3 The manufacturer's submission

The Appraisal Committee (section 9) considered evidence submitted by the manufacturer of eltrombopag and a review of this submission by the Evidence Review Group (ERG; section 10).

3.1 The manufacturer compared eltrombopag within a standard care pathway with the standard care pathway alone, and separately with romiplostim plus standard care. Standard care was defined as a pathway of care without eltrombopag or romiplostim, that is, without thrombopoietin receptor agonists (non-thrombopoietin receptor agonist pathway). It consisted of a sequence of rituximab, azathioprine, mycophenolate mofetil, ciclosporin, dapsone, danazol, cyclophosphamide, vincristine and vinblastine. The manufacturer evaluated the clinical and cost effectiveness of eltrombopag for 2 groups: patients who had had a splenectomy and patients who had not had a splenectomy.

Clinical effectiveness

3.2 The manufacturer presented clinical evidence from 3 randomised controlled trials (RCTs), TRA 100773A, TRA 100773B and RAISE, all of which were placebo-controlled, and from an extension study (EXTEND) that followed patients who had previously participated in the RCTs. The key clinical evidence was obtained from RAISE. The manufacturer also presented a meta-analysis of the results of the 3 eltrombopag RCTs (TRA 100773A, and TRA 100773B and RAISE), and 2 indirect comparisons, 1 between eltrombopag and romiplostim, and the other between eltrombopag and standard care.

3.3 RAISE was a phase IIl multicentre RCT (including 9 UK centres) that evaluated the efficacy and safety of eltrombopag plus standard care compared with placebo plus standard care in adults with a platelet count of less than 30×109 per litre. RAISE was a 6‑month study that followed patients for up to 4 weeks after treatment had been stopped, then at 3 and 6 months. Investigators randomised 197 patients to eltrombopag (n=135) or placebo (n=62), and stratified randomisation by baseline platelet counts (15×109 per litre or less, and more than 15×109 per litre), whether or not a patient had had a splenectomy, and whether or not patients were taking medication for immune thrombocytopenia (ITP) at baseline. Approximately 30% of patients had ITP that was refractory to, or had relapsed after, splenectomy. Patients randomised to either treatment group received standard care (that is, treatment with corticosteroids, non-selective immunosuppressants and rescue medication) as needed, plus either 50 mg eltrombopag or placebo, and investigators adjusted the dose of eltrombopag based on individual platelet counts. Over the 6‑month study period, the mean dose of eltrombopag was 54.7 mg per person per day. At the end of the study, 69% of patients randomised to the placebo group and 55% of those randomised to the eltrombopag group had received concomitant ITP medication.

3.4 The primary outcome in the RAISE trial was the odds of achieving a platelet count of 50–400×109 per litre at any point during the 6‑month study period. Secondary outcomes included use of rescue treatment (defined as a composite of a newly prescribed ITP medication, an increased dose of a concomitant ITP medication, a platelet transfusion or a splenectomy), incidence and severity of bleeding, and health-related quality of life.

3.5 In RAISE, the odds ratio reflecting a response during the 6‑month study period (primary outcome) was 8.2 (99% confidence interval [CI] 3.59 to 18.73; p<0.001). At the end of the study, 52% of patients receiving eltrombopag and 17% of those receiving placebo had platelet counts of 50–400×109 per litre. Once treatment was stopped, the proportions of patients with target platelet counts in the eltrombopag and placebo groups converged, reaching 20% for eltrombopag and 14% for placebo after 4 weeks. The manufacturer reported that the response to eltrombopag did not depend on whether or not the patient had had a splenectomy (p value for interaction was 0.562).

3.6 The manufacturer carried out a post hoc analysis of platelet response in RAISE, that is, an analysis of how long during the study patients maintained platelet counts of 50–400×109 per litre. The manufacturer categorised platelet response into 'sustained' platelet response, when a patient had a platelet count of 50–400×109 per litre for at least 6 of the last 8 weeks of treatment; 'transient' platelet response, when a patient had a platelet response for 4 or more consecutive weeks during the treatment period; and 'overall' platelet response, when a patient had either a sustained or a transient response. The manufacturer performed the analysis on the intention-to-treat population and on the subset of patients treated with study medication for 6 months or more (that is, including patients who continued taking eltrombopag after the study ended). In both groups, a higher proportion of patients receiving eltrombopag had 'sustained' and 'overall' platelet responses than patients receiving placebo, irrespective of whether or not they had had a splenectomy.

3.7 The manufacturer reported results for secondary outcomes in the RAISE trial. Fewer patients randomised to eltrombopag needed protocol-defined rescue treatments than those randomised to placebo (18% and 40% respectively). Among the safety population, the odds of experiencing bleeding (World Health Organization [WHO] grades 1–4) during the study period were 76% lower among patients who took at least 1 dose of eltrombopag than in those who took at least 1 dose of placebo (odds ratio [OR] 0.24; p<0.001; CI not given). At the end of the study, 57% of patients receiving placebo had experienced a grade 1–4 WHO bleed (any type of bleeding) compared with 27% of those receiving eltrombopag (OR 0.25; p>0.001; CI not given). However, a grade 2‑4 WHO bleed (clinically significant bleeding) did not differ between treatment groups (13% and 10% in the placebo and eltrombopag groups respectively). The manufacturer also performed an analysis of the risk of bleeding at least once at any point during the study, and stratified this analysis by whether or not the patient had had a splenectomy. It found that patients randomised to eltrombopag were statistically significantly less likely to have clinically significant bleeding than those randomised to placebo (33% for eltrombopag and 53% for placebo; OR 0.30; p>0.001); the results of the analysis were also statistically significantly different in favour of eltrombopag for patients who had or had not had a splenectomy.

3.8 The manufacturer reported treatment-related adverse reactions for 48 patients (36%) in the eltrombopag group and 18 patients (30%) in the placebo group. The most common adverse reactions experienced by patients receiving eltrombopag were headache (30%), diarrhoea (13%), nausea (12%), nasopharyngitis (10%), upper respiratory tract infection (10%) and fatigue (10%). The manufacturer also reported 2 thromboembolic events in the eltrombopag group and none in the placebo group. A post hoc analysis of patients treated with concomitant medication showed a reduction in corticosteroid-related adverse reactions (including dyspepsia, peripheral oedema and hyperglycaemia) in the eltrombopag group.

3.9 The RAISE trial assessed health-related quality of life at baseline, and at 6, 14 and 26 weeks using the SF‑36 instrument, which consists of 8 subdomains and 2 component summary scores (representing physical and mental health). In addition, investigators used subscales of the Functional Assessment of Chronic Illness Therapy for Patients with Thrombocytopenia (FACIT‑Th) and Functional Assessment of Chronic Illness Therapy (FACIT) instruments. The manufacturer reported that patients receiving eltrombopag improved more from baseline to week 26 across most of the SF‑36 domains for health and wellbeing than those receiving placebo. There were statistically significant differences between treatment groups in the change from baseline in the component summaries for physical role, vitality, emotional role and mental health.

3.10 The manufacturer did a meta-analysis of TRA 100773A, TRA 100773B and RAISE to establish whether treatment with eltrombopag improved platelet counts compared with placebo. It reported the odds ratios for attaining a platelet count of 50×109 per litre or more 6 weeks after the beginning of the study. In this analysis, eltrombopag was associated with higher odds of responding to treatment compared with placebo, with an odds ratio from a fixed effects model of 8.23 (95% CI 4.68 to 14.48) and an odds ratio from a random effects model of 8.16 (95% CI 4.63 to 14.37); there was little evidence of statistical heterogeneity.

3.11 Because there were no head-to-head trials comparing eltrombopag with romiplostim, the manufacturer performed an indirect comparison between the 2 treatments. A systematic review by the manufacturer identified 2 RCTs comparing romiplostim with placebo (both reported in Kuter et al. 2008), which the manufacturer used to compare eltrombopag with romiplostim for efficacy and rates of clinically significant bleeding. Both RCTs evaluated the safety and efficacy of romiplostim in patients with ITP; 1 enrolled 63 patients who had had a splenectomy, and the other enrolled 62 patients who had not. In both studies, patients had platelet counts of 30×109 per litre or less and ITP that was refractory to at least 1 previous treatment. Patients were randomised to either romiplostim plus standard care, or standard care alone, and they received treatment for 6 months. The primary outcome in both studies was the proportion of patients with a durable platelet response (defined as a platelet count of 50×109 per litre or more in 6 or more weekly assessments in the last 8 weeks of treatment), and who did not need rescue medication. The manufacturer combined the results of the 2 studies using standard meta-analytic techniques and then treated them as a single trial to do the indirect comparison.

3.12 The manufacturer used the Bucher method in its indirect comparison between eltrombopag (data from RAISE) and romiplostim (data from the 2 Kuter et al. 2008 trials), using placebo as a common comparator. It performed the comparison for the whole population, and separately for patients who had or had not had a splenectomy. The manufacturer considered 2 main outcome measures: platelet response and clinically significant bleeding. The end points for platelet response differed between the eltrombopag and romiplostim trials. In Kuter et al., the primary outcome was the proportion of patients with platelet counts of 50×109 per litre or more in 6 or more weekly assessments during the last 8 weeks of treatment without using rescue medication (durable platelet response), which the manufacturer equated to 'sustained response' as defined in the post hoc analyses of RAISE (section 3.6). The manufacturer further defined an 'overall response' as having either a durable response or a transient response. There were also differences in the definitions of bleeding between the eltrombopag and romiplostim trials: in RAISE, data on bleeding were collected using the WHO bleeding scale and the Common Terminology Criteria for Adverse Events (CTCAE) scale, whereas in Kuter et al., they were collected using an unnamed scale.

3.13 The manufacturer performed separate analyses for durable response and overall response. The results of the indirect comparison were framed so that odds ratios of more than 1.00 favoured eltrombopag. When eltrombopag was compared with romiplostim, the odds ratio for attaining a durable response was 0.32 (95% CI 0.03 to 3.14) and that for attaining an overall response was 0.22 (95% CI 0.05 to 1.02). For people who had had a splenectomy, the odds ratios were 0.50 (95% CI 0.01 to 17.3) for durable response and 0.09 (95% CI 0.00 to 2.52) for overall response; for people who had not had a splenectomy, the odds ratios were 0.41 (95% CI 0.04 to 4.80) and 0.34 (95% CI 0.06 to 2.14) for durable response and overall response respectively.

3.14 The indirect comparison of rates of bleeding showed that the point estimates favoured eltrombopag in some analyses and romiplostim in others, with no statistically significant differences between the 2 treatments. When eltrombopag was compared with romiplostim, the odds ratio of a clinically significant bleed was 0.60 (95% CI 0.08 to 4.29), and that of a moderate or clinically significant bleed was 1.63 (95% CI 0.4.6 to 5.80).

3.15 The manufacturer highlighted that the indirect comparison showed no statistically significant differences between eltrombopag and romiplostim, and suggested that the differences between individual studies should be acknowledged when interpreting the results. The manufacturer indicated that patients differed between RAISE and the 2 Kuter et al. (2008) trials in terms of duration of ITP, previous use of ITP medications, use of concomitant medication, and whether or not patients had had a splenectomy. It also indicated that the design of the trials was different for timing of platelet count assessments, timeframes in which patients were allowed to reduce concomitant ITP medications, definitions of response and definitions of 'period of rescue medication'. The manufacturer pointed out that 2 published clinical guidelines, the 'International consensus report on the investigation and management of primary immune thrombocytopenia' (Provan et al. 2010) and 'The American Society of Haematology 2011 evidence-based practice guideline for immune thrombocytopenia' (Neunert et al. 2011), do not favour 1 treatment over the other. The manufacturer concluded that its indirect comparison between eltrombopag and romiplostim did not provide evidence of clinical superiority for 1 treatment over the other. In absence of evidence to the contrary, the manufacturer concluded that eltrombopag and romiplostim have 'equal efficacy' and applied this assumption to the cost-effectiveness analysis.

3.16 The manufacturer presented an indirect comparison between eltrombopag and standard care alone (excluding eltrombopag and romiplostim). In this, the manufacturer restricted the treatments used in standard care to those included in the international consensus report (that is, intravenous immunoglobulin G, anti‑D, rituximab, corticosteroids, vinca alkaloids, mycophenolate mofetil, ciclosporin, cyclophosphamide, danazol and dapsone). The manufacturer's systematic review of treatments used in standard care identified 113 studies (including 20 RCTs). However, the manufacturer altered its inclusion criteria after performing the search, which resulted in the exclusion of most of the identified studies. The manufacturer combined results from 37 studies, including 6 RCTs, to calculate weighted averages of response rate, time to response and duration of response for each drug used within the standard care pathway. The manufacturer pooled data regardless of the definition of response, and calculated the efficacy of each intervention using a simple average. The manufacturer highlighted that the results of the weighted averages for each of the included treatments were obtained mainly from non-randomised, highly heterogeneous, older trials; however, it acknowledged that the results largely reflected the response rates outlined in the international consensus report (Provan et al. 2010) and in Romiplostim for the treatment of chronic immune thrombocytopenia (NICE technology appraisal guidance 221).

Cost effectiveness

3.17 The manufacturer developed a de novo economic model to assess the cost effectiveness of eltrombopag in 2 populations of chronic ITP:

  • adults who have not had a splenectomy

  • adults who have had a splenectomy, but whose condition is refractory to previous treatments.

    The manufacturer assumed that patients who have not had a splenectomy reflect those for whom splenectomy is contraindicated.

3.18 The cost-effectiveness model developed by the manufacturer is a state-transition Markov cohort model with a 4‑week cycle length. The model simulates patients with chronic ITP receiving eltrombopag plus standard care, romiplostim plus standard care, or standard care alone. The manufacturer assumed that all patients entering the model have ITP that is refractory to first-line treatment with corticosteroids or immunoglobulins and, if rituximab is considered an appropriate treatment option, patients will have already received it. For patients starting a treatment, the model permits their platelet count to reach 50×109 per litre or more (equal to a response) in the first, second, third or fourth cycle, depending on the time to response associated with each treatment. When the platelet count reaches 50×109 per litre, patients have a treatment-specific probability of losing the response in each cycle, and of receiving rescue therapy when bleeding occurs or a patient is deemed at high risk of bleeding. If the platelet count does not reach 50×109 per litre or patients lose their response, they stop treatment but may receive rescue therapy (intravenous immunoglobulin, anti‑D and corticosteroids), which may result in a temporary platelet response lasting for 1 cycle. During each cycle, a proportion of patients who experience a bleed or whose platelet count does not respond 'exit' the 'non-responder' state and move on to other treatments further down the treatment sequence. Rates of rescue treatment, rates of non-severe bleeds treated in the outpatient setting, and rates of severe bleeds treated in the inpatient setting were lower in patients whose condition responds than in those whose condition does not. Patients in the model who are less likely to bleed are less likely to die.

3.19 The economic evaluation compared 3 treatment sequences: a pathway reflecting standard care without a thrombopoietin agonist (sequence 'a': azathioprine, mycophenolate mofetil, ciclosporin, danazol, dapsone, cyclophosphamide, vinblastine and vincristine), a pathway of eltrombopag with standard care (eltrombopag followed by sequence 'a'), and a pathway of romiplostim with standard care (romiplostim followed by sequence 'a'). The sequence of treatments used as standard care reflects that used by the manufacturer of romiplostim in NICE technology appraisal guidance 221, except that rituximab is removed from the sequence for the base-case analysis in the current submission. This is because, as the manufacturer explained, UK local guidance suggests that clinicians offer rituximab to patients before eltrombopag or romiplostim. The manufacturer discounted costs and benefits at an annual 3.5% rate.

3.20 The manufacturer submitted 3 separate economic evaluations: a base case, an 'alternative' evaluation and a scenario analysis. In the base case, the manufacturer applied a set of assumptions it deemed most relevant to the decision problem, using NICE technology appraisal guidance 221 as its main source of data and assumptions. The only parameters in the base-case model that the manufacturer sourced from the RAISE and EXTEND trials were the thrombopoietin receptor agonist response rates and the thrombopoietin receptor agonist time on treatment. The alternative evaluation applied data from RAISE and EXTEND, along with clinical evidence retrieved from the manufacturer's systematic review for this appraisal. In the scenario analysis, the manufacturer applied all the assumptions and model inputs used in the economic evaluation for romiplostim in NICE technology appraisal guidance 221 (including those that were not used in the base case) to try to replicate as closely as possible the analysis in that technology appraisal.

3.21 In the base case and alternative evaluation, the manufacturer assumed that the response rate (attaining a platelet count of 50–400×109 per litre at any time during the 6‑month study period) for eltrombopag was the same as that observed in the RAISE trial. It also assumed that, if a patient had a platelet response at any time during the 6‑month period, the patient maintained the platelet response while on treatment and had a probability of bleeding and death as if the platelet count had remained elevated. Both the base case and alternative evaluation assumed complete clinical equivalence between eltrombopag and romiplostim, and so a patient in the model taking eltrombopag had the same rate of platelet response as a patient taking romiplostim. The manufacturer assumed that the effectiveness of the 2 treatments was the same because its indirect comparison had not shown that the treatments were different (section 3.15). However, the manufacturer performed sensitivity analyses to test the possibility that romiplostim was more effective than eltrombopag by applying the odds ratio for overall response from its indirect comparison between eltrombopag and romiplostim (0.22).

3.22 For treatments considered to be standard care, the manufacturer took response rates for the base case from NICE technology appraisal guidance 221, in which the response rates were calculated from a systematic review that the manufacturer of romiplostim had done. In the alternative evaluation, the manufacturer estimated a response rate for each treatment from its indirect comparison between eltrombopag and treatments comprising standard care (section 3.16).

3.23 In the base case and alternative evaluation, time to platelet response for eltrombopag was 15 days (standard error 3.75 days), as observed in RAISE. For romiplostim, the time to response was assumed to be 28 days (standard error 7 days), based on the Kuter et al. (2008) trials. For treatments comprising standard care, time to response from NICE technology appraisal guidance 221 was used in the base case and, for the alternative evaluation, it was obtained from the manufacturer's indirect comparison between eltrombopag and standard care (section 3.16).

3.24 Because the manufacturer assumed that eltrombopag and romiplostim were equally effective, it also assumed that time on treatment was the same for eltrombopag and romiplostim. To extrapolate time on treatment over a lifetime horizon, the manufacturer modelled time on treatment as a survival variable using patient-level data on treatment discontinuation from RAISE and EXTEND, and carried out a parametric analysis. The manufacturer found that, among patients whose condition responded to treatment, those who had had a splenectomy spent less time on eltrombopag than those who had not.

3.25 For the time on treatment for therapies included in standard care, the manufacturer took values from NICE technology appraisal guidance 221 for its base case, and from the indirect comparison between eltrombopag and standard care (section 3.16) for its alternative evaluation. The manufacturer assumed that time on treatment for standard therapy followed an exponential distribution.

3.26 The manufacturer assumed that the risk of bleeding in the model is a function of platelet response irrespective of treatment, so patients with platelet counts of 50×109 per litre or more were at risk of non-severe bleeds (treated as an outpatient), and patients with platelet counts of less than 50×109 per litre had a risk of severe (needing inpatient care) or non-severe bleeds. For its base case, the manufacturer applied the rates of bleeding previously used in NICE technology appraisal guidance 221 for romiplostim. For patients who did not have a platelet response, the rate of severe bleeds applied in the base-case model was 4.3% per month. For its alternative evaluation, the manufacturer used the rate of severe bleeds from RAISE and EXTEND (0.8% per month). The manufacturer assumed that patients whose condition is refractory to all previous treatments are twice as likely to bleed as patients whose condition does not respond to treatment but who are between treatments. The manufacturer took this assumption from NICE technology appraisal guidance 221.

3.27 The manufacturer modelled mortality from chronic ITP as a function of severe bleeds in the base case and alternative evaluation. For each bleed for which a patient needed to be hospitalised, the manufacturer applied a mortality rate from Danese et al. (2009), and assumed that this rate doubles for patients whose condition is refractory to all previous treatments. The manufacturer considered that patients would need to be hospitalised for the following categories of bleeds: gastrointestinal haemorrhage, intracranial haemorrhage and haemorrhage resulting from a 'coagulation disorder'.

3.28 For the base case, the modelling assumed that only patients with platelet counts of less than 50×109 per litre receive rescue medication, the types and rates of which were used in NICE technology appraisal guidance 221. The rate of rescue therapy for patients who had had a splenectomy was 68% and, for patients who had not had splenectomy, it was 33%. Rescue medications included intravenous immunoglobulin, anti‑D and corticosteroids, and the proportions in which patients received these medications in the model were based on a survey of 169 UK haematologists that the manufacturer of romiplostim did for NICE technology appraisal guidance 221. To estimate the rate of rescue for patients with platelet counts above and below 50×109 per litre for its alternative evaluation, the manufacturer used data from RAISE and EXTEND limited to countries with healthcare resources comparable to the UK.

3.29 Adverse events in the model were considered as either severe or 'other'. In the base case and alternative evaluation, the manufacturer assumed that the rates of adverse events for eltrombopag and romiplostim were equivalent and used the rates from NICE technology appraisal guidance 221. The manufacturer estimated adverse event rates for treatments included in standard care from the same technology appraisal.

3.30 Although RAISE and EXTEND collected health-related quality-of-life data, the manufacturer chose to use utility data for the base case and alternative evaluation from a study it had identified (Szende et al. 2010). This study developed 6 ITP-related health states that investigators had evaluated using the time trade-off method in 359 members of the UK general public.

3.31 The manufacturer did not identify any resource-use studies relevant to the UK from its systematic review of the literature. Therefore, it used unpublished data to estimate costs including the costs of acquisition and administration of the intervention and comparators, and the costs of the rescue medication, as well as the costs of monitoring. The manufacturer took the list prices of the different drugs from the 'British national formulary' (BNF) edition 63 and applied the patient access schemes for eltrombopag and romiplostim. It calculated the average doses of eltrombopag from RAISE and, after the 6‑month study period, it estimated a stable dose from the EXTEND study. For romiplostim, the manufacturer calculated the average doses from Kuter et al. (2008) and assumed that the dose on which a patient is likely to remain (the stable dose) equals the last dose used in the trials (last dose carried forward). Dosages of drugs other than romiplostim and eltrombopag were taken from Provan et al. (2010), the international consensus report, or NICE technology appraisal guidance 221. Eltrombopag and other oral treatments did not have administration costs. Because romiplostim is injected subcutaneously, it can be administered at home or in hospital; the manufacturer assumed that costs were incurred only when the drug was administered in hospital. The cost of bleeds covered drug costs, hospitalisation and follow-up. The manufacturer assumed that all patients, regardless of treatment, needed monitoring by a haematologist and 2 laboratory tests every 4 weeks.

3.32 In the manufacturer's base-case analysis, eltrombopag dominated romiplostim (that is, was more effective and less costly) for patients who had or had not had a splenectomy. For the comparison of eltrombopag with standard care, eltrombopag dominated standard care for patients who had had a splenectomy, and its incremental cost-effectiveness ratio (ICER) for patients who had not had a splenectomy was £15,105 per quality-adjusted life year (QALY) gained.

3.33 The manufacturer carried out a wide range of sensitivity analyses on the base case, varying 1 parameter at a time. It did not perform one‑way sensitivity analyses on the results of the alternative evaluation or the scenario analysis.

  • For patients who had had a splenectomy, eltrombopag dominated the standard care pathway in all analyses explored. In comparison with romiplostim, eltrombopag dominated in all analyses except when the model incorporated the odds ratio for overall response between eltrombopag and romiplostim from the manufacturer's indirect comparison (0.22, section 3.13). In this scenario, romiplostim gave 0.56 additional QALYs compared with eltrombopag, but at an additional cost of £95,649; the resulting ICER for eltrombopag compared with romiplostim was £171,156 saved per QALY lost (that is, eltrombopag was less effective but also less expensive than romiplostim).

  • For patients who had not had a splenectomy, the ICER for eltrombopag compared with standard care remained below £33,000 per QALY gained in all scenarios except when a 6‑month time horizon was used, in which case the ICER for eltrombopag compared with standard care was £74,250 per QALY gained. For the comparison of eltrombopag with romiplostim, eltrombopag dominated romiplostim in all sensitivity analyses, except when the odds ratio for overall response from the indirect comparison was used to estimate the relative efficacy of eltrombopag and romiplostim (OR 0.22, section 3.13). In this scenario, romiplostim offered 0.46 additional QALYs compared with eltrombopag, but at an additional cost of £51,416. This gave an ICER for eltrombopag compared with romiplostim of £110,983 saved per QALY lost.

3.34 The manufacturer carried out probabilistic sensitivity analyses to summarise the uncertainty in the base-case ICER. This showed that, for patients who had had a splenectomy, there was a 65% probability of eltrombopag being cost effective if the maximum acceptable ICER was £20,000 per QALY gained, and a 70% probability of it being cost effective if the maximum acceptable ICER was £30,000 per QALY gained. For patients who had not had a splenectomy, there was a 54% probability of eltrombopag being cost effective if the maximum acceptable ICER was £20,000 per QALY gained, and a 63% probability of it being cost effective if the maximum acceptable ICER was £30,000 per QALY gained.

3.35 In the manufacturer's alternative evaluation, eltrombopag dominated romiplostim in the analyses for patients who had or had not had a splenectomy. When eltrombopag was compared with standard care, the ICER for eltrombopag was £61,337 per QALY gained for patients who had had a splenectomy and £95,536 per QALY gained for patients who had not had a splenectomy.

3.36 The manufacturer presented a scenario analysis to replicate the analysis for NICE technology appraisal guidance 221. In this, the manufacturer:

  • assumed that eltrombopag and romiplostim are administered before rituximab in the treatment pathway

  • assumed that time on treatment followed an exponential distribution (instead of a log‑normal distribution for the base case)

  • remodelled the response rates for eltrombopag and romiplostim to exclude patients whose condition responded to unlicensed doses

  • calibrated rescue rates to produce rates when the treatment pathway is set to exclude maintenance treatments

  • based utility values on pooled EQ‑5D and vignette utility data as per NICE technology appraisal guidance 221

  • estimated the number of vials of romiplostim needed from NICE technology appraisal guidance 221

  • set administration costs to £262 per cycle for all treatments and assumed that romiplostim did not incur further costs of administration.

3.37 In the scenario analysis, eltrombopag dominated both romiplostim and standard care for all patients.

Evidence Review Group critique and exploratory analyses

3.38 The ERG stated that the manufacturer identified all relevant studies comparing eltrombopag with placebo and presented a suitable meta-analysis. It also considered that the literature review carried out by the manufacturer to estimate the efficacy of standard care was reasonable.

3.39 For the indirect comparison of eltrombopag with romiplostim, the manufacturer used a Mantel‑Haenszel fixed-effect approach to combine the results of the 2 Kuter et al. (2008) trials and then used the Bucher method. The ERG expressed the following concerns about this methodology:

  • Heterogeneity exists between the 2 Kuter et al. trials, and pooling their results may have introduced bias.

  • Although differences exist between RAISE and the 2 Kuter et al. trials (section 3.15), the ERG felt that it was reasonable that the manufacturer had proceeded with the indirect comparison, but advised caution with respect to the results.

  • Because the manufacturer had presented the indirect comparison stratified by splenectomy status, the analyses did not preserve randomisation in RAISE, and the ERG considered them to be observational analyses.

3.40 The ERG performed an exploratory indirect comparison between eltrombopag and romiplostim for the outcomes of durable and overall response, and for clinically significant and moderate bleeds using a Bayesian network meta-analysis to account for the heterogeneity between the 2 Kuter et al. (2008) studies. For durable response and bleeding, the ERG found similar results to those of the manufacturer. For overall response, the manufacturer had found no statistically significant difference between treatments (OR 0.22; 95% CI 0.05 to 1.02), but the ERG found a statistically significant difference in favour of romiplostim (OR 0.15; 95% credible interval 0.02 to 0.84).

3.41 For the indirect comparison of eltrombopag with standard care, the ERG expressed concerns about the methodological rigor of the manufacturer's approach. Because the manufacturer excluded studies from the systematic review after the review had been performed, and had pooled response estimates using a simple weighted average regardless of the definition of response, the ERG considered that bias may exist. The ERG recommended caution when considering the results of this indirect comparison.

3.42 The ERG noted that a major weakness in the base-case analysis was that the manufacturer chose not to use data from the eltrombopag RCTs or from its systematic review of the literature, and instead opted to populate the base-case model with estimates from NICE technology appraisal guidance 221 for romiplostim. Because of this, the ERG considered the alternative evaluation to be more appropriate.

3.43 The ERG had concerns about the manufacturer's assumption that eltrombopag and romiplostim are equally effective, given the uncertainty around the results of the indirect comparison between eltrombopag and romiplostim (section 3.15).

3.44 The ERG noted that the manufacturer did not address the optimal positioning of eltrombopag and romiplostim within the treatment sequence in the model. The manufacturer assumed that eltrombopag and romiplostim followed after rituximab, but preceded other drugs used in standard care. In addition, the ERG pointed out that there is uncertainty about the optimal place of eltrombopag and romiplostim if one is assumed to be more effective than the other. The ERG stated that the manufacturer should have explored additional sequences of treatment.

3.45 The ERG had concerns about the manufacturer's assumption that 'response' and 'platelet response' are the same. The ERG noted that, in RAISE, only 60–80% of patients whose condition responded to eltrombopag had a sustained platelet response of more than 50×109 per litre. Because platelet counts drive bleeding rates and mortality in the model, the ERG stated that the manufacturer's assumption would improve the ICERs for eltrombopag and romiplostim.

3.46 The manufacturer averaged eltrombopag and romiplostim doses from the relevant trials across patients whose condition had responded and those whose condition had not. The ERG noted that, in the Kuter et al. (2008) trials, the median dose of romiplostim in patients whose condition had responded was 40–60% lower than that across the trial as a whole. The ERG stated that eltrombopag and romiplostim doses should be response-specific.

3.47 To model utility, the ERG considered that the manufacturer, in its cost-effectiveness analysis, should have used the SF‑6D health-related quality-of-life data collected from the RAISE and EXTEND trials, which are derived from a validated generic instrument.

3.48 The ERG questioned the manufacturer's assumption that the rate of severe bleeding doubles for patients whose ITP is refractory to all previous treatments, noting that these rates were high.

3.49 The ERG undertook exploratory sensitivity analyses, varying 1 parameter at a time, on both the base case and alternative evaluation; these included the following:

  • Applying the overall response rates from the manufacturer's indirect comparison (60% for eltrombopag and 94% for romiplostim for people who had had a splenectomy [OR 0.09], and 72% for eltrombopag and 88% for romiplostim for people who had not had a splenectomy [OR 0.34]).

    • In the comparison of eltrombopag with romiplostim, eltrombopag was associated with both fewer QALYs and lower costs than romiplostim. For the base-case analysis, the ICERs suggested savings of £174,503 per QALY lost for people who had had a splenectomy when using eltrombopag instead of romiplostim. The ERG did not explicitly report ICERs for people who had not had a splenectomy from its analyses on the base case, nor did it report the ICERs for any of the subpopulations from its analyses on the alternative evaluation.

    • In the comparison of eltrombopag with standard care, the ERG reported costs and QALYs for the base-case analysis only for people who had not had a splenectomy, and for the alternative evaluation both for people who had or had not had a splenectomy. In the base-case analysis, eltrombopag dominated standard care for people who had had a splenectomy. For those who had not had a splenectomy, the ICER for eltrombopag compared with standard care was £15,843 per QALY gained. In the alternative evaluation, the ICERs for eltrombopag compared with standard care were £73,335 and £108,336 per QALY gained for people who had and had not had a splenectomy respectively.

  • Applying the SF‑6D utility data collected from RAISE and EXTEND.

    • In the comparison of eltrombopag with romiplostim, the ERG found that eltrombopag dominated romiplostim for both the base case and alternative evaluation, irrespective of whether or not the person had had a splenectomy.

    • For the comparison of eltrombopag with standard care in the base-case analysis, eltrombopag was dominant for people who had had a splenectomy, and gave an ICER of £18,489 per QALY gained for people who had not had a splenectomy. When the ERG applied the utility values to the alternative evaluation, eltrombopag was associated with ICERs of £90,753 and £133,508 per QALY gained for people who had and had not had a splenectomy respectively.

  • Reducing modelled doses of romiplostim by 40% for people who had had a splenectomy and 60% for people who had not had a splenectomy.

    • In the comparison of eltrombopag with romiplostim, the ERG found that, despite the lower cost of romiplostim, eltrombopag dominated romiplostim in both the base case and alternative evaluation, irrespective of whether or not people had had a splenectomy.

3.50 In response to comments received during consultation on the first appraisal consultation document, the ERG carried out additional exploratory sensitivity analyses on the alternative evaluation, varying 1 parameter at a time, and then varying multiple parameters simultaneously. The ERG stated that, in its opinion, the alternative parameter inputs used in these analyses did not necessarily reflect the most reasonable assumptions. For the following parameters, the ERG:

  • a. applied the odds ratio of 0.22 for overall response from the manufacturer's indirect comparison between eltrombopag and romiplostim (section 3.13). The resulting overall response rates were 60% for eltrombopag and 87% for romiplostim for patients who had had a splenectomy, and 72% for eltrombopag and 92% for romiplostim for patients who had not had a splenectomy

  • b. applied the SF‑6D utility data collected from RAISE

  • c. removed anti‑D treatment from the rescue therapies for patients who had not had a splenectomy

  • d. applied the odds ratio of 0.15 for overall response from the ERG's indirect comparison between eltrombopag and romiplostim, for which the ERG had used a Bayesian approach (section 3.40). The resulting overall response rates were 60% for eltrombopag and 91% for romiplostim for patients who had had a splenectomy, and 72% for eltrombopag and 94% for romiplostim for patients who had not had a splenectomy

  • e. applied a dose of romiplostim of 1.54 vials for patients who had had a splenectomy and 1.10 vials for patients who had not had a splenectomy, as calculated by the manufacturer of romiplostim

  • f. applied a cost per administration of romiplostim equal to £11.50, as suggested by the manufacturer of romiplostim

  • g. applied the above-listed sensitivity analyses b and c simultaneously

  • h. applied the above-listed sensitivity analyses b, c and d simultaneously

  • i. applied the above-listed sensitivity analyses b, c, d and e simultaneously

  • j. applied the above-listed sensitivity analyses b, c, d, e and f simultaneously.

    The ERG found that, when varying 1 parameter at a time, eltrombopag dominated romiplostim (that is, gave the same QALYs as romiplostim but at a lower cost) for patients who had or had not had a splenectomy in all analyses, except when the odds ratio of 0.22 or 0.15 was applied for overall response. In these instances, eltrombopag was associated with fewer QALYs and lower costs compared with romiplostim; when the ERG applied the odds ratio of 0.22, the corresponding ICERs suggested savings of £689,084 and £372,782 per QALY lost for patients who had and had not had a splenectomy respectively; when the odds ratio of 0.15 was applied, the ICERs were savings of £638,042 and £350,685 per QALY lost for patients who had and had not had a splenectomy respectively. The ERG estimated from the analyses in which it varied multiple parameters simultaneously that eltrombopag compared with romiplostim was associated with savings per QALY lost greater than £250,000 in all analyses, irrespective of whether or not the patient had had a splenectomy. The ICER from the sensitivity analysis in which all parameters were varied simultaneously (sensitivity analysis j) was £388,799 saved per QALY lost for patients who had had a splenectomy and £270,694 saved per QALY lost for patients who had not had a splenectomy.

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

  • National Institute for Health and Care Excellence (NICE)