3 The manufacturer's submission
3.1 In the submission the manufacturer compared pemetrexed plus cisplatin (pemetrexed/cisplatin) with gemcitabine plus cisplatin (gemcitabine/cisplatin). The manufacturer justified this choice of comparator with marketing data that suggest gemcitabine plus a platinum drug accounts for 80% of first-line NSCLC treatment, and the fact that according to a meta-analysis and clinical opinion cisplatin is the preferred platinum drug. The manufacturer identified gemcitabine plus carboplatin (gemcitabine/carboplatin) and docetaxel plus cisplatin (docetaxel/cisplatin) as additional comparators. The manufacturer stated that carboplatin is still commonly used in the UK because patients do not need the same hydration that is necessary with cisplatin. It also stated that docetaxel is used occasionally because it requires fewer infusions than gemcitabine.
3.2 For the comparison of pemetrexed/cisplatin with gemcitabine/cisplatin the manufacturer identified one phase III, open-label, non-inferiority, randomised controlled trial (RCT). This trial (known as JMDB) compared 862 patients given pemetrexed/cisplatin with 863 patients given gemcitabine/cisplatin. It included patients with either squamous or non-squamous NSCLC and subgroups were defined by histology type, including adenocarcinoma, large-cell carcinoma and 'not otherwise specified'. Patients received up to six cycles of chemotherapy and were followed for 2.5 years. The trial results demonstrated overall survival (the primary outcome) of 10.3 months for both pemetrexed/cisplatin and gemcitabine/cisplatin for all randomised patients (hazard ratio [HR] 0.94, 95% confidence interval [CI] 0.84 to 1.05, p = 0.259). People with NSCLC of non-squamous histology had a greater overall survival with pemetrexed/cisplatin than with gemcitabine/cisplatin, based on median values (11 months versus 10.1 months respectively; HR 0.84, 95% CI 0.74 to 0.96, p = 0.011). A subgroup analysis based on median values showed that for patients with adenocarcinoma and large-cell carcinoma, overall survival was 11.8 months with pemetrexed/cisplatin compared with 10.4 months with gemcitabine/cisplatin (HR 0.81, 95% CI 0.70 to 0.94, p = 0.005). A similar subgroup analysis showed that patients with not otherwise specified histology had overall survival of 8.6 months for pemetrexed/cisplatin compared with 9.2 months for gemcitabine/cisplatin (HR 1.08, 95% CI 0.81 to 1.45, p = 0.586). The manufacturer concluded that these results together proved the hypothesis that pemetrexed/cisplatin was non-inferior to gemcitabine/cisplatin for overall survival in the overall JMDB trial population. It also stated that these results supported targeting pemetrexed/cisplatin treatment to the subgroup of patients with adenocarcinoma and large-cell carcinoma.
3.3 The difference in median progression-free survival between patients receiving pemetrexed/cisplatin and gemcitabine/cisplatin in all randomised patients was not significant: 4.8 and 5.1 months respectively (HR 1.04, 95% CI 0.94 to 1.15). In patients with non-squamous histology, median progression-free survival was 5.3 months for pemetrexed/cisplatin and 5.0 months for gemcitabine/cisplatin (HR 0.95, 95% CI 0.84 to 1.06). For the manufacturer's target group of patients with adenocarcinoma and large-cell carcinoma the progression-free survival was 5.3 months for pemetrexed/cisplatin and 4.7 months for gemcitabine/cisplatin (HR 0.90, 95% CI 0.79 to 1.02).
3.4 Pemetrexed/cisplatin was associated with statistically significantly fewer grade 3 and 4 adverse events than gemcitabine/cisplatin, specifically neutropenia, febrile neutropenia, thrombocytopenia, anaemia and alopecia. Patients receiving pemetrexed/cisplatin received fewer red blood cell transfusions, and less granulocyte colony stimulating factor and erythropoietin. Patients randomised to pemetrexed/cisplatin experienced statistically significantly more nausea. No quality of life data were measured in the JMDB clinical trial.
3.5 The manufacturer carried out an indirect comparison of pemetrexed/cisplatin with other comparators (gemcitabine/carboplatin and docetaxel/cisplatin). The manufacturer identified two phase II, open-label RCTs that could be mapped to the treatment arms of JMDB: Zatloukal et al. (2003) comparing gemcitabine/cisplatin (n = 87) with gemcitabine/carboplatin (n = 89) and Schiller et al. (2002) comparing gemcitabine/cisplatin (n = 301) with docetaxel/cisplatin (n = 304). All treatments were administered within their licensed indications. The trials were relatively homogenous in terms of patient population and when compared with the JMDB trial. The manufacturer noted that the unadjusted comparison suggested that median overall survival and progression-free survival were improved in patients with squamous and non-squamous NSCLC who were given pemetrexed/cisplatin relative to the other comparators.
3.6 The manufacturer's indirect comparison methodology involved calculating hazard ratios for each of gemcitabine/carboplatin and docetaxel/cisplatin, compared with gemcitabine/cisplatin. The hazard ratios were based on median overall survival and were applied to the hazard rate of the gemcitabine/cisplatin arm in the JMDB trial to produce hazard rates for gemcitabine/carboplatin and docetaxel/cisplatin, adjusted for the JMDB population. This was then used to calculate adjusted median overall survival estimates for the JMDB population. The manufacturer used this method to adjust the hazard rates for the subgroups by using the corresponding hazard rates in JMDB (such as for non-squamous NSCLC). The results of this analysis for the target population of patients with adenocarcinoma and large-cell carcinoma suggested an overall survival advantage for pemetrexed/cisplatin (11.8 months, 95% CI 10.4 to 13.2) versus gemcitabine/carboplatin (9.5 months, 95% CI 8.1 to 13.4) and docetaxel/cisplatin (9.8 months, 95% CI 8.6 to 11.5). Pemetrexed also improved progression-free survival: 5.3 months for pemetrexed/cisplatin compared with 3.8 months for gemcitabine/carboplatin and 4.1 months for docetaxel/cisplatin (no confidence intervals reported).
3.7 The manufacturer developed a Markov model with a 6-year time horizon that compared pemetrexed/cisplatin, gemcitabine/cisplatin, gemcitabine/carboplatin and docetaxel/cisplatin. The efficacy data from the JMDB trial were used for the comparison of pemetrexed/cisplatin with gemcitabine/cisplatin, and the results of the indirect comparison were used for the other comparators. The adverse event states were built into the model as separate mutually exclusive health states. All clinical events were modelled via transition probabilities. Treatment effects considered included overall survival, progression-free survival, response rates, adverse events and HRQoL. All effectiveness data used in the model, apart from HRQoL, were trial-based.
3.8 In the model, patients were given a maximum of four cycles of chemotherapy. A continuation rule stipulated that only patients whose disease had responded to pemetrexed/cisplatin after three cycles continued treatment to a fourth cycle. To reflect treatment discontinuation after the third cycle for patients whose disease did not respond, no further chemotherapy costs were incurred.
3.9 A literature review of utility data for patients with NSCLC identified a number of studies, but the manufacturer considered that none were suitable for inclusion. Instead, a study by Nafees et al. (2008) was used. This was commissioned to study second-line treatment of NSCLC by the manufacturer, but was assumed by the manufacturer to apply to first-line treatment. It involved 100 members of the public interviewed with visual analogue scale and standard gamble techniques to elicit societal values on utilities in lung cancer.
3.10 The base-case analysis compared pemetrexed/cisplatin with gemcitabine/cisplatin. In the population with non-squamous NSCLC, the analysis resulted in an incremental cost of £1364 and 0.041 incremental quality adjusted life years (QALYs). The incremental cost-effectiveness ratio (ICER) for pemetrexed/cisplatin compared with gemcitabine/cisplatin was £33,065 per QALY gained without the continuation rule (see 3.8). With the continuation rule the incremental cost fell to £1252 and the incremental QALY remained the same, resulting in an ICER of £25,967 per QALY gained. When subgroups according to histology were analysed using the continuation rule, pemetrexed/cisplatin compared with gemcitabine/cisplatin in the adenocarcinoma subgroup gave an ICER of £18,442 per QALY gained, and large-cell carcinoma gave an ICER of £8,056 per QALY gained.
3.11 The ERG reviewed the evidence submitted for clinical and cost effectiveness. The ERG report concentrated on the exclusion of vinorelbine, the indirect comparison and the suitability of the chosen cost-effectiveness analysis.
3.12 The ERG noted that vinorelbine had been excluded from the analysis even though the marketing data presented by the manufacturer suggested it accounted for 11% of first-line NSCLC treatment, which was greater than the 4% usage of docetaxel. The ERG considered that vinorelbine should have been included in the manufacturer's decision problem to allow a full assessment of pemetrexed against relevant comparators.
3.13 The ERG noted that in the JMDB trial, baseline characteristics were well balanced between treatment arms and between histological subgroups. The ERG noted that the findings from the per-protocol analysis requested from the manufacturer did not differ much from the findings from the intention-to-treat analysis. The ERG considered that this made the JMDB trial results considerably more robust. On request, the manufacturer reported the p values for the test for interaction as p = 0.0024 for squamous NSCLC compared with non-squamous NSCLC, and p = 0.0059 across all other subgroups. This makes it more likely that there were real differences between the histological subgroups.
3.14 The ERG expressed concerns over the trial selection for the indirect comparison. The ERG believed that all the comparators specified in the scope (pemetrexed, docetaxel, gemcitabine, paclitaxel and vinorelbine) should have been included in the indirect comparison analyses. This would have identified five further phase III RCTs for consideration, and improved the subsequent power and validity of the indirect comparison. The ERG also noted that the manufacturer did not assess validity of the included RCTs.
3.15 The ERG also expressed concern over the statistical approach used in the indirect comparison. It noted that the manufacturer's method may have resulted in under- or overestimation of treatment effects, and loss of statistical power. It also noted that the manufacturer's submission suggested that the treatment-arm-level hazard rates were used; the ERG stated that indirect comparisons should be based on a comparison of relative effects rather than a comparison of single arm estimates, as the former maintains randomisation within a trial. The ERG stated that the key assumption of an indirect comparison is that the relative effects are exchangeable across the trial settings, that is, there are no treatment effect modifiers. Within the JMDB trial, histology is an effect modifier, and this should be accounted for in the indirect comparison. The ERG concluded that, because key comparators were excluded from the indirect comparison analysis, and because of the assumptions underlying the statistical approach used, the findings from this analysis should be interpreted with caution.
3.16 The ERG commented on the submitted cost-effectiveness analysis. It noted that the chosen Markov model structure did not seem to be appropriate because it did not replicate the trial data, which was used to calibrate the model, to an acceptable level of accuracy. The ERG commented that this was noticeable when calculating response and survival. It considered that because overall survival and progression-free survival were the primary outcomes in the JMDB trial, these two outcomes should be accurately replicated in the economic model for each of the subgroups for the trial period. It noted that the manufacturer's model appeared to overestimate overall survival in both arms and almost all patient groups. For progression-free survival, the ERG commented that the model tended to underestimate in the first 6 months and to overestimate thereafter. In addition, the ERG noted that some survival estimates suggested an error in the model's logic.
3.17 The ERG commented on the use of response to treatment in the model structure. It is commonly assumed that response leads to a delay in disease progression and therefore to progression-free survival, this becoming the source of survival gain. Following disease progression it is usually assumed that the natural course of the disease will continue. The JMDB trial data suggested that all the reported survival gain occurred after disease progression, with progression-free survival effectively identical between the pemetrexed/cisplatin and gemcitabine/cisplatin arms. The ERG stated that it was not clear whether objective response determined the extent of health gain and whether the survival gain was restricted to patients whose disease has responded to treatment, or to all patients who had treatment. The ERG considered that this had implications for the design of the model; if response doesn't predict progression-free survival or post-progression survival, then its use as a distinct health state is potentially irrelevant, and could generate misleading results.
3.18 The ERG identified other concerns with the cost-effectiveness analysis, including:
All transition probabilities during the trial period were assumed to arise from constant risk processes (that is, exponential survival distributions), without any justification.
A half-cycle correction appeared to have been disabled for costs and used incorrectly for outcomes.
Cumulative costs and outcome effects of patients having more than one adverse event at any given time (for example, within a single hospital admission) were not taken into account. This omission could have led to overestimation of the costs and harms attributable to treatment.
There may have been an overestimation of mortality because of incorrect use of the febrile neutropenia mortality risk.
3.19 The ERG stated that the evidence submitted by the manufacturer was not sufficiently convincing or robust for it to determine the cost effectiveness of pemetrexed.
3.20 During the consultation for this appraisal, the manufacturer submitted revised cost-effectiveness estimates for pemetrexed/cisplatin compared with gemcitabine/cisplatin. No other comparators were considered. The primary analysis was a modified version of the previously submitted Markov model, but used Weibull distributions to improve its representation of the outcomes of the JMDB trial. The manufacturer responded to the concerns raised by the Committee concerning the use of response, transition probabilities, half-cycle correction, adverse events and mortality due to febrile neutropenia. It also presented two validation models: a trial-based economic analysis conducted using the individual patient survival outcomes and resource use events from the JMDB clinical trial database, and an economic model used for a submission to the Pharmaceutical Benefits Advisory Committee (PBAC) in Australia. The PBAC model was based on the patient-level data from the clinical trial and used Weibull distributions to extrapolate survival beyond the trial period. The manufacturer stated that validation processes included a 'double-build' process for the trial-based model (in which two researchers independently built and analysed the database to make sure data outputs were consistent), and internal and independent external reviews, for both the modified and clinical trial-based models.
3.21 The manufacturer's base-case ICER, using the modified Markov model calculated for a maximum of four cycles of treatment, was £27,565 for the population included in the licence (those with non-squamous histology) and £22,202 for patients with adenocarcinoma or large-cell carcinoma. For the trial-based analysis, the ICER calculated for a maximum of four cycles of treatment was £31,157 for the population included in the licence and £24,224 for patients with adenocarcinoma or large-cell carcinoma. When the number of cycles was increased to six, as specified in the trial, the ICERs increased to £42,306 and £33,730 for the two groups respectively. In the PBAC model, the ICER for only four cycles of chemotherapy for patients with adenocarcinoma or large-cell carcinoma was £23,157 per QALY gained.
3.22 The ERG commented on the manufacturer's additional analysis. It stated that the cost-effectiveness analyses based on the JMDB trial patient-level data without use of projection techniques were very similar to the previous cost-effectiveness models, and used the same unit cost and state utility parameter values.
3.23 However, the ERG noted several limitations with the submitted analyses. These included restricting the number of cycles and corresponding costs, with no corresponding alteration in effectiveness. Therefore, only the estimates using six cycles were valid trial-based estimates. The ERG further noted that new utility values were used in the revised model without explanation. The ERG considered that, as all survival benefit observed for pemetrexed in the JMDB trial occurred after disease progression, the correct utility value for use with the incremental survival is that of the 'progressive disease' state from the original Markov model (that is, 0.47), not that of the pre-progression states of 'stable' (0.65) and 'responding' (0.67).
3.24 The ERG noted that the estimates for the cost of chemotherapy did not consider differences in body surface area, or allow for wastage of part-used vials. The ERG suggested that taking these factors into account increased the cost per cycle of pemetrexed/cisplatin chemotherapy by £81.63 and decreased that of gemcitabine/cisplatin by £3.80.
3.25 The ERG noted that the 'in-trial' analysis did not use discounting on either costs or outcomes, despite trial follow-up extending to more than 2 years for some patients. The ERG stated that this was an important omission, because much of the survival gain occurred after the first 12 months and would therefore be likely to be affected by discounting. Drug costs, however, would be incurred early on. The ERG noted that the 'in-trial' analysis used differential costs per patient for terminal care and for best supportive care (BSC). However, these figures were not derived from an analysis of the trial's individual patient data, but were mean results calculated in the manufacturer's Markov model. This created confusion between observation and modelling, which may have distorted the results of the 'in-trial' analysis. The ERG preferred to include terminal care and BSC costs for all patients, but discounted for a period after the recorded survival date for patients censored in the trial.
3.26 The combined ERG amendments to the in-trial analysis, using the utility derived from disease progression (0.47) and up to six cycles of chemotherapy, produced an ICER for the population included in the licence of £60,130, and £48,055 for the adenocarcinoma and large-cell carcinoma subgroup.
3.27 The ERG noted that the manufacturer's modified Markov model addressed a number of the issues identified by the ERG previously. However, it noted that although the Weibull survival models were better than the original exponential models, they were still not adequate. In particular, they were inaccurate for long-term projection. The ERG also noted that patients having more than one adverse event at a time (for example, during one hospital admission) was not addressed, and chemotherapy costs were based on JMDB trial data and were therefore not representative of UK clinical practice. There was also the issue of reducing six cycles to four, and the effects of this on overall efficacy. In addition the ERG identified new errors in the analysis, including the calculation of adverse event costs, and inappropriate response rates used for the whole population.
3.28 The ERG commented that the PBAC health technology assessment submission was well presented and clearly laid out, thereby simplifying the validation. However, because it was based on the same fundamental assumptions as the manufacturer's Markov analysis, it merely demonstrated that similar assumptions resulted in similar cost-effectiveness results when using a different model structure. The ERG concluded that it did not address some of the major issues with the manufacturer's cost-effectiveness analysis that had been identified previously.
3.29 The ERG stated that the time available to review the new evidence submitted by the manufacturer did not allow detailed modifications to be made to the modified Markov model. Instead it used the information contained in the 'in-trial' analysis, together with the additional exploratory survival analysis, to generate modified cost-effectiveness results.
3.30 The ERG noted that an extract of individual patient data from the JMDB trial was included by the manufacturer in the 'in-trial' cost-effectiveness analysis. This was restricted to the population of patients with NSCLC and included only information relating to chemotherapy treatment cycles and overall survival, that is, the timing of death or censoring. No information was provided about response to treatment or the time of confirmed disease progression. This data made it possible for the ERG to consider what was the most appropriate estimate of survival gain and utility gain attributable to pemetrexed within the JMDB trial, and thus whether it was possible to estimate the likely change in patient outcomes when treatment was limited to four cycles instead of the maximum of six cycles used in the trial. The ERG classified patients according to the last cycle in which they received a dose of pemetrexed or gemcitabine. Initial examination of Kaplan-Meier survival charts by the ERG indicated that patients could be classified into three groups that were mainly homogeneous with respect to prognosis: up to two cycles, three to four cycles and five to six cycles of chemotherapy. In the absence of specific information on disease progression or treatment discontinuation, these divisions should reflect the approximate time when patients leave the stable or response states. The ERG considered that this analysis provided a basis for considering the possible effects of limiting treatment duration.
3.31 The results of the ERG's exploratory analysis suggested that for six cycles of chemotherapy, the ICER for pemetrexed/cisplatin compared with gemcitabine/cisplatin was £28,241 per QALY gained for non-squamous patients and £23,598 per QALY gained for adenocarcinoma and large-cell carcinoma patients. When the number of cycles was reduced to four the ICERs were £20,497 and £17,162 per QALY gained respectively.
3.32 The ERG explored two scenarios to account for the potential consequences of reducing the number of chemotherapy cycles. First, if overall survival is related to tumour response, the overall survival gain lost when chemotherapy is stopped sooner can be estimated from the response rate difference (19%). Secondly, if overall survival is related to drug exposure, the overall survival gain lost when chemotherapy is stopped sooner can be estimated as the proportion of treatment cycles given beyond four cycles (32%).
3.33 For four cycles of pemetrexed/cisplatin in the population included in the licence (those with non-squamous histology), the exploratory analyses described in 3.32 led to an ICER of £25,336 for a 19% reduction in the overall survival gain, and £30,142 for a 32% reduction in the overall survival gain. For the treatment of patients with adenocarcinoma and large-cell carcinoma subgroup, the respective ICERs were £21,214 and £25,239.
3.34 The ERG noted that gemcitabine's patent ended this year (2009), and that generic versions are already being marketed. The ERG explored the potential impact of some market price changes, and noted that they adversely affected the cost-effectiveness estimates for pemetrexed/cisplatin.
3.35 Full details of all the evidence are in the manufacturer's submission and the ERG report.