3 The manufacturer's submission
The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of romiplostim and a review of this submission by the Evidence Review Group (ERG; appendix B).
Clinical effectiveness
3.1 The manufacturer's submission compared romiplostim in addition to standard care with standard care alone for patients with ITP who had undergone splenectomy and, separately, for patients with ITP who had not undergone splenectomy. Evidence was obtained from two double-blind placebo-controlled randomised controlled trials (RCTs) of romiplostim in addition to standard care (defined as prednisone, azathioprine and danazol) compared with standard care alone. One RCT enrolled patients with ITP who had undergone splenectomy (42 patients were assigned to romiplostim and 21 to placebo). The other RCT enrolled patients with ITP who had not undergone splenectomy, but who did not necessarily have a contraindication to splenectomy (41 patients were assigned to romiplostim and 21 to placebo).
3.2 In both RCTs, patients with ITP (defined as the mean of three platelet counts being below or equal to 30 × 109 per litre, with none of the three counts being above 35 × 109 per litre) whose condition was refractory to at least one previous treatment were randomised to romiplostim plus standard care or to standard care alone (placebo) for 24 weeks. The mean platelet count at baseline in the trials was 18 × 109 per litre in the non-splenectomised group and 15 × 109 per litre in the splenectomised group. Romiplostim was given weekly. Platelet counts were monitored and the dose of romiplostim was adjusted to achieve and maintain a platelet count of between 50 × 109 and 200 × 109 per litre. Investigators could increase the dose by 2 micrograms/kg each week (for platelet counts of 10 × 109 per litre or below) or every 2 weeks (after 2 consecutive weeks of platelet counts between 10 × 109 and 50 × 109 per litre). The dose remained stable for platelet counts above 50 × 109 per litre. If a patient's platelet count reached 50 × 109 per litre or more and subsequently fell, the maintenance dose could be increased by 1 microgram/kg each week (if the platelet count fell to 10 × 109 per litre or below) or by 1 microgram/kg every 2 weeks (if the platelet count fell to between 10 × 109 and 50 × 109 per litre for 2 consecutive weeks). The maximum dose permitted in the trial was 15 micrograms/kg, which exceeds the maximum dose of 10 micrograms/kg recommended in the SPC.
3.3 No studies were found that compared romiplostim with a specified sequence of active treatments or rescue therapies for the treatment of ITP. Concurrent active treatments or rescue therapies for ITP in trials were given at the investigators' discretion. Six non-RCTs investigating the safety of romiplostim and one open-label extension study of the phase III RCTs were reported in the manufacturer's submission. In the latter, patients treated with romiplostim or placebo who had completed the phase III study and other clinical studies, and whose platelet counts had fallen below 50 x 109 per litre after stopping romiplostim or placebo, were eligible to enrol in the study and to receive open-label romiplostim. Data from patients going into this extension study were used to calculate time to failure for romiplostim, as this could not be calculated from the phase III studies alone because the interventions ended after 24 weeks.
3.4 The primary endpoint for the two RCTs was the incidence of a durable platelet response, defined as a platelet count of at least 50 × 109 per litre in at least six weekly assessments in the last 8 weeks of treatment without the use of rescue therapies. Rescue therapies included corticosteroids (oral or intravenous), intravenous immunoglobulin (IVIg) and intravenous anti-D immunoglobulin. Other outcomes were: incidence of an overall platelet response (either a durable or a transient platelet response, where a transient response is defined as a platelet count of at least 50 × 109 per litre in at least four weekly assessments during weeks 2 to 25 of treatment, with no weekly response eligible within 8 weeks of the use of rescue therapies, and in the absence of a durable response); time to platelet response (Kaplan–Meier estimated time to first platelet response); duration of platelet response; use of rescue therapies; mortality; adverse events; and health-related quality of life. The primary and secondary outcomes were analysed prospectively. Time to failure (that is, time to stopping romiplostim) and bleeding events were analysed retrospectively.
3.5 In the RCT of splenectomised patients, 16 of 42 patients (38.1%) in the romiplostim group and none of 21 patients in the placebo group had a durable platelet response. Thirty-three patients (78.6%) in the romiplostim group had an overall platelet response. The median time to the first platelet response was 3 weeks. No patients in the placebo group had an overall platelet response. The mean number of weeks with a platelet count equal to or above 50 x 109 per litre (in a study period of 24 weeks) was 12.3 weeks for the romiplostim group and 0.2 weeks for the placebo group (p < 0.0001). Eleven patients (26.2%) in the romiplostim group and twelve patients (57.1%) in the placebo group received rescue therapies during the treatment period (odds ratio 0.3; 95% confidence interval [CI] 0.1 to 0.8; p = 0.02). There were no deaths in the romiplostim group. Three patients in the placebo group died; the causes of death were pneumonia (after the end of the study), pulmonary embolism and cerebral haemorrhage.
3.6 The manufacturer's submission stated that in the RCT of non-splenectomised patients, 25 of 41 patients (61.0%) in the romiplostim group and one of 21 patients (4.8%) in the placebo group had a durable platelet response (odds ratio 24.5; 95% CI 3.3 to 179.2; p < 0.0001). Thirty-six patients (87.8%) in the romiplostim group and three patients (14.3%) in the placebo group had an overall platelet response (odds ratio 34.7; 95% CI 7.8 to 155.4; p < 0.0001). The median time to the first platelet response was 2 weeks. The mean number of weeks with a platelet count equal to or above 50 x 109 per litre (in a study period of 24 weeks) was 15.2 weeks for the romiplostim group and 1.3 weeks for the placebo group (p < 0.0001). Seven patients (17.1%) in the romiplostim group and 13 patients (61.9%) in the placebo group received rescue therapies during the treatment period. One patient in the romiplostim group and no patients in the placebo group died. The cause of death was intracranial haemorrhage 2 weeks after stopping romiplostim. All patients included in both RCTs received at least one dose of either romiplostim or placebo. One non-splenectomised patient randomly assigned to placebo received three doses of romiplostim in error and was included in the safety analysis as a patient given romiplostim and in the efficacy analysis as a patient randomised to placebo.
3.7 The manufacturer's submission reported results for bleeding events, adverse events and health-related quality-of-life outcomes pooled from the two RCTs. These showed that 45 of 84 patients (54%) in the combined romiplostim groups (a revised figure of 48 of 84 [57%] was given in the Evidence Review Group [ERG] report using data provided by the manufacturer following a request for clarification) and 25 of 41 patients (61%) in the combined placebo groups experienced at least one bleeding event of any severity. A serious bleeding event, as defined by the regulatory protocol (which includes, but may not be limited to, any event that: is fatal, is life threatening [puts the person at immediate risk of death], needs in-patient hospitalisation or prolongation of existing hospitalisation, or is a persistent or significant disability or incapacity), was reported for 5 of 84 patients (6%) in the combined romiplostim groups and 4 of 41 patients (10%) in the combined placebo groups. Bleeding of grade 3 or above (rated as severe, life threatening or fatal) occurred in 6 of 84 patients (7%) in the combined romiplostim groups and 5 of 41 patients (12%) in the combined placebo groups. Bleeding events of grade 2 or above (rated as moderate, severe, life threatening or fatal) occurred in 13 of 84 patients (15%) in the combined romiplostim groups and 14 of 41 patients (34%) in the combined placebo groups.
3.8 Data on safety were derived from combined results from the two RCTs and seven case series, which included dose-finding studies, an open-label extension study (that included patients from other romiplostim studies), a study of patients with severe refractory ITP and a bone marrow morphology substudy. Both the incidence and event rates adjusted for study duration for all adverse events that occurred during treatment were calculated. Safety data were submitted as academic-in-confidence information by the manufacturer.
3.9 Data on health-related quality of life from the two phase III RCTs included data from EuroQol 5-D (EQ-5D) and from the ITP Patient Assessment Questionnaire, which is a disease-specific instrument comprising 10 scales. Statistically significant differences between health-related quality of life were not reported in the manufacturer's original submission. In a revised submission, the manufacturer provided a linear regression analysis of combined EQ-5D data from both RCTs, which showed statistically significant differences favouring romiplostim compared with placebo in the mean change in EQ-5D score. Combined data from both RCTs for the change from baseline using the ITP Patient Assessment Questionnaire indicated a statistically significantly greater (p < 0.05) improvement in the 'Symptoms', 'Bother', 'Social Activity' and 'Women's Reproductive Health' scales in the romiplostim group than the placebo group for splenectomised patients. For non-splenectomised patients, those in the romiplostim group had a statistically significantly greater improvement in the 'Activity' scale than those in the placebo group.
3.10 The manufacturer did not identify any RCTs on the effectiveness of comparator treatments used in standard care for ITP compared with romiplostim, as defined in the scope of this appraisal. These included corticosteroids, IVIg, rituximab, immunosuppressive agents (azathioprine, mycophenolate mofetil and ciclosporin), anti-D immunoglobulin and splenectomy for non-splenectomised patients, and corticosteroids, IVIg, rituximab and immunosuppressive agents for splenectomised patients. The manufacturer assessed clinical guidelines, systematic reviews, trials and observational studies for evidence on the effectiveness of comparator treatments and found mostly case series. The manufacturer combined data on efficacy from different studies that examined the same treatment by a method described in the manufacturer's submission as 'taking a weighted average, weighting by sample size'. The number of studies combined in this way varied by treatment.
Cost effectiveness
Original economic model
3.11 The manufacturer submitted an original economic model and, after consultation on the first appraisal consultation document (ACD), a revised model. The manufacturer's original cohort-type economic model used a lifetime horizon and assessed the impact of romiplostim separately for patients with ITP who had undergone splenectomy and those who had not. The model assumed that all patients started with a platelet count below 50 × 109 per litre. Romiplostim was compared with standard care in a model structure in which patients initially either enter 'watch and rescue' (treated as necessary with intravenous immunoglobulin, anti-D immunoglobulin [non-splenectomised patients] or intravenous corticosteroids) or are treated with romiplostim followed by 'watch and rescue'. In the model, patients move through a care pathway consisting of active treatments and 'watch and rescue'. When a patient becomes refractory to an active treatment they receive 'watch and rescue'. The active treatments modelled in the remainder of the care pathway were rituximab, immunosuppressive agents (azathioprine, mycophenolate mofetil, ciclosporin), danazol, dapsone and cytotoxic agents (such as cyclophosphamide and vinca alkaloids).
3.12 The manufacturer built seven health states into the model. The manufacturer conducted a utility survey with the primary objective of directly measuring health utility values for these ITP health states as perceived by members of the public in the UK. Respondents were presented with descriptions of each ITP health state, and utility values for five of the seven health states were derived using the time trade-off method. The utility values for these five health states were: 0.863 for platelet count above 50 × 109 per litre and no bleed; 0.734 for platelet count above 50 × 109 per litre and outpatient bleed; 0.841 for platelet count below 50 × 109 per litre and no bleed; 0.732 for platelet count below 50 × 109 per litre and outpatient bleed; 0.038 for platelet count below 50 × 109 per litre and intracranial haemorrhage. For the remaining two health states included in the model (low platelet count with gastrointestinal bleed and low platelet count with other bleeding events requiring hospitalisation) the manufacturer used a utility value of 0.54 taken from the literature.
3.13 The manufacturer divided use of resources and costs into 4-week cycles. Costs of treatment included those of the romiplostim vials, laboratory testing to check blood counts every cycle (four tests), physician appointments (two sessions), and other drugs. Costs of management of bleeds included those for minor bleeds treated in an outpatient setting, intracranial haemorrhage, gastrointestinal bleeds and other bleeds requiring hospitalisation.
3.14 Other costs for drugs used in treatment and in managing bleeds were taken from the BNF (edition 55) and NHS reference costs.
3.15 The manufacturer's base-case analyses using the original economic model gave incremental cost-effectiveness ratios (ICERs) of £14,633 per quality-adjusted life year (QALY) gained for non-splenectomised patients and £15,595 per QALY gained for splenectomised patients, when using romiplostim as a first-option treatment compared with 'watch and rescue' standard care and with a target platelet count of 50 × 109 per litre. These ICERs reflected new inputs added to the model at the request of the ERG during the clarification step.
3.16 Sensitivity analyses performed by the manufacturer using the original economic model, in response to a request for clarification from the ERG, included the effects of: changes in drug costs to account for vial wastage in practice; the use of EQ-5D data available from the RCTs for serious adverse events; the cost of a bone marrow assessment needed when the condition no longer responds to romiplostim; and the cost of blood film tests (which are needed before treatment with romiplostim can begin).
3.17 The manufacturer, in response to a request for clarification from the ERG, also performed a sensitivity analysis in which it assumed a target platelet count of 30 × 109 per litre (instead of 50 × 109 per litre). This gave ICERs of £14,840 per QALY gained for non-splenectomised patients and £14,655 per QALY gained for splenectomised patients using romiplostim as a first-option treatment compared with 'watch and rescue'.
3.18 The manufacturer estimated that the probability that romiplostim would be cost effective using a target platelet count of 50 × 109 per litre at different acceptability threshold levels of £10,000, £20,000 and £30,000 per QALY gained was 10%, 60% and 81% respectively for non-splenectomised patients (mean ICER £14,633 per QALY gained), and 25%, 55% and 77% respectively for splenectomised patients (mean ICER £15,595 per QALY gained).
3.19 The ERG noted that limited evidence was available for romiplostim and potential comparators for the treatment of patients with chronic ITP, and particularly for long-term outcomes. The ERG also considered that the evidence for comparators did not distinguish between patients who had not undergone splenectomy and splenectomised patients. The ERG was particularly concerned about the methods the manufacturer had used to estimate the efficacy of romiplostim and the comparators, while acknowledging that using formal methods may also have been inappropriate. The ERG was not presented with further evidence that would have significantly altered the results presented in the manufacturer's original submission.
3.20 The ERG performed explorative sensitivity analyses using the manufacturer's original economic model. The one-way sensitivity analysis that had the most effect on the ICER was that in which the cost of romiplostim was adjusted to account for wastage from single-use vials that would occur in practice. On changing the number of vials from 0.93 to 1 for non-splenectomised patients, the ICER increased from £14,633 to £21,214 per QALY gained. For splenectomised patients, a change in the number of vials from 1.4 to 2 increased the ICER from £15,595 to £91,406 per QALY gained. The ERG carried out multivariate analyses which combined sensitivity analyses conducted by the ERG with those provided by the manufacturer. When patients entered the model on active treatment (rituximab) in the comparator arm (rather than 'watch and rescue') the ERG reported ICERs that increased from £14,633 to £21,674 per QALY gained for non-splenectomised patients and from £15,595 to £29,771 per QALY gained for splenectomised patients. When patients entered the model on active treatment (with rituximab) and the cost of romiplostim was adjusted to account for wastage, the ICER increased from £16,633 to £28,556 per QALY gained for non-splenectomised patients and from £15,595 to £109,802 per QALY gained for splenectomised patients. In a multivariate analysis that incorporated EQ-5D data from the RCTs rather than the utility values originally provided by the manufacturer, the cost of romiplostim adjusted to account for wastage, a 50% reduction in serious adverse events and the cost of bone marrow tests and blood film assessments, the ICERs increased from £14,633 to £37,290 per QALY gained for non-splenectomised patients and from £15,595 to £131,017 per QALY gained for splenectomised patients.
Revised economic model
3.21 After consultation on the first ACD, the manufacturer provided revised analyses of the cost effectiveness of romiplostim that addressed a number of the key issues that were raised by the Committee in the first ACD.
3.22 In the manufacturer's revised base-case analysis, it was assumed that 59% of patients in the comparator group start on rituximab and the remaining 41% start on another active treatment in the pathway (that is, immunosuppressive agents, dapsone, danazol and cytotoxic agents). Once an active treatment fails, and before the next active treatment is used, the model assumes that a patient enters a period of 'watch and rescue'.
3.23 The manufacturer took into account potential vial wastage of romiplostim and the fact that patients in the trials received doses that were higher than the maximum dose recommended in the SPC of 10 micrograms/kg. The manufacturer stated that six of the 42 patients in the trial of splenectomised patients were given more than 10 micrograms/kg romiplostim, and one of these showed a response to treatment. In the trial of non-splenectomised patients, two of the 41 patients received more than 10 micrograms/kg romiplostim, one of whom showed a response to treatment. In the cost-effectiveness analyses, the manufacturer provided two base-case scenario analyses: a 'conservative' scenario and a 'realistic' scenario. In both scenarios, all patients who received more than 10 micrograms/kg romiplostim were modelled as 'non-responders'. In the conservative scenario, the costs of continuing treatment at doses above 10 micrograms/kg were included in the model. In the realistic scenario, romiplostim costs were capped at 10 micrograms/kg.
3.24 The manufacturer pooled two sources of utility data: the individual patient level EQ-5D utility values from the two RCTs, and individual person level utility values from the time trade-off study (see section 3.12). The two resulting utility values that differed from the manufacturer's original utility values (based on the time trade-off study) were 0.835 for patients with a platelet count above 50 × 109 per litre and no bleed, and 0.800 for patients with a platelet count below 50 × 109 per litre and no bleed.
3.25 The manufacturer included the costs of bone marrow tests and blood film assessments associated with treatment with romiplostim in the revised analyses. The manufacturer also accounted for reductions in the number of blood counts and patient visits to clinicians.
3.26 The manufacturer submitted a patient access scheme, which is a discount on the 250 microgram vial of romiplostim, and provided four cost-effectiveness analyses: for the use of the 250 microgram vial with and without the patient access scheme, and for the use of a 100 microgram vial with and without the patient access scheme. This document only details the results for the use of the 250 microgram vial with the patient access scheme.
3.27 The manufacturer's revised base-case ICERs for the non-splenectomised group, incorporating all of the above assumptions and the patient access scheme, were £24,795 and £28,278 per QALY gained for the realistic scenario and the conservative scenario respectively. The ICERs for the splenectomised group were £4615 and £16,530 per QALY gained for the realistic scenario and the conservative scenario respectively.
3.28 The ERG reviewed the manufacturer's revised base-case analysis, and noted that uncertainty remained about which active treatment best reflects UK clinical practice. The ERG highlighted that the manufacturer's model describes a defined sequence of treatments, and questioned whether it was reasonable to exclude some treatments in the comparator arm. The ERG performed analyses using the manufacturer's revised base-case model and noted that a patient who enters the comparator arm of the model spends most time on 'watch and rescue' rather than on any other treatment. The ERG noted that any change in the model structure that reduces the amount of time a patient spends in 'watch and rescue' would increase the ICERs for romiplostim.
3.29 The ERG expressed concerns about the methods by which the manufacturer calculated the utility values in its revised base-case analysis. The ERG noted that the manufacturer simply aggregated the two utility measures, without considering whether it was appropriate to combine data from two different tools and two different samples. The ERG questioned whether the EQ-5D data derived directly from the trials might provide the best estimates of utility values.
3.30 The ERG reviewed the approach taken by the manufacturer to account for patients who received doses of romiplostim above the maximum dose stated in the SPC. The ERG noted that patients lost to follow-up in the trials may not have had the same outcomes as patients not lost to follow-up, which could affect the calculation of time to failure. Censored patients were defined as 'lost to follow-up' in the evaluation of time to failure for romiplostim, and were those who had a last observed visit that was not recorded as a withdrawal in the open-label extension study. The ERG expressed further concerns about the assumption that romiplostim extends life, and about the generalisability and applicability of the trial results to a typical NHS population of patients with ITP. The ERG identified no additional evidence that would reduce this uncertainty.
3.31 The ERG conducted an exploratory analysis using the manufacturer's 'realistic' scenario, to test the impact of a 10% increase in the average number of 250 microgram vials of romiplostim used. For splenectomised patients the ICER rose from £4615 to £20,340 per QALY gained, and for non-splenectomised patients the ICER rose from £24,795 to £34,410 per QALY gained. The ERG also noted that if there was lower usage of romiplostim in NHS clinical practice than is reflected in the values included in the manufacturer's model the ICERs for romiplostim would decrease.
3.32 The ERG explored the impact of reducing the duration of the response to romiplostim in the model by 10%, 20% and 30%. When this was reduced by 30%, the ICER rose from £4615 to £6138 per QALY gained for splenectomised patients, and from £24,795 to £25,363 per QALY gained for non-splenectomised patients.
3.33 The ERG conducted one-way sensitivity analyses by varying individual parameters in the revised base-case model to check the impact on the ICERs. These changes included increasing the use of comparator treatments by 25%; increasing the response time for comparators by 50%; increasing response rates for comparators by 25%; reducing the use of rescue therapies to 80% of the base case in both the comparator and romiplostim arms; using alternative utility values; and assuming a 'worst case scenario' in which all patients who were censored in the open-label extension study were assumed to have no longer responded to romiplostim and were treated as withdrawals. The change in the use of rescue therapies had the greatest impact, increasing the ICERs from £24,795 to £35,248 per QALY gained for non-splenectomised patients and from £4615 to £32,190 per QALY gained for splenectomised patients. The ERG noted that when all patients who were censored in the open-label extension study were assumed to have no longer responded to romiplostim and were treated as withdrawals, the ICERs rose from £24,795 to £31,601 per QALY gained for non-splenectomised patients and from £4615 to £18,647 per QALY gained for splenectomised patients.
3.34 The ERG performed a multi-way sensitivity analysis to show the impact on the ICERs of the cumulative effects of varying all individual parameters explored in the one-way sensitivity analyses (see section 3.33). The ICER for splenectomised patients rose from £4615 to £64,646 per QALY gained, and that for non-splenectomised patients rose from £24,795 to £55,470 per QALY gained.
3.35 Full details of all the evidence are in the manufacturer's submission and the ERG report.