quality of life.

3.18 Costs of drug treatment were estimated using the 'British national formulary' (edition 58), with assumptions about the costs of administration and monitoring for the comparators. Fracture costs were estimated using hospital episode statistics for England and Wales in conjunction with the Department of Health's Healthcare Resource Group tariff; assumptions about the proportion of women treated in hospital, with and without surgery, for the different fracture types were informed by a combination of expert opinion, review of the literature and analysis of routine data. Costs associated with severe adverse events (such as gastrointestinal adverse events associated with oral therapies and cellulitis associated with denosumab) were included. Other types of adverse events associated with denosumab and its comparators were not included.

3.19 The results of the manufacturer's base-case analysis (pairwise comparisons) for the primary comparators showed that, for primary prevention, the incremental cost-effectiveness ratios (ICERs) for denosumab were £29,223 per QALY gained compared with no treatment and £9289 per QALY gained compared with raloxifene, and denosumab dominated strontium ranelate (that is, denosumab was less costly and more effective). For secondary prevention, the ICERs for denosumab were £12,381 per QALY gained compared with no treatment, £2046 per QALY gained compared with raloxifene, and denosumab dominated strontium ranelate. ICERs compared with no treatment for primary prevention were £74,239 per QALY gained for raloxifene and £102,592 per QALY gained for strontium ranelate. ICERs compared with no treatment for secondary prevention were £24,524 per QALY gained for raloxifene and £37,123 per QALY for strontium ranelate.

3.20 The results of the manufacturer's base-case analysis (pairwise comparisons) for the secondary comparators showed that denosumab was the lowest-cost treatment. For primary prevention, the ICERs for the other treatments compared with denosumab were £70,900 per QALY gained for zoledronate, £772,424 per QALY gained for teriparatide, and denosumab dominated ibandronate. For secondary prevention, the ICERs for the other treatments compared with denosumab were £29,029 per QALY gained for zoledronate, £451,269 per QALY gained for teriparatide, and denosumab dominated ibandronate.

3.21 The manufacturer presented a subgroup analysis to demonstrate how the cost effectiveness of denosumab varied when using different treatment cut-offs (that is, all women with a T-score at or below −2.5, −3, −3.5 SD and so on). The manufacturer provided further subgroup analyses for women with and without prior fracture by age and T-score. The results of the manufacturer's subgroup analyses showed that the cost effectiveness of denosumab improved as age increases and as T-score decreases, and with the presence of a prior fragility fracture. For primary prevention, in circumstances in which none of the treatments appraised by NICE are recommended, and oral bisphosphonates are unsuitable, the ICER for denosumab compared with no treatment varied between £19,313 and £71,319 per QALY gained. In circumstances in which strontium ranelate is recommended for primary prevention, denosumab dominated strontium ranelate (that is, denosumab was more effective and less costly). For secondary prevention, in circumstances in which no treatment is currently recommended in the NHS, the ICER for denosumab compared with no treatment varied between £12,289 and £22,957 per QALY gained. In circumstances in which strontium ranelate is recommended for secondary prevention, denosumab dominated strontium ranelate, and in circumstances in which raloxifene is recommended for secondary prevention, denosumab dominated raloxifene or had an ICER of £2046 per QALY gained.

3.22 The manufacturer also provided a subgroup analysis using the FRAX algorithm (an internet-based tool developed by the World Health Organization to calculate a 10-year absolute risk of fracture). This showed how cost effectiveness varied depending on T-score and the presence or absence of independent clinical risk factors for fracture. The results demonstrated that the presence of independent clinical risk factors for fracture, particularly rheumatoid arthritis, also improved the cost effectiveness of denosumab compared with the primary comparators (strontium ranelate, raloxifene and no treatment).

3.23 The manufacturer conducted a range of deterministic and probabilistic sensitivity analyses. The results of the deterministic sensitivity analyses showed that alterations to most key parameters had limited impact on comparisons of denosumab with raloxifene, strontium ranelate and no treatment. The impact on comparisons with intravenous ibandronate, zoledronate and teriparatide were most sensitive to changes in assumptions about the cost of denosumab administration. The manufacturer carried out sensitivity analyses that assumed one administration of denosumab in secondary care per year. Under this scenario, the ICER for denosumab compared with no treatment rose to £36,185 per QALY gained in women with no prior fragility fracture, and to £15,720 per QALY gained in women with a prior fragility fracture. This change led to zoledronate dominating denosumab in women with and without a prior fragility fracture. Following a request from the ERG, the manufacturer also carried out a sensitivity analysis in which denosumab treatment was assumed to be started in secondary care and thereafter delivered in general practice. This analysis showed that the additional cost associated with initiating treatment with denosumab in a secondary care setting had a marginal impact on the cost-effectiveness of denosumab compared with both primary and secondary comparators. The ERG also requested that the manufacturer provided further analysis assuming equal efficacy of denosumab and zoledronate for the prevention of wrist fractures. This analysis showed that the ICER for denosumab was moderately sensitive to assumptions about the relative efficacy of the two drugs for the prevention of wrist fractures.

3.24 After consultation, the manufacturer carried out additional sensitivity analyses on the long-term effects of fractures on mortality and nursing home care using the conservative assumption that nursing home admission was zero. These analyses showed no substantial impact on the cost-effectiveness results with the ICER for denosumab compared with strontium ranelate going from denosumab being dominant (that is, denosumab was less costly and more effective than strontium ranelate) to £2040 per QALY gained for primary prevention, and denosumab remaining dominant for secondary prevention. The ICER for denosumab compared with raloxifene went from £9289 (£11,135 costs, 8.0 QALYs) to £12,438 per QALY gained for primary prevention, and from £2,046 (£13,543 costs, 7.9 QALYs) to £5,120 for secondary prevention.

3.25 The results of the manufacturer's probabilistic sensitivity analysis showed that denosumab had approximately 50% probability of being considered cost effective at a threshold of £30,000 per QALY gained compared with the primary comparators (strontium ranelate, raloxifene and no treatment) in the base-case population of women aged 70 years with a T-score at or below −2.5 SD for primary prevention. The probability for secondary prevention was 90%. Against the secondary comparators (ibandronate, zoledronate and teriparatide), denosumab had a 60% probability of being considered cost effective at a threshold of £30,000 per QALY gained in the base-case population of women aged 70 years without prior fracture. The probability in women with prior fracture was 70%.

3.26 The ERG considered that the evidence of clinical effectiveness presented in the manufacturer's submission was derived from a large high-quality trial of adequate duration. The ERG stated that it did not consider the evidence presented in the manufacturer's submission on the effects of drugs on bone mineral density to be relevant because fracture data were available for all drugs. The ERG also noted that the data for morphometric vertebral fractures were not relevant, and so were not used in the modelling.

3.27 The ERG noted that the results for the direct comparison of strontium ranelate with placebo (RR for hip fracture of 0.89 and RR for non-vertebral fracture of 0.88) were similar to the meta-analysis provided in NICE technology appraisal 160 (RR for hip fracture of 0.85 and RR for all non-vertebral fractures of 0.84), which provided some confidence in the results. The ERG expressed concern about the relevant comparator for denosumab (see 3.29) and the methodology of the meta-regression to determine whether mean age and bone mineral density were associated with different effects of treatments.

3.28 The ERG noted that the manufacturer provided multiple comparisons of cost effectiveness using a high-quality validated model that took into account a wide range of costs, such as short-term drug costs and long-term nursing home costs, and that the analysis met the NICE reference case. The ERG considered that the appendices to the manufacturer's submission also provided very detailed accounts of underlying model assumptions and sensitivity analyses.

3.29 The ERG identified several issues with the manufacturer's economic model, specifically:

  • the choice of comparator

  • cost assumptions for denosumab

  • the validity of assumptions used for modelling utilities, costs, persistence and compliance

  • variations in cost effectiveness in subgroups of the cohort modelled

  • omission of underlying fracture risk estimates from the probabilistic sensitivity analysis

  • treatment setting and administration of denosumab.

3.30 First, the ERG believed that zoledronate should be a key comparator. The manufacturer's submission did not consider zoledronate or intravenous ibandronate to be primary comparators for denosumab because they are used by only 0.7% and 0.6% of currently treated women respectively (according to Intercontinental Marketing Services data), and neither comparator had been appraised by NICE. However, the ERG stated that intravenous ibandronate and zoledronate are licensed and used routinely in UK secondary care for treating osteoporosis in postmenopausal women. The ERG noted that intravenous ibandronate and zoledronate were similar in effectiveness but that intravenous ibandronate was given more frequently than zoledronate, and would be associated with greater administration costs. Therefore, given both its effectiveness and the same method of intravenous administration, zoledronate was a key comparator in the ERG's view.

3.31 The second issue raised by the ERG was that the relative cost effectiveness of denosumab compared with zoledronate depended on assumptions made about administration costs. The manufacturer assumed that denosumab would be given twice a year in general practice at the average cost of two standard visits to a GP, whereas zoledronate was assumed to be given once a year in hospital clinics (with some monitoring incorporated into the visit). The ERG believed that this approach made denosumab much less costly than zoledronate. Therefore, the ERG believed that, given the similar effectiveness of denosumab and zoledronate, the cost-effectiveness comparison depended largely on the relative costs used in the model. The ERG carried out additional exploratory analyses assuming that denosumab was given entirely in secondary care, which demonstrated that for primary prevention the ICER for denosumab compared with no treatment was £40,627 per QALY gained. For primary prevention, the ICER for denosumab compared with raloxifene was £25,743 per QALY gained and for denosumab compared with strontium ranelate was £15,866 per QALY gained. For secondary prevention, the ICER for denosumab compared with no treatment was £17,851 per QALY gained, the ICER for denosumab compared with raloxifene was £12,171 per QALY gained, and the ICER for denosumab compared with strontium ranelate was £6606 per QALY gained.

3.32 After comments on the appraisal consultation document, which reported a change in cost of zoledronate and that an alternative relative risk could be used for the effect of zoledronate on wrist fracture, the ERG were requested to carry out additional sensitivity analyses. These demonstrated that the change in the cost of zoledronate (which reduced from £283.74 to £266.72 in January 2010) resulted in ICERs for zoledronate compared with denosumab decreasing from £70,900 to £55,885 per QALY gained for primary prevention, and from £29,029 to £22,966 per QALY gained for secondary prevention. Additional sensitivity analyses were also carried out using an alternate relative risk value of 0.81 for risk of wrist fracture for zoledronate (instead of 1.0 in the manufacturer's base case, and 0.84 in the ERGs original sensitivity analyses). The results showed that the ICER for zoledronate compared with denosumab decreased from £70,900 to £58,764 per QALY gained for primary prevention, and from £29,029 to £24,454 for secondary prevention. The ERG carried out analyses using the manufacturer's assumptions in their economic model by combining both the above zoledronate changes simultaneously. These resulted in an ICER for zoledronate compared with denosumab of £44,804 per QALY gained for primary prevention, and £18,606 per QALY gained for secondary prevention.

3.33 The ERG identified that a simplifying assumption was used for transitions in the model. Women experiencing a vertebral fracture could no longer experience a wrist fracture or other type of fracture (apart from a clinical vertebral fracture or hip fracture). After a hip fracture, women could no longer experience any type of fracture other than a hip fracture. The ERG believed that this assumption was unrealistic because experience of a hip fracture or clinical vertebral fracture would put women at higher risk of further fracture. However, the extent of the effect of these assumptions on the cost-effectiveness estimates was unclear.

3.34 The ERG noted that in the manufacturer's base-case analysis, the assumption that fracture risk would return linearly to baseline levels over the course of 1 year after stopping treatment was conservative and would favour oral therapies. Persistence and compliance were assumed to be 100% for all treatments in the base-case analysis, which was also a conservative assumption. The ERG noted that after initial administration of denosumab, both compliance and persistence would be 100% for 6 months. However, in the long term, persistence with denosumab therapy may be less than 100%. The manufacturer carried out sensitivity analyses that examined variations in persistence for oral therapies and denosumab.

3.35 The ERG noted that the manufacturer's quality-of-life review methodology and the primary studies included in the review suggested that suitable utility multipliers were applied in the model. However, many of the multipliers were derived from observational time-series studies without independent control groups and therefore did not control for all potential confounding factors. The ERG noted that costs and utility losses associated with wrist fractures and other types of fracture were assumed to last for 1 year, whereas hip fractures and clinical vertebral fractures were modelled to incur ongoing costs and utility losses. The ERG also noted that utility loss relative to population norms remained constant in the second and subsequent years after hip fracture or vertebral fracture. This assumption may have slightly overestimated utility loss associated with hip and vertebral fracture if the observed trend towards improved quality of life in the second year after fracture continued in subsequent years.

3.36 The ERG noted that the manufacturer's ICERs varied substantially within subgroups of the cohorts, and that the appropriate comparator also varied by subgroup according to existing NICE guidance. Furthermore, neither raloxifene or strontium ranelate compared favourably with no treatment (ICERs of £74,239 and £102,592 per QALY gained respectively for 70-year-old women with a T-score of −2.5 SD and no prior fragility fracture, and £24,524 and £37,123 per QALY gained respectively for those with a prior fragility fracture), which is consistent with the modelling in NICE technology appraisal guidance 160 and 161. The ERG expressed the view that demonstrating cost effectiveness against these comparators did not allow the conclusion that denosumab is cost effective. The ERG also believed that, for the comparison between denosumab and zoledronate, there was uncertainty about the costs of administering these two drugs and their relative efficacy for the prevention of wrist fracture.

3.37 The manufacturer conducted a range of deterministic and probabilistic sensitivity analyses. The ERG noted that an important omission from the probabilistic sensitivity analysis was the underlying estimates of fracture risk. The manufacturer stated that data limitation meant that distributions could not be estimated for these parameters. The ERG believed that this would have the effect of overestimating the probability of denosumab being considered cost effective at different payment thresholds. It also noted that deterministic sensitivity analysis showed that the cost-effectiveness estimates were sensitive to underlying fracture risk. Following consultation the manufacturer carried out additional sensitivity analyses in which beta distributions were assigned to baseline fracture incidence based on an assumed sample size of 10,000. Probabilistic sensitivity analysis showed that for denosumab compared with no treatment the ICER was £30,422 per QALY gained which was similar to the deterministic ICER of £29,233 per QALY gained.

3.38 The ERG had concerns about the treatment setting and administration of denosumab in the model. The subcutaneous injection of denosumab is simple and could be carried out by a general practitioner, a practice nurse or the woman herself. However, the ERG believed that denosumab treatment would probably not be started in general practice because it is a new biological agent that has effects on other body systems (including the immune system), and that long-term adverse events could not be ruled out. The ERG stated that it would expect at least one outpatient visit to be needed and, in many cases, continued hospital follow-up would be necessary. Additionally, if follow-up was partly or mainly in general practice, the ERG believed that administration of denosumab would probably not be provided in primary care as part of general medical services, but would be regarded as an enhanced service for which an additional payment would be negotiated (the size of which is currently unknown, but may be greater than the manufacturer's assumption of the average cost of two visits to a GP per patient per year). Therefore, the marginal costs per patient of administering denosumab in primary care may be greater than the average cost of two visits to a GP per patient per year as presented in the manufacturer's model.

3.39 The ERG noted that although denosumab could be self-administered by the woman, the average age of women taking medication for the prevention of fracture in the General Practice Research Database dataset was 71.4 years, and many would be older. Such women might not be able to give themselves a subcutaneous injection because of poor eyesight, poor manual dexterity or cognitive impairment. The oldest age groups also have the highest proportion of women treated with oral bisphosphonates. Furthermore, training women to self-administer denosumab might not be regarded as worthwhile because they would have to visit a general practice to obtain the pre-filled pen injection device, and after 6 months some may have forgotten how to administer it (which is unlikely to occur with drugs given daily, such as teriparatide). The ERG also expressed the view that an equality issue exists for women who have had a stroke in the past and who are at increased risk of falls and fracture, together with bone loss because of reduced mobility. Such women might have difficulty swallowing or standing to take oral bisphosphonates, and therefore, intravenous or subcutaneous drugs may be more suitable.

3.40 Full details of all the evidence are in the manufacturer's submission and the ERG report.