3 The manufacturer's submission
The Appraisal Committee (appendix A) considered evidence submitted by the manufacturer of abiraterone and a review of this submission by the Evidence Review Group (ERG; appendix B). The decision problem addressed by the manufacturer considered whether treatment with abiraterone plus prednisolone was clinically effective compared with mitoxantrone (with or without prednisolone) or best supportive care for castration‑resistant metastatic prostate cancer previously treated with a docetaxel‑containing regimen and whether abiraterone treatment reflected a cost‑effective use of NHS resources.
3.1 The manufacturer carried out a systematic literature search to identify all relevant trials and studies of abiraterone and potential comparators for the treatment of castration‑resistant metastatic prostate cancer. The manufacturer identified four studies of abiraterone following previous chemotherapy: one randomised controlled trial (COU‑AA‑301) and three single arm trials (COU‑AA‑304, COU‑AA‑003, COU‑AA‑BMA). Four randomised controlled trials comparing mitoxantrone plus prednisolone with other treatments were also identified by the manufacturer, but there were no other trials to link this evidence to the COU‑AA‑301 trial and enable an indirect comparison of abiraterone and mitoxantrone. The manufacturer's clinical‑effectiveness evidence for abiraterone was derived solely from the COU‑AA‑301 trial (a phase III, placebo‑controlled, randomised, double‑blind, multicentre trial carried out across 130 sites in 13 countries, including the UK). In this trial, patients whose disease had progressed on or after docetaxel therapy and who had an Eastern Cooperative Oncology Group (ECOG) performance score of 0–2 were treated with either abiraterone (four 250‑mg tablets) in combination with prednisone or prednisolone (n=797) or with placebo (four tablets) in combination with prednisone or prednisolone (n=398). Patients in both groups continued treatment until disease progression was documented on the basis of the prostate‑specific antigen (PSA), radiographic imaging, and clinical findings. Study follow‑up was up to 60 months.
3.2 The baseline demographics and disease characteristics were similar between the two treatment groups in the COU‑AA‑301 trial: 93% of patients were white, the median age was 69 years and 28% of patients were 75 years or older. Among patients randomised to abiraterone plus prednisone or prednisolone (hereafter the 'abiraterone group'), 70% had previously received one prior course of docetaxel chemotherapy only (designated as the 'one prior chemotherapy' subgroup), compared with 69% of patients randomised to placebo plus prednisone or prednisolone (hereafter the 'prednisolone group'). The proportion of patients with an ECOG performance score of 2 (reflecting worse performance than a score of 0 or 1) was 10% and 11% in the abiraterone and prednisolone groups respectively. The majority of patients in both treatment groups (approximately 70%) had radiographic progression with or without PSA progression at baseline; 89% of patients in the abiraterone group and 90% of patients in the prednisolone group had bone metastases.
3.3 The primary outcome of the COU‑AA‑301 trial was overall survival, defined as the time from randomisation to death from any cause. A 'primary' analysis was conducted after 552 deaths (12.8 months median follow‑up) for the whole (intention‑to‑treat) population. This was on the basis of a planned interim analysis after 534 deaths (67% of the 797 deaths at the planned final analysis). In the planned interim analysis ('primary' analysis), median survival was statistically significantly longer in the abiraterone group than in the prednisolone group (14.8 months compared with 10.9 months, hazard ratio [HR] 0.65, 95% confidence interval [CI] 0.54 to 0.77). Following this analysis, and because of a significant beneficial effect of abiraterone, the trial stopped. Follow‑up continued, and an 'updated' analysis was conducted after 775 deaths (20.2 months median follow‑up) for the whole population and the one prior chemotherapy subgroup. For the whole population, median survival continued to be statistically significantly longer in the abiraterone group than the prednisolone group (15.8 months compared with 11.2 months, HR 0.74, 95% CI 0.64 to 0.86). Subgroups with an ECOG performance score of 0–1 or 2 and subgroups who had received one or more prior chemotherapy regimens were explored by the manufacturer. For the one prior chemotherapy subgroup, median survival was also statistically significantly longer in the abiraterone group than the prednisolone group (17.0 months compared with 11.7 months, HR 0.71, 95% CI 0.60 to 0.86). The manufacturer stated that statistical testing showed that the relative overall survival benefit of abiraterone was not statistically significantly different between the one prior chemotherapy subgroup and the subgroup with more than one prior chemotherapy.
3.4 Two secondary outcomes in the COU‑AA‑301 trial were radiographic progression‑free survival (time with no radiographically documented disease progression) and 'modified' progression‑free survival (based on time to death or one of the following: PSA progression, radiographic progression, increase in glucocorticoid use, pain progression, a skeletal‑related event, or initiation of a new cancer therapy). Treatment with abiraterone statistically significantly decreased the risk of radiographically documented disease progression or death compared with prednisolone in the primary analysis (HR 0.67, 95% CI 0.59 to 0.78, p<0.0001). A significantly decreased risk was also observed in the updated analysis. The median radiographic progression‑free survival was identical in the primary and updated analyses: 171 days in the abiraterone group and 110 days in the prednisolone group. Treatment with abiraterone also significantly decreased the risk of disease progression based on the criteria for 'modified' progression‑free survival compared with prednisolone in the primary analysis (HR 0.63, 95% CI 0.55 to 0.72, p<0.0001). A significantly decreased risk was also observed in the updated analysis (results provided as academic in confidence).
3.5 The manufacturer indicated that expert opinion had suggested that the endpoints of 'modified' progression‑free survival and radiographic progression‑free survival, as defined in the COU‑AA‑301 trial, did not necessarily represent the time at which trial participants stopped the study drug because of disease progression. On this basis, the manufacturer argued that treatment discontinuation was the most appropriate proxy for progression‑free survival in the economic model. The manufacturer presented rates of time to discontinuation for both the whole population (primary and updated analyses) and the one prior chemotherapy subgroup (updated analysis only). For the whole population, the median time to treatment discontinuation for the abiraterone group was significantly longer than in the prednisolone group in the primary analysis (8 months compared with 4 months), with similar results in the updated analysis. For the one prior chemotherapy subgroup, the median time to treatment discontinuation was also significantly longer for the abiraterone group than the prednisolone group and the difference between the two groups was slightly larger than that observed for the whole population (results provided as academic in confidence).
3.6 Additional outcomes in the COU‑AA‑301 trial included PSA response rates, defined as the proportion of patients with a 50% or greater decrease in PSA confirmed by a second measurement at least 4 weeks later, and objective tumour response rates, defined according to Response Evaluation Criteria in Solid Tumours (RECIST). In the primary analysis for the whole population, confirmed PSA response was statistically significantly greater in the abiraterone group than in the prednisolone group (29.1% compared with 5.5%, p<0.0001).
3.7 The most common adverse reactions (occurring in ≥10% of participants in COU‑AA‑301) reported in both treatment groups were anaemia, vomiting, hot flushes, anorexia, pain in extremities, diarrhoea, musculoskeletal pain, asthenia, dyspnoea, headache, urinary tract infection, weight loss and muscle weakness. For the primary analysis, the most frequently reported grade 3 or 4 adverse reactions in the abiraterone and prednisolone groups were fatigue, anaemia, back pain and bone pain. Adverse reactions relating to mineralocorticoid excess (hypertension, hypokalaemia and oedema), cardiac disorders and hepatotoxicity were more frequent in the abiraterone group than the prednisolone group (55% compared with 44%). Cardiac disorders (primarily grade 1 or 2) were more commonly reported in the abiraterone group than in the prednisolone group (13% compared with 11%, p=0.14). Adverse reactions resulting in death or the need to discontinue study treatment were less frequent in the abiraterone group.
3.8 The manufacturer presented updated analyses for three health‑related quality of life measures for which data were collected in the COU‑AA‑301 trial: the brief pain inventory short form (BPI‑SF), the brief fatigue inventory short form (BFI‑SF), and the functional assessment of cancer therapy‑prostate (FACT‑P). Neither the EQ‑5D nor any other utility measure was collected in the COU‑AA‑301 trial. Analyses indicated that a statistically significantly greater proportion of patients in the abiraterone group compared with the prednisolone group experienced an improvement for all three outcome measures (p<0.001). The manufacturer reported that the proportion of patients who had progression or decline in the outcome measures related to pain, functionality and fatigue did not significantly differ between the two treatment groups. However, patients in the abiraterone group experienced statistically significantly longer time to progression for all three outcome measures compared with the prednisolone group (p<0.05). The manufacturer concluded that evidence from the COU‑AA‑301 trial suggests that patients receiving abiraterone are more likely to experience less pain and fatigue, have improved functional status, and have a longer time before their pain, functional status and fatigue worsen.
3.9 The manufacturer submitted an economic model to compare the cost effectiveness of the three treatments: abiraterone plus prednisolone (hereafter 'abiraterone'), prednisolone alone (hereafter 'prednisolone'), and mitoxantrone plus prednisolone (hereafter 'mitoxantrone'). The manufacturer developed a survival‑based decision model with three health states: pre‑progression, post‑progression and dead. People with castration‑resistant metastatic prostate cancer were assumed to enter the model in the pre‑progression state having already received treatment with docetaxel‑based chemotherapy. The model assumed that people who experienced disease progression would enter the post‑progression state. The model assumed that people who received abiraterone did not take it after their disease had progressed, but would continue taking prednisolone or prednisone. The model assumed that patients in the mitoxantrone and prednisolone treatment groups would also continue taking prednisolone or prednisone until death. For the analysis of abiraterone compared with mitoxantrone, the model further assumed for mitoxantrone treatment a maximum duration of 30 weeks (median duration of 12 weeks) in the base‑case analysis.
3.10 For its base‑case analysis, the manufacturer's preferred population was comprised of people who had received one prior chemotherapy only. The number of people remaining in each health state after each cycle of the model (3 weeks, based on the dosing cycle of mitoxantrone) was calculated directly from the overall survival and progression‑free survival curves from the one prior chemotherapy subgroup of the COU‑AA‑301 trial. Time in the post‑progression state was calculated as the difference between overall survival and progression‑free survival. The model used a lifetime horizon of 10 years. The analysis took an NHS and personal social services perspective and discounted costs and benefits at 3.5%.
3.11 The key parameters of clinical effectiveness in the model were progression‑free and overall survival, which, for the base‑case analysis, the manufacturer derived from data from the updated analysis for the 'one prior chemotherapy' subgroup of the COU‑AA‑301 trial. The manufacturer assumed in the base‑case analysis that progression‑free survival and overall survival did not differ between treatment with mitoxantrone and treatment with prednisolone or prednisone. The manufacturer argued that this assumption was justified because available evidence suggests that mitoxantrone compared with corticosteroids does not extend survival in people without prior chemotherapy, and therefore it would not be expected to extend survival in people who had received chemotherapy. In the base‑case analysis, the manufacturer used data from the abiraterone arm of the COU‑AA‑301 trial to model overall survival for abiraterone, and data from the prednisolone arm to model overall survival for mitoxantrone and for prednisolone. The manufacturer used data from Kaplan–Meier curves up to the point at which 10% of patients remained in the trial. After this, the manufacturer extrapolated the overall survival curves assuming a constant hazard rate (exponential function).
3.12 In the base‑case analysis, the manufacturer modelled progression‑free survival for patients taking abiraterone who had received only one prior chemotherapy. The model assumed that when patients discontinued abiraterone treatment they moved from the pre‑progression to the post‑progression health state. Treatment discontinuation rates were based on data from the COU‑AA‑301 trial for patients who had only one prior chemotherapy, and were used as a proxy for progression‑free survival (see section 3.5). To estimate time spent in the pre‑progression state, the manufacturer took Kaplan–Meier (time to treatment discontinuation) survival data from the one prior chemotherapy subgroup of the COU‑AA‑301 trial up to the point at which 5% of participants remained at risk. Beyond this 5% cut‑off, the manufacturer extrapolated survival curves assuming a constant hazard rate. The manufacturer used data from Kaplan–Meier curves for patients in the prednisolone group who had received one prior chemotherapy to model progression‑free survival for patients taking mitoxantrone or prednisolone. The manufacturer did not extrapolate these data further because just over 2% of patients were still on treatment (that is, their condition had not progressed).
3.13 In the base‑case analysis, the manufacturer also modelled overall survival based on Kaplan–Meier survival curves for the one prior chemotherapy subgroup from the COU‑AA‑301 trial. As for progression‑free survival, the manufacturer extrapolated overall survival using a constant hazard but, because of greater censoring, chose a cut‑off of 10% and applied it to all three treatment groups.
3.14 In the model, the key differences between the mitoxantrone group and the prednisolone group were the treatment durations (maximum duration of mitoxantrone treatment was 30 weeks) and costs (mitoxantrone was associated with more adverse reactions and therefore higher costs). The 30‑week maximum duration of mitoxantrone treatment (ten 3‑weekly cycles) was taken from the TROPIC trial, which compared cabazitaxel plus prednisolone with mitoxantrone plus prednisolone in patients with metastatic prostate cancer who had received docetaxel chemotherapy. Mitoxantrone was associated with less pain than prednisolone, and the manufacturer assumed that this lower level of pain was equal to that for abiraterone. However, because more treatment‑related adverse reactions with mitoxantrone were observed in the TROPIC trial, mitoxantrone was associated with a smaller gain in health‑related quality of life compared with abiraterone.
3.15 Utility values for the pre‑progression and post‑progression health states were identified by the manufacturer in two studies that specifically collected EQ‑5D utility values in men with metastatic prostate cancer (Sandblom et al. 2004; Sullivan et al. 2007). Sandblom et al. estimated utility values in Swedish men (n=1,442) with prostate cancer (metastatic or non‑metastatic) in the year before death. The manufacturer stated that this study provided good estimates of utility for the post‑progression health state but did not provide an accurate estimate of utility for people who had stable disease after receiving further treatment following progression on or after docetaxel treatment (defined as the pre‑progression health state for the purpose of the analysis) in the UK. Sullivan et al. collected utility values for men (n=280) with metastatic prostate cancer in an observational study carried out across Europe, Australia and North America at baseline and at 3, 6 and 9 months follow‑up. The baseline utility was 0.635 for the whole cohort and 0.715 for the UK subgroup (n=29). The manufacturer stated that Sullivan et al. had not provided separate utility values for people whose disease had progressed and those whose disease had not progressed, and therefore the study did not provide suitable utility values for the post‑progression health state.
3.16 Therefore, to estimate utility values for the pre‑progression state, the manufacturer undertook a two‑stage analysis to convert FACT‑P data from the COU‑AA‑301 trial into EQ‑5D utility values. In the first stage, the manufacturer analysed data from a manufacturer‑sponsored cross‑sectional study in five European countries (including the UK) of 291 patients with castration‑resistant metastatic prostate cancer who completed both FACT‑P and EQ‑5D questionnaires. The data were used to develop an algorithm to map FACT‑P data to EQ‑5D using an ordinary least squares (OLS) regression model and the UK EQ‑5D tariff. In the second stage, the manufacturer used this mapping algorithm to convert FACT‑P data from patients in both treatment groups in the COU‑AA‑301 trial who had only one prior chemotherapy to EQ‑5D utility values. The manufacturer then applied a separate regression analysis to derive an effect of treatment on utility. The manufacturer did not explicitly model adverse reactions, but instead assumed that any differences in adverse reactions between patients taking abiraterone or prednisolone were reflected in the overall utility values estimated from the FACT‑P mapping algorithm.
3.17 In the base‑case analysis the manufacturer applied different pre‑progression utility values to the abiraterone and prednisolone groups (utility values provided as academic in confidence). The manufacturer also applied the pre‑progression utility value for abiraterone to the mitoxantrone group. In order to estimate the impact of grade 3 or 4 adverse reactions from mitoxantrone on the utility of the pre‑progression state, the manufacturer conducted a separate regression, which estimated an average utility decrement based on the occurrence of at least one adverse reaction. The manufacturer estimated that, based on the average number of grade 3 or 4 adverse reactions reported for abiraterone (from the COU‑AA‑301 trial) and mitoxantrone (from the TROPIC trial), mitoxantrone increased the occurrence of grade 3 or 4 adverse reactions by 32% compared with abiraterone. This resulted in a small utility decrement for mitoxantrone compared with abiraterone (utility values provided as academic in confidence).
3.18 The manufacturer noted that the COU‑AA‑301 trial did not collect FACT‑P data beyond the point of disease progression (treatment discontinuation). Therefore, the manufacturer used the study by Sandblom et al. to provide an estimate of utility for the post‑progression state. In this study, utility values ranged from 0.58 in men with 8–12 months to live to 0.46 in those with less than 4 months to live. Based on the average utility observed for men in the last 8 months of life, the manufacturer used a utility value of 0.50 for the post‑progression state in the economic model. This post‑progression utility value was applied for all three treatment groups.
3.19 The manufacturer included the costs of drug treatment as costs of drug acquisition, administration and monitoring. Under the terms of the patient access scheme approved by the Department of Health, the cost of a 3‑week cycle of abiraterone, based on a daily dose of 1 g, is commercial in confidence and not reported here. Dosing of mitoxantrone is determined by body surface area. The manufacturer assumed a value of 2.02 m², based on the average body surface area observed in the TROPIC trial. Based on a cost of £100 per 20‑mg vial, and assuming a patient needs two vials, the cost of mitoxantrone used in the model was £200 per 3‑week cycle. The model also assumed that a person receiving treatment with mitoxantrone would need one outpatient visit per 3‑week cycle, resulting in a total cost of £448.45 per 3‑week cycle. The cost of prednisolone was £1.03 for a 28‑tablet (5‑mg) pack (£1.55 per 3‑week cycle at 10 mg daily). Because patients were taking prednisolone with abiraterone and with mitoxantrone, this cost was included for all three treatments and was assumed to continue after treatment with abiraterone and with mitoxantrone was discontinued until death. The costs of mitoxantrone and prednisolone were taken from the BNF (edition 61).
3.20 The model also included the costs of scheduled follow‑up consisting of clinical visits, diagnostic imaging and clinical laboratory tests to monitor the status of the disease. To estimate scheduled UK medical resource utilisation in each treatment group, the manufacturer convened a clinical consensus panel, consisting of five oncologists and three oncology nurse specialists. Based on statistical analysis of data from the COU‑AA‑301 trial, the manufacturer estimated that unscheduled medical resource utilisation (because of unplanned clinical events) would be similar for patients whether taking abiraterone or prednisolone alone. Therefore, the manufacturer applied a one‑off fixed cost of unplanned, event‑related resource utilisation in the pre‑ and post‑progression states to patients taking abiraterone or prednisolone alone. For patients receiving mitoxantrone, the manufacturer assigned the extra costs of treating grade 3 or 4 adverse reactions based on data from the TROPIC trial. The costs of other drugs (including bisphosphonates in the pre‑ and post‑progression states for people taking any of the three treatments, and granulocyte colony‑stimulating factor [G‑CSF] to treat febrile neutropenia in the abiraterone and mitoxantrone groups) were also included in the model. Based on results from the UK subpopulation of the COU‑AA‑301 trial, in which a small proportion of patients received subsequent chemotherapy treatments, including taxanes and anthracenediones, the manufacturer assumed that a proportion of people in each treatment group would receive three cycles of cabazitaxel in the post‑progression state. The manufacturer also estimated resource utilisation and costs for end‑of‑life care based on clinical expert opinion. These costs were applied for the last 3 months of life for patients taking any of the three treatment regimens.
3.21 The manufacturer's original base‑case deterministic cost‑effectiveness results for the one prior chemotherapy subgroup showed that mitoxantrone was extendedly dominated, that is, the incremental cost‑effectiveness ratio (ICER) of mitoxantrone compared with prednisolone was higher than that of the next most effective alternative (abiraterone). The comparison between abiraterone and prednisolone alone resulted in an ICER of £52,851 per QALY gained (incremental costs and incremental QALYs provided as commercial in confidence). The manufacturer's original base‑case probabilistic ICERs were similar. The manufacturer's original base‑case deterministic cost‑effectiveness results for the whole trial population showed that mitoxantrone was extendedly dominated by abiraterone and prednisolone alone whereas the comparison between abiraterone plus prednisolone and placebo plus prednisolone resulted in an ICER of £63,233 per QALY gained (incremental costs and incremental QALYs provided as commercial in confidence).
3.22 The manufacturer conducted a number of one‑way sensitivity analyses on various model parameters which included: altering the time horizon from 10 years to 4, 6, or 8 years, varying the discount rates for costs and benefits from 3.5% each to 0% and 6%, using a Weibull instead of an exponential function to extrapolate survival beyond the cut‑off point of the Kaplan–Meier curve for both progression‑free and overall survival, using alternative estimates of utility for pre‑ and post‑progression health states and for the effect on utility of treatment with abiraterone, and varying a number of cost input parameters (±50%). The results of these one‑way sensitivity analyses indicated that the original base‑case ICERs were fairly insensitive to changes in most, but not all, of the input parameters. Decreasing the baseline utility value for the pre‑progression state to 0.538 increased the ICER significantly, resulting in an ICER for treatment with abiraterone compared with prednisolone of £77,000 per QALY gained. In one scenario analysis, the manufacturer assumed that patients taking mitoxantrone remained in the pre‑progression state for longer than patients taking prednisolone only (HR 0.77). This scenario resulted in an ICER of £21,038 per QALY gained for mitoxantrone compared with prednisolone and an ICER of £62,843 per QALY gained for abiraterone compared with mitoxantrone. Results of the probabilistic sensitivity analysis showed that prednisolone had the highest probability (100%) of being cost effective at a level of £20,000 to £30,000 per QALY gained, whereas abiraterone had the highest probability of being cost effective at over £50,000 per QALY gained.
3.23 The ERG was satisfied with the methodological quality of the COU‑AA‑301 trial and considered that it provided persuasive evidence that abiraterone offers a survival advantage in patients with castration‑resistant metastatic prostate cancer. However, the ERG stated that the one prior chemotherapy subgroup differed from the COU‑AA‑301 trial population and, because it was smaller, it had reduced statistical power for comparison of outcomes between the treatment arms. The ERG commented that the manufacturer had submitted a relatively straightforward economic model comparing the relevant comparators following docetaxel chemotherapy, and had closely adhered to the NICE reference case requirements for economic analysis.
3.24 The ERG commented that the factors with the most influence on the cost effectiveness of abiraterone compared with prednisolone and with mitoxantrone were the differences in the EQ‑5D utility values attached to the pre‑ and post‑progression health states for all three treatments, which the manufacturer had derived from different sources. The ERG noted that the mapping function, developed and used by the manufacturer to determine pre‑progression utility values, had yet to appear in a peer‑reviewed publication. The ERG noted that the pre‑progression utility value for patients taking abiraterone, which was estimated from a function mapping FACT‑P to EQ‑5D, was similar to or higher than EQ‑5D utility values for men of similar age taken from a survey of the general UK population living in the community (Kind et al. 1998). This survey reported average EQ‑5D visual analogue scores of between 0.800 and 0.750 for men aged between 60 and 79 years. The ERG's clinical advisers suggested that, because the COU‑AA‑301 trial may have been oversubscribed, the population selected could have been fitter than generally seen in clinical practice.
3.25 The ERG noted that because most patients in the cross‑sectional study used to derive the mapping function were receiving chemotherapy, both their FACT‑P scores and EQ‑5D utilities would probably be lower than for patients in the COU‑AA‑301 trial who were not taking chemotherapy, but who received abiraterone or prednisolone. The ERG argued that this could have resulted in the FACT‑P mapping function converting FACT‑P values outside the reliable range of the mapping function, which would increase the uncertainty around the manufacturer's derived EQ‑5D utility values. The ERG noted that the manufacturer derived the mapping function from a restricted set of patients who provided FACT‑P scores both at baseline and over the course of the trial, but that over time, a declining proportion of trial participants who remained on treatment provided FACT‑P scores. The ERG further noted that the manufacturer's regression analysis was based on changes from baseline rather than on absolute values of the scores and that the baseline scores among those reporting FACT‑P scores after baseline may have been higher in the prednisolone group than in the abiraterone group. The ERG suggested that patients receiving prednisolone who reported FACT‑P scores after baseline may have had a smaller change in their scores than patients receiving abiraterone because they had less severe disease at baseline.
3.26 The ERG had some concerns about the manufacturer's approach to the regression analysis used to estimate separate pre‑progression utility values for treatment with abiraterone and prednisolone. The ERG noted that the utility values used in the manufacturer's model implied that utility dropped when a patient discontinued treatment (disease progressed) but that the manufacturer assumed a larger utility decrement for discontinuing treatment with abiraterone than for discontinuing prednisolone. The ERG considered this pre‑progression utility value to be an overestimate, which, by increasing the utility decrement when moving from the pre‑progression to the post‑progression state, would exaggerate the benefit of remaining on abiraterone treatment. The ERG also noted that the utility decrement applied when moving from stable to progressive disease was lower in 'Cabazitaxel for hormone‑refractory metastatic prostate cancer previously treated with a docetaxel‑containing regimen' (NICE technology appraisal guidance 255) than that used in this appraisal.
3.27 The ERG explored the manufacturer's methods for extrapolating overall survival and suggested that the manufacturer's decision to extrapolate beyond a cut‑off of 10% of trial participants at‑risk was arbitrary. However, the ERG was aware that when the manufacturer applied a cut‑off of 5% in sensitivity analyses there was little impact on the ICERs for abiraterone compared with prednisolone. In contrast, the ERG's exploratory analyses using cut‑offs ranging from 0% to 20% showed that the cut‑off of 10% used by the manufacturer resulted in a relatively low ICER compared with other cut‑off points. The ERG noted that when extrapolating progression‑free survival, the manufacturer chose different cut‑off points (or none at all) depending on the treatment. The ERG noted that when data from Kaplan–Meier survival plots of the prednisolone group were used in the economic model, 2% of patients remained in the pre‑progression state, which may have slightly underestimated progression‑free survival for the mitoxantrone and prednisolone treatment groups compared with abiraterone. The ERG also noted that the Kaplan–Meier curve for treatment discontinuation had an unusual shape, with many patients discontinuing treatment over a period of a few weeks at approximately 60 days into treatment. In the ERG's view, this was unlikely to represent actual disease progression which would be better represented by fitting a parametric distribution.
3.28 The ERG noted that there was uncertainty around the most appropriate functions for extrapolating overall survival and progression‑free survival in the manufacturer's model. The ERG commented that by using patient‑level trial data to generate Kaplan–Meier curves for overall survival (up to the cut‑off) the manufacturer considered all time points. The ERG argued that patient‑level Kaplan–Meier data, although representing observed data, may be less applicable to patients outside the COU‑AA‑301 trial than well‑fitting parametric distributions. The ERG commented that when extrapolating overall survival using a constant hazard, the manufacturer used only two time points (baseline and cut‑off). The ERG considered that this approach was less reasonable than applying a parametric distribution to the data representing all time points.
3.29 The ERG and manufacturer conducted exploratory analyses to assess the impact of fitting alternative parametric functions to data reflecting overall survival and progression‑free survival from the COU‑AA‑301 trial. In response to requests from the ERG for clarification, the manufacturer indicated that the best‑fitting parametric distributions (according to Akaike and Bayesian information criteria) were: a Weibull distribution for overall survival in the abiraterone group, a log‑normal distribution for overall survival in the prednisolone group, and a log‑logistic distribution for progression‑free survival for both abiraterone and prednisolone. The ERG and the manufacturer agreed that the log‑normal distribution should not be used to extrapolate overall survival because its long tail resulted in implausibly long survival. The ERG also noted that when the log‑logistic distribution was fitted to data for time to treatment discontinuation, the resulting curve crossed the extrapolated part of the Kaplan–Meier overall survival curve from the manufacturer's base case so that some patients appeared to remain in the pre‑progression state after death, which is clinically impossible. In the ERG's view, a Weibull distribution should have been used to estimate overall and progression‑free survival for abiraterone and for prednisolone alone at all time points rather than using extrapolated Kaplan–Meier curves. The ERG acknowledged that both approaches were associated with uncertainty. When the ERG fitted the Weibull parametric distribution to overall survival in both the abiraterone and prednisolone groups and an extrapolated (constant hazard) Kaplan–Meier curve for progression‑free survival, the ICER for abiraterone compared with prednisolone was £56,339 per QALY gained. When the ERG fitted its preferred distribution, that is, a Weibull parametric distribution to overall survival and progression‑free survival in both the abiraterone and prednisolone groups, the ICER for abiraterone compared with prednisolone increased to £58,116 per QALY gained.
3.30 In addition to fitting the Weibull distribution to estimate overall survival and progression‑free survival, the ERG made changes to cost and utility parameters in the economic model. The ERG revised administration costs to reflect the costs of oncology outpatient visits and of administering mitoxantrone, and varied the proportion of patients receiving bisphosphonates following progression. The ERG amended the manufacturer's regression analysis of utility for progression‑free survival for patients who take prednisolone only. With these changes, mitoxantrone continued to be extendedly dominated and the ERG's revised base‑case ICER for abiraterone compared with prednisolone was £60,084 per QALY gained.
3.31 The ERG conducted additional one‑way sensitivity analyses for its revised base‑case analysis. When the ERG varied the costs for unplanned use of medical resources, the ICERs for abiraterone compared with prednisolone ranged from £60,492 to £67,554 per QALY gained. When the ERG varied utility estimates (provided as academic in confidence), ICERs for abiraterone compared with prednisolone ranged from £63,281 to £72,469 per QALY gained. When the ERG extrapolated overall survival for prednisolone using the truncated log‑normal distribution, the ICER for abiraterone compared with prednisolone was £70,217 per QALY gained. Mitoxantrone continued to be extendedly dominated in all sensitivity analyses.
3.32 In response to the appraisal consultation document, the manufacturer revised the confidential discount under the patient access scheme agreed with the Department of Health. The manufacturer also amended the economic model to reflect the changes to costs suggested by the ERG. These included changes to administration costs to reflect the costs of oncology outpatient visits and of administering mitoxantrone, and changes to the proportion of patients receiving bisphosphonates following disease progression. As a result of these changes and the revised discount under the patient access scheme, the manufacturer's deterministic base‑case ICER for abiraterone compared with prednisolone decreased to £46,800 per QALY gained for the one prior chemotherapy subgroup and to £52,851 per QALY gained for the whole population (incremental costs and incremental QALYs provided as commercial in confidence).
3.33 The manufacturer conducted a number of one‑way sensitivity analyses (see section 3.22) using the amended model for the one prior chemotherapy subgroup, resulting in ICERs for abiraterone compared with prednisolone of £42,904 to £51,110 per QALY gained. In all analyses, ICERs were most sensitive to changes to the utility value for the pre‑progression health state and to the statistical approach used to model overall survival. When a utility value of 0.715 (taken from the UK subgroup of the study by Sullivan et al. 2007) was assigned to the pre‑progression state, the ICER increased to £51,110 per QALY gained for abiraterone compared with prednisolone. When a Weibull parametric distribution was used to model overall survival, the ICER increased to £49,911 per QALY gained for abiraterone compared with prednisolone. The probabilistic sensitivity analysis showed that abiraterone had a 67% probability of being cost effective at £50,000 per QALY gained.
3.34 The ERG reviewed the amended model and revised analyses provided by the manufacturer in response to the appraisal consultation document. The ERG noted that the manufacturer's amended model retained extrapolation of the Kaplan–Meier curves using a constant hazard to model overall and progression‑free survival as well as the original utility values for the pre‑ and post‑progression health states. The ERG confirmed that the manufacturer had amended administration costs in the model to reflect the costs of oncology outpatient visits as suggested by the ERG. However, the ERG had concerns about the application of a half‑cycle correction to the drug costs in the model by the manufacturer. The ERG noted that amending this half‑cycle correction in the model would increase the drug costs by approximately half the monthly cost and would slightly increase the ICERs.
3.35 After amending the manufacturer's regression analysis of utility for the pre‑progression health state, the ERG conducted exploratory analyses on the manufacturer's amended model to assess the impact of fitting different curves to estimate survival. For the one prior chemotherapy subgroup, the ERG's base‑case ICER for abiraterone compared with prednisolone was £53,140 per QALY gained when a Weibull parametric distribution was used to estimate overall survival and progression‑free survival. When the one prior chemotherapy subgroup was used with a Weibull distribution to estimate overall survival and a Kaplan–Meier curve for progression‑free survival, the ERG's base‑case ICER was £52,186 per QALY gained. When the whole trial population was used with a Weibull distribution to estimate overall survival and the Kaplan–Meier curve for progression‑free survival, the ERG's base‑case ICER was £60,038 per QALY gained. The ERG also conducted a number of sensitivity analyses in which it varied the utility values for the pre‑ and post‑progression health states using values for metastatic prostate cancer identified by the manufacturer in a literature review plus those used in the cabazitaxel appraisal (TA 255). This resulted in ICERs ranging from £52,362 to £71,358 per QALY gained for the one prior chemotherapy subgroup. The ICERs from all of the ERG's analyses further increased when the ERG amended the half‑cycle correction applied by the manufacturer to the drug costs (ICERs provided as commercial in confidence).
3.36 Full details of all the evidence are in the manufacturer's submission and the ERG report, which are available from www.nice.org.uk/guidance/TA259/history.