3 The manufacturer's submission

The Appraisal Committee (section 8) considered evidence submitted by the manufacturer of enzalutamide and a review of this submission by the Evidence Review Group (ERG; section 9).

Clinical-effectiveness evidence

3.1 The manufacturer conducted a systematic review of studies evaluating the efficacy and safety of enzalutamide and its comparators for treating metastatic hormone-relapsed prostate cancer that had progressed during or after treatment with docetaxel-containing (cytotoxic) chemotherapy. The manufacturer considered abiraterone and best supportive care to be the relevant comparators for enzalutamide. It excluded mitoxantrone as a comparator, even though mitoxantrone being listed as a comparator in the final scope issued by NICE. This was because: mitoxantrone has a small and diminishing market share in the NHS; there are several new treatments available that have displaced mitoxantrone; existing NICE guidance for prostate cancer does not include recommendations on the use of mitoxantrone; clinical evidence does not support using mitoxantrone after docetaxel therapy; and mitoxantrone does not have a marketing authorisation for metastatic hormone-refractory prostate cancer.

3.2 The key clinical evidence for enzalutamide came from 1 randomised controlled trial, AFFIRM, which compared enzalutamide plus best supportive care with placebo plus best supportive care. The manufacturer considered placebo to be equivalent to best supportive care because people in both treatment groups received best supportive care. Another randomised controlled trial, COU‑AA‑301, compared abiraterone plus prednisone with placebo plus prednisone, and the manufacturer used this trial to compare enzalutamide indirectly with abiraterone using placebo as a common comparator. No relevant observational or single‑arm trials of enzalutamide were identified by the manufacturer.

3.3 AFFIRM was a phase III randomised double‑blind placebo‑controlled study that was conducted at 156 sites in 15 countries, including 12 sites in the UK. Eligible patients were adults with metastatic hormone-relapsed prostate cancer who had previously received 1 or 2 cytotoxic chemotherapy regimens, at least 1 of which contained docetaxel. The AFFIRM study excluded patients who had received abiraterone or treatment with any other investigational agents that block androgen synthesis. The study also excluded patients who had disease progression on ketoconazole, or who were using herbal products that may affect prostate-specific antigen (PSA). Randomisation was stratified by baseline Eastern Cooperative Oncology Group (ECOG) performance status and the average pain burden that patients experienced over the 7 days that preceded randomisation measured using the Brief Pain Inventory-Short Form (BPI‑SF) question 3 score. Investigators randomly assigned patients in a 2:1 ratio to either enzalutamide (160 mg orally once daily) plus best supportive care (n=800) or placebo plus best supportive care (n=399). Best supportive care in AFFIRM could include radiopharmaceuticals, analgesics, bisphosphonates, hormonal therapies, corticosteroids, and radiotherapy (which the trial protocol stipulated should be reported as a skeletal-related event end point as defined in section 3.4). Treatment with the study drug continued until disease progressed and the patient was about to start further systemic therapy, or the patient experienced unacceptable adverse reactions, died or withdrew from the study.

3.4 The primary end point in AFFIRM was overall survival, defined as time from randomisation to death from any cause. In AFFIRM, secondary end points included measures of disease progression and response to treatment. Disease progression was defined as radiographic progression or the occurrence of a skeletal‑related event (either radiotherapy or surgery to bone, pathological bone fracture, spinal cord compression, or change of cancer therapy to treat bone pain). Other secondary end points included:

  • time to PSA progression (time to an increase in PSA of 25% or more, and an increase in absolute PSA of 2 nanograms per millilitre or more above the lowest concentration reached since treatment started)

  • radiographic progression-free survival (time to the earliest objective evidence of radiographic soft tissue or bone progression, or death)

  • time to first skeletal-related event (either radiotherapy or surgery to bone, pathological bone fracture, spinal cord compression, or change of cancer therapy to treat bone pain)

  • rate of pain palliation (a reduction of 2 or more points in BPI‑SF question 3 score at week 13).

    In addition, the manufacturer defined 2 further end points after the trial results had been compiled (see section 3.9).

3.5 Of the 1199 patients randomised in AFFIRM, the median age was 69 years (range 41–92 years), 92.7% were white, and 8.5% had an ECOG performance status of 2 (enzalutamide 8.8%, placebo 8.0%). All patients had received docetaxel therapy, and 27% of the trial population had received further cytotoxic chemotherapy, but none of the patients had received abiraterone. Most patients (91.2%) had bone metastases and 25.6% had visceral lung or liver involvement. The manufacturer considered that the 2 treatment groups were well balanced in terms of demographics, disease characteristics at baseline and medical history.

3.6 AFFIRM was designed to detect, with 90% power, a statistically significant difference in overall survival (2‑sided, at the 0.05 level) with a hazard ratio of 0.76 for enzalutamide compared with placebo. A protocol-defined interim analysis was conducted when 520 patients had died (25 September 2011). Based on the results of this analysis, the data monitoring committee recommended that AFFIRM be terminated. The study remained blinded until 576 patients had died and the database was locked (16 December 2011). All the analyses presented in the manufacturer's submission are based on data at the cut-off date for the interim analysis (25 September 2011), except the analyses of overall survival and adverse events, for which the manufacturer reported results at the cut‑off dates for both the interim analysis and the database lock. The manufacturer censored data for patients who had not reached the end points on the date they were last assessed.

3.7 At the cut-off date for the interim analysis (25 September 2011), the maximum follow-up time for any patient was 24.0 months and the median follow‑up time was 14.4 months. Of patients randomised to enzalutamide and placebo, 308 (38.5%) and 212 (53.1%) respectively had died. Median overall survival was 4.8 months longer for enzalutamide than placebo (enzalutamide 18.4 months, placebo 13.6 months). Treatment with enzalutamide reduced the risk of death by 36.9% compared with placebo (hazard ratio [HR] 0.631, 95% confidence interval [CI] 0.529 to 0.752, p<0.001). At the final cut-off date (16 December 2011), the median duration of follow‑up was 15.0 months, by which time 344 patients (43%) randomised to enzalutamide and 232 patients (58%) randomised to placebo had died, lowering the difference in median overall survival to 4.5 months (enzalutamide 17.8 months, placebo 13.3 months). The relative risk reduction associated with enzalutamide was 38.0% (HR 0.618; 95% CI 0.523 to 0.730, p<0.001). More patients randomised to placebo (61.4%) than enzalutamide (42.0%) stopped study medication and moved to other therapies, and the manufacturer suggested that this may have caused the relative effect of enzalutamide on overall survival to be underestimated. In response to a request for clarification, the manufacturer provided estimates of mean overall survival obtained using different parametric functions, with the analysis truncated at 5 years (this assumed that all remaining patients would die at 5 years from baseline). Enzalutamide was associated with longer mean overall survival than placebo (estimates are designated as commercial in confidence by the manufacturer).

3.8 For the secondary end points, the manufacturer reported the following results:

  • Time to PSA progression: of patients randomised to enzalutamide and placebo, 400 patients (50.0%) and 190 patients (47.6%) met the criteria for PSA progression respectively. Median time to PSA progression was 5.3 months longer for enzalutamide than placebo (enzalutamide 8.3 months, placebo 3.0 months), and enzalutamide delayed time to PSA progression (HR 0.248; 95% CI 0.204 to 0.303, p<0.001).

  • Radiographic progression-free survival: enzalutamide prolonged median radiographic progression-free survival by 5.4 months compared with placebo (enzalutamide 8.3 months, placebo 2.9 months), and decreased the risk of radiographic disease progression (HR 0.40; 95% CI 0.35 to 0.47, p<0.001).

  • Time to first skeletal-related event: 35.9% and 40.3% of patients randomised to enzalutamide and placebo, respectively, experienced a skeletal-related event. Median time to first skeletal-related event was 3.4 months longer with enzalutamide than with placebo (enzalutamide 16.7 months, placebo 13.3 months), and enzalutamide delayed time to first skeletal-related event (HR 0.688; 95% CI 0.566 to 0.835, p<0.0001).

  • Rate of pain palliation at week 13: data were available for 68.1% and 58.1% of patients in the enzalutamide and placebo groups respectively. These showed that enzalutamide reduced pain in more patients than placebo (enzalutamide 25%, placebo 14.2%, p<0.001).

3.9 The manufacturer performed 2 post‑hoc analyses (after the trial results had been compiled) on a modified definition of progression-free survival and on time to treatment discontinuation. Modified progression-free survival was defined as the time to radiographic progression, first skeletal-related event or death, whichever occurred first. The median modified progression-free survival was 8.11 months for enzalutamide and 2.79 months for placebo, with a hazard ratio of 0.46 (95% CI 0.40 to 0.53, p<0.001). For time to treatment discontinuation, patients in AFFIRM stopped treatment when investigators confirmed disease progression (radiographic disease progression or a skeletal-related event occurred) and the patient was about to start further systemic therapy, or if the patient experienced an adverse event. The manufacturer reported a median time on study treatment of 8.3 months for enzalutamide and 3.0 months for placebo, a difference of 5.3 months. The proportion of patients who remained on treatment for 12 months or more was higher for enzalutamide than placebo (24.8% and 4.5% respectively), consistent with a hazard ratio of 0.34 (95% CI 0.30 to 0.39, p<0.001).

3.10 In AFFIRM, pre‑specified subgroup analyses were performed by prognostic factors for prostate cancer (including patients who had received 1 compared with 2 or more courses of cytotoxic chemotherapy), demographic characteristics and region. Although it did not report interaction tests for the subgroup analyses, the manufacturer stated that the treatment effect of enzalutamide was consistent across all subgroups, with a median overall survival consistently exceeding placebo by more than 3 months. For patients who had received 1 course of cytotoxic chemotherapy (docetaxel), the median time to death for patients randomised to enzalutamide was not reached, and for patients randomised to placebo, it was 14.2 months (HR 0.59; 95% CI 0.48 to 0.73). The difference in median overall survival for patients who had received 2 or more courses of cytotoxic chemotherapy (including docetaxel) was 3.6 months longer with enzalutamide than placebo (HR 0.74; 95% CI 0.54 to 1.03).

3.11 The manufacturer did not identify any published head-to-head trials comparing enzalutamide with abiraterone. It identified the COU‑AA‑301 trial, which compared abiraterone plus prednisone with placebo plus prednisone, and used it to compare enzalutamide with abiraterone indirectly, using placebo as a common comparator. COU‑AA‑301 was a phase III randomised controlled trial evaluating the efficacy and safety of abiraterone in patients with metastatic hormone-relapsed prostate cancer whose disease had progressed during or after treatment with up to 2 regimens of cytotoxic chemotherapy, 1 of which contained docetaxel. Patients in both groups of COU‑AA‑301 received best supportive care, defined as radiotherapy, bisphosphonates, analgesics and luteinising hormone-releasing hormone agonists, as needed. The primary end point of the trial was overall survival, and the secondary end points included those in AFFIRM. The hazard ratios for abiraterone compared with placebo were reported from COU‑AA‑301 as 0.74 (95% CI 0.64 to 0.86) for overall survival and 0.66 (95% CI 0.58 to 0.76) for radiographic progression-free survival. These were derived from data at the end of the trial when 775 deaths had occurred. At that point, the median follow-up was 20.2 months (Fizazi et al. 2012) compared with 15.0 months in AFFIRM.

3.12 In the manufacturer's opinion, the designs of AFFIRM and COU‑AA‑301 were comparable. However, because abiraterone is taken with corticosteroids but enzalutamide is not, different proportions of patients in the 2 placebo groups used prednisone (45.6% in AFFIRM and 100% in COU‑AA‑301). The manufacturer stated that because there is no evidence to suggest that corticosteroids affect overall survival or progression-free survival, it assumed that differences in corticosteroid use would not affect treatment outcomes. In addition, modified progression-free survival was defined more broadly in COU‑AA‑301 than in AFFIRM, but the manufacturer still considered an indirect comparison between enzalutamide and abiraterone to be possible. The manufacturer performed the comparison using the Bucher method for the following end points: overall survival, radiographic progression-free survival, modified progression-free survival, time to first skeletal-related event, objective response rate (that is, the proportion of patients with a complete or partial radiographic response), PSA response and adverse events.

3.13 The manufacturer expressed the results of the indirect comparison as hazard ratios for overall survival, progression‑free survival and time to first skeletal‑related event, and as odds ratios for objective response rate and PSA response, with hazard ratios less than 1.00 and odds ratios greater than 1.00 favouring enzalutamide. The manufacturer reported a statistically significant difference (p<0.05) in favour of enzalutamide for radiographic progression-free survival and modified progression-free survival, with hazard ratios slightly higher than those reported for enzalutamide compared with best supportive care. The differences between enzalutamide and abiraterone for overall survival and time to first skeletal-related event were not statistically significant (hazard ratios and confidence intervals are academic in confidence). The manufacturer stated that the hazard ratio for overall survival, although not statistically significant, should be interpreted with caution because there is evidence that the hazard ratio for abiraterone compared with placebo varied over time (the treatment effect was not constant), which may have caused the relative treatment benefit of abiraterone to be overestimated. For the objective response rate and PSA response end points, enzalutamide was associated with a statistically significant difference for PSA response only (odds ratios and confidence intervals are academic in confidence).

3.14 For health-related quality of life, investigators collected data during AFFIRM using the EQ‑5D and Functional Assessment of Cancer Therapy‑Prostate (FACT‑P) questionnaires. However, EQ‑5D utility data were collected only at study sites in France, Germany, Italy, Spain and the UK. In total, 179 patients provided EQ‑5D data (15.8% and 13.3% of patients randomised to enzalutamide and placebo respectively). FACT‑P is a 39‑item questionnaire about physical, social, emotional and functional wellbeing. Each item is scaled 0 to 4 (the higher the score the better), and the FACT‑P score is the sum across items. Patients achieved a quality‑of‑life response in AFFIRM if their FACT‑P scores improved by 10 points compared with baseline on 2 consecutive measurements, at least 3 weeks apart. Of patients evaluated using FACT‑P (enzalutamide 81.5%, placebo 62.2%), 42.2% and 14.5% had a quality‑of‑life response in the enzalutamide and placebo groups respectively (p=0.0018).

3.15 Investigators assessed the safety and tolerability of enzalutamide throughout AFFIRM, and collected data on adverse events and serious adverse events for 30 days after the patient's last dose of study treatment or until they started another systemic cancer therapy, if sooner. The manufacturer analysed adverse events for a 'safety population', defined as patients who had taken at least 1 dose of the study drug. The rates of adverse events were similar in the 2 groups. The incidence of adverse events of grade 3 severity or above was lower in the enzalutamide group (45.3% compared with 53.1% in the placebo group), with a median time to the first grade 3 or above adverse event of 12.6 months in the enzalutamide group compared with 4.2 months in the placebo group. The incidence of all grades of fatigue, diarrhoea, hot flushes, musculoskeletal pain and headache was higher in the enzalutamide group than in the placebo group. There were no clinically significant between-group imbalances in the rates of other adverse events, such as high blood glucose concentration, glucose intolerance, weight gain or high lipid concentrations. Abnormal findings in liver function tests were reported as adverse events in 1% of patients receiving enzalutamide and in 2% of patients receiving placebo. By April 2012, 10 patients (0.9%) who received enzalutamide had had a seizure; no seizures were reported in patients who received placebo. Overall, the manufacturer considered that the safety profile of enzalutamide was acceptable and that adverse events could be managed.

3.16 In its indirect comparison between enzalutamide and abiraterone, the manufacturer compared the incidence of adverse events, skeletal-related events and tolerability. The manufacturer stated that the results of the indirect comparison showed that enzalutamide had a similar safety profile to abiraterone. Enzalutamide was associated with a decreased risk of lowering serum potassium concentrations compared with abiraterone. In contrast, the risk of bone pain was higher with enzalutamide. No statistically significant differences were found for other adverse events, skeletal-related events or the rate of treatment discontinuation (used by the manufacturer as a proxy for tolerability) between enzalutamide and abiraterone.

3.17 In response to the second appraisal consultation document (which did not recommend for the use of enzalutamide after abiraterone), the manufacturer provided 12 retrospective studies on the effectiveness of enzalutamide in patients who had received previous treatment with abiraterone. All studies were observational record reviews or audits, and included between 23 and 150 patients who had previously received abiraterone. Outcomes varied across the studies, but in 9 studies reporting change in PSA, 10‑46% of patients had a 50% or more decline in PSA from baseline after treatment with enzalutamide. In the largest study, 39% (58/150) of patients who had previously received abiraterone had a 30% or more decline in PSA from baseline after treatment with enzalutamide. This was less of a decline than that seen in patients who had not previously received abiraterone (55% [18/33] decline in PSA). The manufacturer interpreted these studies as demonstrating that a meaningful proportion of patients benefit from treatment with enzalutamide after abiraterone.

Evidence Review Group critique

3.18 The ERG considered that the manufacturer's systematic review was adequate. It noted that, although there were several randomised controlled trials that evaluated mitoxantrone in the population in the decision problem, none of these compared enzalutamide with mitoxantrone, and none allowed an indirect comparison, because the 2 treatments did not have a common comparator. The ERG therefore considered that the manufacturer was justified in excluding mitoxantrone as a comparator.

3.19 For the indirect comparison, the ERG agreed that patients in AFFIRM and COU‑AA‑301 were generally comparable, but noted slight differences in baseline characteristics:

  • COU‑AA‑301 had more patients aged 75 years or older (27.7% compared with 25.3% in AFFIRM), and more patients with an ECOG performance status of 2 (10.6% compared with 8.5% in AFFIRM).

  • More patients had received 1 or 3 previous regimens in AFFIRM (73.0% and 2.8% compared with 69.7% and 0.0% in COU‑AA‑301), whereas more patients had received 2 previous regimens in COU‑AA‑301 (30.3% compared with 24.3% in AFFIRM).

  • More patients had previously had surgery for cancer in AFFIRM (64.6% compared with 52.1% in COU‑AA‑301).

    In addition, the ERG questioned whether corticosteroids affect pain, quality of life and adverse events. In the manufacturer's response to a request for clarification, the manufacturer indicated that corticosteroids may lessen bone pain associated with disease metastases, improve appetite and reduce weight loss, but that they also cause debilitating adverse reactions. Because the manufacturer could not quantify the effect of corticosteroids, it assumed that their positive and negative effects cancel each other out. Overall, the ERG considered that the manufacturer's indirect comparison of enzalutamide and abiraterone was appropriate and performed correctly.

3.20 The ERG reviewed the 12 retrospective studies provided by the manufacturer in response to the second appraisal consultation document, and considered that the evidence presented was not suitable for assessing enzalutamide's effectiveness when given after abiraterone. Specifically, the ERG noted that the studies prohibited a 'robust analysis' because they all had small patient numbers, included no comparator, provided short periods of follow-up, and presented little data on overall survival or progression-free survival. Using data reported in the abstract of the largest record-review study provided by the manufacturer, the ERG noted that patients treated with enzalutamide (but who had not previously received abiraterone) were 2.3 times more likely to have at least a 50% decrease in their PSA concentration than patients treated with enzalutamide after abiraterone. Overall, the ERG considered that there was insufficient evidence to establish how clinically effective enzalutamide was after treatment with abiraterone.

3.21 The ERG noted that the adverse events associated with enzalutamide were similar to those with abiraterone, and less debilitating than the bone marrow suppression, diarrhoea, physical weakness, hair loss and other adverse events experienced with cytotoxic agents such as mitoxantrone. The ERG viewed enzalutamide as generally safe, with relatively mild adverse events.

Cost-effectiveness evidence

3.22 The manufacturer submitted a de novo economic model to estimate the cost effectiveness of enzalutamide, compared with abiraterone and with best supportive care, in patients with metastatic hormone-relapsed prostate cancer whose disease had progressed during or after docetaxel therapy. The manufacturer conducted the analysis from the perspective of the NHS and personal social services and chose a time horizon of 10 years. The cycle length of the model was 3 weeks, in line with previous models for this indication, and the manufacturer applied a half‑cycle correction except for direct drug costs. Costs and health effects were discounted at an annual rate of 3.5%.

3.23 The manufacturer's model was a state‑transition Markov cohort model simulating 3 states: stable disease, progressive disease and death. All patients entered the model in the stable‑disease state and received enzalutamide, abiraterone or best supportive care. They could then remain in this state, move to the progressive‑disease state or die. Once they moved to a different state in the model, patients could not return to their previous state. Patients who moved to the progressive‑disease state stopped treatment and received palliative care. The manufacturer modelled adverse events (grade 3 and above in AFFIRM) in the stable‑disease state (assuming that only patients receiving treatment could have an adverse event), and skeletal‑related events (spinal cord compression, pathological non‑vertebral bone fracture, radiation to the bone and bone surgery) in the progressive-disease state (assuming that skeletal‑related events result only from disease progression in the bones).

3.24 The manufacturer used clinical data for enzalutamide and best supportive care from AFFIRM, and data for abiraterone from COU‑AA‑301 (Fizazi et al. 2012). To estimate overall survival and progression-free survival, the manufacturer chose best supportive care as a 'reference' treatment. It then fitted alternative parametric functions (exponential, Weibull, log-logistic, log-normal and gamma) to patient‑level data for the placebo group in AFFIRM (representing best supportive care) and extrapolated the curves beyond the end of the trial, using the final cut‑off date (16 December 2011) for overall survival, and the interim analysis cut-off date (25 September 2011) for progression‑free survival. The manufacturer chose the base‑case survival functions for best supportive care based on statistical tests, visually inspecting the curves' fit to the data, and the clinical plausibility of the extrapolation as judged by its clinical experts. The survival functions for enzalutamide and abiraterone were then estimated by applying a hazard ratio derived from their respective trials to the survival function for best supportive care. The manufacturer used these survival functions to establish the transition between states in the model for each treatment.

3.25 The manufacturer chose the Weibull function to estimate overall survival for best supportive care (the 'reference' treatment) because, according to its clinical experts, the shape of its 'tail' was more realistic than other functions, even though the log-logistic function fitted the data best according to statistical tests. The Weibull function produced a median overall survival of 12.96 months for best supportive care. The manufacturer stated that, unlike the hazard ratio for abiraterone compared with placebo, the hazard ratio for enzalutamide compared with placebo remained constant over time (that is, meeting the proportional hazards assumption). Therefore, the manufacturer derived the survival function for enzalutamide by applying the hazard ratio of 0.618 from AFFIRM to the best supportive care function. In contrast, the manufacturer stated that using a constant hazard ratio may overestimate the survival benefit of abiraterone. This was because, over the course of follow-up in COU‑AA‑301, the hazard ratio for abiraterone compared with placebo varied and the Kaplan–Meier curves separated and then converged, implying that the benefit of abiraterone decreased over time. For this reason, the manufacturer fitted a Cox proportional hazards model including a time-dependent covariate, which it chose over piecewise modelling, to capture the changing effect of abiraterone over time. This approach resulted in hazard ratios starting at 0.52 and increasing over time (treatment effect decreasing) to 1.0 at 16.6 months and 1.39 at 25 months. Because no data were published beyond 25 months, the manufacturer assumed that the hazard ratio beyond 25 months remained at 1.39.

3.26 To derive the curves for progression‑free survival, the manufacturer did not use either radiographic progression‑free survival (a secondary end point in AFFIRM) or modified progression‑free survival (defined post hoc) because it considered that neither measure accurately reflected how disease progression would be defined in clinical practice. The manufacturer believed that, in clinical practice, patients would not stop treatment based on a single measure of disease progression (for example, radiographic disease progression). It therefore used time to treatment discontinuation (defined post hoc) as a proxy for progression‑free survival in its base‑case analysis, noting that the criteria for stopping treatment were the same in AFFIRM and COU‑AA‑301 (the manufacturer used the definition of modified progression‑free survival instead of time to treatment discontinuation in a scenario analysis). The manufacturer chose the Weibull function for best supportive care (estimating a median time to treatment discontinuation of 3.71 months), and derived the survival function for enzalutamide by applying enzalutamide's hazard ratio for treatment discontinuation of 0.34. To estimate a hazard ratio for abiraterone, the manufacturer used data from the manufacturer's submission for abiraterone from Abiraterone for castration-resistant metastatic prostate cancer previously treated with a docetaxel-containing regimen (NICE technology appraisal guidance 259). However, these data related to the subgroup of patients who had received only 1 cytotoxic chemotherapy regimen (which included docetaxel). The manufacturer estimated a hazard ratio of 0.52 (95% CI 0.44 to 0.60) for abiraterone, and used this ratio to derive the progression‑free survival function for abiraterone.

3.27 Adverse events in the model (applied in the stable‑disease state) comprised all grade 3, 4 or 5 adverse events that occurred in more than 1% of patients in the enzalutamide or placebo groups of AFFIRM, together with seizures. Skeletal‑related events in the model (applied in the progressive‑disease state) were spinal cord compression, pathological non‑vertebral bone fractures, radiation to the bone or surgery to the bone. The manufacturer noted that a change in cancer therapy to treat bone pain was a skeletal‑related event in AFFIRM, but it did not consider it as such in the model.

3.28 Because a limited number of patients completed the EQ‑5D questionnaire, the manufacturer used a mapping algorithm to transform the FACT‑P responses into EQ‑5D data. However, because mapping utility values introduces uncertainty, the manufacturer used the EQ‑5D data collected from AFFIRM in the model, and applied the mapped values only when it considered the number of EQ‑5D responses insufficient. The manufacturer derived the EQ‑5D utility values by applying UK valuations of health states estimated using the time trade‑off method. For the baseline utility value in the model, the manufacturer presented the EQ‑5D value collected from the trial and the mapped value (both are academic in confidence), but used the former because it considered the sample size to be reasonable (n=209). To capture the benefit of treatment with enzalutamide or abiraterone, the manufacturer applied a value reflecting the increase in utility above the baseline utility value. For patients receiving enzalutamide, the manufacturer estimated a mapped utility increase (academic in confidence) and, for patients receiving abiraterone, it used a value of 0.04 (95% CI 0.032 to 0.048) from the manufacturer's submission for abiraterone to the Dutch Health Care Insurance Board.

3.29 For the decrease in utility associated with disease progression, the manufacturer derived a value from a published study, Sandblom et al. (2004), because no health‑related quality‑of‑life data were collected in AFFIRM after a patient's disease had progressed. Sandblom et al. measured the health-related quality of life of patients with metastatic hormone‑relapsed prostate cancer at the end of life, and reported EQ‑5D utility data for 2 time intervals: 16–8 months before death and 8–0 months before death. The manufacturer stated that the duration of these intervals approximated the time patients in AFFIRM spent in the stable- and progressive‑disease states. The utility decrease used by the manufacturer for the progressive‑disease state was −0.085, reflecting the decrease in utility from 16–8 months before death to 8–0 months before death in the Sandblom et al. study. Although published studies show that utility decreases in the last months of life, the manufacturer did not model this and instead applied a constant utility for the progressive‑disease state. The manufacturer justified this approach because it did not identify reliable data to reflect this utility decrease, and because it believed that health would deteriorate equally among patients on different treatments. The manufacturer sourced the utility decreases associated with adverse events from the published literature, and for skeletal‑related events, it mapped values from FACT‑P to EQ‑5D because it considered that there were insufficient EQ‑5D data for skeletal‑related events from AFFIRM.

3.30 The manufacturer included the following costs in the model: intervention and comparators' costs, resource use costs, and costs of adverse events and skeletal-related events. The price of enzalutamide in the model took into account the patient access scheme discount for enzalutamide, and the manufacturer assumed the patient access scheme discount for abiraterone because the discount was commercial in confidence. The manufacturer applied the resource use costs associated with monitoring patients based on the abiraterone submission for NICE technology appraisal guidance 259. It used NHS reference costs 2011/12 for diagnostic imaging and tests, palliative care, and the management of adverse events and skeletal-related events; the average monitoring cost per 3‑week cycle in the model was £70.90. The frequency of concomitant medications in the model was based on AFFIRM. For drug costs, the manufacturer used the 'British national formulary' (BNF) and the Electronic Market Information Tool (EMIT). Because resource use and associated costs usually peak immediately before death, the manufacturer incorporated a terminal care cost of £3133 per patient as a transition cost to the death state. The manufacturer assumed that all hospitalisation costs would be captured by the costs of adverse events, skeletal-related events and terminal care.

3.31 The manufacturer's deterministic cost‑effectiveness results estimated that enzalutamide provides additional quality-adjusted life years (QALYs) at an additional cost to abiraterone (values are commercial in confidence), resulting in an incremental cost-effectiveness ratio (ICER) of £14,795 per QALY gained for enzalutamide compared with abiraterone. Compared with best supportive care, the ICER for enzalutamide was £43,587 per QALY gained (costs and QALYs are commercial in confidence). In an incremental analysis, abiraterone was extendedly dominated by enzalutamide (that is, a QALY is attained at a higher cost with abiraterone than with enzalutamide because the ICER for abiraterone compared with best supportive care [£102,751 per QALY gained] is higher than that for enzalutamide compared with best supportive care).

3.32 The manufacturer presented deterministic sensitivity analyses in which it varied all the parameters in the model, one at a time. Most parameters were varied to the lower and upper limits of their 95% confidence intervals. In comparing enzalutamide with abiraterone, the ICERs were most sensitive to the hazard ratio used to estimate overall survival for abiraterone. When varied, this resulted in ICERs ranging from £11,843 to £46,022 per QALY gained for enzalutamide compared with abiraterone. The manufacturer advised caution when interpreting these results because 2 parameters (the intercept and the time covariate) were correlated in the hazard ratio for abiraterone and, in the sensitivity analysis, the manufacturer varied only 1 parameter (the intercept), which it suggested may cause the ICERs to be overestimated. For the comparison of enzalutamide with best supportive care, the ICER was most sensitive to enzalutamide's hazard ratio for overall survival (0.618), with ICERs ranging from £34,692 to £58,042 per QALY gained when this was varied. The manufacturer stated that, compared with abiraterone or best supportive care, the cost effectiveness of enzalutamide was most sensitive to the hazard ratios for overall survival and treatment discontinuation, and the degree to which health‑related quality of life improved while taking enzalutamide or abiraterone (that is, the degree to which utility increased 'on-treatment').

3.33 The manufacturer carried out a probabilistic sensitivity analysis to characterise the uncertainty in the base-case ICER, varying parameters simultaneously with values from a probability distribution. The probabilistic ICERs were £14,576 per QALY gained for enzalutamide compared with abiraterone and £43,239 per QALY gained for enzalutamide compared with best supportive care. For enzalutamide compared with abiraterone, there was an 83% probability of enzalutamide being cost effective if the maximum acceptable ICER was £20,000 per QALY gained, and a 98% probability of it being cost effective if the maximum acceptable ICER was £30,000 per QALY gained. Compared with best supportive care, the probability of enzalutamide being cost effective at a maximum acceptable ICER of £30,000 per QALY gained was around 0% (numerical value not provided in the manufacturer's submission), and the probability of it being cost effective at a maximum acceptable ICER of £50,000 per QALY gained was 81%.

3.34 To investigate the structural uncertainty in the model, the manufacturer tested alternative assumptions related to the following in scenario analyses:

  • the parametric function it used to fit the Kaplan–Meier data for overall survival and progression-free survival

  • the overall survival function for abiraterone that it had estimated from time‑dependent hazard ratios

  • how it defined disease progression

  • the utility values it derived for the model

  • the potential that it had double counted the utility decreases for all adverse events and skeletal‑related events

  • the patient access scheme discount for abiraterone.

    The key scenarios are summarised in sections 3.35–3.37.

3.35 In its base case, the manufacturer estimated overall survival for abiraterone by applying a time‑dependent hazard ratio to the survival function for best supportive care (the 'reference' treatment). The manufacturer explored the sensitivity of the model to this method in 2 scenarios:

  • applying the constant hazard ratio of 0.74 from COU‑AA‑301 (Fizazi et al. 2012) to the survival function for best supportive care

  • estimating mean overall survival for abiraterone from COU‑AA‑301, then indirectly comparing this estimate with the mean overall survival for enzalutamide from AFFIRM to derive the model inputs.

    Compared with a base-case ICER of £14,795 per QALY gained for enzalutamide compared with abiraterone, the ICERs from these 2 scenarios were £19,972 and £18,034 per QALY gained respectively.

3.36 In the base‑case model, the manufacturer applied a utility increase to the baseline utility value to reflect the treatment benefit of enzalutamide and abiraterone. The manufacturer investigated the impact of this utility increase:

  • by excluding it from the model

  • by assuming an equal utility increase for enzalutamide and abiraterone.

    For the first scenario, the base‑case ICER increased from £14,795 per QALY gained to £16,720 per QALY gained for enzalutamide compared with abiraterone, and from £43,587 per QALY gained to £51,343 per QALY gained for enzalutamide compared with best supportive care. For the second scenario, the ICERs were £15,652 and £43,587 per QALY gained for enzalutamide compared with abiraterone and best supportive care respectively.

3.37 The manufacturer investigated the effect of using modified progression‑free survival instead of time to treatment discontinuation as a proxy for progression‑free survival. This decreased the ICERs for enzalutamide compared with abiraterone and with best supportive care to £13,476 and £43,396 per QALY gained respectively.

3.38 The manufacturer stated that some clinical experts believed that enzalutamide and abiraterone have the same clinical effect, and that the superiority of enzalutamide demonstrated in its indirect comparison may have reflected differences in the populations of AFFIRM and COU‑AA‑301. Because of this, the manufacturer performed a cost‑minimisation analysis, which assumed equal utility gain and equal rates of adverse events and skeletal-related events for enzalutamide and abiraterone. In this analysis, the acquisition cost of both treatments was set to be equal, but because abiraterone is taken with corticosteroids, and patients on abiraterone need more monitoring, the manufacturer concluded that enzalutamide was less costly and as effective as abiraterone (the manufacturer estimated cost savings of £1007 for enzalutamide compared with abiraterone).

3.39 In response to the first appraisal consultation document, which did not include a recommendation on the use of enzalutamide for patients who had received 2 or more courses of cytotoxic chemotherapy, the manufacturer provided a subgroup analysis comparing enzalutamide with best supportive care for these patients. The median survival in the placebo group was 12.8 months. In this analysis, the manufacturer estimated a hazard ratio for overall survival of 0.66 (95% CI 0.48 to 0.90), which differed from the hazard ratio presented in its original submission (0.74, 95% CI 0.54 to 1.03). The manufacturer stated that this was because the hazard ratio of 0.66 was based on a stratified analysis at the final cut-off date (16 December 2011) whereas the hazard ratio in the original submission was based on an unstratified analysis at the cut-off date for the interim analysis (25 September 2011). Using the increase in utility assumed in the base case for patients on enzalutamide (academic in confidence), the manufacturer estimated an ICER for enzalutamide compared with best supportive care of £45,509 per QALY gained.

Evidence Review Group critique and exploratory analyses

3.40 The ERG considered that the manufacturer's literature review of the cost‑effectiveness evidence was appropriate. The ERG noted that the economic model submitted by the manufacturer was consistent with the structure of other models for the same disease area and used a lifetime time horizon.

3.41 The ERG agreed that it was appropriate for the manufacturer to have chosen the Weibull function for overall survival over the log‑logistic function (which fitted the data best according to statistical tests) because the latter predicted an implausibly high proportion of patients alive at 5 and 10 years (at 5 years: log‑logistic 4.2%, Weibull 0.0%; at 10 years: log-logistic 1.1%, Weibull 0.0%).

3.42 To model overall survival for abiraterone, the manufacturer estimated hazard ratios starting at 0.52 and increasing over time to 1.39 at 25 months (implying that patients receiving or who had received abiraterone have a higher risk of death than those receiving or who had received placebo), and assumed that the hazard ratio beyond 25 months remains 1.39. In the ERG's opinion, using a hazard ratio of 1.39 beyond 25 months was not justified, given the very low number of patients still in the study at 25 months. Instead, the ERG argued that a conservative approach should be adopted by assuming a hazard ratio of 1 beyond 25 months (patients receiving or having received abiraterone or placebo have the same risk of death). The ERG applied this assumption in its exploratory analyses (see section 3.47).

3.43 To derive the progression-free survival function for abiraterone, the manufacturer estimated a hazard ratio (0.52) using data related to the subgroup of patients who had received only 1 previous cytotoxic chemotherapy regimen. The ERG argued that estimating the hazard ratio from the overall population, which included patients who had received more than 1 cytotoxic chemotherapy regimen, was more appropriate and consistent with how the manufacturer estimated overall survival for abiraterone. The ERG therefore preferred a hazard ratio of 0.49 (95% CI 0.37 to 0.63), approximated from the ratio of median values for the whole population reported from COU‑AA‑301 (Fizazi et al. 2012) to model progression‑free survival. The ERG applied this hazard ratio in its exploratory analyses (see section 3.47).

3.44 The ERG agreed that, given the uncertainty around mapping utility values, it was appropriate for the manufacturer to have chosen the baseline EQ‑5D utility value collected from AFFIRM. However, the ERG regarded this value as uncertain, because only a small proportion of patients completed the EQ‑5D questionnaire (209 out of 1199 patients) and of those, a considerable proportion had missing values. In addition, the ERG stated that patients in AFFIRM may have been fitter than patients seen in clinical practice, which may have caused the utility value to be overestimated. The ERG explored the uncertainty around this parameter in a threshold analysis (see section 3.49).

3.45 The ERG could not verify the internal mapping algorithm or the details of the calculations that the manufacturer used to estimate the utility increase for patients taking enzalutamide. In addition, it considered that there was no evidence to support an increase in utility with abiraterone, and so this, too, was highly uncertain. The ERG stated that the difference between the utility increases for enzalutamide and abiraterone is an important determinant of the incremental QALYs of enzalutamide and, in its opinion, there is no strong evidence to justify assuming different utility increases for enzalutamide and abiraterone. Furthermore, the ERG considered that incorporating a utility increase for treatment in the model introduces a risk of double counting and overestimating the benefit of treatment, because the utility decreases for adverse events and skeletal-related events already capture part of this benefit. Overall, the ERG preferred taking a conservative approach and excluding the utility increases for enzalutamide and abiraterone from the model. The ERG used this assumption in its exploratory analyses (see section 3.47).

3.46 The manufacturer derived the decrease in utility associated with disease progression from Sandblom et al. (2004), because the duration of the time intervals of 16–8 months and 8–0 months before death in this study approximated the time patients in AFFIRM spent in the stable- and progressive-disease states. The ERG argued that this may bias the utility value, because some patients assessed in Sandblom et al. may have had stable disease rather than progressed disease 8–0 months before death, and others may have had progressive disease rather than stable disease 16–8 months before death. The manufacturer stated that, although Sandblom et al. does not report the time patients spent in each state, patients are likely to have progressive disease in their last 8 months of life. The ERG considered the manufacturer's utility value from Sandblom et al. (−0.085) to be appropriate. It also explored using an alternative value of −0.07 from Sullivan et al. (2007), which was favoured by the ERG for NICE technology appraisal guidance 259. This increased the base‑case ICERs for enzalutamide compared with abiraterone by approximately £150 per QALY gained, and for enzalutamide compared with best supportive care by £450 per QALY gained.

3.47 To address its concerns about some of the parameters used in the manufacturer's base‑case model, the ERG made the following changes, one at a time:

  • Applying a hazard ratio of 1.0 for abiraterone compared with best supportive care after 25 months. This changed the ICERs for enzalutamide to £15,020 per QALY gained compared with abiraterone and to £43,398 per QALY gained compared with best supportive care.

  • Applying a hazard ratio of 0.49 estimated from the whole population to model progression-free survival for abiraterone. This changed the ICERs for enzalutamide to £12,461 per QALY gained compared with abiraterone and to £43,285 per QALY gained compared with best supportive care.

  • Excluding the utility increases to patients while taking enzalutamide or abiraterone. This changed the ICERs for enzalutamide to £16,464 per QALY gained compared with abiraterone and to £51,014 per QALY gained compared with best supportive care.

    The ERG then applied the changes listed above simultaneously in the manufacturer's base case model (from now on, this modification of the manufacturer's model is referred to as the ERG base case). The ICERs were £14,488 per QALY gained for enzalutamide compared with abiraterone, and £51,124 per QALY gained for enzalutamide compared with best supportive care. In an incremental analysis, abiraterone would be extendedly dominated by enzalutamide in all the ERG's exploratory analyses.

3.48 The ERG provided analyses with mitoxantrone included as a comparator to comply with the scope. Using the manufacturer's base‑case model, the ICER for enzalutamide compared with mitoxantrone was £33,585 per QALY gained. Within the ERG base case, this ICER was £37,840 per QALY gained. In an incremental analysis, mitoxantrone would be extendedly dominated by enzalutamide in both the manufacturer's and the ERG's base case.

3.49 To identify the input value of the model parameter below which enzalutamide would not be cost effective at a maximum acceptable ICER of £50,000 per QALY gained, the ERG performed 2 threshold analyses of:

  • how the manufacturer estimated progression-free survival in the model (that is, by using time to treatment discontinuation as a proxy)

  • the value used by the manufacturer in the model to estimate utility at baseline.

    The ERG considered that, although time to treatment discontinuation was the most reasonable proxy for progression‑free survival given the available evidence, uncertainty existed around the hazard ratio for treatment discontinuation. The ERG found that, for the manufacturer's base case ICER for enzalutamide compared with best supportive care to exceed £50,000 per QALY gained, the treatment discontinuation hazard ratio for enzalutamide compared with best supportive care must be 0.23 or lower (the base case value was 0.34). That is, improving the treatment effect on progression free survival (decreasing the hazard ratio) would increase total costs more than total QALYs for enzalutamide, resulting in a higher ICER. For the utility value at baseline, the ERG found that, for the ICER to exceed £50,000 per QALY gained, the utility value must be 0.58 or lower. The ERG considered that the utility value of 0.58 seemed realistic based on the range of values reported in the literature for this stage of the disease, although these values may represent an average value for patients with both stable and progressed disease, and may reflect between-study differences.

3.50 To reflect the Committee's preferred, but not previously considered analysis described in the first appraisal consultation document, the ERG modelled an overall survival hazard ratio for abiraterone compared with placebo that was constant over time, and assumed equal 'on-treatment' utility increases for enzalutamide and abiraterone of 0.04. This gave an ICER of £22,604 per QALY gained for enzalutamide compared with abiraterone and £45,898 per QALY gained for enzalutamide compared with best supportive care.

3.51 For the subgroup analysis of patients who have received 2 or more previous cytotoxic chemotherapy regimens (of which one included docetaxel) (see section 3.39), the ERG estimated values for median overall survival for the enzalutamide and placebo groups as 15.9 months and 12.3 months respectively (a difference of 3.6 months). The ERG noted that its estimate of median overall survival in the placebo group (12.3 months) differed from the manufacturer's estimate (12.8 months). To explore the uncertainty in the manufacturer's analysis, the ERG applied an alternative 'on-treatment' utility increase for enzalutamide of 0.04 (the 'on-treatment' utility increase used for abiraterone in the manufacturer's base case). This resulted in an ICER of £48,020 per QALY gained for enzalutamide compared with best supportive care.

3.52 Full details of all the evidence can be found on the NICE website.

  • National Institute for Health and Care Excellence (NICE)