3 The manufacturer's submission
The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of vemurafenib and a review of this submission by the Evidence Review Group (ERG; appendix B). The decision problem addressed by the manufacturer considered people with BRAF V600 mutation‑positive melanoma who have not previously received treatment, which is in contrast to the original decision problem that allowed for vemurafenib to be considered in both first‑ and subsequent‑line treatment settings.
3.1 The key clinical evidence came from 1 multicentre, randomised, open‑label, active‑controlled trial (BRIM3) that compared vemurafenib (960 mg twice daily orally; n=337) with dacarbazine (1000 mg per square metre of body surface area by intravenous infusion every 3 weeks; n=338) in adults with previously untreated stage IIIc or IV BRAF V600 mutation‑positive metastatic melanoma, until disease progression or unacceptable toxicity. The randomisation process produced equivalent‑sized groups. However, 14% of patients (48 of 338) randomised to receive dacarbazine did not receive treatment, primarily because they withdrew consent or refused treatment. The median age of patients in the trial was 56 years for people receiving vemurafenib and 52 years for those receiving dacarbazine. About 60% of patients were from western Europe, and the proportion of patients with an Eastern Cooperative Oncology Group (ECOG) performance score of 0 was 68% in both the vemurafenib and dacarbazine groups. At study entry, more than 90% of patients had stage IV disease.
3.2 The primary outcome in the BRIM3 study changed from overall survival to a joint primary outcome of overall survival and progression‑free survival during the study, at the request of the US Food and Drug Administration. Secondary outcomes included confirmed best overall response rate, duration of response and time to response.
3.3 The manufacturer presented 3 analyses for overall survival based on 3 different data cut‑off points (December 2010, March 2011 and October 2011). The Data and Safety Monitoring Board recommended the release of the interim results of efficacy, based on a review of the results of the planned interim analysis of overall survival, and the study was ended and crossover allowed at this time (December 2010). The manufacturer performed 2 additional analyses (using March 2011 and October 2011 data cut‑off time periods) to demonstrate the survival benefit conferred by vemurafenib during follow‑up.
3.4 Results from the December 2010 data cut‑off of the BRIM3 trial showed that treatment with vemurafenib led to a statistically significant reduction in death (hazard ratio [HR] 0.37; 95% confidence interval [CI] 0.26 to 0.55; p<0.001). At 6 months, overall survival was 84% (95% CI 78 to 89) in the vemurafenib group and 64% (95% CI 56 to 73) in the dacarbazine group. People treated with vemurafenib also had a statistically significant reduction in tumour progression (HR 0.26; 95% CI 0.20 to 0.33; p<0.001). The estimated median progression‑free survival (evaluated in 549 patients) was 5.32 months (95% CI 4.86 to 6.57) in the vemurafenib group and 1.61 months (95% CI 1.58 to 1.74) in the dacarbazine group.
3.5 The secondary outcome of confirmed tumour response could be calculated for 439 patients for the December 2010 data cut‑off. In the vemurafenib treatment group, 106 of 219 patients (48%; 95% CI 42 to 55) had a confirmed objective response (including 2 patients with a complete response and 104 patients with a partial response), with a median time to response of 1.45 months. Only 12 of the 220 patients (5%; 95% CI 3 to 9) treated with dacarbazine had a partial response (no patients had a complete response), with a median time to response of 2.7 months.
3.6 Results from the March 2011 data cut‑off included 50 patients (15%) who switched from dacarbazine to vemurafenib. The censored hazard ratio for overall survival was 0.44 (95% CI 0.33 to 0.59). Results from the October 2011 data cut‑off, which included 24% (n=81) of patients who switched from dacarbazine to vemurafenib on disease progression, showed that median overall survival was 13.2 months for the vemurafenib group and 9.6 months for people treated with dacarbazine (censored HR 0.62; 95% CI 0.49 to 0.77).
3.7 In response to consultation, the manufacturer also presented results based on the February 2012 data cut‑off. This included data on 34% of patients who switched over from dacarbazine to vemurafenib and other BRAF inhibitors. Results showed that treatment with vemurafenib led to a statistically significant progression‑free survival benefit (HR 0.38; 95% CI 0.32 to 0.46; p<0.001) compared with dacarbazine. Median overall survival was 13.6 months in the vemurafenib group and 10.3 months in the dacarbazine group (uncensored HR 0.76; 95% CI 0.63 to 0.93; p<0.01). Tumour response rate (defined as at least a 30% reduction in tumour size) was 57% (192 out of 337 patients) in the vemurafenib group compared with 8.6% in the dacarbazine group. In the vemurafenib group, 5.6% of patients had a complete response (that is, the disappearance of all disease), compared with 1.2% in the dacarbazine arm.
3.8 The manufacturer reported results from a range of pre‑specified subgroups, including age, sex, ECOG performance status, tumour stage and geographical regions. The results showed that the survival benefit conferred by vemurafenib treatment was generally maintained across each subgroup.
3.9 The most commonly reported adverse events (grade 2 or more) associated with vemurafenib treatment in the BRIM3 study were cutaneous events, arthralgia and fatigue (December 2010 cut‑off based on 618 patients). People treated with dacarbazine experienced fatigue, nausea, vomiting and neutropenia. A total of 61 people (18%) treated with vemurafenib experienced grade 3 cutaneous squamous‑cell carcinoma, keratocanthoma or both, and were treated with simple excision. Treatment‑related adverse events were recorded for more people who received vemurafenib, which may be explained by the fact that they stayed on treatment longer than those who received dacarbazine (3.1 months for vemurafenib compared with 0.76 months for dacarbazine based on the December 2010 data cut‑off). Adverse events led to dose modification or treatment interruption in 38% of patients in the vemurafenib group (129 of 336 patients) and in 16% of patients receiving dacarbazine (44 of 282 patients). The most common reasons for dose modification were an adverse event or missed cycle. There were more adverse events that led to discontinuation in patients treated with vemurafenib than with dacarbazine (88 compared with 15 patients).
3.10 The manufacturer undertook a systematic literature search but did not identify any economic evaluations of vemurafenib for previously untreated patients with advanced BRAF V600 mutation‑positive metastatic malignant melanoma. Therefore, the manufacturer submitted a de novo 'partitioned survival' economic model in which vemurafenib was compared with dacarbazine. The model comprised 3 health states: progression‑free, progressed disease and death. Hypothetical patients were assumed to enter the model in the progression‑free health state and either remain in that state or progress to a worse health state (that is, progressed disease or death) at the end of each cycle. The model used weekly cycles for a lifetime (30‑year) horizon. The perspective adopted in the economic evaluation was that of the NHS and personal social services, and costs and benefits were discounted at 3.5% per year.
3.11 The proportion of people in each health state in the manufacturer's original model was calculated using progression‑free survival and overall survival data (March 2011 data cut‑off) from the BRIM3 study. The probability of remaining in the progression‑free state was calculated using results observed in the BRIM3 study until month 9 for vemurafenib and month 7 for dacarbazine, after which progression‑free survival for each intervention was extrapolated using exponential functions. Overall survival for patients treated with vemurafenib was estimated directly from the BRIM3 study for the first 9.5 months (March 2011 data cut‑off). A 'stabilised' hazard ratio representing the differences between the vemurafenib and dacarbazine arms up to month 14 was then applied, after which the manufacturer assumed that vemurafenib provided no further treatment benefit (that is, a hazard ratio of 1 was assumed). The estimate of overall survival in the dacarbazine arm was based on 3 different sets of data. The cumulative hazard of overall survival in the BRIM3 study was used directly for 40 weeks (9.2 months), with the longer‑term outcomes up to 46 months derived from a study by Robert et al. (2011), which compared ipilimumab plus dacarbazine with dacarbazine alone in people with previously untreated advanced melanoma. For months 46 and beyond, a long‑term hazard estimate taken from the Surveillance, Epidemiology and End Results (SEER) register was used.
3.12 The manufacturer collected health‑related quality‑of‑life data in the BRIM3 study using the functional assessment of cancer therapy‑melanoma (FACT‑M) questionnaire; however, results were not presented because completion rates were low. Instead, utility values from a study by Beusterien et al. (2009) were used. In this study, standard gamble methods were used to elicit utilities for advanced melanoma health states from members of the general public. These were combined with disutility values associated with adverse events (obtained from Beusterien et al.  and another study by Nafees et al. ). In the manufacturer's base‑case analysis, a utility for progression‑free survival of 0.806 was calculated for people receiving vemurafenib and 0.767 for people receiving dacarbazine. The utility for progressed disease was estimated to be 0.59 based on the study by Beusterien et al. (2009).
3.13 Adverse event rates for vemurafenib and dacarbazine were estimated from the BRIM3 study. The resource costs included in the model were drug acquisition and administration costs, the cost of testing for the BRAF V600 mutation, and the cost of the disease, which included costs related to each health state and of treating adverse events. The average length of a course of treatment with vemurafenib was assumed to be 7 months.
3.14 In the manufacturer's original base‑case analysis (using the March 2011 data cut‑off), the incremental cost‑effectiveness ratio (ICER) for vemurafenib compared with dacarbazine was £56,410 per quality‑adjusted life year (QALY) gained (incremental costs and benefits provided as commercial‑in‑confidence; patient access scheme included). When the October 2011 data cut‑off point was used instead, the ICER increased to £75,489 per QALY gained.
3.15 The manufacturer undertook a series of sensitivity analyses to test the robustness of the results by varying most of the parameters used in the original economic evaluation, including transition probabilities, utilities, costs, discount rate, average age of patients, and BRAF V600 mutation incidence. Taking into account the patient access scheme, the ICERs indicated that vemurafenib was most sensitive to the discount rate (for example, when health benefits were discounted at 1.5% and 0%, the base‑case ICER decreased to £48,249 and £42,054 per QALY gained respectively) and variations to the assumed hazard of death between months 9 and 14. The manufacturer also provided additional scenario analyses that modelled the impact on the ICER of using different utility estimates from Hodi et al. 2010 (which compared ipilimumab plus gp100 with gp100 alone and with ipilimumab alone). When utility values were selected from this study for progression‑free survival (0.80) and progressed disease (0.76) and applied to vemurafenib and dacarbazine in the base case, the ICER fell to £50,052 per QALY gained.
3.16 The ERG considered the BRIM3 study to be well designed and that the clinical‑effectiveness evidence presented by the manufacturer was relevant to the decision problem. The ERG noted that the data from the BRIM3 study demonstrated a statistically significant difference for both overall survival and progression‑free survival for vemurafenib over dacarbazine in patients who had not received previous treatment. It cautioned, however, that the short‑term nature of the results from the BRIM3 study and the heterogeneity of the patient population led to substantial uncertainty when projecting long‑term benefits of treatment.
3.17 The ERG questioned some of the manufacturer's assumptions relating to the costs in the original model, and provided some alternative cost estimates. These included a re‑estimation of costs for dacarbazine therapy based on distributions of body weight and body surface area found in a cohort of UK patients, and the assumption that dacarbazine would be administered as an oncology day case. The ERG also queried long‑term monitoring costs (that is, computed tomography [CT] scan and outpatient visits to an oncologist) for both vemurafenib and dacarbazine with clinical advisers, and found that a programme of 3 to 4 times per year for 2 years, then twice a year for 2 years, and then finally once a year thereafter was more likely than the manufacturer's estimate.
3.18 The ERG acknowledged that the manufacturer adapted an economic model previously used in NICE technology appraisals of cancer drugs in its original submission. It expressed concern that the manufacturer's approach to modelling overall survival was overly elaborate and disagreed with the following methods in the manufacturer's original model:
Survival gains over dacarbazine continued to accrue after vemurafenib treatment was stopped (that is, the vemurafenib group continued to have a lower risk of death) through the application of a hazard ratio estimated from the BRIM3 data to extend the treatment benefit of vemurafenib to 14 months.
The use of a small sample of an arm of the Robert et al. (2011) trial to provide estimates for modelling the outcomes of patients receiving dacarbazine and of those receiving vemurafenib beyond 14 months of survival to 46 months.
Representing the long‑term survival beyond 46 months by a single mortality risk factor parameter calibrated to reconcile data from the study by Robert et al. (2011) with a single value from the SEER database at 10 years (ignoring the SEER hazard profile of more than 1000 patients).
3.19 The ERG explored an alternative approach to modelling overall survival. After examining the Kaplan‑Meier overall survival curves from the BRIM3 study, the ERG proposed that vemurafenib is effective at suppressing disease progression leading to death in the early phase (on average 97 days) but, after a short period, this effect stops and patients revert to the pattern of mortality risk seen in the dacarbazine arm. The ERG suggested that the assumption of a limited window of effectiveness might be supported by the observation that resistance is common with tyrosine kinase inhibitor drugs, reflecting the fact that cancer cells use multiple signalling pathways. The ERG further suggested that there appear to be 2 distinct populations of patients with malignant melanoma: the majority who have a poor prognosis and have a high risk of death within 12 months; and a small group who appear to have good prognosis and can survive for 10 years or more. To address this, the ERG used a simple survival model that included each subgroup split in an unknown ratio and governed by a separate long‑term mortality risk (equivalent to an exponential function). The ERG used a study by Balch et al. (2009) to construct a case‑mix‑adjusted survival curve. This study provided survival curves for each of 4 metastatic melanoma categories (M0: no distant metastases; M1a: distant skin, subcutaneous, or nodal metastases; M1b: metastases to lung; M1c: metastases to all other visceral sites or distant metastases to any site combined with an elevated serum lactate dehydrogenase) based on the American Joint Committee on Cancer Melanoma Staging Database. The ERG constructed the survival curve according to the proportions of patients in the BRIM3 study with each melanoma category (15.9% melanoma stage M0/M1a, 18.8% M1b and 65.3% M1c). The ERG then fitted a 2‑part exponential model to take into account its view of 2 distinct melanoma populations (as described above). The ERG's compound survival model and the BRIM3 case‑mix‑adjusted survival curve showed strong similarities, with the compound survival model indicating that 80.6% of patients would have a mean survival of 11 months (0.91 years) and 19.4% of people with advanced melanoma would have an expected mean survival of more than 12 years (145 months).
3.20 After consultation on preliminary guidance, the manufacturer submitted revised cost‑effectiveness estimates, which incorporated updated survival evidence from the February 2012 data cut‑off of the BRIM3 study. The revised survival estimates were adjusted for patients who switched from dacarbazine to vemurafenib on disease progression (using the rank preserving structural failure time [RPSFT] method). The manufacturer justified using the RPSFT method because it has been previously accepted in a number of NICE technology appraisals. It noted that the RPSFT method did not take into account patients who switched to other BRAF inhibitors (not including vemurafenib) or investigational compounds. Using the RPSFT method, the manufacturer's adjusted estimate of median overall survival in the dacarbazine arm decreased from 10.3 months to 8.9 months (HR 0.64; 5% CI 0.53 to 0.78; p<0.0001). After incorporating the adjusted survival hazard ratio into the model and taking into account the ERG's suggested amendments to discounting, costs and utility for long‑term survivors (see sections 3.15 and 3.17) the manufacturer's revised base‑case ICER was £52,327 per QALY gained.
3.21 After a request from the Committee, the manufacturer provided a full explanation of the assumptions made and parameters used for the RPSFT method, which adjusted the survival estimates for patients who switched from dacarbazine to vemurafenib at disease progression. The manufacturer noted that the RPSFT method attempts to simulate a control arm of people who have not crossed over from dacarbazine to vemurafenib by applying an acceleration factor that 'speeds up' the time for people receiving vemurafenib after disease progression. The manufacturer noted the following key assumptions underlying the RPSFT method:
There is a single underlying acceleration factor associated with the intervention that does not vary with time. The RPSFT method defines an 'average' acceleration factor across all the patients who received the intervention and then applies this to those patients who switched treatments only.
The acceleration factor is valid for patients randomised to the vemurafenib group, and those in the group who switch from dacarbazine to vemurafenib, and also that the treatment is equally effective as a second‑line and as a first‑line treatment.
3.22 The manufacturer stated that the first assumption (the acceleration factor does not vary with time) does not hold for the BRIM3 data. Data from the BRIM3 study suggest that the effect of vemurafenib on mortality is highest within the first few months of treatment and the average acceleration factor of 0.34 estimated from the BRIM3 study was likely to under‑accelerate the survival times of patients who switched from dacarbazine to vemurafenib on disease progression. The manufacturer considered that this would result in a lower calculated survival benefit for vemurafenib than the true benefit. The manufacturer also discussed the plausibility of using alternative approaches to adjust for switching (namely, censoring patients at the point of crossover, inverse probability censoring weighting and the Branson and Whitehead method), but found these methods to be inappropriate in light of the BRIM3 trial data.
3.23 After a request from the Committee, the manufacturer provided a discussion on the use of data from other trials in which no crossover occurred to represent the clinical effectiveness of dacarbazine. The manufacturer compared patient populations in the BRIM3 trial to populations in other trials that included dacarbazine as a comparator. It found some similarity across trials but noted inconsistent reporting of known prognostic factors such as the proportion of patients with elevated lactate dehydrogenase. One trial, Bedikian et al. (2011), evaluated dacarbazine in a malignant melanoma population and had similar patient characteristics in terms of age and stage of disease as the BRIM3 trial. The manufacturer expressed caution about using external data from other trials to model survival in the dacarbazine arm but included an analysis using the Bedikian trial as a sensitivity analysis in its submission (see section 3.25).
3.24 The Committee asked the manufacturer to provide an additional scenario analysis that compared vemurafenib with dacarbazine, in which exponential hazards were applied separately to each arm of the BRIM3 study (using February 2012 data cut‑off) from 14 months onward. The manufacturer declined to provide this scenario analysis. It said that this extrapolation gave a post‑progression survival after treatment with vemurafenib that was 2.2 months shorter than post‑progression survival after dacarbazine, which it considered implausible. It considered that this implausible result may be because of an under‑adjustment of the acceleration factor used in the RPSFT method, and cautioned against using this extrapolation. The manufacturer further justified its decision not to provide the additional scenario analysis on the basis that register data shows that probability of death associated with melanoma reduces over time and an exponential model assumes a constant probability over time. As a result, any calculation based on exponential modelling will have poor external validity.
3.25 The manufacturer provided revised cost‑effectiveness estimates for vemurafenib compared with dacarbazine of £51,757 per QALY gained (using the RPSFT‑adjusted February 2012 data and incorporating the ERG's suggested adjustments), and £44,405 using the Bedikian trial data to represent the dacarbazine arm (including the ERG's adjustments).
3.26 The ERG commented on the manufacturer's additional information about the impact of switching treatment from dacarbazine to vemurafenib after disease progression and responded to the additional scenario analysis. The ERG disagreed with the manufacturer that a 2.2 month shorter post‑progression survival with vemurafenib than dacarbazine was implausible. It commented that vemurafenib may provide only a temporary inhibition to the normal process of disease progression and that overall survival was a more objective outcome than progression‑free survival. It agreed with the manufacturer that the assumption about time invariance was not met in the RPSFT method (that is, BRIM3 shows the treatment effect of vemurafenib changes over time), but disagreed that this would lead to implausible estimates for the scenario analysis that the Committee had requested. The ERG carried out the scenario analysis and reported an ICER of £120,933 per QALY gained.
Details of all the evidence are in the manufacturer's submission and the ERG report, which are available from www.nice.org.uk/guidance/TA269