Cabazitaxel for hormone-relapsed metastatic prostate cancer treated with docetaxel

3.1 TROPIC is a phase III randomised open‑label multicentre trial that compared cabazitaxel with mitoxantrone in men with metastatic hormone‑relapsed prostate cancer. The trial recruited people whose disease had progressed on or after treatment with docetaxel. Patients aged 18 years or older were randomised 1:1 to have either:

3.2 The company stated that mitoxantrone was equivalent to best supportive care. To support this statement, it referred to an analysis that used data from 2 separate trials to compare mitoxantrone plus prednisone with prednisone alone (Green et al. 2015). There was no significant difference in overall survival between mitoxantrone and prednisolone, so the company concluded that mitoxantrone was a reasonable proxy for best supportive care.

3.3 The primary outcome measure in TROPIC was overall survival, defined as the time from the date of randomisation to death from any cause. If it was unknown whether the patient was still alive, the survival time was censored at the last date the patient was known to be alive, or at the data cut‑off date. Secondary outcomes included progression‑free survival defined as the time from randomisation to any one of: tumour progression, prostate‑specific antigen (PSA) progression, pain progression, or death from any cause.

3.4 The company presented an analysis of TROPIC that was published after a median follow‑up of 20.5 months (study cut‑off date: 10 March 2010), at which point 585 deaths had occurred. The trial included 2 analyses: intention to treat and per protocol. The intention‑to‑treat analysis included all randomised patients (n=755); the results are shown in table 1. The per‑protocol analysis for adverse events included only those patients who had at least 1 dose of the study treatment (n=742).

3.5 The company presented a subgroup analysis that was post hoc (not specified up front in the design of the trial) for patients in TROPIC with an eastern cooperative oncology group (ECOG) performance status of 0 to 1 (lower scores reflect better function) who had had 225 mg/m² or more of docetaxel. The company highlighted that in NICE's 2012 technology appraisal guidance on cabazitaxel the committee had considered that this subgroup represented clinical practice in England. The subgroup comprised 632 (83.7%) patients out of a total of 755 randomised patients (table 1).

3.6 In the subgroup analysis (table 1) median overall survival was 15.6 months (95% confidence interval [CI] 13.96 to 17.28) in the cabazitaxel group and 13.4 months (95% CI 11.99 to 14.52) in the mitoxantrone group. The difference was 2.2 months. The risk of death was statistically significantly lower in the cabazitaxel group than in the mitoxantrone group (hazard ratio [HR] 0.69; 95% CI 0.57 to 0.82; p<0.001).

Abbreviations: n, number; CI, confidence interval; ECOG, Eastern Cooperative Oncology Group.

3.7 No trials have directly compared the effectiveness of cabazitaxel with abiraterone or enzalutamide. The company did a network meta‑analysis to compare the effectiveness of these 3 drugs indirectly using a fixed‑effects model. It identified the COU‑AA‑301 and AFFIRM trials from its systematic literature review. AFFIRM compared enzalutamide (with or without oral prednisone) with placebo (with or without oral prednisone). COU‑AA‑301 compared abiraterone plus prednisone with placebo plus prednisone.

3.8 The company noted that the definition of progression in TROPIC was different to the definition in COU‑AA‑301 and AFFIRM because TROPIC used a multiple‑component endpoint. Therefore, the company chose radiographic progression‑free survival to inform its network meta‑analysis, which it defined as the time from randomisation to the first occurrence of tumour progression (based on the Response Evaluation Criteria in Solid Tumors [RECIST] criteria) or death from any cause.

3.9 The network meta‑analysis showed that enzalutamide improved radiographic progression‑free survival, but not overall survival, compared with cabazitaxel. There was no difference between cabazitaxel and abiraterone in either overall survival or radiographic progression‑free survival.

3.10 The company advised that its network meta‑analysis assumed that the trial populations, and control‑group treatments, were similar across all 3 of the included trials. The company noted that these assumptions may not be met, and so the results of the network meta-analysis should be treated with caution.

3.11 The company produced a partitioned survival model to assess the cost effectiveness of cabazitaxel compared with mitoxantrone. In its base case the company modelled the subgroup of patients in TROPIC (see section 3.5) who had an ECOG performance status of 0 to 1 and previously had at least 225 mg/m² of docetaxel.

3.12 The company considered it standard NHS practice to treat hormone‑relapsed prostate cancer with either abiraterone or enzalutamide in the pre‑chemotherapy setting, that is, before docetaxel. Thus, in its main analyses, the company compared cabazitaxel with best supportive care, which it stated was the same as mitoxantrone (see section 3.2). However, in an alternative pathway (without abiraterone or enzalutamide before docetaxel) the company compared cabazitaxel with abiraterone and cabazitaxel with enzalutamide.

3.13 The company's Markov model had 3 states representing disease progression from stable disease through to progressive disease and death. It included a 10‑year time horizon, 3‑week cycle lengths and discounting of costs and health benefits at 3.5%. The company included the costs incurred by the NHS and personal and social services. The base‑case model compared 2 treatments:

3.14 To model time to disease progression and time to death for the subgroup (patients in TROPIC with an ECOG performance status of 0 to 1 who previously had at least 225 mg/m² of docetaxel), the company's original base case used a log‑normal curve for time to progression and a Weibull curve for time to death. In its response to the appraisal consultation document, the company submitted analyses using a piecewise curve to predict overall survival with cabazitaxel (see section 3.29 and section 3.36).

3.15 The company did not collect data on health‑related quality of life in TROPIC, so it took utility values from the UK early access programme (EAP) that allowed the company to provide cabazitaxel to patients before its official launch. The programme measured the health‑related quality of life (using the EQ‑5D) of men who had been treated with cabazitaxel after docetaxel. In the stable disease state of the model, utility increased with successive cycles of cabazitaxel treatment. The utility value was 0.70 during the first cycle and 0.82 during the tenth cycle. In the progressive disease state, the utility was 0.63 until the last 3 months of life in which the company set utility at 0.

3.16 Disutility values for adverse events were not collected in either the UK EAP or in TROPIC. The company derived disutility values associated with experiencing each adverse event from a literature review that was done for NICE's 2012 technology appraisal guidance on cabazitaxel. These studies included breast and lung cancer, but not prostate cancer.

3.17 The company modelled 15 adverse events using the proportions of adverse events in TROPIC, and included all at grade 3 and above that occurred in 2% or more of patients in any TROPIC treatment group. In addition, the company included deep vein thrombosis and peripheral sensory neuropathy as they were classified as important based on clinical expert opinion.

3.18 The company estimated resource use (such as the frequency of hospital admissions and adverse events) using data from: TROPIC; a UK clinical audit; and opinion from experts. It estimated costs using the British national formulary (BNF), NHS reference costs and data from the personal social services research unit.

3.19 In the stable disease state, the company included costs of acquiring drugs (for active treatment, pre‑medications and concomitant medications), costs of administering chemotherapy, costs of managing disease including hospitalisation and testing, and costs of adverse events. Costs for active treatment, pre‑medications and administering chemotherapy were applied for up to 10 cycles for cabazitaxel and mitoxantrone (the maximum number allowed in TROPIC). Mitoxantrone comes in vials and the dose depends on body surface area. The company assumed that the mean body surface area was 1.9 m² (based on clinical opinion; the mean body surface area observed in TROPIC was 2.01 m²). It also assumed that some mitoxantrone would be wasted when a vial was opened but not fully used.

3.20 The dose of cabazitaxel depends on body surface area. Prior to this appraisal, cabazitaxel was only purchased in vials. Because an individual dose may not require a whole vial, and the summary of product characteristics does not permit vial sharing, this meant that some cabazitaxel was wasted. In response to the appraisal consultation document, the company explained that it has set up a new compounding scheme. The company provided the following details of the scheme:

3.21 In the progressed disease state, the company included: acquisition costs for chemotherapy and best supportive care given after disease progression; costs of administering chemotherapy; and costs of managing disease including hospitalisation, imaging and testing.

3.22 In response to the appraisal consultation document, the company increased the level of discount in its patient access scheme. The updated base‑case results using the increased discount are given in section 3.37.

3.23 The company's scenario analyses compared cabazitaxel (including patient access scheme discount) with enzalutamide (at list price) and, separately, abiraterone (at list price). Although both enzalutamide and abiraterone are offered by their respective companies to the NHS with discounts, the enzalutamide discount is confidential and not known to Sanofi. The scenario analyses used the intention‑to‑treat results from TROPIC. The company assumed that patients take enzalutamide and abiraterone until disease progression or death, whereas patients use cabazitaxel for up to 10 cycles.

3.24 The company took the hazard ratios reflecting the effectiveness of abiraterone and enzalutamide compared with cabazitaxel from its network meta‑analysis, and applied these to the parametric distributions modelling overall survival and progression‑free survival with cabazitaxel. The company used a Weibull curve to model progression‑free survival. The company did not report a fully incremental analysis.

3.25 Because of the confidential discounts the ERG recalculated the company's scenario analyses using the patient access scheme discounts for cabazitaxel, enzalutamide and abiraterone.

3.26 The ERG agreed with the company's concerns about the assumptions made in the company's network meta‑analysis (see section 3.10). The ERG noted that in the presence of between‑study heterogeneity, a fixed‑effects model is not appropriate; it advised that instead the company should have used a random‑effects model. The ERG did an analysis using a random‑effects model and a weakly informative prior for the between‑study standard deviation. The results showed no statistically significant difference between any of the treatments in either overall survival or radiographic progression‑free survival.

3.27 The ERG also noted that the company used hazard ratios for the network meta‑analysis which may not have been appropriate. In the COU‑AA‑301 study for abiraterone compared with placebo, the placebo overall survival curve crosses the abiraterone curve at 24 months; this means that the proportional hazards assumption may not hold. Accordingly, the ERG advised that the results of the network meta‑analysis should be treated with caution.

3.28 The ERG noted that 18% of patients in the cabazitaxel group of TROPIC withdrew from treatment because of adverse events, compared with 8% in the enzalutamide group of AFFIRM and 13% in the abiraterone group of COU‑AA‑301. The company, in response to a clarification question before the first committee meeting, presented a fixed‑effects network meta‑analysis of adverse events. The results showed an increase in anaemia and nausea with cabazitaxel compared with best supportive care, abiraterone and enzalutamide. In addition there was an increased incidence of diarrhoea with cabazitaxel compared with best supportive care and abiraterone.

3.29 The ERG noted that in NICE's 2012 technology appraisal guidance on cabazitaxel the committee preferred the piecewise approach for extrapolating TROPIC data rather than the methods presented by the company. This was because some patients in the cabazitaxel group died from neutropenia early in the trial, which may have biased the predicted survival times from a single extrapolation curve. The ERG asked why the company had not used piecewise curves to model overall survival. Piecewise methods use independent distributions to calculate transition probabilities during different time periods; for example, using a Kaplan–Meier curve at the start of the model, then after a cut‑off point using a parametric distribution. In response to a clarification question from NICE before the committee meeting, the company presented results using a piecewise curve for the cabazitaxel arm (specifically, using a Kaplan–Meier curve for the first 2.1 months and a Weibull curve thereafter) and a Weibull curve for the mitoxantrone arm, as unchanged from the base case. The ERG advised that the piecewise curve for overall survival with cabazitaxel is likely to be more appropriate than the single Weibull curve the company used in its base case. The company's new analyses submitted in response to the appraisal consultation document used the piecewise curve for overall survival with cabazitaxel.

3.30 The ERG raised concerns about how the company had modelled patients who stop treatment with cabazitaxel or mitoxantrone. It noted that patients in the stable disease state continued treatment until:

3.31 The company included a disutility in the quality‑adjusted life year (QALY) calculations to account for the assumed reduced quality of life experienced by people with progressive disease in their last 3 months of life. The ERG noted that this disutility was applied to all deaths in the model rather than only people with progressive disease. The company's new analyses submitted in response to the appraisal consultation document removed this disutility. Removing the disutility slightly increased the ICER for cabazitaxel compared with mitoxantrone.

3.32 The ERG advised that for generic drugs it is more appropriate to use prices from the electronic market information tool (eMIT) than the BNF because eMIT is based on the price paid by English hospitals. Using eMIT prices slightly increased the ICER comparing cabazitaxel with mitoxantrone. The company's new analyses submitted in response to the appraisal consultation document used the eMIT price for mitoxantrone.

3.33 The ERG highlighted that 3 different estimates were available for the costs of treatment in the progressed‑disease health state. The most expensive estimate (£1,767.02) was based on the mitoxantrone group in the TROPIC trial. The least expensive estimate (£1,192.81) was based on the cabazitaxel group in TROPIC. The third estimate was from a UK clinical audit (£1,364.07). The company's base case used the estimate from the cabazitaxel group in TROPIC for the costs of treatment after cabazitaxel, and the estimate from the mitoxantrone group in TROPIC for the costs of treatment after mitoxantrone, abiraterone or enzalutamide. In the ERG's opinion, the company should have used the same post‑progression treatment costs for cabazitaxel and each of the comparators. Accordingly, the ERG used the UK clinical audit to estimate the post‑progression treatment costs for cabazitaxel and the comparators. This slightly reduced the ICER for cabazitaxel compared with mitoxantrone. The company's new analyses submitted in response to the appraisal consultation document used the UK clinical audit to estimate post‑progression treatment costs.

3.34 The ERG noted that the company assumed no wasted cabazitaxel. During NICE's 2012 technology appraisal of cabazitaxel, clinical experts advised that because cabazitaxel is supplied in vials, there is likely to be some wastage of cabazitaxel in NHS clinical practice, but there was uncertainty about how much waste would occur. For the committee to consider at its first meeting, the ERG did an analysis, which assumed that a cycle of treatment with cabazitaxel would incur the cost of 1 vial of cabazitaxel and that the NHS would bear the cost of drug wastage from partly used vials. This increased the ICER for cabazitaxel compared with mitoxantrone (the results are confidential and cannot be reported here).

3.35 The ERG's exploratory base case included the following assumptions:

3.36 In response to consultation the company submitted an updated base case, using the committee's preferred assumptions (see sections 4.12 to 4.17) to compare cabazitaxel with best supportive care. The updated base case included the following assumptions:

3.37 The company's deterministic base case estimated that cabazitaxel (with updated patient access scheme discount) compared with mitoxantrone resulted in an ICER of £45,159 per QALY gained (incremental costs £10,682, incremental QALYs 0.237). The probabilistic ICER was £45,982 per QALY gained.

3.39 After consultation, the ERG did a fully incremental analysis comparing cabazitaxel (with updated patient access scheme discount) with enzalutamide, abiraterone and best supportive care (represented by mitoxantrone). Over the course of this appraisal, the patient access scheme for abiraterone changed from a simple discount to a complex scheme (dose capping). The ERG's analyses used the new complex patient access scheme for abiraterone. The ERG used its random‑effects network meta‑analysis (see section 3.26) to estimate the effectiveness of cabazitaxel compared with each treatment. The ERG's incremental analysis showed that cabazitaxel was extendedly dominated by best supportive care and enzalutamide. An intervention is 'extendedly dominated' when it is more costly and less effective than a combination of 2 comparators (in this case, best supportive care and enzalutamide). Abiraterone was also extendedly dominated by best supportive care and enzalutamide.

3.40 The ICERs for cabazitaxel compared with best supportive care were substantially higher in the ERG's fully incremental analysis than the ERG's pairwise comparison of cabazitaxel with mitoxantrone in its base case. The incremental analysis used the network meta‑analysis results to estimate the effectiveness of each treatment, whereas the pairwise comparison used data from TROPIC only. The ERG advised that the network meta‑analysis assumes proportional hazards, but the data may not meet this assumption. Both the ERG and the company stated that the results of the network meta‑analysis should be treated with caution.

3.41 The ERG noted that the company did not compare cabazitaxel with radium‑223 dichloride, as specified in NICE's scope. In response to a clarification question from NICE before the committee's first meeting, the company provided results from ALSYMPCA: a randomised trial that compared radium‑223 dichloride with placebo. In ALSYMPCA, the subgroup of patients treated with radium‑223 dichloride and who had previously had docetaxel had a median overall survival of 14.4 months (95% CI 12.5 to 15.5). For comparison, patients in the cabazitaxel group of TROPIC (intention-to-treat analysis) had median overall survival of 15.1 months (95% CI 14.0 to 16.5). The ERG noted that both overall survival and progression‑free survival with radium‑223 dichloride appeared to be similar to that with cabazitaxel and that if the cost effectiveness of these 2 drugs were compared, drug costs would likely be a key driver. The ERG presented an analysis comparing the costs of cabazitaxel and radium‑223 dichloride (including the confidential patient access scheme discounts for both drugs); the results are confidential and cannot be reported here.