Cost-utility of fingolimod compared with dimethyl fumarate in highly active relapsing-remitting multiple sclerosis (RRMS) in England

Abstract

Objective:

The cost-effectiveness of new oral disease-modifying therapies (DMTs) has not been modeled in highly active (HA) relapsing-remitting multiple sclerosis (RRMS) requiring escalation therapy. This study sought to model the cost-effectiveness of fingolimod compared to dimethyl fumarate (DMF), for which relevant HA RRMS sub-group data were available, from the perspective of the National Health Service (NHS) in England.

Methods:

A cohort Markov model based on Expanded Disability Status Scale scores, similar to previous model designs, was constructed. Published post hoc clinical data in the HA RRMS sub-groups were taken from the pivotal trials for fingolimod and DMF vs placebo. Utility data for each health state and for relapses were used in line with previous similar models. Published costs were inflated to NHS cost year 2013–2014 and UK list prices used for both drugs. Possible Patient Access Scheme (PAS) discount scenarios were investigated.

Results:

In the base case, using list prices for each DMT, the average probabilistic incremental cost-effectiveness ratio (ICER) for fingolimod vs DMF was found to be £14,076, with a 73% chance of fingolimod being cost-effective at a willingness-to-pay threshold of £30,000. Scenario and sensitivity analyses showed that uncertainty in disability progression efficacy was a key model driver. The model was robust to other changes and the majority of PAS permutations do not contradict the base case finding of cost-effectiveness of fingolimod.

Conclusions:

In conclusion, fingolimod remains cost-effective in HA RRMS following the introduction of DMF to the UK market, and this paper supports the evidence that has led fingolimod to be the only oral DMT reimbursed for HA RRMS in England. This model supports the restriction imposed by National Institute for Health and Care Excellence (NICE) on DMF in HA RRMS and highlights the importance of considering different sub-groups of multiple sclerosis when performing health economic analyses.

Keywords::

• Multiple sclerosis
• Fingolimod
• Dimethyl fumarate
• Cost-utility
• Cost-effectiveness
• Highly active relapsing-remitting multiple sclerosis

Introduction

Multiple Sclerosis (MS) is a chronic, immune-mediated disease characterized by inflammation of the central nervous system, which leads to considerable disability1. The most commonly diagnosed type of MS is relapsing-remitting multiple sclerosis (RRMS), which is associated with a repeating pattern of relapses followed by a period of remission2. Although patients’ condition may spontaneously improve for a time, RRMS has an overall progressive trend, leading towards permanent disability3. MS is the most common cause of neurological disability in young adults worldwide4, and the estimated prevalence of MS in the UK is ∼100,000 patients5. MS has a considerable impact on patients’ quality-of-life (QoL), particularly during relapses and as the disease progresses6. It additionally represents a large economic burden for the patient, their carers, and the health service7.

Drug therapies that are able to reduce the number of relapses and/or slow disease progression in RRMS are known as disease-modifying therapies (DMTs); a DMT is typically initiated as soon as a diagnosis of RRMS is confirmed8. A sub-population of RRMS patients exhibit high disease activity despite treatment with at least one DMT; these patients are considered to have highly active (HA) RRMS, a definition first introduced by the European Medicines Agency (EMA)9. Patients with HA RRMS will require escalation therapy and should, therefore, switch to an alternative DMT. The definition of HA RRMS given in the fingolimod Summary of Product Characteristics (SPC)10 is:

Those who have failed to respond to a full and adequate course (normally at least 1 year of treatment) of at least one disease modifying therapy. Patients should have had at least one relapse in the previous year while on therapy, and have at least nine T2-hyperintense lesions in cranial MRI [magnetic resonance imaging] or at least one Gadolinium-enhancing lesion. A ‘non-responder’ could also be defined as a patient with an unchanged or increased relapse rate or ongoing severe relapses, as compared to the previous year.

Whilst the first available DMTs were administered via injection or infusion, three oral DMTs are now available in both the European Union (EU) countries and the US, namely fingolimod, teriflunomide, and dimethyl fumarate (DMF); the licensed indications for fingolimod differ between the US and the EU, but for teriflunomide and DMF the indications are broadly similar, with the EU indication for both being adult patients with RRMS. Fingolimod was the first oral DMT approved in the US in 2010 and in the EU in 2011, following pivotal placebo-controlled (FREEDOMS11 and FREEDOMS II12) and active-controlled (TRANSFORMS13) phase 3 trials. The licensed indication approved in the EU for fingolimod is more restrictive than that in the US, comprising two sub-populations with higher disease activity, rather than all adult patients with active RRMS; namely RRMS patients with HA RRMS or rapidly-evolving severe (RES) RRMS, as defined in the SPC10. Teriflunomide was approved in the US in 2012 and in the EU in 2013, following two placebo- and one active-controlled trials (TEMSO14, TOWER15, and TENERE16, respectively), whilst DMF was approved in the US in 2013 and in the EU in 2014, following two placebo-controlled trials (DEFINE17 and CONFIRM18). Zhang et al.19 have published a cost-effectiveness model comparing these three oral treatments as initial therapies in RRMS in the US, where all three are licensed for this indication; this analysis found DMF to be a cost-effective option. The applicability of this study to countries within Europe is limited, not least given the differing licensed indications issued by the EMA.

In England, the use of fingolimod in HA RRMS was demonstrated to be cost-effective in an appraisal by the National Institute for Health and Care Excellence (NICE)20, and consequently fingolimod was recommended for use by the National Health Service (NHS) in this indication. The cost-effectiveness analysis which informed this appraisal has not been published elsewhere.

Subsequent to this appraisal of fingolimod, both DMF21 and teriflunomide22 were licensed by the EMA for the treatment of RRMS, as noted above. Similarly to Zhang et al.19, the appraisals of teriflunomide23 and DMF24 by NICE were also based on their use as initial DMTs in the whole RRMS population. However, in the NICE appraisals, neither teriflunomide nor DMF was recommended for use in HA (and RES) RRMS patients. As no cost-effectiveness analysis in the HA RRMS population was provided in either submission the aim of this study was to conduct a cost-utility analysis in the HA RRMS population, in order to determine whether fingolimod remains cost-effective amongst oral therapies when evaluated in the appropriate HA RRMS sub-population in England.

Patients and methods

Population and pivotal trials

The patient cohort considered by the economic model was that of HA RRMS, as defined in the fingolimod SPC10. The clinical data sources for the economic model were identified by a systematic literature review. The search strategy for this review combined search terms for randomized controlled trials with terms for ‘highly active RRMS’ and was not restricted by treatment. The databases searched were Ovid Medline, Ovid EMBASE (platforms searched in September 2014), and the Cochrane Library (database searched in August 2014). Additionally, conference proceedings from 2012–2014 were searched for eligible studies, as were the bibliographies of relevant Cochrane systematic reviews. Relevant NICE guidance, entries on clinicaltrials.gov and the European Product Assessment Reports (EPARs) of relevant treatments were also considered. All search results were reviewed by a primary reviewer and checked by a secondary reviewer.

No data were found to assess the efficacy of teriflunomide in HA RRMS patients, precluding an analysis of all three oral therapies in HA RRMS. The clinical efficacy data for fingolimod were taken from a post hoc analysis of the pooled FREEDOMS and FREEDOMS II trials for patients with disease activity and previously treated with any DMT25. The clinical efficacy data for DMF were taken from the relevant sub-group analysis from the EPAR for DMF, available from the EMA26. This is a post hoc analysis of the pooled DEFINE and CONFIRM trials for patients with disease activity and previously treated with beta-interferon. Analyses of fingolimod in HA RRMS patients previously treated with beta-interferon and previously treated with glatiramer acetate show consistent efficacy27 and, therefore, this difference in the available clinical efficacy data is not expected to bias the results—this assumption is also tested, as described later in the paper.

The inclusion and exclusion criteria for the pivotal trials were similar11^,12^,17^,18 and a comparison of the baseline characteristics of the pivotal trials that provide the efficacy inputs is given in the Supplemental Material (Table 5). The baseline characteristics across the pivotal trials were similar to each other and those of patients in the UK NHS MS risk sharing scheme28.

There were some differences in how key outcomes were defined between trials; the detailed definitions of these outcomes are given in the Supplemental Material (Table 6). With respect to relapses, as the inference is on the relative rate scale, it is assumed that a relative change in relapse rate due to treatment or patient characteristics would be the same under the differing definitions of relapse used. The definitions of disease progression were for the most part similar, with differences applicable only to progression from baseline Extended Disability Status Scale (EDSS) states of 0 and 5.5. The more stringent definition of disease progression in the DMF trials would mean that patients would be less likely to be classed as having progressed, which could make the results for DMF more favorable than those for fingolimod in the FREEDOMS trials. Given this variation, the effect of these differences in definitions was investigated by way of scenario analyses, using efficacy data for fingolimod adjusted to match the definition applied in the DMF trials.

The main limitation of the clinical evidence base is that the sub-groups were defined post hoc in both cases and, therefore, the trials were not powered to detect a significant treatment effect in these sub-groups.

Structure of the economic model

In order to determine the cost-utility of oral DMTs in HA RRMS, including consideration of conversion to secondary progressive MS (SPMS), a discrete time, cohort, Markov model based on 10 EDSS scores giving 21 states (10 for RRMS, 10 for SPMS, and the ‘Death’ state) was constructed. The model takes the perspective of the NHS and Personal Social Services (PSS) in England, as per the NICE reference case29. Two cohorts of patients, one for the intervention (fingolimod) treatment sequence and one for the comparator (DMF) treatment sequence, are simulated in the model. DMF is the only comparator modeled, as the other oral DMT licensed for use in highly active RRMS, teriflunomide, has no evidence available in this sub-group.

The Markov structure used, with annual cycles, was based on the models used in all NICE single technology appraisal submissions for DMTs in RRMS submitted prior to 2014, which are themselves all explicitly based on the model produced by the School of Health and Related Research (ScHARR) for the original NICE multiple technology appraisal of DMTs for RRMS30–35. This structure is, at present, the de facto ‘standard’ model of the cost-effectiveness of DMTs in RRMS in the UK. Figure 1 illustrates the main states of the model, which are based on EDSS scores and account for the observed conversion from RRMS to SPMS in patients. While the EDSS comprises 20 categories (from 0–10, at 0.5 intervals between 1 and 10), in the model the “half” states of the EDSS are combined with the state above, except for 9.5 which is combined with 9 (and 8.5), and the death state is treated separately.

Figure 1. Summary of model states and possible transitions between them (death state not shown).

All patients initiate therapy with fingolimod or DMF treatment. The distribution of patients employed at the beginning of the first cycle is set to the HA RRMS population from the pooled FREEDOMS, FREEDOMS II, and TRANSFORMS13 trials (TRANSFORMS was a 12-month trial of fingolimod vs beta-interferon), as this represented the most complete dataset of baseline characteristics in this patient group. For subsequent cycles, the ending distribution of patients of the previous cycle is used. From the starting distribution, the proportion of patients who discontinue treatment due to experiencing an adverse event (AE) is calculated, based on the drug withdrawal probability. If a patient withdraws, they are removed from treatment and transferred to best supportive care (BSC). Mortality is incorporated following withdrawal in each model cycle, and is based on gender-weighted general mortality for a specific age group (published data from the Office for National Statistics36) and the relative risk of death due to MS taken from published sources37^,38. The proportion of females to males in cohorts is assumed to be constant across the span of simulations. The initial characteristics of the patient population entering the model were based on pooled baseline characteristics for HA RRMS patients from TRANSFORMS, FREEDOMS, and FREEDOMS II studies, and are given in the Supplemental Material (Table 7).

Patients who do not die in a given cycle are transferred between EDSS scores within RRMS or to SPMS, according to transition matrices derived from disease natural history studies, modified by the efficacy results of the relevant treatments. Natural history transitions for the EDSS states 0–7 in HA RRMS were derived from the pooled non-responders from the placebo arms of the FREEDOMS I and FREEDOMS II trials and the natural history transitions for EDSS score 8 were based on the longitudinal dataset from London Ontario39. This approach was taken because the London Ontario dataset did not provide natural history transitions specifically for HA non-responder patients, but the FREEDOMS trials could not provide adequate information on the transitions in the higher EDSS states. In SPMS the natural history transitions were taken from the London Ontario dataset39.

As discussed, patients may discontinue treatment and move to BSC as a result of AEs. In addition, after the transitions to new EDSS states take place, patients who have reached the EDSS threshold for discontinuing DMT or have been transferred to SPMS are assumed to receive BSC. The EDSS threshold is by default set to EDSS 6.5, because the guidelines of the Association of British Neurologists recommend DMTs for patients who can walk independently, i.e., those with an EDSS score of 6.5 or less40.

To estimate the average number of patients in a given Markov state, the life table method is used, which averages the number of patients in a given EDSS state at the beginning of the cycle and after the transitions to new EDSS scores are complete41. The resultant mid-year Markov traces are used to calculate the costs and the utilities of the intervention and the comparator treatment.

Clinical inputs

Modeled clinical efficacy parameters are provided in Table 1. The relative efficacy of the treatments was derived from the sub-group analyses for HA RRMS available for each DMT and then adjusted to account for an assumed waning effect. In the base case, the efficacy of both treatment arms is assumed to wane by 75% at 2 years and 50% at 5 years. This represents a hypothetical decrease in efficacy caused by prolonged used of a drug and is based on the waning assumptions adopted by NICE in the appraisal of DMF24. It is assumed that the transition probabilities to SPMS and within SPMS are not affected by the type of drug used25^,26.

Table 1. Clinical parameters for fingolimod and DMF in the HA RRMS sub-group.

Drug	Outcome vs placebo (over 24 months)		Source
Fingolimod	ARR ratio (95% CI)	0.52 (0.40–0.69)	Pooled FREEDOMS & FREEDOMS II HA sub-group25
	HR for 3-month confirmed disability progression (95% CI)	0.66 (0.45–0.96)
DMF	ARR ratio (95% CI)	0.57 (0.39–0.84)	Pooled DEFINE & CONFIRM HA sub-group26
	HR for 3-month confirmed disability progression (95% CI)	1.19 (0.66–2.15)

ARR, annualized relapse rate; CI, confidence interval; HR, hazard ratio.

Previous cost-utility models considered by NICE31^,33^,34 have taken a variety of approaches to selecting which adverse events (AEs) to include. In order to apply a consistent approach across each comparator within this model, events noted as serious for each DMT were included in the model and a scenario analysis was used to test the influence of including these in the model. For fingolimod, the AEs included in the model were those listed in the SPC as most serious10. The SPC for DMF did not specify an equivalent list and, furthermore, the manufacturer’s submission to NICE marked the AE probabilities as confidential21^,31. Therefore, the AEs included for DMF were selected as serious adverse events (SAEs) that occurred in at least one patient in the licensed dose arm of any trial17^,18. AE rates were calculated from the published trials and were assumed to be constant from year 2 onwards in the model. The probabilities of withdrawal from treatment due to AEs (0.031 for fingolimod and 0.074 for DMF) were calculated from the published clinical trials.

Cost inputs

Modeled costs are given in the Supplemental Material (Table 8) and include the costs for each EDSS state, drug acquisition costs, drug administration, and monitoring costs (comprising clinician visits, clinical tests/imaging, and observation/hospitalization costs), costs of adverse events and costs associated with relapses. The model considers costs from an NHS and PSS perspective and takes a cost year of 2013–2014; costs for administration, monitoring, adverse events, and relapses were based on the NHS Reference Costs42 or NHS National Tariff43, as applicable. Costs for each EDSS state were inflated44 from those in the natalizumab manufacturer’s submission to NICE30, in line with the approach taken in another recent model appraised by NICE34. This approach, therefore, provides consistency with other MS models with regards to EDSS health state costs used. These data had been derived from the 2005 UK MS Survey, which has been reported and analyzed in different forms elsewhere6^,45. Details of which costs are included in the overall administration and monitoring costs for each drug can also be found in the Supplemental Material (Tables 9 and 10). The cost of a relapse was derived from the 2013–2014 National Tariff ‘Admitted Patient Care & Outpatient Procedure Tariff, AA30A Multiple sclerosis non-elective tariff’43; a constant cost of relapse across different severity levels is assumed due to the lack of available data on costs for different severities of relapses.

Regarding drug acquisition costs, both fingolimod and DMF are available to the NHS with a patient access scheme (PAS), but the level of discount is confidential in both cases. Therefore, list prices have been used in the base case and alternative price discounts have been investigated as scenario analyses.

Utility inputs

The model considered the utility associated with each EDSS state and with specific events. A full summary of modeled utilities is given in the Supplemental Material (Tables 11 and 12). Utility weights by EDSS score were obtained from Orme et al.46. The Orme et al. data were employed in a number of previous MS models30^,32–34, were derived from the UK population, and are presented for different EDSS scores, making them an appropriate choice as a source of utility variables. Furthermore, these utility data were collected using the EuroQol 5-Dimension (EQ-5D) Survey47 as recommended in NICE guidance on appraisal methods29.

The model applies a utility gain of 0.01 for every 5-year period with MS (or 0.002 per year). This is on the basis that the number of years since diagnosis has a positive effect on utility, which may be due to patients coming to terms with their condition46. In addition, males are expected to experience a further utility gain of 0.017 every year compared to females46.

The disutility associated with experiencing relapse was obtained from Orme et al.46, adjusting the 3-month estimates to 1-year disutility. As Orme et al. presented only the average value for a relapse, this is used for all relapses, regardless of their severity. The disutilities applied to specific AEs associated with fingolimod and DMF were obtained from the respective manufacturers’ NICE submissions, and were assumed whenever values were not available.

Model outcomes

Consistent with previous health economic models in multiple sclerosis, this analysis measures benefit in the form of quality-adjusted life years (QALYs). QALYs and costs are considered over a lifetime horizon (modeled as 50 years) and both costs and benefits are discounted at an annual discount rate of 3.5%, in accordance with the NICE reference case29.

Sensitivity analyses

To investigate uncertainty in the model, a number of approaches to sensitivity analysis were taken. Deterministic sensitivity analysis, to explore the impact of individual parameters on model results, was conducted using the upper and lower confidence intervals presented for the various inputs above; where these were not available an assumed change of ±20% was used. Probabilistic sensitivity analysis (PSA) was additionally performed by running 1000 iterations drawing randomly from the probability distributions associated with each input parameter. This allowed the calculation of the likelihood of fingolimod being cost-effective at a specified willingness-to-pay threshold. Parameters for the sensitivity analyses are given in the Supplemental Material (Table 13).

Scenario analysis investigated the effect of adjusting the model to incorporate the following assumptions: the use of equivalent treatment discontinuation rates for both interventions, the use of different assumptions around waning of efficacy over time, the exclusion of AEs from the model, and the use of natural history data which allow for spontaneous improvement in patients’ conditions. The effects of this latter assumption were assessed by using the Palace et al.48 natural history probabilities in RRMS, which allowed for decreases in the EDSS score. For this analysis, it was assumed that the active treatments did not affect the probabilities of improvement of patients’ conditions. Given the confidentiality of PAS discounts, permutations of PAS discount for DMF and fingolimod were considered to determine the thresholds which would invalidate the base case result.

Finally, further scenario analyses were performed to take account of differences in the clinical data sources between fingolimod and DMF. Fingolimod efficacy data were adjusted to account for the differing definition of progression used in the DMF trials and the effect of using 6-month confirmed disability progression data was explored. Prior treatment was also tested for interactions as a covariate in the hazard model for confirmed disability progression, in order to confirm that the mixture of prior treatments in the fingolimod trials were not introducing bias into the comparison with DMF after prior beta-interferon. Parameter inputs for these scenarios are detailed in the Supplemental Material (Table 14).

Results

Base case and probabilistic sensitivity analysis results

In comparison to DMF, fingolimod was found to be a cost-effective option with an incremental cost-effectiveness ratio (ICER) below £20,000 per QALY in the base case in both the deterministic and probabilistic models (Table 2 and Figure 2). The deterministic ICER for fingolimod vs DMF in HA RRMS under these assumptions was £12,528 and, in the probabilistic model, the ICER for fingolimod vs DMF was found to be £14,076. The probabilistic model showed that fingolimod has a 73% chance of being cost-effective at a willingness-to-pay threshold of £30,000.

Figure 2. (a) Cost-effectiveness scatter plot for fingolimod vs DMF with a willingness-to-pay line at £20,000/QALY, and (b) Cost-effectiveness acceptability curve for fingolimod vs DMF. QALY, Quality-adjusted life year; WTP, willingness-to-pay.

Table 2. Results of the deterministic and probabilistic base case cost-effectiveness analyses.

	Intervention: Fingolimod (list price)		Comparator: DMF (list price)		Incremental cost, £	Incremental QALYs	ICER, £ per QALY	Probability of fingolimod being cost-effective at a WTP threshold of
	Total Cost, £	Total QALYs	Total Cost, £	Total QALYs				£20,000	£30,000
Deterministic	356,323	4.70	346,659	3.93	9,664	0.77	12,528
Probabilistic	357,976	4.68	347,618	3.94	10,358	0.74	14,076	61%	73%

ICER, incremental cost-effectiveness ratio; QALYs, quality-adjusted life years; WTP, willingness-to-pay.

Deterministic sensitivity analysis

Deterministic sensitivity analysis found the model to be most sensitive to changes in the disease progression risk and medication cost for each treatment, in particular the disease progression risk for DMF (Figure 3).

Figure 3. Deterministic sensitivity analysis on the ICER for the fingolimod vs DMF comparison.

Scenario analyses

Results from scenario analyses investigating key model assumptions are presented in Table 3. To test the effect of the different definitions of disability progression used in the fingolimod and DMF trials, one scenario considered fingolimod 3-month confirmed disability progression data calculated in the base case sub-group, but with the definition of progression adjusted to match that used in the DMF EPAR. The results of the scenario analyses demonstrate that the base case results remain stable under all scenarios investigated. The resulting ICERs are seen to be below £20,000 per QALY in all cases.

Table 3. Results of scenario analyses.

Scenario	Fingolimod		DMF		Incremental cost, £	Incremental QALYs	ICER, £ per QALY
	Total cost, £	Total QALYs	Total cost, £	Total QALYs
Scenario: DMF probability of treatment withdrawal equal to the fingolimod probability
Equal probability of withdrawal	356,323	4.70	357,266	3.90	−943	0.80	dominant
Scenarios: considering various efficacy waning assumptions (assumptions applied to both treatments)
Base case: 75% at 2 years and 50% at 5 years	356,323	4.70	346,659	3.93	9664	0.77	12,528
75% at 5 years	354,544	4.82	346,452	3.90	8093	0.92	8842
50% at 5 years	355,310	4.77	346,556	3.91	8754	0.86	10,219
50% at 2 years and 25% at 5 years	357,989	4.59	346,868	3.96	11,121	0.63	17,536
Scenario: no adverse events included in the model
No adverse events in the model	356,214	4.70	346,577	3.93	9637	0.77	12,504
Scenario: using Palace et al.48 natural history transition probabilities when modeling RRMS
Palace natural history data	347,816	6.26	331,883	5.18	15,933	1.08	14,716
Scenario: adjusted fingolimod efficacy, definition of disability progression matched to DMF
Adjusted progression definition	356,718	4.65	346,659	3.93	10,059	0.72	13,968

ICER, incremental cost-effectiveness ratio; QALYs, quality-adjusted life years; CDP, confirmed disability progression; DMF, dimethyl fumarate.

In order to further explore the influence of the definition of disability progression, the total costs and QALYs associated with fingolimod were calculated for different definitions of confirmed disability progression (CDP); these are presented in Table 4. Disability progression confirmed at 6 months, rather than 3 months, is considered a more appropriate and robust measure of efficacy24^,49. A further stringent definition of progression is to consider confirmed disability progression sustained until the last observation. Scenarios presented in Table 4, therefore, compared the use of fingolimod 3-month and 6-month CDP inputs, and also the use or not of confirmed disability progression sustained until the last observation. In all cases, the underlying definition of progression used was adjusted to match that in the DMF EPAR.

Table 4. The effect of the definition of disability progression on total costs and QALYs for fingolimod.

	6-month CDP		3-month CDP		Incremental cost, £	Incremental QALYs	ICER, £ per QALY
	Total cost, £	Total QALYs	Total cost, £	Total QALYs
Comparing 3- and 6-month confirmed disability progression
Adjusted progression definition	354,691	4.91	356,718	4.65	−2027	0.27	Not relevant
Adjusted progression definition; sustained until last observation	353,558	5.07	353,702	5.05	−144	0.02	Not relevant
	CDP sustained until last observation		CDP
Comparing confirmed disability progression with confirmed disability progression sustained until last observation
Adjusted progression definition; 3-month CDP	353,702	5.05	356,718	4.65	−3016	0.40	Not relevant
Adjusted progression definition; 6-month CDP	353,558	5.07	354,691	4.91	−1133	0.16	Not relevant

ICER, incremental cost-effectiveness ratio; QALYs, quality-adjusted life years; CDP, confirmed disability progression.

Confidential PAS discounts

Figure 4 shows the level of DMF PAS discount needed to invalidate the base case finding that fingolimod is cost-effective, for different fingolimod PAS discounts, in steps of 10%. For a willingness-to-pay threshold of £30,000/QALY, fingolimod remains cost-effective for HA RRMS in the UK unless the differential between the percentage PAS discounts is considerable:

• At a minimum, the percentage DMF discount must be 21% or greater above the fingolimod percentage discount to negate the base case finding that fingolimod is cost-effective in this population.

• As the modeled percentage discount on fingolimod rises, so the amount of additional percentage DMF discount required to negate the base case rises, by an additional 4% difference in discount for each 10% increase in fingolimod discount.

Figure 4. The results of the scenario analysis exploring confidential PAS discounts. Percentages in this figure represent the percentage discount that would be required to be offered by the DMF PAS in order to invalidate the finding that fingolimod is cost-effective at the respective PAS for fingolimod at the two willingness-to-pay thresholds typically considered in the UK. DMF, dimethyl fumarate; PAS, patient access scheme [discount]; WTP, willingness-to-pay.

Discussion

Fingolimod is the only oral DMT recommended in England by NICE for HA RRMS20^,23^,24, having been previously demonstrated to be cost-effective in this RRMS sub-population when compared with the historical comparator of Avonex (interferon beta-1a)20. The results of this paper reaffirm the recommendation of cost-effectiveness in the HA RRMS population now that other oral DMT options have become available. The results can be considered robust, as supported by the comprehensive exploration of model assumptions in sensitivity and scenario analyses and by the result from the probabilistic model, which reported an average ICER of £14,076 and a probability of being cost-effective of 73% at a willingness-to-pay threshold of £30,000.

The limitations of the study primarily relate to the clinical input data. First, the lack of any published data for teriflunomide in the relevant patient population precluded its inclusion in the model. With respect to the modeled comparators, the data were, for both fingolimod and DMF, derived from post hoc sub-groups from Phase 3 trials, in line with the HA RRMS sub-group defined by the EMA in granting licensed indications in MS. As the trials were powered to detect significant treatment effects in the whole trial population, these sub-group results are subject to broader confidence intervals than the overall trial results, particularly in the case of DMF. Two key limitations were noted with respect to comparability between the published sub-group results for each DMT: the prior treatment agent, and the definition of progression. This paper has sought to address these limitations explicitly by exploring the use of adjusted fingolimod efficacy data in scenario analyses. Addressing these comparability limitations, we first tested the model used to calculate the hazard ratio for 3-month confirmed disability progression for fingolimod for significant interactions with prior treatment. This was to determine comparability of the fingolimod data (any prior DMT) with that from the sub-group of patients defined in the DMF EPAR (prior interferon); no significant interactions were found, indicating that this difference is unlikely to bias the results. Similarly, the scenario analysis using fingolimod confirmed disability progression data that had been adjusted to match the definition of disability progression used in the DMF trial found an ICER of £13,968. Both these results re-affirm the base case finding that fingolimod remains the cost-effective oral option in HA RRMS.

It was also noted that only 3-month confirmed disability progression results were available for DMF, whereas for fingolimod the more robust 6-month confirmed disability progression was available; a comparison of the results of the economic model for fingolimod using 6-month confirmed disability progression vs 3-month confirmed disability progression demonstrated that taking into account the more robust 6-month data, the results for fingolimod improve with lower overall costs and more QALYs gained. Due to the lack of comparator data for 6-month confirmed disability progression we were unable to perform a direct comparison with DMF using this data.

As noted above, a consequence of the post hoc nature of the input data is uncertainty, most notably reflected in the relatively wide confidence intervals, particularly on the DMF hazard ratio for confirmed disability progression. Whilst comparative randomized-controlled trials explicitly evaluating this sub-group would be ideal to address these limitations, it seems unlikely that these would be conducted. Given the importance of this and related decision problems for the health service, such issues of comparative efficacy in licensed sub-groups should be made the subject of research using observational, or so-called ‘real world evidence’ techniques.

In the UK, HTA processes review the use of treatments within their licensed indication and, thus, the decision problem presented in this paper is of direct relevance to decision-makers. The model sought to use data from the licensed indication in the EU and the post hoc nature of the data available are a result of the population definitions in the EMA licences. It is unfortunate that the post hoc sub-groups are not, by their nature, powered to detect significant differences between each DMT and placebo, but no other data are available to allow a comparison. Given that the modeled choice of comparators presents itself in the real-world health service and is faced by physicians and funders, it is preferable to use what data are available and address the uncertainty in the decision problem in a probabilistic fashion rather than fail to address the question at all.

The deterministic sensitivity analyses show clearly that the effect of uncertainty in disability progression with DMF on the deterministic ICER is considerable. Probabilistic analyses were conducted in order to incorporate this and other sources of uncertainty into the analysis. These results show that fingolimod remains cost-effective when considering the uncertainty associated with the clinical data in this sub-group. An area of further research might be to consider the use of formal techniques to calculate the value of greater certainty in clinical inputs, based on expected value of perfect information (EVPI) and expected value of partial perfect information (EVPPI) analyses.

The second main source of uncertainty is the confidential nature of the PAS discounts and the consequent uncertainty around drug acquisition costs. This has been addressed by investigating the various permutations of possible PAS discount scenarios, demonstrating that, in all cases, DMF would require a substantially larger PAS than fingolimod to invalidate the base case finding of cost-effectiveness of fingolimod. This analysis demonstrates clearly that, with the exception of extreme permutations of PAS discounts, fingolimod remains cost-effective for HA RRMS in England.

Other limitations noted for the model include the fact that both costs and utilities are derived from populations that were not the HA RRMS sub-population, and were treated with other DMTs; rather these inputs were based on the RRMS population more broadly, and this creates the potential for these input values to be biased. The influence the use of these values has on the results was evaluated for both the costs inputs (other than medication cost) and the utility inputs as part of the sensitivity analyses noted above. The results of the sensitivity analyses indicated that variation in these costs and utilities did not substantially affect the model result.

As observed in reports of similar models of RRMS23^,24, DMTs with a higher discontinuation rate are favored by the model. This is clearly shown in the reported scenario analysis in which the higher discontinuation rate for DMF was reduced to match that for fingolimod, resulting in the deterministic ICER for fingolimod dropping from £12,528 to fingolimod being dominant. Given the emerging evidence that adherence on fingolimod may be greater than on DMF50^,51, the impact of this inherent feature of cost-effectiveness models in MS on the results should be considered against the potential clinical benefits of greater adherence.

In the base case, improvement of patients’ condition (transition to a lower EDDS score) was not possible due to difficulties estimating such transitions in the small patient sub-groups available. To both explore this restriction and to test data derived from a much larger sample of patients, a scenario analysis using data published by Palace et al.48 was performed. There are several limitations of the Palace et al. dataset in the context of this model, including that it is derived from a combined RRMS and SPMS patient group: not the HA RRMS sub-group considered in this model. Although in the scenario analysis results the QALYs experienced on each treatment increased substantially, the new natural history data did not affect the comparative results to a great degree, indicating that the structural model assumption of no improvement in EDSS score applied in the base case did not result in bias in the ICER for fingolimod over DMF.

Scenario analysis found that AEs were not a significant driver of the model, as their exclusion did not result in a significant change in the ICER. This result affirms that the approach taken to modeling of AEs did not introduce bias. The waning assumptions that were explored through scenario analysis are arbitrary, but it is apparent from the scenarios presented in this paper that assuming a higher degree of waning increases the ICER of the more effective treatment. This is because the greater the efficacy the larger in absolute terms is a fixed percentage decrease in efficacy. On-going follow-up of patients from the pivotal fingolimod trials has not shown any decrease in efficacy over time52.

This study reaches a different conclusion to that published by Zhang et al.19, which considered the relative cost-effectiveness of fingolimod and DMF from a US perspective and in a wider cohort of RRMS patients rather than the HA RRMS sub-population. The differing conclusions highlight the importance of conducting cost-effectiveness analyses that are relevant to the local setting, and, in particular, the need to consider the specific sub-populations within which therapies are used in a given region. The results of this analysis are considered highly relevant to the setting of the English NHS, being based on cost inputs derived from NHS reference costs, national tariffs, and UK studies as far as possible. Generalizability to other healthcare settings outside the UK may be limited as a result of differing financing structures in other markets, although the clinical inputs for the analysis are based on global trials and, hence, are not restrictive in terms of geographical perspective.

Conclusions

Fingolimod remains cost-effective in HA RRMS following the introduction of DMF to the UK market, and this model supports the evidence that has led it to be the only oral DMT reimbursed for HA RRMS in England. This model supports the restriction imposed by NICE on DMF in HA RRMS and highlights the importance of considering different sub-groups of multiple sclerosis when performing health economic analyses.

Transparency

Declaration of funding

This study was funded by Novartis Pharmaceuticals UK Ltd, Camberley, UK.

Declaration of financial/other relationships

NA is a paid employee of Novartis Pharmaceuticals UK Ltd, Camberley, UK. NB is a paid employee of Novartis Pharma AG, Basel, Switzerland. MM, SM, and MG are paid employees of Costello Medical Consulting Ltd, Cambridge, UK, which was contracted by Novartis to undertake some of the work. JME peer reviewers on this manuscript have no relevant financial or other relationships to disclose.

Acknowledgments

The authors take full responsibility for the content of the paper. No assistance in the preparation of this article is to be declared.