3 The manufacturer's submission
3.1 The manufacturer presented evidence on the clinical effectiveness of pazopanib used in line with the conditional marketing authorisation and the appraisal scope. The main clinical-effectiveness evidence came from patients in the treatment-naive subgroup of a phase III randomised controlled trial (RCT). The RCT, VEG105192, compared the effect of a once-daily dose of pazopanib plus best supportive care (155 patients) with placebo plus best supportive care (78 patients). Best supportive care was defined as the monitoring of progression, symptom control and palliative care without active treatment. The trial was conducted in patients with advanced renal cell carcinoma with predominantly clear cell histology. All patients had a good performance status (ECOG performance status 0 or 1) at the start of the trial. Baseline characteristics of the patients in the two treatment arms were equally balanced.
3.2 The primary outcome in the study was progression-free survival, which was defined as time from randomisation to disease progression or death. Tumour assessments were performed using RECIST (Response Evaluation Criteria in Solid Tumours) and were confirmed by an independent review committee. Once disease progression was confirmed, patients who previously received placebo plus best supportive care could be offered open-label pazopanib plus best supportive care in the open-label extension study (VEG107769) if the treating clinician thought this was in the best interests of the patient. At disease progression, only patients with an ECOG performance status of less than or equal to 2 were permitted to cross over to receive pazopanib. The median progression-free survival was statistically significantly longer in patients receiving pazopanib (p < 0.001). The median progression-free survival was 11.1 months (95% confidence interval [CI] 7.4 to 14.8) for patients receiving pazopanib plus best supportive care and 2.8 months (95% CI 1.9 to 5.6) for patients receiving placebo plus best supportive care (hazard ratio [HR] 0.40, 95% CI 0.27 to 0.60) based on blinded imaging assessment by the Independent Review Committee. This meant there was a 60% reduction in risk of disease progression for patients receiving pazopanib plus best supportive care compared with those receiving placebo plus best supportive care at the final analysis. The manufacturer performed sensitivity analyses of progression-free survival based on actual scan dates and investigators' assessment to confirm the robustness of the findings. The median progression-free survival based on actual scan dates was 10.8 months (95% CI 7.4 to 14.8) for patients receiving pazopanib plus best supportive care and 2.9 months (95% CI 1.9 to 5.4) for patients receiving placebo plus best supportive care (HR 0.36, 95% CI 0.24 to 0.55). These latter estimates were used in the indirect comparison.
3.3 At the time of the interim overall survival analysis (23 May 2008), 31 of 78 (40%) treatment-naive patients randomised to receive placebo in the VEG105192 trial had crossed over to receive pazopanib. At the final analysis (15 March 2010), a total of 40 (51%) treatment-naive patients randomised to placebo had crossed over to receive pazopanib. Overall survival for the treatment-naive, intention-to-treat population, unadjusted for crossover, was 22.9 months (95% CI 17.6 to 25.4 months) for patients randomised to pazopanib plus best supportive care and 23.5 months (95% CI 12.0 to 34.3 months) for patients randomised to placebo plus best supportive care. The hazard ratio for overall survival was 1.01 (95% CI 0.72 to 1.42, p = 0.525). The manufacturer presented a variety of methods to adjust for crossover when estimating median overall survival, including the inverse probability censoring weighted (IPCW) and rank preserved structural failure time (RPSFT) methods. The manufacturer considered the RPSFT weighted method to be the most appropriate because it was considered an acceptable approach for NICE technology appraisal guidance 179 ('Sunitinib for the treatment of gastrointestinal stromal tumours') and the ongoing technology appraisal 'Everolimus for the second-line treatment of advanced renal cell carcinoma'. The RPSFT weighted method estimated the overall survival of patients randomised to receive placebo assuming that they had not crossed over, that is, as if they had remained on placebo for the duration of the trial. The method proportionally 'shrunk' the estimated amount of additional survival given to patients who crossed over to receive pazopanib. The RPSFT analysis suggested that treatment with pazopanib was consistently associated with survival benefit compared with placebo (HR 0.501, 95% CI 0.136 to 2.348). The manufacturer also noted that the 0.501 hazard ratio lies within the range of estimates obtained from the different methods used to adjust for crossover. This hazard ratio was subsequently used for the indirect comparison.
3.4 No head-to-head trials analysing the efficacy of pazopanib compared with other active treatment options were available. Therefore, the manufacturer undertook a search for trials of comparator interventions and carried out an indirect comparison to estimate the relative effect of pazopanib versus the comparators (sunitinib, interferon-α and best supportive care). Seven studies were included in the indirect comparison, including one study of pazopanib compared with placebo (VEG105192), one study of sunitinib compared with interferon-α (Motzer et al. 2009) and five studies that directly compared interferon-α with a non-interferon control therapy (medroxyprogesterone acetate and vinblastine). The populations in the pazopanib (VEG105192) and sunitinib (Motzer et al. 2009) studies were comparable, but with the exception that a higher proportion of patients with a baseline ECOG performance status of 0 were recruited to the sunitinib study than to the VEG105192 trial (approximately 60% versus 40%). Both studies restricted entry to patients with renal cell carcinoma with either clear cell or predominantly clear cell histology. The average age of patients in both studies was 60 years and 83–91% of patients had prior nephrectomy. The patient populations in the five interferon-α studies were generally similar. All patients had renal cell carcinoma, the age range was 60–66 years and 57–100% of patients had prior nephrectomy. Three of the interferon-α studies included some patients with a baseline ECOG performance status of 2.
3.5 Hazard ratios for progression free survival and overall survival from all seven studies were used in the indirect comparison to obtain hazard ratios for pazopanib versus interferon-α and pazopanib versus sunitinib. The hazard ratios for progression-free survival used in the indirect comparison for pazopanib versus placebo plus best supportive care, sunitinib versus interferon-α and interferon-α versus best supportive care were 0.36 (95% CI 0.24 to 0.55), 0.539 (95% CI 0.451 to 0.643) and 0.704 (95% CI 0.0.580 to 0.854). Median progression-free survival estimates derived from the indirect comparison for pazopanib, sunitinib, interferon-α and placebo plus best supportive care were 11.3 months (95% CI 5.1 to 17.5), 10.7 months (95% CI 7.9 to 13.4), 5.4 months (95% CI 5.4 to 5.4) and 5.6 months (95% CI 4.0 to 7.3) respectively.
3.6 The pazopanib overall survival value was estimated using the RPSFT weighted method. The manufacturer used hazard ratios derived from all seven studies in a Weibull survival model to estimate the median overall survival values for pazopanib versus interferon-α and pazopanib versus sunitinib. In the base-case scenario, derived using the RPSFT weighted method, the median overall survival estimates for pazopanib, sunitinib, interferon-α and placebo plus best supportive care were 27.8, 26.8, 15.8 and 12.1 months, respectively. These results indicated that pazopanib was associated with a decreased risk of death (37% reduction) compared with interferon-α, and that pazopanib appeared to have comparable efficacy with sunitinib in terms of overall survival. The manufacturer highlighted that the 95% confidence intervals around the hazard ratios were wide, indicating a level of uncertainty with the estimates. Sensitivity analyses were performed that included varying the hazard ratio for overall survival in the VEG105192 trial by using different methods for adjusting for crossover and varying the interferon-α studies included. The manufacturer concluded that the results of these sensitivity analyses of the indirect comparison were similar to those of the base-case analysis.
3.7 Only the VEG105192 trial, which compared pazopanib and placebo, and the study by Motzer et al. (2009), which compared sunitinib with interferon-α, reported health-related quality of life data. For the VEG105192 trial, there were no statistically significant differences between pazopanib and placebo for any of the instruments used (European Organisation for Research and Treatment of Cancer [EORTC] quality of life questionnaire – Core 30, EQ-5D, EQ-5D-VAS). For the comparison of sunitinib with interferon-α, patients receiving sunitinib had a statistically significantly better quality of life than patients receiving interferon-α, as measured by the EQ-5D, EQ-5D-VAS and the Functional Assessment of Cancer Therapy (FACT) Kidney Symptom Index Disease-related Symptom (FKSI-DRS) Index, FACT-Kidney Symptom Index – 15-item scale (FKSI-15 Index) and the FACT-General Scale (FACT-G). No indirect comparison was made for this outcome.
3.8 In the VEG105192 trial, 91% of patients receiving pazopanib experienced an adverse event. In 87% of patients the adverse events were related to study medication. In the placebo group 74% of patients experienced an adverse event and in 37% of patients these were related to study medication. The most frequent adverse events related to pazopanib treatment were diarrhoea, hair colour change, hypertension, nausea, anorexia and increased liver enzymes. The manufacturer conducted an indirect comparison using one of the interferon-α studies to estimate the adverse event rates of pazopanib compared with sunitinb. The adverse event rates for pazopanib were generally lower than for sunitinib, in particular for dyspepsia, mucositis/stomatitis, fatigue, hand-foot syndrome, skin discolouration, hypophosphataemia, anaemia and altered taste. However, only the difference in fatigue was statistically significant (HR 0.21, 95% CI 0.06 to 0.77).
3.9 The manufacturer submitted an economic model to assess the cost effectiveness of pazopanib in treatment-naive patients. The model compared pazopanib with interferon-α, sunitinib and best supportive care. The manufacturer described the model as a 'partitioned survival' model, characterised by three mutually exclusive health states: alive pre-progression, alive post-progression and dead. Unlike a Markov model, which models transitions between health states explicitly using transition probabilities, the partitioned survival model calculated the proportion of patients in each treatment arm at any time after starting treatment, using parametric survival curves fitted to empirical data on overall survival and progression-free survival over time. The proportion of patients in the 'alive post-progression' health state at any given time was calculated as the difference between overall survival and progression-free survival. In the model, pazopanib was assumed to be given until disease progression or death (if occurring before progression). After starting treatment, patients were assumed to be in an 'alive pre-progression' health state, and to be at risk of disease progression and/or death over time. Patients who experienced disease progression were assumed to discontinue treatment with pazopanib (and receive only best supportive care) and to transition to an 'alive post-progression' health state and to stay in that state until death.
3.10 For the model, a utility value of 0.70 was assumed for patients who had no disease progression and no adverse events, based on the mean EQ-5D utility value among patients without adverse events in the VEG105192 trial. Disease progression was assumed to be associated with a decrement in utility of 15% (that is, a post-progression utility value of 0.59). These values were used for all the interventions in the model. Utility decrements for adverse events were also obtained from the VEG105192 trial. As a result of the lack of published utility data in this patient population, the manufacturer commissioned a health state preference study to generate utility values for progression-free survival and post-progression survival and disutilities for treatment-related adverse events such as anaemia, diarrhoea, fatigue, hand-foot syndrome, nausea, mucositis and hypertension. The utility decrements for adverse events were used in a sensitivity analysis.
3.11 The manufacturer agreed a two-part patient access scheme with the Department of Health. Part A of the patient access scheme provides a 12.5% discount to the list price of pazopanib. Part B of the patient access scheme, the details of which are 'commercial-in-confidence', offers a future rebate linked to the outcome of the head-to-head COMPARZ trial. The manufacturer assumed that there would be no additional costs to the NHS associated with administering the patient access scheme. The costs considered in the economic model included acquisition costs for study medications, drug administration costs for infusions, costs of treating grade 3 or higher adverse events, routine follow-up costs, costs of progression and supportive care costs. In order to account for dose reductions and dose interruptions, the manufacturer adjusted the cost of study medication by using relative dose intensities reported in RCTs of the study treatments. In the model, the manufacturer used a dose intensity for pazopanib of 86%, equivalent to 688 mg per patient per day. Similar dose intensities were used for sunitinib (86%) and interferon-α (84%). Only the costs of treating adverse events that were grade 3 or higher with an incidence of at least 5% were considered for any treatment based on the indirect comparison. The cost per event was assumed to be independent of treatment.
3.12 For part A of the patient access scheme, which provides a 12.5% discount to the pazopanib list price, sunitinib was extendedly dominated by a combination of pazopanib and interferon-α. An option is 'extendedly dominated' when its ICER is higher than that of the next, more effective, option when compared with a common baseline (that is, it is dominated by a combination of two other alternatives). As a result, using the RPSFT weighted method of adjusting for crossover, the ICERs for pazopanib versus sunitinib, interferon-α and best supportive care were £1790, £38,925 and £32,898 per QALY gained respectively. These ICERs were derived from incremental costs of £122, £27,921 and £32,216, and incremental QALYs of 0.068, 0.717 and 0.979 respectively. The manufacturer provided additional cost-effectiveness analyses using alternative methods of adjusting for crossover. The ICERs for pazopanib versus sunitinib, interferon-α and best supportive care ranged from £1790 to £5327, £21,625 to £72,274 and £20,824 to £48,877 per QALY gained respectively. One-way sensitivity analyses showed that the ICER was most sensitive to the efficacy estimates for pazopanib versus interferon-α, which contribute to the relative efficacy of pazopanib and sunitinib. Specifically, the model is sensitive to the method used for adjusting for crossover for overall survival data from the VEG105192 trial. The manufacturer also provided cost-effectiveness estimates based on part B of the patient access scheme. However, these data are 'commercial-in-confidence' and therefore no details can be reported.
3.13 The ERG stated that the evidence base was not ideal for this appraisal as there were no data available from head-to-head comparisons of pazopanib with sunitinib or interferon-α. However, the ERG commented on the substantial amount of evidence on the efficacy and safety of pazopanib and on the considerable effort that had gone into providing these data, and the methods were generally well reported.
3.14 The ERG commented that the choice of model appeared to be appropriate given the decision problem and the data available. The time horizon appeared to be appropriate, although there were concerns that it may overestimate survival because the median age of diagnosis is 60–65 years and constant all-cause mortality was assumed, rather than taking data from life tables, which would have the impact of mortality increasing over time. The ERG stated that the results appeared valid with the methods used. Most of the model analyses performed could be replicated, although this was not true for all sensitivity analyses.
3.15 The ERG noted that in the manufacturer's original submission the base-case analysis was based on estimates from the model using the unweighted unadjusted RPSFT method to adjust for crossover and pooled interferon-α studies. In the updated analysis, provided as an addendum to the original submission, the method used for adjusting for crossover was the RPSFT weighted method, unadjusted. The ERG acknowledged that the manufacturer had presented a set of analyses that comprehensively covered the range of methods available for crossover. The ERG stated that the RPSFT weighted method advocated by the manufacturer was weakened by the lack of an adequately developed method to analyse relatively immature data robustly. The RPSFT weighted method provided a higher hazard ratio than the unweighted method, and although this did affect the results, the change in method had not necessarily been favourable to pazopanib.
3.16 The ERG also highlighted a concern about the manufacturer's assumption that as soon as a patient's disease progressed they stopped treatment. The ERG considered that, in practice, it is unlikely this will happen immediately because patients will only know the status of their disease when they have their next review, which may not be at the exact time the disease progresses. This assumption may create a small bias in favour of the more costly treatments such as pazopanib.
3.17 The ERG also stated that there was some uncertainty around the utility value estimate used. The manufacturer used an estimate that was based on the EQ-5D utility value among all patients without adverse events in the VEG105192 trial. The manufacturer also assumed this value to be similar for all interventions. The ERG noted, however, that given the minimal impact adverse events had on the model (in terms of QALYs and costs) this was not likely to be a major issue.
3.18 The ERG conducted an exploratory analysis to address concerns about the weighted unadjusted RPSFT results for overall survival being used for the base-case analysis. It assessed the potential impact of a robust weighted analysis on the results, particularly with a model adjusted for baseline covariates, for which methods are still in development. The ERG considered the impact of weighting by comparing the unweighted analyses with the weighted analyses when the models were unadjusted for baseline. The ERG concluded that, overall, weighting does have an impact on the hazard ratio, but it is difficult to establish the direction and magnitude of this effect. The ERG performed multivariate sensitivity analyses around utility value estimates for progression-free survival, utility decrement for progression and duration of utility with adverse events. The results of these analyses indicated that they are sensitive to some combinations of changes and that the ICER associated with pazopanib could increase to more than £50,000 per QALY gained. However, the confidence interval around the hazard ratio for overall survival was very wide. The ERG noted that the manufacturer only reported pair-wise probabilistic sensitivity analysis, therefore the ERG used the base case to compare all four treatment options. The ERG concluded that pazopanib is most likely to be the most cost effective option at willingness to pay thresholds between £35,000 and £50,000. However, the ERG stated that such analyses take the base case at face value and that it was not possible to explore the uncertainty of the various methods used to adjust for crossover or the point estimates used in the economic model due to limited data available for some parameters.
3.19 Full details of all the evidence are in the manufacturer's submission and the ERG report.