Raloxifene and teriparatide for the secondary prevention of osteoporotic fragility fractures in postmenopausal women

The Assessment Group for this appraisal (School of Health and Related Research, University of Sheffield [ScHARR]) reviewed data from published randomised controlled trials (RCTs) in postmenopausal women in which fracture or health-related quality of life was an endpoint and where 1 of the 6 drugs of interest was compared with a relevant comparator, such as no treatment, placebo or 1 of the other included interventions. The majority of studies used placebo or no treatment as a control. Most studies ensured that women in all trial arms had normal calcium levels (that is, normal serum concentrations) or adequate supplementation, and some studies used additional dietary supplementation with vitamin D.

For this appraisal, reductions in RR associated with treatment were pooled regardless of the baseline BMD and fracture status of the participants in the studies. It was also assumed that these reductions in RR remained constant at all ages, although little evidence was available for the effectiveness of the drugs in women aged 80 years or older.

For vertebral fractures, some studies used clinical (that is, symptomatic) fractures as their endpoint whereas others used fractures that were identified radiographically. Vertebral fractures identified radiographically, which are termed 'radiographic fractures' or 'morphometric fractures', include both symptomatic and asymptomatic fractures. There are different definitions of a vertebral radiographic fracture, but those definitions that require a 20% reduction in vertebral height are generally recognised as producing more reliable results than those that require a 15% reduction.

For non-vertebral fracture types, individual data on hip, leg, pelvis, wrist, hand, foot, rib and humerus fractures were sometimes provided, whereas some studies only presented data for all non-vertebral fractures grouped together.

Two cohort and 2 case–control studies reported on a potential relationship between acid-suppressive medication (proton pump inhibitors or histamine H2 receptor antagonists) and fracture risk. One of the case–control studies, which used the UK General Practice Research Database (GPRD), found that 1 year or more of acid-suppressive medication was associated with an increase in fracture risk. The other case–control study reported a reduction of fracture risk associated with use of histamine H2 receptor antagonists, and that use of other acid-suppressive medication might increase fracture risk. Both studies, however, were unable to demonstrate convincingly that fracture risk was independent of underlying disease that might determine differences in fracture risk.

A prospective cohort study excluded women taking medication for fracture prevention and reported an increase in non-vertebral fracture in those taking acid-suppressive medication compared with those who were not. Findings appeared similar for users of proton pump inhibitors or histamine H2 receptor antagonists, but differences in fracture risk were not statistically significant for those using proton pump inhibitors compared with those not using acid-suppressive medication. One large retrospective cohort study using the UK GPRD compared women taking acid-suppressive medication plus bisphosphonates with those taking bisphosphonates alone. This GPRD study reported an increase in fracture risk for some fracture sites with concomitant use of acid^-suppressive medication and bisphosphonates, but a reduction in risk for other fracture sites. The information on patients included in this GPRD study was incomplete and details of adjustments for confounders were not reported. The 2 cohort studies were not fully published, and their analysis may have been prone to confounding.

Following the Court of Appeal Order of April 2010, NICE requested an additional submission from the manufacturer of strontium ranelate (Servier), setting out their views on the most appropriate estimate of strontium ranelate's efficacy in reducing the rate of hip fracture.

Servier explained that the pivotal phase 3 RCT (Treatment of Peripheral Osteoporosis Study [TROPOS]) was started before the increased regulatory emphasis on the prevention of hip fracture as a key measure of efficacy of treatments for osteoporosis (because of the significant morbidity associated with hip fracture). TROPOS had not been designed or powered to demonstrate the effect of strontium ranelate treatment on rates of hip fracture. In support of its application for regulatory approval of strontium ranelate, Servier was therefore asked by the European Medicines Agency (EMA) to investigate the efficacy of strontium ranelate in reducing the rate of hip fracture in a post-hoc subgroup analysis of TROPOS participants who met the definition of established osteoporosis (that is, a BMD T-score of -2.5 or below and 1 or more associated fractures). Instead of the requested subgroup, Servier provided the EMA with data for a different subgroup of trial participants whom they identified as being at high risk for hip fracture. This subgroup comprised women aged 74 or older who had a femoral T-score of -2.4 or below. (T-scores were calculated according to trial-specific normative data, using a threshold of -3.0 or below, which is equivalent to a T-score of -2.4 or below when measured according to the standards subsequently adopted by the WHO. The latter classification is used throughout this guidance document.) This subgroup represented 42% of TROPOS participants and had an RR of hip fracture of 0.64 (95% CI 0.412 to 0.997).

Servier described the method used to identify this high-risk subgroup. The placebo arms of 2 RCTs (TROPOS and another trial designed to assess the efficacy of strontium ranelate in reducing vertebral fractures, Spinal Osteoporosis Therapeutic Intervention [SOTI]) were pooled and the influence on fracture rates of 3 of the main risk factors for fragility fracture – age, BMD and prior fracture – was explored. Servier found that, in the pooled placebo arms of these 2 RCTs, prior fracture had no effect on the rate of hip fracture, so this factor was not considered further. To select an age group in which the risk of hip fracture was elevated, Servier investigated various possible age cut-offs, and identified the age at which the difference in the rate of hip fracture between women older and younger than the cut-off was greatest. This process led to the selection of an age cut-off of 74 years. Servier stated that this cut-off was consistent with epidemiological data, in particular a study by Donaldson et al. (1990), which Servier interpreted as showing a rising rate of hip fracture among women in the general population above the age of 74. The selected BMD cut-off was closely aligned to the WHO definition of osteoporosis (a T-score of -2.5 SD or below; see section 2.3). Servier emphasised that, having identified factors related to a high risk of hip fracture by screening the pooled data from the placebo arms of 2 RCTs, a single post-hoc analysis of the effect of strontium ranelate in this subgroup had been performed, without the need for multiple exploratory analyses of fracture risk reduction adopting different criteria for the subgroup selection.

After Servier had submitted data on efficacy in its chosen subgroup to the EMA, the EMA requested further analyses to confirm the effect of strontium ranelate on the rate of hip fracture. Servier provided additional evidence, including data from longer follow-up periods and analyses of trial participants with demonstrated compliance to treatment. Servier indicated that this additional evidence supported the view that an RR of 0.64 is a valid estimate of the efficacy of strontium ranelate in reducing the rate of hip fracture.

In their additional submission to NICE following the Court of Appeal Order in April 2010, Servier also suggested a hypothesis for a possible increased effect of strontium ranelate in older women: most osteoporosis drugs work by reducing the loss of existing bone, but strontium ranelate also stimulates the creation of new bone. Because the creation of new bone is increasingly impaired as women age, Servier stated that it is possible that strontium ranelate is able to provide additional benefit to older women.

Servier argued that the RR of 0.64 derived from the post-hoc analysis of the high-risk subgroup should be used in cost-effectiveness analyses to quantify the effect of strontium ranelate in reducing the rate of hip fracture because, in its view, it represents a more robust estimate of efficacy than the RR for the whole trial population. Servier stated that, unlike the analysis of the whole trial population, the subgroup analysis was suitably powered to demonstrate the effect of strontium ranelate in reducing the rate of hip fracture. Because of this, in Servier's opinion, the estimate was statistically robust.

Servier's view was that the estimate derived from the high-risk subgroup could be assumed to apply to all women taking strontium ranelate, but it acknowledged issues surrounding extrapolation from the high-risk subgroup to a broader population. Servier therefore indicated that it might also be concluded that the RR of 0.64 could only be applied to a population corresponding to the high-risk subgroup.

The DSU was commissioned to review Servier's additional submission, and to comment on the scientific validity of the post-hoc subgroup analysis provided by Servier. The DSU advised that any set of data will show some variation in response to treatment across different subgroups simply by chance. The DSU explained that, because of this, the correct statistical procedure for establishing a subgroup of trial participants with a significantly different response to treatment is via a test for interaction (that is, a formal test, using regression methods, of the hypothesis that the effect is different in 1 group of participants from that observed in the rest of the trial population). The DSU noted that no such test had been reported by Servier.

The DSU stated that the method used by Servier to identify the high-risk subgroup (see section 4.1.45) was logically likely to yield an unduly large relative effect, and the DSU stated that this would lead to a biased estimate of RR. This was because the method used to identify the age cut-off to define the subgroup was 'data-dependent' – that is, most of the data that were used to define the subgroup (the rate of hip fracture in the placebo arm of TROPOS) were also used to estimate the efficacy of strontium ranelate in the selected subgroup. In this way, the rate of hip fracture in the placebo group was certain to be high, relative to other potential age cut-offs, with no guarantee that this was also the case in the strontium ranelate group. Therefore, the DSU stated that the estimate of RR derived from the subgroup was likely to be artificially inflated.

The DSU also noted that, whilst Servier indicated that there were epidemiological data to support the chosen age cut-off (see section 4.1.45), the study by Donaldson et al. (1990) suggested that the rate of hip fracture rises to a notable level after 75 years of age, not 74.

The DSU advised that Servier's argument of enhanced statistical power in the subgroup analysis was incorrect. The DSU explained that, in an analysis of RR, statistical power is dependent on the number of events (in this case, hip fractures) and that choosing a smaller group of participants will tend to reduce, rather than increase, power unless the RR is markedly greater in that subgroup. Because of this, the DSU disagreed with Servier's claim that the subgroup analysis was 'fully powered'.

The DSU was asked to comment on the most appropriate approach, from a statistical viewpoint, to the use of data from the whole trial population of TROPOS and the high-risk subgroup, in determining the relative efficacy of strontium ranelate. The DSU responded that, if the relative effect were to be applied to women in the general population, an intention-to-treat analysis of all randomised trial participants would yield the most appropriate estimate of efficacy. The DSU also commented that, if more than 1 trial is available, a pooled analysis of RRs from the intention-to-treat data of all relevant trials would be preferable. A meta-analysis of the data from SOTI and TROPOS would have provided the most appropriate overall measure of efficacy.

The DSU also advised that even as an estimate of efficacy in the high-risk subgroup, the RR of 0.64 was likely to be too extreme because of the likelihood of selection bias arising from the way in which the subgroup had been identified (see section 4.1.51). The DSU also emphasised that, to estimate the cost effectiveness of strontium ranelate in a particular subgroup, it would not be sufficient simply to adopt an RR of hip fracture from that group. It would also be important to populate the rest of the economic decision model with evidence specific to the subgroup in question.

NICE invited Servier to respond to the DSU's report. Servier provided a document reiterating its previous views that the subgroup analysis performed to evaluate the efficacy of strontium ranelate in reducing the rate of hip fracture was based on sound scientific principles and valid statistical methods. Servier did not respond to other specific issues raised in the DSU report.

For proprietary alendronate, compared with no treatment, the manufacturer's model provided an incremental cost-effectiveness ratio (ICER) of £3,135 per quality-adjusted life year (QALY) gained for 70-year-old women with a T-score below -1.6 SD. The manufacturer's results were more favourable than the results of Assessment Group's 2003 model. This could be because the manufacturer's model was not adjusted for baseline fracture prevalence, or because it used different utilities for vertebral fractures, different efficacy data, different risk groups and a longer time horizon.

For etidronate, compared with no treatment, the manufacturer's model provided an ICER of £18,634 per QALY gained for 70-year-old women with a T-score below -2.5 SD. The manufacturer's model included morphometric vertebral fractures and corticosteroid use as risk factors for further fractures. It is unclear whether the manufacturer's ICER was for women with or without a prior osteoporotic fragility fracture.

For risedronate, compared with no treatment, the manufacturer provided data from 2 models. The ICER derived from the manufacturer's own model was £577 per QALY gained for women aged 74 years. In the second model provided by the manufacturer, which was commissioned from an external body, the ICER was higher, varying from £35,800 per QALY gained in women aged 60 years to £4,800 per QALY gained in women aged 80 years, for women with a prior vertebral osteoporotic fragility fracture and a T-score of -2.5 SD. For women at slightly higher risk of fracture, the ICERs were £18,600 per QALY gained or less for all age groups. The ICER calculated using the manufacturer's own model was difficult to verify from the information given. The ICERs generated by the second model were more consistent with the figures provided by the Assessment Group's 2003 model, although they did differ somewhat. This may be because of different cost and RR inputs.

For raloxifene, compared with no treatment, the manufacturer provided data for different age groups and different risk levels. All of the analyses included the breast cancer benefits. It was not clear how the different risk levels were defined. The ICERs ranged from £12,000 to £22,000 per QALY gained, and were slightly more favourable than the Assessment Group's 2003 analysis, even when the Assessment Group included the breast cancer benefits. In the Assessment Group's 2003 model, the RR for the breast cancer effect was higher (0.38) than the RR for invasive breast cancer used in the manufacturer's model (0.28), and the breast cancer risk was adjusted for the association between low BMD and decreased risk of breast cancer. Additionally, the manufacturer's model was not adjusted for baseline fracture prevalence, and included different utilities for vertebral fractures, different efficacy data, different risk groups, and a longer time horizon than the Assessment Group's model.

For strontium ranelate, compared with no treatment, the manufacturer provided 2 models: 1 developed in-house and the other commissioned from an external body. The first model showed that, for women aged over 75 years with previous fractures and a T-score of -2.5 SD, strontium ranelate was cost-effective at a maximum acceptable incremental cost-effectiveness ratio of £30,000 per QALY gained. The results of this model were comparable with those generated by the Assessment Group's 2005 model. The second model resulted in an ICER of £6,341 per QALY gained for 70-year-old women with a previous vertebral fracture and a T-score of -2.5 SD, decreasing to £5,002 per QALY gained in women aged 80 years. The manufacturer's results were more favourable than the Assessment Group's 2005 results because different modelling assumptions were used. For example, fewer health-state transition possibilities were incorporated. Compared with the Assessment Group's model, the manufacturer's model used more favourable efficacy data for hip fracture from the post-hoc 'high-risk' subgroup of women (see section 4.1.28 and sections 4.1.44 to 4.1.49), and slightly more favourable efficacy data for wrist and proximal humerus fracture. Higher hip-fracture costs were used in the manufacturer's model.

For teriparatide, compared with no treatment, the manufacturer provided ICERs for women aged 69 years. For women with fractures that had occurred more than 6 months previously (historical fracture), the ICER was £35,400 per QALY gained and for women with a more recent fracture the ICER was £28,863 per QALY gained. The manufacturer supplied additional economic analyses with ICERs of £18,845 and £12,106 per QALY gained for historical and recent fracture, respectively, based on changes to the assumptions of sustained efficacy for non-vertebral fractures and of the RR for specific risk groups. The manufacturer's model and the Assessment Group's 2003 model differed in a number of assumptions. The manufacturer's model was not adjusted for baseline fracture prevalence and used different utilities. The Assessment Group's 2003 model used more favourable assumptions on the duration of sustained efficacy after the end of treatment.

The Assessment Group provided a cost–utility model with 2 components (described in detail in the 2005 Strontium Ranelate Assessment Report). As a first step, the model calculated absolute fracture risk from the epidemiological literature on a number of independent clinical risk factors. These data were prepared under the auspices of the WHO and were provided for this appraisal under an academic-in-confidence agreement. As a second step, the model applied RR reductions for fracture taken from the meta-analysis described in section 4.1.22. A single estimate of efficacy was used for alendronate and risedronate based on pooled data for these 2 drugs. Following advice from the original Osteoporosis Guideline Development Group, it was assumed that RRs remained constant across all ages, T-scores and fracture status. The most recent analyses carried out by ScHARR were based on the price of non-proprietary alendronate in February 2008 (£53.56 per year for once-weekly 70-mg tablets; £108.20 per year for daily 10-mg tablets). Note that the remit of the original osteoporosis guideline has since been amended; see NICE's guideline on osteoporosis: assessing the risk of fragility fracture.

All osteoporotic fragility fractures in women aged 50 years or older were included in the modelling. The RR for hip fracture was assumed to apply also to pelvis and other femoral fractures. The RR for non-vertebral fracture was assumed to apply also to proximal humerus, rib, sternum, scapula, tibia, fibula and wrist fractures. Where confidence intervals for RRs spanned unity, it was assumed that there was no effect of treatment, except in the case of strontium ranelate. In this case, an RR of 0.85 for hip fracture was used to acknowledge the effect reported in the high-risk subgroup of the study. The model used UK-specific epidemiological data on femoral neck BMD.

The model assumed an initial utility in the year of fracture and a higher utility in subsequent years. The time horizon for predicting morbidity was 10 years, consisting of 5 years of treatment with sustained efficacy plus 5 years of linear decline to no effect. However, treatment-related decreases in mortality rate extended beyond the 10-year time horizon. For this, the life expectancy for a woman at the threshold T-score for osteoporosis was calculated from standard life tables, and any increase in mortality rate due to fracture would continue until death or an age of 110 years. In the base case, vertebral-fracture utility was assumed to be lower than hip-fracture utility, and a sensitivity analysis was carried out in which the utility for vertebral fracture was assumed to be the same as that for hip fracture. The percentage of women assumed to move from community living to a nursing home following a hip fracture increased with increasing age. An age-dependent gradient of hip-fracture risk was used, and an association between vertebral or proximal humerus fracture and increased mortality in women with osteoporosis was included. No follow-up BMD scans were included in the model; this reflects current clinical practice in the UK.

The model included an assumption about the costs and disutility associated with treatment-related side effects for all drugs, based on the findings of prescription-event monitoring studies in patients treated with alendronate. For the base case, the model assumed 50% persistence with treatment. In addition to the base case, the Assessment Group undertook a number of sensitivity analyses using alternative assumptions, including: persistence with treatment (25% or 75% at 5 years); reduction in the efficacy of the drugs at reducing the risk of fracture associated with risk factors other than age, prior fracture and low BMD to 0% or 50% (with a consequent upward adjustment of the RR for the risk factors of age, prior fracture and low BMD); disutility of vertebral fracture; updated fracture costs; and the disutility and costs of treatment-related side effects. It was assumed that women who experience bisphosphonate-related side effects had 91% of the utility of women who do not have such side effects. In the base case analysis for all of the drugs under consideration this was applied to 2.35% of women in the first treatment month and 0.35% of women thereafter and, in sensitivity analyses for bisphosphonates, to 24% of women in the first treatment month and 3.5% of women thereafter. In the case of strontium ranelate, the effect on VTE was not included in the model. Discount rates of 6% per year for costs and 1.5% per year for health benefits were applied, in accordance with NICE methods relevant to this appraisal.

For raloxifene, 4-year follow-up data from the MORE study were used, and it was assumed that women with low BMD have a lower breast cancer risk than women with normal BMD. The cost effectiveness was modelled excluding the breast cancer benefit, the risk of VTE and the effect on cardiovascular events.

The independent clinical risk factors for fracture used in the model were based on the data prepared under the auspices of the WHO (see section 4.2.7) and included BMI, prior fracture, previous or current use of corticosteroids, parental history of fracture, current smoking, alcohol intake of more than 2 units per year, and rheumatoid arthritis. The study provided prevalence data for the different risk factors, and risk ratios for hip fracture and osteoporotic fracture for each risk factor, including T-score and age. Using these risk ratios, absolute risk of fracture was calculated.

The estimates of cost effectiveness were generated for different levels of absolute risk derived from a large number of combinations of T-score (in bands 0.5 SD wide), age and number of independent clinical risk factors for fracture. For practical reasons relating to the number of potential combinations, single-point RRs of fracture, calculated from the log-normal efficacy distributions, were used in the model. Results were presented for population groups categorised according to age, T-score and number of independent clinical risk factors.

Women with a fracture who present to clinicians require a DXA scan for osteoporosis to be established. Therefore, the Assessment Group also estimated the impact of DXA scanning on the cost effectiveness of the drugs. This required both a calculation of the ICER for treatment, and a calculation of the distribution of risk assessment cost over the population who would benefit from treatment. A net-benefit approach was used to do this. The net-benefit approach is analogous to the more traditional cost per QALY gained approach, but also requires a value of willingness to pay (WTP) for an additional QALY gained. For the calculation of the net benefit of an intervention, the WTP is first multiplied by the incremental QALY gained associated with the intervention, then the incremental cost associated with the intervention is subtracted. For this appraisal, the total net benefit for each age group and DXA scanning approach was calculated by subtracting the cost of DXA scanning from the net benefit of treating all women who can be treated cost effectively.

A stepped net-benefit approach was used to estimate, in reverse order, the cost effectiveness of risk assessment, DXA scanning and treatment of women with a prior fracture. Two WTP values, £20,000 or £30,000 per QALY gained, were applied in the modelling.

• Step 1. ICERs for treatment versus no treatment were calculated for each intervention for various combinations of age, T-score and number of independent clinical risk factors for fracture (see section 4.2.12). The net benefit of treatment per woman was calculated using the following formula: Net benefit = £30,000 (or £20,000) × incremental QALYs – incremental costs.
  
  For women for whom the ICER for treatment was more than £30,000 (or £20,000) per QALY gained, the net benefit was set to zero.

• Step 2. The net benefit per woman was multiplied by the number of women in the population estimated to fall within each combination of age, T-score and number of independent clinical risk factors for fracture (based on the data used to develop the algorithm prepared for the WHO). The net benefits for each group were then added together to give a total net benefit of treatment for women with no, 1, 2 or 3 independent clinical risk factors within each age group.

• Step 3. The cost of DXA scanning all of the women in each age or independent clinical risk factor group was subtracted from the net benefit of treatment for that group (calculated as described in step 2). This provides the net benefit of treatment and DXA scanning for the group, assuming that the number of independent clinical risk factors is known. A positive net benefit indicates that DXA scanning of women in that age or independent clinical risk factor group and treating those groups of women in whom the ICER for treatment is £30,000 (or £20,000) or less provides an ICER for the entire strategy of less than £30,000 (or £20,000) per QALY gained.

• Step 4. When the resulting values of net benefit of treatment and scanning were negative they were set to zero. For each age group, the total net benefit of scanning and treatment was calculated by adding together the net benefits for each age or independent clinical risk factor group. The cost of opportunistic assessment for all women in this age group was then subtracted to give the net benefit of risk assessment, scanning and treatment. A positive net benefit indicates an ICER of less than £30,000 (or £20,000) per QALY gained for risk assessment, DXA scanning and treating women (at a specific T-score related to the ICER for treatment only) of that particular group. Cost per QALY gained data were presented for each strategy.

Following the outcome of the judicial review and the court ruling of March 2009, NICE was able to offer the Assessment Group's executable economic model for consultation. Consultees and commentators who requested the model and returned the necessary confidentiality undertakings received a CD-ROM containing the executable version of the economic model, a document with instructions for running the model and a proforma for commenting on the model. Comments on the Assessment Group's model were received from Servier Laboratories (the manufacturer of strontium ranelate), the Bone Research Society (BRS), the National Osteoporosis Society (NOS) and the Society for Endocrinology. Comments received from each of these consultees are summarised in the sections below.

These 4 consultees expressed the view that the documentation provided with the Assessment Group's model was insufficient, that the model supplied to them was incomplete and that some inputs could not be altered. They also stated that the application of the fracture risk algorithm developed under the auspices of the WHO could not be assessed. They felt that the model could not be validated and that its validity had not been demonstrated in documents made available during development of the guidance.

Servier commented that the fracture risks entered in the Assessment Group's model differed from estimates that Servier calculated using the FRAX fracture risk calculation tool (see section 4.3.48 for further information about the FRAX tool). Servier commented that mortality risk associated with clinical risk factors had been omitted from the model. In Servier's opinion, these differences called into question whether the WHO fracture risk algorithm had been applied correctly in the Assessment Group's model.

Other comments questioned the use of a fixed value for BMI in the model. Consultees commented that no clear explanation was provided of the rationale for the choice of BMI value, that a range of BMI values should have been used, and that the use of a fixed BMI value resulted in underestimation of the cost effectiveness of treatment for some women at risk of fracture.

Servier commented on the selection and weighting of the independent clinical risk factors for fracture used in the Assessment Group's model. Servier, BRS and NOS suggested that the risk associated with alcohol intake was incorrect in the model and that this would have adversely affected estimation of the cost effectiveness of treatment for women at risk of fracture. They suggested that a threshold alcohol intake of 3 or more units per year, as used in the FRAX fracture risk calculation tool, should have been applied. They also stated that the Assessment Group's model and the guidance were inconsistent with each other, and that these differences resulted in the risk of fracture being underestimated in the model. Servier also noted that the Assessment Group's model gave an equal weighting to each of the independent clinical risk factors for fracture. Servier suggested that this was a less precise approach than that used in the FRAX tool, which used different weightings (some higher and some lower than those in the Assessment Group's model) for each fracture risk for specific risk factors. Servier stated that the FRAX tool assesses fracture risk and cost effectiveness more accurately and 'deals more fairly' with variation between women at risk of fracture. Servier also noted that 1 of the risk multipliers for fracture risk included in the Assessment Group's model was not consistent with that given in the assessment report.

Servier and NOS noted that the Assessment Group's model had a time horizon limited to 10 years and criticised how mortality beyond 10 years had been taken into account in the economic evaluation. Servier expressed the view that, as a consequence of this, the model was inaccurate and underestimated the cost effectiveness of treatment for women at risk of fracture. Servier also identified 2 values ('wristbonusat2.5' and 'phbonusat2.5', related to QALY calculations) that were included in the model, but not described in assessment reports.

Servier commented that using the same disutility for side effects associated with strontium ranelate and bisphosphonates was not correct, as the side effects of strontium ranelate are different from those of the bisphosphonates.

BRS and NOS thought that the proportion of women with low BMD in England and Wales was substantially underestimated in the Assessment Group's model. These consultees were also concerned that although both smoking and previous or current glucocorticoid (corticosteroid) use had been included as additional independent clinical risk factors for fracture in the Assessment Group's model, they were not defined as risk factors in the recommendations (see section 1.5). In addition, both consultees thought that interactions between several clinical risk factors were not incorporated in the model, thereby reducing the cost effectiveness of treatment for women at risk of fracture, especially younger women.

All 4 consultees commented on elements of the Assessment Group's economic evaluation that had been considered and agreed by the Appraisal Committee before it directed the Assessment Group to develop the economic model using specific assumptions. These Committee-directed assumptions included the compliance rate, costs associated with fracture, utility values used for vertebral fracture, and the strategy for identifying women at high risk. Servier also commented on the discount rates used in the model.

Servier reported that it had prepared a 'comparative' model which was run using assumptions similar to those in the Assessment Group's model. This model was referred to in a report to support the mathematical foundation of revised analyses discussed as part of Servier's comments on the Assessment Group's model. This report was made available to the Appraisal Committee to inform its consideration of comments by Servier on the DSU report (see below and section 4.3).

The DSU was commissioned to review the comments from consultees on the Assessment Group's executable economic model and report to the Appraisal Committee. The DSU considered issues that were relevant to the economic model. Key issues were grouped under the common themes of model transparency and ability to assess its validity, methodology (approach) and model inputs.

The DSU assessed comments on the transparency and validity of the Assessment Group's model. With regard to the consultees' observation that some model inputs were fixed and that in their view the model provided for consultation was incomplete and not fully executable, the DSU confirmed that certain inputs were intentionally fixed and the ability to alter these inputs was not a feature of the model or necessary for some parameters with minimal uncertainty that are commonly fixed in other economic models. In response to comments on the consultees' inability to assess the application of the WHO algorithm, the DSU explained that the WHO algorithm itself was not embedded within the model. The DSU confirmed that absolute fracture risks were correctly calculated using the WHO algorithm before being entered into the model. The DSU noted that documentation had been provided to consultees in the form of publicly available reports and peer-reviewed manuscripts produced by the Assessment Group, and that instructions on the operation of the Assessment Group's model were also offered to consultees and commentators.

With regard to comments on the modelling approach adopted in the Assessment Group's model, the DSU responded by confirming that alcohol consumption of more than 2 units per year was included in the model, and that the coefficients used in the model were consistent with the WHO algorithm (as supplied to the Assessment Group at the time the model was developed). The DSU also explored how the model considered corticosteroid-related fracture risk, and confirmed that corticosteroid use was included in the model and that the coefficient used for this risk factor was consistent with that calculated using the WHO algorithm. The DSU noted that the fracture risk of women using corticosteroids would have contributed to the overall fracture risk of the whole modelled population and thereby reduced the ICER associated with the treatment of all women at risk of fracture.

The DSU confirmed that each clinical risk factor for fracture was given equal weighting in the model. In response to consultee comments expressing the view that this was a less precise approach than that used in the FRAX tool, the DSU noted 2 points. Firstly, no individual risk calculation tool was publicly available when the model was developed. Secondly, the DSU referred to the 2005 Strontium Ranelate Assessment Report, which compared suggested treatment thresholds (combinations of age, T-score and number of independent clinical risk factors for fracture) from the Assessment Group's model with treatment thresholds indicated by absolute fracture risk. The DSU suggested that the use of absolute fracture risk alone did not accurately predict cost effectiveness, and therefore would not provide a robust basis for the Committee's decision-making.

Consultee comments on the modelling approach also addressed the time horizon and population data used and the grouping of age in 5-year bands. The DSU confirmed that the consequences of fracture were considered beyond 10 years, and provided further explanation of the modelling approach. The DSU additionally undertook exploratory analyses of the impact of mortality after the 10-year time horizon and of incorporating mortality associated with vertebral fracture and proximal humerus fracture. They reported that the change in the results produced by the model was minimal when mortality risk beyond 10 years was doubled. The DSU also confirmed that UK epidemiological data from a study by Holt et al. were used in the Assessment Group's model, and undertook an exploratory analysis around the assumptions of the distribution of T-scores used in the model. For some age bands modelled, the T-scores did not follow a statistically normal distribution, but the DSU noted that the assumption of a normal distribution made it more likely that treatments for women at risk of fracture would be judged to be cost effective. The DSU considered a comment on the calculation of cost-effectiveness estimates averaged for the 5-year age bands implemented in the model. It disagreed with the alternative suggested by the consultee and noted that the Committee had considered and agreed that initial identification by age band was a workable strategy for selecting women at risk of fracture in clinical practice. It also noted that alternative strategies (which did not use age bands) may in fact be more resource-consuming and less likely to be judged as cost effective.

The DSU reviewed consultee comments on inputs used in the Assessment Group's model. It confirmed that the WHO algorithm (as supplied) had been correctly implemented in the model to produce estimates of fracture risk for each T-score band. The DSU suggested that the differences in the estimates of fracture risk obtained using the FRAX fracture risk calculation tool and the Assessment Group's model did not necessarily suggest that the WHO algorithm had been incorrectly applied (see section 4.3.48), and that these differences could occur for a number of reasons. For example, the use of a midpoint age to represent an age band of 5 years could lead to differences in estimates of fracture risk. The DSU confirmed that no increase in mortality associated with clinical risk factors was used in the model. The DSU suggested that inclusion of such mortality effects would be likely to increase the ICERs for women with those clinical risk factors. The DSU explained that this is because fewer QALY benefits would accrue in the model for women who die of causes related to risk factors. In response to a further comment from Servier, the DSU agreed that, for women without clinical risk factors, the inclusion of these mortality effects in the model may have the opposite effect (that is, a decrease in ICERs). Therefore, the overall effect of including the increased mortality associated with clinical risk factors would be small.

The DSU also confirmed that a fixed value for BMI of 26 kg/m² was used in the Assessment Group's model. This was the mean BMI in the UK epidemiological dataset from the Holt study used in the model. In an exploratory analysis using the WHO algorithm, the DSU showed that using a BMI of 26 kg/m² resulted in higher estimated fracture risk than a BMI of 20 or 32 kg/m² when BMD is known, and this was confirmed by the estimates supplied by 1 consultee. The DSU suggested that the BMI value used in the model may favour treatment of women at risk of fracture compared with alternative BMI values. The DSU also pointed out that BMI is a weak predictor of fracture when BMD is known (as specified in the identification strategy in the guidance).

The DSU investigated the risk multipliers used for fracture risk in the Assessment Group's model and the consultee comment that interactions between clinical risk factors had been omitted. It confirmed that the risk multipliers used for fracture risk had been correctly calculated from the WHO algorithm and that all interactions between risk factors had been included. The DSU also noted that the inconsistency between 1 of the risk multipliers for fracture risk included in the Assessment Group's model compared with the assessment report was the result of a typographical error. Accordingly, there was no impact on the results of the model.

The DSU did not respond in detail to comments on assumptions in the model that had already been documented and agreed by the Appraisal Committee and which were available to consultees and commentators earlier in the development of the appraisal guidance. The DSU did, however, list these issues in its report and cited where they had been considered by the Committee or had been available for comment during development of the guidance. Features of the economic evaluation previously discussed and agreed by the Committee included the following:

• discount rates used in model (section 4.2.10)

• treatment compliance (section 4.2.10)

• costs associated with fracture (section 4.2.17)

• strategy for identifying women at high risk of fracture (section 4.2.17)

• utility values used for vertebral fracture (sections 4.2.17 and 4.3.13)

• equal disutility for the side effects of strontium ranelate and bisphosphonates (section 4.2.10)

• sensitivity analyses on disutility (section 4.3.15).

The DSU concluded that, in its view, adequate documentation on the Assessment Group's model had been provided for consultees. It highlighted that the WHO algorithm used to generate estimates of fracture risk was not integrated within the Assessment Group's model; rather, the fracture risks derived from the algorithm were entered into the model. Comparisons with fracture risks derived using the FRAX fracture risk calculation tool were made by several consultees on the basis that the WHO algorithm supplied to the Assessment Group and the FRAX tool are assumed to be identical. The DSU could not verify these analyses without access to the FRAX algorithm. The DSU agreed that some parameters in the Assessment Group's model were fixed. These included those with minimal uncertainty, as well as those that are commonly fixed in other economic models. Sensitivity analyses conducted by the DSU suggested that none of the consultees' suggestions relating to the modelling approach would lead to significant improvements in the cost effectiveness of treatment for women at risk of fracture. The DSU concluded that, in its view, no issues raised by consultees would either affect the validity of the Assessment Group's model or raise significant doubts about the appropriateness of using the model to inform the deliberations of the Committee.

The Committee acknowledged the efforts of the Assessment Group to build on the model used previously, particularly in using epidemiological data and a fracture risk algorithm developed under the auspices of the WHO to calculate transition probabilities and to model the identification approaches. The Committee concluded that the Assessment Group's model was likely to give the best estimates of cost effectiveness because it used data from a wide age range (age 50 to 75 years or older), and was updated to use all osteoporotic fracture sites, more recent utility values, prevalence and risk-factor data, and an adjusted prevalence of fractures in the average population. Although the Assessment Group's model considered a shorter time period (10 years for predicting morbidity, see section 4.2.9) than the manufacturers' models, the Committee thought that this was appropriate considering the age groups involved and the uncertainties around health effects over a longer period.

The Committee discussed the assumptions underpinning the economic modelling undertaken by the Assessment Group. It noted that the most recent modelling explored some of the uncertainties identified by the Committee surrounding the results of the previous modelling; these related to the costs and disutility associated with treatment-related side effects and to non-compliance or non-persistence with treatment in a proportion of patients. The Committee also noted the effect of the recent price reductions for non-proprietary alendronate (70 mg weekly and 10 mg daily doses) on the cost effectiveness of the drug.

The Committee considered the base-case assumptions and those used in additional analyses. The Committee noted that the costs associated with fractures used in the base-case analysis were those used in the original assessment report developed in 2003 and considered that these were likely to be outdated. The Committee agreed that costs based on health resource groups, including home-help costs, were likely to provide the most accurate reflection of the cost of fractures to the NHS and personal social services, and it decided to incorporate these costs into the base-case analysis.

The Committee considered the utility multiplier used in the base-case analysis for the first year after a vertebral fracture and noted that it was based on a hospitalised patient group and not on a typical group of patients with vertebral fractures. Consequently, it was considerably lower than the utility value modelled for a hip fracture. Although the Committee acknowledged that vertebral fracture can lead to greatly reduced quality of life, it considered that its true value would not greatly outweigh the utility decrement associated with a hip fracture. Therefore, the Committee considered it reasonable to assume that the disutility in the first year after a vertebral fracture was equivalent to the disutility in the first year after a hip fracture and decided to include this assumption in the base-case analysis.

The Committee was not persuaded that the drugs under consideration had been unequivocally shown to reduce fracture risk that was attributable to risk factors not mediated through low BMD and age. The Committee concluded that the uncertainty surrounding the efficacy of the drugs on risk factors not mediated through low BMD and age should be factored into its decision-making by using an analysis that assumed 50% efficacy of the drugs on fractures associated with risk factors other than age and low BMD. Although the Committee recognised that 50% was necessarily an arbitrary figure, the use of either 0% or 100% was considered both extreme and less plausible. In the analysis accepted by the Committee, the assumption of 50% efficacy of the drugs on fracture risk associated with other risk factors was adjusted by using a correspondingly greater efficacy of the drugs on fractures associated with the key independent clinical risk factors (age, BMD and prior fracture).

The Committee considered the assumptions used in the modelling for the side effects of bisphosphonates, in which women who experience bisphosphonate-related side effects had 91% of the utility of women who did not have such side effects. In the base case, this was applied to 2.35% of patients in the first treatment month and 0.35% of patients thereafter. Taking into account the persistence data (sections 4.1.36 and 4.1.37) and the comments received from consultees and commentators that about 25% to 30% of women experience gastrointestinal side effects when first taking a bisphosphonate, the Committee agreed that it was important to consider the results of a sensitivity analysis assuming that 24% of women were experiencing bisphosphonate-related side effects in the first treatment month and 3.5% of women thereafter.

The Committee acknowledged that the modelling made assumptions necessary about the value of a QALY gained that could be considered an acceptable use of NHS resources. The Committee considered that women who have already sustained an osteoporotic fracture live with the pain and distress caused by the fracture. The Committee considered that women with an osteoporotic fracture constitute a different population from the primary prevention population and that there were some factors that justified considering a higher ICER range in line with NICE's guide to the methods of technology appraisal.

The Committee discussed a number of concerns surrounding other issues that were not represented in the model but which may have had an impact on the cost-effectiveness estimates. These included: possible long-term adverse effects of bisphosphonates on the formation of new bone; the likelihood that DXA scanning outside a clinical trial environment would not be as effective as in the clinical trials; and the possibility that the proportion of women who experience side effects may exceed the model's base-case assumptions. Finally, the Committee noted that current discount rates used by the Treasury, the Department of Health and NICE result in a cost-effectiveness calculation less favourable to the drugs than the discount rates used in the analysis considered by the Committee. Although a quantitative analysis of the uncertainties surrounding all these issues was not available, the Committee agreed that, for first-line treatment with a bisphosphonate, these uncertainties could be collectively approximated through the sensitivity analysis for side effects (see section 4.3.15). The Committee was persuaded, however, that the results of the sensitivity analysis need only apply to first-line treatment with a bisphosphonate, because many of the factors that led to the adoption of the sensitivity analysis did not apply for second-line treatment.

The Committee considered the results of the economic model following the price reduction for non-proprietary alendronate, the newly included assumptions and the sensitivity analyses (see sections 4.3.10 to 4.3.15). The Committee agreed that, when considering the use of alendronate as a first-line treatment, the sensitivity analysis that captured the uncertainties in the economic model (see section 4.3.14) was the most appropriate. This led the Committee to conclude that alendronate (based on the price of £53.56 per year for once-weekly treatment) would be an appropriate use of NHS resources for secondary preventative treatment in postmenopausal women with fragility fractures and confirmed osteoporosis (that is, a T-score of -2.5 SD or below). The Committee was advised by the clinical specialists from the original Guideline Development Group for the NICE guideline on osteoporosis that, in women aged 75 years or older with a prior fracture, a DXA scan may not be required if the responsible clinician considers it to be clinically inappropriate or unfeasible. This is because a very high proportion of these women would be likely to have a T-score of -2.5 SD or below.

The Committee noted that the prices of the different brands of alendronate vary greatly and concluded that alendronate should be prescribed on the basis of the lowest acquisition cost available.

The Committee noted that risedronate, etidronate, raloxifene and strontium ranelate were dominated by alendronate (based on the price of £53.56 per year for alendronate); that is, these drugs have a higher acquisition cost than alendronate, but are not more efficacious. The Committee was also aware that, for women for whom weekly non-proprietary alendronate could be recommended based on cost effectiveness, the ICERs for risedronate and strontium ranelate were very high, even without inclusion of identification costs (see examples in section 4.2.22).

The Committee considered an approach where the higher costs of risedronate, strontium ranelate and teriparatide were incorporated into the analysis by combining costs based on the estimated use of alendronate, risedronate and strontium ranelate and teriparatide. However, the overall cost effectiveness of such a combined approach for fracture prevention would be less favourable than that of alendronate. As a consequence, some women who would be eligible for treatment with alendronate as recommended in section 1.1 would not be offered treatment using such a combined approach. For this reason, the Committee did not consider the combined approach to be appropriate.

The Committee considered treatment options available for a woman who is intolerant to alendronate or unable to comply with instructions for administration despite reasonable measures to support continuation of alendronate treatment. The Committee noted that all other treatment options have higher acquisition costs and/or different effectiveness profiles, which would reduce the cost effectiveness of preventive treatment if these drugs were used. The Committee observed that the identification costs associated with finding women who could be cost-effectively treated with 1 of the other drugs would be negligible, because they would have already undergone an assessment and had a DXA scan in order to be assessed for first-line treatment with alendronate. Therefore, it agreed that the recommendations for this situation should be based on the modelling that excluded identification costs. The Committee also agreed that, when considering second-line or subsequent treatment, the base-case assumptions for side effects could be applied; that is, a 0.91 utility multiplier should be applied to 2.35% of patients in the first treatment month and 0.35% of patients thereafter.

The Committee considered women who cannot take alendronate because of a contraindication or a disability that prevents them from complying with the instructions for administration. Because such a contraindication or disability would be known before the risk assessment, this would comprise a first-line treatment situation, where identification costs are included. Alternative drugs become cost effective at a higher age and lower BMD in a first-line treatment situation, compared with a second-line treatment situation where identification costs are not included. However, such an approach was considered inappropriate by the Committee because it would unfairly disadvantage women who cannot take alendronate because of a contraindication or a disability. Therefore, the Committee concluded that women who cannot take alendronate for these reasons should have access to alternative drugs in the same way as women who cannot tolerate alendronate (that is second-line treatment, where the analysis excluded identification and assessment costs).

The Committee carefully considered the position of women who cannot take alendronate because of a condition which either makes alendronate contraindicated or which prevents individuals from complying with the instructions for administration for alendronate. In doing so the Committee noted that at least some women in this patient group were likely to be 'disabled' as defined by the Disability Discrimination Act 1995. The Committee was aware of its duties under that Act to avoid unlawful discrimination, to have due regard to the need to promote equality of opportunity for disabled people, and the need to take steps to take account of disabled people's disabilities, as well as its broader legal duties to ensure that its guidance is fair and reasonable.

The Committee noted that the drugs other than alendronate are cost effective only for patients at higher risk of fracture than the risk levels at which alendronate is cost effective. If these other drugs are recommended for use by patients who cannot take alendronate only when those patients meet the criteria at which these alternative drugs become cost effective, these patients will not receive preventative treatment unless they are at higher risk of fracture than the risk levels at which alendronate is recommended. The Committee therefore considered whether, for women who cannot receive alendronate, the other drugs should be recommended at the same risk levels as alendronate (that is using the criteria established as being cost effective for alendronate) in order to provide access to preventative treatment for all patients with the same level of risk. The Committee reviewed the ICERs for risedronate and strontium ranelate within the criteria established to be cost effective for alendronate. The Committee noted that the prices for risedronate and strontium ranelate are approximately 5 to 6 times higher than the price for non-proprietary weekly alendronate, and that the ICERs for these drugs compared with no treatment were very high. For example, the ICER for strontium ranelate for women aged 55 to 59 years with an independent clinical risk factor for fracture was approximately £55,000 per QALY gained (see section 4.2.22). The Committee noted that strontium ranelate would be the most likely choice to be considered for women who are unable to comply with the instructions for administration of alendronate, because the instructions for administration of alendronate and risedronate are similar. The Committee took the view that recommending drugs other than alendronate using the same criteria as alendronate for women who cannot take alendronate would not be justified in this case because of the very high ICERs for the alternative drugs. In reaching this decision the Committee had regard to the fact that the impact of refusing the more favourable recommendation is that there is no generally recommended preventative treatment for a particular group of patients who are at the lower end of fracture risk for which treatment was considered, but that the alternative drugs are recommended when these patients are at higher risk of fracture.

The Committee considered that it is important to maximise the number of patients who are able to take alendronate. Some women will be unable to take alendronate in any circumstances because of contraindication, intolerance or inability to comply with the instructions for administration. However, some women who have a disability that makes it difficult for them to comply with the instructions for administration of alendronate would be able to receive the drug if they received assistance in taking it. The Committee concluded that all reasonable steps should be taken to provide women who have a disability that makes it difficult for them to comply with the instructions for administration of alendronate, with such practical support and assistance with administration (for example through district nurse visits or other home support services), as will enable them to take the drug.

The Committee was aware of the availability of the FRAX internet-based tool, which can be used to calculate a 10-year absolute risk of fracture, developed under the auspices of the WHO. This assessment tool was based on the same epidemiological data that were used in the Assessment Group's model. However, the Committee was not persuaded that recommendations about treatment should be based on absolute risk as calculated using FRAX. Firstly, the Committee did not agree that all clinical risk factors included in the WHO algorithm were appropriate (see section 4.2.12 and section 4.3.9). Secondly, the Committee was aware that absolute fracture risk is not directly related to cost effectiveness, as outlined in the strontium ranelate assessment report issued in 2005. This is because absolute fracture risk is the total for all fracture sites, but different fracture sites have different impacts on quality of life, costs and mortality. Therefore, cost effectiveness is dependent on the contribution from each fracture site to the total fracture risk. Thirdly, the Committee had agreed that treatment benefit had not been proven for fracture risk associated with all independent clinical risk factors (section 4.3.9). Therefore, the Committee concluded that using a combination of T-score, age and number of independent clinical risk factors for fracture is more appropriate for defining treatment recommendations in this appraisal.

The Committee was made aware of data indicating that acid-suppressive medication leads to a small increase in fracture risk and that co-administration of acid-suppressive medication and bisphosphonates may lead to an increased fracture risk compared with bisphosphonate administration alone. The Committee was not persuaded by this evidence; it noted that the data are observational and have not been reported in full, and are different for different fracture sites and for different acid suppressors. Furthermore, the Committee was informed, during consultation, of analyses showing that acid-suppressive medication given in addition to risedronate did not increase fracture risk. The Committee concluded that caution should be exercised when considering the evidence about co-prescription of acid-suppressive medication and bisphosphonates.

The Committee noted sensitivity analyses that included the assumption of an increase in fracture risk for women for whom acid-suppressive medications are co-prescribed (see section 4.2.19). The analysis for treatment strategies did not decrease the T-scores at which the ICERs for alendronate fell below £30,000 to the T-scores established for strategies including strontium ranelate or raloxifene. The Committee also noted that the ICERs for treatment compared with no treatment for an individual woman with a relevant combination of age and T-score were not more favourable for strontium ranelate than for risedronate even if an effect of acid-suppressive medication was assumed. The Committee considered that the evidence for this effect was not sufficiently robust. However, it concluded that the relative positions of alendronate, risedronate and strontium ranelate would remain unchanged even if an effect of acid-suppressive medication was assumed. The Committee therefore concluded that it was not necessary to change its recommendations (section 1) to take account of acid-suppressive medication.

The Committee discussed the effect of calcium and vitamin D on the clinical effectiveness of the drugs considered. In the studies that formed the basis of this guidance, all participants were said to have adequate calcium and vitamin D levels. The Committee appreciated that the general population, particularly the elderly population, cannot be assumed to have an adequate dietary intake of calcium and vitamin D. It was also considered important to note that adequate levels (normal serum concentrations) of calcium and vitamin D are needed to ensure optimum effects of the treatments for osteoporosis. The Committee concluded that calcium and/or vitamin D supplementation should be provided unless clinicians are confident that women who receive treatment for osteoporosis have an adequate calcium intake and are vitamin D replete.

Following the outcome of the judicial review and the court ruling of March 2009, the Appraisal Committee considered the comments received from consultees on the Assessment Group's executable economic model, a report by the DSU reviewing these comments, and responses from the consultees to the DSU report.

The Committee considered the comments from consultees that the Assessment Group's model was not sufficiently transparent, lacked adequate documentation and could not be validated. It noted the number of consultations that took place during the appraisal guidance development, that the consultation documents had included descriptions of the model, and that assumptions and parameter values used had also been provided to consultees. The Committee was aware that instructions on how to run the model were released with the model and that consultees were able to run the model with changed input parameters. The Committee was satisfied with the exploration by the DSU of the functionality and validity of the model. The Committee noted that Servier stated that it had constructed its own economic model in order to validate the Assessment Group's model and to demonstrate the mathematical rationale to support its comments. The Committee noted that the results from Servier's model were very similar to those from the Assessment Group's model when similar assumptions and parameter inputs were used. The Committee was not persuaded by the consultees' doubt about the validity of the model, particularly since differences between the results obtained using Servier's model and the Assessment Group's model were largely because of differences in the assumptions used.

The Committee considered the comments from consultees that some inputs in the Assessment Group's model could not be changed and that it was unclear how fracture risk was calculated. The Committee noted that some of the fixed input parameters were inputs that do not need changing (such as the discount rate and standard mortality rates). Other fixed input values, such as the BMI and issues around the time horizon, were discussed separately (see sections 4.3.57 and 4.3.51 respectively). The Committee concluded that it was reasonable for some inputs in the model to be fixed. The Committee noted that fracture risks were calculated by the Assessment Group using the WHO algorithm in a separate spreadsheet and then entered into the model. It understood that the WHO algorithm itself was provided to the Assessment Group in 2005 as academic in confidence and that at that time NICE did not have permission from the owner of the algorithm to release it to consultees. The Committee understood that although the WHO fracture risk algorithm itself was not embedded in the economic model, the model could not be released because the algorithm could have been back-calculated from the fracture risks entered in the model and because the numbers of women with risk factors from the algorithm were included in the model.

The Committee considered the comments from consultees that the fracture risks entered into the model, calculated using the WHO algorithm, were different from fracture risks estimated using the FRAX tool. The Committee was aware that some differences could be because of the Assessment Group's use of midpoint ages in each 5-year age grouping. It also heard that the Assessment Group had verified the application of the WHO algorithm as provided in 2005, including all interactions between clinical risk factors, and was satisfied that the DSU had adequately assessed its application as being correct in the model. Because neither the DSU nor NICE has access to the algorithm used for the construction of the FRAX tool, the Committee was not in a position to comment further on differences between the 2 ways of estimating fracture risk. It concluded that differences between fracture risk estimates produced using the FRAX tool and those used in the Assessment Group's model were not in themselves a reason to doubt the correct use of the WHO algorithm within the Assessment Group's model.

The Committee considered the comments from consultees that mortality associated with clinical risk factors had been omitted from the Assessment Group's model, and noted the confirmation from the DSU that this was the case. It was persuaded that the inclusion of such additional mortality effects would increase the complexity of the model, and may increase the ICERs for the treatment of women with such clinical risk factors but decrease the ICERs for the treatment of women without such risk factors. The Committee agreed that the overall effect of including mortality associated with clinical risk factors in the model was unlikely to lead to a marked change in the overall results.

The Committee reviewed the consultee comments relating to the fixed BMI value of 26 kg/m² used in the Assessment Group's model. It noted the rationale for selecting this value (see section 4.2.46). It also noted that in the DSU's exploratory analysis using the WHO algorithm, no increase in fracture risk was identified for women with a higher or lower BMI when BMD was known. The Committee was aware of its recommendation to assess BMD in all women under the age of 75 years for whom treatment is being considered, and noted that BMI is a weak predictor of fracture when BMD is known. Therefore, the Committee concluded that the use of a fixed BMI value of 26 kg/m² did not lead to an unfavourable assessment of the cost effectiveness of the interventions.

The Committee considered comments from consultees that the fracture risk associated with alcohol consumption used in the model was incorrect. It noted that the DSU had determined that the WHO algorithm had been correctly implemented, and understood that alcohol consumption of more than 2 units per year was included as a risk factor in the model. The Committee also noted that in its recommendations it had chosen to use a higher level of alcohol consumption in the risk identification strategy, because only alcohol consumption of 4 or more units per year was identified as a statistically significant risk factor for fracture for women – the population considered in the guidance. The Committee also considered a consultee comment that stated that it was unclear whether smoking and corticosteroid use had been included in the model as risk factors. It noted that the DSU had determined that the WHO algorithm had been correctly implemented with regard to both smoking and corticosteroid use in the model. The Committee noted that the effect of smoking in women was not statistically significant when assessing risk of osteoporotic fractures taken as a whole. The Committee was therefore satisfied that risks associated with corticosteroids, smoking and alcohol consumption had been faithfully applied in the Assessment Group's model, and agreed that the levels of alcohol consumption and smoking that should be used in the risk identification strategy were a matter for the Committee to consider and determine. The Committee took the view that it is not appropriate to identify women at high risk of fracture on the basis of risks that were not statistically significant (such as smoking and consumption of fewer than 4 units of alcohol per year) and that, in addition, the impact of these risk factors could arguably be approached by a strategy of smoking cessation and reducing alcohol consumption. The Committee noted comments from the consultees that the Assessment Group's model should have been amended to reflect the Committee's agreed inclusion of risk factors. However, the Committee took the pragmatic view that such amendments would have added unnecessarily to the mathematical complexity of an already complex clinical situation. It noted that women who had taken corticosteroids were included in the model and therefore contributed to the underlying fracture risk, with the effect of reducing the ICERs for the treatment of the population of women considered in the recommendations.

The Committee considered consultee comments that giving equal weighting to different clinical risk factors for fracture in the Assessment Group's model was inaccurate. The Committee considered the complex results presented originally in the 2005 Strontium Ranelate Assessment Report related to the inclusion of different risk factors and combinations of risk factors. The Committee noted that it had previously agreed that a clinically workable risk identification and treatment strategy should include the grouping of risk factors as the only practical way forward. At the time of the model's development, no individual risk calculation tool was available. Even if such a tool had been used in the development of the guidance, the prediction of cost effectiveness from overall absolute fracture risk alone, as suggested by consultees, would not be appropriate, for 2 reasons. Firstly, risk factors have different effects on different fracture types, and the cost effectiveness of treatment depends on the relative contributions of each risk factor to fracture risk. Secondly, the effectiveness of the drugs in reducing fracture risk was limited to only some of the clinical risk factors (age, T-score of -2.5 SD or below and prior fracture). The Committee heard from the DSU that there was considerable uncertainty about the cost effectiveness of treating women based on absolute risk alone (see section 4.3.48). Therefore, the Committee concluded that, when developing the guidance, simplification of the model was justified to in order to produce workable recommendations.

The Committee reviewed a comment from a consultee that the methods used to model effects beyond 10 years were not adequately described. It noted that the DSU confirmed that consequences beyond 10 years were considered in the Assessment Group's model, and an expanded description of the methods used was provided in an annex to the DSU report. The Committee also noted that the DSU carried out a sensitivity analysis in order to establish the impact of any possible underestimation of the mortality risk after the 10-year time horizon. It noted that doubling of the mortality risk led to only very small changes in the results. The Committee therefore concluded that mortality effects beyond the 10-year time horizon had been reasonably accounted for in the model and that sufficient description of these methods had been made available to consultees.

The Committee considered comments from consultees that the population data on the distribution of BMD (T-score) were not appropriate. It noted the DSU response confirming that the UK epidemiological dataset from the Holt study had been correctly implemented in the Assessment Group's model, and that the assumptions about the normality of the distributions used were likely to favour treatment for women at risk of fracture. The Committee also noted that the particular UK epidemiological dataset used in the model had been originally suggested by consultees for this appraisal. The Committee concluded that the population data had been used appropriately in the model.

The Committee considered a comment from a consultee that using a single estimate of cost effectiveness for 5-year age groupings of women at risk of fracture could exclude women from being offered treatment. It noted that this identification method was a Committee decision, and that identification strategies based on other factors could make treatments less cost effective.

The Committee reviewed comments from a consultee that the application of the same disutility for the side effects associated with strontium ranelate and bisphosphonates was not correct, as the side effects of strontium ranelate are different from those of the bisphosphonates. The Committee was aware that the side effects observed for strontium ranelate in the clinical trials did not include gastrointestinal effects, but did include an increased risk of VTE. Because the increased risk of VTE was not included in the Assessment Group's model, the Committee had agreed that it was appropriate to include a disutility equivalent to the bisphosphonate base-case side-effect disutility to take account of this adverse effect.

The Committee reviewed comments from consultees about model assumptions or inputs that the Committee had directed the Assessment Group to use. It noted that issues such as treatment compliance, discount rates, costs of fracture, utility values for vertebral fracture and side-effect profiles used in the model had been considered and agreed by the Committee and reported in the guidance. The Committee also agreed that it had considered identification strategies for women at risk of fracture and, noting the advice of clinical specialists, it had recommended that women should have their BMD assessed by DXA scanning, except in certain circumstances as defined in the guidance. The Committee concluded that views expressed by consultees on the choice of modelling assumptions, input parameters and risk identification strategy were not about the operation of the Assessment Group's model, but were about Committee decisions that had already been discussed during development of the guidance.

The Committee also considered the consultees' view that the FRAX tool provides a 'mechanism to compute cost-effectiveness' according to clinical risk factors and that each of the current recommendations covers a wide range of absolute risk values, depending on the individual risk factors involved. The Committee understood that the FRAX tool is not an economic model, but a tool to estimate fracture risk. The Committee acknowledged that the current set of recommendations involved necessary simplifications from the more complex algorithm used to develop the Assessment Group's model. It was also aware that a direct prediction of cost effectiveness from absolute fracture risk alone would be inappropriate (see section 4.3.48).

The Committee concluded that the Assessment Group had provided an executable economic model and had implemented the WHO algorithm (as supplied) correctly. The Committee agreed with the DSU's comments that alterations to the modelling approach, as suggested by consultees, would not lead to significant improvements in the cost effectiveness of treatment for women at risk of fracture. The Committee confirmed that the model provided a suitable framework to allow it to make recommendations on the cost-effective use of treatment for women at risk of fracture. The Committee noted that assumptions used in the Assessment Group's model had been considered and agreed by the Committee in developing the guidance. It agreed that it would not be useful to request further analysis from the Assessment Group at this stage. The Committee further agreed that any exploration of how absolute fracture risk could be used in making treatment decisions would require a new assessment and appraisal. Therefore, the Committee concluded that the recommendations based on the Assessment Group's model were appropriate, and that the recommendations should remain unchanged.

The Committee noted the comments from some consultees that the guidance should be reviewed soon because the price of some of the appraised drugs had changed. The Committee noted that any possible price reductions could be offset by the use of the currently applicable discount rate, and that any review should also take into consideration how NICE might assess diagnostic tools such as absolute fracture risk prediction tools.