Enzalutamide for treating metastatic hormone-relapsed prostate cancer before chemotherapy is indicated

Clinical effectiveness

3.1 PREVAIL was a randomised, double‑blind, placebo‑controlled trial comparing enzalutamide 160 mg once daily with placebo in adults with asymptomatic or mildly symptomatic metastatic hormone‑refractory prostate cancer in whom immediate chemotherapy was not yet clinically indicated. In total, 1717 people were randomised ('intention to treat population'); 872 to enzalutamide and 845 to placebo. A total of 1715 patients had at least 1 dose of the study drug ('safety population'); 871 had enzalutamide and 844 had placebo. The study was done at 207 sites in 22 countries; 153 patients were from the UK. People were eligible to participate if they were asymptomatic or mildly symptomatic (that is, had a score of less than 4 on the Brief Pain Inventory [BPI] question 3), had an Eastern Cooperative Oncology Group (ECOG) performance status of 0–1, and had an estimated life expectancy of 6 months or more. The mean age of the study population was 71 years (range 42–93 years). Most people in both arms had an ECOG status of 0 (enzalutamide 67%; placebo 69%).

3.2 The co‑primary endpoints in PREVAIL were overall survival (OS) and radiographic progression‑free survival (PFS). Radiographic PFS (rPFS) was defined as time from randomisation to the first objective evidence of radiographic disease progression, based on imaging review by central (trial) radiologists, or death due to any cause within 168 days of stopping treatment, whichever was first. It was planned that, to show a statistically significant treatment effect, the probability (p) value for OS should be less than 0.049 and the p value for rPFS should be less than 0.001 at their final analyses. The study was powered on target hazard ratios of 0.83 for OS (equal to 80% power, based on 765 deaths), and 0.57 for PFS (>99% power). The company planned 1 ('final') analysis for PFS when 410 patients had evidence of radiographic progression; this was done on 6 May 2012, at which point the disease had progressed in 439 people. The company planned 2 analyses of OS; 1 interim analysis at 516 deaths (two‑thirds of deaths used in sample size calculations) and 1 final analysis (at 765 deaths). The interim analysis for OS was done on 16 September 2013 at which point there had been 540 deaths. To account for the increased risk of false positive results, the statistical plan stipulated that the p value should not exceed 0.012 to be considered statistically significant at the interim analysis for OS. The company did another (post‑'final', post hoc) analysis of rPFS at the same time as the interim OS analysis. After this, the Independent Data Monitoring Committee recommended unblinding the study and allowing people in the placebo arm to switch to enzalutamide. The study was unblinded on 3 December 2013. However, the company continued to follow the patients and presented an analysis of OS done on 30 June 2014.

3.3 Patients remained on the study drug until their disease progressed, which was radiographically confirmed or a skeletal-related event, and then began either cytotoxic chemotherapy or an investigational agent for prostate cancer. After stopping the study drug, people could have docetaxel, hormonal treatments, abiraterone, enzalutamide, cabazitaxel or sipuleucel‑T. The company stated that, in current practice, after disease progression, clinicians would offer cytotoxic chemotherapy. However, more than 25% of patients in the placebo arm and more than 15% of patients in the enzalutamide arm had treatments that would not normally be given to patients at this stage of the treatment pathway in the UK. The company has stated that the number of patients having treatments that are not available at this stage in the UK treatment pathway is academic in confidence and cannot be reported here.

3.4 The company stated that at the first planned analysis for OS in September 2013, 241 people (27.6%) in the enzalutamide arm and 299 people (35.4%) in the placebo arm had died. OS with enzalutamide was longer than with placebo (median 32.4 months and 30.2 months respectively; hazard ratio [HR] 0.706; 95% confidence interval [CI] 0.596 to 0.837; log‑rank test p<0.001). OS was also longer with enzalutamide compared with placebo in the data analysis done in June 2014 after study unblinding (the company has stated that the results of this analysis are academic in confidence and so cannot be published here). The company applied 2 statistical methods to adjust the OS estimates for people switching after their study drug to an active drug that would not be given at this position in the treatment pathway in clinical practice in the UK and which can prolong survival (see section 3.3). These were the inverse probability of censoring weights (IPCW) and a 'two‑stage method'. Applying these adjustments was associated with a larger OS benefit with enzalutamide relative to placebo than seen with the unadjusted estimates; of the 2 methods, the IPCW was associated with the greatest benefits. The company stated that the data are academic in confidence and cannot be reproduced here.

3.5 In the planned final analysis for rPFS (6 May 2012), 118 people (14.2%) randomised to enzalutamide and 321 people (40.1%) randomised to placebo experienced radiographic progression as determined by a central review team (HR 0.186; 95% CI 0.149 to 0.231; log rank p<0.0001). Progression continued to be measured after May 2012 but this was done by a study investigator rather than the central review team. The company did an additional analysis on 16 September 2013 and by this time the disease had progressed in 287 people (44.4%) in the enzalutamide arm and 502 people (59.4%) in the placebo arm (HR 0.307; 95% CI 0.267 to 0.353; log rank p<0.0001).

3.6 In PREVAIL, patients continued treatment with the study drug until:

3.7 The company measured quality of life using the Functional Assessment of Cancer Therapy – Prostate (FACT‑P) and European quality‑of‑life 5‑domain scale (EQ‑5D) questionnaires at baseline and again at weeks 5, 13 and then every 12 weeks until disease progression as defined by radiographic evidence or a skeletal‑related event. These outcomes were exploratory because they had not been specified in the study protocol. People in both the enzalutamide and placebo arms showed a decrease in FACT‑P scores from baseline (meaning a worsening of quality of life). However, the company stated that a 'clinically meaningful deterioration', which it defined as a decrease in FACT‑P score of more than 6 points, was seen only in the placebo group. To estimate a treatment effect for enzalutamide relative to placebo, the company produced a mixed repeated measures model to estimate the change from baseline in utility value (derived from EQ‑5D) in people who remained on treatment. Over the course of the study, the utility value for people taking enzalutamide was 0.02 higher than for people taking placebo.

3.8 The overall incidence of adverse events with enzalutamide and placebo was similar (96.9% compared with 93.2%) across grades. The time on study treatment was longer in the enzalutamide arm than the placebo arm because patients randomised to enzalutamide took longer to have disease progression. There were 279 people (32.0%) in the enzalutamide arm and 226 people (26.8%) in the placebo arm who had a serious adverse event. The overall incidence of adverse events grade 3 or higher was 42.9% in the enzalutamide arm and 37.1% in the placebo arms. The incidence of grade 3 or higher adverse events in the first year of treatment was 32.0% with enzalutamide and 35.1% with placebo. Statistically significantly higher rates of grade 3 or higher hypertension measurements were seen with enzalutamide (6.8% compared with 2.3% for placebo, relative risk (RR) 3.01; 95% CI 1.81 to 5.00). The rate for cataracts was 1.3% in the enzalutamide arm compared with 0.1% in the placebo arm (RR 10.66; 95% CI 1.38 to 82.38). Other grade 3 or higher adverse events that were seen in 0.5% or more people in the enzalutamide arm than in the placebo arm respectively were: nausea 1.0% compared with 0.5%; general physical health deterioration 2.1% compared with 1.2%; pneumonia 1.3% compared with 0.8%; fall 1.4% compared with 0.7%; spinal cord compression 3.8% compared with 2.8%; and syncope 1.6% compared with 0.9%. Forty-nine people (5.6%) taking enzalutamide and 51 (6.0%) taking placebo stopped treatment because of an adverse event. Thirty-seven people (4.2%) in the enzalutamide arm died because of an adverse event compared with 32 (3.8%) in the placebo arm (RR 1.12; 95% CI 0.70 to 1.78).

3.9 There are no head‑to‑head trials comparing enzalutamide with abiraterone. The company therefore compared enzalutamide and abiraterone indirectly using data from PREVAIL and COU‑AA‑302 because both had placebo arms. COU‑AA‑302 was a double‑blind, randomised controlled trial of abiraterone 1000 mg daily plus prednisone 10 mg daily (n=546) compared with placebo plus prednisone 10 mg daily (n=542) in men with metastatic hormone‑relapsed prostate cancer, who were asymptomatic or mildly symptomatic and in whom chemotherapy was not yet clinically indicated. COU‑AA‑302, like PREVAIL, also had a co‑primary endpoint combining OS and rPFS (time from randomisation to the first evidence of radiographic disease progression, progression of soft tissue lesions measured by CT or MRI as defined in modified Response Evaluation Criteria in Solid Tumors criteria or death from any cause, whichever was first).

3.10 As in PREVAIL, COU‑AA‑302 had interim and final analyses, but unlike PREVAIL, it was unblinded early without meeting the pre-specified criterion for a statistically significant difference in OS at an interim analysis. The company used data from the September 2013 cut‑off from PREVAIL (enzalutamide follow‑up 22.2 months; placebo 22.4 months) and from the third analysis of COU‑AA‑302 (planned when 55% of events had been reached; follow-up median 27.1 months) in an indirect treatment comparison using a fixed-effect model. The HRs for OS and rPFS for abiraterone compared with placebo at the third interim analysis in COU‑AA‑302 were 0.79 (95% CI 0.66 to 0.95) and 0.53 (95% CI 0.45 to 0.61) respectively. In its indirect treatment comparison the company assumed that the treatment effect in the control arm of COU‑AA‑302 was the same as that in the control arm of PREVAIL. However, the company noted that the proportion of people taking corticosteroids in the control arm of COU‑AA‑302 (100% taking prednisone) differed from that in PREVAIL (30% taking corticosteroids throughout the trial; 4% of people taking corticosteroids at baseline). The company considered that this may bias an indirect comparison of the 2 trials because of the potential effect of prednisone on the outcomes, but also the extent of prednisone's effect was unknown. The company has stated that the results of its indirect treatment comparison are academic in confidence and cannot be reported here.

3.11 The ERG considered that the PREVAIL population represented the population that would have enzalutamide before chemotherapy in clinical practice in the UK. Clinical advisers to the ERG stated that there were no subgroups of patients in PREVAIL that would have been eligible to start docetaxel at the point that they entered the trial. The ERG stated that both arms of the trial were balanced in terms of demographics, baseline disease characteristics and medical history.

3.12 The ERG noted that the company stated TTD is the most appropriate endpoint to assess disease progression because it is standard practice to stop treatment once progression is diagnosed. The ERG noted that at the September 2013 cut‑off, median TTD was comparable with median time to rPFS. The ERG commented that in the PREVAIL study there were about 2 months between patients stopping treatment with enzalutamide or placebo and starting second‑line treatment. The ERG noted that the company used different data cut‑off results for different variables in its model. The ERG commented that the company had used data up to June 2014 for TTD in its modelling, but that the earlier unblinding of the data in December 2013 might have influenced the decision on whether to continue or stop study treatment.

3.13 The ERG commented that the company considered its indirect treatment comparison was biased because the control groups in PREVAIL and COU‑AA‑302 differed in corticosteroid use. The ERG agreed that the control groups were different, but did not think that comparing the active arms of the 2 trials would give more accurate results. The ERG stated that there was a lack of transparency in reporting the methods the company used to do its indirect treatment comparison, but it checked the results using standard methods (Bucher) and produced similar results to the company.

Cost effectiveness

3.14 The company produced a new Markov model to assess the cost effectiveness of enzalutamide compared with abiraterone or best supportive care in adults with metastatic hormone‑relapsed prostate cancer who were asymptomatic or mildly symptomatic after androgen deprivation therapy failed and in whom chemotherapy was not yet indicated. The company assumed that the placebo arm of PREVAIL represented best supportive care because patients randomised to placebo could have, when needed: luteinising hormone‑releasing hormone analogues, corticosteroids, blood transfusions, bisphosphonates, radiotherapy, analgesics and palliative surgery to treat skeletal-related events. The modelled population had the same characteristics as the PREVAIL population at baseline. The model ran over a lifetime horizon (10 years), and had a cycle length of 1 week with half‑cycle correction. A 3.5% discount rate was applied for health effects and costs.

3.15 The model had 3 main health states: stable disease, progressed disease and death. People entered the model with stable disease after androgen deprivation therapy. Within the progressed health state, there were 3 further health states to reflect that after progressing on enzalutamide, abiraterone or best supportive care, the disease may progress on subsequent treatments. These health states were:

3.16 The company took estimates of survival and TTD from PREVAIL for enzalutamide and best supportive care, and from COU‑AA‑302 for abiraterone. The company used TTD as a proxy for progression for first‑line treatments because it said that this reflected clinical practice. In its base case, to compare the effectiveness of enzalutamide and abiraterone, the company used results from its naive comparison rather than from its indirect treatment comparison. The company used data for TTD and OS for enzalutamide and best supportive care from the 30 June 2014 cut‑off. By this time, the PREVAIL trial had been unblinded for 6 months and less than half of people in both arms had died (the company stated that the exact proportions of people who had died at this time is academic in confidence and cannot be reported here). For abiraterone, the company used OS and rPFS estimates from the third interim analysis from COU‑AA‑302 (55% deaths). Because there were no published TTD data from COU‑AA‑302 the company assumed that rPFS was equivalent to TTD for abiraterone because rPFS and TTD were similar for enzalutamide in PREVAIL. The OS estimates for enzalutamide and best supportive care were adjusted for treatment switching using the IPCW method resulting in an adjusted HR and weighted Kaplan–Meier curves. The company stated that it was not possible to adjust abiraterone OS data for treatment switching.

3.17 To extrapolate the rates of stopping the primary treatment or dying after the end of the trials, the company tested whether the HRs were proportional, and determined they were not. This meant that the company needed to find out which curves had the best fit to data for each treatment arm. The company tested 5 parametric models (exponential, Weibull, log logistic, log normal and generalised gamma) on each of the enzalutamide and placebo arms from PREVAIL and on the abiraterone arm from COU‑AA‑302 to extrapolate the survival curves for OS and TTD. The company considered that the exponential, log-normal and log-logistic curves gave implausible estimates for 5- and 10‑year survival. The Weibull and gamma extrapolation of enzalutamide and best supportive care resulted in curves that crossed. Because the Weibull curve crossed later than the gamma curve, the company selected the Weibull curve in its base case to extrapolate the enzalutamide and best supportive care OS trial data. The company also extrapolated the OS curve for abiraterone using a Weibull distribution. TTD curves for enzalutamide, best supportive care and abiraterone were extrapolated using a gamma distribution.

3.18 The company chose exponential curves to reflect TTD for second- and third‑line treatments. The company estimated the TTD for people having docetaxel from Tannock et al. (2004; TAX 327, a trial of docetaxel with prednisone compared with mitoxantrone with prednisone for advanced hormone‑refractory prostate cancer). The company estimated the TTD for people having third‑line enzalutamide or third‑line abiraterone using the median number of administrations of enzalutamide and abiraterone in AFFIRM and COU‑AA‑301 respectively. AFFIRM and COU‑AA‑301 were placebo-controlled trials of enzalutamide and abiraterone respectively, taken after docetaxel for metastatic hormone‑relapsed prostate cancer.

3.19 To estimate the changes from baseline EQ‑5D score during the trial among people remaining on their first‑line treatment in PREVAIL, the company developed a mixed repeated measures model. The company used the results from this model to determine a baseline utility value (0.844) using UK tariffs for people in the stable disease health state having best supportive care. The company applied an additional utility increment for people having enzalutamide (0.022), from its modelled estimate of a 'treatment effect' of enzalutamide on quality of life from PREVAIL. The treatment effect was the difference between the degree to which quality of life decreased over time with enzalutamide and with placebo. The company assumed that abiraterone would have the same on-treatment utility benefit as enzalutamide.

3.20 As the investigators in PREVAIL collected EQ‑5D only from people on treatment (enzalutamide or placebo before chemotherapy) who by definition did not have progressed disease, the company estimated utility values in the progressed health states from the published literature. The company used a weighted average from 2 publications that had assessed the quality of life of people who were on chemotherapy, who had previously had chemotherapy, and who had metastatic hormone‑relapsed prostate cancer. The company used this to estimate a utility value of 0.658 for post‑progression state 1 (when the disease had progressed on enzalutamide, abiraterone or best supportive care and people were having docetaxel) and 0.612 for post‑progression state 2 (when the disease had progressed on best supportive care and docetaxel and people were having enzalutamide). In line with NICE's technology appraisal on enzalutamide for metastatic hormone-relapsed prostate cancer previously treated with a docetaxel-containing regimen, the company applied an on‑treatment utility gain of 0.04 for enzalutamide after docetaxel in people who had best supportive care before docetaxel. This on‑utility gain for enzalutamide was assumed to be the same as that for abiraterone in an evidence submission the manufacturer of abiraterone had made to the Dutch Healthcare Insurance Board. The company estimated a utility value of 0.500 for people who had palliative care after their disease progressed on active treatment (Sandblom et al. 2004).

3.21 The company incorporated the rates of skeletal-related events seen in PREVAIL for people randomised to enzalutamide or placebo (using data from the September 2013 data cut‑off). The model included the rates of adverse events of grade 3 or higher from PREVAIL and COU‑AA‑302. Adverse events while on docetaxel came from Tannock et al. (2004). The company assumed that the rates of adverse events for third‑line enzalutamide and abiraterone were the same as for first‑line treatment. To estimate the disutility associated with adverse events, the company sourced values from the published literature for adverse events of grade 3 or above. Because no data on the rates of adverse events were available for the period people were taking abiraterone in COU‑AA‑302, the company assumed that these were the same as for enzalutamide. The disutility associated with a skeletal-related event was applied for 1 month and was derived from EQ‑5D data from PREVAIL.

3.22 Both enzalutamide and abiraterone have confidential patient access schemes (price discounts) established when NICE appraised each of the drugs for use after docetaxel. At the request of NICE, the company provided its base‑case results incorporating the list prices for enzalutamide and abiraterone. NICE requested that the ERG provide the results of the company's modelling and its own exploratory analyses, including both the list price and the discounts. The company assumed that the same proportion of people would have corticosteroids plus enzalutamide or best supportive care as in PREVAIL and that all people having abiraterone would also have corticosteroids. The company used the price of docetaxel listed in the electronic market information tool from the Department of Health (£47.30 per 160‑mg infusion vial). The dosing regimen for docetaxel was once every 3 weeks and the modelled administration cost was £301.56 (NHS reference cost).

3.23 The ERG commented that adjusting OS for treatment switching using the IPCW method resulted in reduced estimates for OS compared with the unadjusted results in the placebo arm, but increased estimates for OS compared with the unadjusted results in the enzalutamide arm. This effect was found when using either the September 2013 data cut‑off or the June 2014 data cut‑off, but the difference was greater when using the June 2014 data (as used by the company in its base case). The ERG considered that for OS, it preferred the June 2014 data cut‑off with IPCW adjustment rather than the September 2013 cut‑off because the later data provided more endpoints. The ERG stated that modelled curves had long tails that were not consistent with the trial Kaplan–Meier data and resulted in a large modelled survival gain after most people in the trial had died or were censored, which was not justified.

3.24 The ERG commented that the company had modelled TTD estimates for enzalutamide and best supportive care using PREVAIL data from the June 2014 cut‑off, 6 months after unblinding the study. The ERG considered that unblinding the study might have influenced a clinician's or a patient's decision to stop or continue with treatment. The ERG considered that the choice of curve (gamma) to extrapolate TTD was appropriate, but that using the data from the September 2013 cut‑off was more appropriate for modelling.

3.25 For abiraterone, the ERG noted that in the model the TTD curve (extrapolated with a gamma distribution) crossed the OS curve (extrapolated with a Weibull distribution); this was also seen with enzalutamide, but at a later time point. The ERG noted that this implied that patients die before disease progression. To account for this, the company assumed that after the curves crossed, the time of death reflected the time at which patients stop abiraterone. However, this meant the company could not model subsequent treatments after abiraterone from the point at which the curves crossed. The ERG noted that using a Weibull distribution rather than a gamma distribution to extrapolate the abiraterone TTD curve meant that the curve did not cross over the OS curve. The ERG noted that, although the enzalutamide TTD and OS curves also crossed, this occurred later and had less of an effect on the incremental cost-effectiveness ratio (ICER) estimates than did abiraterone's earlier-crossing curves.

3.26 The ERG commented that in the model, a patient's probability of dying at a particular time point was the same regardless of their health state. The ERG considered this to be implausible because it meant that people with stable, asymptomatic or mildly symptomatic disease on their first treatment had the same risk of dying as people with progressive disease on palliative care after up to 3 lines of active treatment had failed.

3.27 The ERG discussed how the company had modelled the quality‑of‑life data from PREVAIL using the mixed model with repeated measures approach. The ERG stated that the increment for enzalutamide compared with best supportive care (0.022) was based on quality of life decreasing from a baseline of 0.844 with best supportive care (by 0.064), but decreasing less so with enzalutamide (by 0.042). The ERG thought that it would have been more appropriate for the company to apply the decrease in quality of life from an average baseline utility for placebo and enzalutamide rather than adding the utility increment to a baseline value.

3.28 The ERG noted that while the company had modelled quality of life separately for enzalutamide and best supportive care, it had analysed the impact of having skeletal-related events by pooling both treatment arms. Therefore, the impact of skeletal-related events on quality of life might have already been captured in the analysis of quality of life by treatment arm and already reflect any reduction in the rates of skeletal-related events with enzalutamide compared with best supportive care.

3.29 The ERG noted that the company based drug costs on the number of people having the drug at the end, rather than the start, of each cycle. The company assumed that clinicians prescribe enzalutamide and abiraterone weekly, rather than monthly, as implied by the package size. The ERG assumed that clinicians would prescribe a 1‑month course of tablets at a time.

3.30 The ERG noted that the company chose higher monitoring costs for abiraterone (monitoring visits every 4 weeks) than for enzalutamide (monitoring visits every 8 weeks). The ERG noted that the summary of product characteristics for abiraterone stipulates the frequency of monitoring for patients taking abiraterone, but the summary of product characteristics for enzalutamide does not state this. The ERG stated that its clinical experts had advised that the frequency of monitoring of people taking enzalutamide and abiraterone would be expected to be the same. The ERG also noted that monitoring for people having enzalutamide in the company's model was less frequent than for those having best supportive care (every 8 weeks and every 6 weeks respectively). The ERG stated that it expected the monitoring frequency for a person having enzalutamide or best supportive care to be the same.

3.31 The ERG used its preferred assumptions applied to the company's model to produce an ERG exploratory base case:

3.32 Currently, Enzalutamide is available to the NHS for patients after treatment with docetaxel through a patient access scheme (PAS). During the course of the appraisal, the company submitted a revised simple PAS with an increased discount. The company also submitted a revised base case that compared 2 treatment pathways:

3.33 The company's revised base case included the following assumptions incorporating:

3.34 The company's revised base case resulted in an ICER for enzalutamide compared with best supportive care of £27,036 per quality‑adjusted life year (QALY) gained. The company did not present the comparison between enzalutamide and abiraterone with its new PAS because the Committee had concluded in a previous meeting that best supportive care was the key comparator to enzalutamide (section 4.2).

3.35 The one‑way sensitivity analyses that had the greatest effect on the company base case assumed that either no one or everyone in the best supportive care arm goes on to have docetaxel. This increased the ICER to £37,453 or decreased the ICER to £24,361 per QALY gained respectively. The company also presented a scenario in which the utility value for people having enzalutamide after docetaxel was 0.688. This scenario increased the ICER to £28,208 per QALY gained.

3.36 The ERG agreed that it was appropriate to use the existing PAS for calculating the cost of enzalutamide taken after docetaxel in the best supportive care arm of the model (see section 3.32). The ERG noted that company's revised base case included some, but not all, of the ERG's preferred assumptions. It noted that the revised base case did not include:

3.37 The ERG noted that the company had presented new data from AFFIRM to adjust the utility value for post‑docetaxel best supportive care by the decrease in utility seen in the trial. The ERG noted that the uplift in utility the company had applied for enzalutamide was not based on data from the AFFIRM trial, but rather came from a separate source (a submission for abiraterone to the Dutch Healthcare Insurance Board, see section 3.20). The ERG noted that the uplift in utility for enzalutamide calculated from the AFFIRM trial was greater than that used in the company's base case (the exact value of the uplift cannot be reported here because it is commercial in confidence).

3.38 The ERG carried out an additional scenario analysis in which it used the two‑stage method for adjusting OS for subsequent treatments that prolong life, but which is not used in the NHS. Applying the two‑stage method, including the ERG's assumptions on luteinising hormone‑releasing hormone analogue, docetaxel and outpatient costs and monitoring, but keeping the company's assumption on post‑docetaxel utility value resulted in an ICER of £34,759 per QALY gained.