The clinical effectiveness and cost-effectiveness of cetuximab (review of technology appraisal no. 176) and panitumumab (partial review of technology appraisal no. 240) for previously untreated metastatic colorectal cancer: a systematic review and economic evaluation

Incidence and prevalence

In terms of incidence, CRC is the fourth most common cancer in the UK behind breast, lung and prostate cancer, accounting for 13% of all new cancer cases. 3 It is the third most common cancer in both men (14% of the total for men) and women (11% of the total) separately. 3 Table 2 summarises the number of new cases and incidence rates in the UK.

Approximately two-thirds (66%) of cancer cases affect the colon and one-third (34%) affect the rectum, although this distribution varies by sex. 3 The crude incidence rates show that there are 46 and 41 new colon cancer cases per year for every 100,000 men and women in the UK, respectively, and around 29 and 17 new rectal cancer cases per year for every 100,000 men and women in the UK, respectively. 3

Approximately 25% of people present with metastases at initial diagnosis and almost 50% of people with CRC will develop metastases. 4

Prevalence refers to the number of people who have previously received a diagnosis of cancer and who are still alive at a given time point. Some people will have been cured of their disease and others will not. In the UK, > 143,000 people were still alive at the end of 2006, up to 10 years after being diagnosed with CRC (Table 3). 3

Risk factors

Risk factors for CRC include age and family history. In the UK between 2009 and 2011, an average 43% of bowel cancer cases were diagnosed in people aged ≥ 75 years and 95% of cases were diagnosed in those aged ≥ 50 years. 3 The lifetime risk of developing bowel cancer in the UK is 1 in 14 for men and 1 in 19 for women. 3

Mortality

Colorectal cancer is the second most common cause of cancer death in the UK (2012 data), accounting for 10% of all deaths from cancer. 5 In 2012 there were 16,187 deaths from CRC in the UK (Table 4). The crude mortality rates show that there are 28 CRC deaths per year for every 100,000 men in the UK and 23 per year for every 100,000 women. 5

Around 6 in 10 (61%) CRC deaths are due to cancers of the colon and around 4 in 10 (39%) are due to cancers of the rectum. 5 Almost one-fifth (18%) of CRC deaths occur in people aged 60–69 years. 5

Survival and prognosis

Approximately 77% of men survive CRC for at least 1 year, which is predicted to fall to 59% at ≥ 5 years, as shown by age-standardised net survival for people diagnosed with CRC during 2010–11 in England and Wales. 6 Survival for women at 1 and 5 years is slightly lower, with 74% surviving for ≥ 1 year and 58% predicted to survive for at least 5 years. 6

Survival is, however, highly dependent on the stage of disease at diagnosis. Survival by stage is not yet routinely available for the UK because of inconsistencies in the collecting and recording of staging data in the past. However, published estimates suggest that approximately 90% of people diagnosed at the earliest stage will survive for > 5 years, whereas < 10% of people diagnosed with distant metastases will survive for > 5 years. 7 In general, the earlier the diagnosis the higher the chances of survival. 7

Personalised treatment

Normal cell behaviour in multicellular organisms is controlled by a complex network of signalling pathways that ensures that cells proliferate only when they are required to, for example in wound healing. 19 Cancer occurs when normal growth regulation breaks down, usually because of defects within these signalling mechanisms. 19 The rat sarcoma (RAS) genes play an important role in the epidermal growth factor receptor (EGFR) pathway, a complex signalling cascade that is involved in the development and progression of cancer (Figure 2). 20 Signals are passed from protein to protein along several different pathways. Disruption of the signals through mutation of the RAS gene is involved in many tumour types.

FIGURE 2.

Epidermal growth factor receptor signalling pathway. PI-3K, phosphoinositide 3-kinase; PLC-γ, phospholipase C gamma; STAT, signal transducer and activator of transcription. Adapted by permission from Macmillan Publishers Ltd on behalf of Cancer Research UK: British Journal of Cancer. Lo HW, Hung MC. Nuclear EGFR signalling network in cancers: linking EGFR pathway to cell cycle progression, nitric oxide pathway and patient survival. British Journal of Cancer 2006;94(2):184–8. Copyright 2006. 21

The three RAS genes – Kirsten rat sarcoma (KRAS), Harvey rat sarcoma (HRAS) and neuroblastoma rat sarcoma (NRAS) – are the most common oncogenes in human cancer. 19,20 All three are widely expressed, with KRAS expressed in almost all cell types. 19 Published research has demonstrated that mutations in codons 12 and 13 of exon 2 of the KRAS gene are predictive of a reduced response to anti-EGFR therapies in mCRC. 22–29 For this reason, only people with KRAS exon 2 wild-type (WT) tumours were initially approved for treatment with this class of agents. 30–32

More recently, it has been shown that other mutations in genes of the RAS family (mutations in codon 61 of exon 3 and codons 117 and 146 of exon 4 of KRAS, and mutations in exons 2, 3 and 4 of NRAS) are also associated with a reduced response to anti-EGFR therapy. 4,26,28,29,33,34 These developments led the European Medicines Agency (EMA) to update the marketing authorisations for cetuximab and panitumumab in 2013 by restricting the indication in mCRC to the treatment of people with RAS WT tumours (see Interventions considered in the scope of this assessment). 35–40

Exon 2 mutations in the KRAS gene occur in approximately 40% of mCRC cases and other KRAS and NRAS mutations occur in approximately 10% of mCRC cases (Figure 3). 22,26,33,41–44 Approximately 50% of people do not have RAS mutations and are classified as RAS WT.

FIGURE 3.

Grouping of molecular characteristics of tumours: research progress. ID, identification; MT, mutant.

National Institute for Health and Care Excellence TA176: cetuximab for the first-line treatment of metastatic colorectal cancer

National Institute for Health and Care Excellence TA176 states that:

Cetuximab in combination with 5-fluorouracil (5-FU), folinic acid and oxaliplatin (FOLFOX), within its licensed indication, is recommended for the first-line treatment of mCRC only when all of the following criteria are met:

1. the primary colorectal tumour has been resected or is potentially operable

2. the metastatic disease is confined to the liver and is unresectable

3. the person is fit enough to undergo surgery to resect the primary colorectal tumour and to undergo liver surgery if the metastases become resectable after treatment with cetuximab

4. the manufacturer rebates 16% of the amount of cetuximab used on a per patient basis.

Reproduced with permission from National Institute for Health and Clinical Excellence (2009) TA176 Cetuximab for the first-line treatment of metastatic colorectal cancer. Available from: www.nice.org.uk/guidance/TA176. 10 Content accurate at time of going to press

Cetuximab in combination with 5-FU, folinic acid and irinotecan (FOLFIRI), within its licensed indication, is recommended for the first-line treatment of mCRC only when all of the following criteria are met:

1. the primary colorectal tumour has been resected or is potentially operable

2. the metastatic disease is confined to the liver and is unresectable

3. the patient is fit enough to undergo surgery to resect the primary colorectal tumour and to undergo liver surgery if the metastases become resectable after treatment with cetuximab

4. the patient is unable to tolerate or has contraindications to oxaliplatin.

Reproduced with permission from National Institute for Health and Clinical Excellence (2009) TA176 Cetuximab for the first-line treatment of metastatic colorectal cancer. Available from: www.nice.org.uk/guidance/TA176. 10 Content accurate at time of going to press

Patients who meet these criteria should receive treatment with cetuximab for no more than 16 weeks. 10 At 16 weeks, treatment with cetuximab should stop and patients should be assessed for resection of liver metastases. 10

National Institute for Health and Care Excellence TA240: panitumumab for the first-line treatment of metastatic colorectal cancer

The appraisal of panitumumab in combination with chemotherapy for the treatment of mCRC (TA240) was ended because no evidence submission was received from the manufacturer or sponsor of the technology. 11 Therefore, NICE was unable to make a recommendation about the use in the NHS of panitumumab in combination with chemotherapy for the treatment of mCRC. 11

National Institute for Health and Care Excellence TA212: bevacizumab in combination with oxaliplatin and either 5-fluorouracil plus folinic acid or capecitabine for the treatment of metastatic colorectal cancer

Bevacizumab in combination with oxaliplatin and either 5-fluorouracil plus folinic acid or capecitabine is not recommended by NICE for the treatment of mCRC. 12

Cetuximab

Cetuximab is a recombinant monoclonal antibody that blocks the human EGFR and therefore inhibits the proliferation of cells that depend on EGFR activation for growth. 35

Previously, cetuximab was indicated for use in people with EGFR-expressing KRAS WT mCRC. 30,31,52,53 In November 2013, in response to new biomarker data, the Committee for Medicinal Products for Human Use (CHMP) changed the indication to clarify the particular genetic make-up of the cancer that must be present before treatment with cetuximab is initiated. 37,39 Based on this recommendation, cetuximab is now indicated for the treatment of people with EGFR-expressing RAS WT mCRC:

• in combination with irinotecan-based chemotherapy

• in first line in combination with FOLFOX

• as a single agent in people who have failed oxaliplatin- and irinotecan-based therapy and who are intolerant to irinotecan. 35

In this label change, the combination of cetuximab with oxaliplatin-containing chemotherapy is now contraindicated for people with RAS mutant mCRC or for whom the RAS status is unknown. 35

Premedication with an antihistamine and a corticosteroid at least 1 hour prior to the administration of cetuximab should be given. This premedication is recommended before the initial and subsequent infusions. Cetuximab is administered once a week; the initial dose is 400 mg/m2 of body surface area, with subsequent weekly doses of 250 mg/m2 of body surface area. 35

One common AE related to cetuximab treatment is the development of skin reactions, which occurs in > 80% of people and mainly presents as an acne-like rash or, less frequently, as pruritus, dry skin, desquamation, hypertrichosis or nail disorders (e.g. paronychia). 35 The majority of skin reactions develop within the first 3 weeks of treatment. 35 The Summary of Product Characteristics (SPC) notes that, if a person experiences a grade 3 or 4 skin reaction, cetuximab treatment must be stopped, with treatment being resumed only if the reaction resolves to grade 2. 35 Other common AEs of cetuximab include mild or moderate infusion-related reactions such as fever, chills, nausea, vomiting, headache, dizziness or dyspnoea that occur soon after the first cetuximab infusion. 35

Panitumumab

Panitumumab is a recombinant monoclonal antibody that targets the EGFR, thereby inhibiting the growth of EGFR-expressing tumours. 36

In June 2013, the CHMP changed the indication for the use of panitumumab for the treatment of mCRC,38,40 restricting use to the treatment of adults with RAS WT mCRC:

• in first line in combination with FOLFOX or FOLFIRI

• in second line in combination with FOLFIRI for people who have received first-line fluoropyrimidine-based chemotherapy (excluding irinotecan)

• as monotherapy after failure of fluoropyrimidine-, oxaliplatin- and irinotecan-containing chemotherapy regimens. 36

In this label change, the combination of panitumumab with oxaliplatin-containing chemotherapy is now contraindicated for people with RAS mutant mCRC or for whom the RAS mCRC status is unknown. 36

The recommended dose of panitumumab is 6 mg/kg of bodyweight given once every 2 weeks. 36 Before infusion, panitumumab should be diluted in 0.9% sodium chloride to a final concentration not exceeding 10 mg/ml. 36

Panitumumab is contraindicated in people with a history of severe or life-threatening hypersensitivity reactions to the active substance or to any of the excipients. 36 The most common AEs observed (incidence ≥ 20%) are skin toxicities (i.e. erythema, dermatitis acneiform, pruritus, exfoliation, rash and fissures), paronychia, hypomagnesaemia, fatigue, abdominal pain, nausea, diarrhoea and constipation. 36

As noted, recent research (see Personalised treatment) has resulted in the CHMP adopting a change to the licensed indication for both cetuximab and panitumumab, restricting use to people with RAS WT mCRC. These developments and changes to the licensed indications provide the rationale for this MTA.

Study characteristics

The 2009 STA (TA17610) identified two RCTs investigating the effectiveness of the addition of cetuximab to either oxaliplatin- (FOLFOX) or irinotecan-based (FOLFIRI) chemotherapy [CRYSTAL (Cetuximab Combined with Irinotecan in First-Line Therapy for Metastatic Colorectal Cancer)24 and OPUS (Oxaliplatin and Cetuximab in First-line Treatment of Metastatic Colorectal Cancer)23]. As research into the impact of KRAS and NRAS tumour mutations on the effectiveness of EGFR inhibitors developed, the ITT populations from the pivotal trials were re-evaluated, forming the basis for the revision of the licensed population.

In total, three subgroup analyses from three randomised, open-label trials {OPUS,65 CRYSTAL43 and FIRE-3 [5-FU, Folinic Acid and Irinotecan (FOLFIRI) Plus Cetuximab versus FOLFIRI Plus Bevacizumab in First Line Treatment of Colorectal Cancer]28} were included in the update review. Of note, in the FIRE-328 trial a protocol amendment was made, restricting eligibility for the ITT population to those with KRAS WT exon 2 tumours, because of the emerging evidence on the negative predictive value of KRAS exon 2 mutations and the subsequent changes to the licence for cetuximab. However, in all of the included trials the extended RAS subgroup analysis of interest to this review was conducted retrospectively.

Of the included trials, two evaluated the addition of cetuximab to background chemotherapy (FOLFOX65 or FOLFIRI43) and one evaluated the addition of cetuximab or bevacizumab to background chemotherapy (FOLFIRI)28 (Table 10). All of the trials evaluated the same dose of cetuximab and administration method.

All of the included trials measured the following outcomes: ORR, PFS, OS, secondary resection of liver metastases with curative intent and safety and tolerability (including the incidence and type of AEs). 28,43,65

In two of the included trials the primary end point was the proportion of participants who had an ORR. 28,65 In the OPUS trial65 tumour response was assessed by an independent review committee according to modified WHO criteria, whereas in the FIRE-3 trial28 tumour response was measured according to the Response Evaluation Criteria in Solid Tumours (RECIST) version 1.0, as assessed by the study investigators. The independent review committee conducted a blinded review of images and clinical data. In the CRYSTAL trial,43 the primary end point, PFS time, was defined as the time from randomisation to disease progression or death from any cause (within 60 days following randomisation or the last tumour assessment). No data were identified for HRQoL for the RAS WT population from any of the included trials.

Median follow-up was not reported in the OPUS trial65 or the CRYSTAL trial. 43 In the FIRE-3 trial28 the median follow-up was 33.0 months [interquartile range (IQR) 19.0–55.4 months] in the cetuximab plus FOLFIRI arm and 39.0 months (IQR 22.5–56.9 months) in the bevacizumab plus FOLFIRI arm.

Population characteristics

The baseline demographic and disease characteristics for the RAS WT subgroup are reported in Table 11.

For the ITT population, for each of the included trials the baseline demographic and disease characteristics were well matched between the groups. In all studies, existing deoxyribonucleic acid samples from KRAS exon 2 WT tumours were reanalysed for other RAS mutations in four additional KRAS codons (exons 3 and 4) and six NRAS codons (exons 2, 3 and 4). Mutation status was evaluable in 796 (73.0%) of 1090 trial participants with KRAS exon 2 WT tumours (see Table 10). The proportions of study participants evaluated to be RAS WT in the different trials are summarised in Table 10. In all trials, the baseline and disease characteristics were comparable with those seen for the KRAS WT population (see Appendix 4 for baseline and disease characteristics for the KRASWT population).

Participants were similar in terms of age, sex distribution and site of primary cancer (see Table 11). However, as is usually the case with cancer trials, the study populations were significantly younger than the general population presenting with mCRC, with a median age of 59–65 years for the study populations (see Table 11) compared with a peak in number of cases in the UK at between 70 and 79 years of age for men and 75 and 85 years for women.

Study characteristics

The appraisal of panitumumab in combination with chemotherapy for the treatment of mCRC (TA24011) was suspended as no evidence submission was received from the manufacturer or sponsor of the technology. As such, all data included in this update review for panitumumab were identified by the Assessment Group’s searches. It is also important to consider that, as for cetuximab, the ITT population from the pivotal trials for panitumumab were re-evaluated in line with research developments on the impact of RAS mutations on the effectiveness of EGFR inhibitors.

For this MTA review, a total of two subgroup analyses of the RAS WT population from two RCTs29,44 evaluating panitumumab were eligible for inclusion. In the PEAK [Panitumumab Plus mFOLFOX6 vs. Bevacizumab Plus mFOLFOX6 for First Line Treatment of Metastatic Colorectal Cancer (mCRC) Patients With Wild-Type Kirsten Rat Sarcoma-2 Virus (KRAS) Tumors] study29 the extended RAS subgroup analysis was prespecified. In the PRIME (Panitumumab Randomized trial In combination with chemotherapy for Metastatic colorectal cancer to determine Efficacy) study44 the extended RAS subgroup analysis was noted alongside a protocol amendment restricting the analysis of the ITT population to compare PFS and OS according to KRAS status.

Of the two included trials, one evaluated the addition of panitumumab to background chemotherapy (FOLFOX4)44 and one evaluated the addition of panitumumab or bevacizumab to background chemotherapy [modified FOLFOX6 (mFOLFOX6)]. 29 All trials evaluated the same dose of panitumumab and administration method (Table 12). No clinical evidence assessing the effectiveness of panitumumab in conjunction with FOLFIRI was identified.

Both of the included trials measured the following outcomes: ORR, PFS, OS, secondary resection of liver metastases with curative intent, and safety and tolerability (including the incidence and type of AEs). 29,44 The primary end point in both trials was PFS, defined as the time from randomisation to disease progression or death from any cause within 60 days after the last tumour assessment or after randomisation. No data were identified for HRQoL for the RAS WT population from the included trials.

Median follow-up in the PRIME trial was 22.31 months (IQR 10.12–35.65 months) for the panitumumab plus FOLFOX4 treatment group and 17.71 months (IQR 8.74–32.20 months) for the FOLFOX4 alone treatment group. 44 In the PEAK trial, median follow-up was 14.97 months (IQR 8.83–22.81 months) in the cetuximab plus mFOLFOX6 treatment group and 14.93 months (IQR 8.76–21.39 months) in the bevacizumab plus mFOLFOX6 treatment group. 29

Population characteristics

Baseline demographic and disease characteristics for the RAS WT subgroup are reported in Table 13.

In all studies, existing deoxyribonucleic acid samples from KRAS exon 2 WT tumours were reanalysed for other RAS mutations in four additional KRAS codons (exons 3 and 4) and six NRAS codons (exons 2, 3 and 4). Mutation status was evaluable in 682 (65.6%) out of 1345 trial participants with KRAS exon 2 WT tumours (see Table 12). The proportions of study participants evaluated to be RAS WT for the different studies are summarised in Table 12. In all trials, the baseline demographic and disease characteristics were comparable to those seen for the KRAS WT population (see Appendix 4 for baseline and disease characteristics for the KRAS WT population).

Participants were similar between the arms in terms of age, sex distribution and site of primary cancer (see Table 13). However, as is usually the case with cancer trials, the study populations were significantly younger than the general population presenting with mCRC, with a median age of 60–62 years for the study populations (see Table 13) compared with a peak in the number of cases in the UK at between 70 and 79 years of age for men and 75 and 85 years for women.

Treatment allocation

The method of random allocation, including the method of sequence generation, was clearly stated and adequate for all of the included trials. All trials used a stratified permuted block procedure. Stratification factors varied between the studies but were predominantly based on Eastern Cooperative Oncology Group (ECOG) performance status (0 or 1 vs. 2) and region (Eastern or Western Europe vs. outside of Europe and Western Europe, Canada and Australia vs. rest of the world).

However, data for those with RAS WT mCRC were available only from subgroup analyses and not from the ITT trial population for any of the included trials. In response to research developments demonstrating a treatment interaction of RAS and EGFR inhibitors (specifically the negative impact of RAS mutations on the effectiveness of EGFR inhibitors), tumour samples from participants of the original RCTs were re-evaluated for RAS status. None of the included studies stratified randomisation by RAS status; this was because the impact of RAS mutations on the effectiveness of EGFR inhibitors was not known at the protocol development phase. For four of the trials (OPUS,65 CRYSTAL,43 FIRE-328 and PRIME44) the subgroup analyses were retrospective. However, for two of these trials (PRIME and FIRE-3) protocol amendments were made in line with research developments. The only trial in which the extended RAS WT subgroup analysis was prespecified was the PEAK trial. 29

Tumour samples from participants in the ITT population identified as KRAS exon 2 WT were re-evaluated for RAS mutations and allocated to either the RAS WT subgroup or the RAS mutant subgroup. The methods used to detect RAS mutations varied between studies, minimising the potential for ascertainment bias. The ascertainment rate for RAS testing of participants with KRAS WT status was 78% [excluding the PRIME trial in which all participants were tested (n = 512/1060)] and the missing data largely resulted from unavailable tumour samples or inconclusive RAS test results. Of note, none of the included subgroup analyses reported the results of a test for treatment interaction.

Similarity of the groups

Three of the included trials fully reported baseline characteristics for the RAS WT population (OPUS,65 CRYSTAL43 and PEAK29). Although the other two trials (PRIME44 and FIRE-328) did not report baseline characteristics for the subgroup of interest in the trial publication, we were able to confirm these through Merck Serono and Amgen [19 March 2015, personal communication (via NICE)]. Of note, however, baseline characteristics provided by the manufacturer for the PRIME study were for a total of 505 participants, whereas the study publication44 reports a total of 512 participants in the RAS WT subgroup.

Given the use of subgroup data, all comparisons were made without protection by stratification/randomisation, increasing the risk of selection bias. However, from the evidence provided (published and unpublished) we were able to confirm that the treatment groups were similar at baseline on a range of prognostic factors for the RAS WT population. Moreover, the characteristics were similar to those for both the ITT population and the KRAS WT population, suggesting a low risk of selection bias in the RAS-tested trial population.

Implementation of masking

The trials had an open-label design and, as such, participants and outcomes assessors were not blinded. However, a blinded retrospective review of radiological assessment and clinical data was carried out for progression and best ORR in two of the studies43,65 and for ORR in one study. 44 In addition, in one study44 an Independent Data Monitoring Committee reviewed interim analyses of safety and one descriptive interim analysis of PFS. No independent assessment was performed in either the PEAK trial29 or the FIRE-3 trial. 28

Completeness of trial data

With regard to the reporting of a priori outcomes, all included trials were rated as unclear. This was because the original trial reports for the ITT population failed to explicitly state if all outcomes defined in the study protocol were reported. Therefore, we were by default unable to assess if all a priori outcomes had been reported for the RAS WT population. Summary data, including event numbers and denominators, were reported for the majority of expected outcomes for the RAS WT population and, when not reported, we were able to confirm data (predominantly ORRs and resection rates) using secondary sources, for example EMA documents30–32,38–40,52,53 or through the manufacturers [Merck Serono and Amgen, 19 March 2015, personal communication (via NICE)].

Withdrawals and dropouts were adequately reported in all of the original trial publications (by providing numbers and reasons by treatment group in the form of a Consolidated Standards of Reporting Trials flow diagram) for the ITT population. Loss to follow-up was, however, unclear. With respect to the RAS WT population, missing data largely resulted from unavailable tumour samples or inconclusive RAS test results.

Currently, available data on the effectiveness of both cetuximab and panitumumab in the RAS WT population are from subgroup analyses and not from the ITT trial population and, as such, ITT analysis was not conducted. Because of the retrospective nature of the RAS analysis, a low number of samples was available for analysis, reducing the power of the studies to show statistical significance.

Applicability to the NHS in England

The population evaluated is in line with that specified in the licensed indication and the NICE final scope. 14 The study arm populations had median/mean ages of between 59 and 65 years and the majority of participants had an ECOG performance status of < 2, meaning that participants were younger and fitter than the UK population of people with mCRC. This is a recurrent problem, however, in the findings of trials of therapies for mCRC in the UK population. All of the included studies were multicentre studies (including European centres) and evaluated the study drugs in line with their licensed indications. Importantly, however, data for the RAS WT population were available only from subgroup analyses rather than ITT analyses and, as such, sample sizes were often small and the results are subject to a high degree of uncertainty.

The rationale for the use of subgroup data is based on research developments that demonstrated that genotype is an important determinant of both the response to treatment and the susceptibility to adverse reactions for a wide range of drugs. 60,61 In CRC response to EGFR inhibitors has been shown to be dependent on gene expression; studies have demonstrated a treatment interaction between RAS status and the effectiveness of EGFR inhibitors. 62–64 It was in line with these research developments evaluating the negative impact of RAS mutations on the effectiveness of EGFR inhibitors that tumour samples from trial populations supporting the original licensed indications were evaluated retrospectively for RAS status. Therefore, data are not from the ITT population for any of the included studies, but are from a subgroup of people contained within the original RCTs.

Although subject to the uncertainties outlined above, these subgroup data are currently the only available data for the RAS WT subpopulation. The Assessment Group did not identify any RCTs with an ITT population by RAS WT status and only one of the included trials29 prespecified the extended RAS analysis. Of note, the EMA’s recent change to the licensed indication was based on subgroup data from trials that inform this current assessment and, although subgroup analyses were defined post hoc, the rationale was based on research developments into tumour biology and results were in line with the expected direction of effect and consistent across included studies. Hence, the extent to which the results of the included trials can provide a reasonable basis for generalisation to the UK NHS population of people with mCRC is unclear.

Cetuximab

Panitumumab

Cetuximab

All of the included cetuximab trials reported AEs. 28,43,65 All trials reported grade 3 or 4 events,28,43,65 with one trial also reporting grade 1 or 2 events. 43 Two trials also reported summary AE information (any AEs or any serious AEs). 43,65

As RAS mutation status refers to the tumour only, the EMA concluded in its report that there were no good reasons to postulate differences in safety profiles related to RAS status other than from the perspective that people with RAS WT tumours would be treated for longer periods of time. Taking small sample sizes into account, the assumption that safety is independent of tumour RAS status was considered to be in line with reported data. 39

Panitumumab

Data were available for AEs from both the PRIME trial44 and the PEAK29 trial. Both trials reported any AEs, AEs with a worst grade of 3, AEs with a worst grade of 4, AEs with a worst grade of 5, any grade 1 or 2 AEs, any grade 3 or 4 AEs and any serious AEs. AEs with a worst grade of 1 or 2 and AEs with a worst grade of 3 or 4 were available from the PEAK trial29 but not the PRIME trial. 44

The EMA concluded that no new safety concerns were identified for the safety profile of panitumumab in patients with RAS WT tumour status as these patients were indistinguishable from patients with KRAS WT tumour status.

Progression-free survival

All three RCTs contributed to the estimation of PFS. The NMA found no evidence to suggest that cetuximab plus FOLFOX is any more effective than panitumumab plus FOLFOX at increasing the time to progression (TTP) or death (HR 0.74, 95% Crl 0.36 to 1.49) (Table 30). Nevertheless, as the upper 95% CrI for cetuximab plus FOLFOX compared with all of the other treatments is > 1, it is possible that cetuximab plus FOLFOX could be associated with greater progression or death than FOLFOX, bevacizumab plus FOLFOX or panitumumab plus FOLFOX.

The direct evidence from the PRIME26 and PEAK29 trials suggests that panitumumab plus FOLFOX is more effective than FOLFOX (HR 0.72, 95% CrI 0.58 to 0.90) and bevacizumab plus FOLFOX (HR 0.65, 95% CrI 0.44 to 0.96).

Overall survival

All three RCTs contributed to the estimation of OS. The analysis suggests that there is no evidence that panitumumab plus FOLFOX is more effective than cetuximab plus FOLFOX (HR 1.22, 95% CrI 0.71 to 2.11), as the upper 95% CrI is > 1 (Table 31).

The direct evidence from the PRIME trial26 suggests that panitumumab plus FOLFOX is more effective than FOLFOX (HR 0.77, 95% CrI 0.64 to 0.93).

Objective response rate

All three RCTs contributed to the estimation of ORR. The ORR was measured at either 6- or 8-week intervals (according to methods reported in the primary publications). However, because of differences in the reporting of the timing of the ORR in each study, it is unclear whether or not the timings are entirely comparable across studies. Given this uncertainty, results reported for the RAS WT population for this outcome should be treated with caution.

The NMA suggests that there is little evidence that cetuximab plus FOLFOX is any more effective than panitumumab plus FOLFOX for ORR (OR 1.90, 95% CrI 0.72 to 5.02) (Table 32).

Resection rates

Only data from the PRIME and PEAK trials were available to analyse resection rates; therefore, a comparison with cetuximab plus FOLFOX could not be made. The data suggest that there is little difference in resection rates between the treatments, as the 95% CrIs all include a value of 1 (Table 33).

Adverse events

The indirect evidence suggests that there is no difference in the ORs for any grade 3/4 AEs or any serious AEs between cetuximab plus FOLFOX and panitumumab plus FOLFOX (Tables 34 and 35). However, panitumumab plus FOLFOX is estimated (from direct evidence) to be associated with more grade 3/4 AEs than FOLFOX or bevacizumab plus FOLFOX. The evidence is less clear for cetuximab plus FOLFOX compared with FOLFOX or bevacizumab plus FOLFOX as the 95% CrIs include a value of 1 (see Table 34).

The results of analyses of specific grade 3/4 AEs are shown in Tables 36–39. The available information allows estimation of the ORs for cetuximab plus FOLFOX compared with panitumumab plus FOLFOX for neutropenia (see Table 36), paraesthesia (see Table 37), rash (see Table 38) and skin conditions (see Table 39). The estimated ORs (and 95% CrIs) suggest that there is little difference in the number of individuals experiencing these AEs between cetuximab plus FOLFOX and panitumumab plus FOLFOX. Note that for the outcomes of rash and skin conditions, the 95% CrIs are very wide because of the low number of events reported in all three RCTs.

For the remaining AEs, the OPUS trial33 did not provide the required information and so no comparisons could be made between cetuximab plus FOLFOX and panitumumab plus FOLFOX for diarrhoea, hypokalaemia, hypomagnesaemia, mucositis/stomatitis, mucosal inflammation, fatigue, peripheral neuropathy or asthenia. Instead, analyses are reported to allow an indirect comparison to be made between bevacizumab plus FOLFOX and FOLFOX (see Appendix 5). Note that, because of the small numbers of events for hypomagnesaemia, mucositis/stomatitis and mucosal inflammation, the 95% CrIs are wide.

Subgroup analysis by liver metastases at baseline

Restricting the evidence to the subgroup of people with liver metastases at baseline has little impact on the overall conclusions: there is limited evidence to suggest any difference between cetuximab plus FOLFOX and panitumumab plus FOLFOX for PFS (Table 40), OS (Table 41) and ORR (Table 42) as the 95% CrIs all include a value of 1.

Only data from two RCTs were available for the analysis of surgical resection rates for the liver metastases subgroup (Table 43). As the OPUS trial did not report this outcome, no comparison could be made between cetuximab plus FOLFOX and panitumumab plus FOLFOX. The available data suggest that there is little evidence of a difference in surgical resection rates between FOLFOX, bevacizumab plus FOLFOX and panitumumab plus FOLFOX.

For complete resection, all three RCTs reported relevant evidence and so a comparison could be made between panitumumab plus FOLFOX and cetuximab plus FOLFOX (Table 44). However, there was very little evidence to say that any one treatment was associated with a greater number of complete resections than any other, although these analyses were based on a small number of participants.

Progression-free survival

The NMA suggests that bevacizumab plus FOLFIRI is more effective than FOLFIRI at increasing TTP or death (HR 0.60, 0.41 to 0.88) and cetuximab plus FOLFIRI is more effective than FOLFIRI at increasing TTP or death (HR 0.56, 0.41 to 0.76) (Table 45). Evidence from the FIRE-3 RCT suggests that cetuximab plus FOLFIRI is no more effective than bevacizumab plus FOLFIRI at increasing TTP or death (see Table 45).

Overall survival

The NMA suggests that there is no evidence that bevacizumab plus FOLFIRI is more effective than FOLFIRI at increasing OS; however, the evidence indicates that cetuximab plus FOLFIRI is more effective than both FOLFIRI and bevacizumab plus FOLFIRI at increasing OS (Table 46).

Objective response rate

Two RCTs contributed to the estimation of ORR in the FOLFIRI network. 28,43 However, because of differences in the reporting of the timing of ORR in each study, it is unclear whether or not the timings are entirely comparable across studies. Given this uncertainty, the results reported for the RAS WT population for this outcome should be treated with caution.

The NMA suggests that bevacizumab plus FOLFIRI and cetuximab plus FOLFIRI are both more effective than FOLFIRI for the outcome of ORR; however, the evidence that cetuximab plus FOLFIRI is any more effective than bevacizumab plus FOLFIRI for ORR is uncertain because of the wide 95% CrI (OR 1.28, 95% CrI 0.83 to 1.99) (Table 47).

Adverse events

The NMA suggests that bevacizumab plus FOLFIRI and cetuximab plus FOLFIRI are associated with more grade 3/4 AEs than FOLFIRI (Table 48) and that cetuximab plus FOLFIRI is associated with more skin conditions than FOLFIRI or bevacizumab plus FOLFIRI (Table 49). For diarrhoea, the evidence is unclear regarding whether or not one treatment is associated with more cases than any other treatment (Table 50).

Sensitivity analyses

Addition of FIRE-3 data from the manufacturer’s submission (15 June 2015, version 2; see also Appendix 5) to the analysis of HRs for progression or death (Table 51), HRs for OS (Table 52) and ORs for ORR (Table 53) had very little effect on the overall conclusions for the FOLFIRI network.

Cetuximab

• Two trials provided evidence for the effectiveness of cetuximab in combination with chemotherapy (FOLFOX or FOLFIRI) compared with chemotherapy alone (FOLFOX or FOLFIRI). 43,65 The evidence consistently suggested a treatment effect in favour of the addition of cetuximab to chemotherapy compared with chemotherapy alone.

  • Median PFS ranged from 11.4 months in the CRYSTAL study to 12 months in the OPUS study for the experimental arms, and from 5.8 months in the CRYSTAL study to 8.4 months in the OPUS study for the control arms.

  • Median OS ranged from 19.8 months in the CRYSTAL study to 20.4 months in the OPUS study for the experimental arms, and from 17.8 months in the CRYSTAL study to 20.2 months in the OPUS study for the control arms.

  • Tumour response rates ranged from 58% in the OPUS study to 66% in the CRYSTAL study in the experimental arms, and from 29% in the OPUS study to 60% in the CRYSTAL study in the control arms.

  • In people with liver metastases at baseline, the results in terms of improvement in OS and PFS were consistent with the results for the overall RAS WT population. Of people with liver metastases at baseline, 13.3% in the OPUS study and 16.3% in the CRYSTAL study had complete resection in the experimental arms.

  • Overall, clinical safety data for the RAS WT population were consistent with the results for the KRAS WT population in both of the trials. The most common AEs were diarrhoea, haematotoxicity, neutropenia and skin reactions.

• One trial provided evidence for the effectiveness of cetuximab in combination with chemotherapy (FOLFIRI) compared with bevacizumab in combination with chemotherapy (FOLFIRI). 28

  • The proportion of people who achieved an objective response was similar between the cetuximab plus FOLFIRI arm and the bevacizumab plus FOLFIRI arm. However, the association with longer OS suggests a benefit for cetuximab plus FOLFIRI compared with bevacizumab plus FOLFIRI (HR 0.70, 95% CI 0.53 to 0.92).

Panitumumab

• One trial provided evidence for the effectiveness of panitumumab in combination with chemotherapy (FOLFOX4) compared with chemotherapy alone (FOLFOX4). No evidence was identified comparing panitumumab plus FOLFIRI with FOLFIRI. The evidence consistently suggested a treatment effect in favour of the addition of panitumumab to FOLFOX4 compared with FOLFOX4.

  • Median PFS was 10.1 months in the experimental arm and 7.9 months in the control arm.

  • Median OS was 25.8 months in the experimental arm and 20.2 months in the control arm.

  • In people with liver metastases at baseline, the results in terms of improvement in OS and PFS were consistent with the results for the overall RAS WT population.

  • Overall, clinical safety data for the RAS WT population were consistent with results for the KRAS WT population in all of the trials. The most common AEs were diarrhoea, haematotoxicity, neutropenia and skin reactions.

• One trial provided evidence for the effectiveness of panitumumab in combination with chemotherapy (mFOLFOX6) compared with bevacizumab in combination with chemotherapy (mFOLFOX6).

  • The proportion of people who achieved an objective response was similar between the panitumumab plus mFOLFOX6 arm and the bevacizumab plus mFOLFOX6 arm. For PFS, the addition of panitumumab to mFOLFOX6 was associated with a 35% reduction in the risk of progression compared with bevacizumab plus mFOLFOX6. In addition, a trend towards an OS benefit with panitumumab plus mFOLFOX6 was observed (HR 0.63, 95% CI 0.39 to 1.02).

Summary of the network meta-analysis

• A NMA was also conducted based on the trials identified. It was not possible to construct a complete network and so two discrete networks were generated, one evaluating FOLFOX-containing chemotherapy regimens and the other comparing FOLFIRI-containing chemotherapy regimens.

Network meta-analysis

Amgen performed a NMA to compare panitumumab in combination with FOLFOX with other identified comparators in the scope (Amgen submission, section 2.1, p. 51).

The company conducted a systematic review: the search strategy combined ‘drug names’ with ‘disease terms’ and ‘study design terms’ (the search strategy was provided as an appendix). Inclusion criteria for the NMA were in line with the PICO (population, intervention, comparator, outcome) criteria specified in the NICE scope14 (Amgen submission, section 2.1, p. 51).

Evidence informing the NMA came from a total of 21 RCTs [reported in 22 publications28,29,33,34,44,78–94 and one unpublished article (Amgen Ltd. Data on File: Supplemental CSR 20050203 RAS/BRAF Analysis. 15 April 2013)]. Four trials84,87,88,92 were excluded from the primary analysis because of population differences or differences in the treatment regimen administered. Based on the 17 RCTs, Amgen built one network. Studies excluded from the company’s primary analysis were included in a sensitivity analysis. The sensitivity analysis included clinically similar chemotherapy regimens [FOLFOX/XELOX and FOLFIRI/XELIRI (capecitabine + irinotecan)] and relevant comparators (FOLFOX, XELOX, XELIRI and cetuximab plus FOLFOX/FOLFIRI). There were too few data to perform a NMA comparing panitumumab plus FOLFOX or FOLFIRI with the comparators of interest in the subgroup of people with liver metastases.

The study designs of the included studies were comparable; however, not all studies reported all outcomes of interest (OS, PFS or ORR) and, hence, not all studies contributed to the analysis for each outcome (see Amgen submission, appendix 8, pp. 27–35). In addition, disease progression and response rate were not assessed using the same method in all of the included studies; however, it was assumed that this had no impact on the comparative treatment effect on the PFS or ORR end points. Population characteristics were assumed to be the same between the studies; however, the studies evaluating a non-EGFR inhibitor included those with mixed or unknown RAS status.

The company used meta-analysis techniques (random effects with fixed effects examined in sensitivity analysis) to pool direct comparisons using SAS version 9.2 software (SAS Institute Inc., Cary, NC, USA). For indirect comparisons, the company used the method of Bucher et al. 95 The indirect estimate of panitumumab compared with the comparator was calculated using the results of the direct comparisons with a common control. Both fixed- and random-effects meta-analyses were used. Each indirect comparison was estimated separately within the indirect comparison framework. Within the indirect comparison, the underlying assumptions of homogeneity, similarity and consistency were reviewed according to guidelines by Song et al. 96 Details of the implementation of the meta-analysis and indirect comparison were provided in the Amgen submission (appendix 8).

For the NMA, a Bayesian framework with Markov chain Monte Carlo simulation was used, using methodology outlined by Ades et al. 97 Analyses were performed using SAS version 9.3. Non-informative priors were used. Analyses were run with an initial burn-in of 10,000 iterations followed by an additional 50,000 iterations. To address the potential for autocorrelation, it was necessary to thin the samples generated through SAS (a thinning factor of 40 was used). The posterior mean/median and 95% CrI were reported together with the probability that each treatment was better (more effective) than the others. Within the indirect comparison, the underlying assumptions of homogeneity, similarity and consistency were reviewed according to guidelines by Song et al. 96 Convergence of the models was examined and Amgen noted that, in some cases, the models for the treatment arm-level analyses did not converge to a stationary distribution, showing a high level of autocorrelation between draws of the Markov chain, even with thinning factors of ≥ 100 and a burn-in period of > 1,000,000 iterations attempted. The results for these models were not shown; the company note that this was because of their unsuitability. Details of the implementation of the mixed-treatment comparison are provided in the Amgen submission (appendix 8).

Point estimates for relative effectiveness (including 95% CrIs and the probability of each treatment being the better treatment) were reported in full in the Amgen submission (appendix 8, pp. 41–2 and pp. 87–97) and full results (including the results of the sensitivity analyses conducted) were reported (appendix 8, pp. 87–97). Amgen’s NMA was not used to analyse liver resection rates or AEs.

The following limitations of the NMA were acknowledged: (1) data for non-EGFR inhibitors were from populations with mixed or unspecified RAS status and (2) data for the RAS WT population were not from the protocol-defined population for any of the EGFR inhibitor studies and the results were not for the ITT population but a retrospective subgroup.

Comparison with the Assessment Group’s network-meta-analysis

Of the studies included in Amgen’s NMA [n = 21, reported in 22 publications28,29,33,34,44,78–94 and one unpublished article (Amgen Ltd. Data on File: Supplemental CSR 20050203 RAS/BRAF Analysis. 15 April 2013)], 18 (including the unpublished article) were not included in the Assessment Group’s NMA. 78–94 The reason for their exclusion was that they did not evaluate the effectiveness of the interventions in the RAS WT population. In addition to the abstracts for the OPUS33 and CRYSTAL34 trials included in the Amgen NMA, the Assessment Group identified the full publications. 43,65

Evidence from the included studies enabled the company to construct a complete network. The study by Badulescu et al. 78 compared FOLFOX and FOLFIRI and enabled the complete network approach based on the assumption that there was little difference between FOLFOX and FOLFIRI in terms of effectiveness. The NMA conducted by the Assessment Group included two separate networks (FOLFOX and FOLFIRI) as none of the included studies provided evidence to link the two networks for the RAS WT population.

Assumptions regarding the similarity between the included trials in terms of study design were considered by the Assessment Group to be appropriate. However, in terms of population characteristics, although data included in the NMA for panitumumab and cetuximab were restricted to the RAS WT population in line with the population specified in the NICE scope,14 data for non-EGFR inhibitors were not available for the RAS WT population, given that efficacy is not contingent on the expression of the RAS genotype. Although the Assessment Group consider this to be a logical approach, it should be noted that data included in the NMA for non-EGFR inhibitor treatments came from study populations with mixed or unspecified RAS status, the likely impact of which would be to increase the uncertainty surrounding the effect estimates.

Analyses were conducted for the following outcomes: PFS, OS, ORR, CR and PR. Time-to-event data were analysed using study-level data (HR) and response rate data were analysed using study-level data (relative risk). The company noted that there were too few data to perform a NMA for panitumumab plus FOLFOX compared with cetuximab plus FOLFOX or cetuximab plus FOLFIRI in the subgroup of people with liver metastases.

The methods used in Amgen’s NMA were in line with guidance set out in the publication by Ades et al. 97

Despite the broader approach taken, the results for panitumumab plus FOLFOX compared with FOLFOX were similar to those of the Assessment Group’s NMA for OS and PFS. The effect estimates for this comparison for all outcomes showed a greater effect of panitumumab plus FOLFOX than FOLFOX, but the 95% CrIs were wider in the Assessment Group’s results. There was no evidence to suggest that panitumumab plus FOLFOX was any more or less effective than cetuximab plus FOLFOX in terms of TTP or death or time to death. All of the results, however, are subject to uncertainty as a result of the acknowledged limitations. As the Assessment Group’s NMA focused entirely on the RAS WT population, no comparison could be made with Amgen’s analysis of panitumumab plus FOLFOX compared with XELOX. Also, given that the Assessment Group’s approach to the NMA resulted in two networks, the results could not be compared with Amgen’s NMA for panitumumab plus FOLFOX compared with either FOLFIRI or cetuximab plus FOLFIRI.

Network meta-analysis

Merck Serono performed a NMA to compare cetuximab plus chemotherapy (FOLFOX or FOLFIRI) for the treatment of RAS WT mCRC with other comparators specified in the NICE scope14 (Merck Serono submission, section 2, p. 51).

The company conducted a systematic review, with the search strategy combining ‘drug names’ with ‘disease terms’ and ‘study design terms’ (the search strategy was provided as an appendix). Inclusion criteria for the NMA were in line with the PICO criteria specified in the NICE scope14 (see section 2.1, p. 51).

Six trials were included in the NMA. 28,29,43,44,101 Evidence from these trials enabled one complete network to be established for the outcomes of OS and PFS. This was possible as the CALGB-80405 trial compared cetuximab plus FOLFOX or FOLFIRI with bevacizumab plus FOLFOX or FOLFIRI, reporting separate Kaplan–Meier curves for each of the possible combination therapies. 101 Within the global network, a sensitivity analysis was also conducted with FOLFOX and FOLFIRI grouped as generic chemotherapy. The complete network approach was not possible for ORR as neither the PEAK trial29 nor the CALGB-80405 trial101 reported ORR and, as a result, only a FOLFIRI network was possible for this outcome. It was also not possible to include the CALGB-80405 trial101 in any safety outcome network because of lack of reporting. Therefore, two separate networks, one for FOLFOX and one for FOLFIRI, were created to allow an indirect treatment comparison for safety outcomes.

The study designs of the included studies were comparable. Although Merck Serono noted that disease progression was not assessed using the same method in all of the included studies, it was assumed that this had no impact on the comparative treatment effect for the PFS end point. For the safety outcomes, in the absence of reported data for the RAS WT population in the PRIME trial,44 Merck Serono used data reported for the KRAS WT population. Although the company pre specified safety outcomes of interest, not all could be analysed because of limited reporting in several trials.

Population characteristics were assumed to be the same, although for some trials the baseline characteristics for the RAS WT population were not reported44 or very little published information was available101 and data from the KRAS WT population were used as a proxy. Merck Serono highlighted differences with respect to disease progression (ECOG performance status ≤ 2 in four of the trials28,43,44 vs. 0 or 1 in two of the trials29,101). However, the proportion of participants with an ECOG performance status of 2 in the OPUS and PRIME44 studies was low and, as such, was not considered to have an impact on the comparative treatment effect. It was assumed that both FOLFOX regimens (FOLFOX4 and mFOLFOX6) have a comparable effect.

Network meta-analyses were undertaken using a Bayesian approach with Markov chain Monte Carlo simulation in WinBUGS. Non-informative prior distributions were used. For the analysis of PFS, OS and ORR, models with a normal likelihood and identify link were used. In addition, survival data extracted from the Kaplan–Meier curves were also analysed using a binomial or log-likelihood and log-link using a fractional polynomial model. Analysis of AEs used a model with a binomial likelihood and logit link. Analyses were run with an initial burn-in of 10,000 iterations (30,000 for the fractional polynomial models), followed by an additional 80,000 iterations (30,000 for fractional polynomials) and convergence of the samples was examined visually. Monte Carlo error was checked to ensure that it was ≤ 5% of the posterior standard deviation (SD) for the parameters examined. Both fixed- and random-effects models were used. The deviance information criterion was used to compare the fixed- and random-effects models to determine goodness of fit (deviance information criterion values were reported for both models); when a difference of < 5 was observed, a fixed-effects model was reported and the results of the random-effects model were reported in appendix B (Merck Serono submission). The posterior mean/median and 95% CrI were reported together with the probability that each treatment was better (more effective) than the others.

Point estimates for relative effectiveness (including the 95% CrI and the probability of each treatment being the better treatment) were reported in full in the Merck Serono submission (pp. 51–82). Table 59 summarises the results for OS, PFS and ORR for cetuximab plus FOLFOX and cetuximab plus FOLFIRI compared with relevant comparators. In terms of AEs (not shown here), cetuximab plus FOLFIRI was associated with more events than FOLFIRI alone for grade 3–4 venous thromboembolism, skin reactions, acne-like rash, mucositis, neutropenia, hypokalaemia, hypomagnesaemia and paronychia. Compared with bevacizumab plus FOLFIRI, cetuximab plus FOLFIRI was worse for skin reactions, acne-like rash, hypokalaemia, hypomagnesaemia and paronychia. However, cetuximab plus FOLFIRI was better than bevacizumab plus FOLFIRI for nausea (all grades) and vomiting (all grades). For the FOLFOX network, cetuximab plus FOLFOX was worse than FOLFOX alone for grade 3–4 pulmonary embolism and skin reactions. Compared with panitumumab plus FOLFOX, cetuximab plus FOLFOX was worse for grade 3–4 skin reactions.

The following limitations of the NMA were noted: (1) because of the retrospective nature of the RAS analysis, for some studies there was a low number of samples available for analysis, reducing the power of the studies to show statistical significance and (2) limited data were available on safety for the CALGB-80405 trial,101 resulting in many of the indirect comparison analyses having very wide CIs and making interpretation of the indirect comparison difficult.

Comparison with the Assessment Group’s network meta-analysis

Of the studies included in the NMA, only the CALGB-80405 trial101 was not included in the NMA conducted by the Assessment Group. The CALGB-80405 trial compared cetuximab plus FOLFOX or FOLFIRI with bevacizumab plus FOLFOX or FOLFIRI; however, participants were randomised only to the cetuximab or bevacizumab component of the treatment, with the background chemotherapy (FOLFOX or FOLFIRI) chosen at the discretion of physicians. In addition, data from the CALGB-80405 trial are currently available only in abstract form. For these reasons this study was excluded from the Assessment Group’s systematic review and NMA. No trials were identified analysing the effectiveness of panitumumab plus FOLFIRI compared with any of the comparators specified in the NICE scope. 14

Using the CALGB-80405 trial101 enabled Merck Serono to construct a complete network for the outcomes of PFS and OS. The company conducted two analyses. One analysis used data from participants in the trial according to chemotherapy received; however, in this approach randomisation is broken and this could introduce bias into the analysis. The second analysis was a sensitivity analysis in which the results for FOLFOX and FOLFIRI were pooled as generic chemotherapy, based on the assumption that there was little difference between FOLFOX and FOLFIRI in terms of effectiveness, as reported in the study by Colucci et al. 103 The analysis of ORR and safety outcomes required two separate networks (one for FOLFOX and one for FOLFIRI). The Assessment Group’s NMA used two separate networks (FOLFOX and FOLFIRI) for the analysis of all outcomes in the RAS WT population, as none of the included studies provided evidence to link the two networks.

Assumptions regarding the similarity of the trials in terms of study and population characteristics were considered by the Assessment Group to be appropriate.

The absence of reported data for the PRIME44 and PEAK29 trials meant that analysis of ORR and analysis of all-grade AEs could not be conducted for the FOLFOX network. The Assessment Group, however, had access to unpublished data from the PRIME and PEAK trials and was able to analyse safety outcomes for any grade 3/4 AEs and serious AEs as well as grade 3–4 AEs by type, occurring in ≥ 5% participants in either treatment arm. The Assessment Group also conducted NMA for the outcome of resection rates and also for the subgroup of patients with liver metastases at baseline.

The methods used in Merck Serono’s NMA were in line with guidance from the NICE Decision Support Unit. 66

Despite the slight differences in approach between the Merck Serono NMA and the Assessment Group’s NMA, the overall results were similar, with both analyses subject to significant uncertainty.

History of submission

We received Merck Serono’s original submission on 6 May 2015. We requested an explanation of the discrepancies between the model and the report, as well as how to reproduce the liver metastases subgroup analysis within the model.

Merck Serono submitted a new executable model and report on 15 June 2015, which had one significant change. Merck Serono claimed that it had detected another error of its own in the cost of cetuximab and it adjusted this value accordingly. Some other discrepancies between this model and the previous version were identified, but checks revealed that these were unlikely to have a big impact on the cost-effectiveness results: implementing the changes that we could identify into the original model gave very similar results to the new model (ICERs differed by < £3 per QALY). This also suggested that no major wiring errors had been introduced into this new model. As such, the model methods and results described in this section refer to the version of the model that we received on 15 June 2015.

Merck Serono also submitted an additional executable model for the liver metastases subgroup on 16 June 2015. On request, Merck Serono submitted on 26 June 2015 a list of the parameters that had been altered in the ‘overall population model’ to create this subgroup analysis. The ICERs for this subgroup had again been updated.

Even with the list of parameters, we were unable to reconcile the overall population model and the liver-limited disease subgroup model. We also noted that OS had been hard-coded into this subgroup model, which we believe was in error, as this meant that survival did not alter when different interventions and comparators were selected.

As we could not reconcile this subgroup model with the model for the overall population, and as Merck Serono submitted its independent model for the liver metastases subgroup at a late stage in this HTA assessment, we have not critiqued the liver-limited disease subgroup model. We therefore present the results for this subgroup without comment.

Description of methods

Model parameters

Resection rates

Resection of liver metastases is an important component of both our model and Merck Serono’s model, as cost-effectiveness is sensitive to it.

Merck Serono used the resection rates from the RCTs to estimate the rates for use in its model (Table 66).

TABLE 66 - Liver metastases resection rates assumed in the Merck Serono model

BEV, bevacizumab; CET, cetuximab.

Treatment	All RAS WT patients
FOLFIRI network
CET + FOLFIRI	7.3% (Merck Serono data from the CRYSTAL trial129)
FOLFIRI	2.1% (Merck Serono data from the CRYSTAL trial111)
BEV + FOLFIRI	7.3% (no justification given)
FOLFOX network
CET + FOLFOX	7.3% (derivation explained in accompanying text)
FOLFOX	2.1%111

Merck Serono did not discuss the derivation of their estimate of the rate of resection for cetuximab plus FOLFOX, 7.3%. We assumed that Merck Serono set the resection rate as equal to its rate for cetuximab plus FOLFIRI, which we believe is unreasonable. It estimated the rate of resection for FOLFOX (2.1%) from Tournigand et al. 111 This is substantially lower than our estimate (see Chapter 6, Model parameters, Resection rates). The study by Tournigand et al. 111 was of second-line treatment not restricted to RAS WT, whereas our estimate was taken from the first-line treatment of RAS WT patients; therefore, we prefer our value.

Time of liver resection

Merck Serono simulated liver resection at cycle 3 in its model. Notably, the timing of liver resection was not clearly stated in the submission. As detailed in table 20 (Merck Serono submission, section 3.2.2, p. 49), resection was modelled at cycle/month 4; however, in table 21 the submission states that at 3 months in its model some patients can be referred for curative-intent resection of liver metastases.

Merck Serono’s assumption regarding the timing of liver resection surgery was based on the study by Adam et al. ,113 as indicated in table 20 of the submission (section 3.2.2, p. 49). This assumption seems reasonable, based on advice from our clinical experts and the values used in TA176. 10

Post liver resection: progression-free survival and overall survival

In its submission, Merck Serono stated that it assumed that all patients who underwent curative liver resection for initially unresectable colorectal liver metastases, turned resectable by systematic chemotherapy, and who are cured of the disease ‘remain in a progression free state until death and do not require second-line treatment’ (Merck Serono submission, section 3.2.2, p. 47). However, elsewhere in the submission and in the executable model there exists a progressive disease state, including treatment, for patients post liver resection.

Merck Serono model PFS and OS after liver resection surgery according to data from Adam et al. 113 We also used these data as we understand that they are the most appropriate data available. Further discussion of this study can be found in Chapter 6 (see Model parameters, Post liver resection: progression-free survival and overall survival).

Merck Serono fitted a log-logistic distribution to both PFS and OS post resection (Figure 8). Technically, these data are taken from rows 95 and 96 of Merck Serono’s worksheet ‘Survival models’. Importantly, Merck Serono did not explain its choice of distribution, or indeed how it estimated the curve fits.

FIGURE 8.

Merck Serono’s PFS and OS post-resection fit to the empirical data. Source: figure produced from data available in the Merck Serono submission (May 2015).

The fits appear reasonable up to end of study follow-up at 10 years, which is also the time horizon of Merck Serono’s model. However, after about 11 years, PFS in the Merck Serono model is greater than OS, which is clearly impossible. We believe that this renders the results from Merck Serono’s model for time horizons > 11 years incorrect.

In common with us, for those patients who had a successful resection, Merck Serono assumed that PFS and OS were independent of first-line treatment.

Based on its 10-year time horizon, which we believe is far too short, we calculate that Merck Serono estimates a mean PFS of 2.8 years and mean OS of 4.1 years.

First-line progression-free survival: unresected patients

Merck estimated first-line PFS for unresected patients directly from the pivotal RCTs: CRYSTAL,24 FIRE-328 and OPUS. 23 It compared pairs of treatments independently and did not perform simultaneous comparisons of multiple treatments. Therefore, unlike us, it did not perform indirect comparisons for first-line PFS for unresected patients.

Merck Serono estimated PFS for unresected patients from data for all patients (resected and unresected) in the RCTs. We believe that this is an important mistake. Given that it modelled PFS for resected patients separately, as described in the previous section, it effectively double counted PFS for resected patients. It overestimated PFS for unresected patients because PFS for resected patients (our estimate 4.5 years) is far greater than that for unresected patients (e.g. our estimate for cetuximab plus FOLFIRI 1.0 years).

In our analysis, as explained in the previous section, we also estimated PFS for resected patients from the study by Adam et al. 113 However, we estimated PFS for unresected patients using the RCT data for PFS for all patients and then subtracting PFS for resected patients (see Chapter 6, Model parameters, First-line progression-free survival: unresected patients).

Merck Serono’s choices of statistical distributions and estimates for mean PFS for first-line unresected patients are provided in Table 67.

TABLE 67 - Merck Serono-modelled PFS rates for unresected patients

BEV, bevacizumab; CET, cetuximab.

We estimated mean PFS in the Merck Serono model from the ‘Results’ worksheet, setting the discount rate to 0% and the resection rates in the ‘Setup’ worksheet to 0%.

Treatment	Distribution	Mean PFS (months)a
CET + FOLFOX	Log-normal	13.4
FOLFOX	Log-normal	9.0
CET + FOLFIRI (vs. FOLFIRI)	Weibull	12.5
CET + FOLFIRI (vs. BEV + FOLFIRI)	Weibull	12.8
FOLFIRI	Weibull	8.9
BEV + FOLFIRI	Weibull	10.8

We believe that the PFS curve fits; however, as already stated, we believe that these are overestimates of PFS for unresected patients.

All other things being equal, Merck Serono’s approach results in cetuximab plus FOLFOX/FOLFIRI appearing to be better value for money than we believe, given that a greater proportion of patients in the cetuximab plus FOLFOX/FOLFIRI arms compared with the FOLFOX/FOLFIRI arms are resected and that PFS for resected patients is substantially greater than that for unresected patients.

Probability of postoperative death

Merck Serono states in table 21 of the submission (section 3.2.2, p. 50) that the rate of postoperative death was set to 0%, based on data from the CRYSTAL trial. However, in the executable model Merck Serono assumes a probability of postoperative death of 1% for all treatment regimens. As Merck Serono used data from the study by Adam et al. 113 to model the cohort post resection, we think that it would be more appropriate to use the value of 0.7% reported by Adam et al. 113 for operative mortality within 2 months.

Time on first-line drug treatment

The mean times on first-line drug treatment are extremely important quantities because, in Merck Serono’s model, they affect the total mean cost of drug acquisition and administration per person. In Merck Serono’s model, the former, in particular, is a critical driver of cost-effectiveness. Therefore, treatment duration is worthy of close scrutiny.

Despite its importance, Merck Serono mentioned treatment duration only very briefly.

Merck Serono estimated the mean duration of cetuximab use in England as 24–25 weeks, ‘depending on chemotherapy backbone and disease progression’, citing the source as ‘data on file’ (Merck Serono submission, table 3, p. 17). The submission stated that ‘The period of treatment with cetuximab plus chemotherapy used in the model were obtained from the relevant clinical trials. As stated in the clinical evidence section, the period of treatment in the clinical trial represents clinical practice as Merck Serono research indicates that the period of cetuximab treatment is 25 weeks on average’ (section 3.7.2, p. 64).

In its model, Merck Serono assumed that all patients take first-line drug treatment while in PFS, up to a certain cut-off time, which varies slightly by treatment arm. After the cut-off time, patients take no first-line drug. The cut-off times were:

• cetuximab plus FOLFOX – 5.5 months

• FOLFOX – 5.5 months

• cetuximab plus FOLFIRI (vs. FOLFIRI) – 5.8 months

• cetuximab plus FOLFIRI (vs. bevacizumab plus FOLFIRI) – 4.8 months

• FOLFIRI – 5.9 months

• bevacizumab plus FOLFIRI – 5.3 months.

Under its method of modelling treatment duration, we calculate that Merck Serono estimates the following mean treatment durations:

• cetuximab plus FOLFOX – 4.9 months

• FOLFOX – 4.6 months

• cetuximab plus FOLFIRI (vs. FOLFIRI) – 5.3 months

• cetuximab plus FOLFIRI (vs. bevacizumab plus FOLFIRI) – 4.5 months

• FOLFIRI – 5.2 months

• bevacizumab plus FOLFIRI – 5.1 months.

Below, we argue that these are underestimates (see Critique of the Merck Serono model, Model parameters, Time on treatment).

Second-line progression-free survival: unresected patients

Both we and Merck Serono assumed that all patients have second-line FOLFIRI after first-line FOLFOX-based treatment and second-line FOLFOX after first-line FOLFIRI-based treatment.

Merck Serono modelled second-line PFS using data from the study by Tournigand et al. 111 Inspection of its model revealed that it assumed a log-logistic distribution, and we calculated a mean of 0.31 years in second-line PFS for patients who start on second-line treatment. Merck Serono assumed that this value was independent of first-line treatment (whether FOLFOX or FOLFIRI based).

Given the lack of data to the contrary, we also assume that PFS on second-line FOLFOX or FOLFIRI is independent of first-line treatment.

Although not stated in its report, and in common with us, inspection of its model revealed that Merck Serono assumed that patients take FOLFOX or FOLFIRI for the entire duration of second-line PFS.

Third-line survival: unresected patients

In common with us, Merck Serono modelled third-line survival using data from the study by Jonker et al. 112 Inspection of its model revealed that it assumed a Weibull distribution, and we calculated a mean of 0.74 years’ survival for patients who start on third-line treatment. Merck Serono also assumed that this value was independent of first- or second-line treatment.

Merck Serono assumed that most patients receive BSC in third line, with 17% receiving capecitabine or cetuximab. It further assumed that patients would not be retreated with cetuximab.

Utilities

The utilities used in Merck Serono’s model are reported in Table 68. We noted that there were differences between the utilities in the main report and those reported in appendix B. The values in the appendix correspond to those used in the model.

TABLE 68 - Health state utilities reported by Merck Serono

NR, not reported.

Source: Merck Serono submission, table 20, pp.50–1; appendix B, table 1, p. 1.

Health state	Merck Serono main report	Merck Serono model (and appendix B)	Source
First line	0.77	0.778	Bennett et al.115
Second line	0.73	0.769	Bennett et al.115
Third line	0.68	0.663	Wang et al.116
PFS post resection	NR	0.789	Petrou and Hockley 2005117
Progressive disease post resection	NR	0.682	Average of second- and third-line utilities, weighted by time spent in second and third line

No RAS WT utility data were identified by Merck Serono or reported by the included trials. Merck Serono used the study by Bennett et al. 115 to obtain estimates of utilities in first- and second-line treatment. Bennett et al. 115 report utilities for first- and second-line KRAS WT mCRC populations. Further discussion of this source can be found in Chapter 6 (see Model parameters, Utilities). For first-line utility, Merck Serono used the estimate of first-line utility reported at baseline for the panitumumab plus FOLFOX population (0.778). For second-line utility, Merck Serono used the second-line baseline results for panitumumab plus FOLFIRI (0.769).

Merck Serono used an estimate of 0.663 from Wang et al. 116 for third-line treatment. This utility was for a previously treated KRAS WT mCRC population who were receiving BSC. This source is also discussed further in Chapter 6 (see Model parameters, Utilities).

Merck Serono used a general population estimate for the utility of PFS post resection. The source of this value was the study by Petrou and Hockley,117 which used Health Survey for England data from 1996. More recent data and approaches for using these data are available. 130,131

For post-resection progressive disease states, the utility was assumed to be a weighted average of second-line and third-line health states, adjusted for time in each state.

Costs

RAS mutation testing

Merck Serono reported a cost of £200 for RAS mutation testing from the All Wales Genetic Laboratory (Merck Serono submission, appendix B, table 2), which was applied to all arms of the model, regardless of treatment.

Drug acquisition

Merck Serono assumed costs for drug acquisition per month as shown in Table 69.

TABLE 69 - Drug acquisition costs per month in Merck Serono’s model

BEV, bevacizumab; CET, cetuximab.

Regimen	Cost per month of drug acquisition (£)
CET + FOLFOX4	5083
FOLFOX4	1546
FOLFOX6 (second line only)	1616
XELOX	1950
CET + FOLFIRI	4876
BEV + FOLFIRI	3345
FOLFIRI	1339

These monthly costs were calculated based on the pharmaceutical costs shown in Table 70, all of which are list prices and do not include any discounts that may be obtained by the NHS.

TABLE 70 - Costs of pharmaceuticals in Merck Serono’s model

BNF, British National Formulary.

Version of BNF used not stated.

Source: Merck Serono executable model.

Agent	Cost (£)	Source
Cetuximab	20-ml vial (5 mg/ml): 178.10	Merck Serono
	100-ml vial (5 mg/ml): 890.50
Bevacizumab	4-ml vial (25 mg/ml): 242.66	BNF132
	16-ml vial (25 mg/ml): 924.40
Oxaliplatin	10-ml vial (5 mg/ml): 155.00	BNF132
	40-ml vial (5 mg/ml): 622.38
Fluorouracil	10-ml vial (50 mg/ml): 6.40	BNF132
	50-ml vial (50 mg/ml): 32.00
Leucovorin	10-tablet (15-mg) pack: 19.41	BNF132
Irinotecan	2-ml vial: 46.50	BNF132
	5-ml vial: 114.00
	25-ml vial: 601.25
Capecitabine	60-tablet (150-mg) pack: 40.00	BNF132
	120-tablet (500-mg) pack: 295.65
Doxycycline	8-tablet (100-mg) pack: 1.11	BNFa
Ondansetron	30-tablet (4-mg) pack: 5.37	BNFa
Dexamethasone	50-tablet (2-mg) pack: 7.05	BNFa

For each agent in each regimen, the target dosage was calculated based on an assumed constant body surface area or body mass (Table 71) and then wastage was considered by using the minimum number of vials to achieve the minimum wastage; for example, for a target cetuximab dose of 895 mg, two 500-mg vials would lead to wastage of 105 mg, whereas one 500-mg vial and four 100-mg vials would lead to wastage of 5 mg (in which case the latter was assumed). Wastage was not minimised based on cost, but if the average cost per milligram was the same across vial sizes (or very similar), this method will minimise cost. It was assumed that for all regimens there would be 2.17 cycles per month, which is accurate for 14-day cycles.

TABLE 71 - Methodology used by Merck Serono to calculate the monthly costs of regimens

CET, cetuximab; BEV, bevacizumab.

Regimen	Agent	Cycles per month	Dosage per cycle	Cost per cycle (£)	Monthly cost (£)
CET + FOLFOX4	CET	2.17	500 mg/m2	1616.37	3507.52
	FOLFOX4	See below			1546.45
	Doxycycline	2.17	200 mg	1.11	2.41
	Ondansetron	2.17	8 mg	7.05	15.30
	Dexamethasone	2.17	8 mg	5.37	11.65
	Total				5083.33
FOLFOX4	Oxaliplatin	2.17	85 mg/m2	622.38	1350.56
	Leucovorin	2.17	200 mg/m2	58.23	126.36
	5-fluorouracil	2.17	1600 mg/m2	32.04	69.53
	Total				1546.45
FOLFOX6	Oxaliplatin	2.17	100 mg/m2	622.38	1350.56
	Leucovorin	2.17	200 mg/m2	58.23	126.36
	5-fluorouracil	2.17	2800 mg/m2	64.02	138.92
	Total				1615.85
XELOX	Capecitabine	2.17	28,000 mg/m2	245.94	533.69
	Oxaliplatin	2.17	130 mg/m2	652.90	1416.79
	Total				1950.50
CET + FOLFIRI	CET	2.17	500 mg/m2	1616.37	3507.52
	FOLFIRI	See below			1339.04
	Doxycycline	2.17	200 mg	1.11	2.41
	Ondansetron	2.17	8 mg	7.05	15.30
	Dexamethasone	2.17	8 mg	5.37	11.65
	Total				4875.92
BEV + FOLFIRI	BEV	2.17	5 mg/kg	924.40	2005.95
	FOLFIRI	See below			1339.04
	Total				3344.99
FOLFIRI	Irinotecan	2.17	180 mg/m2	456.00	989.52
	Leucovorin	2.17	400 mg/m2	97.05	210.60
	5-fluorouracil	2.17	2800 mg/m2	64.02	138.92
	Total				1339.04

Merck Serono’s model allowed for both weekly and fortnightly administration of cetuximab, but we present only the parameter values for fortnightly administration because we believe that this is a more appropriate base case as it reflects current clinical practice more closely.

Merck Serono assumed premedication with doxycycline, ondansetron and dexamethasone prior to cetuximab administration, but these did not significantly contribute to overall costs.

Merck Serono did not include any adjustments for mean dose intensity; in practice, some patients would likely require reductions in their target dose (often because of side effects).

Drug administration

The drug administration unit costs used in Merck Serono’s economic model are provided in Table 72. The report differed from the model in that appendix B appeared to report inpatient and outpatient costs the other way around.

TABLE 72 - Merck Serono’s drug administration unit costs

OP, outpatient.

Administration setting	Visit number	Unit cost (£)	Source
Inpatient chemotherapy administration	First visit	287	NHS Reference Costs 2012–2013:133 SB14Z [OP]
	Subsequent visits	255	NHS Reference Costs 2012–2013:133 SB15Z [OP]
Outpatient chemotherapy administration	First visit	226	NHS Reference Costs 2013–2014:133 SB14Z [OP]
	Subsequent visits	314	NHS Reference Costs 2013–2014:133 SB15Z [OP]

It was not stated in Merck Serono’s report how these unit costs were used and so it was necessary to determine this from the executable model.

Merck Serono assumed that the ‘first visit’ cost applied to the whole of the first cycle and that the ‘subsequent visits’ cost applied to all subsequent cycles, that is, even if a patient had multiple attendances per cycle, only one attendance was costed. Drug administration costs were consistent across all regimens per cycle and all regimens were assumed to have 2.17 treatment cycles per month (including XELOX).

Merck Serono also assumed that drug administration for first-line treatment was carried out 100% of the time in the outpatient setting and for second-line treatment was carried out 100% of the time in the inpatient/day-case setting.

In summary, total drug administration costs per month in Merck Serono’s model were £633.38 (first month) or £681.38 (subsequent months) for first-line treatments, and £585.35 (first month, except XELOX) or £553.35 (subsequent months, all months for XELOX) for second-line treatments.

Medical management

The executable model submitted by Merck Serono used resource use and unit costs for medical management as shown in Table 73. As can be seen, Merck Serono assumed no medical management costs in three health states (first-line progression free, second-line and post-resection progression free), a cost of £315 per month for post-resection progressive disease and a cost of £1040 per month for third-line treatment (mainly BSC).

TABLE 73 - Medical management costs in the model submitted by Merck Serono

CEA, carcinoembryonic antigen; CA 19–9, carbohydrate antigen 19–9; CT, computerised tomography.

Merck Serono stated that these were intended to be the resource use values for the first month only, but they were applied throughout in the executable model submitted.

Health state	Item	Unit cost (£)	Resource use (per month)	Monthly cost (£)
First-line progression free				0
Second line				0
Third line	BSC costs			997
	Capecitabine monotherapy	Per month per patient receiving: 246	17.5% of patients	43
	Total			1040
Post-resection progression free				0
Post-resection progressive disease	Evaluation of tumour markers: CEA	60	1a	60
	Evaluation of tumour markers: CA 19–9	60	1a	60
	Liver function tests	28	1a	28
	Hepatic ultrasonography	51	1a	51
	Oncology outpatient attendance	333	0.25a	83
	Abdominal CT scan	90	0.125a	11
	Lung CT scan	90	0.125a	11
	Large bowel CT scan	90	0.125a	11
	Total			315

Resection cost

Merck Serono specified in the submission (appendix B, table 2) that the average cost of liver resection surgery assumed in the model was £2707. Merck Serono stated that this cost was derived from NHS reference costs 2013/14134 [Healthcare Resource Group (HRG) unit costs for hepatobiliary and pancreatic surgery in malignant gastrointestinal disorders]. It represents the average of the HRG unit costs weighted by the number of finished consulting episodes (Merck Serono submission, appendix B, table 2). The relevant HRG unit costs are detailed in table 3 of appendix B.

Notably, national average unit costs for the HRGs used to estimate the average cost of liver resection in Merck Serono’s model (Merck Serono submission, appendix B, table 3) are not consistent with the NHS reference costs 2013/2014. 134 The average cost of liver resection based on the actual average unit costs reported for these HRG codes is £2467.

Costs post resection

Follow-up consultations

Merck Serono assumed a cost of £333 per oncological outpatient attendance. In the executable model it reported the source of this cost as NHS Reference Costs 2012–2013,133 but we could not confirm this cost.

The frequency of follow-up consultations in Merck Serono’s model was one visit per 4 months, as in the study by Adam et al. 113 We agree that this is appropriate.

Blood tests

Merck Serono modelled the following blood tests in patients post resection: liver function test and the tests for the tumour markers carcinoembryonic antigen and carbohydrate antigen 19–9 (appendix B, table 2).

The submission states that the cost of the liver function test was £28.76 (in 2013 UK pounds). However, in the executable model Merck Serono used a cost of £27.60 per test (in 2013 UK pounds). This cost was based on the TA17610 (appendix B, table 2) and we believe that this source is appropriate.

Merck Serono assumed that each tumour marker test cost £59.87, based on information from the Information Services Division (ISD) Scotland. We were unable to identify this source and so cannot comment on its relevance.

In the manufacturer’s model, the blood tests are conducted during the first month after resection and then every 4 months, based on the study by Adam et al. 113 Based on advice from our clinical expert (Dr Mark Napier, 2015, personal communication), we believe that this cost should be incurred every 3 months.

Despite the difference in cost and frequency of blood tests between our estimates and those of Merck Serono, we found that using our values in place of Merck Serono’s resulted in a minimal change to the cost-effectiveness results.

Imaging tests

Merck Serono modelled hepatic ultrasonography and computerised tomography (CT) scans in patients post resection. The cost of the hepatic ultrasonography test was assumed to be £51 (Merck Serono submission, table 2, appendix B) and it was assumed that this test was conducted during the first month after the surgery and then every 8 months. Merck Serono modelled abdominal, lung and large bowel CT scans separately, at a cost of £90 per test (Merck Serono submission, table 2, appendix B). The tests were assumed to be performed every 8 months.

Merck Serono stated that the above cost estimates were based on NHS reference costs 2012/13;133 however, we could not confirm these estimates.

We note that, although Merck Serono calculated different costs for the first month after resection and for subsequent months, based on changes in resource use, it did not implement these correctly in the model and instead use the first month costs throughout.

Adverse events

Merck Serono modelled costs and disutilities of grade 3/4 AEs. The probability of an AE was taken directly from each of the relevant trials and for some AEs these came from a KRAS WT rather than a RAS WT population. Merck Serono assumed that all AEs lasted for 1 month.

The costs and disutilities associated with each AE are reported in Table 74. Periphery sensory neuropathy and vomiting have disutilities but no costs.

TABLE 74 - Adverse event utilities and unit costs used in the Merck Serono model

BNF, British National Formulary; CC, complications and comorbidities.

Source: Merck Serono submission, appendix B, table 1, p. 1, and table 4, p. 5.

AE	Cost (£)	Source provided	Utility decrement	Source
Hypertension	622	NHS Reference Costs 2013–2014:134 non-elective inpatient stay – EB04Z – hypertension	–0.069	Doyle et al.135
Gastrointestinal perforation	2693	NHS Reference Costs 2013–2014:134 FZ38K – gastrointestinal bleed with single intervention with a CC of score 5–7	–0.195	Tolley et al.136
Arterial thromboembolism	777	NHS Reference Costs 2013–2014:134 deep-vein thrombosis with a CC score of 3–5 – QZ20D	–0.195	Tolley et al.136
Venous thromboembolism	777	NHS Reference Costs 2013–2014:134 deep-vein thrombosis with a CC score of 3–5 – QZ20D	–0.195	Tolley et al.136
Skin reactions	13.09	BNF132	–0.03248	Nafees et al.137
Neutropenia	877	NHS Reference Costs 2013–2014:134 non-elective inpatient stay – PA45Z – medical oncology	–0.09	Nafees et al.137
Diarrhoea	153	NHS Reference Costs 2013–2014:134 general medicine outpatient visit – service code 300	–0.103	Lloyd et al.138
Leukopenia	153	NHS Reference Costs 2013–2014:134 general medicine outpatient visit – service code 300	–0.03248	Assumption: equal to disutility for neutropenia
Periphery sensory neuropathy			–0.116	Lloyd et al.138
Fatigue	153	NHS Reference Costs 2013–2014:134 general medicine outpatient visit – service code 300	–0.115	Lloyd et al.138
Vomiting			–0.103	Lloyd et al.138
Neurological toxicities	1400	NHS Reference Costs 2013–2014:134 WA17A medical oncology neoplasm-related admission with a CC score of 3+	–0.116	Assumption: equal to disutility for peripheral sensory neuropathy
Hypokalaemia	153	NHS Reference Costs 2013–2014:134 general medicine outpatient visit – service code 300	–0.115	Assumption: equal to disutility for fatigue

The cost sources for AEs were poorly reported. We were unable to confirm the source of the costs for hypertension, arterial thromboembolism, venous thromboembolism, neutropenia or neurological toxicities.

The disutility estimates for AEs were better reported and came from a range of published literature. 135–138 All of these sources were UK-based studies, using EQ-5D vignettes, but none was conducted on a CRC population, and some studies reported on the EQ-5D visual analogue scale and some reported on the EQ-5D time trade-off scale.