Fingolimod for the treatment of highly active relapsing–remitting multiple sclerosis

3 The manufacturer's submission

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

3.1 The manufacturer presented three populations in its submission:

3.2 The manufacturer undertook a systematic literature review and identified two randomised controlled trials, the FREEDOMS trial and the TRANSFORMS trial, which both assessed the efficacy and safety of fingolimod in adults with relapsing–remitting multiple sclerosis. The FREEDOMS trial was a phase III, multicentre, double-blind trial in which 1272 adults with relapsing–remitting multiple sclerosis were randomised to receive daily doses of oral fingolimod 0.5 mg (425 patients), oral fingolimod 1.25 mg (429 patients) or placebo (418 patients) for 24 months. In the FREEDOMS trial, 90 patients treated with fingolimod 0.5 mg and 79 patients treated with placebo were considered by the manufacturer to meet the criteria for population 1b. The TRANSFORMS trial was a phase III, multicentre, double-blind trial in which 1292 adults with relapsing–remitting multiple sclerosis were randomised to receive oral fingolimod 0.5 mg (431 patients) or oral fingolimod 1.25 mg (426 patients) once a day, or intramuscular Avonex 30 micrograms (435 patients) once a week for 12 months. In the TRANSFORMS trial, 191 patients who were treated with fingolimod 0.5 mg and 183 patients who received Avonex met the criteria for population 1b in the decision problem. Only data relating to fingolimod 0.5 mg are presented in the remaining sections of this document.

3.3 Patients were eligible for inclusion in the FREEDOMS and TRANSFORMS trials if they had an Expanded Disability Status Scale (EDSS) score between 0 and 5.5 (the EDSS ranges from 0 to 10 with 0.5-unit increments representing higher levels of disability), and at least one documented relapse during the previous year or at least two documented relapses during the 2 years preceding study enrolment. The primary outcome of the trials was annualised relapse rate. Secondary outcomes included disability progression confirmed after 3 months, time to first relapse and radiological outcomes, such as the number of new or enlarged lesions. In the FREEDOMS trial, patient-reported outcomes were assessed using the EuroQoL 5-dimension survey (EQ-5D). Quality-of-life data were collected in the TRANSFORMS trial using the Patient-Reported Indices for Multiple Sclerosis – Quality of life (PRIMUS–QoL), the Patient-Reported Indices for Multiple Sclerosis – Activities (PRIMUS–Activities) and the Unidimensional Fatigue Impact Scale (UFIS).

3.4 Results from the FREEDOMS and TRANSFORMS trials showed that the annualised relapse rates were statistically significantly reduced for all patients treated with fingolimod compared with placebo (0.18 compared with 0.40; p < 0.001) and those treated with fingolimod compared with Avonex (0.16 compared with 0.33; p < 0.001). Treatment with fingolimod also reduced the annualised relapse rates (primary outcome) for patients in population 1b in the manufacturer's submission (see section 3.1), compared with placebo (0.21 compared with 0.54; p < 0.001) and for those who received fingolimod compared with Avonex (0.25 compared with 0.51; p < 0.001). In the TRANSFORMS trial, 94.1% of all patients treated with fingolimod had no disability progression after 3 months (95% confidence interval [CI] 91.8 to 96.3) compared with 92.1% of all patients treated with Avonex (95% CI 89.4 to 94.7); however, this difference was not statistically significant (p = 0.25). Among the whole population of the FREEDOMS trial, 82.3% of patients treated with fingolimod had no disability progression after 3 months compared with 75.9% of all patients treated with placebo (p = 0.03). No statistically significant difference in disability progression between the treatment groups was reported for population 1b. The manufacturer pointed to the European Public Assessment Report of the European Medicines Agency which stated that the results in the subgroups with highly active disease were consistent with those obtained in the overall trial population.

3.5 Fingolimod was well tolerated by patients in the clinical trials. It was considered to have a comparable safety profile to placebo and to be associated with fewer adverse reactions than Avonex. The incidence of serious adverse reactions after treatment with fingolimod was low across both studies, with the most common being infections, macular oedema and transient atrioventricular block at treatment initiation. In the TRANSFORMS trial, adverse reactions leading to treatment discontinuation in population 1b were reported in 3.1% of patients treated with fingolimod compared with 1.6% of patients treated with Avonex. In the FREEDOMS trial, the rate of treatment discontinuation because of adverse reactions in population 1b was lower in patients receiving fingolimod (2.2%) compared with placebo (7.6%). There were no treatment-related deaths reported with fingolimod or Avonex treatment in the TRANSFORMS trial. In the FREEDOMS trial, no treatment-related deaths were reported among patients receiving fingolimod or placebo.

3.6 In the TRANSFORMS trial, patients who received fingolimod showed significantly less deterioration in their ability to perform daily activities according to the PRIMUS–Activities scale compared with patients receiving Avonex (change from baseline 0.08 in patients treated with fingolimod compared with 0.43 in patients treated with Avonex; p < 0.05). In addition, a slight, non-significant improvement in health-related quality of life, as measured on the PRIMUS–QoL scale, was observed in patients treated with either fingolimod or Avonex. After 6 months of treatment with fingolimod, patients showed a statistically significant improvement in UFIS score compared with patients treated with Avonex; however, at 12 months this difference between the groups was no longer statistically significant. In the FREEDOMS trial, no statistically significant changes from baseline for EQ-5D measures were observed in patients with relapsing–remitting multiple sclerosis treated with fingolimod or placebo.

3.7 To estimate the relative effectiveness of all of the comparators in the absence of direct evidence, the manufacturer conducted a mixed treatment comparison of 18 trials that assessed annualised relapse rates, disability progression and treatment discontinuation because of adverse reactions, for the following interventions: fingolimod, natalizumab, beta interferon-1a (Avonex, Rebif-22 and Rebif-44), beta interferon-1b (Betaferon), glatiramer acetate and placebo. The manufacturer acknowledged that although the populations in the included trials had a clinical diagnosis of relapsing–remitting multiple sclerosis, they did not fully meet the criteria for highly active disease described in the marketing authorisation for fingolimod. There was also considerable clinical heterogeneity between the trials with respect to permitted and actual prior use of disease-modifying treatments, treatment duration and the criteria used to define the trial end points. As a consequence, the manufacturer did not use the results of the mixed treatment comparison to inform the economic model. Instead, an indirect comparison using the TRANSFORMS and FREEDOMS trials provided an estimate of the relative efficacy of Avonex and placebo for population 1b in the model.

3.8 The manufacturer undertook a systematic search but did not identify any economic evaluations of fingolimod for the treatment of relapsing–remitting multiple sclerosis. The manufacturer submitted a de novo economic model that is structurally similar to models used in previous NICE technology appraisal guidance on treatments for multiple sclerosis (Natalizumab for the treatment of adults with highly active relapsing–remitting multiple sclerosis [NICE technology appraisal guidance 127] and Beta interferon and glatiramer acetate for the treatment of multiple sclerosis [NICE technology appraisal guidance 32]). The model is based on a Markov cohort approach and estimates disease progression through 21 disability states that are defined by EDSS score (ranging from 0 to 10) and account for disability for patients with relapsing–remitting multiple sclerosis (10 states), patients with secondary progressive multiple sclerosis (10 states) and death. In each cycle of the model, a patient with relapsing–remitting multiple sclerosis can progress to a worse EDSS state or remain in the same state. Patients can also convert from relapsing–remitting multiple sclerosis to secondary progressive multiple sclerosis; however, once a patient reaches this point in the disease course they cannot convert back to relapsing–remitting disease. Only people with relapsing–remitting multiple sclerosis and an EDSS score of 6 or less are assumed to receive disease-modifying treatment in the model. People with an EDSS score greater than 6, or with secondary progressive multiple sclerosis, are assumed to receive best supportive care.

3.9 Disability progression rates in the model were defined in a natural history transition matrix and derived from a longitudinal data set from patients with multiple sclerosis in Ontario, Canada. The manufacturer excluded patients with less progressive forms of relapsing–remitting multiple sclerosis from the data set and adjusted the natural history transition matrices to represent patients for whom treatment with disease-modifying therapies would be suitable. The probability of relapse in each model cycle was determined using published sources to estimate the natural history of relapses by disease type and EDSS stage.

3.10 All patients were individually followed through the model from treatment initiation for a maximum of 50 years. Probabilities for all-cause mortality for the general population were derived using current statistics for England and Wales, and were then adjusted for patients with multiple sclerosis using mortality ratios from published sources. A published equation was also used to predict the excess mortality for individual EDSS states.

3.11 The relative risks of annual relapse rate and of disability progression for fingolimod treatment compared with best supportive care were calculated from the FREEDOMS trial. The corresponding relative risk value for Avonex was calculated indirectly from the TRANSFORMS trial. The relative risks associated with disease progression and relapse were constant over the entire on-treatment period. Discontinuations because of adverse reactions were included in the model based on trial data. The relative risks for disease progression were not applied to patients with secondary progressive multiple sclerosis (who receive best supportive care in the model); instead, patients entering this disease state followed the natural history of the disease (as predicted by data from Ontario, Canada).

3.12 Although quality-of-life data were collected in the TRANSFORMS and FREEDOMS trials, the manufacturer did not use these to estimate utilities for the model. Instead, published EDSS-based EQ-5D scores were used, in line with those from NICE technology appraisal guidance 127 and 32. Health-related quality of life was assumed to remain constant over time for each EDSS score, but a single adjustment (0.01 utility gain for each 5-year period) was made to reflect the time since diagnosis. Utility decrements attributable to adverse reactions were applied for the whole duration of the treatment period. The model also incorporated caregiver's disutility for each EDSS score in the base case, in line with estimates from NICE technology appraisal guidance 127. The maximum disutility for a caregiver of a person with multiple sclerosis was assumed to be 0.14 (EDSS 9).

3.13 The resource costs included in the model were drug acquisition costs, administration and monitoring costs, and the cost of the disease, which included the cost of each EDSS state, the cost of relapse and the costs associated with serious adverse reactions. Costs associated with non-serious adverse reactions were not considered in the model. The model assumes that when patients discontinue treatment with disease-modifying therapy and receive best supportive care, the only costs incurred are the disease costs by EDSS states.

3.14 The original base-case incremental cost-effectiveness ratio (ICER) for fingolimod compared with Avonex was £55,634 per QALY gained for population 1b (patient access scheme not included). Cost-effectiveness analyses for population 1a and population 2 (defined in section 3.1) were not provided by the manufacturer. One-way sensitivity analyses suggested that the ICER for fingolimod compared with Avonex was most sensitive to the relative risks of disease progression assumed for fingolimod and Avonex, and the relative risk of relapse for Avonex. Uncertainty in all other parameter values led to only small changes in the ICER. Results of a probabilistic sensitivity analysis showed that there was a 12% probability that the base-case ICER was less than £20,000 per QALY gained, and a 26% probability that it was less than £30,000 per QALY gained.

3.15 The manufacturer explored uncertainty in the model caused by structural assumptions, including possible waning of treatment effect and the assumed time horizon. When treatment efficacy was assumed to be reduced by 50% or 75% after the first 2 years of treatment, the ICER increased to £73,191 and £85,266 per QALY gained respectively. When the time horizon was shortened to 10 years and 20 years, the ICER increased to £97,159 and £64,280 per QALY gained respectively.

3.16 The manufacturer acknowledged that there was considerable overlap between the populations defined in the marketing authorisation for fingolimod, and provided an analysis for a subgroup of population 1b that excluded patients who also met the criteria for population 2 (that is, it excluded those with rapidly evolving severe multiple sclerosis). The relative treatment effects estimated from the trials for this subgroup were significantly different from those estimated for the whole of population 1b. In particular, the risk of disease progression with Avonex was estimated (by indirect comparison) to be higher than with placebo. The manufacturer's ICER for fingolimod compared with Avonex in this subgroup was £18,741 per QALY gained (patient access scheme not included). No sensitivity analyses were conducted for this subgroup.

3.17 In its response to the first appraisal consultation document, the manufacturer included a patient access scheme, which was agreed with the Department of Health, to apply a simple confidential discount to the list price of fingolimod. The manufacturer's deterministic base-case ICER for fingolimod in population 1b reduced to £10,839 per QALY gained compared with Avonex when the discounted price of fingolimod was included in the model. The probabilistic ICER determined by the manufacturer, including the patient access scheme, was £15,825 per QALY gained. Probabilistic sensitivity analyses suggested that there was a 58% chance that the ICER for fingolimod would be less than £30,000 per QALY gained when the discounted price was included.

3.18 Sensitivity analyses provided by the manufacturer in response to the first appraisal consultation document suggested that the ICER for fingolimod compared with Rebif-44 was £27,774 per QALY gained (patient access scheme included). The manufacturer noted that the data to inform this analysis were from patients with relapsing–remitting multiple sclerosis, rather than from those who had a suboptimal response to disease-modifying therapy (that is, population 1b), and therefore the true ICER for population 1b was likely to be lower. The manufacturer also compared fingolimod with Rebif-22 and Betaferon using adjusted data from the mixed treatment comparison. Efficacy rates for each treatment were scaled down by 13.25% to account for the fact that the clinical effects seen in the trials for people with relapsing–remitting multiple sclerosis were likely to be reduced in population 1b. The ICERs from this analysis for fingolimod were £23,587 per QALY gained compared with Rebif-22, and £27,660 per QALY gained compared with Betaferon (patient access scheme included).

3.19 In its response to the second appraisal consultation document, the manufacturer provided an updated economic model incorporating the following assumptions which the Committee had concluded to be more plausible:

3.20 In response to the second appraisal consultation document, the manufacturer also explored the directional effect on the ICER of changing the natural history transition matrix to slow disability progression. The manufacturer noted that in the FREEDOMS and TRANSFORMS trials, the greatest disability progression was three EDSS states within a 12-month period. In a scenario analysis which assumed that people could not progress more than one EDSS state each year in the relapsing–remitting multiple sclerosis natural history matrix, the manufacturer's probabilistic ICER for fingolimod compared with Avonex increased to £21,244 per QALY gained. When it was assumed that people could not progress more than one EDSS state each year in the secondary progressive multiple sclerosis natural history matrix, the probabilistic ICER increased to £19,774 per QALY gained. In the manufacturer's view, these scenario analyses were based on extreme assumptions that did not reflect the available clinical data for patients with relapsing–remitting multiple sclerosis.

3.21 The ERG considered that the TRANSFORMS and FREEDOMS trials were well designed to assess the efficacy of fingolimod in patients with relapsing–remitting multiple sclerosis. The ERG noted that the populations in the clinical trials were broader than those defined in the marketing authorisation for fingolimod, but considered that the manufacturer's post-hoc subgroup analyses provided a reasonable approximation to the populations in the marketing authorisation. The ERG noted that population 1b comprised only 43.6% of patients in the TRANSFORMS trial and 19.7% of patients in the FREEDOMS trial. The ERG was concerned that because of the smaller number of patients, the power of the trials to assess fingolimod relative to the comparators in the populations covered by the marketing authorisation was reduced. However, the ERG noted that the SPC for fingolimod states that 'further analyses of clinical trial data demonstrate consistent treatment effects in the highly active subgroups of relapsing–remitting multiple sclerosis'. The ERG was also concerned that there was considerable overlap between the populations and requested separate analyses from the manufacturer for population 1a, population 2, and populations 1a and 1b with patients who also met the criteria for population 2 excluded. The manufacturer provided analyses only for population 1b excluding patients who also met the criteria for population 2.

3.22 The ERG was concerned by the manufacturer's approach of using only Avonex as the comparator treatment for population 1b. The ERG noted that population 1b constitutes patients with highly active disease that has remained unchanged or worsened despite treatment with beta interferon. In the ERG's view, a comparison with Avonex may represent continued use of a treatment that is suboptimal in this group of patients, and may also cause adverse reactions. The ERG also noted that the results from the manufacturer's mixed treatment comparisons did not yield clear differences between the beta interferons in patients with relapsing–remitting multiple sclerosis in terms of disease progression and annualised relapse rates. It cautioned that a comparison solely with Avonex could underestimate the ICER of fingolimod and therefore reasoned that a comparison including best supportive care would have been more appropriate.

3.23 The ERG considered the additional cost-effectiveness analysis from the manufacturer for the subgroup consisting of population 1b without those who also met the criteria for population 2 (section 3.16). The ERG noted that the ICER for fingolimod compared with Avonex was more favourable for this subgroup than for the whole of population 1b. The ERG considered that this difference was largely attributable to the revised relative efficacy estimates for Avonex from the manufacturer's indirect comparison for this subgroup. This suggested that Avonex provides less benefit than placebo (that is, that Avonex was associated with an increased risk of disease progression compared with placebo).

3.24 The ERG was concerned about the resources and costs assumed in the manufacturer's original model. The ERG was unclear why the costs associated with only some severe adverse reactions were included in the model, and why the costs associated with non-serious adverse reactions were not included. The ERG was also unclear whether costs associated with relapsing–remitting multiple sclerosis were different from those associated with secondary progressive multiple sclerosis. In addition, the ERG noted that the cost of relapse used in the model was significantly different from the cost from other data sources and in NICE technology appraisal guidance 127. In the ERG's view, the manufacturer had not adequately justified the administrative and monitoring costs for fingolimod and Avonex. In particular, it was unclear why the manufacturer assumed that patients treated with Avonex would need two more neurology visits in the first year of treatment than patients who received fingolimod. The ERG noted that in response to consultation on the first appraisal consultation document the manufacturer provided additional justification for the resource use and cost assumptions included in the model, and showed that alternative assumptions only slightly increased the ICER.

3.25 The ERG noted that although the manufacturer had included a probabilistic model in its original submission, the cost-effectiveness results presented in the original submission were deterministic. The ERG provided a probabilistic analysis for the manufacturer's original base case that gave an ICER of £69,787 per QALY gained for fingolimod compared with Avonex (patient access scheme not included). This ICER was noted to be substantively higher than the manufacturer's original deterministic estimate of £55,634 per QALY gained.

3.26 The ERG noted that the manufacturer had presented adjusted hazard ratios in its original submission to describe the relative effect on disease progression of treatment with fingolimod compared with Avonex. However, these estimates were not employed in the model, and instead relative risks from unadjusted trial data were used. The ERG analysed the manufacturer's original base case (population 1b) using hazard ratios instead of relative risks (patient access scheme not included) and noted that, in an incremental analysis, the probabilistic ICER for fingolimod compared with best supportive care was £94,094 per QALY gained. In an incremental analysis, Avonex was extendedly dominated (that is, the ICER for Avonex was higher than the ICER for the next more effective alternative [fingolimod]). For population 1b, excluding those who also met the criteria for population 2, the ICER for fingolimod compared with best supportive care was £81,369 per QALY gained and Avonex was dominated by best supportive care (Avonex was less effective and more expensive). The ERG concluded that the incremental analysis shows that in both populations Avonex is either dominated or extendedly dominated. The ERG therefore advised that the cost effectiveness of fingolimod should be derived from this incremental analysis. The ERG acknowledged that the manufacturer had provided an additional analysis in response to the first appraisal consultation document. In this analysis, hazard ratio values were applied as relative risks in the model, and this reduced the deterministic base-case ICER from £55,634 (section 3.14) to £52,906 per QALY gained for population 1b. In the ERG's view the manufacturer's additional analyses did not address its initial concerns, because it considered that the hazard ratio values used should have been applied as hazard ratios, rather than relative risks, in the probabilistic (not the deterministic) model.

3.27 The ERG was concerned that the manufacturer provided insufficient justification in its original submission for choosing published EDSS-based EQ-5D scores rather than the trial outcomes to derive utility data. The ERG cautioned that although the published utility data had been used in previous NICE technology appraisal guidance on treatments for multiple sclerosis, these data had been criticised for coming from studies with low response rates, possible selection bias and unrepresentative populations. The ERG suggested that because the manufacturer's base case targeted a very specific patient population (population 1b), it would have been more appropriate to use utility data for these patients, which were available directly from the FREEDOMS and TRANSFORMS trials. The ERG conducted an exploratory analysis to assess the impact of using the average utilities for each EDSS score in the trial (up to EDSS 6) and then using published sources to impute the missing utility data for EDSS scores of 7 and above. In this analysis, the probabilistic ICER for fingolimod compared with best supportive care in population 1b increased to £106,824 per QALY gained (patient access scheme not included) when the missing utility estimates for EDSS scores 7 to 10 were imputed using values from NICE technology appraisal guidance 127. Based on these results, the ERG cautioned that changing the utility values of only three EDSS scores has a significant impact on the ICER for fingolimod. The ERG acknowledged that in the manufacturer's response to the first appraisal consultation document an additional analysis was provided in which utility data from the trials were used for EDSS states up to 6, and then data from a study by Orme et al. (2007) were used for the remaining 13 states. Using data from the FREEDOMS and TRANSFORMS trials reduced the manufacturer's original deterministic base-case ICER for fingolimod compared with Avonex to £52,982 per QALY gained and £52,866 per QALY gained respectively. In its critique of the manufacturer's original submission, the ERG had previously explored a number of alternative scenarios for incorporating trial utility data into the model, which were shown to both increase and decrease the ICERs. The ERG cautioned that the model predictions are highly sensitive to the utility estimates and therefore it is important to fully justify the data sources and imputation methods used.

3.28 The ERG was concerned about the representativeness of the initial EDSS score distribution used in the manufacturer's original model. The ERG examined a number of scenarios and showed that the cost effectiveness of fingolimod varies depending on the initial distribution of patients across EDSS states. The ICER for fingolimod compared with best supportive care in population 1b was £78,338 per QALY gained when it was assumed that all people enter the model with an EDSS score of 4, and £102,718 per QALY gained when all people enter the model with an EDSS score of 2 (patient access scheme not included). The ERG considered that its analyses highlighted that the model was highly sensitive to the initial population EDSS distribution assumed.

3.29 The ERG noted from market share data provided by the manufacturer that Rebif-44 is commonly prescribed in the NHS for the treatment of multiple sclerosis. The ERG conducted two exploratory analyses that included Rebif-44 as an additional comparator. The first analysis used direct evidence on the effectiveness of Rebif-44 and Avonex, and the second used the results from the mixed treatment comparison provided by the manufacturer (patient access scheme not included). Deterministic results were calculated using relative risks from the direct evidence and showed that Rebif-44 dominated Avonex in population 1b and in the subgroup of population 1b that excluded patients who also met the criteria for population 2. However, for population 1b, Rebif-44 was extendedly dominated (that is, the ICER for Rebif-44 was higher than the ICER for the next more effective alternative [fingolimod]) in an incremental analysis. The ICER for fingolimod compared with best supportive care was £91,059 per QALY gained for population 1b, and £79,315 per QALY gained for population 1b without those who also met the criteria for population 2. When data from the manufacturer's mixed treatment comparison were used instead, Avonex was dominated by Rebiff-44 for both populations. The ICER for fingolimod compared with best supportive care was £119,213 per QALY gained for population 1b and £119,746 per QALY gained for population 1b without those who also met the criteria for population 2.

3.30 The ERG noted that the baseline relapse rates in the manufacturer's original model were dependent on EDSS state but were then adjusted by the relative risk of relapse with a particular disease-modifying therapy compared with best supportive care. The ERG was concerned that these estimates for relative effect were taken from different data sets and therefore had no implicit correlation. In addition, the ERG cautioned that the impact of disease-modifying therapy could be double-counted in the model. To explore this, the ERG re-ran the original model (patient access scheme not included) and excluded all direct treatment effects on relapse rates. For population 1b, the ICER for fingolimod compared with best supportive care increased to £112,294 per QALY gained compared with the ERG's base-case estimate of £94,094 per QALY gained. Avonex was extendedly dominated by best supportive care and fingolimod. For population 1b without those who also met the criteria for population 2, the ICER for fingolimod compared with best supportive care was £98,019 per QALY gained compared with £81,369 per QALY gained in the ERG's base case, and Avonex was dominated by best supportive care.

3.31 The ERG was concerned that the underlying progression rates predicted in the manufacturer's original model were higher than the rates seen in the TRANSFORMS and FREEDOMS trials, but the manufacturer did not explain the differences between the model predictions and the trial observations. The ERG conducted four scenario analyses to examine the sensitivity of the manufacturer's model to natural history progression rates. These included reducing natural history progression transitions by 50%, 25% and 10%, and increasing them by 10%. Reducing the natural history progression rates substantially increased the ICER for fingolimod compared with best supportive care. Assuming a 50% decrease in natural history progression rate increased the ICER to £252,147 per QALY gained for population 1b, and to £191,027 per QALY gained for population 1b without those who also met the criteria for population 2 (patient access scheme not included). The ERG considered that the model predictions were highly sensitive to the natural history progression data used in the model.

3.32 The ERG noted that the manufacturer's original model assumed a constant and continued treatment effect in patients who receive disease-modifying therapy, as long as they remain on treatment, over the time horizon of the model. In the ERG's view this assumption, which was informed by trials of only 12 months' and 24 months' duration, was optimistic. The ERG conducted an exploratory analysis (expanding on the manufacturer's sensitivity analysis; patient access scheme not included) to evaluate the possible waning of treatment effect over time. Treatment efficacy was modelled to wane by 50%, 75% or 100% of the original level after 2 years and 5 years. In all scenarios, the more the efficacy was reduced, the higher the ICER. This is because the costs of treatment are still incurred but less health benefit is obtained. When the treatment effect was assumed to wane after 2 years, the ICERs for fingolimod compared with best supportive care were £140,282 per QALY gained (50% efficacy reduction), £177,674 per QALY gained (75% efficacy reduction) and £249,735 per QALY gained (100% efficacy reduction). When it was assumed that the treatment effect does not wane until after 5 years, the ICERs for fingolimod compared with best supportive care were £114,532 per QALY gained for a 50% efficacy reduction, £131,135 per QALY gained for a 75% efficacy reduction and £143,869 per QALY gained for a 100% efficacy reduction.

3.33 The ERG reviewed the revised model provided by the manufacturer in response to the first appraisal consultation document. It noted that the manufacturer only adjusted the drug acquisition cost in the model in line with the patient access scheme. The ERG noted that the manufacturer had not updated the model to reflect the assumptions that the Committee had considered to be most plausible during the first Committee meeting. The ERG ran the manufacturer's updated model including the patient access scheme and produced a probabilistic base-case ICER for fingolimod compared with Avonex of £14,997 per QALY gained. The ERG also produced an incremental analysis using the manufacturer's updated model which showed that the probabilistic ICER for fingolimod was £58,024 per QALY gained compared with best supportive care, and Avonex was extendedly dominated by fingolimod and best supportive care with an ICER of £176,357 per QALY gained. The ERG cautioned that despite the discounted drug acquisition cost, the remaining uncertainties around the model and its inputs still remained.

3.34 The ERG reviewed the revised model provided by the manufacturer in response to the second appraisal consultation document. The ERG confirmed that the manufacturer had correctly revised its model to incorporate the assumptions that the Committee had considered important for exploring the uncertainty surrounding the base-case ICER during its second meeting. The ERG confirmed that it was able to approximate the manufacturer's probabilistic cost-effectiveness results for fingolimod compared with Avonex and for fingolimod compared with Rebif-44. However, it was unable to fully verify the manufacturer's ICER for fingolimod compared with the weighted average of the comparators, because results for some of the comparators (best supportive care and Betaferon) could not be reproduced from the manufacturer's model. The ERG urged caution in using a weighted average for the comparators and considered that a fully incremental analysis was a more appropriate way to explore the cost effectiveness of fingolimod relative to multiple comparators.

3.35 Full details of all the evidence are in the manufacturer's submission and the ERG report, which are available on the NICE website.