Cetuximab, bevacizumab and panitumumab for the treatment of metastatic colorectal cancer after first-line chemotherapy

The manufacturer identified 1 RCT (the E3200 trial) of bevacizumab as second-line treatment for metastatic colorectal cancer. The trial investigated the effectiveness of bevacizumab plus an oxaliplatin-containing chemotherapy regimen, which is outside the scope of this appraisal. People with metastatic colorectal cancer (n=829) who had previously been treated with a fluoropyrimidine with or without irinotecan were randomised to receive bevacizumab plus FOLFOX4 (oxaliplatin plus 5-fluorouracil 400 mg, fluorouracil 600 mg and folinic acid), FOLFOX4 alone or bevacizumab alone. The primary endpoint was overall survival.

Median overall survival was 12.9 months in people randomised to bevacizumab plus FOLFOX4 (n=286), 10.8 months in people randomised to FOLFOX4 alone (n=291), and 10.2 months in people randomised to bevacizumab alone (n=252). The incremental overall median survival for bevacizumab plus FOLFOX4 compared with FOLFOX4 was 2.1 months, with a hazard ratio (HR) of 0.75 (95% confidence interval [CI] not provided, p=0.001). Median progression-free survival was 7.3 months in the bevacizumab plus FOLFOX4 arm, 4.7 months in the FOLFOX4 alone arm, and 2.7 months in the bevacizumab alone arm. The incremental median progression-free survival for bevacizumab plus FOLFOX4 compared with FOLFOX4 alone was 2.6 months (p<0.0001). The bevacizumab alone arm of the study closed early after an interim analysis suggested inferior overall survival compared with the other arms.

The manufacturer identified 3 RCTs (Hurwitz et al. 2004, Kabbinavar et al. 2005, Saltz et al. 2008) that investigated the effectiveness of bevacizumab as first-line treatment for metastatic colorectal cancer. However, the scope for this appraisal specifies that bevacizumab should be considered as second-line and subsequent treatment. The manufacturer stated that, because no RCTs for metastatic colorectal cancer in the second-line setting studied bevacizumab in combination with chemotherapy regimens not containing oxaliplatin, these RCTs in the first-line setting only provide evidence to 'indicate that bevacizumab is safe and effective in combination with irinotecan in second-line metastatic colorectal cancer'. One study (Hurwitz et al. 2004) compared bevacizumab plus irinotecan, bolus 5-fluorouracil and folinic acid (n=402) with placebo plus irinotecan, bolus 5-fluorouracil and folinic acid (n=411). The primary endpoint was median overall survival, which was 20.3 months for people randomised to bevacizumab plus irinotecan, bolus 5-fluorouracil and folinic acid compared with 15.6 months for people randomised to placebo plus folinic acid (HR 0.66, p<0.001). The median progression-free survival was 10.6 months for people randomised to bevacizumab plus irinotecan, bolus 5-fluorouracil and folinic acid and 6.2 months for people randomised to placebo plus folinic acid (HR 0.54, p<0.001). The second RCT (Kabbinavar et al. 2005) evaluated bevacizumab plus bolus 5-fluorouracil and folinic acid (n=105) compared with placebo plus bolus 5-fluorouracil and folinic acid (n=104). Median overall survival was 16.6 months in people randomised to bevacizumab plus bolus 5-fluorouracil and folinic acid compared with 12.9 months in people randomised to placebo plus 5-fluorouracil and folinic acid (HR 0.79, p=0.16). The third RCT (Saltz et al. 2008) compared bevacizumab plus oxaliplatin-based chemotherapy (FOLFOX or capecitabine plus oxaliplatin [XELOX], n=699) with oxaliplatin-based chemotherapy alone (FOLFOX or XELOX, n=667). Progression-free survival (the primary endpoint) was significantly greater in people randomised to bevacizumab plus chemotherapy compared with chemotherapy alone (median progression-free survival 9.4 months versus 8.0 months; HR 0.83, p=0.002). There was no significant difference in overall median survival between the 2 arms (21.3 months versus 19.9 months for the bevacizumab plus chemotherapy arm and chemotherapy alone arm respectively; HR 0.89, p=0.77).

The manufacturer identified 2 observational cohort studies that investigated the effectiveness of bevacizumab as second-line treatment for metastatic colorectal cancer in people with progressed disease. One study, which used data from the 'Bevacizumab Regimens: Investigation of Treatment Effects and Safety' (BRiTE) registry compared overall survival in people treated with bevacizumab as first- and second-line therapy (n=642), with people treated with bevacizumab as first-line therapy but with other chemotherapy treatments in the second-line setting (n=531). Overall survival was significantly greater in people who received bevacizumab as a second-line treatment compared with those receiving other second-line chemotherapy regimens (median overall survival post progression 31.8 months compared with 19.9 months; HR 0.48, p<0.001). The other observational study used data from the ARIES registry and compared overall survival in people with metastatic colorectal cancer who received bevacizumab as first- and second-line treatment (n=208) with people treated with bevacizumab in the second-line setting only (n=255). Overall survival was significantly greater in people who received bevacizumab as first- and second-line treatment compared with those who received it in the second-line setting only (median overall survival post progression 21.7 months (95% CI 17.8 to 27.0) compared with 17.5 months (95% CI 15.9 to 21.5). The manufacturer noted that the data from the registries were not available by type of chemotherapy, and therefore specific information about the effect of bevacizumab in combination with non-oxaliplatin-based chemotherapy could not be provided.

The Assessment Group did not identify any trials that met the inclusion criteria for its review (that is, bevacizumab plus non-oxaliplatin-based chemotherapy for the treatment of people with metastatic colorectal cancer whose disease had progressed after first-line chemotherapy).

The manufacturer and the Assessment Group identified 1 RCT (the CO.17 trial), which compared cetuximab plus best supportive care with best supportive care alone in people (n=572) with metastatic colorectal cancer who had previously been treated with a fluoropyrimidine, irinotecan and oxaliplatin or who had contraindications to these treatments. This trial was mainly a trial of third-line and subsequent therapy, because 96% to 98% of people had received both irinotecan and oxaliplatin. Nearly half of the participants had received 4 or more chemotherapy regimens. The primary endpoint was overall survival.

The median overall survival in the whole trial population (irrespective of KRAS mutation status of the tumour) was 6.1 months with cetuximab plus best supportive care and 4.6 months with best supportive care alone, with an HR of 0.77 (95% CI 0.64 to 0.92; p=0.005). Approximately 7% of people randomised to best supportive care alone were given cetuximab after crossover.

A total of 394 (68.9%) tumour specimens were retrospectively examined for KRAS mutation status after completion of the trial (Karapetis et al. 2008). Limiting analyses to people with KRAS wild-type status, the median overall survival was 9.5 months for people randomised to cetuximab plus best supportive care compared with 4.8 months for people randomised to best supportive care alone (HR 0.55, 95% CI 0.41 to 0.74; p<0.001). In an analysis that adjusted for both randomisation and potential prognostic factors (age, Eastern Cooperative Oncology Group performance status, previous chemotherapy), the HR increased to 0.62 (95% CI 0.44 to 0.87; p=0.006).

To compare cetuximab plus best supportive care with cetuximab plus irinotecan in people with KRAS wild-type status, the manufacturer presented a published retrospective observational analysis (De Roock et al. 2008, referred to as the De Roock analysis) that analysed data on Belgian participants combined from 4 studies (EVEREST n=50, BOND n=44, SALVAGE n=17 and BABEL n=2). Approximately one-quarter of people had been treated with cetuximab monotherapy, and three-quarters with cetuximab (at varying dosages) plus irinotecan. The phase 2 EVEREST trial investigated the effect of cetuximab dose escalation on EGFR expression in people with metastatic colorectal cancer whose disease had not responded to prior treatment with irinotecan. The BOND study was a randomised open-label multicentre phase 2 RCT of cetuximab plus irinotecan versus cetuximab monotherapy in people with metastatic EGFR-expressing colorectal adenocarcinoma. The SALVAGE study was an uncontrolled trial of monotherapy with cetuximab in people with metastatic colorectal cancer, whose tumours expressed EGFR and were refractory to irinotecan, oxaliplatin and fluoropyrimidines. The BABEL study was an uncontrolled trial of cetuximab monotherapy in people with KRAS wild-type metastatic colorectal cancer. The De Roock analysis provided data for a total of 113 people (67 with KRAS wild-type status, 46 with the KRAS mutation) with irinotecan refractory metastatic colorectal cancer who had been treated with cetuximab monotherapy. The Assessment Group excluded the De Roock analysis from its review because it judged the analysis to have a number of limitations: the people selected may not have been representative of people treated in UK clinical practice, and 2 of the studies (BABEL and SALVAGE) were single arm (that is, uncontrolled) studies. Only the BOND study compared cetuximab plus irinotecan with cetuximab monotherapy.

The manufacturer and the Assessment Group identified 1 RCT (the 'Amgen' trial) that compared panitumumab plus best supportive care with best supportive care alone in 463 people with metastatic colorectal cancer that had progressed after standard first- and second-line chemotherapy (a fluoropyrimidine, irinotecan and oxaliplatin). The primary endpoint of the trial was progression-free survival. Overall survival was analysed as a secondary endpoint. No statistically significant difference was observed in overall survival in the whole population (irrespective of KRAS mutation status; HR 1.00, 95% CI 0.82 to 1.22).

Tumour samples from 427 (92%) people in the Amgen trial were retrospectively obtained for KRAS mutation status testing after the end of the trial. In the KRAS wild-type population, median progression-free survival was 12.3 weeks in people randomised to panitumumab plus best supportive care compared with 7.3 weeks in people randomised to best supportive care alone. When calculating overall survival in people randomised to the best supportive care arm, the manufacturer excluded people with wild-type KRAS who crossed over to receive panitumumab. Overall survival was estimated using 2 mutually exclusive time points: mean time to disease progression and mean time from progression to death. Survival estimates for best supportive care were based on people randomised to best supportive care with a KRAS mutation or wild-type KRAS for the time until disease progression (before any treatment crossover occurred), and people randomised to best supportive care with a KRAS mutation for the time from disease progression to death. Mean times to disease progression and from progression to death were estimated by fitting survival models to patient-level data from the clinical trial and then estimating the area under the best-fit curves and the mean survival for each distribution. The median overall survival from an intention-to-treat analysis in the KRAS wild-type population was 8.1 months for people randomised to panitumumab plus best supportive care compared with 7.6 months for people randomised to best supportive care alone. No statistically significant difference in median overall survival after disease progression between panitumumab and best supportive care was shown in this KRAS wild-type population (HR 0.99; 95% CI 0.75 to 1.29).

There was significant crossover in the Amgen trial; of 219 people randomised to best supportive care alone, 166 (76%) crossed over after disease progression to receive treatment with panitumumab. Because panitumumab lengthened progression-free survival, and many people randomised to best supportive care also received panitumumab after progression, the estimates of effectiveness from intention-to-treat analyses (see section 4.2.11) were considered by the manufacturer to underestimate the effectiveness of panitumumab. Therefore, in an attempt to adjust for this bias, the manufacturer adjusted the overall survival results by including people with KRAS mutations randomised to best supportive care in the analysis, regardless of whether they crossed over to receive panitumumab treatment after disease progression. The manufacturer's rationale for this method was that the trial showed that people with a KRAS mutation did not benefit from treatment with panitumumab. Therefore, people with a KRAS mutation in the best supportive care arm who crossed over to receive panitumumab after disease progression would also be less likely to benefit from it. The average survival gain adjusted for crossover was between 2.74 months (overall survival estimated by splitting response rates) and 3.13 months (overall survival estimated by aggregating survival across response rates) for panitumumab compared with best supportive care. The Assessment Group judged that the manufacturer's approach and assumptions to adjust for crossover were reasonable.

The Assessment Group and the manufacturers did not identify any RCTs that directly compared each of the technologies included in this appraisal. Both the Assessment Group and the manufacturer of cetuximab carried out a mixed treatment comparison using the Bucher approach to estimate the relative effectiveness of the technologies relevant to the decision problem. Without clinical evidence for the use of bevacizumab as specified by the scope, there were 4 treatments or comparators:

• best supportive care

• monotherapy with cetuximab plus best supportive care

• monotherapy with panitumumab plus best supportive care, and

• cetuximab plus chemotherapy plus best supportive care.
  
  The manufacturers of panitumumab and bevacizumab did not submit a mixed treatment comparison because they did not consider it possible to conduct a robust mixed treatment comparison of the 3 technologies based on the evidence available.

To compare the clinical effectiveness of cetuximab plus chemotherapy with panitumumab plus best supportive care and best supportive care alone, the manufacturer (Merck Serono) used data from the CO.17 trial and from the Amgen trial. Although the scope of this appraisal covers cetuximab plus chemotherapy, the only evidence available was for cetuximab plus irinotecan. The manufacturer used data from 80 people in a retrospective analysis of the De Roock analysis to compare cetuximab plus best supportive care with cetuximab plus irinotecan in the KRAS wild-type population. The manufacturer did not identify any relevant evidence for bevacizumab.

The resulting HR for overall survival for cetuximab plus irinotecan and best supportive care compared with best supportive care alone was 0.29 (95% CI 0.14 to 0.59). Following advice from clinical specialists, the manufacturer concluded that the parametric model it had fitted to the Kaplan–Meier curve for overall survival (Weibull function) did not match the original data. The manufacturer therefore obtained additional data from the retrospective analysis of the De Roock analysis for 364 people. The resulting HR for overall survival for cetuximab plus irinotecan compared with best supportive care changed to 0.32 (confidence interval not reported). The manufacturer used the 95% CI from the original retrospective analysis of the De Roock analysis (that is, 95% CI 0.14 to 0.59). The resulting HR for overall survival for cetuximab plus best supportive care compared with panitumumab plus best supportive care was 0.56 (95% CI 0.37 to 0.83).

The Assessment Group expressed concerns about the validity of the manufacturer's approach to calculating progression-free survival and overall survival hazard ratios from the mixed treatment comparison because:

• it combined randomised and non-randomised evidence, subjecting it to bias and confounding

• it did not assess whether the populations in the chosen studies were comparable

• for calculating the overall survival HR for cetuximab plus irinotecan compared with best supportive care, the manufacturer used data from a non-comparative study to adjust the HR to fit the model, but the statistical fit of the model was determined by clinical specialists rather than statistical testing, and the manufacturer did not clarify how adjustments were made to fit the data to the model

• the manufacturer used unadjusted HRs from the cetuximab CO.17 RCT instead of values adjusted for patient characteristics, which may have overestimated the effectiveness of cetuximab

• the BOND study, which was included in the observational retrospective analysis that combined 4 studies (the De Roock analysis) did not account for crossover, and this could have led to an underestimation of overall survival gain for cetuximab plus best supportive care compared with cetuximab plus irinotecan.

The Assessment Group also carried out a mixed treatment comparison for the 4 treatments: best supportive care, cetuximab monotherapy plus best supportive care, panitumumab monotherapy plus best supportive care, and cetuximab plus irinotecan and best supportive care. The Assessment Group used data from the 2 RCTs used by the manufacturer of cetuximab in its mixed treatment comparison: the CO.17 trial (cetuximab plus best supportive care compared with best supportive care alone) and the Amgen trial (panitumumab plus best supportive care compared with best supportive care alone). It also used data from the retrospective analysis of 4 studies (the De Roock analysis). The Assessment Group assumed that best supportive care was equivalent between the CO.17 trial and the Amgen trial. Unlike the manufacturer's analysis, the Assessment Group adjusted HRs in its mixed treatment comparison for the patient characteristics in the KRAS wild-type population. However, the HRs obtained from the indirect comparison were not adjusted using data from the retrospective analysis (De Roock analysis).

Results of the mixed treatment comparison showed that patients who received cetuximab plus best supportive care would be expected to have significantly longer overall survival than those receiving panitumumab plus best supportive care (unadjusted HR 0.56, 95% CI 0.37 to 0.83; adjusted HR 0.63, 95% CI 0.41 to 0.97). The Assessment Group highlighted that the HR for overall survival for panitumumab from the Amgen trial may have underestimated the effectiveness of panitumumab relative to best supportive care because most people randomised to best supportive care also received treatment with panitumumab after their disease had progressed. The Assessment Group reported an overall survival estimate of 16.2 months for cetuximab plus irinotecan in an appendix to the assessment report.

Amgen and Roche Products did not submit health economic models. Roche Products submitted calculations outlining the treatment costs for bevacizumab plus FOLFIRI compared with cetuximab plus FOLFIRI. The treatment cost for cetuximab plus FOLFIRI was estimated to exceed that for bevacizumab plus FOLFIRI by £5,408, with an incremental cost for KRAS testing of £462, additional drugs costs of £3,357 and additional administration costs of £1,589.

Merck Serono provided a Markov model to make 4 comparisons:

• cetuximab plus best supportive care compared with best supportive care alone

• cetuximab plus irinotecan plus best supportive care compared with best supportive care alone

• cetuximab plus best supportive care compared with panitumumab plus best supportive care

• cetuximab plus irinotecan plus best supportive care compared with panitumumab plus best supportive care.

The population modelled by Merck Serono included people with EGFR-expressing KRAS wild-type metastatic colorectal cancer who had received second- or subsequent-line chemotherapy for metastatic disease. The model had a 10-year time horizon and a UK National Health Service (NHS) perspective. The cycle length was 1 week and a half-cycle correction was not applied. The model had 3 health states: progression-free disease, progressive disease and death. Merck Serono based the transitions between health states on parametric approximations of Kaplan–Meier estimates of progression-free survival and overall survival from the relevant arms of the RCTs (with time spent by a patient in progressive disease defined as the difference between the 2).

To compare cetuximab plus best supportive care with best supportive care alone, Merck Serono estimated separate probabilities for time to disease progression and time to death for people in the progression-free disease and progressive disease health states using patient-level data. Merck Serono chose different functions on the basis of goodness-of-fit measures for each transition (log-normal for time to progression; log-logistic for death from the health state of pre-progression; Weibull for death from the health state of progressive disease).

For the comparison of cetuximab plus irinotecan and best supportive care with best supportive care alone, Merck Serono modelled progression-free survival and overall survival using a 2-stage process. First, it simulated progression-free survival and overall survival for people treated with best supportive care alone using a Weibull curve and then validated this curve using data from the best supportive care arm of the CO.17 trial. The corresponding values for progression-free survival and overall survival for cetuximab plus irinotecan and best supportive care were then estimated by applying the overall survival HRs for cetuximab plus irinotecan and best supportive care with the HR for best supportive care being drawn from the mixed treatment comparison. Merck Serono obtained estimates of utility for each health state using the Health Utility Index (HUI) scale (a generic preference-based measure of quality of life) by reanalysing data by health state in the CO.17 trial. These utility values were then applied to cetuximab plus irinotecan and best supportive care and panitumumab plus best supportive care. The manufacturer used utility values of 0.809 for progression-free disease, 0.789 for progressive disease and 0.000 for death.

The following assumptions were made in the model: the mean time on treatment with cetuximab plus best supportive care is 2.6 months and the mean time for cetuximab plus irinotecan is 4.4 months; active treatment stops at set cut-off time points, that is, 13 weeks for cetuximab plus best supportive care and 24 weeks for cetuximab plus irinotecan plus best supportive care, even if a patient's disease has not progressed.

Merck Serono produced a series of pairwise comparisons of cost effectiveness in people with KRAS wild-type metastatic colorectal cancer:

• Cetuximab plus best supportive care compared with best supportive care alone produced a base-case incremental cost-effectiveness ratio (ICER) of £47,095 per quality-adjusted life year (QALY) gained.

• Cetuximab plus irinotecan plus best supportive care compared with best supportive care alone produced a base-case ICER of £43,887 per QALY gained.

• Cetuximab plus irinotecan plus best supportive care compared with panitumumab plus best supportive care produced a base-case ICER of £21,819 per QALY gained.
  
  When cetuximab plus best supportive care was compared with panitumumab plus best supportive care, panitumumab was associated with higher costs and fewer QALYs (-0.193 incremental QALYs and £2,629 incremental costs). Merck Serono completed one-way sensitivity analyses on all the model parameters and the only factor found to significantly change the ICERs was varying the cost of cetuximab (which included changes to the price of the drug, its administration costs, and/or the duration of treatment).

Merck Serono's probabilistic sensitivity analyses indicated that, compared with best supportive care alone, cetuximab plus best supportive care and cetuximab plus irinotecan had a 64.7% and a 68% chance respectively of being cost effective at £50,000 per QALY gained. Compared with panitumumab plus best supportive care, cetuximab plus best supportive care had a 100% chance of being cost effective at £15,000 per QALY gained. Cetuximab plus irinotecan compared with panitumumab plus best supportive care has a 73.8% chance of being cost effective at £30,000 per QALY gained and a 93% chance of being cost effective at £50,000 per QALY gained.

From a literature review the Assessment Group identified 1 cost-effectiveness analysis of bevacizumab in previously untreated metastatic colorectal cancer, which was not relevant to this appraisal. It also identified a study by Mittman et al. (2008) which was a trial-based cost-effectiveness analysis that used data from the cetuximab CO.17 trial. The Assessment Group also identified 3 studies (Annemans et al. 2007, Norum et al. 2006, Starling et al. 2007) which assessed the cost effectiveness of cetuximab plus irinotecan compared with best supportive care. The base-case ICER in the Annemans et al. study was €40,273 per life year gained (based on 12 weeks of treatment). The base-case ICER was €205,536 per life year gained in the Norum et al. study and £57,608 per QALY gained in the Starling et al. study.

The Assessment Group noted that the Merck Serono model did not attempt to compare cetuximab plus irinotecan plus best supportive care with cetuximab plus best supportive care. Moreover, Merck Serono assessed the cost effectiveness of cetuximab only as third-line treatment and did not consider it as second-line treatment, but the scope for this appraisal allows any of the technologies to be considered as second-line treatment. The Assessment Group questioned the validity of the utility values obtained from the CO.17 trial by Merck Serono because they exceeded the values produced by the health economic evaluation that accompanied the CO.17 trial (Mittman et al. 2008).

The Assessment Group provided an area under the curve model that compares cetuximab plus best supportive care with best supportive care alone, cetuximab plus irinotecan plus best supportive care with best supportive care alone, and panitumumab plus best supportive care with best supportive care alone in people with EGFR-expressing KRAS wild-type metastatic colorectal cancer who had received at least second-line chemotherapy for metastatic disease. The Assessment Group did not include bevacizumab in the economic analysis because no clinical effectiveness evidence was available for bevacizumab plus chemotherapy without oxaliplatin in people who had received previous chemotherapy. The model had a 10-year time horizon and a UK NHS perspective. The cycle length was 1 month and a half-cycle correction was applied.

The model had 3 health states: progression-free disease, progressive disease and death. The Assessment Group used an 'area under the curve' or 'cohort partition' method to determine the number of people in each health state at each cycle of the model, rather than using transition probabilities. The Assessment Group obtained estimates of utility from the Mittman et al. (2008) study that used individual patient-level data and HUI data from the CO.17 trial. The Assessment Group used utility values for progression-free disease of 0.81 for cetuximab plus best supportive care, 0.75 for best supportive care, 0.75 for cetuximab plus irinotecan, and 0.87 for panitumumab plus best supportive care; a utility value of 0.69 was used for progressive disease (for all treatments).

The Assessment Group's model differed from the Merck Serono model in the following ways for the comparison of cetuximab plus best supportive care versus best supportive care alone:

• the estimates of mean time on cetuximab varied: Assessment Group 4.8 months, Merck Serono 2.6 months

• the estimates of drug costs varied because of differences in estimates of treatment duration: Assessment Group £14,400, Merck Serono £8,200

• the estimates of drug administration costs varied because of differences in estimates of treatment duration: Assessment Group £5,500, Merck Serono £2,000

• the estimates of utility were taken directly from Mittman et al. (2008) by the Assessment Group; reanalysed estimates from the CO.17 trial were used by Merck Serono

• the Assessment Group model included an adjustment for crossover for the overall survival HR for panitumumab compared with best supportive care whereas the Merck Serono model did not.

For the comparison of cetuximab plus irinotecan versus best supportive care, the main differences between the Assessment Group's model and the Merck Serono model were the:

• estimates of mean time on cetuximab plus irinotecan varied: Assessment Group 8.8 months, Merck Serono 4.4 months

• estimates of drug costs varied because of differences in estimates of treatment duration: Assessment Group £32,000, Merck Serono £17,400

• estimates of drug administration costs varied because of differences in estimates of treatment duration: Assessment Group £12,700, Merck Serono £3,800.

The Assessment Group produced a series of pairwise comparisons of cost effectiveness in people with KRAS wild-type metastatic colorectal cancer:

• Cetuximab plus best supportive care compared with best supportive care alone produced a base-case ICER of £98,000 per QALY gained.

• Cetuximab plus irinotecan plus best supportive care compared with best supportive care alone produced a base-case ICER of £88,000 per QALY gained.

• Panitumumab plus best supportive care compared with best supportive care alone produced a base-case ICER of £150,000 per QALY gained.
  
  The Assessment Group completed deterministic one-way sensitivity analyses on model parameters and the only factor found to substantially change the ICER was the estimate of overall survival. The ICER for cetuximab plus best supportive care compared with best supportive care alone was more than £70,000 per QALY gained in all scenarios, the ICER for cetuximab plus irinotecan compared with best supportive care was more than £55,000 per QALY gained, and the ICER for panitumumab plus best supportive care compared with best supportive care was more than £110,000 per QALY gained. When the unadjusted progression-free survival estimates from the Amgen trial were used, the ICER for panitumumab plus best supportive care compared with best supportive care was reduced to £109,000 per QALY gained. In an additional analysis conducted in response to comments received from the manufacturers during consultation on the assessment report, the overall survival estimate for best supportive care was increased from 6.8 months to 7.2 months. This gave an ICER of £119,000 per QALY gained for panitumumab plus best supportive care compared with best supportive care.

The Assessment Group's probabilistic sensitivity analyses indicated that below £60,000 per QALY gained, none of the drugs appraised is the most cost-effective treatment for second-line or subsequent chemotherapy of metastatic colorectal cancer. Above £90,000 per QALY gained, cetuximab plus irinotecan is likely to be the most cost-effective treatment compared with best supportive care. Cetuximab monotherapy or panitumumab are never the most cost-effective option when compared with best supportive care.

The Committee discussed the clinical effectiveness of bevacizumab in people with metastatic colorectal cancer who have received first-line chemotherapy. The Committee discussed the results of the 3 RCTs (see section 4.2.4) presented in the Roche Products submission, which investigated the effectiveness of bevacizumab as first-line treatment for metastatic colorectal cancer. The Committee agreed that these trials demonstrated that bevacizumab is an effective first-line treatment for metastatic colorectal cancer, but recognised that the scope of this appraisal was to consider bevacizumab in the second- and third-line setting. The Committee understood that Roche Products, the regulatory authorities and the clinical specialists considered that if bevacizumab plus a non-oxaliplatin-containing regimen is effective in the first-line setting, the combination would also likely be effective in second- and third-line settings, despite not having been tested in these situations. The Committee heard that this assumption was the basis of the regulatory approval for bevacizumab as a second-line therapy. However, the Committee agreed that people receiving bevacizumab as second-line therapy would have more advanced disease than people receiving bevacizumab for first-line therapy. Therefore, the Committee concluded that people receiving bevacizumab as second-line therapy would likely have smaller gains in survival than people who have not previously received chemotherapy.

The Committee noted the results of 2 registry-based observational studies, BRiTE and ARIES. The Committee understood that Roche Products could not provide data from these registries specifically for bevacizumab in combination with non-oxaliplatin chemotherapy. In line with this, the manufacturer also informed the Committee that these registries were unlikely to inform the Committee's decision regarding the use of bevacizumab as second-line or subsequent therapy in combination with non-oxaliplatin chemotherapy. The Committee concluded that the observational evidence presented in the manufacturer's submission could not be used to establish the magnitude of the overall survival gain with bevacizumab plus non-oxaliplatin chemotherapy for people with metastatic colorectal cancer which had not responded to first-line chemotherapy.

The Committee then discussed the E3200 RCT presented in the Roche Products submission (see section 4.2.2), which investigated the effectiveness of bevacizumab plus an oxaliplatin-containing chemotherapy regimen as second-line treatment compared with placebo plus folinic acid for metastatic colorectal cancer. The Committee acknowledged that NICE's technology appraisal guidance on bevacizumab in combination with oxaliplatin and either fluorouracil plus folinic acid or capecitabine for the treatment of metastatic colorectal cancer has already evaluated the clinical and cost effectiveness of bevacizumab plus oxaliplatin-containing chemotherapy in metastatic colorectal cancer, and that the remit of the current appraisal is to appraise bevacizumab plus non-oxaliplatin chemotherapy. The Committee agreed that the results of the E3200 trial could not be used to establish the overall survival gain with bevacizumab plus non-oxaliplatin chemotherapy as second- or third-line treatment for people with metastatic colorectal cancer who had not responded to first-line or second-line chemotherapy. The Committee noted that this conclusion was supported by 1 of the clinical specialists, who pointed out that the effectiveness of biological drugs plus oxaliplatin differs from the effectiveness of biological drugs plus irinotecan. The Committee acknowledged that there is an ongoing RCT of bevacizumab plus FOLFIRI compared with panitumumab plus FOLFIRI as second-line treatment of metastatic colorectal cancer, which is due to finish in August 2012. However, it agreed that it was not aware of any currently available evidence on which to base a decision about the clinical effectiveness of bevacizumab plus non-oxaliplatin chemotherapy in people with metastatic colorectal cancer who had previously received chemotherapy.

The Committee discussed the likely cost effectiveness of bevacizumab plus non-oxaliplatin chemotherapy compared with best supportive care and noted the Assessment Group's view that lack of relevant evidence on clinical effectiveness meant that it was not feasible to carry out a cost-effectiveness evaluation of bevacizumab. The Committee also heard from Roche Products that it had not submitted an economic model because it did not believe it would be possible to establish that bevacizumab plus non-oxaliplatin chemotherapy is cost effective as a second-line treatment for metastatic colorectal cancer. The Committee noted that previous NICE technology appraisal guidance on bevacizumab in combination with oxaliplatin and either fluorouracil plus folinic acid or capecitabine and bevacizumab and cetuximab had not found bevacizumab to be a cost-effective first-line or second-line treatment for metastatic colorectal cancer. In view of its previous judgement that bevacizumab is likely to be less effective as second-line therapy than as first-line therapy, on balance, the Committee felt that it was unlikely that bevacizumab would be a cost-effective treatment for people with metastatic colorectal cancer who had received first-line therapy. The Committee therefore concluded that the available clinical and cost-effectiveness evidence does not justify a positive recommendation for bevacizumab plus non-oxaliplatin chemotherapy as second-line treatment for metastatic colorectal cancer.

The Committee discussed the clinical effectiveness of cetuximab (monotherapy or combination chemotherapy) in people with KRAS wild-type metastatic colorectal cancer that has progressed after first-line chemotherapy. The Committee noted that the people in the CO.17 trial had previously received oxaliplatin- and irinotecan-based therapy, that is, cetuximab was used in the third-line or later setting. The trial had shown a median overall survival of 9.5 months for cetuximab plus best supportive care compared with 4.8 months for best supportive care alone. The Committee was aware that only 68.9% of people in the CO.17 trial were tested for KRAS mutation status, and was concerned that there may have been selection bias related to the KRAS mutation testing if those people tested were fundamentally different in a way which influenced their response to treatment. The Committee noted the Assessment Group's comment that the age of people in the trial, including people whose tumour displayed KRAS mutations and those not tested, averaged 63 years (range 28 to 88 years), which was on average 10 years younger than people with metastatic colorectal cancer typically seen in clinical practice in the NHS. The Committee heard from the clinical specialists that age at of the start of treatment was unlikely to affect clinical response. The Committee therefore agreed that although the trial population did not fully represent people seen in clinical practice in the NHS, the evidence of clinical effectiveness for cetuximab monotherapy was generalisable to UK clinical practice. The Committee concluded that there was sufficient evidence to show that cetuximab plus best supportive care gave greater benefit in terms of both progression-free survival and overall survival than best supportive care alone.

The Committee discussed the evidence available for the clinical effectiveness of cetuximab plus irinotecan chemotherapy. The Committee noted that there were no head-to-head trials of cetuximab plus irinotecan compared with best supportive care in KRAS wild-type colorectal cancer. The Committee therefore discussed the results of the manufacturer's mixed treatment comparison that compared cetuximab plus chemotherapy with panitumumab or best supportive care and cetuximab monotherapy with panitumumab in the KRAS wild-type population. The Committee noted the Assessment Group's concerns about the validity of the mixed treatment comparison (see section 4.2.17). The Assessment Group was particularly concerned about the reliance on the retrospective observational analysis (the De Roock analysis) for the effectiveness estimate for cetuximab plus irinotecan, which combined data from RCTs and non-RCTs (single-arm trials), not all of which included treatment with cetuximab plus irinotecan. The Committee therefore agreed that the results of the mixed treatment comparison should be interpreted with caution. The Committee concluded that the estimates of overall survival for cetuximab plus irinotecan were subject to a high degree of uncertainty.

The Committee discussed the economic model submitted by the manufacturer of cetuximab, and the Assessment Group's comments on this model. The Committee concluded that using best supportive care as 1 of the comparators in the model was appropriate. However, the Committee was concerned that the manufacturer had not submitted an economic comparison of cetuximab plus best supportive care versus cetuximab plus irinotecan plus best supportive care, despite having submitted estimates of clinical effectiveness, and had not given a reason for this. The Committee discussed 2 concerns about the total cost estimated by the manufacturer for cetuximab; that is, the administration costs and costs associated with duration of treatment. The Committee was aware that the Assessment Group model had used estimated administration costs for cetuximab that were 2 to 3 times higher than those estimated by the manufacturer. The Committee discussed its concerns about the assumptions in the manufacturer's model about the duration of treatment; particularly whether in clinical practice people would receive cetuximab for a fixed treatment period (as modelled) rather than until disease progresses (as specified in the SPC). The Committee heard from the clinical specialists that clinicians would offer people treatment with cetuximab monotherapy until their disease progressed, but would likely offer cetuximab plus irinotecan for a fixed period in view of the increased toxicity of combined treatment. The Committee therefore concluded that it did not accept the assumption in the manufacturer's model that a fixed treatment period for cetuximab represented UK clinical practice. The Committee also noted a comment made by Amgen during consultation that NICE's guide to the methods of technology appraisal states a preference for the use of the public list price for a technology and not the negotiated price to the NHS. The Committee noted that the lower NHS price for cetuximab was previously used in NICE technology appraisal guidance on cetuximab (now replaced by NICE's technology appraisal guidance on cetuximab and panitumumab) rather than the list price. It agreed that the most relevant price to be considered in this appraisal is the 1 that is nationally available and in the public domain, and therefore considered it appropriate to use the NHS price for cetuximab in the economic model.

The Committee discussed the utility estimates in the manufacturer's model, which were obtained from the CO.17 trial. The Committee was aware that using HUI deviated from the NICE reference case, which encourages the use of EQ-5D. However, it agreed that the HUI was a valid measure of utility and that values obtained from the trial population were likely to be generalisable to patients in the UK. The Committee noted that the utility estimates for each of the disease states were not consistent with the expectation that quality of life worsens with progression of disease. The Committee was aware of the Assessment Group's concern that the values of utility recalculated by the manufacturer from the CO.17 trial were higher for progression-free disease than those of Mittman et al. from the same trial. The Committee also noted that the utility estimates used in the model (for example, 0.81 for progression-free disease for cetuximab plus best supportive care) were similar to those expected for people of the same age without metastatic colorectal cancer. The Committee concluded that the utility values in the manufacturer's model were highly uncertain.

The Committee discussed the results of the manufacturer's sensitivity analyses and noted that the estimate of cost effectiveness was most sensitive to the estimate of overall drug costs, which was determined by the time on cetuximab treatment. The Committee heard from the manufacturer that the estimates of time on treatment in the model were based on clinical opinion rather than direct estimates from the CO.17 trial. The Committee agreed that the assumption of a fixed treatment period for cetuximab in the manufacturer's model did not represent UK clinical practice (see section 4.4.11).

The Committee considered the Assessment Group's economic model for cetuximab. The Committee heard that the utility estimates in the Assessment Group's model had been obtained from a published cost–utility study of the CO.17 trial (Mittman et al. 2008) and were in general lower than those used in the manufacturer's model. The Committee agreed that the utility values used by the manufacturer were implausibly high because they were similar to those of the general population of the same age without metastatic colorectal cancer. The Committee also noted that because the manufacturer did not provide an estimate of the average length of cetuximab treatment in the CO.17 trial, the Assessment Group contacted Dr Mittman to obtain this estimate after the assessment report had been submitted to the Committee. This estimate was provided to the Committee as an addendum, and is not given in this document because it is considered academic-in-confidence. The Committee agreed that this estimate of time on treatment was more appropriate because it was derived from trial data rather than from an assumption.

The Committee noted that 1 of the main factors affecting the cost effectiveness of cetuximab was the assumption about the duration of treatment. The Committee agreed that using the values provided as academic-in-confidence in the Assessment Group's analyses gave the most plausible ICER for cetuximab plus best supportive care of £90,000 per QALY gained and for cetuximab plus irinotecan plus best supportive care of £88,000 per QALY gained, both compared with best supportive care. The Committee was aware of another cost−utility analysis of the CO.17 trial (Mittman et al. 2008) that had estimated an ICER of £101,000 per QALY gained for cetuximab plus best supportive care compared with best supportive care. The Committee was also aware that the manufacturer, Merck Serono, noted in its comments during consultation that cetuximab 'is not cost effective under the usual threshold range for acceptability'. The Committee concluded that the most plausible ICERs for cetuximab monotherapy and cetuximab in combination chemotherapy did not represent a cost-effective use of NHS resources.

The Committee discussed the clinical effectiveness of panitumumab monotherapy in people with KRAS wild-type metastatic colorectal cancer that has progressed after first-line chemotherapy. It noted that the only trial evidence available applied to people who had previously received both oxaliplatin- and irinotecan-based therapy, that is, in the third-line or subsequent setting (the Amgen trial). The Committee heard from the manufacturer that over 90% of people in the trial were assessed for the KRAS mutation and concluded that selection bias associated with testing was unlikely. The Committee noted that although a benefit in progression-free survival of 5 weeks was associated with panitumumab monotherapy relative to best supportive care, no statistically significant effect on overall survival was observed in the trial. The Committee heard from 1 of the clinical specialists that in trials of metastatic colorectal cancer, gains in progression-free survival cannot reliably translate to gains in overall survival. The Committee was aware that most people in the study who had been randomised to receive best supportive care crossed over to receive panitumumab. The Committee noted that 1 consultee proposed that analyses adjusting for crossover did not adjust for the adverse reactions related to panitumumab treatment. It heard the Assessment Group's view that the manufacturer's analyses to adjust for crossover reflected a reasonable approach. The Committee concluded that panitumumab provided a survival benefit relative to best supportive care, but that the magnitude of this benefit was uncertain.

The Committee discussed the results of the Assessment Group's economic analysis for panitumumab, which was based on the HRs for panitumumab from the Assessment Group's mixed treatment comparison, adjusted for crossover. The Committee also noted the analyses carried out by the Assessment Group after public consultation on the assessment report, which used unadjusted HRs from the Amgen trial directly. These analyses resulted in a decrease in the ICER from £150,000 to £109,000 per QALY gained when panitumumab plus best supportive care was compared with best supportive care alone. The Committee also noted that the results of the Assessment Group's one-way sensitivity analyses showed that increasing the mean overall survival estimate for panitumumab plus best supportive care from the base-case value of 6.8 months to 7.2 months (based on an increase of 2 standard errors) resulted in an ICER of £110,000 per QALY gained. The Committee concluded that it was not possible to specify a precise ICER for panitumumab plus best supportive care compared with best supportive care alone, but that the most plausible ICER was likely to be between £110,000 and £150,000 per QALY gained.

The Committee considered supplementary advice from NICE that should be taken into account when appraising treatments that may extend the life of patients with short life expectancy and that are licensed for indications that affect small numbers of people with incurable illnesses. For this advice to be applied, all the following criteria must be met:

• The treatment is indicated for patients with a short life expectancy, normally less than 24 months.

• There is sufficient evidence to indicate that the treatment offers an extension to life, normally of at least an additional 3 months, compared with current NHS treatment.

• The treatment is licensed or otherwise indicated for small patient populations.
  
  In addition, when taking these criteria into account, the Committee must be persuaded that the estimates of the extension to life are robust and that the assumptions used in the reference case of the economic modelling are plausible, objective and robust.

The Committee discussed whether the technologies appraised fulfil the criteria for consideration as life-extending, end-of-life treatments. For metastatic colorectal cancer that has progressed after first-line treatment, the Committee agreed that the technologies fulfil the first criterion related to life expectancy, because estimates of life expectancy from people randomised to best supportive care in the second-line setting were less than 12 months.

For bevacizumab, the Committee agreed that there was no evidence to show how much bevacizumab plus non-oxaliplatin chemotherapy given as second-line treatment extends survival. In addition, the Committee understood that it should take into account all populations which are covered by all indications of the marketing authorisation for a given technology when considering the size of the patient population. The Committee noted that bevacizumab has a marketing authorisation for a number of indications and therefore does not fulfil the criterion of being indicated for a small patient group. The Committee concluded that bevacizumab plus non-oxaliplatin chemotherapy does not meet all of the criteria for a life-extending, end-of-life treatment.

For cetuximab, the Committee acknowledged that cetuximab plus best supportive care prolonged life by a median of 4.7 months in the third-line or later setting relative to best supportive care alone and therefore met the second end-of-life criterion. The Committee was aware from the manufacturer's data that approximately 7600 people have EGFR-positive, KRAS wild-type metastatic colorectal cancer in England and Wales, and only a small proportion of these (approximately 260 to 390 people) would be fit enough for third or subsequent lines of treatment. However, the Committee noted that cetuximab has a marketing authorisation for people with any stage of EGFR-positive KRAS wild-type metastatic colorectal cancer, and also for people with locally advanced and recurrent and/or metastatic head and neck cancer, which has previously been estimated to be a population of about 3,000 (NICE's previous technology appraisal guidance on cetuximab for the treatment of recurrent and/or metastatic squamous cell cancer of the head and neck, now replaced by NICE's technology appraisal guidance on cetuximab). The Committee discussed the decisions from 2 previous NICE technology appraisal appeals and noted that the Appeal Panel recognised that the criterion in the supplementary advice for end-of-life treatments for small patient populations indicated that 'Sufficient regard should be given to recognition of the desirability of developing new treatments in smaller disease areas and that higher prices, and therefore reduced cost effectiveness, were more likely to be justified given the need to recoup costs of development of the product from more limited licences'. The Appeal Panel had concluded that it was appropriate, according to the supplementary advice, to add together the potential patient populations covered by the marketing authorisation for different indications rather than on the basis of actual or recommended use. The Committee therefore concluded that the true size of the cumulative population covered by the marketing authorisation for cetuximab was likely to be over 10,000 patients and was not small, and that cetuximab does not meet all of the criteria for a life-extending, end-of-life treatment.

The Committee considered whether panitumumab provides a life extension of about 3 months. It noted that the manufacturer estimated that the mean life extension (after adjusting for crossover) was between 2.7 and 3.2 months, and that the Assessment Group judged the method used to derive this estimate to be reasonable. The Committee also noted that the progression-free survival benefit for panitumumab was similar to that for cetuximab and therefore there was sufficient evidence to indicate that panitumumab offers an extension to life of approximately 3 months compared with best supportive care alone. The Committee noted that panitumumab has a marketing authorisation for people with KRAS wild-type and EGFR-expressing metastatic colorectal cancer in whom both irinotecan- and oxaliplatin-containing chemotherapy has failed. The Committee agreed that this represents a small patient population. However, the Committee was aware that the Committee for Medicinal Products for Human Use recently recommended an extension of the marketing authorisation for panitumumab in combination therapy for KRAS wild-type metastatic colorectal cancer to first-line and second-line settings. Therefore, it is expected that in the near future panitumumab will be licensed for the treatment of metastatic colorectal cancer in a patient population of similar size to that estimated for cetuximab. The Committee noted the most plausible ICER for panitumumab monotherapy lies between £110,000 and £150,000 per QALY gained. Therefore, the Committee concluded that, even if panitumumab monotherapy met all the criteria for a life-extending, end-of-life treatment, the additional weight that would need to be assigned to the QALY benefits would be too great to justify it as an appropriate use of limited NHS resources.

The Committee noted that testing tumour characteristics, such as the KRAS mutation, allowed clinicians to identify people who were more likely to respond to treatment with cetuximab or panitumumab, and agreed that this was an innovation in the treatment of metastatic colorectal cancer. The Committee also heard from the clinical specialists that in the future, the identification of further KRAS and also BRAF mutations will allow even better identification of people who are likely to benefit from therapy. The Committee considered whether any of the technologies in this appraisal could be considered innovative. It concluded that it had not been presented with a case, substantiated by data, that the treatments add demonstrable and distinctive benefits of a substantial nature that had not already been adequately captured in the QALY measure.