The appraisal committee (section 7) considered evidence submitted by Novartis and a review of this submission by the evidence review group (ERG; section 8). See the committee papers for full details of the evidence.
3.1 The company conducted a systematic literature review for clinical trials investigating ruxolitinib that included patients with primary myelofibrosis, post polycythaemia vera myelofibrosis or post essential thrombocythaemia myelofibrosis. Two randomised controlled trials were identified that met the inclusion criteria: COMFORT‑I and COMFORT‑II. The company also included supportive evidence from 4 non-randomised controlled studies of ruxolitinib in patients with intermediate‑1 risk myelofibrosis or a low platelet count (ROBUST, JUMP, Study 258 and EXPAND).
3.2 COMFORT‑I is a multicentre (USA, Canada and Australia), phase III, randomised, double-blinded trial that compared ruxolitinib (15 mg or 20 mg twice daily, n=155) with placebo (n=154) in people with primary myelofibrosis (45.2% of ruxolitinib group; 54.5% of placebo group), or myelofibrosis secondary to polycythaemia vera (32.3% of ruxolitinib group; 30.5% of placebo group) or essential thrombocytopenia (22.6% of ruxolitinib group; 14.3% of placebo group). Patients who enrolled on the trial had resistant or refractory myelofibrosis, or available therapy was contraindicated or not tolerated. All patients on the trial had intermediate‑2 risk or high-risk myelofibrosis, a platelet count of at least 100×109/litre and a palpable spleen length of at least 5 cm. The duration of the study was 24 weeks, after which patients could enter an open-label extension phase. In COMFORT‑I, patients were eligible to crossover to ruxolitinib treatment. Before week 24, patients on placebo needed to have symptom worsening and 25% or more spleen volume increase from baseline. After week 24, patients needed to have 25% or more spleen volume increase from baseline.
3.3 COMFORT‑II is a multicentre (Europe, including sites in the UK), phase III, randomised, open-label trial that compared ruxolitinib (15 mg or 20 mg twice daily, n=146) with best available therapy (n=73) in people with primary myelofibrosis (53% of ruxolitinib group; 53% of the best available therapy group), or myelofibrosis secondary to polycythaemia vera (33% of ruxolitinib group; 27% of best available therapy group) or essential thrombocythaemia (14% of ruxolitinib group; 19% of best available therapy group). Best available therapy comprised a range of treatments. The most frequently used were hydroxycarbamide, prednisolone and epoetin alfa. Other treatments used as best available therapy included lenalidomide and thalidomide. All patients on the trial had intermediate‑2 or high-risk myelofibrosis, a platelet count of at least 100×109/litre and a palpable spleen length of at least 5 cm. The company stated that the trial population may have been healthier than the general population with myelofibrosis because the trial excluded people with uncontrolled hypertension, unstable angina and a life expectancy of less than 6 months. The duration of the trial was 48 weeks, after which patients could enter an open-label extension phase. In COMFORT‑II, patients were eligible to crossover to ruxolitinib treatment. Patients on best available therapy whose disease progressed (defined according to the study protocol as either 25% or more increase in spleen volume from on‑study nadir, including baseline, or needing splenectomy, splenic irradiation or leukaemic transformation) could crossover to have ruxolitinib at any time.
3.4 The primary outcome for both COMFORT‑I and COMFORT‑II was the proportion of patients achieving a spleen volume reduction of 35% or more from baseline, assessed by MRI or CT scan. The primary efficacy outcome was measured at 24 weeks in COMFORT‑I and at 48 weeks in COMFORT‑II.
3.5 Secondary outcomes for the COMFORT‑I trial included maintenance of reduction in spleen volume, reduction in palpable spleen length, change in total symptom score (measured using the modified myelofibrosis symptom assessment form [MF‑SAF] v2.0 diary), overall survival, and health-related quality-of-life measures. Secondary outcomes for the COMFORT‑II trial included outcomes from the COMFORT‑I trial, as well as the time to achieve a spleen volume reduction of 35% or more, progression-free survival, leukaemia-free survival and transfusion dependency. In COMFORT‑II, additional overall survival analyses were carried out at 3.5 years follow‑up.
3.6 The intention-to-treat (ITT) population was used for all efficacy end points. Patients who stopped treatment or crossed over before 24 weeks (in COMFORT‑I), or did not have a 48‑week assessment of spleen volume (in COMFORT‑II because of discontinuation and entering the open-label extension phase) were counted as patients whose disease did not respond (for change in spleen volume and symptom score).
3.7 In COMFORT‑I, a statistically significantly greater proportion of patients in the ruxolitinib group achieved a reduction in spleen volume of 35% or more from baseline, compared with the placebo group at 24 weeks (41.9% versus 0.7%; p<0.001). In COMFORT‑II, a statistically significantly greater proportion of patients in the ruxolitinib group achieved a reduction in spleen volume of 35% or more from baseline, compared with the best available care group at 48 weeks (28% versus 0%; p<0.001). In COMFORT‑I, a statistically significantly greater proportion of patients in the ruxolitinib group achieved a reduction in total symptom score of 50% or more from baseline, compared with the placebo group at week 24 (45.9% versus 5.3%; p<0.001). This outcome was not collected in COMFORT‑II.
3.8 Overall survival was a secondary end point in both COMFORT trials and neither was designed to be sufficiently powered to detect a statistically significant difference in overall survival between treatment groups.
3.9 In COMFORT‑I, overall survival was statistically significantly improved with ruxolitinib over placebo at a median follow‑up of 51 weeks; 91.6% compared with 84.4% (hazard ratio [HR] 0.50; 95% confidence interval [CI] 0.25 to 0.98) and 102 weeks (HR 0.58; 95% CI 0.36 to 0.95). At a median follow‑up of 3 years, 42 patients in the ruxolitinib group and 54 patients in the placebo group had died and the difference in overall survival was no longer statistically significant (HR 0.69; 95% CI 0.46 to 1.03). Because crossover was permitted during the treatment period of the study the company provided an analysis that adjusted for crossover using the rank preserving structural failure time (RPSFT) method. Ruxolitinib was associated with a 64% reduction in the risk of death compared with placebo (HR 0.36; 95% CI 0.20 to 1.04).
3.10 In COMFORT‑II, overall survival was not statistically significantly different between ruxolitinib and best available therapy at a median follow‑up of 61 weeks. It reached borderline statistical significance at a median of 112 weeks of follow‑up: 86% compared with 78% (HR 0.52; 95% CI 0.27 to 1.00). At median follow‑up of 3 years, 20% (29 patients) in the ruxolitinib group and 30% (22 patients) in the best available therapy group had died, and ruxolitinib was associated with a 52% reduction in the risk of death compared with best available therapy (HR 0.48; 95% CI 0.28 to 0.85). The probability of survival at 144 weeks was 81% in the ruxolitinib group and 61% in the best available therapy group.
3.11 The company provided the results of a further analysis done at median follow‑up of 3.5 years, which included additional survival information for 15 of 41 patients who were previously deemed lost to follow‑up. At 3.5 years of follow‑up, 27% (40 patients) in the ruxolitinib group and 40% (30 patients) in the best available therapy group had died. Ruxolitinib was associated with a 42% reduction in the risk of death compared with best available therapy (HR 0.58; 95% CI 0.36 to 0.93); median overall survival had not yet been reached. The probability of survival at 3.5 years was 71% in the ruxolitinib group and 54% in the best available therapy group (p=0.02).
3.12 The company was asked during the clarification stage to provide an overall survival analysis with adjustment for crossover using the RPSFT method for the COMFORT‑II trial. Ruxolitinib was associated with a 65% reduction in the risk of death compared with best available therapy in the RPSFT analysis (the corrected hazard ratio is confidential and is therefore not presented here).
3.13 Because median overall survival was not reached in the ruxolitinib group, it was not possible to directly calculate the median (or mean) survival benefit associated with ruxolitinib compared with best available therapy, and therefore estimated values needed to be modelled. The company included a summary of an indirect comparison made between the ruxolitinib treatment group of COMFORT‑II and the Dynamic International Prognostic Scoring System (DIPSS) cohort. The DIPSS database is a multicentre database and includes 519 people with primary myelofibrosis, who were not having any experimental drug or haematopoietic stem cell transplantation. Matched patients (n=350) were compared with patients with primary myelofibrosis enrolled in COMFORT‑II (n=100). The number of observed deaths in the 2 cohorts was 30 (30%) on ruxolitinib and 256 (86%) on best available therapy, generating estimates of median survival of 5 years from diagnosis (95% CI 2.9 to 7.8) on ruxolitinib compared with 3.5 years (95% CI 3.0 to 3.9) for the DIPSS cohort.
3.14 Adverse event data were collected in COMFORT‑I at 24 weeks and at 48 weeks in COMFORT‑II. Anaemia was the most common grade 3 or 4 adverse event in COMFORT‑I (45%) and COMFORT‑II (42%). In COMFORT‑II, the most common adverse event was diarrhoea, and it was more frequently reported with ruxolitinib compared with best available therapy (23% compared with 12%). There were more grade 3 or 4 adverse events with ruxolitinib than with best available therapy (42% compared with 25%). There was a similar number of people with grade 3 or 4 thrombocytopenia with ruxolitinib compared with best available therapy (8% compared with 7%). Treatment was stopped in 12 people (8.2%) in the ruxolitinib group and 4 people (5.5%) in the best available therapy group because of adverse events.
3.15 Although symptom reduction was not specifically assessed in the COMFORT‑II trial, the company carried out a post hoc exploratory analysis of health-related quality of life and symptom analyses on the primary analysis data set (at 48 weeks) from COMFORT‑II. Of the 9 symptom scores assessed by the Global Health Status (EORTC QLQ‑C30), 6 symptom scores (appetite loss, dyspnoea, fatigue, insomnia, pain and diarrhoea) were improved with ruxolitinib compared with best available therapy.
3.16 Health-related quality of life was assessed in the COMFORT trials using the Global Health Status (EORTC QLQ‑C30) and Functional Assessment of Cancer Therapy for patients with Lymphoma (FACT‑Lym) questionnaires. There were statistically significant gains in favour of ruxolitinib in the average change in health-related quality of life in the COMFORT‑I trial, and there were improvements in all health-related quality-of-life subscales in favour of ruxolitinib in the COMFORT‑II trial.
3.17 The ROBUST study was a phase II study that was done in the UK (n=48). It included patients with intermediate‑1, intermediate‑2 and high-risk disease. At week 48, 40% of patients achieved reduction in spleen length of at least 50% and 21% achieved a reduction in total symptom score of at least 50% (as assessed using MF‑SAF). Treatment success, defined as a 50% or more decrease in spleen length and/or total symptom score at week 48, was achieved by 50.0% of the overall population and 57.1%, 38.5% and 52.4% of the intermediate‑1 risk, intermediate‑2 risk and high-risk disease groups respectively. Consistent with findings from the COMFORT trials, the most common haematological adverse events were anaemia (45.8% of patients) and thrombocytopenia (37.5%).
3.18 The phase III expanded-access, Janus-associated kinase (JAK)-inhibitor ruxolitinib in myelofibrosis patients (JUMP) trial was also designed to assess the safety and efficacy of ruxolitinib in patients with high-risk, intermediate‑2 risk or intermediate‑1 risk disease. As of September 2014, 2138 patients had been enrolled in 25 countries and data had been reported for an analysis of 1144 patients who had ruxolitinib for a median of 11.1 months. At week 48, 61% of patients achieved at least a 50% reduction from baseline in palpable spleen length. Clinically meaningful improvements in symptoms, as assessed using the FACT‑Lym total score, were seen as early as week 4 and were maintained during the study. Ruxolitinib was generally well-tolerated, with only 14% of patients stopping treatment as a result of adverse events. The most common grade 3 or 4 haematological adverse events were anaemia (33.0%), thrombocytopenia (12.5%) and neutropenia (3.9%); each of these rarely led to discontinuation of ruxolitinib. The incidences of grade 3 or 4 non-haematological adverse events were low.
3.19 The JUMP study included patients with low platelet counts (at least 50 to under 100×109/litre). In this patient population, ruxolitinib was started at a dose of 5 mg twice daily. This could be increased to 10 mg twice daily at week 4 in patients whose disease had not responded adequately, if platelet counts were at least 50×109/litre and there had been no treatment-related toxicities that resulted in dose reduction, interruption or discontinuation during initial treatment. Results of an interim analysis for 6 months of therapy in the first 50 patients with low platelet counts were reported. At this time point, 82% of patients (31 of 38 patients starting therapy on 5 mg twice daily) remained on the 5 mg twice daily dose and 18% had a dose escalation to 10 mg twice daily. At week 24, 38.2% (13 of 34 evaluable patients) achieved a reduction of at least 50% from baseline in palpable spleen length; overall, 44.7% of patients (21/47) achieved at least a 50% reduction from baseline in spleen length at any time. Clinically meaningful improvements in symptoms, as assessed using the FACT‑Lym total score, were seen as early as week 4 (mean change from baseline, 8.2) and were still seen at week 12 (change from baseline, 9.6). However, the reduction in splenomegaly and improvements in symptoms seen in this subgroup of patients were not as good as those achieved for the overall JUMP population. The adverse effect profile was consistent with previous studies in patients with platelet counts under 100×109/litre. The most common grade 3 or 4 haematological adverse events were thrombocytopenia (30%) and anaemia (28%): 3 patients (6%) stopped treatment because of thrombocytopenia and 1 patient stopped because of anaemia. Grade 1 or 2 haemorrhages were reported in 4 (8%) patients and grade 3 or 4 haemorrhages in 2 (4%) patients. Rates of grade 3 or 4 non-haematological adverse events were low. Nine patients (18%) stopped therapy because of adverse events. The company commented that this analysis suggested that ruxolitinib doses of 5 to 10 mg twice daily were generally well tolerated and efficacious in patients with myelofibrosis who have platelet counts of at least 50 to under 100×109/litre.
3.20 Study 258 was a phase II dose-finding study investigating the efficacy and safety of ruxolitinib in patients with low platelet counts (50 to 100×109/litre). Patients were started on a dose of 5 mg twice daily, with the option to increase to 10 mg twice daily if platelet counts remained adequate. An interim analysis of data from this study reported that by week 24, 62% of patients achieved stable doses of at least 10 mg twice daily. A median percentage reduction in spleen volume of 24.2% was achieved and 20% of patients achieved a reduction in spleen volume of at least 35.0% at 24 weeks. When evaluated by titrated dose (average dose over the last 4 weeks of the study, up to week 24), median percentage reductions from baseline in spleen volume at week 24 were 16.7% for patients who had 5 mg once or twice daily (n=7), and 28.5% for patients who had 10 mg twice daily (n=20). Decreases in total symptom score were also seen in patients who completed 24 weeks of therapy (n=32). The median percentage reduction from baseline in total symptom score for patients who completed 24 weeks of therapy was 43.8%. The study reported a mean change in Global Health Status (EORTC QLQ‑C30) score from baseline of approximately 13 at week 24.
3.21 Thrombocytopenia was the most frequently reported grade 3 or 4 adverse event, occurring in 56% of patients. Grade 3 or 4 anaemia was reported in 42% of patients. Most other adverse events were grade 1 or 2 and no other grade 3 or 4 adverse events were reported in more than 2 (4%) of patients. Thrombocytopenia that needed dose reductions and dose interruptions occurred in 12 (24%) and 8 (16%) of patients respectively, and occurred mainly in patients with baseline platelet counts of 75×109/litre or less. Two patients stopped treatment as a result of adverse events: in 1 patient, this was because of grade 4 thrombocytopenia; and the reason was not reported for the other patient. The company stated that the results of this study indicated ruxolitinib, started at a dose of 5 mg twice daily, can benefit patients with low platelet counts.
3.22 EXPAND is an open-label, phase Ib, dose-finding study, which investigated the optimum dose of ruxolitinib in patients with low baseline platelet counts. In this ongoing study 15 mg of ruxolitinib twice daily in patients with platelet counts of 75 to 99×109/litre and doses of up to 10 mg twice daily in patients with lower platelet levels are used. Results of a preliminary analysis of data for 34 patients have shown that most (97%) patients achieved reductions in palpable spleen length and 50% of patients achieved a reduction in spleen length of at least 50% as their best response. Improvements in symptoms, as assessed using the MF‑SAF total symptom score, were also observed; a reduction from baseline of at least 50% in total symptom score at any time was achieved by 43% (6/14) of patients with platelet counts of 75 to 99×109/litre and 66.7% (8/12) of patients with platelet counts of 50 to 74×109/litre. The reported adverse effects were consistent with the known safety profile of ruxolitinib.
3.23 The company submitted an individual patient discrete event simulation model comparing ruxolitinib with best available care. The company considered this design to be more flexible and transparent compared with a Markov cohort approach. The model had a lifetime horizon of 35 years. Although the model did not use time cycles, it effectively had a cycle length of 1 week, because this was the shortest unit of time in the model. The company based the analysis on an NHS and personal social services perspective, and costs and benefits were discounted at an annual rate of 3.5%. There were 4 health states in the model: on ruxolitinib, on best available therapy, on supportive care and death.
3.24 Hypothetical patients in the best available therapy group were assumed to begin in the best available therapy health state. In this health state, patients had a selection of treatments considered to be best available therapy, which reflects the treatment had by patients in the control group of the COMFORT‑II trial. Best available therapy was assumed to give patients some control of symptoms but no control of splenomegaly and little improvement in health-related quality of life. Patients could continue to have best available therapy until death or they could stop (after exhausting possible options) and progress to the supportive care health state. In this health state, patients experienced a gradual worsening of the disease (symptoms and haematological parameters) and health-related quality of life until death. No formal stopping rule was applied to patients having best available therapy and discontinuation was modelled based on discontinuation seen during the COMFORT‑II trial.
3.25 Hypothetical patients who entered the model on ruxolitinib were categorised into 4 groups based on their outcomes at 24 weeks in the COMFORT trials. There were 4 categories of response in the model: response, no response, early discontinuation of therapy or early death. Patients whose disease did not respond to treatment were subject to a stopping rule. The company's choice of 24‑week stopping rule and decision was based on the British Committee for Standards in Haematology guideline for the diagnosis and management of myelofibrosis (2012), and also took into account the Summary of Product Characteristics of ruxolitinib which states that treatment should be stopped after 6 months (26 weeks) if there has been no reduction in splenomegaly or improvement in symptoms since starting therapy. The definition of response was based on the International Working Group-Myeloproliferative Neoplasms Research and Treatment/European LeukemiaNet's criteria for treatment response in myelofibrosis guidelines, and defined in terms of either a spleen response or a symptom response. This stopping rule was not applied in the COMFORT‑I or COMFORT‑II trials.
3.26 Clinical-effectiveness data used in the model were mainly taken from the COMFORT‑II trial, which enrolled patients with intermediate‑2 or high-risk myelofibrosis. Additional data were used from the COMFORT‑I trial, which enrolled intermediate‑2 and high-risk patients whose disease did not respond to other therapies.
3.27 Ruxolitinib dosing was subject to dose-intensity adjustment and varied according to platelet count, patient's tolerance of therapy and efficacy. To reflect this, individual patient data from the COMFORT‑II trial were used to estimate dose given. Based on this data, the dose of ruxolitinib used in the model varied between 5 mg and 25 mg twice daily, or 5 mg and 35 mg once per day. For a small proportion of treatment days (1.38%), dose interruptions were also accounted for, that is, 0 mg dose. The most common doses used in the model were 5 mg twice daily (14.50% of treatment days), 10 mg twice a day (25.93% of treatment days), 15 mg twice daily (20.14% of treatment days) and 20 mg twice daily (30.66% of treatment days).
3.28 The comparator in the model, best available therapy, consisted of several different treatments for myelofibrosis based on data from the COMFORT‑II trial. Dose intensity, duration of treatment or order of treatment were not recorded in the COMFORT‑II trial. To account for this lack of data, several assumptions were made when calculating the cost of best available therapy.
3.29 In the model, the proportions of patients gaining a spleen response, stopping ruxolitinib treatment, and dying early were based on data from the COMFORT‑I and COMFORT‑II trials. The proportion of patients gaining symptom response was based on the COMFORT‑I trial. Because there were no data to model overall survival and discontinuation rates in a response group that included both patients whose spleen decreased in length by 50% or more and whose symptoms improved, the company assumed that overall survival and discontinuation rates were the same for both arms.
3.30 For patients starting best available therapy, death could occur either while on treatment or after stopping best available therapy, when patients had moved to the supportive care state. The number of patients dying on best available therapy was based on data from the COMFORT‑II trial and time to death for this group was based on time to stopping therapy.
3.31 For all patients starting on ruxolitinib, mortality rates were the same during the initial treatment phase (24 weeks). After 24 weeks, mortality rates differed according to whether their disease responded to treatment, did not respond to treatment, or they stopped treatment during the initial treatment phase. As with best available therapy, patients whose disease responded to ruxolitinib treatment could die either while on treatment or after they had stopped treatment. Data for both ruxolitinib and best available therapy were obtained from the COMFORT‑II trial. In the baseline model, the mortality rate for patients whose disease responded to ruxolitinib was assumed to be 0.0%, that is, no patients die while on ruxolitinib. For patients stopping ruxolitinib (both during the initial 24‑week period and for patients whose disease responded after this initial period), duration alive after discontinuation was modelled based on survival seen in the COMFORT‑II trial.
3.32 Patients whose disease did not respond to ruxolitinib were assumed to move to best available therapy after 24 weeks. Mortality was modelled in the same way as patients starting on best available therapy except that patients whose disease did not respond to ruxolitinib were assumed to have a mortality benefit of an additional 24 weeks of life.
3.33 For patients starting on ruxolitinib, the model used 2 alternative discontinuation rates: one for the initial 24‑week treatment phase of the model (see section 3.25); and one that was applied after 24 weeks to patients who had a 50% or more reduction in spleen length and improvement in symptoms from baseline, and who continued treatment. Both rates were taken from the COMFORT‑II trial. After 24 weeks, the rate of discontinuation was based on analysis of time to discontinuation for patients who had a reduction in spleen length of 50% or more. A range of parametric survival models were considered to extrapolate beyond the observed data, and based on the Akaike information criterion and Bayesian information criterion, a Gompertz distribution was considered the most appropriate. Scenario analyses using the alternative distributions were also presented. A single rate of discontinuation was used for patients on best available therapy, based on data from the COMFORT‑II trial, because no stopping rule was applied. As with discontinuation from ruxolitinib, several parametric survival models were tested and the Gompertz distribution was considered to be the most appropriate. The company also presented scenario analyses using alternative distributions.
3.34 The model included the possibility of leukaemic transformation. It did this by allowing this to occur as an adverse event with disutility and cost applied. The company used the same rate of leukaemic transformation from the COMFORT‑II trial for patients in both the ruxolitinib and best available therapy groups.
3.35 The COMFORT‑I and COMFORT‑II trials did not include a generic measure of health-related quality of life (such as the EQ‑5D). However, the company explained that although it would have been possible to do so, it was not considered appropriate to use a mapping algorithm to develop health-related quality of life based on EQ‑5D. Instead, a condition-specific preference-based measure for myelofibrosis, the MF‑8D, was developed using existing measures, the MF‑SAF and EORTC QLQ‑C30. The model used changes in health-related quality of life on a continuous scale according to different phases of the myelofibrosis disease state. Patients were assumed to experience constant benefits with ruxolitinib and best available therapy, but health-related quality of life was assumed to steadily decline in the supportive care health state.
3.36 The costs associated with managing myelofibrosis were obtained from the Haematological Malignancies Research Network (HMRN) audit and the ROBUST study. The HMRN audit provided information on the number of hospital nights, outpatient visits and laboratory tests. The ROBUST study provided data on resource use. Data from the JUMP study were used to represent the reduction in resource use associated with the use of ruxolitinib. These data were supplemented by information from the COMFORT trials and assumptions when appropriate.
3.37 The company presented base-case cost-effectiveness results with and without a patient access scheme (PAS). The deterministic incremental cost-effectiveness ratio (ICER) for ruxolitinib compared with best available therapy with the original PAS was £44,905 per quality-adjusted life year (QALY) gained (incremental costs £112,843, incremental QALYs 2.51).
3.38 The company carried out a series of deterministic one-way sensitivity analyses. Most inputs had minimal impact on the ICER estimate, with the exception of post-ruxolitinib discontinuation survival, and the overall survival estimate for best available therapy. However, the estimated ICER did not exceed £50,000 per QALY gained in any of the sensitivity analyses.
3.39 The company conducted a series of scenario analyses:
varying the model time horizon; assuming the best available therapy discontinuation rate followed an exponential, Weibull or log-normal distribution
varying the duration on best available therapy, using the ITT overall survival estimate from the COMFORT‑II trial
changing the post-best-available-therapy discontinuation survival (survival after best available therapy discontinuation) to follow a shape of 1 (compared with 0.63 in the base case)
impact of different response criteria
discontinuation rate for patients on ruxolitinib achieving a spleen response was assumed to follow alternative distributions and assuming all patients remain on treatment for a maximum duration of 3.5 years, 5 years, 7.5 years and 10 years.
None of these scenarios were found to significantly impact the ICER.
3.40 After consultation on the appraisal consultation document (ACD), the company requested permission which was granted by NICE, to present new evidence and a revised version of the model, which was updated in line with the ERG's critique on the original model. The changes to the assumptions of the original model were:
Increased PAS discount.
Correction of the errors, identified by the ERG, on the formula used for including leukaemic transformation.
Excluded lenalidomide from the basket of treatments that made up best available therapy and replaced with thalidomide.
Upgraded baseline utility by 10%.
People whose disease responds to ruxolitinib spend 30% of their time on best available therapy, after stopping ruxolitinib.
This resulted in a revised company base-case ICER of £31,229 per QALY gained (incremental costs £89,428; incremental QALYs 2.86).
3.41 The company also did exploratory analyses that incorporated the impact of ruxolitinib treatment on the quality of life of carers. A 0.1 utility decrement was applied for carers of patients on best available therapy to the utility of the general population. Using the revised company model, this analysis resulted in an ICER of £28,060 per QALY gained (incremental costs £87,633; incremental QALYs 2.84).
3.42 In response to a request from NICE, the company carried out further exploratory analyses to calculate the cost-effectiveness results separately for the 2 subgroups reflected in the model (that is, for the intermediate‑2 and high-risk subgroups). This model incorporated separate patient-level data from the COMFORT trials for the intermediate‑2 and high-risk subgroups for the following inputs:
Overall survival on the best available therapy arm, adjusted for crossover using RPSFT method.
Number of outpatient visits on best available therapy.
Probability of discontinuation for people whose disease responds (spleen response only) on ruxolitinib.
Survival after stopping ruxolitinib.
Change in health-related quality of life.
The results showed that the ICER for the intermediate‑2 risk subgroup was £25,896 per QALY gained (incremental costs £124,429; incremental QALYs 4.80) and the ICER for the high-risk subgroup was £37,985 per QALY gained (incremental costs £59,119; incremental QALYs 1.56).
3.43 The ERG was satisfied that all relevant studies had been included in the company's submission. The ERG stated that the COMFORT trials were of good quality and appropriate for addressing the decision problem.
3.44 The ERG commented that the COMFORT trials only included patients with splenomegaly and intermediate‑2 or high-risk myelofibrosis, who had a platelet count of 100×109/litre or more and an absolute neutrophil count of 1×109/litre or more. Also, patients suitable for allogeneic haematopoietic stem cell transplantation (allo‑HSCT) at the time of study enrolment were excluded from the trials. Therefore, the population represented in the trials was narrower than that covered by the marketing authorisation.
3.45 The ERG stated that overall survival was a secondary end point in both the COMFORT trials and that neither trial had sufficient power to detect a statistically significant difference in overall survival between treatments. The ERG noted that all methods to adjust for crossover have limitations, but the methods used by the company were appropriate.
3.46 The ERG considered the economic model in the company submission and the updated model received during consultation (see section 3.40), which included an updated PAS and revised assumptions. It agreed that the increased PAS discount had been correctly applied. The ERG commented that using an individual patient discrete event simulation model can be considered novel because most oncology models are cohort Markov structures. The ERG stated that using this type of modelling approach appears justified given the progressive nature of the disease and has the advantage of increased flexibility and is appropriate for the decision problem.
3.47 The ERG noted that the population in the model pragmatically reflected the patients in COMFORT‑II, which represents a subset of the population specified in the marketing authorisation for ruxolitinib, that is, patients with intermediate‑2 or high-risk myelofibrosis. The ERG commented that the modelling presented therefore reflects the cost effectiveness of ruxolitinib in this more restricted population.
3.48 The ERG had concerns about the composition of best available therapy used in the original model. The clinical adviser to the ERG indicated that lenalidomide is rarely used in the UK, and the HMRN audit appeared to confirm this. The ERG also noted that in the revised model, the company replaced lenalidomide with thalidomide in the basket of therapies that made up best available therapy and agreed that this assumption was plausible. The ERG stated that it was also clear from the published literature that other treatments are used in the UK, which were not included as part of best available therapy. In particular, the British Committee for Standards in Haematology guideline for the diagnosis and management of myelofibrosis (2012) indicates that allo‑HSCT is a potential therapy for myelofibrosis and is the only curative treatment for patients. The ERG was of the opinion that allo‑HSCT should have been considered either as part of best available therapy or as an alternative comparator because significant survival benefits have been seen using allo‑HSCT. However, the ERG recognised that this treatment option would not be suitable for all patients and has a different treatment goal (curative as opposed to managing symptoms).
3.49 The ERG considered the assumption of no drug wastage for ruxolitinib did not accurately reflect drug usage in clinical practice. The ERG had concerns about drug wastage, given that most adverse events are managed by dose reduction or interruption, leading to additional costs.
3.50 The ERG considered the company's assumption of 0% mortality with ruxolitinib treatment to be unrealistic. During clarification, the company acknowledged that this assumption may be optimistic. It therefore provided additional scenario analyses assuming either the same probability of death on discontinuation used for the best available therapy group, or assuming a probability equal to 10%.
3.51 The ERG considered the assumption of upgrading the baseline utility value by 10%. It noted that upgrading the utility value was in line with its critique on the original model, and that this scenario may represent a more realistic estimate of the ICER. However, the ERG raised concerns that a 10% upgrade in utility, based on clinical opinion was associated with substantial uncertainty and considered the following to be equally plausible:
the unadjusted baseline values
the 5% upgrading factor, presented in a scenario analysis by the company
and the 10% upgrading factor.
3.52 The ERG commented on the revised structure of the model and the assumption that patients whose disease responded on the ruxolitinib arm of the model would spend 30% of their time alive on best available therapy, instead of moving directly to supportive care. The ERG acknowledged that in clinical practice, it is likely that a proportion of patients would spend time on best available therapy after stopping ruxolitinib treatment, but it noted that there is minimal evidence available to support it. Regarding the 30% value, the ERG commented that this value seems to be arbitrary. It also commented that the assumption that all patients would move to best available therapy after stopping treatment with ruxolitinib is unlikely to reflect clinical practice. It also considered the assumption that best available therapy would be as effective for patients who stopped ruxolitinib as for patients starting best available therapy to be highly optimistic. The ERG therefore considered the revised structure of the model to be subject to high uncertainty.
3.53 The ERG stated that the results presented by the company for the extensive sensitivity and scenario analyses, based on the original and the revised models, showed the estimated ICER to be largely robust to a range of input values and assumptions made in the model.
3.54 The ERG noted that the new evidence and the revised model (with the updated PAS) received during consultation addressed several issues raised by the ERG, including the composition of best available therapy and upgrading baseline utility values. It did further exploratory analyses on the revised model focusing on the discontinuation rate for best available therapy and assumptions around the revised model structure.
3.55 The ERG carried out exploratory analyses on the revised model, in which it incorporated all of its preferred assumptions into the company's revised cost-effectiveness model. The assumptions of the 2 scenarios were:
Scenario 1 (ERG preferred base case):
Time on ruxolitinib is part of the time on best available therapy for people whose disease does not respond to ruxolitinib.
No adjustment to baseline utility.
Discontinuation rate for best available therapy is reduced by 20%.
People move to supportive care after stopping ruxolitinib, in line with the company's original model structure.
Scenario 2 (all the above assumptions plus the company's preferred base case):
Upgrade baseline utility values by 10%.
People whose disease responds to ruxolitinib spend 30% of their time alive after stopping ruxolitinib on best available therapy, in line with the company's revised model structure.
The results of the exploratory analyses using these scenarios were an ICER of £35,632 per QALY gained for Scenario 1 and an ICER of £31,676 per QALY gained for Scenario 2.
3.56 The ERG then critiqued on the company's exploratory analyses. It did not consider the carer scenario to be robust because the company used data from studies that were done with a non-UK patient population. It also commented that the utility decrement applied for the carer was based on a study from a different disease area (glioma and other types of cancer) and that the results of this study did not represent a statistically significant change in quality of life. Finally, the ERG considered that the assumption that the quality of life of carers returns to that of the general population during treatment with ruxolitinib to be overly optimistic. Using its revised assumptions in Scenario 1 (see section 3.53), it calculated the cost-effectiveness result of the exploratory analysis on carers' quality of life. This resulted in an ICER of £31,855 per QALY gained.
3.57 The ERG also commented on the exploratory analyses that calculated the ICERs separately for the 2 subgroups: intermediate‑2 and high-risk disease. It considered that the implementation of the assumptions was appropriate and that there were significant differences in the ICERs between the subgroups. These were because of the substantial differences between the subgroups in the response and prognosis of patients having ruxolitinib and the minimal differences in prognosis for patients having best available therapy. The ERG also noted that there was no difference between the 2 subgroups in overall survival for patients on the best available therapy arm.