3.1 The manufacturer's decision problem compared gefitinib with gemcitabine and carboplatin, paclitaxel and carboplatin, vinorelbine and cisplatin, and gemcitabine and cisplatin. The decision problem defined the population as patients with locally advanced or metastatic NSCLC who are previously untreated and who test positive for an EGFR-TK mutation (EGFR-TK mutation-positive patients). Outcomes were defined as overall survival, progression-free survival, objective tumour response rates, health-related quality of life and adverse events associated with treatment. In the economic evaluation the incremental cost per quality-adjusted life year (QALY) gained was presented. A lifetime horizon was used, and costs were considered from the perspective of the NHS and personal social services (PSS).
3.2 The manufacturer's submission presented clinical effectiveness data from the Iressa Pan Asian Study (IPASS), a randomised controlled trial (RCT) set in East Asia. IPASS was a multicentre, open-label RCT in clinically selected patients older than 18 years who had the following characteristics: histologically or cytologically confirmed stage IIIb (locally advanced disease such as pleural effusion not amenable to local therapy) or stage IV (metastatic) NSCLC with adenocarcinoma histology (including bronchoalveolar carcinoma), had never smoked (or had smoked fewer than 100 cigarettes per lifetime) or had been light smokers (stopped smoking at least 15 years previously and had smoked no more than 10 pack-years), had no prior chemotherapy, biological or immunological therapy, and had a WHO performance status of 0, 1 or 2 (on a scale of 0 to 4, with low values reflecting better health).
3.3 IPASS included 1217 patients from 87 East Asian centres. Patients were randomised to receive 250 mg of gefitinib once daily or paclitaxel (200 mg/m2 body surface area) immediately followed by carboplatin (at a dose corresponding to an area under the curve [AUC] of concentration versus time of 5.0–6.0 minute.mg/ml) in 3-weekly cycles. Treatment was continued until disease progression (according to Response Evaluation Criteria in Solid Tumours [RECIST], which used tumour measurement rather than investigator assessment), unacceptable adverse events, a patient or clinician request to discontinue, severe non-adherence to the protocol, or until six chemotherapy cycles were reached. Following disease progression, all patients in the gefitinib arm were offered treatment with paclitaxel and carboplatin; if the patient declined or the combination was considered unsuitable, the clinician chose an approved therapy. Following disease progression on paclitaxel and carboplatin treatment, choice of treatment was at the clinician's discretion.
3.4 The manufacturer's submission focused on a subgroup of 261 EGFR-TK mutation-positive patients from the overall IPASS population. This subgroup accounted for 21% of the IPASS population. Of these patients, 80.8% were women. Most patients (94.3%) had never smoked, 5.4% had been light smokers and 0.4% were ex-smokers. On a scale of 0 (good) to 4 (poor), most patients (65.9%) had a WHO performance status of 1; 26.4% had a WHO performance status of 0 and 7.7% had a WHO performance status of 2. Most patients had tumours with histology indicating adenocarcinoma (94.6%); 5.4% had histology indicating bronchocarcinoma and none had unknown histology. At study entry most patients had metastatic disease (81.6%); 18.4% had stage IIIb locally advanced disease. Baseline characteristics in the subgroup were similar between both treatment arms.
3.5 The primary outcome examined in IPASS was progression-free survival, which was assessed from the date of randomisation to disease progression (determined by RECIST) or death from any cause. Secondary outcomes included overall survival, objective tumour response rate, health-related quality of life, symptomatic improvement, safety and tolerability. Estimates of overall survival in the overall population were based on an interim analysis after 450 deaths (37% of study participants), as well as modelled values reflecting median overall survival. The final analyses are due in the second quarter of 2010. Health-related quality of life was assessed by the Functional Assessment of Cancer Therapy–Lung (FACT–L) and the Trial Outcome Index (TOI), calculated from the domain scores from FACT–L representing physical and functional wellbeing, and lung cancer symptoms (LCS).
3.6 To assess the non-inferiority of gefitinib compared with paclitaxel and carboplatin, analysis of progression-free survival used a Cox proportional hazard model adjusting for baseline covariates in the intention-to-treat population.
3.7 In the overall study population, patients randomised to receive gefitinib had a statistically significantly longer progression-free survival compared with patients randomised to receive paclitaxel and carboplatin. The hazard ratio (HR) for progression-free survival (gefitinib compared with paclitaxel and carboplatin) was 0.74 (95% confidence interval [CI] 0.65 to 0.85, p < 0.0001). The objective tumour response rate was statistically significantly higher for gefitinib compared with paclitaxel and carboplatin (43.0% versus 32.2%; odds ratio [OR] 1.59, 95% CI 1.25 to 2.01, p = 0.0001). The estimates of overall survival in the overall study population were similar for both groups (HR for gefitinib compared with paclitaxel and carboplatin 0.91, 95% CI 0.76 to 1.10).
3.8 In the subgroup of EGFR-TK mutation-positive patients (n = 261), progression-free survival in patients randomised to receive gefitinib was statistically significantly longer than for patients randomised to receive paclitaxel and carboplatin (HR 0.48, 95% CI 0.36 to 0.64, p < 0.0001). Median progression-free survival was 9.5 months for patients randomised to receive gefitinib and 6.3 months for patients randomised to receive paclitaxel and carboplatin. The objective tumour response rate was statistically significantly higher for patients randomised to receive gefitinib compared with patients randomised to receive paclitaxel and carboplatin (71.2% versus 47.3%; OR 2.75, 95% CI 1.65 to 4.60, p = 0.0001). There was no statistically significant difference in the estimates of overall survival for patients randomised to receive gefitinib compared with patients randomised to receive paclitaxel and carboplatin (HR 0.78, 95% CI 0.50 to 1.20).
3.9 In the subgroup of EGFR-TK mutation-negative patients (n = 176), progression-free survival in patients randomised to receive gefitinib was statistically significantly shorter than for patients randomised to receive paclitaxel and carboplatin (HR 2.85, 95% CI 2.05 to 3.98, p < 0.0001). Median progression-free survival was 1.5 months for patients randomised to receive gefitinib and 5.5 months for patients randomised to receive paclitaxel and carboplatin (that is, EGFR-TK-negative patients randomised to receive gefitinib had shorter progression-free survival than patients randomised to receive conventional chemotherapy). The objective tumour response rate was statistically significantly lower with gefitinib than with paclitaxel and carboplatin (1.1% versus 23.5%; OR 0.04, 95% CI 0.01 to 0.27, p = 0.0013). There was no statistically significant difference in the estimates of overall survival for patients randomised to receive gefitinib compared with those randomised to receive paclitaxel and carboplatin (HR 1.38, 95% CI 0.92 to 2.09).
3.10 In the overall study population, statistically significantly more patients randomised to receive gefitinib experienced a clinically relevant improvement in health-related quality of life and disease symptoms, assessed by the FACT–L and TOI, than patients randomised to receive paclitaxel and carboplatin (FACT–L – OR 1.34, 95% CI 1.06 to 1.69, p = 0.0148; TOI – OR 1.78, 95% CI 1.40 to 2.26, p < 0.0001). Rates of symptomatic improvement were measured using the lung cancer symptoms (LCS) domain of the FACT–L and were similar for patients randomised to receive gefitinib and patients randomised receive to paclitaxel and carboplatin.
3.11 Similarly in the subgroup of EGFR-TK mutation-positive patients, statistically significantly more patients randomised to receive gefitinib experienced a clinically relevant improvement in health-related quality of life and disease symptoms than patients randomised to receive paclitaxel and carboplatin (FACT–L – OR 3.01, 95% CI 1.79 to 5.07, p < 0.0001; TOI – OR 3.96, 95% CI 2.33 to 6.71, p < 0.0001; LCS – OR 2.70, 95% CI 1.58 to 4.62, p = 0.0003). Time to worsening of health-related quality of life and disease-related symptoms was longer for patients randomised to receive gefitinib than for patients randomised to receive paclitaxel and carboplatin (median range 11.3–16.6 months for gefitinib and 2.9–3.0 months for paclitaxel and carboplatin).
3.12 In the subgroup of EFGR-TK mutation-negative patients, statistically significantly more patients randomised to receive paclitaxel and carboplatin had a clinically relevant improvement in health-related quality of life and disease-related symptoms than patients randomised to receive gefitinib (FACT–L – OR 0.31, 95% CI 0.15 to 0.65, p = 0.0021; TOI – OR 0.35, 95% CI 0.16 to 0.79, p = 0.00111; LCS – OR 0.28, 95% CI 0.14 to 0.55, p = 0.0002). Time to worsening of health-related quality of life and disease-related symptoms was similar or shorter for patients randomised to receive gefitinib compared with patients randomised to receive paclitaxel and carboplatin (median 1.4 months for gefitinib versus 1.4–4.2 months for paclitaxel and carboplatin).
3.13 The manufacturer's submission did not provide an analysis of adverse events according to EGFR-TK mutation status. The manufacturer's submission stated that in the overall populations gefitinib was associated with fewer grade 3 or 4 adverse events than paclitaxel and carboplatin (28.7% versus 61.0%).
3.14 The manufacturer identified two additional trials (First-SIGNAL [n = 42] and the North East Japan Gefitinib Study Group [NEJGSG] trial [n = 198]) that compared gefitinib with chemotherapy for the treatment of chemotherapy-naive patients with predominantly adenocarcinoma histology and EGFR-TK mutations. The manufacturer considered including these studies in a meta-analysis along with data from IPASS. However, the manufacturer excluded the First-SIGNAL study on the basis that it examined only a small number of EGFR-TK mutation-positive patients (n = 42) and because the comparator (gemcitabine and cisplatin) differed from IPASS. The NEJGSG trial, which compared gefitinib with paclitaxel and carboplatin, was considered suitable by the manufacturer for inclusion in the meta-analysis and used as supporting evidence for IPASS. In the NEJGSG trial, patients randomised to receive gefitinib had a statistically significant longer progression-free survival than those randomised to receive paclitaxel and carboplatin (HR 0.357, 95% CI 0.25 to 0.51, p < 0.001). The meta-analysis incorporating progression-free survival from the IPASS and the NEJGSG trial demonstrated a statistically significant improvement in progression-free survival for EGFR-TK mutation-positive patients who were randomised to receive gefitinib compared with mutation-positive patients who were randomised to received paclitaxel and carboplatin (fixed effects model: HR 0.43, 95% CI 0.34 to 0.53, p < 0.001).
3.15 The manufacturer carried out a systematic review and mixed-treatment comparison of RCTs comparing chemotherapy in chemotherapy-naive patients with NSCLC. The manufacturer chose paclitaxel and carboplatin as a baseline comparator for all analyses. The systematic review identified 29 trials, and 28 studies were included in the network that formed the basis for the mixed-treatment comparison. In response to a request from the ERG for clarification, the manufacturer provided an updated mixed-treatment comparison, which included treatment with pemetrexed and cisplatin as a comparator (29 studies in the updated mixed-treatment comparison). The manufacturer extracted and analysed data for clinical efficacy (progression-free survival, overall survival and objective tumour response) and tolerability (anaemia, diarrhoea, fatigue, febrile neutropenia, nausea and vomiting) for use in the economic evaluation. The manufacturer calculated the relative effect of alternative chemotherapy (other than paclitaxel and carboplatin) compared with paclitaxel and carboplatin in an unselected population with NSCLC (that is, without regard to EGFR mutation). The manufacturer then applied the relative estimates for clinical efficacy to a baseline event rate in EGFR-TK mutation-positive patients who had been randomised to receive paclitaxel and carboplatin in IPASS.
3.16 The manufacturer used a Markov economic model to assess the cost effectiveness of gefitinib compared with chemotherapy in the first-line treatment of EGFR-TK mutation-positive patients with NSCLC. Patients entered the model with stable disease. The model had four distinct health states: response to treatment, stable disease, disease progression and death. The model had a cycle length of 21 days and a 5-year time horizon (assumed to be a lifetime horizon).
3.17 The manufacturer obtained data for effectiveness from a variety of sources. The hazard ratio for progression-free survival for EGFR-TK mutation-positive patients for gefitinib relative to paclitaxel and carboplatin was derived from the manufacturer's meta-analysis (HR 0.43) The hazard ratio for overall survival for EGFR-TK mutation-positive patients for gefitinib relative to paclitaxel and carboplatin was estimated from IPASS; estimates of hazard ratios for progression-free survival and overall survival for the chemotherapy regimens were derived from the manufacturer's mixed-treatment comparison. The manufacturer chose a Weibull model for extrapolating costs and outcomes beyond the IPASS follow-up period. Covariates in the model included: mutation status, gender, performance status (0 or 1 versus > 1) and smoking history (never-smoker or ever-smoker).
3.18 The characteristics of the population modelled in the manufacturer's economic evaluation were based on the IPASS population, which comprised chemotherapy-naive EGFR-TK mutation-positive patients who were eligible to receive chemotherapy. The comparator technologies were paclitaxel and carboplatin, gemcitabine and cisplatin, gemcitabine and carboplatin, and vinorelbine and cisplatin.
3.19 Utility estimates in the manufacturer's model were adopted from a single UK study in which utility values were derived from a survey of 105 members of the general public who were asked to value descriptions of health states of second-line chemotherapy for patients with NSCLC. This study did not provide utility estimates associated with the mode of delivery of treatment (oral versus intravenous), so the manufacturer used utility values previously applied in NICE technology appraisal guidance 162 ('Erlotinib for the treatment of relapsed non-small cell lung cancer'), which examined second-line chemotherapy for patients with NSCLC and included utilities related to oral (erlotinib) and intravenous treatment.
3.20 Resource use in the model included: medication, delivery of chemotherapy, EGFR-TK mutation testing, patient monitoring, NHS transport service, management of grade 3 or 4 adverse events, best supportive care and active treatment after progression. Resource use was estimated from a range of secondary sources (such as references costs, British national formulary, previous NICE technology appraisal submissions and the ERG reports for NICE technology appraisal guidance 162). The manufacturer's model incorporated details of a patient access scheme.
3.21 In the manufacturer's base case, the incremental cost-effectiveness ratios (ICERs) for EGFR-TK mutation-positive patients ranged from £19,402 per QALY gained (gefitinib compared with paclitaxel and carboplatin) to £35,992 per QALY gained (gefitinib compared with vinorelbine and cisplatin) using a 16.6% prevalence for EGFR-TK mutation.
3.22 The manufacturer undertook a range of one-way sensitivity analyses and noted that the results of the cost-effectiveness analysis were sensitive to five key parameters: the overall survival for EGFR-TK mutation-positive patients randomised to receive gefitinib; the overall survival for EGFR-TK mutation-positive patients randomised to receive gemcitabine and carboplatin; the progression-free survival for EGFR-TK mutation-positive patients randomised to receive gemcitabine and carboplatin; the progression-free survival for EGFR-TK mutation-positive patients randomised to receive gefitinib; and the maximum number of chemotherapy cycles, which varied from four to eight.
3.23 The manufacturer also carried out a number of scenario analyses, and none led to any substantial change in the ICER. The manufacturer's probabilistic sensitivity analysis showed that vinorelbine and cisplatin was the most cost-effective regimen for the first-line treatment of EGFR-TK mutation-positive patients up to a threshold of £35,100 per QALY gained. Beyond this threshold, gefitinib was the most cost-effective option for the first-line treatment of EGFR-TK mutation-positive patients. At a threshold of £30,000 per QALY gained the probabilities of each treatment being the most cost effective in EGFR-TK mutation-positive patients were: vinorelbine and cisplatin (75%); gefitinib (18%); gemcitabine and carboplatin (4%); gemcitabine and cisplatin (3%); and paclitaxel and carboplatin (0%).
3.24 The ERG considered that the evidence of clinical effectiveness presented in the manufacturer's submission was derived from a high-quality trial that used robust randomisation techniques and was suitably powered to demonstrate the primary objectives of the trial for the overall population. The ERG stated that the trial provided convincing evidence of the efficacy and benefits to health-related quality of life of gefitinib in EGFR-TK mutation-positive patients compared with paclitaxel and carboplatin.
3.25 The ERG highlighted several areas of concern about the clinical evidence submitted by the manufacturer. The ERG was concerned about the generalisability of the clinical results from IPASS to the UK population given the characteristics of the people in the trial (predominantly women, East Asians, non-smokers), the histological type of NSCLC (adenocarcinoma accounts for approximately 25% of the population with NSCLC in the UK), and the comparator used (it is estimated that 5% of patients in the UK receive paclitaxel and carboplatin for the first-line treatment of NSCLC).
3.26 The ERG noted that the licensed indication for gefitiib was in locally advanced or metastatic NSCLC in patients with activating mutations of EGFR-TK, and questioned the feasibility of conducting EGFR-TK mutation testing within the NHS given that this is not routinely carried out. The ERG was concerned that making the service operational throughout England and Wales may require substantial investment in time and resources.
3.27 The ERG highlighted that in IPASS the measurement of the primary outcome of progression-free survival may be unreliable because it was assessed without blinding. The ERG was also concerned that the hazard ratios for this outcome may have been inappropriately calculated using the Cox proportional hazards method. This was because this method is valid only if the hazard ratio for the two groups being compared remains constant over time, and the ERG believed that this criterion was not met in the manufacturer's intention-to-treat analysis of IPASS. The ERG had major concerns about the immaturity of the overall survival data (that is, that relatively few deaths had occurred) because the interim analysis in the manufacturer's submission was based on 450 deaths of 1217 participants (death of 37% of participants). The ERG highlighted that confounding may have occurred in IPASS because of crossover of treatment after disease progression. Therefore, any changes in overall survival may not result from the treatment to which trial participants were originally assigned.
3.28 The ERG highlighted that the manufacturer's meta-analysis could have appropriately included the First-SIGNAL trial because the comparator used in the First-SIGNAL trial (gemcitabine and cisplatin) was not substantially different in terms of clinical benefit and tolerability from the comparator used in IPASS. The ERG noted that an indirect comparison or mixed-treatment comparison including all three studies (IPASS, NEJGSG and First-SIGNAL) would have been more appropriate. The ERG emphasised a number of weaknesses in the manufacturer's mixed-treatment comparison, such as the extraction of unreported outcome statistics for some studies from two published meta-analyses. Different methods were used to estimate the unreported hazard ratios and this may have led to bias regarding the selection of studies included in the mixed-treatment comparison. The ERG was also concerned that the mixed-treatment comparison assumed that EGFR-TK mutation status did not affect outcomes in patients receiving chemotherapy.
3.29 The ERG noted that assessment of gefitinib is more complex than a simple comparison of two treatment options as presented in the manufacturer's submission, because it involves both a specific diagnostic test to identify the presence of EGFR-TK mutations and the consequent choice of treatment following the test result (gefitinib or chemotherapy). The accuracy (that is, analytical validity) of the amplification-refractory mutation system (ARMS) test for identifying EGFR-TK mutations is very high, but the power of the test result to predict a good response to treatment with gefitinib is lower. The ERG suggested that the average benefit for patients receiving gefitinib in IPASS involved a trade-off between those who would get a good outcome (EGFR-TK mutation-positive patients who correctly tested positive for the mutation) and those who would get no benefit at all (EGFR-TK mutation-negative patients who tested positive for the mutation). The ERG also noted that performance characteristics of the diagnostic test should have been incorporated by the manufacturer within the model.
3.30 The ERG expressed concern that the prevalence of EGFR-TK mutations (that is, the proportion of EGFR-TK mutation-positive patients within the tested population) would determine the volume and cost of EGFR-TK tests, and that this would contribute to the incremental cost of adopting a 'test and treat' policy. The ERG highlighted that the results from the manufacturer's economic model for EGFR-TK mutation-positive patients receiving gefitinib were dependent on the prevalence of EGFR-TK mutations. The ERG noted that varying the prevalence of EGFR-TK mutations from the 16.6% stated in the manufacturer's submission (producing an ICER of £20,010 per QALY gained based on a 6-year time horizon) to between 5.0% and 25.0% produced ICERs ranging from £32,685 to £18,174 per QALY gained. The results of the economic model varied depending on the combination of a specific test (ARMS) and gefitinib treatment, and might not be valid if tests other than ARMS were used.
3.31 The ERG believed that the time horizon in the manufacturer's model should have been 6 years instead of 5 years because this more closely approximated the length of life for EGFR-TK mutation-postive patients with locally advanced or metastatic NSCLC. The ERG also highlighted that the chemotherapy costs used in the model were not accurate. The ERG made adjustments to the costs of first-line chemotherapy comparators, which had a modest impact on cost effectiveness. However, the reduction in dose level of comparator chemotherapy because of the higher proportion of female patients in the population of EGFR-TK mutation-positive patients compared with the general lung cancer population, combined with lower BNF prices for generic paclitaxel, led to an increase in the ICER of gefitinib compared with paclitaxel and carboplatin from £20,010 per QALY gained (based on the 6-year time horizon) to £38,063 per QALY gained.
3.32 The ERG was concerned that IPASS allowed a maximum of six chemotherapy cycles whereas in their view patients in the UK usually receive four cycles with up to a maximum of six allowed if their disease responds well. This adjustment to the model by the ERG increased the ICER to more than £32,000 per QALY gained when gefitinib was compared with gemcitabine and carboplatin or paclitaxel and carboplatin, and to £44,000 per QALY gained when gefitinib was compared with vinorelbine and cisplatin or gemcitabine and cisplatin. Furthermore, the ERG noted that the economic model assumed that all patients received prescribed medication up to a maximum of six cycles and that this overestimated the mean number of cycles of chemotherapy administered per patient. When corrected, the ICER for gefitinib increased from £20,010 to £25,427 per QALY gained compared with paclitaxel and carboplatin, which was broadly representative of all chemotherapy regimens.
3.33 The ERG expressed concern about the manufacturer's method of extrapolating survival data beyond the period of IPASS. This involved a two-parameter Weibull formulation for modelling both progression-free survival and overall survival. The ERG digitised the Kaplan–Meier curves for EGFR-TK mutation-positive patients in IPASS and used these to estimate the cumulative hazard for each outcome. The ERG highlighted that in a Weibull survival model the cumulative hazard of an event increases exponentially over time, but that the results from IPASS did not support this. The ERG noted that the parametric model corresponded poorly to the IPASS data, particularly at the beginning and end of the trial. The ERG stated that it obtained a better fit to the data by fitting a linear regression line to obtain a 'spline' model (that is, in this case, two exponential models spliced together at a time when the risk profile of patients changes). The ERG stated that this method reflected the IPASS data accurately across the whole period of the study and with greater accuracy than the Weibull models, which overestimate progression-free survival for both treatment arms. The reanalysis by the ERG reduced estimates of progression-free survival and increased estimates of overall survival, but in both cases reduced the incremental gain attributable to gefitinib by approximately 1 month. This represented a reduction in modelled outcome gains of approximately 25% from those reported in the manufacturer's submission.
3.34 The ERG noted that the manufacturer's economic analysis used differential hazard ratios for the four chemotherapy regimens derived from the mixed-treatment comparison. However, the ERG felt that the four chemotherapy regimens were equally clinically effective. Furthermore, the mixed-treatment comparison depended upon the assumption of proportional hazards, and data from IPASS indicated that this may not be a valid assumption because the hazard ratios within IPASS varied over time. Because the hazard ratios for gefitinib compared with paclitaxel and carboplatin are the main factors determining outcomes in the model and affect results for all comparators as a result of their use in the mixed-treatment comparison, the ERG expressed concern about all the estimates of cost-effectiveness generated by the manufacturer's model.
3.35 The ERG identified several technical errors in the manufacturer's model and carried out amendments and corrections to address these issues. The ERG also incorporated docetaxel and cisplatin, and pemetrexed and cisplatin (using results for pemetrexed and cisplatin from the manufacturer's updated mixed-treatment comparison) into the economic analysis.
3.36 The ERG's revised base-case analysis indicated that ICERs ranged from £59,016 to £72,908 per QALY gained depending on the comparator used. The ERG highlighted that it appeared from this analysis that gefitinib was dominated by pemetrexed and cisplatin (that is, gefitinib was both more expensive and less effective).
3.37 Additional analyses were provided by the manufacturer in response to NICE's request for further clarification on the clinical and cost effectiveness of gefitinib presented in the appraisal consultation document. The manufacturer responded to most concerns raised by the Committee about: alternative probability distributions (models) for the extrapolation of survival data beyond the IPASS; how the models related to observational evidence on long-term survival; independent survival curves and approaches to applying the hazard ratio to incorporate other comparators (with gefitinib or paclitaxel and carboplatin as the baseline); and updated analyses to include amendments to the number and cost of chemotherapy cycles, lower first-line chemotherapy dosing in female patients, and variations in the prevalence of EGFR-TK mutations and costs for EGFR-TK mutation testing.
3.38 The manufacturer provided additional analyses examining alternative probability distributions, with consideration given to model fit to the early trial data and the shape of the curves at the tail of the distribution. The manufacturer examined five distributions – Weibull, log-normal, log-logistic, Gompertz and exponential. The models were fitted to data from IPASS (taken predominantly from EGFR-TK mutation-positive patients) in three ways: to each treatment arm separately (stratified); to the whole population using a stratified model but in the absence of other covariates; and to the whole population using an unstratified model (which assumed proportional hazards between treatments for distributions with this property, that is, the Weibull, Gompertz and exponential models). The manufacturer evaluated the model fit using the Akaike Information Criterion (AIC) and Cox–Snell residuals. The manufacturer's analyses showed that for progression-free survival and overall survival the Weibull models consistently provided the best fit according to AIC, although the log-logistic distribution also provided a good fit to the overall survival data. The manufacturer provided evidence that the proportional hazards assumption was satisfied (that is, there was a constant ratio of the hazards between the two treatments across all points in time) and stated the 'spline' model proposed by the ERG was therefore not appropriate. The manufacturer provided, as academic in confidence, unpublished observational evidence on long-term survival of patients with NSCLC from the NEJGSG study, and published evidence, which showed that for overall survival the data supported the choice of the Weibull or log-logistic distributions.
3.39 The ERG considered that the manufacturer had provided data to address most of the Committee's concerns raised in the appraisal consultation document. However, the ERG noted that the manufacturer had not carried out sensitivity analyses to determine the robustness of the ICERs to alternative survival distributions. Furthermore, the ERG considered that there were several limitations with the manufacturer's submitted analyses because four of the requested amendments to the modelling (a mid-cycle correction, corrected costs for first- and second-line chemotherapy, and adjusted costs to take account of patient drug exposure) had not, in the ERG's view, been implemented correctly in the manufacturer's revised analyses. The ERG therefore carried out an additional analysis to adjust for this. The ERG's additional analysis resulted in ICERs for gefitinib ranging from £30,368 per QALY gained compared with six cycles of gemcitabine and carboplatin to £40,048 per QALY gained compared with four cycles of gemcitabine and carboplatin.
3.40 The ERG also adjusted the manufacturer's two-way sensitivity analyses that varied the prevalence of the EGFR-TK mutation (from 5% to 17%) and the assumed costs of EGFR-TK mutation testing (from £157.50 per test to £210.00 per test). This resulted in ICERs ranging from £27,457 to £49,323 per QALY gained for gefitinib compared with gemcitabine and carboplatin (based on six cycles of chemotherapy). Assuming a cost of £157.50 for EGFR-TK mutation testing in line with advice from consultees and the clinical specialists resulted in ICERs ranging from £31,800 per QALY gained (with a 10% prevalence of EGFR-TK mutation) to £27,500 per QALY gained (with a 17% prevalence of EGFR-TK mutation) for gefitinib compared with gemcitabine and carboplatin (based on six cycles of chemotherapy).
3.41 Applying the ERG's corrections to the manufacturer's additional analyses also showed that, when using paclitaxel plus carboplatin as the baseline, the ICER for gefitinib compared with pemetrexed plus cisplatin was £23,615 per QALY gained for a maximum of six cycles (mean 5.4) and £64,481 per QALY gained for a maximum of five cycles (mean 4.6). When using gefitinib as the baseline, gefitinib dominated pemetrexed plus cisplatin (that is, pemetrexed plus cisplatin was both more expensive and less effective than gefitinib) regardless of whether the model assumed a maximum of five or six cycles.
3.42 The ERG identified an anomaly in the manufacturer's updated economic model relating to Kaplan–Meier estimates of survival in IPASS. However, following clarification with the manufacturer it was established that this anomaly reflected a typographical error and did not affect the manufacturer's cost-effectiveness calculations.
3.43 Full details of all the evidence are in the manufacturer's submission and the ERG report.