The Appraisal Committee considered evidence submitted by the manufacturer of pemetrexed and a review of this submission by the Evidence Review Group (ERG).
The manufacturer's submission contained evidence on the clinical effectiveness of pemetrexed maintenance therapy compared with best supportive care. The manufacturer stated that pemetrexed is the only chemotherapy currently licensed for the maintenance treatment of non-small-cell lung cancer in the UK and worldwide. Therefore, the comparator used in the clinical trial was placebo plus best supportive care.
The manufacturer identified one phase 3 multicentre, double-blind randomised control study (the JMEN trial) which evaluated the efficacy of maintenance treatment with pemetrexed monotherapy in people with advanced or metastatic (stage IIIB and IV) non-small-cell lung cancer whose disease had not progressed following treatment with platinum-based first-line chemotherapy. All patients had an Eastern Cooperative Oncology Group (ECOG) performance status of 0 or 1. The trial randomised 663 patients with squamous and non-squamous non-small-cell lung cancer to pemetrexed plus best supportive care (n=441) or placebo plus best supportive care (n=222). Patients in both arms of the trial received concomitant medication with folic acid, vitamin B12 and dexamethasone. Patients in the pemetrexed arm received pemetrexed 500 mg per m2 on day 1 of each 21-day cycle, administered as a 10-minute infusion, plus best supportive care, until disease progression. Patients in the placebo arm received normal saline (0.9% sodium chloride) on day 1 of each 21-day cycle, administered as a 10-minute infusion, plus best supportive care, until disease progression. The manufacturer presented evidence for the subgroup of non-squamous non-small-cell lung cancer (n=481) in accordance with the licensed indication. Of this subgroup, 325 patients received pemetrexed plus best supportive care and 156 received placebo plus best supportive care.
The mean number of pemetrexed cycles for the non-squamous population was 8.0 (standard deviation 8.62) and the median was 6.0 cycles (25th to 75th percentile 2.5 to 10.0). There were a few patients who received between 20 and 55 cycles (7% to 11% of patients received more than 20 cycles).
The primary outcome of the JMEN trial was initially overall survival, but this was changed to progression-free survival during the trial. Median progression-free survival was significantly longer with pemetrexed plus best supportive care compared with placebo plus best supportive care (4.5 months versus 2.6 months, hazard ratio [HR] 0.44, 95% confidence interval [CI] 0.36 to 0.55, p<0.00001). A subgroup analysis for patients with adenocarcinoma (a type of non-squamous non-small-cell lung cancer) reported similar improvement in progression-free survival with pemetrexed plus best supportive care compared with placebo plus best supportive care (4.7 months versus 2.6 months, HR 0.45, 95% CI 0.35 to 0.59, p<0.00001). Secondary outcomes of the JMEN trial included tumour response, disease control rate and time to worsening of symptoms. The JMEN trial demonstrated a statistically significant median overall survival benefit of 5.2 months for the non-squamous population in favour of pemetrexed compared with placebo (15.5 months versus 10.3 months, HR 0.70, 95% CI 0.56 to 0.88, p=0.002). Similar results were reported for the adenocarcinoma subgroup. For the non-squamous population, 1-year overall survival in the pemetrexed plus best supportive care arm was 60% compared with 42% in the placebo arm. The difference in overall survival was smaller at 2 years (28% for pemetrexed compared with 22% for placebo). The trial reported similar results for the 1- and 2-year overall survival in the adenocarcinoma subgroup. Statistically significant improvements in tumour response, disease control rate and time to worsening of symptoms were reported for pemetrexed plus best supportive care compared with placebo plus best supportive care. The manufacturer's submission noted the absence of trial-based health-related quality-of-life data because many of the patients did not complete quality-of-life surveys.
The manufacturer's submission reported higher rates of grade 3 and 4 adverse events with pemetrexed plus best supportive care than with placebo plus best supportive care (6.3% versus 2.3%). Fatigue and neutropenia were the most commonly reported adverse events. There were significantly higher percentages of patients in the pemetrexed arm who discontinued treatment, required transfusion, erythropoiesis-stimulating agents or hospitalisation because of drug-related toxicity, or withdrew from the study.
The manufacturer developed a trial-based model which included three health states (not progressed, progressed and terminal state). Patients entered the model at the start of maintenance treatment, which was assumed to begin after four cycles of first-line chemotherapy (consisting of a platinum doublet with gemcitabine, paclitaxel or docetaxel) in patients who had no evidence of disease progression. Patients in the placebo arm received 'watch and wait' treatment and best supportive care, and patients in the pemetrexed arm received treatment plus best supportive care in 21-day cycles until disease progression. After disease progression patients were eligible for second-line treatment.
The economic model had a time horizon of 72 months (29-month overall survival data from the JMEN trial extrapolated to 72 months using an exponential survival function). Treatment effects that were included in the model were overall survival, adverse events and health-related quality of life. All effectiveness data used in the model, apart from health-related quality of life, were trial based. Trial data on progression-free survival were not used in the economic model. The number of treatment cycles in the trial was used as a proxy for the time to progression in the pemetrexed arm. The disutility of adverse events was not included in the base-case model but was captured in the sensitivity analyses.
In the JMEN trial, patients received pemetrexed treatment until their disease progressed. Although this resulted in patients receiving up to 55 cycles (with a mean of 8 cycles), the manufacturer's submission noted that clinical specialists suggested that if maintenance treatment were introduced to UK clinical practice, patients would receive a maximum of 10 cycles of pemetrexed maintenance treatment. The manufacturer therefore incorporated a 'capping rule' in which the maximum number of cycles of pemetrexed was set at 1 standard deviation above the mean, equivalent to a maximum of 17 cycles (with a new mean of 5.84) for the non-squamous population, and a maximum of 18 cycles (with a new mean of 6.16) for the adenocarcinoma population. The new means were used in the manufacturer's base case.
In the absence of data on health-related quality of life from the JMEN trial, utility data were taken from literature estimates. The manufacturer mainly used a study on the second-line treatment of non-small-cell lung cancer by Nafees et al. (2008). It involved 100 members of the public interviewed with visual analogue scale and standard gamble techniques to generate societal values on utilities in lung cancer. In addition, the manufacturer also used data from a study by Berthelot et al. (2000). Based on these two studies, the manufacturer assigned a utility of 0.66 to patients on pemetrexed and 0.58 to patients on placebo.
The manufacturer's base-case analysis compared pemetrexed plus best supportive care with placebo plus best supportive care in the non-squamous population. The incremental cost-effectiveness ratio (ICER) for pemetrexed compared with best supportive care in the non-squamous population was calculated to be £33,732 per QALY gained, based on an incremental cost of £9,137 and an incremental QALY of 0.27. The ICER for the adenocarcinoma subgroup was £39,364 per QALY gained, based on an incremental cost of £9,554 and an incremental QALY of 0.24.
The manufacturer also presented the ICERs for 36 one-way sensitivity analyses and a number of scenarios that explored the effect of per-vial costing and cycle capping, and included a best-case and worst-case scenario. Most of the results in the one-way sensitivity analyses had little effect on the base-case ICERs. However, two results did have a large effect:
When the incremental survival of pemetrexed was reduced from 5.3 months in the base case to 1.15 months, the ICER increased to £105,826 per QALY gained.
When the overall survival advantage was reduced by 9.5%, to allow for the patients excluded with the base-case capping rule, the ICER increased to £48,290 per QALY gained.
The ERG reviewed the evidence submitted for clinical and cost effectiveness, focussing on the non-squamous population in accordance with the licensed indication. The ERG stated that the JMEN trial was reasonably well designed, incorporating blinding, placebo control and independent monitoring of investigator assessments. The clinical outcomes reported from the trial addressed the outcomes that were relevant to the appraisal (overall survival, progression-free survival, tumour response, adverse events and health-related quality of life).
The ERG raised concern about the conduct of the trial, its generalisability to the UK patient population and the uncertainty around the estimates of cost effectiveness. The ERG noted that the inclusion criteria of the JMEN trial were restricted to younger patients with a good performance status (ECOG 0 or 1) and with few comorbidities. Only a relatively small proportion of the total number of non-small-cell lung cancer patients treated in clinical practice in the UK has an ECOG performance status of 0 or 1.
The ERG did not consider that adequate justification was given for changing the primary endpoint of the JMEN trial from overall survival to progression-free survival. It considered that this decision had the effect of truncating the data available for analysis of overall survival, which was of critical importance to the economic evaluation. The ERG also considered the high rate of missing data on health-related quality of life to be a limitation. It was not clear how patients' quality of life would be affected by maintenance treatment with pemetrexed.
The ERG noted that 53% of patients in the pemetrexed arm and 36% of patients in the placebo arm of the JMEN trial received second-line treatments that are not used in UK clinical practice. This may have influenced the overall survival estimates observed in the trial and may mean that the results of the trial do not reflect the survival benefits that might be expected in UK clinical practice.
The ERG was concerned that the key clinical evidence was derived from a histological subgroup of the trial population, but that histology was not included in the stratification for the randomisation procedure.
The ERG assessed the manufacturer's cost-effectiveness analysis. It commented on the version of the model which used the exponential (rather than Weibull) projection as the basis for comparison (this being the manufacturer's base case). The ERG noted that the capping of pemetrexed treatment at 17 cycles was much less than the maximum of 55 cycles in the JMEN trial. The ERG considered that this limited the costs of maintenance treatment with no similar limitation on the benefits accrued from the use of pemetrexed, which led to bias in favour of pemetrexed. The ERG considered that the most appropriate base case should have included the full costs and benefits of maintenance treatment based on the number of cycles received in the JMEN trial. The ERG conducted an analysis in which the number of treatment cycles was not capped. This increased the ICER from £33,732 per QALY gained to £43,179 per QALY gained.
The ERG considered that the discounting applied in the model was based on inappropriate assumptions. All maintenance chemotherapy cycles were assumed to occur in the first year (consistent with the imposed maximum cycles limit but not with the trial data), all second-line chemotherapy took place in the first year, all best supportive care was assumed to occur only in years 1 or 2 and all terminal care was assigned to year 3.
The ERG did not consider the additional monitoring of patients on pemetrexed chemotherapy (who were assessed every two cycles) to be consistent with UK clinical practice. It considered the appropriate follow up to be at 3, 6 and 12 months and every 6 months thereafter until progression for the best supportive care arm, and every four cycles (12 weeks) until progression in the pemetrexed arm. The ERG also noted that the body surface area distribution used in the model was not representative of the UK population because 35% of the trial population was Asian (from China, Korea, Taiwan and India).
The ERG noted that no direct use was made in the model of the primary trial outcome (progression-free survival) and the duration of maintenance therapy was used as a proxy. The ERG also expressed concerns that in the model the overall survival of patients who received second-line treatment was assumed to be the same as those who did not.
The ERG did not consider it appropriate for patients entering the model at randomisation who were in the same health state (without disease progression) to be assigned different utility values (0.66 for patents in the pemetrexed arm and 0.58 for patients in the placebo arm). This was not consistent with data from the JMEN trial in which the rate of grade 3 or 4 fatigue was noticeably higher in the pemetrexed arm (3.66%) than in the placebo arm (0.64%). When the ERG used utility values which incorporated the disutility associated with adverse events (0.6568 in the pemetrexed arm and 0.6628 in the placebo arm) the ICER increased from the base case of £33,732 per QALY gained to £36,798 per QALY gained.
The ERG considered that the manufacturer did not adequately justify the choice of parameters and parameter values used in the one-way sensitivity analyses. The ERG also expressed concern that a probabilistic sensitivity analysis had not been undertaken. When the ERG conducted an approximate probabilistic sensitivity analysis based on the overall survival gain and the mean number of treatment cycles from the individual patient data the cost-effectiveness acceptability curve showed that pemetrexed maintenance treatment would have zero probability of being cost effective at a threshold of £30,000 per QALY gained and 50% probability of being cost effective at a threshold of approximately £51,000 per QALY gained.
The ERG identified other concerns with the cost-effectiveness analysis, including:
The half-cycle correction applied to survival estimates appeared to be inappropriate. The ERG considered that the correct approach would be to use the area under the curve from the trial analysis unaltered, and then calculate 'mid-cycle' corrected estimates for the remainder of the model duration derived from a parametric model.
Post progression costs and survival values had been double discounted.
A minor error in the calculation of the proportion of patients assumed to receive docetaxel or erlotinib as second-line treatment. When this was corrected, the manufacturer's base-case ICER increased slightly.
The ERG investigated the impact of unlimited cycles of treatment, revised utility values, revised discounting assumptions, and increased cost of monitoring based on a model populated with individual patient data. The cumulative effect of these changes was an increase in the ICER for pemetrexed maintenance treatment from the manufacturer's estimated base case of £33,732 per QALY gained to £51,192 per QALY gained. The number of treatment cycles and utility revision had the most impact on the ICER.
The manufacturer presented a revised cost-effectiveness analysis to address the concerns raised by the Committee. The revised analysis included a probabilistic sensitivity analysis with an exponential extrapolation survival function and presented six scenarios in which the duration of treatment and the utility values in the pemetrexed and placebo arms were varied (Three different treatment durations were presented, each with two possible utility assumptions, giving a total of six scenarios). The different treatment durations considered were: 1 year (a maximum of 17 cycles), 2 years (a maximum of 35 cycles) and treatment until disease progression in accordance with the JMEN trial (a maximum of 55 cycles). The survival benefits modelled for each treatment duration were consistent with those seen in trial patients. Utility was either the same in both arms (0.66) or a lower utility was assigned to the pemetrexed arm (0.657) compared with the placebo arm (0.663). The ICERs for pemetrexed compared with best supportive care ranged from £46,137 per QALY gained to £50,286 per QALY gained, with a 46% to 58% probability of being cost effective at a threshold of £50,000 per QALY gained.
The ERG commented on the manufacturer's revised analysis and examined scenario 5 in detail. This scenario represented treatment until disease progression, used the entire trial population and incorporated a utility of 0.663 for the placebo arm and 0.657 for the pemetrexed arm. The ERG noted that most of the changes made by the manufacturer were those required to accommodate a probabilistic sensitivity analysis. The ERG also noted that the changes were implemented appropriately.
The ERG conducted a probabilistic sensitivity analysis on scenario 5 which incorporated all of the amendments suggested in their original analysis (see section 3.23). The ERG also presented the results of a probabilistic sensitivity analysis using an exponential and a Weibull extrapolation of the trial data. The ICER for pemetrexed compared with best supportive care using the exponential survival function was £56,903 per QALY gained using deterministic analysis and £47,168 per QALY gained using probabilistic analysis, with a 57.71% probability of being cost effective at a threshold of £50,000 per QALY gained. When the Weibull function was applied, the ICER for pemetrexed compared with best supportive care was £57,082 per QALY gained using deterministic analysis and £50,673 per QALY gained using probabilistic analysis, with a 49.70% probability of being cost effective at a threshold of £50,000 per QALY gained.
Full details of all the evidence are in the manufacturer's submission and the ERG report.